Functional structure design of new high …...Functional structure design of new high-performance materials via atomic design and defect engineering (ADDE) edited by Prof. David Rafaja

Functional structure design of new

high-performance materials via atomic

design and defect engineering (ADDE)

edited by

Prof. David Rafaja

Imprint

Copyright: Technische Universität Bergakademie Freiberg, Spitzentechnologiecluster ADDE

Publishing: SAXONIA Standortentwicklungs- und -verwaltungsgesellschaft mbH

Technical editing: Alexander Eisenblätter, Dr. Uta Rensch

Layout: Alexander Eisenblätter, Susann Müller

Print: SDV Direct World GmbH, Dresden

All rights reserved

No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recpording or otherwise, without written permission from the Publisher.

The authors are responsible for the content of their publication as well as completeness and correctness of literature references cited. The publisher has performed only editorial changes to the original manuscripts.

ISBN 978-3-934409-68-2

Preface

Modern industry requires instantly new materials with tailored properties and energy efficient technologies for their production. Technical University Bergakademie Freiberg is one of the European universities, which are permanently active both in the materials science and engi-neering and in the development of modern technologies for the advanced materials produc-tion. Based on this long-time tradition of the Freiberg University, the development of modern high-performance materials with high functionality and efficiency for applications in the fields of communication, mobility, energy and environment was naturally the main goal of the Centre of Excellence “Functional structure design of new high-performance materials via atomic de-sign and defect engineering (ADDE)”, which was established in 2009 and funded until the end of 2014 by the European Regional Development Fund (ERDF) and by the Ministry of Science and Art of Saxony.

The main idea of the Cluster of Excellence ADDE was to control the crystal structure and mi-crostructure defects in order to tailor the properties of materials. This book presents an over-view of the results of 19 interdisciplinary projects, which dealt with the generation and manip-ulation of desired microstructure features in functionalised materials like metastable phases, controlled phase decomposition, precipitation and nano-sized structures, or with the elimina-tion of unwanted defects in large crystals intended for special electronic applications like for-eign atoms or dislocations. The main aim of the Cluster of Excellence ADDE was to understand the interactions between individual crystal defects and microstructure features as a first step towards targeted defect engineering. The choice of the materials was stimulated by the topics, which are established at the TU Bergakademie Freiberg and at the principal cooperation part-ners, which were the Helmholtz Centre Dresden Rossendorf and the Leibnitz Institute for Solid State and Materials Research Dresden. As the education and training of young professionals and academics for the Saxonian industry and research was one of the central tasks of the Centre of Excellence ADDE, a close cooperation with the local industry played a very important role.

The contributions in this book are divided into four groups. The first one is devoted to the technologies for production of thin silicon solar cells. It comprises the growth and processing of silicon ingots, the methods for their characterisation and the technologies for recycling of sawing slurries. In the second group of the contributions, selected materials for microelectron-ics, information storage and sensor technology are presented like the wide-gap semiconductor GaN, the dielectrics TiO2, SrTiO3, ZrO2 and TaZrOx, the azulene and anthraquinone structures for information storage and the conductive coordination polymers for transducers in sensor applications. In the third part of this book, protective coatings for wear reduction and corrosion protection are discussed with a special focus on the use of metastable phases. The last group of the contributions is dedicated to the mechanical properties and thermodynamics of light metal alloys.

David Rafaja November 2015Coordinator of the ADDE project

Contents

TECHNOLOGIES FOR PRODUCTION OF SILICON SOLAR CELLS

Ultra-short time processing of silicon solar cellsS. Prucnal, F. L. Bregolin, K. Krockert, H. J. Möller, W. Skorupa

Growth and characterization of multi-crystalline silicon ingotsE. Schmid, C. Funke, Th. Behm, S. Würzner, O. Pätzold, V. Galindo, M. Stelter, H. J. Möller

Experimental and numerical investigations of the formation of surface defects during machining of silicon wafersM. Budnitzki, T. Behm, M. Kuna, H. J. Möller

Development of a new process for recycling of used sawing slurries from solar industryI. Nitzbon, A. Obst, U. Šingliar, M. Bertau

MATERIALS FOR MICROELECTRONICS AND SENSOR TECHNOLOGY

Defect engineering in GaN layers grown by hydride vapor phase epitaxyG. Lukin, O. Pätzold, M. Stelter, M. Barchuk, D. Rafaja, C. Röder, J. Kortus

Strontium titanate – Breaking the symmetryH. Stöcker, J. Hanzig, F. Hanzig, M. Zschornak, E. Mehner, S. Jachalke, D. C. Meyer

Atomic layer deposition of dielectric thin films in the ternary system TiO2-SrTiO3B. Abendroth, S. Rentrop, W. Münchgesang, H. Stöcker, J. Rensberg, C. Ronning, S. Gemming, D. C. Meyer

Synthesis and characterization of Ge nanocrystals embedded in high-k materials for alternative non-volatile memory devicesD. Lehninger, P. Seidel, M. Geyer, F. Schneider, A. Schmid, V. Klemm, D. Rafaja, J. Heitmann

Novel molecular materials for information storage – Synthesis, electronic properties and electrode designM. Mazik, E. Weber, N. Seidel, S. Förster, E. Kroke, J. Wagler, A. Kämpfe, J. Kortus, T. Hahn, S. Liebing, Y. Joseph, R. Dittrich

Development of an electrically conductive coordination polymer based transducer for sensor applicationsM. Günthel, J. Hübscher, F. Katzsch, R. Dittrich, Y. Joseph, E. Weber, F. Mertens

6

26

42

64

78

100

118

134

146

166

PROTECTIVE COATINGS AND HARD MATERIALS

On the thermal stability of nanoscaled Cr/ta-C multilayersU. Ratayski, Ch. Schimpf, T. Schucknecht, U. Mühle, C. Baehtz, M. Leonhardt, H.-J. Scheibe, D. Rafaja

Defect engineering in Ti-Al-N based coatings via energetic particle bombardment during cathodic arc evaporationCh. Wüstefeld, M. Motylenko, D. Rafaja, C. Michotte, Ch. Czettl

Experimental and numerical assessment of protective coatings deposited by high velocity oxygen fuel flame spraying: Spraying process and thermo-mechanical behaviorS. Roth, M. Hoffmann, C. Skupsch, M. Kuna, H. Biermann, H. Chaves

Synthesis, properties and potential applications of rocksalt-type aluminium nitride (rs-AlN)K. Keller, M. R. Schwarz, S. Schmerler, E. Kroke, G. Heide, D. Rafaja, J. Kortus

MECHANICAL PROPERTIES AND THERMODYNAMICS OF LIGHT METAL ALLOYS

Influence of multi-pass roll-bonding on the mechanical properties of twin roll cast magnesium sheetsF. Schwarz, St. Reichelt, L. Krüger, R. Kawalla

Mg-Al composite wiresE. Knauer, J. Freudenberger, A. Kauffmann, L. Schultz

A unified approach to identify material properties from small punch test experimentsM. Abendroth

Atomistic modeling of defects in the framework of the modified embedded-atom methodS. Groh

Thermodynamic investigations in the ternary Al-Ti-Cr systemM. Kriegel, O. Fabrichnaya, D. Heger, D. Chmelik, D. Rafaja, H. J. Seifert

List of authors

184

200

224

242

260

278

288

306

328

346

Ultra short time processing of silicon solar cells

S. Prucnal 1, F. L. Bregolin 1, K. Krockert 2, H. J. Möller 2 and W. Skorupa 1

1 Institute of Ion Beam Physics and Materials Research, Helmholtz-Zentrum Dresden-Rossendorf, P.O. Box 510119, 01314 Dresden, Germany2 Institute of Experimental Physics, TU Bergakademie Freiberg, Leipziger Str. 23, 09599 Freiberg, Germany

6

AbstractThe research highlights for the further development of silicon based solar cell technologies focus on the cost reduction by applying inexpensive materials such as Solar Grade Multicrys-talline Silicon (SoG mc-Si) and/or the simplification of the production process. Replacement of standard diffusion based doping by ion implantation reduces two of the solar cell produc-tion steps: elimination of the phosphosilicate glass (PSG) cleaning and edge isolation steps. For this propose the highly efficient plasma based ion implantation system for phosphorous doping was installed at the HZDR. Although ion implantation doping got very recently dis-tinct consideration for doping of monocrystalline solar material, efficient doping of multic-rystalline solar material remains the main challenge to reduce the costs. The usefulness of the plasma immersion ion implantation system (PIII) combined with advanced flash lamp anneal-ing (FLA) was already validated [1]. We have shown that within the millisecond annealing time, implanted phosphorous is electrically activated and silicon is recrystallized. Simultane-ously, the diffusion of metal impurities and their activation is suppressed. The PIII togeth-er with FLA in the ms-range is demonstrated as a very promising technique for the emitter formation at an overally low thermal budget. Within this chapter we are going to validate the application of ion implantation and flash lamp annealing for the doping of solar cell materi-als, which are compact and technology compatible solutions for efficient silicon based solar cell formation. Finally, a new concept for the silicon based solar cell fabrication is presented. Keywords: PIII, FLA, Silicon, solar cells, ion implantation

7

Introduction – state of the art The global photovoltaic (PV) solar industry has been growing about 70 % per year over the past 5 years [2]. At present, the main effort in the solar cell industry is directed to the “costs per watt” reduction of the solar panels fabri-cation. According to the “Roadmap for Ener-gy 2020”, the average price per kWh should be reduced from the present 0.20 €/kWh to 0.07 €/kWh in 2020. It can be done by reduction of the overall wafer thickness, increase of the so-lar cell efficiency and/or simplification of the production complexity.

Another approach towards cost reduction of PV panels is realized by implementing thin film semiconducting absorbers. The main-stream PV technology improvements aim at cost reduction of the modules, whose costs roughly split in half between solar cell mod-ule production and balance of system fabri-cation. Half of the fabrication costs consist of the silicon substrate price. The other half is shared between cell processing (20 %) and module processing (30 %) [3]. The efficiencies of commercial silicon PV cells with standard cell structures are in the range of 16-19 % for monocrystalline and up to 17 % for multic-rystalline silicon substrates.

The simplest method for cost reduction re-lates to the use of less expensive silicon sub-strates for solar cell fabrication, since half of the fabrication costs are spent on it. This can be realized by using thinner and thinner sili-con substrates. Between 1979, when the first silicon solar cell was constructed, and now-adays, the thickness of the silicon substrate decreased from 500 to 160 µm [2] today, and it is predicted to decrease further to 100 µm within a decade. Reducing the silicon thick-ness obviously leads to a strong advantage for cost reduction. Replacement of the standard silicon wafers (160 µm thick) by much thin-ner ones into present production lines re-quires overcoming a number of technological challenges. Surprisingly, solar cells made of thinner wafers exhibit better device perfor-mance. The highest VOC of 747 mV for single silicon junction was obtained for 60 µm thick wafers, while 50 µm thick silicon solar cells show efficiencies up to 21.5 % [4, 5]. Howev-er, due to high brittleness of such thin silicon substrates, a new method of wafer handling at module level during solar cell processing has to be developed. In addition, increas-ing the average wafer size has a positive in-

fluence on the total production costs due to the reduction of the amount of handling and the total number of cells required for a giv-en power. The cell sizes have increased from 75 x 75 mm2 in 1979 (first silicon based solar cell) to 150 x 150 mm2 today, and this tenden-cy seems to continue. Moreover, a significant cost reduction can be obtained by replacing monocrystalline wafers by much cheaper multicrystalline ones, e.g. Solar Grade mc-Si. However, due to transition metal impurities, point defects and grain boundaries, the best polycrystalline silicon solar cells, as the ones developed by Fraunhofer ISE, show energy conversion efficiencies of 20.3 % [6], around 5 % less than the maximum efficiency ob-tained from monocrystalline silicon solar cells. Industrial polycrystalline cells offer ef-ficiencies of 15-17 %, which are around only 1 % lower than for the monocrystalline cells fabricated on the same production line; and at the module scale, the efficiency of both types of solar cells is comparable, due to the higher packing factor for mc-Si. Taking into account the substrate prices, the mc-Si solar cells could become more and more popular.

Another approach towards cost reduction is realized by using thin silicon films deposited on glass, ceramics or flexible substrates, such as polymers or metal foils. Thin-film silicon solar cells with a nominal absorber thick-ness in the range of a few microns could be cheaper than conventional devices because they use far less material. Most of the thin sil-icon films used for the solar cell formation are polycrystalline, with grain sizes in the range of 0.1 – 100 µm. The grain size determines the minority carrier lifetime and VOC due to high-er carrier recombination velocity at the grain boundaries. Relatively low recombination ve-locities and hence high VOC can be obtained by defect passivation at the grain boundaries by hydrogen. So far, the highest conversion efficiency is achieved by thin film polycrys-talline silicon cells made by the solid phase crystallization (SPC) technique (10.5 %), with grain sizes in the order of 1-2 µm [7]. The main methods for polycrystalline silicon thin film formation on non-silicon substrates from

amorphous precursors are: SPC, aluminium induced crystallization, zone-melt recrystalli-zation, e-beam recrystallization, high temper-ature chemical vapour deposition, or general-ly, crystallization by ultra-short annealing i.e. laser or flash lamp annealing. The thin amor-phous silicon layers are mainly obtained by vacuum systems e.g. various chemical vapour deposition (CVD) or sputtering processes, which drive up the cost of silicon and dielec-tric film deposition. The costs of amorphous Si precursors could be significantly reduced if these films are produced using a liquid pre-cursor, since film coating and annealing pro-cesses are generally more efficient and much simpler than CVD or sputtering processes [8]. The hydrogenated polycrystalline silicon films can be formed by spin-coating or ink-jet printing of liquid silicon, followed by thermal annealing on any substrate. Recently, Masuda et al. have shown thin film amorphous silicon solar cell fabrication by employing a solu-tion-based process using polydihydrosilane solutions [9], for which the power conversion efficiency is low but promising. Even though the conversion efficiency at the current devel-opment stage is low, the presented technique seems to be promising for low-cost thin sil-icon film formation on any glass or flexible substrates. Another interesting approach for the thin silicon film PV is the epitaxy-free layer transfer method, proposed by IMEC [10]. A few micrometres quasi-epitaxial lay-er is made by high temperature annealing of a highly porous material in a hydrogen-rich atmosphere. The porosity is well controlled by deep-UV lithography and dry etching, while the layer thickness depends on the pitch depth and diameter.

Besides the costs reduction by using cheaper substrates (reduction of wafers thickness, thin silicon film, lower quality material), an en-hancement of the power conversion efficien-cy of the solar cells contributes to fulfil the EU Roadmap for Energy 2020 generated by renewable sources. Although the theoretical power conversion efficiency predicted for the monocrystalline silicon substrate is around 30 %, the highest efficiency, recently an-

8

nounced by Solar3D, Inc. and obtained from homojunction containing multiple micro 3D photovoltaic structures is 25.47 % [11], a 250 % improvement over the first silicon solar cell fabricated (1979). Simultaneously, the module price decreased from 70 $/kWh (current US dollars in 1979) to 0.20 €/kWh today. The cell efficiency is dependent on the VOC, JSC and, finally, on the fill factor (FF). The open circuit voltage and short circuit cur-rent depend on the minority carrier recom-bination, which is determined by the bulk properties of the silicon substrate: surface passivation quality, back surface field (BSF) quality, surface texturization, antireflective layer quality and emitter quality. Concur-rently, the fill factor mainly depends on the front metal contact quality and area. Hence, the key technologies for high efficiency sili-con solar cells focus on minimizing photon, carrier and electrical losses. One of the most advanced cells, besides 3D PV, are the passi-vated emitter rear localized cells (PERL), the heterojunction with intrinsic thin layer (HIT) and back-contact back-junction cells charac-terized by the power conversion efficiencies of 25.0 %, 23.0 % and 23.4 %, respectively [12-14]. In the case of polycrystalline silicon, the maximum power conversion efficiency is much lower when compared to monocrys-talline Si and it is in the range of 20.3 % for the PERL design [6]. Surprisingly, conversion efficiencies at the industrial module scale of solar cells made of both monocrystalline and mc-Si did not change such significant over the last decade and oscillate around 19.5 ± 2 % of module efficiency. The highly efficient solar cells are simply too complicated to be imple-mented into the current solar cell technology. The main development at the production level is to improve and simplify individual process-ing steps rather than the implementation of a high-efficiency but complicated technology to the mass production. Therefore, the main challenge in the future manufacture of solar cell modules is to find a compromise between high efficiency and low-cost technology. The solar-grade (SoG) mc-Si produced by upgrading metallurgical-grade silicon is an attractive material for low-cost solar cells.

The remaining impurities after the purifica-tion process, mainly transition metals, are the main obstacle towards highly efficient solar cells, effectively limiting the minority carrier lifetime. Here we propose a novel method for the purification of SoG silicon by a millisec-ond range low thermal budget internal getter-ing process. The solar cells were produced by Plasma Immersion Ion Implantation (PIII) of phosphorous using PH3 gas source and mil-lisecond range Flash Lamp Annealing (FLA). We have shown that diffusion of metal im-purities into the space charge region can be avoided by an one step FLA, only. It will be presented that the implanted phosphorous is electrically activated and all defects intro-duced into silicon during the ion implanta-tion process are removed while metal impu-rities are kept far away from the p-n junction region. The effect of hydrogen co-implanted with phosphorous on solar cell performance will be explored. FLA is demonstrated here as a very promising technique for the emitter formation in SoG silicon using a low thermal budget. Moreover, we present a new concept of the simplification of the Si-based solar cell processing by employing (only one) single step milliseconds annealing for entire solar cell fabrication. A tremendous advantage of our technology is that it can be directly trans-ferred to in-line production process leading to significant cost reduction and drop of the amount of chemicals used during solar cells manufacturing.

Methods and materials The p–n junctions were formed in mono- and multicrystalline p-type silicon wafers by phosphorus implantation and millisecond range FLA. Phosphorus was implanted with fluences up to 1×1016/cm2 by either a con-ventional ion beam implanter or by PIII. The ion beam implantation was performed at an energy of 120 keV through a 120 nm pro-tecting SiO2 layer. The oxide was deposited using the plasma enhanced chemical vapour deposition technique (PECVD). In the case of PIII, silicon wafers were implanted with-out the capping layer and phosphorous ions

9

were extracted from plasma using an electric field of 20 keV. The plasma was generated by a 500W RF power supply and consists of 5 % PH3 and 95 % H2. Therefore H is co-implanted into the silicon wafers together with phospho-rous. In both cases the projected range of the implanted phosphorus is about 30 nm from the silicon surface, while hydrogen can pene-trate silicon up to 300 nm. The depth profiles of the implanted P and H was calculated us-ing the SRIM 2003 code and experimentally confirmed by Rutherford backscattering spec-troscopy using 1.7MeV 4He+ projectiles at a backscattering angle of 170o and secondary ion mass spectrometry (SIMS), respectively. In SIMS a beam of 10 keV O2 ions was scanned over a surface area of 150×150 µm and sec-ondary ions were collected from the central part of the sputtered crater. Crater depths were subsequently measured using a Dektak 8 stylus profilometer, and a constant erosion rate was assumed for depth calibration. The H-concentration calibration was performed using an implanted reference sample. The sys-tematic H quantification was performed by the ERDA technique at the Ion Beam Center of the Helmholtz-Zentrum Dresden-Rossen-dorf. Hydrogen atoms recoiled from the sam-ple by a 1.7 MeV He+ beam provided by a 2 MV Van der Graaff were detected by a solid state Si barrier detector equipped with an Al stopper foil.

After implantation the samples were annealed at different temperatures ranging from 800 up to 1200 oC for the time scale from 1.3 ms up to 30 min. The millisecond range annealing was performed with the FLA system described in the next chapter. The FLA was done without or with preheating. The preheating was used in order to reduce an internal thermal stress in the annealed wafers appearing during the FLA process due to the strong temperature gradient between top and bottom of the wa-fer [15]. The temperature and preheating time was up to 600 oC and up to 5 min, respective-ly. For the annealing in the second or minute range the standard rapid thermal annealing (RTA) and furnace annealing (FA) was used. Structural and optical properties of silicon wa-fers were investigated by means of µ-Raman

spectroscopy (RS) and temperature depend-ent photoluminescence (PL). The µ-Raman spectra were recorded at room temperature in the backscattering geometry in the range 100 to 1000 cm-1 using 325 and 446 nm He-Cd la-ser and 532 nm NdYAG laser. The PL spectra were recorded at room temperature using the Jobin Yvon Triax 550 monochromator and a cooled InGaAs detector for detection of the near infrared emission or a photomultiplier (Hamamatsu H7732-10) in the visible spectra region. For the sample excitation during PL measurements a 405 nm or 532 nm laser light with a powers of 20 and 200 mW was applied, respectively. For the sheet resistance (SR) measurements the four-point probe resistor system (CMT-SR 3000 Advanced Instrument Technology) was used. Minority carrier dif-fusion length measurements were done by surface photovoltage spectroscopy using six lasers with different wavelengths (785, 809, 850, 904, 950, and 980 nm), allowing analysis penetration depths from 10 to 105 mm. The voltages at each wavelength and measurement point were recorded six times and averaged. Subsequently, a Goodman plot was con-structed from the average values in order to obtain the average minority carrier diffusion length (LD). The microstructural properties of fabricated solar cells were investigated by bright-field transmission electron microsco-py (TEM), high-resolution TEM (HRTEM), high-angle annular dark-field scanning TEM (HAADF-STEM), and energy dispersive X-ray spectroscopy (EDXS) in cross-sectional geometry employing an image-corrected FEI Titan 80–300 microscope operating at an ac-celerating voltage of 300 kV.

Flash Lamp Annealing system The FLA system employed in our experi-ments is illustrated schematically in Fig. 1. It consists of two annealing systems. At the top there are twelve 30 cm long Xe lamps spaced by about 3 cm representing together with the reflector the FLA system, and at the bottom a lower bank of halogen lamps allows the wa-fer to be preheated to a selected temperature – a type of rapid thermal annealing system. The shielding of the annealing chamber is

10

made of aluminium and it is water cooled. The top and bottom of the chamber is closed by quartz plates fully transparent in the visi-ble spectrum range. The FLA process is per-formed in an inert ambient, typically of ni-

Fig. 1: Schematic illustration of the flash lamp annealing system.

The energy of the light pulse deposited at the wafer surface (final temperature) is con-trolled by varying the capacitor charge. The maximum sample size which can be annealed in our system is limited to 300 mm. Usually, prior to the FLA process, the preheating step to an intermediate temperature of the flashed wafers is performed. This process is necessary to increase the uniformity of the heat redistri-bution over the entire wafer, thus minimizing the unwanted thermally induced stress due to the ultrafast temperature elevation. The maximum preheating temperature which can be obtained in our system is slightly above

950 oC and the system is calibrated for silicon, germanium and SiC wafers. In the case of sil-icon the final temperature during FLA can easily reach the melting point of silicon and the annealing time can vary from 0.8 up to 20 ms in a single shot with temperature vari-ation below 10 % through 200 mm diameter wafer. The maximum temperature in our FLA system was obtained during annealing of SiC and it reaches 2000 oC for 20 ms annealing. The schematic representation of the wafer temperature during FLA process with pre-heating and spectrum of Xe lamps are shown in Fig. 2 a and b, respectively.

trogen, argon or forming gas (95 % of N2 and 5 % of H2). The Xe lamps are energized by discharging a capacitor/inductor unit in the millisecond time scale.

Fig. 2: Change of wafer temperature during FLA process with preheating (a). The time of FLA is defined by the FWHM of flash pulse. The spectrum of Xe lamps mounted in our FLA system is shown in (b).

11

Depending on the application the FLA pro-cess can be performed with or without pre-heating and the annealing time is defined by the full width at half maximum (FWHM) of the flash pulse (see Fig. 2a). The maximum energy which can be delivered to the sample during a single 20 ms shot is 250 kJ. The spec-trum of Xe lamps used in our system is shown in Fig. 2b. The main emission falls in the visi-ble spectral range in between 350 and 800 nm. For effective annealing, FLA treated materials should have a high absorption coefficient in this spectral region, e.g. like silicon, perfectly suited for FLA processes.

The influence of the FLA on the microstruc-tural properties of implanted Si An alternative road to high temperature dif-fusion for silicon doping is ion implantation followed by short time thermal annealing. In the past, ion implantation had negligible meaning in solar cell production due to low efficiency in large scale mass production. Recently, Varian Solion offers a dual-magnet ribbon beam implant architecture for cost-ef-fective high efficiency solar cell production [16]. The key benefits of ion implantation in solar cell manufacture is cost reduction per watt through process simplification by elimi-nating phosphorous silicate glass (PSG) clean and edge isolation steps. An efficiency of 19 % from p-type silicon wafers implanted with phosphorous was announced at begin-ning of 2011 by Suniva [17]. However, after phosphorous implantation silicon is partial-ly or totally amorphised depending on the implantation fluence. To recrystallize silicon and activate the dopants samples are an-nealed by RTA or FA techniques at a temper-ature usually above 900 oC for a few minutes, which activates impurity metals in mc-Si as well. Therefore, ion implantation processing is successfully applied only to n- or p-type monocrystalline silicon.

To extend the application of ion implantation technique to mc-Si the diffusion of metal im-purities during recrystallization of silicon and electrical activation of phosphorous has to be suppressed. Here we present a new technique,

which allows electrical activation of implant-ed elements by a short time light pulse using a FLA tool. The FLA systems are successfully used for silicon recrystallization not only in research laboratories [18-21]. Also leading edge companies as Intel andformer AMD have been using similar flash annealing tools opti-mized for 300 mm in diameter semiconduc-tor grade wafers for the activation of highly activated ultra-shallow junction formation in their advanced micro-processor production for more than 10 years [22]. The main advan-tages of FLA over different annealing tech-niques are: (i) the FLA is a large scale facility, (ii) within a millisecond time range with a single energy pulse any kind of damage intro-duced to the crystal structure can be removed, and (iii) the thermal diffusion of foreign at-oms in the host is suppressed. Fig. 3 shows the depth distribution of phosphorous implanted into silicon using PIII and post-implantation thermal treatment. For comparison the depth distribution of P in silicon after standard ion beam implantation is shown as well. In both cases the implantation energy was fixed at 20 keV and the ion fluence was 1×1016 cm-2. Af-ter PIII the samples were annealed using FLA for 20 ms and RTA for 30 s at temperature of 1200 oC and 1000 oC, respectively.

Fig. 3: Depth distribution of phosphorous implanted into silicon using PIII before and after annealing obtained by SIMS. For com-parison the distribution of ion beam implanted phosphorous is shown as well (black curve). The inset shows the cross-section HRTEM image of a P implanted sample after FLA at 1100 oC for 20 ms.

12

During the PIII process, the final kinetic en-ergy of accelerated ions depends also on the charge transfer collisions in the dynamic plas-ma sheath [23]. This phenomenon leads to a broad energy distribution with an average en-ergy lower than the maximum by a few keV. In our system the average energy of phosphorous ions implanted at 20 keV is 16 ±4 keV (90 % of implanted ions) with a long tail on the low en-ergy part. Therefore the depth distribution of as-implanted P in silicon has a box-like shape and it starts just at the sample surface. While using an ion implanter the ion beam is mo-no-energetic and the distribution of implant-ed ions in as-implanted state is Gaussian like (see Fig. 3 black curve). The broadening of the depth distribution using ion beam implan-tation is mainly due to collision cascades in solids and interaction of implanted ions with hosts atoms. Independently on the used im-plantation techniques implanted ions intro-

Micro-Raman study The crystal quality of the silicon wafer after phosphorous implantation and annealing was controlled by Raman spectroscopy. Presented samples were implanted using PIII at an ener-gy of 20 keV and ion fluence of 1.6×1015 cm-2. The recrystallization evolution with depth af-ter various post-implantation annealing steps was investigated by µ-Raman spectroscopy using different lasers: a He-Cd laser with ex-citation wavelengths at 325 and 442 nm, and a Nd:YAG laser operating at 532 nm. Due to different absorption coefficients for different

duce defects and generate damage to the crys-tal lattice. Finally, for a certain fluence and ion energy a thin amorphous layer can be formed. In order to restore the initial structure and ac-tivate dopants the post-implantation thermal treatment is used. The green curve in Fig. 3 shows the P distribution after RTA. The 30 s annealing at 1000 oC is sufficient to activate electrically P but simultaneously P diffuses deep into silicon. This significantly decreases the local doping concentration and makes it impossible to precisely control the p-n junc-tion depth, which is crucial for highly efficient silicon based solar cells. In contrary to RTA the FLA for 20 ms leads only to a slight diffu-sion of the implanted P and, at the same time, a full recrystallization of the implanted layer takes place (see inset Fig. 3). Moreover, the FLA is a low thermal budget technique mak-ing it attractive for the low cost production of solar cells.

photon energies in silicon, the Raman spectra obtained after excitation with different wave-lengths are recorded from different depth regions. The penetration depth of laser light for 325, 442 and 532 nm in silicon is in the range of 8, 200 and 900 nm, respectively [24]. Hence, after excitation with ultraviolet (UV) (325 nm), blue (442 nm) and green (532 nm) lasers the Raman spectra deliver information about the surface quality, the damaged region due to ion implantation and the bulk material, respectively.

Fig. 4: µ-Raman spectra from implanted and annealed samples after different laser excitation. Annealing conditions: FA-900 oC for 30 min, RTA-1000 oC for 30s, FLA-1000 oC for 3 or 20 ms with preheating at 400 oC for 1 min. The excitation wavelengths during Raman measurements are indicated in the figures and the P fluence was 1.6×1015 cm-2.

13

Independent of the excitation wavelength, the Raman spectra of as-implanted samples show a broad band at around 470 cm-1 due to amor-phous silicon formed during ion implantation [25]. The projected range for plasma implant-ed phosphorus at 20 keV is around 30 nm be-low the silicon surface. Hence, taking into ac-count the energy distribution of P atoms, the degradation layer of the silicon due to ion im-plantation can extend up to 100 nm. Because the penetration depth of the blue and green laser light in amorphous silicon is slightly longer than the implantation depth the small peak visible at 520 cm-1 in the as-implanted sample is due to photons scattered from the crystalline silicon substrate. After annealing there is no significant difference in the Raman spectra between samples annealed at different conditions both for low and high implanta-tion doses. The silicon peak in Raman spec-tra (520 cm-1) obtained after UV excitation is not symmetric at the low frequency side. The asymmetry of the 520 cm-1 peak can be due to several reasons:

(i) Silicon was not fully recrystallized during annealing and a residual amorphous phase exists near the surface region.(ii) The surface region contains parti- ally disordered silicon due to a high amount of phosphorous incorporated into the silicon matrix.(iii) Fano interaction as a consequence of interference between discrete phonons and the continuum of elec- tronic states in heavily phosphorous doped Si [26].

Taking into account that the regrowth rate of damaged silicon is in the range of 1 nm/s at 600 oC and increases rapidly with tem-perature, the FA (900 oC, 30 min) and RTA (1000 oC, 30 s) samples should be fully recrys-tallized up to the surface [27]. Therefore the existence of amorphous silicon after annealing can be neglected. Here the asymmetry of the silicon peak is assigned to the Fano effect due to high doping and activation level of implant-

ed phosphorous. The broad peak observed at around 618±3 cm-1 is designated to the Si-B vibration mode (see Fig 4a-c) [28]. Its inten-sity and position depends on the free carrier concentration i.e. boron atoms in substitution position. With increasing boron concentra-tion, the peak position shifts towards higher wavenumbers [28]. For UV excited samples, only the surface region up to 10 nm contrib-utes to the Raman spectra. For this the sub-stitutional boron related peak is detected at 615 cm-1 and a strong asymmetry of the 520 cm-1 Si peak on the low energy side of the maximum is observed. By increasing the ex-citation wavelength to 442 or 532 nm when more bulk material contributes to the Raman spectra the peak position of the Si-B vibration mode shifts to 618 cm-1 or 620 cm-1, respec-tively. This means that within the implanted region P atoms are in substitutional position after annealing overcompensating positive intrinsic carriers. Furthermore, these results confirm that the FLA in the ms range is suf-ficient to recrystallize silicon and activate implanted phosphorous. Fig. 5a shows the influence of preheating temperature on the recrystallization process of implanted sili-con investigated by µ-Raman spectroscopy. According to the obtained data annealing at 400 oC for 3 min (standard preheating param-eters in our experiments) is enough to start the recrystallization of damaged silicon but the main part of silicon remains amorphous. Real recrystallization appears after annealing at 600 oC for at least 3 min which is proved by a single peak at 520 cm-1 corresponding to the single crystalline silicon (see Fig. 5a). Similar data were obtained for FLA already at 800 oC for 20 ms even without preheating suggesting that such a low flash energy is suf-ficient to recrystallize implanted silicon via a solid phase epitaxial regrowth (see Fig. 5b). The best crystal quality, highest carrier con-centration and the lowest resistivity were ob-tained after annealing at 1200 oC for 20 ms. However one has to keep in mind that the µ-Raman spectroscopy is only a local probe technique limited to the area of about 10 µm2 (in our experiments) and local lattice dis-tortions or some defects are not detected by

14

Raman. More information about the crystal-line quality of annealed silicon can be gained from photoluminescence spectroscopy where the existence of any kind of defects is crucial for the PL emission.

Fig. 5: The influence of preheating temperature (a) and FLA (b) on the recrystallization of P implanted silicon. The presented spectra were vertically shifted for clarity.

Opto-electrical properties of implanted and annealed silicon

The degradation level of the implanted layer due to defect formation as well as the recrys-tallization process strongly depend on the im-plantation and annealing parameters and can be determined using PL spectroscopy. In our study we have used a cw green laser operat-

ing at 532 nm with 200 mW excitation power. The laser spot was about 1 mm in diameter and the measurements were performed at room temperature (RT). Fig. 6 shows the RT PL spectra obtained from P implanted sample after different annealing.

Fig. 6: RT PL spectra obtained from crys talli-ne silicon implanted with P+ ions at a fluence of 1.6×1015 cm-2 and annealed at different tempera-tures and time. The preheating was done at 400 or 600oC for 3 min, while the FLA was performed at 800 or 1200 °C for 20 ms with and without pre-heating.

15

The broad peak at around 1140 nm corre-sponds to the band-to-band (B-B) transition in silicon and the narrow peak at 1064 nm is due to elastically scattered laser light from the sample [29]. The side band of the B-B with the maximum at around 1220 nm corresponds to the luminescence centers formed by phospho-rous ions. The PL intensity strongly depends on the crystal quality of the silicon matrix. Defects, such as atomic displacement, vacan-cies, dangling bonds etc., work as non-radi-ative channels for the PL de-excitation and hence reduce the PL efficiency. It is worth noting that many of these defects cannot be detected in the Raman spectrum but are easily visible in PL measurements. The as-implant-ed sample (not shown here) and the sample annealed at 400 oC for 3 min do not show PL emission at all (see Fig. 6). After FLA at 800 oC

for 20 ms, the PL measurements reveal only weak B-B luminescence. This means that the implanted region still contains many defects which cannot be detected by Raman spectros-copy. Independent of preheating, the highest PL intensity was obtained from samples an-nealed at 1200 oC for 20 ms, where the peak intensity only slightly depends on the preheat-ing. After ion implantation and annealing, the optical property of the wafer is homogeneous over the whole wafer area for the crystalline silicon but not for multicrystalline silicon. The structural and optical properties of the mc-Si depend on the grain size, local crystal-lographic orientation, grain boundaries and different impurities present in the Solar Grade material. The implanted mc-Si does not show photoluminescence at RT when the annealing temperature is lower than 600 oC.

Fig. 7: RT PL spectra obtained from mc-silicon implanted with P+ ions and annealed at different conditions. Annealing conditions: FA-900 oC for 30 min, RTA-1000 oC for 30s, FLA-1000 oC for 3 or 20 ms with preheating at 400 oC for 1 min.

The broad peak at around 1140 nm with the side band with a maximum at around 1220 nm corresponds to the B-B transition in sil-icon and luminescence centers formed by substitutional phosphorous atoms, respec-tively. The PL intensity strongly depends on the crystal quality of the silicon matrix and activation of different kinds of impurities act-ing as non-radiative channels for PL de-ex-citation and hence reducing the PL efficiency. It is worth noting that many of these defects cannot be detected in the Raman spectrum

but are easily visible from PL measurements. Independent of the implantation fluence the highest B-B PL intensity was observed from samples annealed for 3 ms and the lowest one for RTA. In our experiments, Solar Grade mc-Si with a relatively high concentration of met-al impurities were used. During long time an-nealing (RTA or FA) metals diffuse from the bulk towards the space charge region and are active there as non-radiative recombination centers and carrier traps. Both lead to a de-creasing PL intensity and finally kill the pho-

16

Figure 8a presents the average sheet resist-ance value as a function of different anneal-ing conditions for SoG silicon implanted by PIII. The lowest SR values were measured for FA and RTA treated sample and it was 48 and 62 Ω/sq, respectively. In case of FLA samples the SR is in the range of 80 ± 10 Ω/sq. The standard deviation of the average SR is in the range of 1.5 % for long term annealing and slightly above 2 % for flashed samples. Figure 8b shows the correlation between PL intensity and sheet resistance measured on monocrys-talline Si wafers. The SR strongly depends on the concentration of phosphorous activated in silicon during annealing and the defect densi-ty remaining after the annealing process. The defects have also a huge influence on the op-tical properties of the solar cell-like processed

tovoltaic effect. During the millisecond range annealing the thermal budget introduced to the sample is sufficient to activate implanted phosphorous but too short for the diffusion of metal impurities.

An important parameter for the solar cell per-formance is the sheet resistance of the emitter. From the application point of view it should

Fig. 8: Sheet resistance as a function of annealing parameters (a) obtained from mc-Si and comparison of PL intensity and SR for crys-talline samples annealed at different temperature with and without preheating (b). Samples were implanted using the PIII system with a P fluence of 2×1015 cm-2 and an energy of 20 keV.

silicon wafer. Comparing the integrated PL intensity (B-B transition in silicon) and sheet resistance measurements, we have found that the PL intensity is inversely proportional to the SR value. The maximum PL intensity was obtained from samples annealed at 1200 oC for 20 ms with 400 oC preheating for 3 min, at the same time the SR shows the minimum value (16 Ω/sq). This is a direct proof of high electrical activation efficiency of P in silicon and perfect recrystallization of the layer dam-aged by ion implantation.

The average minority carrier diffusion length (LD) is particularly sensitive to the quali-ty of the surface passivation, where high LD values are required for high performance solar cells. Fig. 9 a-d show the SPV scan of

be as low as possible in order to provide an ohmic contact between the metal electrode and silicon. However, low resistive material means high carrier concentration which caus-es strong Auger recombination and degrada-tion of the solar cell performance.

17

Fig. 9: SPV measurements obtained from SoG silicon implanted with P after different annealing processes: FA – (a), RTA – (b), FLA (20ms) – (c) and FLA (3ms) – (e).

3×3 cm2 area from samples implanted with a fluence of 1.6×1015 P+/cm2 and annealed at 900 oC for 30 min (a), 1000 oC for 30 s (b) and 1000 oC for 20 ms (c) and from 3 ms (e) with 400 oC preheating for 1 min. The average LD for 3×3 cm2 area is 9.55 µm for FA, 9.13 µm for RTA and 67.92 µm for 3 ms FLA. The inho-mogeneity of the average LD determined from SPV measurements is in the same range for all samples and does not exceed 10 %. It is worth noticing that the SPV measurements were per-

formed on non optimised samples. Further in-crease of the LD can be obtained by SiN layer deposition as antireflection coating. The ap-plication of the FLA technique in the low-cost solar cell production is an alternative route for the metal-impurity engineering proposed by Buonassisi et al. [30] or standard phosphorous gettering [31]. Buonassisi et al. have shown an influence of the cooling rate on the metal im-purity distribution in mc-Si.

18

Slowly cooled samples exhibit a low density of micrometre size defect clusters and a fourfold increase of LD (up to 50 µm) in comparison with fast cooled samples [30]. In our case, the FA and RTA annealed samples show the LD in the same range as that obtained by Buonassisi et al. for fast cooled samples (around 10 µm) and up to two times higher values after FLA

Hydrogen engineering via plasma immersion ion implantation and flash lamp annealing An additional implantation of H using PIII for the passivation of solar cell emitters exhib-ited promising results [32]. However, a more interesting approach of introducing hydro-gen into the Si-substrate is the simultaneous co-implantation during doping via PIII. By using phosphine (PH3) as a gas precursor in the PIII process, hydrogen is co-implanted with phosphorous used for the emitter for-mation in the Si substrate. After implanta-tion such samples were FLA treated. Due to

comparable to those of slowly cooled mc-Si. Based on the presented data we can conclude that the combination of Plasma Immersion Ion Implantation and ultra-short Flash Lamp Annealing enable a precise control of the dop-ing level, charge carrier distribution and emit-ter thickness – parameters which are crucial for the solar cell performance.

the ultra-short annealing times combined with the H co-implantation via PIII, a precise control of the hydrogen content in the silicon substrate is possible. Since hydrogen will be introduced during the doping step, it enables a wide choice of antireflection coatings pro-duced by different deposition methods, dis-regarding its H content. Data presented here is based on the solar cell emitter formed in p-type monocrystalline Si substrates by PIII of PH3 and subsequent FLA.

Fig. 10: SIMS depth distribution of H obtained from as-implanted and FLA treated samples at 1200 °C for 20 ms in Ar. H was co-implanted with phosphorous during the PIII doping process. Ex-planation of regions 1,2,3, see text.

Fig. 10 shows the depth distribution of H in silicon in the as-implanted state and after FLA for 20 ms at 1200 oC. It is worth noting that the plasma used for the doping is composed of 95 % of H2 and 5 % of PH3. Moreover, the PH3 molecules dissociate in the plasma into PH+

x (x€(0-3) and H+1-3 which ends with a strong

energy distribution for H ions. Therefore, the H depth profile is much wider than expected or can be calculated using the SRIM-Code for

hydrogen implanted with an energy of 20 keV [33]. The obtained depth distribution of H can be divided into 3 parts: (1) surface region with H peak concentration (Cp) in the order of 4.5×1021 H/cm3, (2) a second region with Cp in the range of 2.2×1021 H/cm3 due to hy-drogen co-implanted with P where the source of H are PH3 molecules and, (3) a region with Cp in the range of 3.8×1022 H/cm3. The pro-jected range of H located in region 3 fits to

19

the Rp for H2 implanted by PIII with an energy of 20 keV. Moreover, a part of H2 molecules can dissociate already in the plasma which leads to atomic H ion implantation. Overall, this leads to wide distribution of hydrogen in PIII treated silicon. The regions 1 and 2 are very important for the surface defects and grain boundaries within the emitter region, while the hydrogen implanted into region 3 can be used for the bulk defect passivation. After annealing the highest concentration of H is detected within the emitter region (re-gion 2), where the highest defect density due to implantation related damage is expected. That hydrogen is trapped mainly at the grain boundaries.

Fig. 11: Influence of the annealing ambient on the total H concen-tration in implanted and flash lamp annealed silicon wafers. FLA was performed at 1200 oC for 3 and 20 ms.

In order to systematically characterize the samples submitted to different preparation parameters, the ERDA technique was used to quantify the total amount of H present in the samples. The advantages of using ERDA when compared to more traditional meth-ods of H profiling such as resonant nuclear reaction analysis (r-NRA) are the possibili-ty of simultaneous multi-elemental analysis and throughput, at the expense of sensitivity and depth resolution [34]. As can be seen in Fig. 11, after the FLA treatments, the residual concentration of H still present in the samples strongly depends on the annealing time and atmosphere. For samples annealed for 3 ms, a significant drop in the H concentration is seen for the ones annealed in Ar. After annealing

in N2, the H loss is less pronounced due to the formation of a thin SixNy layer during the FLA that acts as a diffusion barrier, as indicated by visual inspection and the presence of a faint N signal in the ERDA spectra. For samples annealed in forming gas, the H concentration did not change significantly with the anneal-ing temperature. Since the forming gas has 5 % of H2 and 95 % of N2 in its composition, it can attenuate the H loss by acting as an H source during the FLA. No SixNy layer was de-tected in the samples annealed in forming gas, indicating that it is the hydrogen that is pre-dominantly bound at the surface during the passivation process. We have also investigated an influence of annealing ambient on the mi-nority carrier diffusion lengths. In the case of samples annealed for 20 ms at 1200 oC with preheating, i.e. if the maximum phosphorous activation takes place, the annealing ambient only slightly affects the LD. The best results were obtained for samples annealed in N2 and FG (~ 150 µm) while Ar has a slightly negative influence (~130 µm) [32].

OFOCell concept In order to meet the needs of EU for renewa-ble energy sources in 2020, the average price per kWh by PV generated electricity has to be as low as 0.07 €/kWh. It requires high-ef-ficiency solar cells (23 %) with low produc-tion costs per unit area. The cell efficiency is dependent on the VOC, JSC and, finally, on the fill factor (FF). The open circuit voltage and short circuit current depend on the minority carrier recombination, which is determined by the bulk properties of the silicon substrate: surface passivation quality, back surface field (BSF) quality, surface texturization, antire-flective layer quality and emitter quality. Con-currently, the fill factor mainly depends on the front metal contact quality and area. Hence, the key technologies for high efficiency sili-con solar cells focus on minimizing photon, carrier and electrical losses. To meet-in-the-middle we have developed a new concept for the solar cell fabrication which bases on the plasma immersion ion implantation and flash lamp annealing. The combination of PIII and

20

ultra-short FLA enables a precise control of the doping level, charge carrier distribution and emitter thickness – parameters which are crucial for the solar cell performance. Imple-mentation of the high-tech technology used in advanced microelectronics into PV man-ufacturing followed by the simplification of the solar cell production steps should provide significant cost reduction and allow achieving the objectives necessary for the high-efficien-cy low cost silicon solar cells. The developed

one-flash-one-solar-cell (OFOCell) concept enables a significant reduction of the overall thermal budget required for crystalline sil-icon solar cell manufacturing. The OFOCell concept allows the use of inexpensive metals for contact formation into c-Si solar cells, e.g. nickel instead of silver. Furthermore, since the AR layer is deposited after metallization, it can be freely chosen. The schematic overview of the OFOCell is shown is Fig 12.

Fig. 12: Schematic representation of the OFOCell concept with main production steps.

The main innovation is directly related to the single step millisecond range flash lamp an-nealing process. The annealing is performed for 20 ms at a temperature around 1100 °C. During such single step thermal treatment:

(i) the silicon layer amorphised during P implantation is recrystallized, (ii) phosphorous is electrically activated, (iii) the back and front ohmic contacts are formed, (iv) aluminium (used for back contact for mation) partially diffuses into silicon where it becomes electrically active leading to the back surface field formation,(v) the front ohmic contact between i.e. Ni and n-type silicon is generated.

Using our concept for the silicon solar cell fabrication the processing path is reduced by three steps compared to the standard process: Phosphorous thermal diffusion, PSG etching and edge isolation are no longer needed.

21

Fig. 13: EDXS spectra of the elemental composition of emitter region measured just below the Ni contact after FLA (a) and aluminium distribution on the back side of silicon wafer (b). The insets show the cross-section TEM micrographs of the emitter just below the metal contact (in a) and back side of Si wafer after Al diffusion (in b).

The formation of the emitter using PIII and FLA was described in detail previously (chap-ter 5). Therefore, here we will focus only on the metal contact formation, back surface field and metal impurity distribution in SoG mc-Si using the OFOCell concept. Fig. 13a shows the EDXS spectra obtained in the cross-section geometry from an area just below the Ni con-tact deposited through the mask. The width of Ni stripes was 300 µm. After FLA silicon even below the Ni stripe fully recrystallized due to the lateral heat transfer (see inset Fig. 13a) and the EDXS spectrum taken just below the Ni does not reveal Ni diffusion into silicon. It is worth mentioning that as the diffusion bar-

rier a 3 nm thick oxide layer was used. Due to phosphorous implantation the presented oxide is conductive and does not disturb the electrical properties of fabricated solar cell. In the standard solar cell fabrication process, the BSF is formed during the so called co-firing process. In our case, the aluminium should also diffuse into silicon and should be electri-cally activated within a single flash pulse. In fact, the EDXS investigation performed in the cross-section geometry reveals a strong diffu-sion of Al into silicon and Al was still detected 1 µm below back silicon surface.

Fig 14: Internal Quantum Efficiency (IQE) mapping by LBIC of Si solar cells, which were fabricated using the OFOCell concept (a) and I-V characteristic of an illuminated solar cell (b).

22

Summary In summary, we have shown that the FLA is a unique technique applicable for both monocrystalline and multicrystalline solar grade silicon photovoltaic cell production. Within the millisecond annealing time im-planted phosphorous is electrically activated and silicon is recrystallized. Simultaneously the diffusion of metal impurities and their ac-tivation is suppressed. Moreover, we validated the simultaneous implantation of phosphorus and hydrogen via PIII for the doping and in-troduction of a passivation agent in a single step. During FLA, phosphorus was activated and hydrogen passivates the defects present in the matrix caused by the implantation pro-cess. The synergistic combination of plasma immersion ion implantation and flash lamp annealing provided an innovative solution for solar cell emitter fabrication offering desira-ble possibilities in terms of cost reduction and high throughput, while at the same time re-placing time and energy consuming processes such as POCl3 deposition and diffusion, PSG cleaning, and edge isolation. Further, since H was introduced into the matrix during the doping process, alternative anti-reflective layers, such as SiO2, or cheaper and more ef-ficient deposition techniques can be used dis-regarding their H content, allowing greater flexibility in the design of next generation ad-vanced solar cells. Finally, a new concept for an efficient and cost-effective solar cell fabri-cation was presented.

Finally the proof-of-concept was performed using standard monocrystalline p-type Si wafers which were implanted using PIII to a fluence of 2×1015cm-2 and flash annealed at 1200 oC for 20 ms with preheating at 500 oC for 3 min. Fig. 14a shows the IQE mapping of the laser beam induced current (LBIC) investigation. Due to lack of texturisation and without optimisation of AR coating this

sample shows relatively high light reflection in the visible range which is above 11 %. Nev-ertheless the average IQE was above 70 %. According to the I-V characteristics obtained using a Sun Simulator at an illumination den-sity of 1000 W/m2 the pilot samples show an promising efficiency in the range of 6 % – 8 % (Fig. 14b).

Acknowledgements This work was performed within the Clus-ter of Excellence “Structure Design of Novel High-Performance Materials via Atomic De-sign and Defect Engineering (ADDE)” that is financially supported by the European Union Regional Development Funds and by the Min-istry of Science and Art of Saxony (SMWK).

23

References

1. S. Prucnal, B. Abendroth, K. Krockert, K. Koenig, D. Henke, A. Kolitsch, H. J. Moeller, and W. Skorupa, Millisecond annealing for advanced doping of dirty- silicon solar cells, J. Appl. Phys. 111, 123104 (2012)2. R. M. A. Swanson, A vision for crystalline silicon photovoltaics. Progress in Photovoltaics, v. 14, n. 5, p. 443-453, Aug 2006. 3. T. Sagata, Advances in crystalline silicon solar cell technology for industrial mass production. Npg Asia Materials, 2, 96-102, (2010). 4. D. Fujishima, et al. High-Performance HIT solar cells for thinner silicion wafers. 35th IEEE Photovoltaic Specialists Conference. Honolulu, HI. Jun 20-25, 2010. p. 3137-3140.5. J. H. Petermann, et al. 19%-efficient and 43mm-thick crystalline Si solar cell from layer transfer using porous silicon. Progress in Photovoltaics, 20, 1-5, (2012). 6. O. Schultz, S. W. Glunz, G. P. Willeke, Multicrystalline silicon solar cells exceeding 20% efficiency. Progress in Photovoltaics, 12, 553-558, (2004). 7. M. J. Keevers, T. L. Young, U. Schubert, M. A. Green. 10% Efficient CSG minimodules. 22nd European Photovoltaic Solar Energy Conference, Milan, September 2007. 8. H. Tanaka, et al. Spin-on n-type silicon films using phosphorous-doped polysilanes. J.Appl. Phys., 46, L886-L888, (2007). 9. T. Masuga, N. Sotani, H. Hamada, Y. Matsuki and T. Shimoda, Fabrication of solution- processed hydrogenated amorphous silicon single-junction solar cells. Appl. Phys. Lett. 100, 253908 (2012). 10. V. Depauw, Y. Qiu, K. Van Nieuwenhuysen, I. Gordon, J. Poortmans, Epitaxy-free monocrystalline silicon thin film: first steps beyond proof-of-concept solar cells. Progress in Photovoltaics: Research and Applications, 19, 844–850, (2010).11. http://www.solar3d.com/view_news.php?Id=39 Initial Tests Indicate the Company’s Breakthrough Silicon Solar Cell Design Will Produce 25.47% Conversion Efficiency.12. M. A. Green, The Path to 25% Silicon Solar Cell Efficiency: History of Silicon Cell Evolution. Progress in Photovoltaics, 17, 183-189, (2009). 13. T. Mishima, et al. Development status of high-efficiency HIT solar cells. Solar Energy Materials and Solar Cells, 95, 18-21, (2011). 14. Atteq ur Rehman and Soo Hong Lee, Advancements in n-Type Base Crystalline Silicon Solar Cells and Their Emergence in the Photovoltaic Industry, . The Scientific World Journal, 2013, 470347 (2013). 15. Skorupa, W.; Gebel, T.; Yankov, R. A.; Paul, S.; Lerch, W.; Downey, D. F.; Arevalo, E. A. J. Advanced thermal processing of ultrashallow implanted junctions using flash lamp annealing. J. Electrochem. Soc. 2005, 152, G436–G440.16. J. Benic, N. Bateman, M. Hermle, Presented at the 25th European PV Solar Energy Conference and Exibition, 5-10 September 2010, Valencia, Spain.17. Rohatgi, D. L. Meier, B. McPherson, Y.-W. Ok, A. D. Upadhyaya, J.-H. Lai and Francesco Zimbardi, Energy Procedia, 15, 10-19 (2012).18. K. Ohdaira, T. Fujiwara, Y. Endo, K. Shiba, H. Takemoto and H. Matsumura, Jpn. J. Appl. Phys. 49, 04DP04 (2010).19. F. Terai, S. Matunaka, A. Tauchi, Ch. Ichimura. T. Nagatomo and T. Homma, J. Electrochem. Soc. 153, H147 (2006).20. Ch. H. Poon, A. See, Y. Tan, M. Zhou, and D. Gui, J. Electrochem. Soc. 155, H59 (2008).21. W. Anwand, S. Z. Xiong, C. Y. Wu, T. Gebel, Th. Schumann, G. Brauer, W. Skorupa, Acta Phys. Pol. A. 113, 1273 (2008).

24

22. S. Govindaraju, C. -L. Shih, P. Ramanarayanan, Y. -H. Lin, K. Knutson, ECS Trans. 28, 81-90 (2010).23. Shu Qin, Chung Chan and Nicol E McGruer, Energy distribution of boron ions during plasma immersion ion implantation, Plasma Sources Sci. Technol. 1, 1-6 (1992).24. D. E. Aspens and A. A. Studna, Phys. Rev. B, 27, 985-1009 (1983).25. J. E. Smith Jr, M. H. Brodsky, B. L. Crowder, M. I. Nathan, and A. Pinchuk, Phys. Rev. Lett. 26, 642-646 (1971).26. U. Fano, Phys. Rev., 124, 1866-1878(1961).27. N. G. Rudawski, K. S. Jones, S. Morarka, M. E. Law, and R. G. Elliman, J. Appl. Phys. 105, 081101(2009).28. R. Beserman and T. Bernstein, J. Appl. Phys. 48, 1548-1550 (1977). 29. S. Prucnal, T. Shumann, W. Skorupa, B. Abendroth, K. Krockert, and H. J.Moeller, Acta Phys. Polonica A 120, 30–34 (2011).30. T. Buonassisi, A. A. Istratov, M. A. Marcus, B. Lai, Z. Cai, S. M. Heald and E. R. Weber, Nature Materials, 4, 676-679 (2005).31. M. Loghmarti, K. Mahfoud, J. Kopp, J.C. Muller and D. Sayah, Phys. Stat. Sol.(a), 151, 379-386 (1995).32. Y. Y. Chen, J. Y. Chen, R. J. Hsu, W. S. Ho, C. W. Liu, W. F. Tsai, and C. F. Ai, J. Electrochem. Soc. 158(9), H912–H914 (2011).33. F. L. Bregolin, K. Krockert, S. Prucnal, L. Vines, R. Hübner, B. G. Svensson, K. Wiesenhütter, M. H. J. and W. Skorupa, J. Appl. Phys. 115, 064505 (2014).34. L. S. Wielunski, D. Grambole, U. Kreissig, R. Grotzschel, G. Harding, and E. Szilagyi, Nucl. Instrum. Methods Phys. Res. Sec. B 190, 693–698 (2002).

25

Growth and characterization of multi-crystalline silicon ingots

E. Schmid 1, C. Funke 2, Th. Behm 2, S. Würzner 3, O. Pätzold 1, V. Galindo 4 , M. Stelter 1, H. J. Möller 2, 3

1 Institute of Nonferrous Metallurgy and Purest Materials, TU Bergakademie Freiberg 2 Institute of Experimental Physics, TU Bergakademie Freiberg 3 Fraunhofer Technology Center for Semiconductor Materials THM, Am St. Niclas Schacht 13, 09599 Freiberg, Germany4 Institute of Fluid Dynamics, Helmholtz-Zentrum Dresden-Rossendorf, Dresden, Germany

26

AbstractThis paper summarizes studies in the field of growth and characterization of multi-crystal-line (mc) silicon ingots performed within the Cluster of Excellence “Structure Design of Novel High-Performance Materials via Atomic Design and Defect Engineering (ADDE)”. Experimen-tal results on the interaction between impurities, inclusions, dislocations and grain boundar-ies in multi-crystalline (mc) silicon ingots grown from well-mixed and poorly mixed melts in graphite-containing and graphite free configurations are presented. The ingots were grown in a high-vacuum induction furnace by the vertical Bridgman (VB) method and the degree of impurity mixing within the melt was modified by changing the growth configuration and the growth rate. Vertical and horizontal slices were prepared from the ingots and analyzed by Fourier transform IR spectroscopy, as well as reflected-light and IR transmission microscopy to measure the axial carbon concentration and the distribution of dislocations or inclusions, re-spectively. The correlation between individual inclusions and dislocations has been investigated by correlative reflected-light/IR transmission and scanning electron microscopy in both setups. The influence of the melt mixing on the segregation of carbon is demonstrated and discussed with respect to the consequences for the formation of inclusions and dislocation clusters in multi-crystalline silicon. Additionally the alignment of dislocations in samples from VB-grown ingots and wafers from edge-defined film-fed (EFG) growth are investigated. Crystallographic orientations of single grains and dislocation structures are analyzed by electron backscatter diffraction and by the “traces on two parallel surfaces” method. The influence of the growth and cooling conditions on the final alignment of dislocations in mc-Si is discussed and explained.

mono-casting [6], notched crucibles [7], an additional argon flow [8] or by post-growth annealing [9-10]. In the last years, the for-mation and interaction of the microstructure in mc-Si ingots grown under different con-ditions have attracted considerable interest. The behavior of impurities in growth from a well-mixed melt in contact with atmospheres of different CO concentrations was studied in [11-12]. The influence of the growth rate and the melt flow on the impurity segregation as well as on the formation of SiC- and Si3N4-in-clusions was discussed in [11], [13-15]. Nev-ertheless, a detailed knowledge about the cor-relations between structural defects in mc-Si is still missing.

This paper summarizes results of experi-mental studies on the defect interaction in mc-Si, which were performed within the Cluster of Excellence “Structure Design of Novel High-Performance Materials via Atomic Design and Defect Engineering (ADDE)” [15-18]. It deals with the correlations between impurities, inclusions, dislocations, and grain

IntroductionMulticrystalline silicon (mc-Si) ingots, grown from the melt by a directional solidification process, are widely used to produce substrates for solar cells [1]. However, the cell efficien-cy is limited by a relatively poor structural quality of the substrates, because defects like inclusions or precipitates, grain boundaries, and dislocations degrade the minority carri-er lifetime by acting as recombination centers [2-3]. Therefore, it is of great interest to study and understand the influence of growth pa-rameters on the formation of the microstruc-ture in mc-Si ingots. The general motivation behind is to develop methods of defect engi-neering on solidification, which are aimed at an improvement of mc-Si with respect to an increased carrier lifetime.

Several techniques to control and optimize the grain size and the type of grain bound-aries were systematically investigated. It was shown, that a microstructure with large grains joined by recombination-inactive grain boundaries like twins can be achieved by using spot cooling [4], dendrite casting [5],

27

boundaries under different growth conditions i.e., different setups (graphite-free vs. graph-ite-containing setup), as well as different degrees of melt mixing and growth or cooling rates. Advanced and newly developed charac-terization methods based on IR transmission and scanning electron microscopy (SEM)

were used to gain a deeper insight into the de-fect interaction. The discussion of the results is focused on the formation of inclusions and dislocations in correlation with the segrega-tion of impurities, and on the influence of the cooling rate on the final dislocation structure in mc-Si.

Experimental

Growth setup

Growth experiments were carried out in a high-vacuum induction furnace using the vertical Bridgman (VB) technique shown in Fig. 1. The induction coil has a diameter of 200 mm, a height of 150 mm and consists of 10 windings. The maximum power is 20 kW generated at a frequency of 10 kHz. Investiga-tions can be carried out in an inert gas atmo-sphere or under vacuum up to 10-6 mbar. An axisymmetric setup was used with the com-mercially available electronic-grade silicon in a Si3N4 coated silica crucible (inner diameter: 105 mm), which was surrounded by susceptor and insulation. For the graphite-containing

setup, a graphite susceptor and graphite insu-lation were used, while for the graphite free setup a molybdenum susceptor and Al2O3 in-sulation were applied. The setup is placed on a water-cooled translation stage in a central po-sition in the induction coil. A pyrometer and several thermocouples were used for measur-ing the temperatures at different positions in the system. Further details of the furnace and the growth configuration are given in [11]. After melting of the feedstock and homogeni-zation of the melt, the setup was lowered out of the coil with different, constant translation rates, resulting in directional crystallization with different solidification or growth rates.

Fig. 1: High-vacuum induction furnace and growth configuration.

28

Fig. 2 shows a scheme of the growth setup and the flow resulting from natural buoyan-cy and from the Lorentz force induced by the induction coil. It is important to note that the melt flow in our setup for both configurations is dominated by the Lorentz force. This was demonstrated by calculating the force densi-ties and the related flow patterns numerically. The simulation of the electro-magnetic fields was carried out using the finite element code Opera 3D (© Cobham plc) taking into ac-count the influence of the conducting parts of the furnace as well as the parameters and

Fig. 2: Growth configuration (left) and numerically simulated melt flow (right) resulting from the induced Lorentz force for graphite-containing setup (a) and graphite free setup (b). The axial position of the setup relative to the induction coil corresponds to the experimental situation during homogenisation of the melt prior to the directional solidification.

dimensions of the induction coil. The soft-ware library OpenFOAM was used to cal-culate the induced melt flow by solving the Navier-Stokes equation with the Lorentz force term. Fluctuations of the flow velocity were treated in the frame of the low Reynolds k-Ω SST turbulence model. The maximum flow velocity was found to be about 0.14 m/s for graphite-containing setup and 0.016 m/s for graphite free setup. The higher flow velocity for the graphite-containing setup results from a lower screening of the magnetic field by the graphite susceptor.

Experiments with translation rates of 5 mm/h and 20 mm/h for both growth setups were performed to modify the degree of impurity

mixing within the melt and to influence gen-eration of inclusions and dislocations.

29

Sample preparation and characterizationThe grown crystals were cut into vertical and horizontal samples as sketched in Fig. 3. The samples were grinded, polished, and etched in a Secco solution [19] at ambient tempera-ture to reveal inclusions and dislocations. As a criterion of the dislocation density, etch pit density (epd) mappings of individual disloca-tions were created from high-resolution re-flected-light microscopy images, as described in [17]. By means of high-resolution optical microscopy individual etch pits on the sample surfaces were detected. The microscope was equipped with a CCD camera and an auto-focus system. Sample areas up to 1.3 x 4 cm2 were measured by stitching individual micro-scopic images of 170 x 140 µm2. A semi-auto-mated image processing system programmed

Fig. 3: As-grown mc-Si crystal (left) and sketch of sample preparation.

in the Olympus a4i software package was used to identify and count the pits. Finally, the noncommercial software Discorr [20] was used to calculate the etch pit density (epd) topogram of the analyzed surface area. The distribution of inclusions was detected by IR transmission microscopy. Inclusions and dislocations in selected sample areas were characterized by correlative reflected-light/IR transmission and scanning electron micros-copy (SEM) [15]. Axial concentration profiles of carbon were measured by Fourier trans-form infrared (FTIR) spectroscopy. This mea-surements were performed with a TENSOR 27 FTIR spectrometer, (BRUKER Optics).

Crystallographic orientations of single grains, which are described by Euler angles, were analyzed by electron backscatter dif-fraction. The measurements were carried out using a Philips XL30 scanning elec-tron microscope (SEM) with TSL OIM™ data collection from Edax/Ameteks Inc. The alignment of dislocations was analyzed on the basis of stereographic projections (pole figures) and the related traces of crystallo-graphic lattice planes. For a correct identifica-tion of the planes the so-called “traces on two parallel surfaces” method was used, which is

based on a combination of orientation mea-surements and IR transmission microscopy [16]. The combination of transmission mi-croscopy and orientation determination by EBSD technique allows gaining additional re-sults compared to each individual technique. This methodical development has been based on Randle´s overview [21] about the “Appli-cation of EBSD on the analysis of interface planes” and description of the “two surface” sectioning method, firstly performed in [22]. There, two plane traces on two perpendicular surfaces of a metallic specimen were charac-terized by two angles α and β, see Fig. 4.

®

30

Results and Discussion

Impurity segregation and formation of inclusions

Axial carbon concentration profiles in in-gots grown with translation rates of 5 mm/h and 20 mm/h in both growth configurations are compared in Fig. 5. In crystals grown in the graphite-containing configuration no axial macrosegregation is observed. The carbon concentration for both translation rates is found to be almost constant at about 3x1017 At/cm3, which is close to the solubil-ity limit. Obvisiouly, the carbon level in the melt is pinned by the SiC layer at the surface [11]. No inclusions were found in the volume of the crystals due to the intensive melt flow (Fig. 5a-b).

In contrast, a pronounced axial macroseg-regation profile was found in crystals grown in graphite free configuration. The concen-tration profile in the slowly grown ingot fits well with the known Scheil’s law valid for an enclosed melt with a uniform solute concen-tration. In Fig. 5 it is drawn using the equi-librium partition coefficient of carbon in sil-icon (k0,C = 0.07) and a melt concentration c0 = 7.6 1017 cm-3, which was estimated from the initial carbon concentration in the crystal.

Fig. 4: Characterization of aboundary plane trough anglesand : is the anglefrom specimen x-axis (sx) to grainboundary trace on xy-specimenplane. is thetilt angle between the investigatedboundary and negative z-axis (-sz).

Randle [23] gives the maths to calculate fromthe measured angles α, β or ϕ, and themeasured three Euler angles (φ1, θ,φ2), seeGottstein [24], the plane normals in thecrystal coordinate system as well as in thespecimen coordinate system. To use thistechnique, two traces of the boundary planein question on perpendicular surfaces or atleast surfaces with known angle ofintersection have to be prepared. Thisrequires a high experimental effort in termsof metallurgical target preparation for eachinvestigated boundary or plane.

It is much easier to customize opticaltransmission microscopy instead. Trans-mission microscope examination of polishedspecimens gives information about the tracesof a boundary plane both on the upper(green in Fig. 4) and lower (red in Fig. 4) xy-plane of the specimen. Together with thespecimen thickness t and the lateral distanced of associated traces the angle ϕ can becalculated according equation dt=tan(ϕ).With the determination of ϕ the same mathas proposed by Randle [23] is sufficient now.By this, the “Two surface sectioningmethod” has been expanded to the “Traceson two parallel surface” method.

.

31

Apparently, carbon was almost completely mixed within the melt during solidification in this experiment. In growth with a translation rate of 20 mm/h, the axial concentration pro-file differs considerably from the Scheil curve due to a pronounced increase of the carbon concentration in the centre of the melt. This behaviour reflects the known influence of the growth rate on the axial segregation of impu-rities in directional solidification processes [25]. For a semi-quantitative evaluation, the profile can be fitted using a modified Scheil approach for a partly mixed melt [15], [26], which bases on the assumption of a variable, effective partition coefficient keff > k0:

This result clearly indicates that carbon was poorly mixed during growth, leading to a stronger axial segregation than in growth from a well-mixed melt like in the graph-ite-containing configuration. As an important consequence, the melt in front of the solid-liq-uid interface eventually becomes supersat-urated with carbon on solidification. This is indicated by the appearance of inclusions in the upper part of the vertical sample of the rapidly grown ingot, as evident from the cor-responding IR transmission image in Fig. 5d.

with a,b – fit parameters and g = z/L – solid-ified fraction (L – central length of vertical samples). Assuming again c0 = 7.6 1017 cm-3, the fit curve provides 0.09 < keff < 0.26 in the range of 0<g<0.7 (see Fig. 5).

Fig. 5: Axial carbon concentrations in graphite-containing and graphite free setups. (a-b) IR transmission images of crystals of graphite-containing configuration with translation rates of 5 and 20 mm/h; (c-d) IR transmission images of crystals of grown in the graphite free configuration with translation rates of 5 and 20 mm/h.

In growth with translation rate of 5 mm/h, carbon exceeds the solubility limit only in the very last solidified part of the ingot (Fig. 5c). The related IR transmission image reveals no inclusions apart from some small structures close to the top of the crystal. These struc-tures were not analyzed in detail here, but are known from [13], [14] to be associated with Si3N4 inclusions (needle-like structures) and SiC inclusions (filament-like structures).

32

Fig. 6: IR transmission images and epd mappings of horizontal samples from the bottom and upper parts of the slow-ly grown and rapidly grown ingots in graphite-containing (a-b) and graphite free configurations (c-d). The boxes indicate the position of the epd maps on the samples.

Correlation between inclusions and dislocations

Fig. 6 shows IR transmission micrographs and epd mappings from horizontal half slices cut from the bottom and upper parts of the ingots in both growth configurations. Gener-ally, the distribution of dislocations is found to be very inhomogeneous. Highly dislocat-ed sample areas are adjacent to areas with at least several orders of magnitude lower dis-location density (note the logarithmic scale in Fig. 6). Besides, the formation of disloca-tions seems to be influenced by the different growth configuration and the growth rate. In graphite-containing setup solidification with the translation rate of 20 mm/h results in numerous, narrow dislocation clusters with

an epd up to 107 cm-2 on the surface of the bottom sample (Fig. 6b). On the other hand, no such clusters can be seen on the bottom sample of the slowly grown crystal (transla-tion rate 5 mm/h) (Fig. 6a). Several extend-ed dislocated areas have been detected on the surfaces of the top samples of both crystals. This indicate the known phenomena that the dislocations spread vertically and laterally during growth, and that the dislocation densi-ty in mc-Si crystals increases with increasing crystal height [27]. Nevertheless, the mean epd of the samples from the slowly grown crystal remains relatively low, which is obvi-ously associated with the absence of disloca-tion clusters on the related bottom sample.

33

IR-TM measurements were carried out to clarify the origin of the narrow dislocation clusters observed. In the IR-TM images of the fast grown crystal the dark spots of vanishing optical transmission represent agglomera-tions of inclusions, which were detected to be SiC by means of SEM-EDX (Fig. 7). The dis-tribution of the inclusions depends on the so-lidification rate. They are mainly concentrated at the edge of the bottom sample in growth with a translation rate of 5 mm/h, whereas randomly distributed over a large sample area in growth with a translation rate of 20 mm/h.

It can be seen that the ‘epd section’ of the bottom sample of the fast grown crystal in-cludes several inclusions just at the positions of the dislocation clusters. This gives clear evidence that the detected dislocation clus-ters are closely related to the existence of

This result can be attributed to an increased SiC nucleation rate due to a pronounced su-persaturation of carbon near the crucible bottom in growth with a high solidification rate. On the other hand, lowering of the solid-ification rate is expected to reduce the super-saturation and hence, the probability of het-erogeneous nucleation, which is confirmed by the low number of inclusions detected on the bottom sample of the crystal grown with a translation rate of 5 mm/h.

Fig. 7: Microscopic images and SEM image with EDX spectrum of selected area of the bottom sample of the mc-Si crystal grown with translation rate of 20 mm/h in graphite-containing setup showing a high dislocation density in the vicinity of SiC inclusions.

inclusions. To confirm this result, relevant sample areas were inspected in more detail by optical microscopy. Fig. 7 shows enlarged microscopic images revealing man inclu-sions of different morphology. In the vicini-ty of the inclusions, the dislocation density

34

is always significantly enhanced. This can be explained by local thermo-mechanical stress fields due to the different thermal expansion coefficients of SiC and the surrounding Si ma-trix, which lead to a preferred formation and multiplication of dislocations during growth. The situation becomes completely different in the crystals grown in the graphite free setup [15]. The samples of the slowly grown ingot and the bottom sample of the rapidly grown ingot reveal no inclusions at all, whereas the sample from the upper part of the rapidly grown ingot shows an extended area with inclusions (Fig. 6c-d). Independent of the growth rate, the dislocation densities on the bottom samples are found to be very low with an epd of about 104 cm-2. In the area with in-clusions of the rapidly grown ingot, however, numerous dislocation clusters with a much higher epd up to about 107 cm-2 were detected, whereas the dislocation density in the upper part of the slowly grown ingot remains at a relatively low level. There is obviously a direct correlation between the existence of inclu-sions and the formation of dislocations. This corresponds to results of the graphite-con-taining setup [17], in which large SiC inclu-sions with characteristic dimensions of sever-al microns were identified to be an important source of dislocations in mc-Si ingots. On the other hand, some dislocation clusters appear also in the upper part of the slowly grown ingot, which is free of inclusions. This indi-cates that inclusions are a major, but not the only source of dislocations, which confirms previous studies [27] demonstrating that for instance grain boundaries can be relevant for the formation of dislocations in mc-Si.

Fig. 8 shows details of the microstructure within the area of inclusions of the rapidly grown ingot in the graphite-free configura-tion.

Reflected-light/IR transmission micrographs and the correlative SEM image of inclusions with characteristic lateral dimensions of some 10 µm are presented. The inclusions are composed of nearly stoichiometric SiC- and Si3N4-crystals, as detected by SEM/EDX. The EDX spectrum of a Si3N4 inclusion is shown in Fig. 8 on the right. This corresponds to former results on the existence of C- and N-related inclusions in mc-Si crystals (e.g., [13-14]). Thus, the formation of inclusions is not only associated with the supersatura-tion of carbon, but also of nitrogen during growth. Like carbon, nitrogen can be ex-pected to show a stronger segregation with increasing growth rate, too. Though not di-rectly measured in this study, it is therefore not surprising that nitrogen plays an im-portant role in the formation of inclusions in the upper part of the rapidly grown ingot. In the vicinity of the inclusions shown in Fig. 8, an increased dislocation density ap-pears. Assuming a stress-induced formation of dislocations to be dominant [28], this can be explained by local thermo-mechanical stresses due to different thermal expansion coefficients of SiC/Si3N4 and the surrounding Si matrix. However, it has to be noted that no complete correlation between individual in-clusions and dislocations is observed. Typical-ly, inclusions are surrounded by pronounced dislocation clusters, but there are also inclu-sions without dislocation clusters in the ad-jacent matrix. Some examples are shown in [15]. This might indicate a significant differ-ence in the effective stress fields established around inclusions. Possible reasons for such differences could be the dimensions and mor-phologies of the inclusions, the local orien-tation and arrangement of grains and grain boundaries, or the number and distribution of dislocations and inclusions already existing in the vicinity of a certain inclusion, when it is grown-in on solidification. On the basis of the presented results, however, these explanations remain speculative. For a better understand-ing of the interaction between inclusions and dislocations in mc-Si, further investigations are necessary.

35

Fig. 8: Reflected-light, IR transmission, SEM micrographs of SiC/Si3N4 inclusions, and EDX spectrum of a Si3N4 inclusion detected in the upper part of the ingot grown with a translation rate of 20 mm/h in the graphite-free setup (see Fig. 5).

Investigation of dislocation structures in in-got - and wafer-grown multicrystalline silicon

Dislocation structures detected in ingot and Dislocation structures detected in ingot and edge-defined film-fed (EFG)-grown crystals are presented in Fig. 9. It was found that the crystallographic alignment depends on the growth and cooling conditions [18]. Fig. 9a shows a typical structure in a mc-Si sample grown by VB technique with low growth rate. A significant number of dislocations visible in the micrographs is linearly arranged, parallel to each other and perpendicular or inclined to the twin boundaries. The twin boundaries (structures A) were identified by the ”traces on two parallel surfaces” method as 111 or, more exactly, as (111) planes. The relevant poles and plane projections are marked in the corresponding 111 pole figures of the O1 grains in Fig. 9a. It can be seen that the projections are indeed parallel to the twin boundaries under consideration. It is noted

that the investigation of O2-oriented grains would lead to the same conclusion, because the positions of the relevant poles in O1 and O2 grains are identical. The dislocations struc-tures investigated by the ”traces on two paral-lel surfaces” method are labelled by B and C. It turns out that dislocations are mainly aligned in (121) planes (structures B and C in Fig. 9a). An exceptional example for a dislocation structure in ingot-grown mc-Si is presented in Fig. 9b. The sample was prepared from the very last solidified part of an ingot. Due to the existence of a solid SiC layer on the Si melt during growth in our furnace [12], this part is formed by a rapid solidification of the residual melt. The dislocations in this region are main-ly aligned in perpendicular directions forming a distinct cell-like structure. The plane trace constructions in the 111 and 211 pole fig-ures give evidence that the structures A and B correspond to an arrangement of disloca-tions in (111) and (211) planes, respectively.

36

Fig. 9: a) Optical micrograph of a mc-Si sample from a VB-grown ingot and 111 and 211 pole figures of grains with O1 and O2 orientations. The analyzed twins and dislocation structures are marked by A, B, C. Asterisks and solid lines in the pole figures indicate relevant poles and the corresponding plane projections, respectively. b) Optical micrograph, secondary electron image of a mc-Si sample from a rapidly grown part of an ingot and 111 and 211 pole figures. The analyzed dislocation structures are marked by A, B and the relevant plane traces in the pole figures by solid lines. c) Secondary electron image of a typical microstructure in an EFG-grown mc-Si sample and 111 and 211 pole figures of grains with O1 and O2 orientations. The analyzed twin structures in the O1 grains are marked by ’A’ and the relevant poles additionally by asterisks in the 111 pole figures. Solid lines indicate projection of the related crystallographic plane (thick line) and its normal direction (thin line).

Fig. 9c shows a typical microstructure of a mc-Si sample prepared from an EFG wafer. Twin structures, labelled by A, adjoined to a dislo-cated area can be recognized. The dislocations form mainly a linear structure parallel to the twin boundaries. In this way, twin boundar-

The results presented in Fig. 9 show that the dislocations in mc-Si are mainly aligned either in 111 or 211 planes. Because 111 planes are known to be the preferred slip planes of dislocations in a fcc material [29], dislocation structures parallel to 111 planes can be at-

ies were clearly identified to be parallel to the (111) plane by the method of “traces on two parallel surfaces”. Apparently, the dislocations of grain O2 are mainly aligned in the same plane, indicating a preferred movement of dislocations in the (111) plane during growth.

tributed to plastic deformation under local thermo-mechanical stress. In contrast, dislo-cation structures parallel to a 211 plane are assumed to reflect the subsequent recovery of the material to minimize the energy stored in a deformed structure.

37

This explanation is supported by results ob-tained in metal physics, where linear arrange-ments of dislocations perpendicular to the slip planes are generally ascribed to recovery [29]. Related processes are, for instance, strain field-induced or point defect-assisted anni-hilation and polygonization of dislocations. During polygonization, dislocations can align themselves perpendicular to a slip plane forming a small-angle grain boundary with minimized energy due to an optimal overlap of the dislocation strain fields. The typical dis-location structures in EFG- and ingot-grown mc-Si were found to be different. In wafers grown by EFG with a very high growth rate, dislocations are mainly aligned in 111 slip planes (Fig. 9c), whereas in VB-grown ingots an alignment out of slip planes in 211 planes predominates (Fig. 9a) This indicates that the final arrangement of dislocations depends on the solidification or cooling conditions the crystals are exposed to during growth. Thus, a deformed structure is frozen in EFG-grown mc-Si, because the dislocations become im-mobile during plastic deformation due to a rapid cooling of the crystals. On the other hand, the microstructure of slowly grown mc-Si ingots typically represents an annealed dislocation arrangement formed by plastic deformation and subsequent recovery. Some samples from mc-Si ingots reveal significant dislocation alignment in 111 and 211 planes (Fig. 9b). Apparently, both plastic de-formation and recovery play an important role for the final arrangement of dislocations in this case. For explanation, it has to be not-ed that the investigated samples were from

an ingot grown with a relatively high average growth rate. They were cut from the very last solidified part of the ingot. Though not exactly known, the actual growth rate in these regions was certainly higher than the average value. With increasing growth rate, however, recov-ery processes are increasingly impeded by the limited mobility of dislocations. So it appears that the dislocation structures in the samples under consideration reflect an incomplete recovery during growth, so that a significant number of dislocations persists in a deformed structure parallel to 111 slip planes.

The dislocation structure in vertical Bridg-man-grown ingots was found in former works to remain essentially unchanged during post-growth annealing [30-31]. Compared with a EFG-grown material, mc-Si ingots are usually processed with very low growth and cooling rates. The investigations of as-grown material in this paper show to a large extend already annealed dislocation structures aligned per-pendicular to slip planes, at least in samples from slowly grown ingot parts. It is assumed that recovery processes generally play a domi-nant role for the final arrangement of disloca-tions in slowly grown mc-Si crystals. This can be attributed to the relatively long, high-tem-perature period during growth, where dis-locations are sufficiently mobile to form a microstructure with minimal stored energy. Such a structure will not change significant-ly by an additionally annealing treatment, which corresponds to the results presented in [30-31].

38

Summary

Experimental investigations of the defect interaction in mc-Si were presented with focus on the correlations between carbon impurities, SiC inclusions, and disloca-tions in VB-grown ingots. To influence the concentration and mixing of impurities in the melt, ingots were grown in graph-ite-containing or graphite-free setups with different translation or growth rates. The graphite-containing setup leads the melt to be saturated with carbon prior to solidi-fication. In growth with a high growth rate, SiC inclusions appear near the bottom of the ingots due to a temporary supersaturation of carbon in the melt at the very beginning of solidification. The bulk, however, is free of in-clusions, because the carbon concentration in the melt remains at a constant level near the solubility limit during growth. This excep-tional behaviour can be explained by an in-tense, Lorentz force-driven melt flow leading to a uniform mixing of the melt, and by the existence of a SiC layer at the melt surface, which provides nucleation sites to balance the enrichment of carbon by segregation. When using the graphite-free setup with a molybdenum susceptor, there is no SiC sur-face layer and the initial carbon contamina-tion of the melt is strongly reduced. Therefore, no SiC inclusions are formed at the beginning of growth, and a continuously increasing car-bon concentration due to segregation is ob-served in the solidified ingot. On the other hand, electro-magnetic fields and the Lorentz force are effectively screened due to the high electrical conductivity of the susceptor, which results in a significant decrease of the melt flow intensity. In experiments with a high growth rate, this is associated with a poor mixing of carbon and other impurities in the melt. With successive impurity enrichment in front of the interface by segregation, the solubility limit is reached or exceeded during growth, so that SiC and other inclusions can be formed in the bulk of the ingots, as actually observed.

Inclusions are identified to be the origin of narrow dislocations clusters with an epd up to 107 cm-2. During growth, disloca-tions tend to propagate through the ingot, finally leading to extended, highly-dislo-cated areas. In the setups under consid-eration, the formation of inclusions and related dislocation clusters can be drasti-cally reduced by lowering the growth rates. Furthermore, dislocation structures in VB- and EFG-grown mc-Si were analysed and compared. In VB-grown ingots, the dom-inant dislocation alignment is found to be in 211 planes, which are perpendicular to 111 planes of the glide system in Si. Based on standard knowledge in metal physics, this structure has been attributed to recovery pro-cesses after plastic deformation. Furthermore, hybrid structures with dislocations aligned in both 111 and 211 planes have been detected in those parts of a VB ingot grown with higher rates. It can be concluded that a direct correlation between the growth or cooling conditions and the dislocation struc-ture in mc-Si exists. Moreover, the growth rate appears to be the most important parameter for the final arrangement of dislocation in the crystals. The higher the growth rate, the more dislocations survive in energetically unfavor-able, deformed structures, because recovery processes are increasingly suppressed. In as-grown EFG wafers the dislocations are mainly aligned in 111 slip planes indicating signifi-cant plastic deformation of the material.

39

References

1. „EPIA Fact Sheet: Sustainability of Photovoltaic Systems. Materials Availability,“ EPIA Publications, 2013.

2. M. Rinio, A. Yodyungyong, S. Keipert-Colberg, D. Borchet and A. Montesdeoca-Santana, Physica Status Solidi A208, p. 760–768, 2011.

3. O. Breitenstein, J. Bauer, P. Altermatt and K. Ramspeck, Diffus. Defect Data B: Solid State Phenom.156–158, pp. 1-10, 2009. 4. T. Wang, S. Hsu, C. Fei, K. Yei, W. Hsu and C. Lan, Journal of Crystal Growth 311, pp.

263-267, 2009. 5. K. Fujiwara, W. Pan, K. Sawada, M. Tokairin, N. Usami, Y. Nose, A. Nomura,

T. Shishido and K. Nakajima, Journal of Crystal Growth 292, p. 282–285, 2006. 6. M. Trempa, C. Reimann, J. Friedrich, G. Müller and D. Oriwol, Journal of Crystal

Growth 351, p. 131–140, 2012. 7. T. Li, K. Yeh, W. Hsu and C. Lan, Journal of Crystal Growth 318, p. 219–223, 2011. 8. T. Li, H. Huang, H. Tsai, A. Lan, C. Chuck and C. Lan, Journal of Crystal Growth 340,

p. 202–208, 2012. 9. K. Hartman, M. Bertoni, J. Serdy and T. Buonassisi, Applied Physics Letters 93,

p. 122108, 2008. 10. M. I. Bertoni, D. M. Powell, M. L. Vogl, S. Castellanos, A. Fecych and T. Buonassisi,

Physica status solidi (RRL) - Rapid Research Letters 5, p. 28–30, 2011. 11. L. Raabe, J. Erig, S. Würzner, O. Pätzold, M. Stelter and H. Möller, Diffusion and Defect

Data B: Solid State Phenomena 156-158, pp. 49-54, 2009. 12. L. Raabe, O. Pätzold, I. Kupka, J. Ehrig, S. Würzner and M. Stelter, Journal of Crystal

Growth 318, pp. 234-238, 2011. 13. C. Reimann, M. Trempa, J. Friedrich and G. Müller, Journal of Crystal Growth 312,

pp. 1510-1516, 2010. 14. M. Trempa, C. Reimann, J. Friedrich and G. Müller, Journal of Crystal Growth 312,

pp. 1517-1524, 2010. 15. C. Funke, E. Schmid, G. Gärtner, S. Reißenweber, W. Fütterer, A. Poklad, L. Raabe,

O. Pätzold and M. Stelter, Journal of Crystal Growth 401, pp. 732-736, 2014. 16. C. Funke, T. Behm, R. Helbig, E. Schmid and S. Würzner, Journal of Microsopy 246,

pp. 70-76, 2012. 17. E. Schmid, S. Würzner, C. Funke, V. Galindo, O. Pätzold and M. Stelter, Journal of

Crystal Growth 359, pp. 77-82, 2012. 18. E. Schmid, C. Funke, T. Behm, O. Pätzold, H. Berek and M. Stelter, Journal of Crystal

Growth 382, pp. 41-46, 2013. 19. F. S. d. Aragona, Journal of The Electrochemical Society 119, pp. 948-951, 1972. 20. M. Rinio, „http://nbn-resolving.de/urn:nbn:de:swb:105-6884856,“ 2004. [Online]. 21. V. Randle, Journal of Microscopy 230, pp. 406-413, 2008. 22. V. Randle and D. Dingley, Scripta Metallurgica 23, pp. 1565-1570, 1989. 23. V. Randle, The Measurement of Grain Boandary Geometry, Bristol and Philadelphia:

Institute of Physics Publishing, 1993. 24. G. Gottstein, Physikalische Grandlagen der Materialkande, Berlin, Heidelberg, New

York.: Springer-Verlag, 2007.

40

25. G. Müller, Materials Science Forum 276-277 , pp. 87-108, 1998. 26. M. Czapelski, Journal of Crystal Growth 187 , pp. 138-139, 1998. 27. B. Ryningen, G. Stokkan, M. Kivambe, T. Ervik and O. Lohne, Acta Materialia 59,

pp. 7703-7710, 2011. 28. J. Voelkl, in Handbook of Crystal Growth Vol. 2b, Elsevier Science, 1994. 29. P. Haasen, Physical Metallurgy, Cambrige: Cambridge University Press, 1996. 30. G. Stokkan, C. Rosario, M. Berg and O. Lohne, Solar Power, InTech, 2009. 31. T.Ervik, M. Kivambe, G. Stokkan, B. Ryningen and O. Lohne, Acta Materialia 60,

pp. 6762-6769, 2012.

41

Experimental and numerical investigations on the formation of surface defects during machining of silicon wafers

M. Budnitzki 1, T. Behm 2, M. Kuna 1, H. J. Möller 2 1 Institute of Mechanics and Fluid Dynamics, Lampadius Str. 4, 09599, Freiberg, Germany2 Institute of Experimental Physics, Leipziger Str. 23, 09599 Freiberg, Germany

42

AbstractThe strength of silicon wafers is governed by surface defects introduced mainly during the wire-slicing processing step. Depending on whether loose or fixed abrasive is used, the elementary material removal process is either chipping due to penetration with hard pointed asperities or scratching. We investigated the resulting damage patterns on the macro-scale by means of sa-wing experiments and confocal microscopy as well as on the micro-scale using a combination of indentation and scratching experiments supported by numerical modeling using a dedica-ted constitutive law. The results highlight the importance understanding the displacive phase transitions in silicon in order to improve abrasive machining techniques for this important semiconductor material.

1 IntroductionSemiconductor substrates form the material basis for most of the electronic and optical components of microelectronic and photo-voltaic devices. They are manufactured by mechanical processing steps, such as sawing, grinding, lapping or polishing, from mono- or multi-crystals. Semiconductor materials are very expensive and can easily break un-der mechanical load because of their brittle-ness. The loss of the expensive material due to breakage and the mechanical treatment of the wafers is a substantial cost factor, especially in photovoltaics, where wafers are generally thinner.

Previous fracture investigations have shown that the surface damage, which is generated through the mechanical wafer processing, mainly determines the fracture properties, whereas bulk properties are of minor im-portance for today’s high quality, defect poor semiconductor crystals. Although all tech-nologies can be optimized to some extent by experimental practice, further improvements require a thorough understanding of the mi-cromechanical phenomena, which control the material removal process and induce the sur-face damage. Therefore, many fundamental scientific questions remain open, particularly because new processes have been developed in the last few years.

Multi-wire sawing is the first step of wafer processing and induces the strongest surface damage. The standard method is the cutting

with an abrasive slurry, mainly silicon carbi-de powders dispersed in polyethylene glycols (loose abrasive sawing - LAS). Although this technique is already mature, improvements are still possible for instance by the use of wa-ter based fluids or structured steel wires. Since a few years a new technique is also under de-velopment, the slicing with diamond coated wires (fixed abrasive sawing - FAS) [1 - 5].

In both cases the induced surface damage mainly consists of microcracks in the micro-meter range. If the stress in a deformed wafer exceeds a critical value, which depends on the local microcrack length, the wafer will break. The longest cracks are the most critical; there-fore, it is the goal of the process improvement to reduce microcrack lengths. In addition, other quality factors such as the surface profi-le (roughness, waviness, flatness, etc.), the to-tal thickness variations (TTV), or the crystal-linity at the surface are important depending on the application of the wafers.

Considering the basic mechanisms slurry based and fixed abrasive sawing are different. In the first case the abrasive (SiC) particles move freely in the carrier fluid and indent randomly into the crystal surface [6-8]. In the second case the (diamond) particles are fixed on the wire and scratch over the crystal sur-face. Whereas the first mechanism is basically the same as in lapping and polishing proces-ses, the second case is typical for grinding processes.

43

In the present study the micromechanical me-chanisms for both cases have been investiga-ted in detail both experimentally and by nu-merical simulations. The results are compared with investigations from sawing and fracture tests and are finally discussed in view of the significance for the sawing process.

2 Loose Abrasive - Crystal Interaction

2.1 Sawing experiments

Wafers were sawn on an industrial multi-wire saw (Meyer&Burger DS 265), which has been equipped with a force sensor to measure the forces normal (sawing direction) and parallel to the wire (wire direction). In most cases a slurry consisting of polyethylene glycol (PEG 200) and about 43 vol% of SiC has been used. The wires were 120 µm thick and used to cut wafer thicknesses between 100 - 150 µm in most cases.

Material removal occurs by the random inter-action of the abrasive particles with the crys-tal surface. Previous results have shown that it is necessary to reduce the average forces Fp on the interacting particles if one wants to reduce the surface damage (microcrack lengths, roughness and TTV). The average particle force is estimated from the total force on a wire Ftot by the relationship Ftot = m Fp , where m is the number of interacting partic-les below a single wire. Since the force sensor in the machine measures Ftot, it was possible to identify sawing conditions, which lead to a low subsurface damage.

Another factor, which has been identified to be important is the slurry temperature. Due to the friction in the sawing channel the slurry and the ingot temperature increase along the wire direction. This has been measured by a thermography camera on the side face of an ingot. Temperature increases up to 70 o C have been measured depending on the sawing con-ditions. Generally it has been found that high-er temperatures at the wire exit can lead to an instability of the wire motion and the occur-rence of a high roughness or even grooves on

the wafer surfaces (saw marks [9]). This also correlates with a high subsurface damage.

Sawing trials on an industrial machine are ex-pensive and time-consuming. Therefore, bet-ter understanding of the micromechanical in-teraction of single abrasive particles with the crystal surfaces can be very helpful to guide the optimization process. Experimentally one can study the interaction by a careful analysis of as-sawn wafer surfaces and by indentation experiments with single particles.

The as-sawn wafer surfaces have been eva-luated by measuring the surface profile with white light interferometry (globally) and con-focal microscopy (locally). This allows one to determine the total thickness variations (TTV) and roughness from line scans over the surface.

In addition, a new surface characterization method has been developed, which gives in-formation about the two-dimensional surface structure. An as-sawn wafer surface consists of small planar surface elements in the micro-meter range, which are inclined to the wafer normal under various angles. This is because the surface structure is produced by many in-dividual fracture events, which mainly occur along (111) planes in silicon, see Fig. 1. The ori-entations of the surface elements can be deter-mined from the three dimensional surface in-formation of a digitized confocal microscope image. Such an image consists of a net of 1024 x 1024 points and the corresponding surface heights. We developed a method to determi-ne the surface element orientation relative to the wafer surface from the height informati-on at three neighboring measurement points (a triangle). The orientations of the normal of all triangles are then plotted in an orientation density plot (ODS). An example is given in Fig. 2. Such plots show if preferential orien-tations of surface elements occur for instance due to the sawing direction or a preferential etching.

2.2 Wafer surface characterization

44

Figure 1: Differentially etched wafer surface (Confocal microscopy image)

Figure 2 : Orientation density plots calculated from the corresponding confocal microscopy images in Fig. 1

The microcracks below the surface (subsur-face damage) can be observed on wafer cross sections in the scanning electron microscope and optically at etched polished sections of beveled surfaces. Both the microcrack lengths distributions and orientations have been de-termined for different sawing conditions.

A microcrack analysis is rather time-consu-ming. Since the roughness of an as-sawn wa-fer is however an indicator for the subsurface damage it can be used to determine if sawing conditions are favorable in view of the micro-

crack depth distribution. Fig. 3 shows some results for the roughness variation across the wafer surface in the wire motion direction for different average grit sizes. It is generally ob-served that the roughness decreases from the wire inlet to outlet region. It also depends on sawing parameters such as the wire speed. The most important factors are the particle size distribution and the particle shape. This has a direct influence on force and thus the local micromechanical interaction between partic-les and crystal and was therefore investigated in single particle indentation experiments.

Fig. 3: Dependence on the average grit size (F500: 13 µm, F800: 6.5 µm, F1000: 4.5 µm, F1200: 3.9 µm): Roughness measurements along the sawing direction and force per wire as a function of wire velocity (PEG 200 and SiC volume fraction of 23 %).

45

Understanding the interaction of a silicon surface with a pointed asperity or sharp par-ticle is a crucial step towards a quantitative description of the abrasion processes during mechanical machining.

During a contact loading – unloading cycle silicon undergoes a series of stress driven pha-se transitions. As many as 12 distinct crystal-line or amorphous phases of silicon have been experimentally observed at various stress levels [29, 31, 33, 34]. Experiments emplo-ying diamond anvil cells in order to impose hydrostatic loading conditions have revealed that diamond-cubic silicon (cd-Si, space group Fd3m) transforms to the (metallic) β-tin struc-ture (β-Si, space group I41/amd) at ~11.3 GPa, leading to ~20 % densification [31, 35]. It is known from literature that non-hydrostatic conditions lower the transformation stress, which is in accordance with theoretical con-siderations [30]. This transition is not rever-sible and amorphous silicon (a-Si) is formed during rapid decompression [29, 32].

The stresses during loading and the residual stresses during unloading of the contact cause the formation and growth of several crack systems, which – in part – contribute to the residual damage.

Indentation experiments on polished mono-silicon surfaces were carried out with single particles under defined forces. Both the par-ticle shape and the resulting crack systems were analyzed. A typical result is shown in Fig. 4 for a well-defined sapphire particle and for an irregular shaped SiC particle. An important observation is that SiC particles frequently brake during indentation. Consi-dering the situation during sawing one can expect that the particle size distribution chan-ges along the sawing channel from the wire inlet to the wire exit. This has actually been observed and taken into account when analy-zing sawing result on the wafer surface.

The indentation investigations show a certain sequence of microcrack evolution. In the loa-ding phase mainly vertical (median / radial) cracks are generated, whereas during unloa-ding lateral cracks (parallel to the surface) oc-cur, which finally lead to chipping of material. The vertical cracks remain to some extend in the bulk and form the subsurface damage, which is observed on as-sawn wafers.

2.3 Indentation experiments 2.3.1 Microcrack development

Fig. 4: Microcracks at particle indentations. (left) Vickers indentation with sapphire tip, (right) indentation with a SiC particle in the center.

46

Cracking occurs when the fracture stress is exceeded. The stress distribution below an in-denter is however very complex, particularly for irregularly shaped particles. In addition, one has to take into account the local crystal-lography and that the material may undergo phase transformations. Quantitative predic-tions are therefore difficult and require more detailed information about the influence of the various factors. The approach here was to investigate the stress state in nanoindentation experiments and by corresponding numerical simulation.

2.3.2 Nanoindentation experimentsNanoindentation experiments have been per-formed using the UNAT (Asmec GmbH) in-dentation device, for which the manufacturer specifies a noise level for force measurement of ≤6µN and a noise level for displacement measurement of ≤1 nm. A load cycle consists of a 10 s loading portion, a 5 s hold segment (which, since silicon does not creep, serves as a measure for the thermal drift) followed by 5 s of unloading. Care was taken to minimize thermal drift. During processing, the data was corrected for the effect of finite instrument compliance and the point of contact was de-termined via backward extrapolation to zero force.

2.3.3 Characterization of the residual imprints

Atomic force microscopy (AFM) imaging was carried out using a Nanite B (Nanosurf GmbH) mounted on the frame of the nanoin-denter. The scan head has a maximum scan range of 110 µm and a maximum vertical scan range of 22 µm. Imaging was performed in tapping mode. We used cantilevers of the type Tap190-Al-G (BudgetSensors), which are monolithic silicon probes with resonant frequency of 190±60 kHz and a tip radius of less than 10 nm. 512 lines with 512 points each were recorded. The time per line was set to 3 s.

Fig. 7a shows a three-dimensional view of the residual imprint, while Fig. 7b shows a “bot-tom” view of the same indent, nicely illustra-ting its pyramidal shape. The in-plane shape of the imprint is best seen in a clipped density plot view (Fig. 7c), showing a concave triang-le, which is typical for materials with a large ratio of hardness to elastic modulus. Finally, Fig. 7d shows the height profile through the residual imprint following the path indicated by the dashed arrow in Fig. 7c.

0.00 0.05 0.10 0.15 0.20 0.25 0.30

Displacement h (in µm)

0

2

4

6

8

10

12

14

16

Force

P(inmN)

measured curves

averaged curve

Fig. 5 shows ten load – displacement cur-ves for the Berkovich indenter tip (shown in Fig. 6) into a (111) single crystal silicon sur-face as well as their average. This illustrates the very high repeatability of indentation tests in silicon. The β-Si → a-Si transition during unloading reflects as a kink (elbow) on the load – displacement curve.

Fig. 5: Force – displacement cur-ves for 15 mN Berkovich indents in (111) silicon.

47

Fig. 6: The Berkovich indenter tip.

500 nm

500 nm0.0 0.5 1.0 1.5 2.0 2.5 3.0

µm

1.36

1.38

1.40

1.42

1.44

µm

Fig. 7: AFM surface topography data of 15 mN indent with Berkovich tip in (111) silicon.

(a) Top View. (b) „Bottom“ view.

(c) Clipped density plot 8d) Height profile along the dashed line in Fig. (c)

48

Based on previous results [27, 25, 24], we developed a constitutive model [26] that captures the cd-Si → b-Si, a-Si → hda-Si and (b-Si, hda-Si) → a-Si transitions within the framework of thermodynamics with internal variables. Since both the elastic and inelastic strains are moderately large, a finite defor-mation framework is employed based on the multiplicative decomposition of the deforma-tion gradient. The constitutive equations are reformulated in rate form, employing only symmetric second order tensor valued and scalar variables. The model was implemented as a user material subroutine for the finite ele-ment code Abaqus/Std. in analogy to pressure sensitive, rate-independent, non-associated, non-smooth multisurface plasticity within a “hyperelastic-plastic” framework . Material parameters were determined from load – dis-placement curves obtained using a Berkovich indenter in (111) silicon by the solution of an inverse problem. The resulting fit of the ex-perimental data is shown in Fig. 8.

2.4 Numerical calculations

Fig. 8: Experimental load – displacement curve and best fits using our constitutive model and von Mises plasticity.

0.00 0.08 0.16 0.24 0.32

Displacement h (in µm)

0

2

4

6

8

10

12

14

16

Force

P(inmN)

experimental reference

our model

von Mises plasticity

We modeled the sample as an elastic half-space that is subjected to penetration by a compliant diamond indenter with elasticities E=1140 GPa, ν=0.07. The indenter was con-structed from the projected area function obtained in indentation experiments with reference materials [28].

As discussed in Sec. 2.3.1, the initiation and propagation of median and radial cracks un-der contact loading is dominated by the ap-plied load and the geometry of the asperity/indenter. During unloading residual stresses not only cause these cracks to propagate fur-ther and coalesce but initiate and drive lateral cracks (more or less parallel) to the sample surface. The residual stresses strongly depend on the material behavior and are the subject of investigation of this section. In the following, we make a qualitative and quantitative com-parison between the results predicted by our phase transformation model and ideal von Mises plasticity used as a reference.

49

Fig. 9: First principal stress distribution in the cross section after unloading.

Fig. 10: S33 stress distribution in the cross section after unloading.

Fig. 9 shows the first principal stress contours in „side view’’ on one third of the Si sample (see schematic in the lower right corner). The region of high tensile principal stresses pre-dicted by our model spans from the surface of the sample to the axis of the indent, while the plasticity model predicts a concentration of similar stress magnitude right underneath

the sample surface at the boundary of the im-print. In both cases the first principal stress is oriented almost perpendicular to the depict-ed planes. We can conclude that the phase transformation model predicts the growth of both median and radial cracks as well as their possible coalescence, while only the growth of shallow radial cracks is predicted by von Mis-es plasticity.

50

Fig. 10 shows the normal stress component in the direction of the indenter axis. This is the residual stress driving lateral cracks. The location of maximum S33 stress predicted by both constitutive models is typical for the for-mation of shallow lateral cracks. However, the phase transformation model predicts larger stresses with a concentration about 30% clos-er to the sample surface.

The Yoffe [36] model suggests that it is worth-while to consider radial stresses in a spherical coordinate system (with its origin located at the center of the indent on the level of the original surface) as the driving force for deep lateral cracks. Our model predicts a radial stress maximum underneath the transformed zone (see Fig. 11). This is consistent with ex-perimental observations in [23]. The plasticity model also predicts a stress concentration un-derneath the transformed zone; however, it is much smaller in magnitude and located about twice as deep.

Fig. 11: Srr stress distribution in the cross section after unloading.

This suggests that the chipped volume due to deep lateral cracks is severely overestimated by plasticity based models. Interestingly, the von Mises plasticity model predicts the global radial stress maximum at a different location. It is found directly at the sample surface on the perimeter of the imprint (see Fig. 11). This suggests the formation of cone cracks, which however have never been observed in Berk-ovich indents in silicon.

51

3 Fixed Abrasive - Crystal Interaction 3.1 Sawing experimentsSawing with diamond coated wires (fixed abrasive sawing - FAS) requires different con-ditions. The wafers were sawn in the oscillat-ing mode (contrary to the previous LAS case) using a commercial diamond coated wire with a total thickness of 140 µm. The height of the diamond particles and coating layer was about 10 µm. The density of the particles on the wire surface has been estimated to be about 5 x 104 cm-2. The line density along the wire length is about 102 cm-1 (Fig. 12). Wafer thickness-es were around 180 µm. The multi-wire saw used here (Meyer&Burger DS 264) was also equipped with a force sensor to measure the forces normal (sawing direction) and parallel to the wire direction. A water based coolant fluid was used.

Figure 12: Diamond particles on a wire (Confocal microscopy image)

Figure 13: Height profile of an LAS (a) and FAS sawn wafer (b). Profile height are between 0 - 5 µm.

Roughness, grooves and microcrack distri-bution on the surface were determined by the same methods as described in Sec. 2.2. Fig. 13 shows an example in comparison with a slurry sawn wafer surface. One can observe a different surface morphology, which indi-cates already that the material removal pro-cess is different here.

The corresponding microcrack distribution is depicted in Fig. 14. The main feature is the arrangement of a periodic crack pattern in wire motion direction. It consists of a mi-crocrack parallel to the scratch direction and one or two cracks to the side. The length of the deepest crack corresponds to the average repetition distance of the pattern.

3.2 Wafer surface characterization

52

Figure 14: Microcrack traces along scratches on a beveled, etched surface of an as-sawn wafer. Enlargement shows the microcracks pattern in single scratch grooves.

One striking result is that the microcrack depths on mono- and multicrystalline silicon differ for the same sawing conditions. Such a difference has not been observed on LAS sawn wafers. A closer investigation of the microc-rack depths on single grains on a multicrys-talline wafer shows a dependence on the grain orientation (Fig. 15). Particularly for grains with a [110] orientation parallel to the wafer surface normal two different results have been observed. We could show that a larger crack length is obtained, when the scratch direc-tion is parallel to the trace of one of the (111) fracture planes perpendicular to the wafer surface. In all other cases the fracture planes are always inclined, which obviously leads to shorter microcracks.

Figure 15: Maximal microcrack lengths in grains of different orientations in a multicrystalli-ne silicon wafer (FAS sawn).

Such detailed investigations are however time-consuming. Therefore further quan-titative investigations were carried out on single scratches, which were generated with the scratch-indenter apparatus developed by Fraunhofer Technology Center for Semicon-ductor Materials.

53

The scratch test with pointed asperity is a physical model for the microscopic processes during wire sawing with fixed abrasive.

3.3 Single scratching experiments

In the scratch-indenter a single particle is mounted on the rim of a rotating wheel. The specimen is pushed shortly upwards, so that a single scratch is produced. Scratch velocities up to 20 m/s can be produced. During contact the forces in vertical and horizontal direction are measured. The scratch tests have been carried out on monocrystalline surfaces with particles from a diamond-coated wire.

A typical example of the surface damage is shown in Fig. 16 (top). It consists of a heav-ily damaged core region and a high density of microcracks to both sides. These deeper microcracks are shown in a series of images taken at increasing depths below the surface

3.3.1 High speed scratch tests

Figure 16: (top) SEM images of a scratch produced with a diamond wire particle at a speed of 0.1 m/s. The force acting on the particle is 420 mN. (bottom) SEM images of microcracks below the same scratch at increasing depths below the surface (from top to bottom). The depth increases from left to right. The scale is the same in all images. The maximal microcrack depth is 13 μm.

(Fig. 16, bottom). They consist of a vertical crack directly in the scratching direction and cracks to the side approximately along the <110> directions.

The periodic pattern can also be seen on a cross section of the wafer along the scratch direction (Fig. 17). We assume that the per-io-dicity is equal to the extension of the re-gion, which is chipped away near the surface in one single event. The results indicate the following sequence of events during scratch-ing. At the beginning, when the moving tip is blocked by material, it induces a plastic zone below and mainly in front of the tip (Fig. 18). One can expect regions of tensile and compressive stresses. If the applied forc-es induce stresses, which exceed the fracture stress, the material breaks and cracks occur. Lateral cracks will lead to chipping and the re-moval of material. After chipping the tip can move freely until it is blocked by material and stresses build up again.

54

Figure 17: SEM image of deep microcracks on a wafer cross section along the core region of scratch.

Figure 18: Schematic diagram of the forces and deformations at the scratching tip before cra-cking.

The quantitative force measurements are summarized in Fig. 19 for different particle shapes and can be described by the following equation

( )0 minn

totc c F F= − , (1)

where n varies between 0.38 - 0.49. A threshold force of Fmin = 140 mN was taken in all cases. One can observe for the needle tests that there is only a slight difference if the scratching direction is reversed. This indicates that the crystal orientation plays a minor role here compared to the applied forces.

55

Figure 19: Calculated total forces Ftot - Fmin on a single scratching particle vs. ma-ximal crack lengths. Solid lines calculated from equa-tion 1. Thick point: result from wafers sawn with a diamond wire.

The results have also been compared with the microcrack lengths obtained from diamond wire sawn wafers. The total vertical and lateral forces on a single wire have been determined to be about 1.5 and 3 N, respectively. The damage on the wafer surfaces results from the particles on the side of the wire. We assume that all forces on the constant area of the wire are of the same magnitude. Due to the size

distribution of the protruding diamond par-ticles we also estimate that along the entire wire length on the side only about 1% of the particles are scratching at the same time [14]. This yields about 15 particles and thus forc-es FV and FR on a single particle 100 mN and 200 mN, respectively. The resulting value is included in Fig. 19 and fits into the curve for the scratching test with the diamond particles.

The response of silicon to scratching at low loads was investigated using the UNAT testing device, which allows to apply lateral displace-ments superimposed to force or displacement based loading vertical to the surface. All forc-es and displacements can be recorded during the experiment with noise levels of ≤10 µN and ≤1.5 nm for the lateral head.

All experiments were performed in displace-ment control. Fig. 20 shows the AFM height profile of such a 10 µm long scratch with Berkovich indenter tip and a penetration

3.3.2 Low load, low speed scratch tests

force of 10 mN before scratching. Lateral force – displacement curves for 1 µm and 10 µm long scratches are given in Fig. 21. It is in-teresting to note, that the lateral force quickly saturates, suggesting that that we operate in a steady state ploughing regime.

In summary, one can conclude that with sin-gle scratch tests the elementary process dur-ing diamond wire sawing can be studied from the bottom up, providing quantitative infor-mation about the mechanical interactions.

56

0.0 2.5 5.0 7.5 10.0

Lateral displacement (in µm)

0.0

0.5

1.0

1.5

2.0

2.5

Lateral

force(inmN)

We applied the constitutive model for silicon, which was previously fitted to indentation load – displacement curves to the simulation of scratching with a Berkovich indenter tip. Based on symmetry considerations we mod-eled and meshed only one half of the sample

3.4 Numerical calculations

Fig. 20: AFM micrograph of a 10 µm scratch with 10 mN penetration force.

Fig. 21: Lateral force - displace-ment for a 10 µm scratch with 10 mN penetration force.

and indenter. Specifically, the sample mesh consists of a fine homogeneous region em-bedded into a coarser mesh, which in turn is embedded in a layer of halfspace elements (see Figs. 22 and 23).

57

Fig. 22: Top view of the FE mesh for scratch simula-tions.

Figures 24 and 25 show the evolution of the β-Si zone over the length of a 1 µm scratch. Fig. 26 presents the maximum principle stress distribution in top view (indenter removed in this representation). The region of large tensile stresses is located in the wake of the indenter, while the orientation is somewhat tilted relative to the scratch direction. Cracks would open in the same way as is observed experimentally (see Fig. 16). The simulation does however not include the presumed stick-slip behavior discussed in Sec. 3.3.1.

Fig. 23: Side view of the FE mesh for scratch simula-tions.

The predicted lateral force – displacement curve is shown in Fig. 27 along with experi-mental data for the 1µm and 10 µm scratches as well as a simulated curve using von Mis-es plasticity. We note that once the force has saturated the match between the experimen-tal and the simulated curves is nearly ideal. Deviations only occur in the initial loading portion, in which however, the experimen-tal curves show considerable spread. Most interestingly, the force – displacement curve predicted using von Mises plasticity shows an entirely different behavior; the force never saturates, but keeps increasing with a nearly constant slope throughout the scratching pro-cess.

58

Fig. 24: Transformed zone after initial penetration into the material.

Fig. 25: Transformed zone at the end of a 1 µm scratch.

Fig. 26: Maximum principal stress distribution underneath the indenter tip during scratching.

59

Fig. 27: Lateral force – displacement curves. Comparison between experiments and nu-merical predictions.

0.0 0.5 1.0 1.5 2.0

Lateral displacement (in µm)

0.0

0.6

1.2

1.8

2.4

3.0

Lateralforce(inmN)

Experiment: 10 µmExperiment: 1 µmSimulation: plasticity

Simulation: phase transition

The present work deals with phenomena rel-evant for the understanding and characteri-zation of material removal mechanisms and residual damage during multi-wire sawing with fixed and loose abrasive particles. The experimental investigations were carried out on various length scales, from the macro-scopic sawing process down to the single par-ticle interaction on micro scale, for different process parameters and tightly linked with a characterization of residual defects and their distributions.

The investigations at the micro scale are sup-ported by numerical simulations with a newly developed constitutive model for silicon un-dergoing stress – induced phase transforma-tions. After a calibration using indentation ex-periments the model was able to predict other results, including the surface topography of indents. Moreover, load – displacements curves for scratch tests could be simulated and verified experimentally. We believe that the availability of proper material modeling supported by experimental investigations is a crucial step towards the understanding of abrasive machining of silicon.

4 Summary

The final goal is to optimize the fabrication processes whilst ensuring minimal surface defects. In order to bridge the gap between the elementary microscopic abrasion process-es and macroscopic sawing, meso scale sim-ulations considering an ensemble of particles or asperities interacting with silicon are re-quired. Some work carried out using the dis-crete element method exists in literature (see e.g. [37]). The results obtained in the present study at the micro scale will provide essential information to these meso scale simulations, which will be the subject of future investiga-tions.

60

5 References

1. A. Bidiville, K. Wasmer, R. Kraft, C. Ballif, Proc. 24th EU PVSEC (2009) 14002. D. Kray, M. Schumann, A. Eyer, G. P. Willeke, R. Kubler, J. Beinert, G. Kleer, IEEE 4th

World PV Co. (2006)3. Y. Kondo, N. Watanabe, D. Ide, T. Matsuki, H. Takato, I. Sakata, Proc. 23rd EU PVSEC

(2008) 12974. Lombardi, G. Fragiacomo, C. Zehetmeier, J.-I. Bye, Ø. Nielsen, C. Rohr, B. Gäumann, A.

Künzli, Proc. 24th EU PVSEC (2009) 12565. J.-I. Bye, S. A. Jensen, F. Aalen, C. Rohr, Ø. Nielsen, B. Gäumann, J. Hodsden, 24th EU

PVSEC (2009) 12696. Bidiville, K. Wasmer, J. Michler, P. M. Nasch, M. Van der Meer, C. Ballif, Prog. Photov.

Res. Appl (2010) 7. H. J. Möller, Advanced Eng. Materials 6 (2004) 5018. H. J. Möller, phys. Stat. Sol. 203,4, (2006) 6579. H. J. Möller, S. Retsch, R. Rietzschel, On the origin of wafer saw marks in slurry based

multi - wire sawing, Proc. 28th EU PVSEC, (Paris, 2013) p. 92710. C. Funke, O. Sciurova, W. Fütterer, H. J. Möller, Proc. 21th EU PVSEC (2006) 125611. Y. Gogots, C. Baek, F. Kirscht, Semicond. Sci. Technology 14 (1999) 936.12. Evans, E. Charles, J. Am. Cer. Soc. 59 (1976) 371.13. J. Lankford, D. Davidson, J. Mater. Sci. 14 (1979) 166214. J. Hagan, J. Mater. Sci. 14 (1976) 2975.15. H. J. Möller, in Crystal Growth Technology (Eds. H. J. Scheel, P. Capper, Wiley, Wein-

heim, 2008) 45716. Th. Behm, C. Funke, H. J. Möller, Surface orientation characterisation of rough mc-

silicon surfaces by confocal microscopy and EBSD, Surface and Interface Analyses 2012 (Wiley & Sons) Wiley Online Library

17. H. J. Möller, R. Buchwald, S. Winstanley, Basic mechanism of diamond wire sawing, Proc. 6th Int. Conf. Crystalline Silicon for Solar Cells, (Aix-les-bains 2012) 83

18. R. Buchwald, K. Fröhlich, S. Würzner, T. Lehmann, K. Sunder, H. J. Möller, Analysis of the sub-surface damage of mc- and cz-si wafers sawn with diamond-plated wire, Energy Procedia, Volume 38, 2013, 901

19. R. Buchwald, K. Fröhlich, S. Würzner, T. Lehmann, K. Sunder, H. J. Möller, Analysis of the sub-surface damage of mc- and cz-si wafers sawn with diamond-plated wire, Proc. 28th EU PVSEC, (Paris, 2013) p. 1502

20. S. Würzner, S. Retsch, R. Buchwald, H. J. Möller, A new view on the microcrack structure (sub-surface damage) by single scratch tests with diamond particles on monocrystalline silicon wafers, Proc. 29th EU PVSEC, (Amsterdam, 2014), accepted

21. R. Buchwald, S. Würzner, K. Fröhlich, M.Fuchs, S. Retsch, T. Lehmann, H. J. Möller, Ana-lysis of the topography and the sub-surface damage of Cz- and mc- silicon wafers sawn with diamond wire, 40th IEEE Conf. , Denver, 2014, accepted

22. H. J. Möller, Wafer Processing, Handbook of Crystal Growth, Second Edition, Vol 2, chapter 18, (Elsevier, 2014) 40 Seiten

23. J. Bradby, J. Williams, J. Wong-Leung, M. Swain, and P. Munroe. Me- chanical deformati-on in silicon by micro-indentation. Journal of Materials Research, 16(5):1500–1507, 2001.

24. M. Budnitzki and M. Kuna. Modeling silicon under contact loading condi- tions: aspects of non-associated flow. Technische Mechanik, 32(2):146–154, 2012.

25. M. Budnitzki and M. Kuna. A thermomechanical constitutive model for phase transfor-mations in silicon under pressure and contact loading condi- tions. International Journal of Solids and Structures, 49(11-12):1316–1324, 2012.

61

26. Michael Budnitzki. Constitutive modeling and experimental investigations of phase tran-sitions in silicon under contact loading. PhD thesis, TU Bergakademie Freiberg, 2014.

27. Michael Budnitzki and Meinhard Kuna. A novel constitutive law for silicon under contact loading. Proceedings in Applied Mathematics and Mechanics, 11(1):357–358, 2011.

28. T. Chudoba. Nanostructured Coatings, chapter Measurement of hardness and Young’s modulus by nanoindentation, pages 216–260. Springer, February 2007.

29. V. Domnich and Y. Gogotsi. Phase transformations in silicon under contact loading. Reviews on Advanced Materials Science, 3:1–36, 2002.

30. J.J. Gilman. Shear-induced metallization. Philosophical Magazine Part B, 67(2):207–214, 1993.

31. JZ Hu, LD Merkle, CS Menoni, and IL Spain. Crystal data for high-pressure phases of silicon. Physical Review B, 34(7):4679–4684, 1986.

32. T Juliano, Y Gogotsi, and V Domnich. Effect of indentation unloading conditions on phase transformation induced events in silicon. Journal of Materials Research, 18(5):1192–1201, 2003.

33. MI Mcmahon and RJ Nelmes. New high-pressure phase of Si. Physical Review B, 47(13):8337–8340, 1993.

34. MI Mcmahon, RJ Nelmes, NG Wright, and DR Allan. Pressure-dependence of the imma phase of silicon. Physical Review B, 50(2):739–743, 1994.

35. Welber, C.K. Kim, M. Cardona, and S. Rodriguez. Dependence of the indirect energy gap of silicon on hydrostatic pressure. Solid State Communications, 17(8):1021–1024, 1975.

36. E.H. Yoffe. Elastic stress fields caused by indenting brittle materials. Philo- sophical Ma-gazine A, 46(4):617–628, 1982.

37. B. Nassauer, A. Hess, M. Kuna. Numerical and experimental investigations of micro-mechanical processes during wiresawing. International Journal of Solids and Structures, 51(14):2656-2665, 2014.

62

Development of a new process for recycling of used sawing slurries from solar industry

I. Nitzbon, A. Obst, U. Šingliar, M. Bertau

Department of Industrial Chemistry, TU Bergakademie Freiberg

64

AbstractThe wire sawing process is one step of the production of photovoltaic silicon wafers. With a suspension (so-called slurry) consisting of silicon carbide used as the abrasive and polyeth-ylene glycol used as a coolant, the wafers are cut from a solar grade silicon ingot. During this operation silicon abrasion accumulates in the sawing slurry. Over time the contaminated slurry can no longer be used in the sawing process due to the accumulation and has to be replaced periodically. Thus, large amounts of waste slurry are generated in the global production of wafers. Because of economic and ecological reasons recycling of all slurry components should be the goal. Therefore a new process is developed, which is based on the separation of polyeth-ylene glycol by a acetone assisted filtration and the recovery of silicon carbide by the reaction of silicon/silicon carbide mixture with chloromethane. This new process enables the reuse of polyethylene glycol as well as silicon carbide in the wafer sawing process. The obtained chlo-rosilanes are precursors for further products, especially to obtain silicones.

So far there are only few technical options for the recycling of used slurries from the wire sawing process. Published processes are based on mechanical separation methods of the all slurry components or on chemical reaction for silicon/silicon carbide separation. The main problem for size sorting processes is the small particle size of silicon carbide as well as silicon (< 1-10 µm) in the used slurries. The SiC Processing GmbH developed a method for the separation based on the application of a liquid-solid centrifugal separator. The resulting solid-containing fractions are re-fined by a battery of up to five hydrocyclones. The product consists of reusable silicon car-bide, 2-5 % Si and Fe fines. This reveals that a complete separation of the silicon and silicon carbide is not possible with this method [1]. The patent of Kanuit et al. [2] describes the possible separation of the solid fraction under use of the centrifugal separator. In this pro-cess the slurry is heated to reduce the viscosity of polyethylene glycol. In both cases [1, 2] the resulting liquid fraction is treated by a dis-tillation process to obtain pure polyethylene glycol. Another method for the separation of silicon and silicon carbide in spent slurry is based on density differences [3]. Therefore the density of the slurry has to be adjusted with a liquid (e. g. methanol) to 1.6 and 1.7 g/cm3. After stirring or centrifugation the silicon car-bide settles and the silicon remains in the flu-id. The declared purity of the resulting solids is 99 % (silicon as well as silicon carbide), but the grain sizes of the solids in the used slurry

are not specified. So it is doubtful to reach a sufficient selectivity with this method if the grain size is in the range of < 1-10 μm.

The separation of silicon and silicon carbide with chemical methods is based on the re-action of silicon to liquid or gaseous prod-ucts while silicon carbide does not react and remains as a solid [4, 5, 6]. The reaction is carried out in form of direct chlorination with hydrogen chloride [4, 5, 6]. However, the principal purpose of these methods is to form chlorosilanes which are precursors for further products e. g. to obtain high-purity silicon for the reuse in the wafer production process. Up to now there does not exist a known technical application for these referred methods.

In addition to the separation processes men-tioned above there are still a number of pro-posals in various publications. There are stud-ies of particle sedimentations in an electric field, studies based on phase transfer separa-tion and methods based on selective melting of the silicon/silicon carbide mixture. How-ever, the silicon obtained in these methods shows only a low purity and industrial pro-cesses on the basis of the investigations are not known [7, 8, 9].

For development of an alternative method for the slurry recycling focus was put on the con-version of all slurry components including sil-icon carbide as the preferred product as well as silicon and polyethylene glycol.

State of the art

65

Three sawing slurry samples were provided by the Solarworld AG. Two of them were used and the third, called fresh slurry, was recycled by SiC Processing GmbH, Bautzen.

The used slurry consisted of 48.7 wt.% solid. After separation of polyethylene glycol the amount of silicon in the dried solid phase (16.9 wt.%) was determined with the help of a carbon/sulfur analyser. The remaining sol-id consisted of 81.5 wt.% silicon carbide and 1.6 wt.% other metal impurities. The chem-ical composition was determined by ETV- ICP-OES. The results are shown in Table 1.

The varying metal compositions of the used slurries for example 339 ppm aluminum in the used slurry 1 compared to 192 ppm in the used slurry 2 might be based on differ-ent usage times of the saw wire used during the sawing process. The high iron concentra-tion in the used slurry is a result of abrasion during wire sawing. The iron content in the fresh slurry presumably indicates an incom-plete separation during the recycling process.

For further comparisons of the slurries par-ticle size measurement were performed. The results are shown in figure 1.

Table 1: Summary of the ETV-ICP-OES analysis of the main metal impurities in the fresh slurry and two different spent slurries (used slurry 1, 2)

element used slurry 1 used slurry 2 fresh slurry [ppm] [ppm] [ppm]

Fe 15659 8571 1611

Cu <LOD 289 18

Mn 115 29 9

Al 339 192 315

Mg 246 29 21

P <LOD 36 13

Figure 1: The particle size distri-bution of fresh and two differ-ent spent slurries

Characterisation of investigated SiC-based sawing slurries

66

The fresh slurry shows a monomodal distri-bution with a maximum at 10 μm. The used slurry shows a bimodal distribution. Both an-alysed samples of used slurry show only small differences in the grain size distribution. The used slurries contain more fine particles which can be explained with the abrasion of silicon and crushed silicon carbide during the sawing process. Figure 2 shows scanning elec-tron microscope images of recycled and spent slurries.

A comparison of the images illustrates the im-pact of the sawing process on the grain size of the silicon carbide particles. The sawing grain was very heavily used, so that a very high pro-portion of fine particles are present at the end of the sawing process. There are no significant differences between the two samples of the used slurries. In contrast, the recycled slur-ry (figure 2, left image) consists of uniformly sized particles without fines.

Figure 2: Scanning electron microscope images of fresh slurry (left) and used slurry 2 (right) at 1000x magnification

Process development for separation and re-cycling of polyethylene glycol and characteri-sation of the different PEG samples

The separation of polyethylene glycol was investigated by centrifugation, vacuum filtra-tion and pressure filtration. Due to the high viscosity and the small particle size of the solid content in the slurries the separation process was difficult. Centrifugation experi-ments, performed at 15000 rpm for at least 20 minutes at temperatures between 20 °C and 30 °C, were not successful. The fine grain re-mained in the suspension in each case, so that complete separation of the solid could not be achieved. Also the use of vacuum filtration is not sufficient, since the separation was not possible within a reasonable time frame. In order to achieve a short filtration time the separation of the solid parts was realised with a stainless steel pressure filtration device. The filtration was tested with 0.2 micron reversed

cellulose filter as well as 0.5 μm, 1 μm and 2 μm glass fiber filters. Still the filtration was not practicable. For example, the filtration of 150 g slurry with a 1 μm glass fiber filter and a pressure of 6.5 bar had to be cancelled after 30 minutes because of filter clogging. To establish an adequate filtration process is was necessary to reduce the viscosity of the slurry. A complete heating of the used slur-ry was tested but a major improvement of the filtration time was not achieved. Also a com-plete heating of large amounts of used slurry is not preferable due to the presumable high energy consumption. Nevertheless, in order to reduce the viscosity acetone was added to the slurry. There are some advantages in the use of acetone compared to other chemicals.

67

It is nontoxic and it is non-reactive to polyeth-ylene glycol. The boiling point is lower than water and thus the removal after filtration is easier and less energy-intensive. It was pos-sible to achieve good results in terms of the filtration time with acetone contents of 15-20 wt.%. Under these conditions the filtration of 150 g slurry using a 1 μm glass fiber filter at pressures of 2-3 bar was possible in less than two minutes. However, the fine grain of the slurry was not retained with 0.5 μm to 2 μm filters despite repeated filtration (filter cake filtration) and leads to a second filtration step. For this purpose a 0.2 μm reversed cellulose filter was used, which allowed the complete retention of fine grain without an additional dilution of polyethylene glycol. The second filtration step could be finished at a pressure of 2 bar in less than 2 minutes. The obtained coarse grain from the first filtration step was dried after washing with acetone in an oven at 80 °C. The amount of fine grain obtained in the second filtration step makes up less than 0.01 wt.% of the original slurry so that further treatment does not seem reasonable.

Different methods for the separation of ac-etone from polyethylene glycol were tested. The removal of acetone by distillation (70 °C), vacuum distillation (40 °C, 100 - 270 mbar) and a stripping method with nitrogen (40 °C)

were investigated. Distillation at 70 °C did re-sult in a complete separation of the acetone but it took 3 hours to remove 12 g of ace-tone out of a mixture of 60 g, which proved this method to be impractical. With vacuum distillation no complete separation could be achieved. Hence an additional purification step e. g. stripping with nitrogen would be needed. The best results were achieved with the single stripping step. Complete separation of the acetone (20 wt.% polyethylene glycol) could be achieved at 40 °C with a nitrogen flow rate of 20 l/h in 2 hours.

The polyethylene glycol contained in the fresh and used slurries, the recycled polyethylene glycol as well as a PEG-200 sample from Alfa Aesar were characterised in terms of pH, wa-ter amount and refractive index. The density and dynamic viscosity of the sample from Alfa Aesar as well as the established process sample were determined in addition to the other parameters. For the determination of the dynamic viscosity the Ubbelohde method was used. Karl Fischer titration was used to measure the water content of the samples.

The specification of polyethylene glycol from Brenntag AG, which is used in the sawing process at Solarworld AG, serves as compar-ison.

Table 2: Characterisation of different polyethylene glycol samples in terms of density ρ, pH of 10 % PEG in water, dy-namic viscosity Vdyn at 20 °C, water content and refractive index nD20

sample ρ pH Vdyn

H2O content nD20

[g/cm³] (solution 10 %) [mPa/s] 20°C [%]

PEG (Brenntag) 1.12 4-7 60-70 ns. ns.

PEG (Alfa Aesar) 1.12 5.8 66 0.2 1.460

PEG (fresh slurry) - 4.8 - 1.1 1.59 (0.57% in suspension)

PEG (used slurry 2) - 4.8 - 2.1 1.459 (1.0% in suspension)

PEG (pressure, 1.12 4.4 63 1.9 1.458 filtration, stripping)

68

Table 2 shows that the fully processed poly-ethylene glycol (stripping method) corre-sponds to both of the reference substances (Alfa Aesar, Brenntag) in terms of density, pH value and viscosity. The determined refractive index of 1.458 matches the literature value of PEG-200 [10]. In respect to density, pH, vis-cosity and refractive index it can be assumed that the material does not undergo changes during the recycling process.

The amount of water in the used slurry is higher than that of the fresh slurry. Because polyethylene glycol is highly hygroscopic the amount of water rises during the sawing pro-cess and handling. If the amount of water in the slurry is too high the viscosity will change and may have negative affect on the sawing

process and furthermore promotes the cor-rosion of the wire equipment. Therefore it is necessary to reduce the amount of water in the recycling process as much as possible. A significant reduction of the water content in the reprocessed polyethylene glycol could not be achieved by using the pressure filtra-tion/stripping method compared to the wa-ter content of the polyethylene glycol in the used slurry. The use of drying agents like mole sieves should be considered to minimise the water content in the reprocessed slurry.

The following scheme (figure 3) represents the overall process scheme for the separation of polyethylene glycol.

Figure 3: Process scheme for recovery of polyethylene glycol from used silicon/silicon carbide slurry

The developed process offers numerous ben-efits in comparison to processes in literature. On the one hand the filtration step runs rap-id even with low pressure and on the other hand the small fines can be retained com-pletely. The filtration aid acetone can easi-ly be separated and recycled by stripping.

This is of great benefit towards the usage of water, which produces high costs for clarifi-cation and/or removal. Stripping methods for the recycling of acetone is less energy-inten-sive than distillation steps for the separation of water [1, 2, 3, 11, 12].

69

Rochow synthesis as a well-known technique is an interesting new way for separation of the silicon from the Si/SiC mixture. The Rochow synthesis is based on the reaction (1) of ele-mental silicon with methyl chloride to form methylchlorosilanes using a copper based cat-alyst.

The main product is dichlorodimethylsilane, which can usually be obtained in yields of 83-93 %. The main byproducts are trichloro-methylsilane (3-10 %) and chlorotrimethyl-silane (1-5 %). Other byproducts are H-con-taining silanes, tetramethylsilane and disilanes [14].

Figure 4: Reactor type I (fixed bed), II (stirred bed with zigzag stirrer) and III (stirred bed with anchor stirrer): 1. CH3Cl, N2 (inlet), 2. methylchlorosilanes (exit), 3. heater, 4. Si/SiC, 5. G3-frit, 6. stirrer

Process development for the conversion of a silicon/silicon carbide mixture with chloro-methane as well as the characterisation of gas-eous and solid products

The reaction is usually carried out in a flui-dised bed reactor in which the reaction gas flows from the bottom through the bed. The used silicon has a particle size of up to 500 μm [14]. In contrast the silicon carbide in the slurry of the wire sawing process consists of very small particles (about 10 μm). In addi-tion, small silicon particles are introduced into the slurry during sawing so that the waste mixture covers a particle size range of 0.1 to 10 μm. Geldart group C showed that, it is not possible to form a stable fluidised bed with-in this particle range due to the existing co-hesion forces [15]. Therefore it was necessary to adapt the reactor design. So different con-struction types of reactors were tested (figure 4).

In the fixed bed reactor (reactor type I) no fluidisation could be achieved with any of the tested flow rates (up to 800 ml/min). Due to the lack of movement of the contact mass hot zones were formed which led to the forma-tion of large amounts of hydrochloric acid through pyrolysis of chloromethane as shown in the red infrared spectrum in figure 5 (hy-

drochloric acid signals between 2650 cm-1 and 3100 cm-1). In order to improve the mixing of the contact mass during the process, two fused silica stirrers (reactor type II and III) were tested. One stirrer formed as a zigzag-shaped rod (reactor type II) and could not move the contact mass completely. The other stirrer, an anchor stirrer (reactor type III), turned out to

Cl)CH(SiCl)CH(SiClSiCHClCHnSi 33223333 ++→+

(1)

70

Figure 5: Comparison of the infrared spectra of the pyrolysis of chloromethane and formation of hydrogen chloride (frequencies between 2650-3101 cm-1) during Rochow synthesis in different reactor setups (red: reactor type 1- fixed bed, blue: reactor type III- anchor stirrer)

be successful. The entire contact mass could be moved at a flow rate of 100-150 ml/min and a speed of 35 rpm. By the use of these flow rates only low discharge of fine particles was observed. No pressure rise could be de-tected in the reactor during the whole process

and the pyrolysis of chloromethane could be significantly reduced as shown in the blue in-frared spectrum in figure 5. In this spectrum only weak H-Cl vibrations between 2650 cm-1

and 3100 cm-1 could be detected compared to the red spectrum.

As already mentioned the reaction (1) was carried out using a copper based catalyst. Various catalytic active copper compounds are described in literature, for example cop-per(I)oxide, copper(II)oxide, copper(I)chlo-ride and copper(II)chloride [14, 15, 16, 17]. For the laboratory experiments anhydrous copper(II)chloride (5 wt.% and 10 wt.%) was used. In addition zinc powder (0.25 wt.% and 0.5 wt.%) is used as the promoter to improve the activity of the catalyst as well as the selec-tivity of the reaction to form dichlorodimeth-ylsilane, the preferred product of the Rochow synthesis [20].

The experiments were carried out at 360 °C up to 400 °C and with varying chloromethane/nitrogen ratios (0,3 - 0,8/1). Process monitor-ing was performed using an infrared gas flow

cell. The formation of silanes was observed at a temperature of at least 360 °C. This reaction temperature is similar to the temperature of 340 °C which is described for silane forma-tion using a copper(II)chloride catalyst [16]. Overall, no significant changes in the reaction rate could be achieved with different reaction parameters. After approximately 7 hours the complete conversion of silicon to silanes was finished and no more Si-H bonds could be de-tected in the infrared spectrum.

The characterisation of the gaseous products was carried out by infrared online measure-ments as well as gas chromatography-mass spectrometry detections (GC MS/MS). Figure 6 shows the infrared spectrum of the gaseous product at 400 °C (red) compared to the spec-trum of chloromethane (blue) at 400 °C.

71

Figure 6: Comparison of infrared spectra of chloromethane (blue) and the product gas (red) composi-tion of the reaction of silicon and chloromethane at 400 °C

Figure 7: Chromatogram of the composition of the product gas collected with a cold trap and dissolved in dichloromethan

The infrared spectrum of the gaseous product shows in contrast to the precursor chloro-methane (figure 6, blue line) Si-Cl, Si-H and Si-C-H bonds (figure 6, red line). Strong Si-H stretching vibrations occurred at 878 cm-1 and 2214 cm-1. Si-C-H bending vibrations could be detected at 1270 cm-1. Si-C-H bending as well as Si-C stretching vibrations occurred be-tween 811 cm-1 and 878 cm-1. Frequencies of Si-Cl and Si-Cl2 vibrations have been detected

between 450 cm-1 and 577 cm-1 [21, 22, 23]. Thereby the formation of chlorosilanes during the reaction was proven. For further differen-tiation of the formed chlorosilanes the prod-uct gas was collected in a cold trap which was placed after the reactor exit. The samples were dissolved in dichloromethane and analysed via gas chromatography-mass spectrometry (GC MS/MS).

72

As shown in figure 7 dichlorodimethylsilane and trichloromethylsilane could be detected as expected. Additionally trichlorosilane and tetrachlorosilane were present in the product. Tetrachlorosilane could have been formed by reaction of the HCl from pyrolysis with sili-con. The different formed chlorosilanes were not quantified in this project. In addition the optimisation of the process needs further investigations especially concerning the pre-ferred product dichlorodimethylsilane.

Through X-ray analysis it was possible to show that no more silicon could be observed in the solid product (figure 8). Still number of reflexes in the X-ray diffractogram could not be assigned to specific compounds. Based on

the formation of copper chloride (Nantokite) during the reaction, these reflexes are most likely different metal salts, which are formed by the reaction of chloromethane with the metal impurities (e. g. chlorides of iron and zinc) in the silicon/silicon carbide mixture. In addition it seems like the copper salt, which is used as catalyst reacts with chloromethane and forms elemental copper during the reac-tion. To separate copper chloride and the oth-er impurities the product was treated with wa-ter and HNO3. Due to the good solubility of the metal chlorides a separation can already be achieved with distilled water. An almost complete separation of the metallic copper which is formed during the reaction is possi-ble by treatment with a 69 % HNO3 solution.

Figure 8: X-ray diffractogram of the solid product after reaction with chloromethane

A further undesirable component in the solid product is carbon which results from pyroly-sis of chloromethane. To separate carbon the same reactor type as for the reaction of the sili-con/silicon carbide mixture with chlorometh-ane was used. Due to the fact, that the carbon oxidation takes place in a range from 370 °C up to 404 °C as shown in a thermogravime-

try analysis (figure 9) the solid product was treated with pure oxygen at a temperature of 450 °C until no more carbon dioxide forma-tion could be detected by infrared spectros-copy. At the end of this process step a product consisting of pure silicon carbide was ob-tained.

73

Figure 9: TGA of carbon oxidation from obtained silicon carbide

The pure silicon carbide obtained at the end of the complete recycling process consists main-ly of particles in the range of 4 µm up to 10 µm (figure 11). Consequently there is not much difference to the so called fresh slurry (from SiC Processing GmbH, Bautzen). Scanning electron microscopy image (figure 10) as well

as the particle size distribution show fine par-ticles in the recycled product, which are most likely silicon carbide particles crushed in the sawing process. For reuse of the reprocessed silicon carbide in the sawing process a classi-fication step should follow the main recycling process.

Figure 10: Scanning electron micro-scope image of recycled silicon carbide at 1000x magnification

74

Figure 11: Particle size distribution of used silicon/silicon carbide mixture, recycled sil-icon carbide and the fresh slurry

In summary the suggested method for re-cycling of the silicon/silicon carbide sawing slurry is shown in figure 12.

Figure 12: Process scheme for recycling of silicon/silicon carbide sawing slurries

75

At first, the used slurry underwent a pres-sure filtration process to remove the PEG from the solid. Removal of the silicon in the solid phase took place in a stirred bed reac-tor with a chloromethane-nitrogen mixture at a temperature of at least 360 °C. The con-tact mass consisted of silicon/silicon carbide, anhydrous copper(II)chloride as catalyst and zinc powder as promoter. The catalyst was re-moved in the following step by leaching with 69 % HNO3 at 25° C. In a last step, carbon as a result of chloromethane pyrolysis was re-

moved by oxidation to carbon dioxide by oxy-gen at temperatures above 450 °C.

At the end of the complete recycling process all components of the waste slurry are reus-able. Polyethylene glycol and silicon carbide, after a classification step, are suitable for reuse in the wafer sawing process according to their characteristics. The obtained chlorosilanes can be used as precursor for further products, especially to obtain silicones.

76

1. G. Fragiacomo (SIC Holding Geschäftsführungs GmbH), WO 2006/137098 A1, 2006.2. P. Kanuit, G. Fritsch (AKW Apparate und Verfahren GmbH, SIC Processing AG),

WO 2011/003782 A1, 2011.3. A. Menzel, A. Voitsch, M. Schmidt, S. Anders, T. Schmiady (SCHOTT Solar AG),

DE 10 2010 025 606 A1, 2010.4. F.-W. Schulze, F. Schaaf (PV Silicon Forschungs- und Produktions GmbH),

WO 2010/127669 A1, 2010.5. O. Pikhard, M. Scholz, T. Melin (RWTH Aachen), WO 2011/051334, 2011.6. P. Bakke, R. Gibala, J. M. Svalestuen, G. V. Ol (Norsk Hydro ASA), WO 2008/133525 A1,

2008.7. T. H. Tsai, Separation and Purification Technology: 78 (2011), S.16-20.8. Y. C. Lin, C. Y. Tai, Separation and Purification Technology: 74 (2010), S. 170-177.9. T. Y. Wang, Y. C. Lin, C. Y. Tai, R. Sivakumar, D. K. Rai, C. W. Lan, Journal of Crystal

Growth: 310 (2008), S 3403-3406.10. CSID:7908, http://www.chemspider.com/Chemical-Structure.7908.html (accessed 12:59,

Nov 4, 2014).11. C. Zavattari, G. Fragiacomo, E.Portaluppi (MEMC Electronic Materials, SpA), US

7223344B2, 2007.12. H. Müller, W. Schuldes (Scholz Recycling GmbH, NL Erfurt), DE 10/2008/022237 A1,

200913. K. Rissler, Chromatographia: 49, 11-12 (1999), S. 615-620.14. W. Kalchauer, P. Pachaly, Müller-Rochow Synthesis: The Direct Synthesis to Methylchlo-

rosilanes in Handbook of heterogenous catalysis, 2008, Wiley-VCH, Weinheim, S. 2635-2644.

15. D. Geldart, Powder Technology: 7 (1973), S. 285-292.16. H. Ehrich, D. Born, J. Richter-Mendau, H. Lieske, Applied Organometallic Chemistry: 12

(1998), S. 257-264.17. A. D. Gordon, B. J. Hinch, D. R. Strongin, Catalysis Letters: 133 (2009), S.14-22.18. L. Lewis, W. V. Ligon, J. C. Carnahan, Silicon Chemistry: 1 (2002), S. 23-33.19. H. Lieske, R. Zimmermann, Catalysis Letters: 33 (1995), S. 413-420.20. L. Rösch, P. John, R. Reitmeier: Silicon Compounds, Organic in Ullmann’s Encyclopedia

of industrial Chemistry Vol. 32, 2012, Wiley-VCH, Weinheim, S.637-674.21. J. P. Lambert, H.F. Shurvell, R.G. Cooks: Introduction to Organic Spectroscopy, 1987,

Macmillan, USA, 174-177.22. J. Coates: Interpretation of Infrared Spectra, A practical Approach in Encyclopedia of

Analytical Chemistry, 2000, John Wiley & Sons Ltd., Chichester, 10815-10837.23. P. J. Launer: Infrared Analysis of Organosilicon Compunds: Spectra-Structure Correla-

tions in Silicon Compounds: Register and Review, 1987, Petrarch Systems, 100-103.

References

77

Defect engineering in GaN layers grown by hydride vapor phase epitaxy

G. Lukin,1 O. Pätzold,1 M. Stelter,1 M. Barchuk,2 D. Rafaja,2 C. Röder 3 and J. Kortus 3

1 TU Bergakademie Freiberg, Institute of Nonferrous Metallurgy and Purest Materials, Leipziger Str. 34, 09599 Freiberg, Germany

2 TU Bergakademie Freiberg, Institute of Materials Science, Gustav-Zeuner-Str. 5, 09599 Freiberg, Germany

3 TU Bergakademie Freiberg, Institute of Theoretical Physics, Leipziger Str. 23, 09599 Freiberg, Germany

78

AbstractThe impact of kinetic effects on the homoepitaxial and heteroepitaxial growth of GaN by Hydride Vapor Phase Epitaxy (HVPE) has been studied. It was shown that kinetically limited growth conditions strongly affect the defect formation and stress evolution in GaN, and can be used as an in-situ method of defect engineering in order to control the material properties by HVPE. Application of kinetically limited growth during the homoepitaxy of GaN induces the generation of inverse pyramids (V-pits), which affect the threading dislocation density in deposited layers. The heteroepitaxial, kinetically limited growth of GaN in the temperature range of 750 - 900 °C resulted in a novel approach to deposit GaN layers directly on sapphire substrates by HVPE. The two-step deposition process includes the growth of GaN nucleation layers at in- termediate temperatures (750 - 900 °C) and a subsequent high-temperature overgrowth. The results of first experiments demonstrate the possibility to grow 10 - 15 µm thick, crack-free GaN layers of high crystalline quality direct on sapphire. Furthermore, the correlation between the residual stress and the density of threading dislocations was investigated. It was found that the increase of the threading dislocation density with the increasing compressive residual stress is different in dependence on the nucleation procedure. Still, some differences in the character of the dislocations were observed for the studied sample groups. The structural properties of grown GaN layers were characterized by scanning and transmission electron microscopy. The residual stress was determined using micro-Raman spectroscopy and high-resolution X-ray diffraction. The density of threading dislocations was concluded from the broadening of the reciprocal lattice points that was measured using high-resolution X-ray diffraction as well. The fitting of the reciprocal space maps allowed the character of the threading dislocations to be described quantitatively in terms of the fractions of edge and screw dislocations.

Keywords: gallium nitride, HVPE, kinetically limited growth, microstructure defects, residual stress

Introduction Gallium nitride (GaN) based materials are widely used for opto- and microelectronic applications, e.g. light emitting diodes, blue lasers, solar cells [1]. Due to the lack of na-tive substrates, GaN is typically grown he-teroepitaxially on foreign materials, such as sapphire (Al2O3), silicon carbide or silicon. The heteroepitaxial growth still remains one of the main factors limiting the performance of GaN-based devices. Owing to the high growth rate, the Hydride Vapor Phase Epitaxy (HVPE) is ascribed a great potential for pro-ducing free-standing GaN substrates to over-come this problem [2]. Several routes to ob-tain thick, high-quality HVPE layers starting from sapphire substrates are currently under investigation, such as (i) HVPE overgrowth of GaN templates produced by Metalorganic Va-por Phase Epitaxy (MOVPE) on sapphire [3], and (ii) the deposition of GaN layers direct-

ly on sapphire in a closed HVPE process [4]. An inherent drawback of the MOVPE/HVPE method is the combination of different growth techniques involving separate reactors and sample preparation. Furthermore, reproduc- ible quality of HVPE layers depends strongly on the parameters of the MOVPE templates, such as their stress level, which is difficult to be specified precisely [5]. Nevertheless, MOVPE/HVPE grown GaN substrates are already commercially available, although at very high prices.

Concerning the direct deposition of GaN on sapphire, the large lattice misfit [6] promotes the nucleation and growth of individual and isolated islands instead of the growth of a closed film. Böttcher et al. correlated the av- erage diameter of the islands with the residual stress in the layer [7]. In-situ wafer curvature

79

measurements performed at growth tem- peratures revealed the presence of tensile stress in GaN up to hundreds of MPa in many cases [8], which can result in the formation of cracks in the GaN [9]. The tensile stress was explained by the coalescence of GaN islands at initial growth stages [9]. Due to the dif-ferent thermal expansion of GaN and Al2O3 [10], the low tensile residual stress in GaN turns to be compressive upon cooling from the deposition temperature to room tem- perature. For GaN layers with a thickness between 1 and 2 µm and grown by MOVPE at 1050°C and cooled down to room tem- perature, Hearne et al. [8] and Böttcher et al. [7] reported the contribution of the thermally induced stress of about -(660 ± 100) MPa and -(710 ± 100) MPa, respectively. In thick GaN layers, the residual stress can be compensated by structural defects. A lot of studies reveal that polar GaN heteroepitaxial layers grown by HVPE possess a huge number of threading dislocations (TDs) [6, 11].

To overcome the problems related to the large lattice misfit, e.g. formation of incoherent nu-clei as well as threading dislocations, multiple step HVPE processes consisting of the growth of a low-temperature (LT) nucleation layer on sapphire followed by HVPE overgrowth have been applied (e.g. Ref. 12–17). The nucleation layer acts as a buffer to compensate the lat-tice misfit and provides more or less coherent GaN nucleation sites leading to a rapid coa- lescence of the overgrowing layer. Usually, nu-cleation layers are grown at 450 - 600 °C. In this temperature range, a high nucleation density with a narrow lateral size distribution of the nuclei is typically achieved [12, 13, 16, 17]. The poor crystalline quality of the LT layers is im- proved by an additional high-temperature (HT) treatment at 950 - 1080 °C, which is assumed to result in a complete recrystalliza-tion of the layers, i.e. in the formation of fully hexagonal structures of the deposited nuclei [18]. In a final step, thick GaN layers are ob- tained by HT HVPE overgrowth. Despite of large effort, an established LT nucleation proce-dure to deposit high-quality HVPE layers is not yet available [15]. As reported by Prazmowska

et al. [16], the quality of HT HVPE layers is strongly impacted by small variations of nucleation layer growth conditions such as growth rate and growth time. A typical growth rate of LT nucleation layers is lower than 5 µm/h [12, 17] and is hard to be controlled during the HVPE process. Despite of the tremendous progress in HVPE growth of GaN in the last years, some important aspects of this growth method are not well studied and understood. One of these issues is the role of a kinetically limited growth regime in HVPE process and its impact on the defect for- mation in different stages of the GaN growth.

This paper describes the homoepitaxial HVPE growth on MOVPE templates as well as the heteroepitaxial HVPE growth on sapphire under kinetically limited growth conditions, and their impact on the defect formation and stress evolution in GaN layers. The kineti-cally limited growth mode by HVPE can be achieved in the temperature range between 750 °C and 930 °C without a significant loss of the material quality. This temperature range can be understood as intermediate between the temperature ranges of the LT nucleation and HT GaN growth mentioned above. Re-garding studies of the GaN growth by HVPE in this temperature range to our knowledge, no reports can be found in literature. From a technological point of view, this effort may result in a novel approach (i) to a closed pro-cess for depositing high-quality GaN layers on sapphire which can be more suitable for the HVPE growth process as the LT nucle-ation as well as (ii) to in-situ defect and stress engineering during the HVPE deposition. Hereinafter, the homoepitaxial deposition and the formation of nucleation layers are studied in dependence on the process temperature and duration of growth. The surface morphol-ogy, crystal quality, and residual stress of the layers are investigated in detail. The results on HVPE overgrowth of nucleation layers formed under intermediate temperatures are presented. Furthermore, we describe the in-terplay between the layer thickness, the den-sity of threading dislocations and the residual stress in grown GaN layers. The residual stress

80

(1)

was determined at room temperature by means of high-resolution X-ray diffraction (HRXRD) and micro-Raman spectroscopy. The density of threading dislocations was obtained from the XRD line broadening that was measured via reciprocal space mapping using HRXRD.

Experimental HVPE sample preparation The GaN layers were grown in a commercial, vertical HVPE reactor from Aixtron using the conventional HVPE process with elemental gallium, hydrogen chloride (HCl), and am-monia (NH3) as precursors. All experiments were performed at 900 mbar. The homoep-itaxial growth was performed in nitrogen atmosphere on (0001) oriented, 2 inch com-mercial MOVPE templates. The kinetically limited growth conditions were achieved using reduced substrate temperatures as well as high V/III ratios. The heteroepitaxial GaN layers were grown on exactly (0001) orient-ed, 2 inch sapphire substrates. Nitrogen or a N2/H2 mixture with a ratio of nearly 1:1 was used as carrier gas during the nucleation step. Prior to nucleation, the substrates were hea-ted up to 1050 °C and annealed for 5 min- utes under carrier gas atmosphere. Nucleation layers with a thickness up to 2 µm were depos-ited at different temperatures between 765 °C and 900 °C under a constant HCl flux as well as under a constant V/III ratio of 20. Mean growth rates during the nucleation calcu- lated from the GaN mass gain were higher than 200 µm/h. The overgrowth of nucleation layers was performed at 1040 - 1050 °C using N2/H2 mixture as carrier gas. Sample characterization The surface morphology of the GaN layers was studied by scanning electron microscopy (SEM). The high-resolution X-ray diffraction (HRXRD) measurements were carried out at a triple-axis diffractometer (Seifert/FPM) with an Eulerian cradle which was equipped

with a sealed X-ray tube with copper anode and two perfect, i. e. dislocation-free, (111) oriented Si crystals. The first Si crystal was used as a monochromator in the pri- mary beam, the second one as an analyzer of the diffracted beam. The cross section of the pri- mary X-ray beam was reduced by a set of slits to 0.09 × 2 mm2. The instrumental line broaden-ing of the diffractometer was below 10 arcsec. The penetration depth of the X-rays was estimated as 5 µm. The density of threading dislocations (TDs) was determined from the reciprocal space maps (RSMs) that were recorded in copla-nar diffraction geometry on the symmetri-cal reflection 0004 and on the asymmetrical reflections 1014 and 1015. During the recip- rocal space mapping, a set of radial (2θ/ω) scans was measured for different values of ω. The quantity 2θ denotes the detector angle, ω the angle between the primary beam and the sample surface. As a result of the reciprocal space mapping, a distribution of the meas-ured intensity in the angular (ω,2θ) space was obtained for each diffraction line that was converted into the (qx,qz) representation of the reciprocal space using the transformation

where λ = 0.154056 nm refers to the wave-length of the X-ray beam. In addition to RSMs, radial (2θ/ω) and azimuthal (ω) scans were performed through the intensity maxima of the symmetrical diffractions 0002, 0004, and 0006 as well as asymmetrical diffractions 1014, 1015, and 1124 or 2024 for each sample. These radial and azimuthal scans were used in order to obtain exact line positions in the qx and qz coordinates (see Eq. (1)) and inter-planar spacings that are needed for the residual stress calculation (see below).

The threading dislocations were visualized on the cross section of the samples by means of transmission electron microscopy (TEM).

81

During the cross section preparation, a thin lamella was cut out of the samples and dimpled in a precision ion polishing system. The TEM experiments were performed on a JEM2200FS from JEOL that was equipped with an illumi-nation system corrected for spherical aberra-tion (CS correction), an ultra-high resolution objective lens (CS = 0.5 mm), and an in-col-umn energy filter. The acceleration voltage was 220 kV. As detector, a 2k × 2k CCD camera from Gatan was used.

The Raman measurements were performed at room temperature in backscattering ge-ometry using a Labram HR 800 Horiba Jobin Yvon (Villeneuve d’Asq, France) spectrome-ter with a thermoelectrically cooled charge- coupled device (CCD) detector. The spectral calibration was realized by employing a mer-cury vapor lamp. Raman scattering was ex- cited with the 532 nm (2.33 eV) line of a frequency-doubled Nd:YAG laser. By pass-ing the laser beam through a 100x Olympus microscope objective, the linearly polarized laser light was focused on the surface of the GaN specimens. This provided a lateral resolution of 1 µm. The scattered light was collected by the same objective and contains both the z(yx)z and z(yy)z configurations with the z directions oriented parallel to the c axis of the samples which enabled the detec-tion of the GaN E2 and A1(LO) modes [19]. The spectral resolution of the Raman mea-surements was better than 1 cm−1. Since GaN is transparent in the visible spectral range, it is possible to obtain depth dependent in-formation by using the confocal technique [20, 21]. Moving the focal plane within the GaN layer allows monitoring the spectral position of the observable Raman modes as function of the distance from the GaN surface. The diam-eter of the confocal hole was adjusted in order to have a depth resolution of about 2.5 µm.

Calculation of the threading dislocation density and residual stress

Several X-ray based methods appropriate for the determination of the threading dis-location density have been developed in the past [22–24]. Recently, a Monte Carlo method established by Holy et al. [25] for simulation of the intensity distribution in the reciprocal space was applied to determine the TD den-sity in a series of polar GaN layers grown by MOVPE on sapphire substrates [26]. Later on, this technique was successfully applied to more complicated AlGaN two-layer systems [27]. The approach from Barchuk et al. [26] is based on fitting of RSMs simulated by the Mon-te Carlo method [25] to the measured RSMs. The free parameters of the model used for the Monte Carlo simulation are the densities of the edge and screw TDs with given Burgers vectors. During the refinement of the free pa-rameters, first the density of screw TDs (ρs) with the Burgers vector [ 22] is determined from the symmetrical RSM measured on the reflection 0004, which is not affected by the edge dislocations with the Burgers vector [22]. Sub-sequently, the asymmetrical RSMs measured on the reflections 1014 and 1015 are employed to determine the density of edge TDs (ρe). During the fitting of the asymmetrical RSMs, the density of screw TDs obtained from the symmetrical reflection is taken into account but not refined. All calculated RSMs were convoluted with the resolution function which takes into account the azimuthal divergence of the primary beam and the angular acceptance of the detector [28] as well as an additional line broadening caused by the sample bowing. The crystal truncation rod (CTR) and the effect of the surface stress relaxation on the RSMs were neglected in the calculations in order to reduce the computation time [29].

82

(2)

(3)

For calculation of the residual stress in HVPE GaN from the XRD data, a modification of the sin2 ψ method [30, 31] was employed that was derived for hexagonal single-crystalline layers (with the point group 6mm) assuming that the residual stress acts perpendicular to the <0001> direction, i.e. in the plane of the GaN layers. For this orientation of the resid-ual stress, a straightforward application of the Hookes law reveals that the residual stress (σ) causes elastic lattice deformation (εψ ), which is a linear function of sin2 ψ,

In Eq. (2), ψ refers to the angle between the diffracting lattice planes and the sample surface, i.e. the lattice planes (0001); S11 = 3.086 TPa−1, S12 = -0.996 TPa−1 and S13 = -0.557 TPa−1 are single-crys-talline elastic constants of GaN as tak-en from Polian et al. [32]. The elastic lattice deformations were determined for three symmetrical (0002, 0004, and 0006) and three asymmetrical diffraction lines (1014, 1015, and 1124 or 2024) using

where dψ = λ/(2 sin θ) denotes the interpla-nar spacing measured in the ψ direction and

refers to the intrinsic (stress-free) interplanar spacing calculated using the lattice parameters a = 0.31895 nm and c = 0.51861 nm [33].

(4)

According to Davydov et al. [37], the intrinsic position of the E2(high) Raman mode in un-strained bulk GaN is , the linear stress coefficient assuming biaxial stress in the c plane is The precise spectral position of the E2(high) mode, ,was obtained within an error of ±0.03 cm−1 by fitting the measured data by a Lorentzian function. As mentioned above, the use of the confocal technique in conjunction with Raman spectroscopy allowed the resid-ual stress to be determined in different depths under the sample surface. However, in order to be able to compare residual stress values obtained from XRD and Raman spectros- copy, the residual stress calculated from the Raman shift was averaged over the uppermost 5 µm, which corresponds approximately to the penetration depth of X-rays.

.

As a complementary method, confocal mi-cro-Raman spectroscopy has been used to determine the residual stress. This technique allows the residual stress to be measured in different depths below the surface of the GaN samples. Thus, it can detect possible depth gra-dients of the residual stress. The residual stress measurement using the Raman spectroscopy is based on the analysis of the peak position shift of observable Raman modes, which is directly proportional to the lattice strain. For elastic deformation, the change in the wave-number is consequently directly proportional to the residual stress [34–36]. Although three optical phonon modes, i.e. E2(low), E2(high), and A1(LO), are allowed in GaN c plane back-scattering according to the selection rules [19], we selected the non-polar E2(high) phonon for our study, because it is sensitive to the strain in the basal plane (c plane) and thus directly to the in-plane residual stress in the (0001) ori-ented GaN layers:

83

Results and Discussion

Homoepitaxial HVPE growth under ki-netically limited growth conditions

The deposition of high-quality GaN layers typ-ically requires transport limited process condi-tions with a sufficiently fast surface kinetics, which is determined mainly by the substrate temperature. Moreover, the surface diffusivi-ty is also affected by other process parameters such as growth rate, V/III ratio or carrier gas composition. Its reduction can be realized in different ways. For instance, in case of nitro-gen-rich growth conditions (high V/III ra-tios) and nitrogen atmosphere the surface dif-fusivity is decreased owing to the enhanced

growth rate. In the present work the influence of reduced surface kinetics on the surface morphology and on the defect formation in HVPE GaN layers was studied. For this pur-pose, MOVPE templates were used for the homoepitaxial HVPE growth of GaN using process conditions promoting the limitations of the surface kinetics. The experimental pa-rameters as well as the results of rocking curve measurements of the investigated samples are summarized in Table I.

Table I: Growth conditions and full width at half maximum (FWHM) of 0002 and 1015 rocking curves for GaN layers grown homoepitaxially on MOVPE templates.

Figure 1a shows the surface morphology of sample S1 which was grown at a temperature of 1050 °C, but in comparison to the standard process at a higher V/III ratio and with an enhanced growth rate. The sample reveals a smooth, mirrored surface without inverse pyr-amids. However, a noticeable step bunching which is well recognizable in the SEM image in-dicates a reduced surface diffusivity during the growth. Further slowdown of the surface ki-netics by decreasing the growth temperature results in an enhanced formation of V-pits on the layer surface, as it is clearly seen in Fig. 1b of sample S2. In comparison to S2, sample S3 grown at 880 °C (Fig. 1c) is more than twice thinner, but its surface is completely built of inverse pyramids with much smaller lateraldimensions. The full width at half maximum (FWHM) of the rocking curves (0002) and

(1015) is similar to that of used MOVPE tem-plates for samples S1 and S3, but slightly lower in case of sample S2. The excellent FWHM values of all samples reveal that the applied kineti-cally limited growth conditions do not lead to degradation of the crystal quality. Moreover, the decreased FWHMs of sample S2 indicate the reduction of the threading dislocation density. Similarly, the impact of V-pits formed during coalescence of ELO islands at 1030 °C on GaAs was observed by Motoki [38]. Taking stress relations into account, this effect can be used as in-situ method for the additional reduction of the threading dislocation density during HVPE growth.

Sample Thickness(µm)

Temperature(C)

V/IIIratio Growth rate(µm/h)

FWHM (0002)(arcsec)

FWHM (1015)(arcsec)

S1 50 1050 40 150 238 182

S2 12 930 320 60 196 158

S3 5 880 640 65 241 155

84

Figure 1: SEM images of sample (a) S1, (b) S2, and (c) S3 grown on MOVPE templates.

Furthermore, we consider some issues of the V-pit structure. Figure 2a shows the plane view of a typical V-pit in HVPE growth which consists of a six facetted 1101 surfaces with a V-shape angle of about 56°. On the contrary, considering sample S2 in more detail (Fig. 2b) it is recognizable that all V-pits appear nearly round. A cross section SEM image of sample S2 in Fig. 2c supplies the V-shape an-gle value of about 90°. According to Du et al. [39] a rounding of the facetted corners occurs due to the fast growth of corners between, e.g. two slow 1101 facets at kinetically limited growth conditions. Thus, varying the growth conditions, from the form of V-pits conclu-sions about the dominating growth regime can be drawn. Apart from a lot of V-pits, the surface of sample S2 reveals some small 3D is-lands which are marked by red arrows in Fig. 2b. These islands indicate a sufficiently re-duced surface diffusivity as well.

Growth of nucleation layers by intermedi-ate temperatures

Similar growth conditions ensuring a reduced surface diffusivity have been applied for the direct deposition of GaN on sapphire. Table II shows growth parameters of the experiments which resulted in the deposition of non-closed nucleation layers A-D and some identical closed layers grown under same growth condi-tions and denoted as nucleation layer E. Fur-ther details of these experiments are described in the article by Lukin et al. [40]. SEM images of two non-closed nucleation layers A and B grown at 900 °C and 780 °C under nitrogen atmosphere and otherwise identical conditions are shown in Figs. 3a and 3b. Owing to the short growth time, the sapphire substrate is only partly covered by deposited GaN islands. Due to the large lattice misfit between sapphire and GaN, the 3D growth dominates.

Figure 2: SEM images of inverse pyramids (V-pits) on the surface of HVPE layers. (a) Typical, hexagonal pit usually appearing on the surface of HVPE layers. (b) Round shaped V-pits on the surface of S2. Red arrows mark 3D small islands. (c) Cross section of a V-pit on S2 reveals the V-shape angle of 90°.

85

Table II: Process parameters and FWHM of the 0002 rocking curve of the deposited nucleation layers.

Two in-plane crystal directions in GaN [1100] and [1120] are shown in Fig. 3a as deter-mined from the flat direction of the sapphire substrate. Mainly hexagonally shaped GaN is-lands with edges parallel to the <1120> crystal direction in GaN are observed, indicating a high degree of crystallinity. XRD measure-ments confirmed the predominant hexagonal lattice structure with the (0001) orientation of the islands. However, a large FWHM of the rocking curve (0002), which was equal to 2180 arcsec and 1580 arcsec as measured for the nucleation layers A and B, respectively, indicated a high mutual tilting and/or mosaicity of the crystallites.

The density and size distribution of the GaN islands were found to depend strongly on the growth temperature as well. At 900 °C relatively extended islands with varying size and mutual distance are observed (Fig. 3a). When growing at 780 °C (Fig. 3b), the islands are smaller, whereas the nucleation density is higher and the lateral and size distributions are more uniform. Regarding the different widths of the rocking curves discussed above, this re-sult supports the hypothesis that the mosaicity increases with increasing lateral size of the is-lands and that the width of the rocking curves is mainly controlled by the mosaicity of indi-vidual islands. The high nucleation density and

Figure 3: Surface morphology of nucleation layers (SEM images) grown for the same duration: (a) at 900 °C using N2 carrier gas, (b) at 780 °C using N2 carrier gas, (c) at 780°C using N2/H2 mixture, (d) at 780 °C using N2/H2 mixture and increased V/III ratio.

Nucleation layer Carrier gas Temperature (C) V/IIIratio FWHM (0002)(arcsec)

A N2 900 20 2180

B N2 780 20 1580C N2/H2 780 30D N2/H2 780 60

E N2 780 20 1134

86

the high grade of the island size homogeneity of nucleation layer B favor the coalescence of islands and hence, the formation of a closed GaN layer. Decreasing the growth temperature to about 765 °C, in turn, leads to a further de-crease of the island size and to an increase of the defect density by coalescence (mainly thread-ing dislocations) [7] in subsequent stages of growth. The addition of hydrogen to the car-rier gas at similar growth parameters results in a drastical reduction of the nuclei density (see surface morphology of sample C in Fig. 3c). A sufficiently high nucleation density can be partially restored by increasing the V/III ratio as in the case of nucleation layer D (Fig. 3d). Therefore, further growth experiments have been performed at temperatures around 780 °C using nitrogen as carrier gas.

Figure 4: SEM images of a non-closed nucleation layer B show the surface morphology of the type 1 and type 2 islands as well as the surface evolution at the very first stage of lateral growth (dashed circle in (b)) of the second layer. The dashed arrow in (a) indicates a side facet of a second layer island. The solid arrow in (b) displays the inverse pyramids (V-pits) appearing during the coalescence of second layer nuclei.

In Fig. 4 the surface morphology of a non-closed nucleation layer B is presented in more detail. The structure of GaN islands and their evolution to closed layers at temperatures around 780 °C reveal some interesting features. The results and the developed structure model have been published recently by Lukin et al. [40]. According to this structure model the island shapes can be attributed to two differ-ent types (Fig. 4a). Except for the predominant type 2, some islands of type 1 are observed. Grown at 780 °C, type 2 islands possess a com-plex polyhedron form, hexagonal in projection on the substrate plane. For simplicity of the following discussion we denote this island part as the first layer. The top [0001] surface of such first layer islands might act as a preferred nucleation site for the subsequent deposition.

The dashed arrow in Fig. 4a indeed indicates, e.g. a separated hexagonal structure with smooth 1101 side facets on the top sur-face of the first layer islands. We denote these structures as the second layer islands. Several of these structures (marked by dashed circle in Fig. 4b) seem to be already coalesced forming a first fragment of a closed second layer. Obvi-ously, the second layer can rapidly grow in the lateral directions in contrast to the previous 3D growth modus during direct deposition of GaN on sapphire. Thus, the surface pattern shown in Fig. 4 is assumed to represent a very early stage of closed GaN layer formation by lateral growth.

The coalescence of second layer nuclei is asso-ciated with the appearance of distinct, facetted pits having the form of an inverse pyramid. Such kind of defects are already observed at the very beginning of lateral growth as indicated by the solid arrows in Fig. 4b, whereas the first layer islands reveal no pits at all on coalescing (see Fig. 4c). On the basis of results reported by Liliental-Weber et al. [41], second layer pit formation is supposed to be initiated by threading dislocations (or bundles of threading dislocations) occurring in the grain boundaries of underlying clus-ters of nuclei. This is also suggested by the ar-rangement of pits detected in subsequent layer growth (see Fig. 6 below).

87

Figure 5 shows a cross section of two coalesced islands of the non-closed nucleation layer in a bright-field TEM image. The dark contrast between the islands indicated by the dashed arrow shows the common grain boundary. The horizontal patterns visible in the TEM micro-graph of the individual islands stems from the diffraction contrast on local strain fields and can be explained by a series of laterally extend-ed stacking faults. The streaks along the [0001] direction and twin spots clearly seen in the

Figure 5: Bright-field cross section TEM image of a non-closed nucle-ation layer B grown at 780 °C showing the microstructure of two coalesced islands. The direction of the prima-ry electron beam was [1120]. The dashed arrow indicates the grain boundary. Islands are primarily free of threading dislocations, but con-tain a lot of stacking faults. The corresponding diffraction pattern reveals the reflex splitting and the streaks due to basal stacking faults.

corresponding diffraction pattern confirm the presence of basal stacking faults [42]. Hence, initial islands on sapphire obviously contain a high density of stacking faults, whereas they appear to be free of threading dislocations. On the contrary, dislocations are concentrated in the grain boundaries between the islands. In addition to the threading dislocations, a signifi-cant density of partial dislocations terminating the stacking faults in the basal planes of GaN [43] is assumed to exist in the grain boundary.

Rapid lateral coalescence of the second layer was confirmed by experiments with a longer growth time. As an example, Fig. 6a shows the sur-face morphology of nucleation layer E when the growth time at 780 °C was three times lon-ger as compared to sample B. The FWHM of the (0002) rocking curve was 1134 arcsec indi-cating a reduced mosaicity in comparison with the early stages of nucleation layer growth (see Fig. 3b). An almost closed GaN layer with a thickness of about 0.8 µm has formed with the only obvious indication of non-complete-ly buried type 2 island being marked by the dashed arrow in Fig. 6a. The enhanced lateral growth of the second layer seems to be analog to the rapid coalescence of GaN islands ob-served on a AlN buffer layer by MOVPE [44].

In the case of the nucleation at 780 °C, the role of the buffer layer plays, obviously, the first layer. On the basis of the non-closed nucleation layer structure discussed above, we propose a

model (see Fig. 6b) which can explain the for-mation of the surface morphology shown in Fig. 6a. The islands of the second layer grow mainly in lateral directions forming pits by co-alescence. In contrast to the individual facetted pits, already described in the context of Fig. 4 above, a lot of pits with mainly continuously curved sides (not facetted) are found to be the dominating surface defects. Often, pits are linearly arranged as indicated by solid arrows in Fig. 6a. These rows of pits probably partly represent the grain boundaries of underlying type 2 islands, because the boundaries consist of threading dislocations, which, in turn, lead to the formation of pits as already mentioned. Assuming that a pit is caused by just one threading dislocation, a minimal threading dislocation density of approx. 1.5×10

9 cm−2 can

be estimated for the nucleation layer shown in Fig. 6a.

88

Figure 6: Surface morphology of closed nucleation layers E (SEM images) at 780 °C: Solid arrows in (a) show rows of pits, which probably represent grain boundaries between underlying nuclei. The dashed arrow indicates a hexagonally shaped nucleation island that is non-completely buried during growth. (b) Schematic model representing the structure of closed nucleation layer.

The micro-Raman spectroscopy has been used to determine the residual stress of the nucle-ation layers, which is an important criterion for evaluation of the relaxation behavior. The SEM image in Fig. 7a illustrates the spot size of the focused laser beam in relation to the surface morphology of the investigated closed nucleation layer (cf. Fig. 6a). In Fig. 7b, the volume-averaged Raman shift of the E2(high) mode at several neighboring positions on the surface of the closed nucleation lay-er is plotted. The zero stress frequency is marked by the horizontal dashed line.

Figure 7: (a) Surface morphology of the closed nucleation layer E deposited at 780 °C. The SEM image illustrates the spot size of the focused laser beam in relation to the surface morphology of the sample. (b) Raman shift of the E2(high) mode from different positions of the sample surface. Raman shift of unstrained GaN is marked by the dashed horizontal line.

It was found that the layer is nearly unstrained on average with a mean E2(high) frequency of (567.63 ± 0.12) cm−1. The significant variations of the frequency of up to 0.37 cm−1 represent a maximum residual stress of about 137 MPa. It is important to note that Raman investigations of non-closed and closed nucleation layers give similar results independent of growth time and layer thickness of up to 2 µm as well as growth temperatures in the range of 765 - 800 °C. The FWHM of the E2(high) Raman mode in the range of 3.7 - 4.7 cm−1 indicates a low crystal quality and/or a high defect density.

89

To explain the results obtained from the closed nucleation layers, we refer to the strain-stress relations in MOVPE-grown templates sys-tematically studied, e.g. by Hearne et al. [8]. Comparing stress values of about -660 MPa for closed, 1 - 3 µm thick MOVPE grown lay-ers with the samples under consideration here, the nearly zero residual stress of the nucle-ation layers would correspond to an unrealistic high level of tensile stress at growth tempera-tures. The residual stress relaxation is assumed to arise from the lateral growth of islands.

Stresses can be strongly reduced as direct con-nection to the substrate is limited to the center of the domains. As a consequence, threading dislocations are virtually only present near/at domain boundaries. The significant stress inho-mogeneity reflects this. Further investigation is required to understand potential relaxation mechanisms which reduce the thermal stress on cooling the samples down from growth temperatures including the impact of disloca-tions, mosaicity, and surface roughness.

Nucleation layer overgrowth

The results on the HT overgrowth of our HVPE nucleation layers are presented in Table III. The non-closed nucleation layer B and almost closed layers E, respectively, have been used as tem-plates for the overgrowth by HVPE, which was carried out at 1040 - 1050 °C resulting in closed GaN layers with a thickness of up to

Table III: Results of overgrowth experiments using nucleation layers as templates. Sample S4 is marked by an asterisk because it was grown at a three-time lower V/III ratio than the other samples.

15 µm. Thereby, sample S4 was grown at three-time lower V/III ratio than the other samples. The overgrowth of nucleation layer D grown using hydrogen/nitrogen mixture as carrier gas provided similar results as for nucleation layer B, and will be not discussed here.

As can be seen from the SEM image (Fig. 8a), the deposition on the non-closed nucleation layer gives a rough surface with a typical thickness variation of 1 - 2 µm. Furthermore, a high defect density dominated by small pits of up to 1 µm in lateral size and some small holes left after HT coalescence at the GaN/sapphire interface are found. The FWHM of the (0002) and (1015) rocking curve was de-tected to be 770 arcsec and 575 arcsec, re-spectively. In contrast to sample S7 grown on the non-closed nucleation layer, samples S5 - S6 grown on closed nucleation layers appears

smooth and mirroring, e.g. Fig. 8b shows a cross section of sample S5. Besides, they re-veal a very good crystal quality (see FWHM values in Table III), which is even comparable with typical values of HVPE layers grown on MOVPE templates (e.g. Ref. 14 and 45). Conse-quently, the full coalescence of the second layer and the formation of the pit-like surface mor-phology (Fig. 6) by nucleation at about 780°C provide the fast smoothing of the GaN layers by HT overgrowth (Fig. 8b), and results in high crystal quality.

Sample Nucleation layer Thickness(µm)

FWHM (0002)(arcsec)

FWHM (1015)(arcsec)

S4∗ E 13 247 195

S5 E 9 257 142S6 E 15 276 152

S7 B 8 770 575

90

Figure 8: SEM images of overgrown GaN layers. Image (a) shows the layer produced by HT HVPE overgrowth of the non-closed nucleation layer B. Image (b) shows the cross section of sample S5.

Threading dislocation density

Further, a comprehensive study of threading dislocations density and residual stress was performed on grown samples. According to the nucleation procedure, the studied sam-ples can be divided in three groups. Samples S1 - S3 were grown homoepitaxially on 2 inch (0001) oriented MOVPE GaN templates. The second sample series comprising samples S4 - S6 was deposited on closed HVPE GaN tem-plates. The last sample (S7) was deposited at similar deposition conditions like S5 and S6,

Table IV: Characteristics of the samples under study: template type, average sample thickness (t), screw TD density (ρs), edge TD density (ρe), total TD density (ρtot), residual stress determined from XRD data (σXRD) and Raman measurements (σRaman). The densities of screw and edge TDs from Monte Carlo simulation were determined with an error of ±15%. The values of residual stress from X-ray and micro-Raman experiments possess an error of approx. ±50 MPa. The asterisks mark those samples, which were grown at a lower V/III ratio.

but the non-closed HVPE GaN nucleation layer grown at 780 °C was employed directly as template for the further growth. Addition-ally, a commercial, nearly unstressed HVPE layer with a thickness of 900 µm (S0) grown on a MOVPE template was used for comparison. This thick layer was deposited at low V/III ra-tio as well as sample S4. The results are sum-marized in Table IV. Additional results and details can be found in Barchuk et al. [46].

91

Sample Templatet

(µm)ρs

(10−8 cm−2)ρe

(10−8 cm−2)ρtot

(10−8 cm−2)σXRD

(MPa)

σRaman

(MPa)S0∗ MOVPE 900 0.13 0.38 0.5 -55.1 -44.4

S1 MOVPE 50 0.28 1.04 1.3 -190.6 -274.1S2 MOVPE 12 0.09 0.84 0.93 -504.0 -222.0S3 MOVPE 5 0.77 1.9 2.7 -566.4 -496.3

S4∗ E 13 0.68 3.3 4.0 -231.2 -203.7S5 E 9 0.90 8.9 9.8 -496.6 -444.4S6 E 15 0.86 5.4 6.3 -366.5 -348.1S7 B 8 6.7 17.6 24.3 -658.9 -507.5

For all samples under study, the intensities of the diffuse scattering in the vicinity of the dif-fraction maxima, which were measured us- ing the ω scans and/or extracted from the qx scans, can be described by qx

−n function with n being between 2 and 3. One example of the I vs. qx scans, which was measured on the reflec-tions 0004 and 1014 in sample S5 and displayed in the double-logarithmic representation, is shown in Fig. 9a. Such dependence indicates the presence of mixed (screw and edge) dislo-cations [24, 25]. The refinement of the density of screw and edge TDs using the Monte Carlo method described above revealed the dislo-cation densities summarized in Tab. IV. The

Figure 9: Reciprocal-space maps simulations of the reflections 0004 and 1014 for S5. (a) Comparison of experimental (dots) and simulated (solid lines) qx cuts in the reciprocal space. (b) Experimental (solid blue lines) and simulated (dot-ted red lines) reciprocal-space maps. The step of intensities is 100.5 in a logarithmic scale.

quality of the refinements is illustrated in Fig. 9b, where the RSMs measured on the symmet-rical 0004 and the asymmetrical 1014 diffrac-tions as well as the corresponding simulated RSMs are shown. An excellent agreement be-tween the measured and simulated RSMs was achieved for the qx cuts in the reciprocal space (cf. also Fig. 9a) that are directly influenced by the TDs density. Some disagreements in the intensity levels between experimental and sim-ulated data (mainly in the qz direction) stem possibly from the neglected strain relaxation towards the sample surface and from the cor-relation of the dislocation positions that was not regarded in our calculations.

Depending on the template type and on the GaN layer thickness, the TD densities range between 107 and 2×109 cm−2 (cf. Tab. IV). In general, the TD densities are lower for MOVPE GaN templates (samples S0 - S3) than for HVPE GaN templates (samples S4 - S7). Furthermore, the TD densities decrease with increasing thickness of the GaN layers (Fig. 10). The dislocation density in sample S2 was strongly affected by the V-pits formation. It is first of all noticeable for the density of screw dislocations. According to Motoki et al. [38], V-pits provide annihilation of TDs due to their interaction with the facets of V-pits. Both, screw and edge threading dislocation density of sample S2 indicate that the annihilation of

screw TDs proceeds on the V-pits facets faster than that of edge TDs. The reduction of the TD density in thicker GaN layers can be ex-plained by dislocation bunching [47, 48]. These studies showed that the dislocation bunching takes place typically in the first ten microme-ter above the GaN/template interface. Since the penetration depth of X-rays is about 5 µm, only the near-surface region of the thick GaN layers is probed by XRD. In this region, the TDs are already bunched. As XRD quantifies the TD density from the extent of the local strain fields caused by TDs, the decrease of the TD density with increasing GaN layer thickness and thus with advancing TD bunching can be interpret-ed as a gradual decay of the mean strain field

92

around bunched TDs. The reasons for this phenomenon can be (i) the interaction of TDs within the TD bunches and the consequent interference of their strain fields, (ii) too large atomic displacements within the TD bunches

Our TEM experiments confirmed the presence of bunched edge TDs in the samples under study (not shown here).

The determined TD densities (Fig. 10) reveal that the density of edge dislocations is larger than the density of screw dislocations. This effect was already reported for other GaN lay-ers grown by different processes [4, 11, 38]. However, the character of the threading dis-locations, which is given by the ratio between edge and screw TDs, depends significantly on the template type. For the GaN layers grown on MOVPE templates (samples S0, S1, S3) ρe/ρs approaches 3. In the GaN layers grown on closed HVPE GaN templates (samples S4 - S6), mainly the density of edge TDs increases (ρe/ρs ≈ 10) raising the total TD density in these GaN layers. The dependence ρs t−0.32

where t refers to the thickness of the GaN lay-er was determined for screw TDs in samples S0, S1, and S3. The density of screw TDs in samples S4 - S6 approaches this relation which is plotted by the solid line in Fig. 10. The higher density of edge TDs observed in GaN

Figure 10: Densities of screw (squares) and edge (circles) thread-ing dislocations as well as the total threading dislocation density (tri-angles) plotted as function of the GaN layer thickness.

that produce widespread diffuse scattering, which is not fully recognized in the HRXRD experiment, and (iii) the annihilation of TDs at the vertical boundaries of the mosaic GaN blocks [47, 48] or at the bottom of the pits [38].

layers grown on closed HVPE templates is related to a higher mosaicity of these layers. The boundaries of the mosaic blocks contain a large number of edge TDs that are responsi-ble for a mutual disorientation of the adjacent blocks. The highest density of edge and screw TDs was found in the GaN layer grown on the non-closed HVPE GaN nucleation layer (sample S7). Due to the high density of screw TDs, the ratio of the TD densities was com-parable with the ρe/ρs ratio in the GaN layers grown on the MOVPE templates (ρe/ρs ≈ 3). Such unusually high density of screw TDs is obviously related to the non-closed nature of the nucleation layer B that was used as a template in this particular case. Probably, the island coalescence at high temperatures results in the enhanced formation of screw threading dislocations. The screw dislocations probably compensate the lattice misfit between the ac-tual HVPE GaN layer and the uncovered sap-phire substrate as well as the larger surface roughness of the HVPE GaN nucleation layer, which is similar to the phenomena observed by Kawamura et al. [47] in staircase structures.

93

In contrast to screw TDs, the density of edge TDs measured in S7 is comparable with the densities of edge TDs in samples S4 - S6.

Residual stress

The confocal micro-Raman spectroscopy and the XRD measurements revealed that all in-vestigated GaN layers are under compressive residual stress (see Table IV). The comparison of residual stresses in individual samples shows that the compressive residual stress decreases with increasing GaN layer thickness. As com-pared to the general trend, the compressive residual stress is smaller in samples S0 and S4, which were deposited at the lowest V/III ratio, and slightly higher in sample S7, which was deposited on a non-closed HVPE GaN nucleation layer. The low stress value for the sample S2 obtained by micro-Raman spectros-copy must be treated carefully. Obviously, large V-pits on the surface lead to the surface relax-ation and to a very low stress values measured by micro-Raman spectroscopy. In contrast, the HRXRD intensity was collected from the much larger area and provides a stress value comparable to other samples grown at higher V/III ratios.

The residual stresses for other samples obtained using micro-Raman spectroscopy and HRXRD are in a very good agreement, i.e. within the error bars in most samples. The largest discrep-ancies were observed for GaN layers with a steep depth gradient of the residual stress (as recognized by confocal micro-Raman spectros-copy), although the micro-Raman data were averaged over the penetration depth of X-rays as described above. Whereas sample S5 grown on the closed HVPE GaN template possessed no stress gradient, the residual stress in sam-ple S7, which was grown on the non-closed HVPE GaN nucleation layer, decreased from -590 MPa at the GaN/template interface to -475 MPa at the surface of the GaN layer (see Fig. 7 in Ref. 46). For such a gradient of the residual stress, the assumptions of Eq. (2) are not valid any more. Furthermore, a steep depth gradient of the residual stress mimes sheer stress components [30], which are not considered in Eq. (2).

Correlation between threading dislocation density and residual stress

The simultaneous decrease of the dislocation density and residual stress with increasing GaN layer thickness, which is moreover de-scribed by similar power function of sample thickness for S0, S1, S3, indicates some cor-relation between the total TDs density and the residual stress in the HVPE GaN layers. This correlation is shown in Fig. 11 and dis-cussed below in more details. For all template types, the compressive residual stress increas-es with increasing total TD density, which was calculated as a sum of the densities of edge and screw TDs. For individual templates, the de-pendence of the residual stress on the total TD density can be described by power functions, where the power n is characteristic for each template. For the MOVPE GaN template, n = 1.43, for the closed HVPE GaN template, n = 0.80, and for the non-closed HVPE GaN nucleation layer, n ≈ 0.69.

This correlation between the residual stress and the total TD density leads to the hy-pothesis that the compressive residual stress is caused by uncompensated strain fields of the dislocations. However, the uncompensated strain fields cannot be generated by non-in-teracting dislocations, as they cause strain fields with zero mean atomic displacement (see, e.g., Ref. 49 and/or 50) and thus with no macroscopic change of the interplanar spac-ing, which is recognized as zero residual stress.On the contrary, the strain fields of interact-ing dislocations overlap, which can break the symmetry of the strain fields, especially if the dislocations are not randomly distrib- uted. Consequently, the mean value of lattice deformation is non-zero, which leads to a shift of the diffraction maxima from their intrinsic positions that is recognized as residual stress. The effect of the broken symmetry of the strain fields of TDs on the X-ray diffuse scattering was shown by Holy et al. [25], who simulat-ed the simultaneous broadening and shift of diffraction lines caused by threading disloca-tions that were non-uniformly distributed in a limited diffracting volume. A special case

94

Figure 11: The dependence of the residual stress on the total density of threading dislocations. The resid-ual stresses were determined using the sin2 ψ method (black squares) and micro-Raman spectroscopy (red circles).

of the inhomogeneous distribution of TDs is the dislocation bunching, which leads to the formation of local lateral gradients of the TD density within individual crystallites [48, 51, 52]. Such gradients of TD density would be responsible for the formation of compressive residual stresses in HVPE GaN layers. However, the dislocation bunching must be accompanied by the annihilation of the dislocations during the HVPE GaN layer growth, as both the TD density and the resid-ual stress decreases with increasing GaN lay-er thickness. The annihilation of TDs at the vertical boundaries of the mosaic GaN blocks was reported by Datta et al. [48] and Kawamura et al. [47], the annihilation of TDs at the bottom of the pits by Motoki et al. [38]. If the residual stresses result from the uncompensated strain fields of bunched TDs, it can be expected that the kind of the dislocation bunching strongly influences the dependence of the residual stress on the TD density. From Fig. 11, it can be concluded that the kind of the TD bunching is characteristic for each template. In the HVPE GaN layers grown on MOVPE GaN templates (samples S0 - S3), the compressive residual stress increases steeply with increasing TD density, although the TD density is relatively low in these GaN layers. This effect can be explained by an in-tense dislocation bunching. The HVPE GaN layers grown on closed HVPE GaN templates

(S4 - S6) possess similar compressive residual stress like samples S0 - S3, despite a higher TD density. Some part of TDs remains probably “unbunched”, i.e. more or less randomly distributed over the crystallites in samples S4 - S6.

Conclusion

The impact of the kinetically controlled HVPE growth on the defect formation in GaN layers has been studied. Application of the kinetically limited growth during the ho-moepitaxy of GaN induce the generation of inverse pyramids (V-pits), which promote the reduction of the threading dislocations densi-ty in grown layers. The density of screw thread-ing dislocations will be decreased considerably more than that of edge threading dislocations. The initial stages of HVPE growth on sapphire at intermediate temperatures in the range of 750 - 900 °C are accompanied by the formation of a lot of stacking faults in nucleation islands, and results in the rapid formation of a closed uniform structured GaN layer based on a form of epitaxial lateral overgrowth, which can be used as a nucleation/buffer layer for further high-temperature HVPE growth of smooth, high quality GaN layers. The homoepitaxial-ly and heteroepitaxially grown samples were investigated by using high-resolution X-ray diffraction and micro-Raman spectroscopy.

95

Acknowledgement

The authors gratefully acknowledge the fruitful cooperation with Freiberger Compound Ma- terials GmbH. This work was performed within the Cluster of Excellence “Structure Design of Novel High-Performance Materials via Atomic Design and Defect Engineering (ADDE)” which is financially supported by the European Union (European regional de-velopment fund) and by the Ministry of Sci-ence and Art of Saxony (SMWK).

In all samples, the density of edge threading dislocations was higher than the density of screw dislocations. The residual stresses in the GaN layers were always compressive; their amount increased with increasing thread-ing dislocation density. The correlation between the residual stress and the dislocation density was explained by the formation of uncompensated strain fields around bunched threading dislocations, and was strongly af- fected by the nucleation procedure. For the same layer thickness, the lowest density of edge and screw threading dislocations was achieved in the HVPE GaN layers grown on MOVPE GaN templates. The dislocation density in the GaN layers grown on HVPE GaN templates was 2-3 times higher. Furthermore, the characteristic of the template affected the dislocation bunching and thus the depen-dence of the residual stress on the threading dislocation density. The interplay between the dislocation bunching and the residual stress formation restrained the compressive re- sidual stresses to be below 600 MPa in most GaN layers under study.

96

1. S. Nakamura and G. Fasol, The Blue Laser Diode (Springer, New York, 1997).2. Y. Oshima, T. Eri, H. Shibata, M. Sunakawa, and A. Usui, Phys. Status Solidi (a) 194, 554

(2002).3. K. Fujito, S. Kubo, H. Nagaoka, T. Mochizuki, H. Namita, and S. Nagao, J. Cryst.

Growth 311, 3011 (2009).4. H. Ashraf, R. Kudrawiec, J. L. Weyher, J. Serafinczuk, J. Misiewicz, and P. R. Hageman,

J. Cryst. Growth 312, 2398 (2010).5. J. Napierala, D. Martin, N. Grandjean, and M. Ilegems, J. Cryst. Growth 289, 445 (2006).6. L. Liu and J. H. Edgar, Mat. Sci. Eng. R 37, 61 (2002).7. T. Bottcher, S. Einfeldt, S. Figge, R. Chierchia, H. Heinke, D. Hommel, and J. S. Speck,

Appl. Phys. Lett. 78, 1976 (2001).8. S. Hearne, E. Chason, J. Han, J. A. Floro, J. Figiel, J. Hunter, H. Amano, and I. S. T.

Tsong, Appl. Phys. Lett. 74, 356 (1999).9. E. V. Etzkorn and D. R. Clarke, J. Appl. Phys. 89, 1025 (2001).10. M. Leszczynski, T. Suski, H. Teisseyre, P. Perlin, I. Grzegory, J. Jun, S. Porowski, and T. D.

Moustakas, J. Appl. Phys. 76, 4909 (1994).11. J. Q. Liu, J. F. Wang, Y. F. Liu, K. Huang, X. J. Hu, Y. M. Zhang, Y. Xu, K. Xu, and H.

Yang, J. Cryst. Growth 311, 3080 (2009).12. P. R. Tavernier, E. V. Etzkorn, Y. Wang, and D. R. Clarke, Appl. Phys. Lett. 77, 1804

(2000).13. E. Richter, S. Gramlich, A. Klein, E. Nebauer, I. Rechenberg, U. Zeimer, and M. Weyers,

Phys. Status Solidi (a) 188, 439 (2001).14. R. Korbutowicz, J. Koz-lowski, E. Dumiszewska, and J. Serafinczuk, Cryst. Res. Technol. 40,

503 (2005).15. D. Martin, J. Napierala, M. Ilegems, R. Butt e, and N. Grandjean, Appl. Phys. Lett. 88,

241914 (2006).16. J. Prazmowska, R. Korbutowicz, R. Paszkiewicz, A. Szyszka, J. Serafinczuk, A. Pod-

horodecki, J. Misiewicz, and M. T-lacza-la, Vacuum 82, 988 (2008).17. T. Bohnen, A. E. F. de Jong, W. J. P. van Enckevort, J. L. Weyher, G. W. G. van Dreumel,

H. Ashraf, P. R. Hageman, and E. Vlieg, J. Cryst. Growth 311, 4685 (2009).18. F. Meng, I. Han, H. McFelea, E. Lindow, R. Bertram, C. Werkhoven, C. Arena, and S.

Mahajan, J. Cryst. Growth 327, 13 (2011).19. H. Harima, J. Phys: Condens. Matter 14, R967 (2002).20. V. V. Strelchuk, V. P. Bryksa, K. A. Avramenko, M. Y. Valakh, A. E. Belyaev, Y. I. Mazur,

M. E. Ware, E. A. DeCuir Jr, and G. J. Salamo, Phys. Status Solidi (c) 8, 2188 (2011).21. C. Roder, F. Lipski, F. Habel, G. Leibiger, M. Abendroth, C. Himcinschi, and J. Kor-

tus, Journal of Physics D: Applied Physics 46, 285302 (2013).22. T. Metzger, R. Hopler, E. Born, O. Ambacher, M. Stutzmann, R. Stommer, M. Schuster,

H. Gobel, S. Christiansen, M. Albrecht, and H. P. Strunk, Phil. Mag. A 77, 1013 (1998).23. V. Ratnikov, R. Kyutt, T. Shubina, T. Paskova, E. Valcheva, and B. Monemar, J. Appl. Phys.

88, 6252 (200).24. V. M. Kaganer, O. Brandt, A. Trampert, and K. H. Ploog, Phys. Rev. B 72, 045423

(2005).25. V. Holy, T. Baumbach, D. Lubbert, L. Helfen, M. Ellyan, P. Mikulık, S. Keller, S. P.

DenBaars, and J. Speck, Phys. Rev. B 77, 094102 (2008).26. M. Barchuk, V. Holy, B. Miljevi c, B. Krause, T. Baumbach, J. Hertkorn, and F. Scholz, J.

Appl. Phys. 108, 043521 (2010).

References

97

27. S. Lazarev, M. Barchuk, S. Bauer, K. Forghani, V. Holy, F. Scholz, and T. Baumbach, J. Appl. Cryst. 46, 120 (2013).

28. M. Barchuk, V. Holy, B. Miljevic, B. Krause, and T. Baumbach, Appl. Phys. Lett. 98, 021912 (2011).

29. S. J. Shaibani and P. M. Hazzledine, Phil. Mag. A 44, 657 (1981).30. I. C. Noyan and J. B. Cohen, Residual Stress - Measurement by Diffraction and Interpreta-

tion (Springer- Verlag, New York, Berlin, Heidelberg, 1987).31. K. Kusaka, T. Hanabusa, and K. Tominaga, Thin Solid Films 281-282, 340 (1996).32. A. Polian, M. Grimsditch, and I. Grzegory, J. Appl. Phys. 79, 3343 (1996).33. V. Darakchieva, B. Monemar, and A. Usui, Appl. Phys. Lett. 91, 031911 (2007).34. C. Kisielowski, J. Kruger, S. Ruvimov, T. Suski, J. W. Ager, E. Jones, Z. Liliental-Weber,

M. Rubin, E. R. Weber, M. D. Bremser, and R. F. Davis, Phys. Rev. B 54, 17745 (1996).35. J. M. Wagner and F. Bechstedt, Phys. Rev. B 66, 115202 (2002).36. R. Korbutowicz, M. T-lacza-la, J. Kovac, G. Irmer, and R. Srnanek, Cryst. Res. Technol. 43,

1339 (2008).37. V. Y. Davydov, Y. E. Kitaev, I. N. Goncharuk, A. N. Smirnov, J. Graul, O. Semchinova,

D. Uffmann, M. B. Smirnov, A. P. Mirgorodsky, and R. A. Evarestov, Phys. Rev. B 58, 12899 (1998).

38. K. Motoki, T. Okahisa, S. Nakahata, N. Matsumoto, H. Kimura, H. Kasai, K. Takemoto, K. Uematsu, M. Ueno, Y. Kumagai, A. Koukitu, and H. Seki, Mat. Sci. Eng. B 93, 123 (2002).

39. D. Du, D. J. Srolovitz, M. E. Coltrin, and C. C. Mitchell, Phys. Rev. Lett. 95, 155503 (2005).

40. G. Lukin, C. Roder, E. Niederschlag, Y. Shashev, U. Muhle, O. Patzold, J. Kortus, D. Rafaja, and M. Stelter, Cryst. Res. Technol. 47, 121 (2012).

41. Z. Liliental-Weber, Y. Chen, S. Ruvimov, and J. Washburn, Phys. Rev. Lett. 79, 2835 (1997).

42. J. Narayan, P. Pant, A. Chugh, H. Choi, and J. C. C. Fan, J. Appl. Phys. 99, 054313 (2006).43. V. Potin, P. Ruterana, and G. Nouet, J. Phys.: Condens. Matter 12, 10301 (2000).44. I. Akasaki, H. Amano, Y. Koide, K. Hiramatsu, and N. Sawaki, J. Cryst. Growth 98, 209

(1989).45. H. Ashraf, J. L. Weyher, G. W. G. van Dreumel, A. Gzregorzyck, and P. R. Hageman, J.

Cryst. Growth 310, 3957 (2008).46. M. Barchuk, C. Roder, Y. Shashev, G. Lukin, M. Motylenko, J. Kortus, O. Patzold, and

D. Rafaja, J. Cryst. Growth 386, 1 (2014).47. F. Kawamura, M. Tanpo, N. Miyoshi, M. Imade, M. Yoshimura, Y. Mori, Y. Kitaoka, and

T. Sasaki, J. Cryst. Growth 311, 3019 (2009).48. R. Datta, M. J. Kappers, M. E. Vickers, J. S. Barnard, and C. J. Humphreys, Superlat-

tices and Microstructures 36, 393 (2004).49. M. A. Krivoglaz, X-Ray and Neutron Diffraction in Nonideal Crystals (Springer, Berlin,

1996)50. T. Ungar, Mat. Sci. Eng. A 309–310, 14 (2001).51. P. Rudolph, C. Frank-Rotsch, U. Juda, M. Naumann, and M. Neubert, J. Cryst.

Growth 265, 331 (2004).52. A. Sakai, H. Sunakawa, A. Kimura, and A. Usui, Appl. Phys. Lett. 76, 442 (2000).

98

Strontium titanate Breaking the symmetryHartmut Stöcker, Juliane Hanzig, Florian Hanzig, Matthias Zschornak, Erik Mehner, Sven Jachalke, Dirk C. Meyer

TU Bergakademie Freiberg, Institute of Experimental Physics, Leipziger Str. 23, 09599 Freiberg

100

AbstractPure strontium titanate exhibits a cubic perovskite-type structure at room temperature. Many approaches to break this high degree of symmetry are accessible. The present review summari-zes the possible methods by discussing the effects of ion implantation, temperature treatment and external electric fields. Since oxygen vacancies are the most prominent and most mobile defect species in strontium titanate, they play a major role also for structural changes due to external influences. Here, an overview on the involved microscopic models will be given.

tetragonal at a temperature of 105 K accom-panied by a change of the space group from cubic Pm3m to tetragonal I4/mcm (see Fig. 2). The reorganization of the unit cell is referred to as an anti-ferrodistortive phase transition. With a further decrease in temperature, po-lar soft modes in the far infrared have been found [13]. Cooling down below a transi-tion temperature of 40 K, thermodynamic calculations [14] suggest the initiation of the phase transition to a ferroelectric phase, which is, however, not completed even at 0 K. Oxygen vacancies, stoichiometry changes and temperature are only some of the influences that may lead to breaking of the high symme-try in the cubic perovskite-type structure of SrTiO3. Further approaches and the connect-ed physical phenomena will be discussed in the present review.

Fig. 1: Quasi-binary equilibrium phase diagram of the system SrO–TiO2, after [7].

IntroductionStrontium titanate SrTiO3 is a model materi-al for perovskite-type transition-metal oxides providing applications ranging from high-k dielectrics [1] and sensors [2] to resistive switching random access memories (RRAM) [3, 4] and nanoscale batteries [5]. In partic-ular, resistive switching in transition-metal oxides is based on the change of electric re-sistance over several orders of magnitude, which is closely related to local changes of the oxygen stoichiometry. Therefore, migration of oxygen vacancies in thermal and electric fields determines device performance [6].

The equilibium phase diagram of SrO–TiO2 (see Fig. 1) [7] indicates possible secondary phases [8, 9] that may be formed by crystal or layer growth and during device operation due to ion migration which leads to local stoichiometry changes. For ambient pressure and temperatures up to the melting point, at a composition of 50 mol.% of SrO and TiO2 each, the perovskite SrTiO3 is formed. For the analysis of diffusion-induced stoichiometry changes on the structure, we also have to con-sider the ternary phase diagram of the system Sr–Ti–O [10]. According to theoretical calcu-lations [11, 12], possible structural reactions to oxygen deficiency are vacancy ordering in planes as in brownmillerite or clustering of oxygen vacancies. In all oxygen-deficient cases, mixed valence states of titanium exist. For the centrosymmetric equilibrium struc-ture of SrTiO3 at ambient conditions, nei-ther ferroelectricity, pyroelectricity, piezo-electricity, nor flexoelectricity are allowed. Stoichiometric strontium titanate undergoes a structural phase transition from cubic to

101

Fig. 2: Structural models of the SrTiO3 structures at temperatures above 105 K (left) and below (right). The tetragonal low-temperature phase (right) is shown along the c direction of the unit cell. The rotation of oxygen octahedra against each other is highly exaggerated.

Sr Ti O

Preparation of strontium titanate and Ruddlesden-Popper phases

Since SrTiO3 has a relatively high melting temperature of 2040 °C (see Fig. 1), single crystals are mostly grown using the Verneuil method (flame fusion) that does not need a crucible. In this method, the powdered start-ing materials melt in a hydrogen flame and drop on the growing single crystal with a di-ameter of approx. 5 cm. Subsequent annealing steps are needed to produce transparent crys-tals [15, 16]. The growth of strontium titanate layers on a substrate is possible using virtually any physical or chemical deposition method.

Under Ti-deficient conditions during syn-thesis, additional SrO planes are introduced as ordered SrO-OSr stacking faults, which occur as Ruddlesden-Popper (RP) phases SrO(SrTiO3)n with a body-centered tetrago-nal unit cell in space group I4/mmm [8] (see Fig. 3). These defect structures are stable up to decomposition temperatures above 1600 °C for RP phases with n = 1 and 2 (see Fig. 1), although growth of single crystals from the melt is prohibited by peritectic decomposi-tion. Further, they exhibit a strong anisotropy of the dielectric response due to the ordered array of SrO (001) excess planes [17–20].

Experimental studies of such ordered stack-ing faults by high-resolution transmission electron microscopy and X-ray diffraction have been reported [9, 21–27].

Besides ceramic bulk materials [21, 28–32] or small anisotropic crystals [33] the availability of RP phases in the form of thin films is of particular importance for technical applica-tions. In this context we have reported on the oriented growth of SrO(SrTiO3)n thin films by chemical solution deposition [9, 27]. The ap-plied approach is based on a modified Pechini route, which has been proven to be a feasible low-temperature method for the prepara-tion of RP phases [34–36]. Thin films of RP phases have also been prepared following so-phisticated layer-by-layer deposition growth techniques, e.g. molecular beam epitaxy or pulsed-laser deposition [23, 24, 37, 38]. As promising candidates for thermoelectric energy conversion materials, also rare-earth doped, electrically conducting ceramics and thin films of SrO(SrTiO3)n have been reported [39–41].

102

Fig. 3: Structural models of the homologous series of Ruddlesden-Popper phases SrO(SrTiO3)n with the members SrO (n = 0), Sr2TiO4 (n = 1), Sr3Ti2O7 (n = 2), Sr4Ti3O10 (n = 3) and SrTiO3 (n = ∞) from left to right.

Structural stability as well as electronic, mi-croscopic, and elastic properties of SrTiO3- and SrO-based layered compounds have been theoretically studied within the last de-cade [42–45]. Various atomistic simulations, Hartree-Fock and density functional theory (DFT) calculations are available for RP phases with n ≤ 3, but the stability of the phases has been discussed controversially so far, with respect to both the tendency within the ho-mologous series and their absolute formation energies [22, 46, 47]. Especially for the forma-tion of the RP phase with n = 3, no clear trend could be obtained from previous calculations.For an extended stability discussion accord-ing to the formation reaction

the bulk SrO(SrTiO3)n RP homologous series (n = 0–5, ∞) has been studied by DFT with the ABINIT code [48]. To assess the influence of the error of the exchange-correlation func-

tional both the generalized-gradient approx-imation (GGA) with the Perdew-Burke-Ern-zerhof functional [49] and the local density approximation (LDA) in Teter-Pade param-etrization [50] have been used for compari-son in our calculations. Extended norm con-serving Teter potentials [51] were employed, which allow for an explicit treatment of semi-core and valence states. All structures have been fully relaxed with respect to cell geom-etry and atomic positions.

The resulting energies of formation of the ho-mologous RP series are shown in Fig. 4 as a function of the SrO/SrTiO3 ratio. The choice of LDA or GGA exchange-correlation func-tional and the simulation cell size (body-cen-tered tetragonal or primitive cell) have no effect. A gain of formation energy up to the phase n = 3 is found, then a saturation thresh-old of 215 meV is reached, which remains constant for phases of higher order.

Sr

Ti

O

SrO + n SrTiO3 SrO(SrTiO3)n

103

Fig. 4: Calculated energies for the formation of RP phases. The integration of the SrO stacking fault into SrTiO3 shows exothermic character for all phases and saturates at n = 3. Choice of LDA or GGA exchange-correlation functional and unit cell sampling (tetragonal or primitive cell) do not influence the results.

These results imply a consecutive driving force going from phases of lower to higher n up to the RP phase with n = 3. They confirm the trends obtained in calculations by Le Bacq et al. [47] and explain the experimental ob-servations that the RP phases with n = 2 and n = 3 occur preferentially at higher heating rates [9, 27].

Because of the plateau for RP phases with n > 3, the maximum range of interaction be-tween two neighboring stacking faults can be estimated to about 12 Å. An endothermic formation of the RP phase n = 3 [46] or an alternating formation energy within the ho-mologous series [22] could not be confirmed.

104

Fig. 5: After N+ implantation a distinct increase of the spectral B peak is observed in the X-ray absorption spectrum at the Ti-K edge. Furthermore, a dependence on the angle of X-ray incidence ω relative to the surface is observed, which is related to the thickness of the distorted layer.

Ion implantation in strontium titanate

Ion implantation is a straightforward way to tune the real structure of any crystalline ma-terial. It can be used to introduce dopants or vacancies far from thermal equilibrium and for amorphization of the material.

Pre-edge maxima within X-ray absorption near-edge spectra (XANES) have proven to be a useful indicator for local symmetry distortions, especially when an atom is dis-placed from a high symmetry position, e.g. in PbTiO3 [52], for phase transitions in BaTiO3 [53, 54], in SrTiO3 thin films [55], and in studies of these and related perovskites under pressure [56, 57]. These maxima even reveal the origin of hybridization and orbital chang-es on the absorbing atomic species, as has been shown for rutile [58], SrTiO3 [59] or 3d transition metal compounds in general [60].

Several modification pathways for strontium titanate SrTiO3 single crystals have been test-ed with respect to changes of the short-range ordering of the perovskite-type structure, among them donor and acceptor doping during crystal growth, vacuum annealing, and ion implantation using several species. Only for nitrogen ion implantation using N+ ions of 40 keV kinetic energy with total flu-ences between 5 × 1016 and 2 × 1017 cm−2, an altered Ti-K absorption edge fine structure was observed (see Fig. 5) [61]. In the exper-imental XANES spectra the second pre-edge feature, typically labelled peak B [56, 62, 63], depends on the angle of incidence and, hence, provides clear evidence of a distorted surface layer.

105

106

Oxygen vacancies Oxygen vacancies are the most common de-fects in SrTiO3 with the lowest formation en-thalpy and a higher diffusivity and mobility than oxygen [64]. An oxygen exchange with the surrounding atmosphere occurs if stron-tium titanate is stored at elevated tempera-tures. Reducing conditions at temperatures above 750 °C cause the introduction of elec-trons and, accordingly, lead to n-type conduc-tivity. The reaction, using Kröger-Vink nota-tion [65], can be expressed as given in Tab. 1, reaction (A). The process is reversed under oxidizing atmosphere (see Tab. 1, B).

Description Reaction

Vacancy creation / reduction (A)

Vacancy annihilation / oxidation (B)

Formation of RP phases (C)

Formation of SrO (D)

Ionisation of oxygen vacancies (E)

Ionisation of strontium vacancies (F)

Electron-hole equilibrium (G)

Schottky-Wagner equilibrium (H)

Donor on strontium position (I)

Donor on titanium position (J)

Acceptor on titanium position (K)

The distorted surface layer created by N+ ion implantation is characterized by a shift of the Ti-K edge from the Ti4+ state of SrTiO3 towards Ti3+ in the layer phase. Simulating XANES spectra of undisturbed SrTiO3 clus-ters and structures with local Ti displacement shows that the measured spectrum of the dis-torted layer corresponds best with a Ti atom displaced in [001] direction [61]. The strong increase of the additional resonance in the peak B region in the distorted phase with an accompanied shift in energy gives evidence of a remarkable static displacement of the Ti atom relative to the surrounding oxygen octa-hedron of at least 0.3 Å. This can be explained by the introduction of oxygen vacancies, which are most effectively created by nitrogen ions with similar mass as oxygen, and the re-lated creation of distorted titanium coordina-tion octahedra in the modified surface layer.

Tab. 1: Possible defect reactions in SrTiO3

107

The oxygen ion diffusion coefficient reach-es a value of 1.0 × 10−10 cm2/s at 900 °C [66]. Migration of oxygen vacancies also causes an enhanced cation and electron migration since charge neutrality must be maintained [67]. Oxygen migration takes place via a vacancy mechanism. Therefore, the diffusion coeffi-cient of oxygen in perovskite-type compounds depends on three parameters: formation, mi-gration and association of oxygen vacancies [68]. By electrocoloration experiments, the vacancy migration can be visualized [6]. Since the migration of oxygen vacancies results in local changes [69] also of the cation stoichi-ometry, secondary phases should be expect-ed specifically near surfaces and interfaces to electrodes, and the stoichiometry changes will be most significant in these regions. For SrTiO3 either the formation of RP phases (see Tab. 1, C) or the precipitation of SrO (see Tab. 1, D) can be expected.

Since SrTiO3 is a mixed conductor, modeling of the conductivity includes the interaction between electronic and ionic charge carriers (see Tab. 1, E–H). As for every semiconduc-tor, doping increases the number of electrons or holes (see Tab. 1, I–K). From the knowledge of mass action constants and the assumption of charge neutrality, the concentrations of all species can be calculated. Including the charge carrier mobilities then leads to the conductivity of the material.

The introduction of additional charge car-riers into SrTiO3 by temperature treatment is possible above approx. 800 °C. Annealing under vacuum of approx. 10-6 mbar leads to the introduction of free electrons (see Tab. 1, A). These can be detected optically by Fourier transform infrared spectroscopy as a reduc-tion of sample transmission. Modeling of the spectra using a Drude term yields the electron concentrations (see Fig. 6). Further details can be found in Ref. 70.

Fig. 6: Infrared transmission spectra of SrTiO3 single crystals annealed for different times in vacuum at 900 °C. Symbols indicate measured points and lines show fits using a Drude term with given electron concentrations.

108

External electric fields

Oxygen vacancies are the most import-ant defects in transition metal oxides and are able to move in an external electric field due to their charge. From the defect chemistry point of view the question aris-es how these mobile defects interact with the external electric field and how defect migration in transition metal oxides influ-ences the stability of the crystal structure.

If an external electric field of 106 V/m is ap-plied to a SrTiO3 single crystal coated on both sides with Ti electrodes over a time of 12 h, one can measure a typical formation curve as shown in Fig. 7. The electric current through

the bulk first increases to a maximum and then decreases to the initial value. One can as-sume that a field-driven redistribution of ox-ygen ions and oxygen vacancies, respectively, takes place with oxygen vacancies V being attracted to the cathode (negative pole) and oxygen ions moving to the anode (positive pole). At the point of maximum current as many V as possible are travelling to the cath-ode. When more and more vacancies arrive at the cathode, the current decreases again. This redistribution of oxygen vacancies caus-es a concentration gradient through the bulk crystal.

o

Fig. 7: Time-dependent current I(t) through a SrTiO3 single crystal during formation. A selection of eight forma-tion cycles (F1–F8) is shown, in which the polarity of the applied voltage varies between +100 V and −100 V with relaxation times of 12 h. The current was measured in darkness.

o

109

The redistribution process is connected to structural changes observable at the anode. To investigate these changes, the 002 SrTiO3 reflection is recorded in situ during electro-formation. Fig. 8 shows X-ray diffraction scans of the initial state (orange line), record-ed during 12 h of electroformation (black lines) and of the status at the end of formation (green line). At the anode, where an oxygen excess is expected, the 002 reflection is shift-ed and broadened towards smaller diffraction angles. The main peak corresponds to un-changed bulk SrTiO3, whereas the shoulder is caused by crystal volumes with enlarged lattice constants d. This indicates a stretching of the unit cell in [001] direction, i.e. in the di-rection of the applied electric field. Increasing the applied electric field yields a voltage-de-pendent elongation of the lattice constant, illustrated by a shoulder spanning to even smaller diffraction angles. While the evolu-tion of the shoulder takes several hours, the disappearance happens on a time scale of sec-onds, which means that at least two different processes are involved. By means of Raman spectroscopy the polar character of the anode region was proven. Because SrTiO3 exhibits no first order phonon spectrum, a weakening of selection rules during formation suggests a breaking of the centrosymmetry in the orig-inally cubic structure [71]. In contrast, no structural changes are obtained at the cath-ode. This asymmetry is attributed to local stoichiometry changes due to the migration of oxygen vacancies in the external electric field, where deplete at the anode and accumulate at the cathode.

Fig. 8: Comparison of the structural response at anode and cathode of a SrTiO3 single crystal in an external elec-tric field of 1.0 × 106 V/m. The development of a shoulder at the 002 reflection during formation is observed at the anode (top). No changes of intensity, position or broade-ning of the 002 reflection during formation are found at the cathode (bottom).

V o

110

This effect has been first observed by Meyer et al. [72, 73] and described by the intergrowth of secondary phases. Present data show that the observations can be explained by a mere deformation of the SrTiO3 unit cells near the surface [74]. Nevertheless, a coupling of these structural changes to mechanical properties [75] and the Ti valence [69] have been report-ed previously.

To explain the experimental findings, the fol-lowing model is established: The presence of an oxygen vacancy in a SrTiO3 double unit

cell leads to opposite displacements of Ti and O ions (see Fig. 9 a). Consequentially, dipole moments exist in both unit cells next to the vacancy, which are amplified or weakened when an electric field is applied (see Fig. 9 b). Additionally, oxygen vacancies become mo-bile and leave point defect free unit cells with dipole moments, stabilized by the electric field with thermal energy being insufficient to overcome the field energy (see Fig. 9 c, d). Therefore, a point defect free dipole structure is established, which is called migration-in-duced field-stabilized polar (MFP) phase [74].

Fig. 9: Formation of the migration-induced field-stabilized polar phase in SrTiO3 single crystals. (a) Initial state (E = 0): oxygen vacancies initiate atomic displacements. (b) Applying an electric field (E > 0): additional displacements of nearest atoms in the vicinity of an oxygen vacancy break the inversion symmetry. (c) Vacancy migration leaves a point-defect-free elongated unit cell (see orange circle) with remaining atomic displacements. (d) Traces of MFP phase are formed as a result of vacancy migration [74].

111

Ferroelectricity in strontium titanate

Intensive investigations concerning the for-mation of a ferroelectric phase transition in perovskites have been reviewed recently [76]. There are some possibilities to favor the for-mation of a ferroelectric phase in strontium titanate, namely symmetry breaking and low temperatures. A polar phase was found in SrTiO3 ceramics, where grain boundaries [77] are assumed to cause local polarization, which is attributed to a symmetry breaking and an enhancement of point defects compared to the averaged volume. Ferroelectric microre-gions [78] have been observed in strontium titanate single crystals as a function of the impurity concentration. Epitaxial SrTiO3 thin films exhibit strain-induced ferroelectricity. There, tensile strain leads to an in-plane po-larization, whereas an out-of-plane polariza-tion can be achieved by compressive strain [14, 79]. Haeni et al. [80] even observed a ferroelectric transition in strained strontium titanate thin films close to room temperature.

In the previous section, we reported on the so-called MFP phase [74] in strontium ti-tanate caused by defect separation in an ex-ternal electric field. The polar character of this non-centrosymmetric structure can be demonstrated using the Sharp-Garn method [81], which utilizes a sinusoidal temperature excitation to detect pyroelectricity. An elec-troformation with an external electric field of 106 V/m, as previously described, super-imposed by a temperature excitation of 1 K amplitude has been conducted, resulting in a time dependent current characteristic I(t) as shown in Fig. 10. Initially, the current re-sponse of a SrTiO3 single crystal shows the in-phase temperature behavior (see Fig. 10, insets (a) and (b)) typical for semiconduc-tors, where heating results in a current in-crease and cooling leads to a current decrease. Hence, charging and discharging of band gap states, caused by intrinsic defects like oxygen vacancies, take place when temperature os-cillates and, thus, contribute to the thermally stimulated current ITSC.

When more and more oxygen vacancies ar-rive at the cathode, a decrease of the current response I(t) is detectable (see Fig. 10, inset (c)). When defect migration finishes, an out-of-phase temperature behavior is observed, exhibiting a temperature dependence of an inherent polarization of the formed MFP phase [74]. After establishment of the MFP phase in the SrTiO3 single crystal the current response reveals a major phase shift by 180°, referring to metallic conductivity (see Fig. 10, inset (d)), i.e. heating leads to a current decrease and cooling to a current increase. This phenomenon can be ascribed to the ac-cumulation of oxygen vacancies, which form a highly conductive area, known as virtual cathode [6].

With the presented measurement, it is possi-ble to track oxygen vacancies during migra-tion in an external electric field. Analyzing the phase shift between temperature excitation and measured current response I(t) [82], a pyroelectric coefficient of pMFP = 30 µC/m2K [83] is obtained. Since a requirement for py-roelectricity is a crystallographic point group without inversion symmetry but with polar axis, the breaking of centrosymmetry of the originally cubic SrTiO3 structure by the MFP phase is verified.

Resistive switching

The role of surfaces of transition metal oxides and their interfaces to metals comes especially into focus when applications are considered. The functionality with the greatest potential is presently the so-called resistive switching that can be used in non-volatile RRAM mem-ories with high packing densities and writing speeds [3, 4]. This effect manifests itself as a hysteresis in current-voltage measurements and is mainly investigated for thin films but also present in single crystals (see Fig. 11) [84].

112

Fig. 10: Current response I(t) of a SrTiO3 single crystal during electroformation using an external electric field of 106 V/m superimposed by a sinusoidal temperature excitation: (a, b) in-phase current temperature behavior during oxygen vacancy migration, (c) disappearance of temperature induced modulation of cur-rent response, (d) reversed-sign in-phase current temperature behavior when defect separation is finished and the MFP phase is established. Same current scaling for all insets emphasizes the amplitude changes.

Fig. 11: Characteristic current-voltage curves before and after formation of a SrTiO3 single crystal in an electric field of 1.0 × 106 V/m. Inset shows the contacting scheme for voltage application and current mea-surement.

113

At the moment, several pictures of possible switching mechanisms in SrTiO3 exist in the literature [4, 6], the most common ones being:

(1) formation and degradation of conducting filaments [3, 85],(2) phase transition between insulating and conducting state [86–88],(3) alteration of the potential barrier at the interface [89],(4) charge trapping and electron emission from trap states [90, 91].

In the connected micromechanical pictures, oxygen vacancies often play an important role, e.g. as constituents of filaments or trap-ping sites. Structurally these mechanisms rely on a local breaking of the cubic symmetry of the host material.

This can also be related to defects of different dimensionality:

0D point defects, especially oxygen vacancies,1D line defects, in the case of conducting filaments,2D planar defects, when regarding the metal/oxide interface,3D volume defects, if phase transitions occur.

In any of the four models (1–4) given above, the functionality of the transition metal oxi-de material relies on the defects of the ideal structure and a breaking of the high symmet-ry, respectively.

Conclusion

The cubic perovskite structure of SrTiO3 with its high symmetry can be influenced in many ways, which always leads to a breaking of the symmetry. In the present review, several ap-proaches including ion implantation, tem-perature treatment and external electric fields have been summarized. In all cases, the oxy-gen vacancy turns out to be the most import-ant defect species in the material system. The vacancy migration in external electric fields leads to a breaking of cubic symmetry togeth-er with the appearance of pyroelectricity and is the main driving force for resistive switch-ing. Since the perovskite-type structure is the basis for many transition metal oxides, the re-sults obtained for strontium titanate translate easily to other functional oxide materials.

Acknowledgments

This work was performed within the Clus-ter of Excellence “Structural Design of Nov-el High-Performance Materials via Atomic Design and Defect Engineering (ADDE)” which is financially supported by the Euro-pean Union (European regional development fund) and by the Ministry of Science and Art of Saxony (SMWK). Parts of this work were carried out within the BMBF-funded proj-ect CryPhysConcept (03EK3029A), the ESF young researcher group PyroConvert (SAB 100109976) and the HGF-funded Virtual Institute MEMRIOX – “Memory Effects in Resistive Ion-beam Modified Oxides” (VH-VI-422).

114

References1. J. Robertson, J. Vac. Sci. Technol. B 18, 1785 (2000).2. W. Menesklou, H. J. Schreiner, K. H. Härdtl, and E. Ivers-Tiffée, Sens. Actuators B 59, 184 (1999).3. K. Szot, W. Speier, G. Bihlmayer, and R. Waser, Nat. Mater. 5, 312 (2006).4. A. Sawa, Mater. Today 11, 28 (2008).5. J. Hanzig, M. Zschornak, M. Nentwich, F. Hanzig, S. Gemming, T. Leisegang, and D. C. Meyer, J. Power Sources 267, 700 (2014).6. R. Waser, R. Dittmann, G. Staikov, and K. Szot, Adv. Mater. 21, 2632 (2009).7. E. M. Levin, C. R. Robbins, H. F. McMurdie, and M. K. Reser, Phase Diagrams for Ceramists (The American Ceramic Society, Inc. Columbus, Ohio, 1964).8. S. N. Ruddlesden and P. Popper, Acta Crystallogr. 10, 538 (1957), Acta Crystallogr. 11, 54 (1958).9. M. Zschornak, S. Gemming, E. Gutmann, T. Weißbach, H. Stöcker, T. Leisegang, T. Riedl, M. Tränkner, T. Gemming, and D. C. Meyer, Acta Mater. 58, 4650 (2010).10. T. Tanaka, K. Matsunaga, Y. Ikuhara, and T. Yamamoto, Phys. Rev. B 68, 205213 (2003).11. D. D. Cuong, B. Lee, K. M. Choi, H. S. Ahn, S. Han, and J. Lee, Phys. Rev. Lett. 98, 115503 (2007).12. N. Shanthi and D. D. Sarma, Phys. Rev. B 57, 2153 (1998).13. A. S. Barker Jr. and M. Tinkham, Phys. Rev. 125, 1527 (1962).14. Y. L. Li, S. Choudhury, J. H. Haeni, M. D. Biegalski, A. Vasudevarao, A. Sharan, H. Z. Ma, J. Levy, V. Gopalan, S. Trolier-McKinstry, D. G. Schlom, Q. X. Jia, and L. Q. Chen, Phys. Rev. B 73, 184112 (2006).15. J. Bednorz, J. Crystal Growth 41, 5 (1977).16. K. Nassau and A. E. Miller, J. Crystal Growth 91, 373 (1988).17. C. J. Fennie and K. M. Rabe, Phys. Rev. B 68, 9 (2003).18. A. H. Reshak and S. Auluck, I. Kityk, Jpn. J. Appl. Phys. 47, 5516 (2008).19. N. D. Orloff, W. Tian, C. J. Fennie, C. H. Lee, D. Gu, J. Mateu, X. X. Xi, K. M. Rabe, D. G. Schlom, I. Takeuchi, and J. C. Booth, Appl. Phys. Lett. 94, 3 (2009).20. S. Clima, G. Pourtois, N. Menou, M. Popovici, A. Rothschild, B. Kaczer, X. P. Wang, C. Adelmann, M. A. Pawlak, K. Tomida, A. Rothschild, B. Govoreanu, R. Degraeve, M. Schaekers, M. Zahid, A. Delabie, J. Meersschaut, W. Polspoel, S. Clima, G. Pourtois, W. Knaepen, C. Detavernier, V. V. Afanas’ev, T. Blomberg, D. Pierreux, J. Swerts, P. Fischer, J. W. Maes, D. Manger, W. Vandervorst, T. Conard, A. Franquet, P. Favia, H. Bender, B. Brijs, S. van Elshocht, M. Jurczak, J. Van Houdt, and D. J. Wouter, Microelectric Eng. 86, 1936 (2009).21. R. J. D. Tilley, J. Solid State Chem. 21, 293 (1977).22. M. A. McCoy, R. W. Grimes, and W. E. Lee, Phil. Mag. A 75, 833 (1997).23. J. H. Haeni, C. D. Theis, D. G. Schlom, W. Tian, X. Q. Pan, H. Chang, I. Takeuchi, and X.-D. Xiang, Appl. Phys. Lett. 78, 3292 (2001).24. W. Tian, X. Q. Pan, J. H. Haeni, and D. G. Schlom, J. Mater. Res. 16, 2013 (2001).25. T. Hungria, I. MacLaren, H. Fuess, J. Galy, and A. Castro, Mat. Lett. 62, 3095 (2008).26. S. Sturm, M. Shiojiri, and M. Ceh, J. Mater. Res. 24, 2596 (2009).27. E. Gutmann, A. A. Levin, M. Reibold, J. Müller, P. Paufler, and D. C. Meyer, J. Solid State Chem. 179, 1864 (2006).28. G. McCarthy, W. B. White, and R. Roy, J. Am. Ceram. Soc. 52, 463 (1969).29. J. H. Sohn, Y. Inaguma, M. Itoh, and T. Nakamura, Mater. Sci. Eng. B 41, 50 (1996).30. P. L. Wise, I. M. Reaney, W. E. Lee, T. J. Price, D. M. Iddles, and D. S. Cannell, J. Eur. Ceram. Soc. 21, 2629 (2001).

115

31. V. Berbenni, A. Marini, and G. Bruni, J. Alloys Compd. 329, 230 (2001).32. T. Hungria, A. B. Hungria, and A. Castro, J. Solid State Chem. 177, 1559 (2004).33. Y. F. Liu, Y. N. Lu, M. Xu, and L. F. Zhoun, J. Am. Ceram. Soc. 90, 1774 (2007).34. K. Nishio, K. Fukuda, T. Imai, H. Takenouchi, H. Mae, M. Fujimoto, T. Iida, N. Hamada, T. Kineri, and T. Watanabe, Mat. Res. Soc. Sym. Proc. 1044, 205 (2008).35. N. Zhou, G. Chen, H. Z. Xian, and H. J. Zhang, Mat. Res. Bull. 43, 2554 (2008).36. U. Balachandran and N. G. Eror, J. Mater. Sci. 17, 2133 (1982).37. M. Okude, A. Ohtomo, T. Kita, and M. Kawasaki, Appl. Phys. Exp. 1, 081201 (2008).38. Y. Iwazaki, T. Suzuki, S. Sekiguchi, and M. Fujimoto, Jpn. J. Appl. Phys. 38, L1443 (1999), Jpn. J. Appl. Phys. 39, L303 (2000).39. K. H. Lee, Y. F. Wang, S. W. Kim, H. Ohta, and K. Koumoto, Int. J. Appl. Ceram. Technol. 4, 326 (2007).40. K. H. Lee, A. Ishizaki, and S. W. Kim, J. Appl. Phys. 102, 033702 (2007).41. J. F. Wang, K. H. Lee, H. Ohta, and K. Koumoto, Ceram. Int. 34, 849 (2008).42. S. Gemming and G. Seifert, Acta Mater. 54, 4299 (2006).43. J. Albina, M. Mrovec, B. Meyer, and C. Elsässer, Phys. Rev. B 76, 165103 (2007).44. U. Schwingenschlögl and C. Schuster, Chem. Phys. Lett. 467, 354 (2009).45. D. Music and J. M. Schneider, J. Phys. Condens. Matter 20, 055224 (2008).46. C. Noguera, Phil. Mag. Lett. 80, 173 (2000).47. O. Le Bacq, E. Salinas, A. Pisch, C. Bernard, and A. Pasturel, Phil. Mag. 86, 2283 (2006).48. X. Gonze, J. M. Beuken, R. Caracas, F. Detraux, M. Fuchs, G. M. Rignanese, L. Sindic, M. Verstraete, G. Zerah, F. Jollet, M. Torrent, A. Roy, M. Mikami, Ph. Ghosez, J.-Y. Raty, and D.C. Allane, Comput. Mater. Sci. 25, 478 (2002).49. J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996).50. S. Goedecker, M. Teter, and J. Huetter, Phys. Rev. B 54, 1703 (1996).51. M. Teter, Phys. Rev. B 48, 5031 (1993).52. V. Kraizman, A. Novakovich, R. Vedrinskii, and V. Timoshevskii, Phys. B: Cond. Matter 208-209, 35 (1995).53. B. Ravel, E. A. Stern, R. I. Vedrinskii, and V. Kraizman, Ferroelectrics 206, 407 (1998).54. A. I. Frenkel, Y. Feldman, V. Lyahovitskaya, E. Wachtel, and I. Lubomirsky, Phys. Rev. B 71, 024116 (2005).55. J. C. Woicik, E. L. Shirley, C. S. Hellberg, K. E. Andersen, S. Sambasivan, D. A. Fischer, B. D. Chapman, E. A. Stern, P. Ryan, D. L. Ederer, and H. Li, Phys. Rev. B 75, 140103 (2007).56. J. Itié, B. Couzinet, A. Dhaussy, A. Flank, N. Jaouen, P. Lagarde and A. Polian, High Press. Res. 26, 325 (2006).57. J. P. Itié, A. M. Flank, P. Lagarde, S. Ravy, and A. Polian, in: High-Pressure Crystallography, Springer, 2010, p. 51–67.58. B. Poumellec, R. Cortes, G. Tourillon, and J. Berthon, phys. status solidi (b) 164, 319 (1991).59. M. Fujita, H. Nakamatsu, S. Sugihara, J. I. Aihara, and R. Sekine, J. Comput. Chem. Jpn. 3, 21 (2004).60. L. A. Grunes, Phys. Rev. B 27, 2111 (1983).61. H. Stöcker, M. Zschornak, C. Richter, J. Hanzig, F. Hanzig, A. Hinze, K. Potzger, S. Gemming, and D. C. Meyer, Scripta Mat. 86, 1 (2014).62. D. Cabaret, B. Couzinet, A. M. Flank, J. P. Itié, P. Lagarde, and A. Polian, AIP Conf. Proc. 882, 120 (2007).63. R. V. Vedrinskii, V. L. Kraizman, A. A. Novakovich, P. V. Demekhin, and S. V. Urazhdin, J. Phys.: Cond. Matter 10, 9561 (1998).

116

64. P. Pasierb, S. Komornicki, and M. Rekas, J. Phys. Chem. Solids 60, 1835 (1999).65. F. A. Kröger and H. J. Vink, in: Solid State Physics, edited by F. Seitz and D. Turnbull, Vol. 3 (Elsevier, Amsterdam, 1956), p. 307–435.66. A. E. Paladino, J. Am. Ceram. Soc. 48, 476 (1965).67. R. Meyer and R. Waser, J. Eur. Ceram. Soc. 27, 1743 (2001).68. D. Neagu and J. T. S. Irvine, Chem. Mater. 23, 1607 (2011).69. T. Leisegang, H. Stöcker, A. A. Levin, T. Weißbach, M. Zschornak, E. Gutmann, K. Rickers, S. Gemming, and D. C. Meyer, Phys. Rev. Lett. 102, 087601 (2009).70. J. Hanzig, B. Abendroth, F. Hanzig, H. Stöcker, R. Strohmeyer, D. C. Meyer, S. Lindner, M. Grobosch, M. Knupfer, C. Himcinschi, U. Mühle, and F. Munnik, J. Appl. Phys. 110, 064107 (2011).71. W. G. Nilsen and J. G. Skinner, J. Chem. Phys. 48, 2240 (1968).72. D. C. Meyer, A. A. Levin, S. Bayer, A. Gorbunov, W. Pompe, and P. Paufler, Appl. Phys. A 80, 515 (2005).73. D. C. Meyer, A. A. Levin, T. Leisegang, E. Gutmann, P. Paufler, M. Reibold, and W. Pompe, Appl. Phys. A 84, 31 (2006).74. J. Hanzig, M. Zschornak, F. Hanzig, E. Mehner, H. Stöcker, B. Abendroth, C. Röder, A. Talkenberger, G. Schreiber, D. Rafaja, S. Gemming, and D. C. Meyer, Phys. Rev. B 88, 024104 (2013).75. H. Stöcker, M. Zschornak, T. Leisegang, I. Shakhverdova, S. Gemming, and D. C. Meyer, Cryst. Res. Techn. 45, 13 (2010).76. Y. I. Yuzyuk, Phys. Solid State 54, 1026 (2012).77. J. Petzelt, T. Ostapchuk, I. Gregora, I. Rychetský, S. Hoffmann-Eifert, A. V. Pronin, Y. Yuzyuk, B. P. Gorshunov, S. Kamba, V. Bovtun, J. Pokorný, M. Savinov, V. Porokhonskyy, D. Rafaja, P. Vaněk, A. Almeida, M. R. Chaves, A. A. Volkov, M. Dressel, and R. Waser, Phys. Rev. B 64, 184111 (2001).78. H. Uwe, H. Yamaguchi, and T. Sakudo, Ferroelectrics 96, 123 (1989).79. N. A. Pertsev, A. K. Tagantsev, and N. Setter, Phys. Rev. B 61, R825 (2000).80. J. H. Haeni, P. Irvin, W. Chang, R. Uecker, P. Reiche, Y. L. Li, S. Choudhury, W. Tian, M. E. Hawley, B. Craigo, A. K. Tagantsev, X. Q. Pan, S. K. Streiffer, L. Q. Chen, S. W. Kirchoefer, J. Levy, and D. G. Schlom, Nature 430, 758 (2004).81. L. E. Garn and E. J. Sharp, J. Appl. Phys. 53, 8974 (1982).82. I. Lubomirsky and O. Stafsudd, Rev. Scientic Instruments 83, 051101 (2012).83. J. Hanzig, E. Mehner, S. Jachalke, F. Hanzig, M. Zschornak, C. Richter, T. Leisegang, H. Stöcker, and D. C. Meyer, New Journal of Physics 17, 023036 (2015).84. H. Stöcker, M. Zschornak, J. Seibt, F. Hanzig, S. Wintz, B. Abendroth, J. Kortus, and D. C. Meyer, Appl. Phys. A 100, 437 (2010).85. M. Janousch, G. I. Meijer, U. Staub, B. Delley, S. F. Karg, and B. P. Andreasson, Adv. Mater. 19, 2232 (2007).86. A. Beck, J. G. Bednorz, C. Gerber, C. Rossel, and D. Widmer, Appl. Phys. Lett. 77, 139 (2000).87. M. Hamaguchi, K. Aoyama, S. Asanuma, Y. Uesu, and T. Katsufuji, Appl. Phys. Lett. 88, 142508 (2006).88. S. Q. Liu, N. J. Wu, and A. Ignatiev, Appl. Phys. Lett. 76, 2749 (2000).89. A. Baikalov, Y. Q. Wang, B. Shen, B. Lorenz, S. Tsui, Y. Y. Sun, Y. Y. Xue, and C. W. Chu, Appl. Phys. Lett. 83, 957 (2003).90. S. F. Alvarado, F. L. Mattina, and J. G. Bednorz, Appl. Phys. A 89, 85 (2007).91. Y. Watanabe, J. G. Bednorz, A. Bietsch, C. Gerber, D. Widmer, A. Beck, and S. J. Wind, Appl. Phys. Lett. 78, 3738 (2001).

Atomic layer deposition of dielectric thin films in the ternary system TiO2-SrTiO2

B. Abendroth 1, S. Rentrop 1, W. Münchgesang 1, H. Stöcker 1, J. Rensberg 2, C. Ronning 2, S. Gemming 3, D. C. Meyer 1

1 Institute of Experimental Physics, TU Bergakademie Freiberg, Leipziger Str. 23, 09599 Freiberg2 Institute of Solid State Physics, Friedrich Schiller Universität Jena, Helmholtzweg 5, 07743 Jena3 Helmholz-Zentrum Dresden-Rossendorf, Bautzner Landstraße 400, 01314 Dresden

118

AbstractThis work summarizes the atomic layer deposition of TiO2 and SrTiO3-related dielectric thin films from commercially available precursors bis(tri-isopropylcyclopentadienyl)-strontium, Tetrakis-(dimethylamido)titanium and H2O. The layers are analyzed with respect to the forma-tion of crystalline phases, stoichiometry and the resulting optical and electric properties. For the deposition of pure TiO2 the layers are independent of the substrate temperature initially amorphous. Crystallization occurs at a critical layer thickness, which decreases with increas-ing deposition temperature. For the ALD of ternary Sr Ti O oxides the layer composition is controlled by the metal-precursor pulse ratio. By variation of the layer composition, the re-fractive index and optical gap can be adjusted between the values of amorphous TiO2, SrTiO3 and SrOH. Electrical leakage current behavior and resistance switching of the ALD oxides is investigated using TiN/ALD oxide/Au capacitor stacks with 15 nm oxide layer thickness and 100 × 100 µm2 contact area.

Many of these applications are based on thin layers of only a few nanometers thickness on plane substrates or, more likely, on three dimensionally structured substrates. Since atomic layer deposition (ALD) facilitates the deposition of thin layers with monolayer control of the thickness and allows a homo-geneous deposition on 3D substrates, it is a widely used technique for the fabrication of thin oxide layers in microelectronics. It is how-ever, also of growing interesting for the depo-sition of catalytic active materials on porous substrates with large specific surfaces, e.g. in photocatalytic hydrolysis and batteries [14]. All these applications make use of the op-tical and electronic properties of TMOs, which are closely related to the microstruc-ture and composition. For TiO2 as an exam-ple, the non-crystalline state as well as the stable rutile and metastable anatase are of technical relevance. The metastable brook-ite polymorph is observed sometimes as a by-product but is not of technical relevance. Amorphous titania, anatase and rutile are all wide gap semiconductors with a direct gap ranging between 3.2 eV (anatase) and 3.0 eV (rutile). Due to their crystallographic symmetry the crystalline modifications pos-sess uniaxial anisotropic refractive indices.

1 Introduction

Transition metal oxides (TMO) are a class of materials with extremely versatile electronic properties. Due to the covalent character of the metal-oxygen bond these materials are wide gap semiconductors or insulators and hence transparent in the visible and near-in-frared spectral range. Due to the large order number of the transition metal the refractive indices are generally high. Therefore, many transition metal oxide have been traditionally used as gemstone, examples are rutile (TiO2), zirconia (ZrO2) and taussonite (SrTiO3). Modern applications of transition metal ox-ides span a wide range from the field of renew-able energy technologies to microelectronics. TiO2 is one of the most versatile transition metal oxide in the field. It is being utilized in dye sensitized solar cells [1, 2], as active ma-terial for photocatalytic hydrolysis [3, 4] or as dielectric in microelectronic capacitor struc-tures [5, 6]. In biomedical applications TiO2 is applied for self-sterilizing surfaces. Due to its high permittivity, strontium titanate is widely studied as a model system for ternary high k materials in microelectronic devices (e.g. references 7 and 8). Both oxides have been intensively investigated for resistance switch-ing metal-insulator-metal (MIM) capacitor structures for non-volatile data memories. Examples can be found in references 9, 10 and 11 for TiO2-based and references 12 and 13 for SrTiO3-based resistive switching devices.

x y z

119

The structural motif of the ambient crystalline phases is the TiO6 coordination polyhedron. In the anatase structure each TiO6 octahedron is linked by four edges to the neighboring oc-tahedra, resulting in a tetragonal symmetry in the space group I41/amd. In the rutile struc-ture, each TiO6 polyhedron shares two oppo-site edges and two corners with neighboring units forming one-dimensional chains. The chains are connected via the two remaining corners of the octahedra. The rutile struc-ture is also tetragonal with the space group P42/mnm. The amorphous state of TiO2 is as well dominated by the TiO6 octahedral co-ordination, however these are linked irregu-larly by common edges and corners [15, 16]. SrTiO3 is a typical representative of the per-ovskite-type structure with the room tem-perature cubic structure with the space group Pm3m [17]. Again, this structure hosts the Ti cation in octahedral oxygen coordination. The cubic unit cell contains one central TiO6 octahedron which is connected via six sharing corners to the TiO6 units in the neighboring unit cells. The Sr cations fill the large spaces in-between the TiO6 octahedra, occupying the corners of the cubic unit cell. Bulk SrTiO3 has a refractive index of 2.4 and an indirect band gap at 3.25 eV and direct band gap at 3.75 eV [18]. In this paper the ALD of dielectric thin films in the ternary system TiO2-SrTiO3 is re-viewed. The ALD processing and relevant characterization techniques are comprised in section 2. The intended application of these layers is the use as resistance switching mate-rial in MIM capacitor structures for non-vol-atile data storage. As resistance switching utilizes a soft and reversible dielectric break-down of the insulator, it is of particular inter-est to control the factors initiating the electric breakdown of the oxide. Intrinsic electronic properties of the material that depend on the crystalline phase and the composition are the permittivity and the width of the band gap. It is shown how these properties can be con-trolled by the deposition process. In section 3.1.2 the evolution of crystalline phase during the ALD of TiO2 is investigated in depen-

dence on deposition temperature and layer thickness. Further, the layer by layer depo-sition approach can be utilized to produce ternary layers with stoichiometries far off the equilibrium chemical composition of crys-talline oxide materials. In the ternary system Sr-Ti-O the layer composition can be adjust-ed in a wide range between SrO and TiO2 in-cluding the ideal composition of strontium titanate, SrTiO3. Going along with the com-position, also refractive index, band can be tuned between the values of the end-members SrO and TiO2. These results are comprised in section 3.2. Resulting electrical characteristics such as leakage current and resistance switch-ing of MIM structures with binary and terna-ry oxide layers are presented in section 3.3.

2 Experimental

2.1 Atomic layer deposition

ALD of metal oxide thin films is a meth-od derived from common chemical vapour deposition (CVD). Similar to CVD, the oxide is formed from the reaction of a volatile met-al-organic compound and an oxidant. Unlike CVD, the reaction does not take place in the gas phase but is restricted to the surface. Using fast switching pneumatic valves, only small amounts of the vapour phase of a heated pre-cursor are admitted at one time into the reac-tion chamber. The precursor molecules occu-py by chemisorption all available surface sites. Any non-adsorbed precursor is purged by an inert gas. In the second step, the oxidant is ad-mitted into the reaction chamber. The organic groups are now separated from the metal atom by oxidation leaving behind a metal-OH ter-minated surface. After a second purging step the process is repeated again. Arising from the self-limiting growth, the distinguished advantage of ALD as compared to CVD is the very homogeneous deposition on flat as well as on 3D surfaces with large aspect ratios. ALD is predestined for the oxide deposition into porous material or into deep trenches.

120

The layer thickness is, ideally, only controlled by the number of repeated pulse/purge cycles. Hence, the value of growth per cycle (GPC) can be used as a measure of ideal ALD behav-ior of the process. Deviations from the ideal ALD are shown for example in section 3.1.1 Typically the layer growth in one cycle is less than one complete monolayer of metal oxide, since the surface coverage is limited by steric hindrance of the organic groups of the pre-cursor molecules. Consequently growth rates are generally very low as compared to CVD. For this work, all thin films were deposited in a Savannah S100 or Savannah S200 ALD tool (Cambridge Nanotech). The system possesses three precursor lines. Each precursor reservoir and line can be heated up to 150 °C. Precursor pulse times can be adjusted down to 10 ms. N2 was used as carrier and purge gas at a flow rate of 20 sccm resulting in a working pressure of 30 Pa. Silicon wafers with 100 orientation were used as substrates. These substrates were cleaned by standard cleaning procedure [19] to re-move organic and metallic contaminations. For process optimization experiments silicon substrates with a native oxide of approximate-ly 2 nm were used. For metal-insulator-metal (MIM) capacitor structures a bottom elec-trode layer of 50 nm TiN was used. The TiN was produced by reactive dc magnetron sput-tering from a Ti target in Ar-N2 atmosphere at a deposition temperature of 450 °C. Gold layers of 70 nm thickness have been deposited by e-beam evaporation as top contacts. Gold was deposited without a Ti adhesion layer to exclude influences of redox reactions at the electrode interface. Subsequent electrode pads were structured by lift-off lithography. All ALD processes use deionized water as oxidant. For the deposition of TiO2, Tetrakis- (dimethylamido)titanium (TDMAT) was used as Ti precursor. Ternary SrxTiyOz oxides were deposited from bis(tri-isopropylcyclo-pentadienyl)-strontium, Sr(iPr3Cp)2 (Absolut Sr™ by Air Liquide), TDMAT and H2O. For the deposition of ternary compounds the process is divided in subcycles of titanium dioxide and strontium oxide, respectively.

The layer stoichiometry is adjusted by varia-tion of the subcycle ratio. For TiO2 the depo-sition temperature Ts was varied from 150 to 330 °C. For the ALD of SrxTiyOz the proper-ties of both metal-organic precursors must be accounted for. As a result the temperature window of the process is very narrow. For Ts < 300 °C the SrO half cycle reaction is incom-plete leading to significant carbon concentra-tions in the layers [21] and for Ts > 330 °C the TDMAT molecule decomposes in the gas phase leading to a CVD type growth with strongly enhanced growth rates as the self-limiting ALD characteristics is then lost [20] . All data shown here for SrxTiyOz are ob-tained for Ts = 320 °C. Further details of the deposition process are described for TiO2 in ref. 20 and for SrTiO3 in ref. 21 and 22.

2.2 Layer characterization

Film thickness and optical properties of the layers were determined by spectrosco- pic ellipsometry (SE) in the spectral range of 0.8 - 5 eV at 70° and 75° angle of incidence using a Woollam M2000 spectroscopic el-lipsometer. For ellipsometry data analysis, the Fresnel coefficients of the sample are cal-culated for an optical layer model including the substrate, native SiO2 or the TiN bottom electrode and the ALD oxide layer. Each component is described by the correspond-ing optical constants (refractive index n and extinction coefficient k) and layer thickness d. For the silicon and silicon oxide the opti-cal constants were taken from reference 23. Optical constants of the ALD oxides are de-scribed by the Tauc-Lorentz expression for the complex refractive index N = n+i k which applies for amorphous semiconductors in the transparent and interband transition region [24]. A TiN layer of 50 nm thickness is still sufficiently thin to be transparent for visible light and the layer thickness can be evalu-ated by SE. A superposition of a Drude and two Lorentz oscillator contributions was used for the TiN model dielectric function to ac-count for the free electron and 3d electron absorption in the near infrared and visible

121

3 Results

3.1 ALD of TiO2

3.1.1 Optimization of deposition parameters ALD was tested for substrate temperatures Ts = 120-330 °C. The pulse duration times of the TDMAT precursor and H2O were set to 0.15 s and 0.015 s, respectively, for all tem-peratures, which is sufficient to produce a homogeneous deposition over the 100 mm diameter of the reactor chamber. It was found that the inert gas purge time following the H2O pulse is a critical parameter to obtain a linear growth regime, specifically at Ts ≥ 250 °C. Figure 1a) shows the effect of H2O purge time in-creased from 5 s to 15 s at TS = 300 °C on the layer thickness d as measured by SE. Insufficient H2O purge times lead to stagnating deposi-tion rates after approximately 500 ALD cycles or correspondingly, for layer thicknesses of more than 25 nm. Similarly, in reference [25], decreasing deposition rates were observed for TDMAT/water ALD of TiO2 for depo-sition temperatures increasing from 150 °C to 210 °C at constant purge times. For the present ALD process, an increase of the H2O purge time to 15 s resulted in a linear growth

spectral range. The model dielectric function parameters and layer thickness are fitted to reproduce the measured ellipsometric data. Complementary, x-ray reflectivity (XRR) measurements were carried out to confirm the SE film thickness and to get additional in-formation on the layer density. XRR analysis was performed with a Philips X’Pert PW3710 diffractometer using Cu Kα radiation with a primary beam diversion set to 0.25° and a secondary monochromator. Measurements were taken with parallel beam geometry at an angle of incidence of 0° - 4°. The analysis of the XRR data is carried out in analogue to SE data analysis based on calculation of the Fresnel reflection coefficients from a lay-er model. The fit parameters in the model for x-ray reflectivity include SiO2, TiN and TiO2 or SrxTiyOz layer thickness and density, as well as surface and interface roughnesses. For the identification of crystalline phases in the TiO2 layers, grazing incidence X-ray diffraction (GIXRD) under an angle of in-cidence of 1° was carried out on a Bruker D8 Advance diffractometer with Cu K α ra-diation, equipped with a Goebel mirror parallel optics in the primary beam and an equatorial Soller collimator with an opening angle of 0.2° and a solid state point detector. The composition of the main constituents Sr, Ti and O has been quantified independently by two different methods. Firstly, wavelength dispersive x-ray fluorescence spectrosco-py (WDXRF) was measured using a Bruker AXS S8 spectrometer, with an internal cali-bration. The Bruker AXS software ML Quant was used to calculate the stoichiometry of the TiO2 and SrxTiyOz layers. Based on a lay-er model, this software package calculates the reabsorption and secondary fluorescence contributions to the primary fluorescence yield for a layered and hence vertically inho-mogeneous sample. For SrxTiyOz thin films Rutherford backscattering spectroscopy (RBS) was carried out using 900 keV 4He+-ions. Backscattered projectiles were detected at an angle of 170°. To increase the resolution and to avoid any possible ion channelling ef-

fects, all samples were measured off-normal under a tilting angle of 60°. The collected charge amounted to 15 μC. From the integral backscattering yield of each RBS signal cor-responding to O, Ti and Sr, respectively, the areal density NzO,Sr,Ti of each atomic species and thus the layer composition were calcu-lated. Details are described elsewhere [22]. In contrast to WDXRF, the layer stoichiometry can be calculated directly from RBS signals without the need of calibration on standard samples. Results of WDXRF and RBS agree well within the accuracy of both methods. All layers are free of residual carbon and nitro-gen as measured by photoelectron spectros-copy on an Escalab 250Xi system, equipped with an Ar cluster ion source for low energy surface cleaning and depth profile sputtering [22].

122

Fig. 1: TiO2 layer thickness as function of the number of ALD cycles. Panel a) shows the evolution of d at Ts = 300 °C for 5, 10 and 15 s H2O purge time. Panel b) displays the layer growth at Ts of 150, 250, 300 and 320 °C using optimized H2O purge times Reprint with permission of [20].

Table 1: Optimized process parameters for TiO2 ALD from TDMAT and H2O.

Figure 1b demonstrates that the deposition rate varies with Ts. Previously, a decreasing deposition rate for increasing temperatures has been observed in references 26 and 27 for constant H2O purge times. Similarly decreas-

Sustrate TDMAT TDMAT H2O H2Otemperature pulse (s) purge (s) pulse (s) purge (s)

120 °C - 250 °C 0,15 8 0,015 5

250 °C - 300 °C 0,15 8 0,015 15

300 °C - 330 °C 0,15 8 0,015 25

ing growth rates are observed in reference 25, however, in this work also a narrow ALD window between 120 °C and 150 °C has been reported.

at Ts = 300 °C. Hence, the H2O purge times were increased progressively with increasing deposition temperatures up to 25 s at 320 °C (see Table 1) until a constant growth per cy-cle (GPC) was achieved for each temperature. Figure 1b shows the linear layer growth as a function of the number of ALD cycles for deposition temperatures ranging from 150 °C

up to 320 °C. Dashed lines are linear regres-sions to the data, which can be extrapolated through the origin for all substrate tempera-tures. No incubation period delaying the TiO2 growth during the first few ALD cycles is observed. An effect of the purge time after TDMAT pulses on the deposition rate was not observed at any temperature.

123

Figure 2: TiO2 layer density (a) for 500 deposition cycles as function of the substrate temperature. The error bars on the values of the layer density result from the standard deviation from the least square fitting procedure during XRR data analysis. Panel b) shows the coverage per cycle as function of the deposition temperature. The atomic aerial deposition is calculated as total deposition of n(Ti + 2 O) atoms per cycle. Error bars for result from the uncertainty of the density determination Reprint with permission of [20].

2 The thickness here was obtained by SE. In general, layer thicknesses yielded by SE and XRR data fitting agree within 1-2 %.

If the densities of the layers change with in-creasing deposition temperature, a constant deposition rate, which is assumed for an ideal ALD window, cannot be expected. To account for an increasing layer density, an atomic deposition rate or surface coverage of TiO2 per cycle is more appropriate. Assuming stoichiometric layers with a molar weight of 79.87 g/mole and neglecting carbon or nitrogen contaminations, an atomic areal deposition per cycle was calculated from the layer density and thickness2. The results in Figure 2b) clearly point out that the increase in density alone cannot explain the changes in deposition rate, since the rate of surface cov-erage moderately decreases with increasing temperatures up to 250 °C and then strong-ly increases for further increased deposition temperatures. The minimum of the depo-sition rate at 250 °C indicates that different chemical reaction mechanisms are active for the TiO2 growth from TDMAT and H2O. In

this temperature range, the thin film surface changes from an amorphous TiO2 to anatase and rutile termination and at the same time the decomposition mechanism of TDMAT changes from surface dominated to surface plus gas phase decomposition [30, 31]. Ana-tase and rutile offer different sites for the chemisorption of water and OH groups, with higher binding energies on the rutile surface [20, 32]. It is suggested that the combination of a thermally activated and hence delayed desorption of H2O from a rutile surface and a gas phase decomposition of TDMAT leads to an unwanted chemical reaction in the gas phase during the TDMAT pulse if the H2O purge time is too short.

To further investigate the growth mechanism, the deposition rate is correlated with the lay-er density. Figure 2a) summarizes the results of XRR data analysis for the TiO2 deposition of 500 cycles at temperatures ranging from 120 °C to 330 °C. The layer density is low for deposition temperatures below approximately 200 °C. For temperatures between 200 °C and 250 °C the layer density equals the density of

anatase, 3.96 g/cm3 [28]1. For higher tempera-tures, the density increases further to about 4.1 g/cm3 but does not reach the value of ru-tile of 4.23 g/cm3 [29]. The deposition rate decreases continuously from a GPC value of 0.57 Å/cycle at 120 °C to a minimum of 0.33 Å/cycle at 250 °C. For substrate temperatures exceeding 250 °C the GPC then increases to 0.6 Å/cycle at 330 °C.

1 The density of anatase varies in the literature in the range of 3.79-3.98 g/cm3, used here is the value for nano-crystalline powder material.

124

3.1.2 Evolution of crystalline TiO2 phases

Many reports exist in the literature that for the formation of crystalline phases a critical layer thickness of 15 nm [33] to 25 nm [34] is necessary, quite independent on the spe-cific ALD process. Analysis of the crystalli-zation behavior in this work investigates the critical layer thickness in dependence of the deposition temperature. Further the influence of a crystalline substrate layer is investigat-ed here. It is known from numerous publi-cations that purely rutile phase ALD layers can be grown epitaxially on RuO2 which has also rutile structure and offer a small lattice mismatch to rutile TiO2 (one example is giv-en in ref. 5). On the other hand, it has been reported that a TiN substrate surface could promote the TiO2 anatase phase [35]. In this work, the ALD of TiO2 on TiN surfaces is in-vestigated as TiN is a possible electrode mate-rial in metal-insulator-metal (MIM) capacitor devices. TiN is chemically inert against oxi-dation at typical ALD process temperatures and additionally offers suitable properties as an ohmic contact material to TiO2. The TiN is produced by reactive dc magnetron sput-tering from a Ti Target in Ar/N2 atmosphere at a deposition temperature of 450 °C. In contrast to the amorphous native silicon ox-ide surface, TiN layers are nanocrystalline where the TiN has rock salt structure with a lattice parameter a = 4.24 Å (e.g. ref. 36). Figure 3 summarizes grazing incidence XRD data for the deposition on native silicon ox-ide and TiN surfaces. For Si/SiO2 substrates data are shown for Ts = 300 °C and a layer thickness of 23 nm (a) and for Ts = 250 °C for layer thickness increasing from 8 to 38 nm in the 2 -range of the most prominent re-flections of anatase (101, 2 = 25.44°, [28]), rutile (110, 2 = 27.41°, [29]) and brookite (121, 2 = 30.83°, [37]). Figure 3c shows for comparison the GI diffraction patterns of TiO2 layers grown on TiN surfaces for TiO2 layer thicknesses of 16-18 nm and Ts increas-ing from 150 °C to 300 °C. In the case of Ts = 320 °C, the layer thickness amounts to 35 nm.

The positions of the ICSD reference data of the TiO2 polymorphs and TiN are indicated above the experimental diffraction patterns. The assignment of the measured reflections to either anatase or brookite is not unambig-uous. With the exception of the brookite 121 reflection at 30.8°, all measured reflections that could be assigned to brookite overlap with reflections, which can be also be attribut-ed to anatase. Also the most intense anatase reflection at 25.4°, coincides with the 021 and 111 reflections of brookite. The theoretical relative intensities for the brookite 021, 111 and 121 reflections are 100 %, 75 % and 94 % respectively [37]. It is assumed, that the mea-sured reflections of brookite in the ALD thin films appear with comparable relative inten-sities. If no significant intensity is measured for the brookite 121 reflection around 30.8°, the intensity at 25.4° is assigned to anatase only. Other crystalline phases have not been detected.

As determined by Gaussian profile fitting, the full width at half maximum (FWHM) is 0.62° ± 0.06° for the anatase 101 reflection. Within the statistics of the measurement the line broadening does not change with tem-perature or the number of deposition cycles, whereas the FWHM of the rutile 110 reflec-tion shows no dependence on deposition temperature but reduces from 1.2° to 0.75° FWHM when the number of ALD cycles is increased from 200 to 700. The Scherrer for-mula FWHM = is widely used to cal-culate the crystallite size from the broadening of XRD reflections. λ is the incident wave-length (here 1.542 Å for Cu Kα radiation), L the size of crystallites in one dimension, 0 is the diffraction angle of the considered re-flection, and ks is a correction factor, whose value is near unity [38]. Here we use ks = 1. Applying the Scherrer formula to our results, crystallite sizes of 15 ± 2 nm were obtained for anatase and 10 ± 2 nm for rutile for 200 and 300 deposition cycles increasing to 13 ± 2 nm for 700 deposition cycles.

θθ

θθ

θ

125

Figure 3: GIXRD data for Ts = 300 °C (a) and 250 °C (b) for the deposition on native silicon oxide surface. Panel a) displays the intensities measured in the 2 -range of 20°-80° for a 23 nm thick TiO2 layer. Panel b shows the 2 region of the most prominent reflections of anatase, rutile and brookite for various layer thicknesses. Data for TiO2 deposited on TiN for Ts = 150-300 °C with d = 16-18 nm and for Ts = 320 °C with d = 35 nm are displayed in panel c. Reference reflection positions for anatase [28], brookite [37] and rutile [29] and TiN [36] are indicated above the data.

For both types of substrate surfaces inves-tigated, anatase is the first crystalline phase observed. The critical minimum layer thick-ness for nucleation is 4 nm at Ts = 300 °C and increases to 19 nm at Ts = 200 °C. Below a deposition temperature of 160 °C no crys-tallization occurs and films are amorphous. For deposition on amorphous SiO2 surfaces anatase and rutile are observed together for Ts ≥ 250 °C and layer thickness exceeding ap-proximately 10 nm. On TiN surfaces the onset of anatase crystallization is similar as in the case of SiO2 substrates. The formation of ru-tile is, however, largely suppressed. Based on the measured line intensities and broadening of the anatase and rutile reflections a phase transformation is excluded. Instead, it is sug-gested that rutile nucleates at some point on an anatase surface and both phases continue

growing coexistently. Anatase (112) twin in-terfaces could offer these sites for nucleation due to their structural similarity to the rutile (100) plane [39]. As determined from XRD texture measurement, the crystalline TiN lay-er is characterized by a 001 fiber texture. This surfaces leads to a strong preferential orienta-tion of the anatase with [001] being parallel to the substrate surface, which is not observed for the deposition in native SiO2. In result, this prohibits the exposure of anatase (112) faces to the atmosphere and consequently, the for-mation of rutile is suppressed. Analysis of the preferential orientation and phase formation is subject of ongoing research.

The line broadening of the TiO2- related re- ctions is for deposition on TiN comparable to the case of TiO2 growth on silicon native oxide surfaces.

θ

θ

126

3.2 ALD of SrxTiyOZ

3.2.1 Deposition rate The chemical composition of ternary oxide layers can be adjusted by the variation of the precursor pulse sequence. The process is di-vided into part- or half cycles consisting of a metal precursor pulse followed by a purge time, an oxidation step and a second purge time, successively run for each metal com-ponent. Details of the deposition process are given in refs. 21 and 22. We tested the possible variations of SrxTiyOz compositions by vary-ing the ratio of SrO/TiO2 half-cycles between 1:8 and 5:1. Beside the stoichiometry, the GPC is a measure for the successful adsorption of the metal precursors on surfaces with varying Me-oxide and Me-hydroxide composition. The GPC of ternary SrxTiyOz layers is normal-ized to the total number of Me-O cycles or so called super-cycles by: (eq. 1)

Figure 4 summarizes the deposition rate be-havior for SrxTiyOz ALD with varying SrO:-TiO2 cycle ratios. For pure TiO2 the result-ing GPC is 0.49 ± 0.06 Å/super-cycle. For a SrO:TiO2 cycle ratio of 1:1 a GPC of 1.23 ± 0.1 Å/super-cycle was calculated. As the lattice parameter of SrTiO3 is 3.9 Å [40] a complete monolayer is deposited by three super cycles. The growth of layers deposited with more than one subsequent Sr pulse does not in-crease linearly with increasing number of cy-cles indicating a non-ALD behavior. Further there is no run-to-run reproducibility for very SrO-rich ALD sequences. This might be ex-plained by the hygroscopic nature of SrO and the formation of SrOH. Hence the discussion of the results concentrates mainly on layers with a cation ratio of Sr:Ti < 1.

Figure 4: Thickness as a function of the number of ALD super cycles for SrxTiyOz deposited by different cycle ratios at TS = 320 °C. The error bars are within the size of the symbols. The arrow indicates increasing Sr content in the layers as obtained from RBS measurements Reprint with permission of [22].

With a = number of SrO half cycles, b = number of TiO half cycles and c is the number of repetitions of the complete a (SrO) + b(TiO2) deposition cycles.

127

3.2.2 Composition and properties

For the variation of the ratio of SrO/TiO2 half cycles between 1:8 to 5:1 figure 5a summariz-es the Sr and Ti concentrations of the layers, which were obtained from RBS. The oxygen signal was also evaluated, but is not includ-ed in here. For a cycle ratio of SrO:TiO2 = 1:1

Figure 5: a) Atomic concentrations of Sr and Ti as a function of the Sr:Ti half-cycle ratio. The element compositions were determined by RBS. The horizontal line marks the Sr, Ti concentration of stoichiometric SrTiO3. Panel b) shows the optical gap energy Eg and refractive index n of the layers as a function of the Sr:Ti layer composition obtained from SE data analysis Reprint with permission of [22].

a layer composition of Sr:Ti:O = 18.6 ± 0.4 : 22.8 ± 0.5 : 58.6 ± 1.2 at% (error results from fitting procedure) was determined, which equals approximately 1.0 : 1.2 : 3.1 with a cor-responding Sr/Ti ratio of 0.82.

For the half cycle ratios of SrO:TiO2 = 2:1, 1:1 and 1:2 the Sr and Ti concentrations are equal within the limit of the accuracy of the method. However, the RBS spectrum of layers deposit-ed with SrO:TiO2 = 2:1 show a decreasing Sr concentration from the interface to the surface, whereas for half-cycle ratios of 1:1 and 1:2 ho-mogenous Sr and Ti distributions throughout the layer thickness are measured (see ref. 22). A composition closest to ideal, stoichiometric SrTiO3 could be achieved for a cycle ratio of SrO:TiO2 = 1:2 resulting in a composition of Sr:Ti:O = 1:1.02:3.3 and a Sr/Ti ratio of 0.98. Figure 5 b shows the dependence of the re-fractive index n and the optical gap Eg of SrxTiyOz on the Sr/Ti layer composition. The composition is obtained from RBS, optical parameters result from SE data analysis. For near stoichiometrich SrTiO3 layers Eg was measured to be 3.87 eV which corresponds well to the direct band to band transition in

bulk crystalline SrTiO3 [18]. With an increas-ing Sr content, the optical gap energy increas-es to a maximum of 4.38 eV at Sr/Ti = 2.75. Concomitantly, n decreases from 2.3 to 1.7 with increasing the Sr content from 6.3 ± 0.2 at% to 22.9 ± 0.5 at%. For Sr rich layers the values of the refractive index do not approx-imate the value known for bulk SrO (1.81 at 2 eV [41]). Instead it approaches bulk Sr(OH)2 (1.59 at 2 eV [42]), which gives evidence, that for more than one subsequent SrO half-cycle deposition on each other, also Sr(OH)2 is in-corporated in the layers. With increasing Ti content n and Eg approach the values of pure ALD TiO2 for which we determined the re-fractive index to be n = 2.5 and Eg = 3.3 eV (for ALD at Ts = 320 °C, ref. 20).

128

Figure 6: Current-voltage characteristics (a) of TiN-Ti/Sr oxide-Au capacitor structures with varying oxide stoichiometry. The contact size is 100 x 100 µm2 and oxide thickness is 15-16 nm, Reprint with permission of [22].

As the SrxTiyOz thin films are intended for the use as resistance switching oxide for data stor-age devices, the initial electric leakage current of the as deposited layers have been deter-mined from current-voltage (I-V) character-istics. For TiN-SrxTiyOz-Au MIM devices with an oxide thickness of approximately 15 nm and contact size of 100 x 100 µm2 I-V charac-teristics are shown in Figure 6 (voltage is ap-plied at the TiN bottom electrode). In compar-ison to pure amorphous TiO2, all amorphous ternary oxide layers show leakage currents being two orders of magnitude lower than that of TiO2 at comparable layer thickness. The I-V measurements were carried out as hysteresis starting at 0 V in pos-itive direction, whereas the voltage was applied at the TiN bottom electrode. For stoichiometric and layers with Sr ex-cess a typical schottky diode characteristic is observed. The current difference between forward and backward direction of the mea-surement is associated with the charging of the stack. The positive voltage branch or re-versed-bias direction is dominated by a schott-

ky barrier. At negative voltage or forward-bias direction the conductivity of the semiconduc-tor dominates at higher voltages. As it can be expected from the work functions of SrTiO3 (4.2 eV [43]), TiN (4.0-4.8 eV [44],45,46) and Au (5.3 eV [47]) this characteristic can be ex-plained by a schottky contact at the interface of SrxTiyOz/Au and an ohmic contact at the TiN/ SrxTiyOz interface. For layers with Ti ex-cess the current on the negative branch does not show a logarithmic increase characteristic of the schottky barrier. Either, this can be explai-ned by a strong change of the work function of SrxTiyOz which effects a second schottky barri-er at the interface to the bottom electrode. Or, as the conductivity of the oxide layer decre-ases with increasing Ti content, the conduc-tivity of the oxide layer becomes the limiting factor rather than the barrier at the interface. From the I-V characteristic a leakage current density JL was determined at +0.5 V, values are listed in Table 2. The leakage current strong-ly increases from 5.6 ± 1.4 fA for layers with Ti excess to 113.0 ± 2.9 fA for Sr excess sam-ples.

3.3 Electrical characterization

3.3.1 Leakage current in amorphous SrxTiyOz layers

129

In general, no breakdown was observed for 15 nm thick SrxTiyOz layers up to 6 V and in turn, no resistance switching could be deve-loped. Hence, in the as-deposited amorphous state, SrxTiyOz layers are not suitable as resis-tance switching oxide. Further work inves-tigates the potential of crystallization or ion beam modification to purposefully introduce structural inhomogenities and defects. Their effect on leakage current and possible resis-tance switching is subject of ongoing research.

In contrast to SrTiO3, titanium dioxide crys-tallizes readily at temperatures around 200 °C and the crystalline structure of the ALD layers can be adjusted by the substrate temperature Ts (see section 3.1.2).

MIM capacitor structures are produced identically to the above case of SrxTiyOz in-sulator layers. In a similar way, I-V curves before any electroforming are characterized by a schottky barrier at the TiO2/Au interface and an ohmic contact at the TiN/TiO2 inter-face. For thin oxide layers of approximately

Table 2: Leakage current density measured in reverse direction at 0.5 V for SrxTiyOz layers of different chemical composition.

Sr:Ti J [A/cm2] at 0.5 V

1.9 1.13x10-6 ± 2.88x10-8

0.95 3.51x10-7 ± 2.09x10-8

0.5 5.63x10-8 ± 1.43x10-8

3.3.2 Resistance switching of TiN-TiO2-Au MIM structures

15-18 nm the leakage current density JL in reverse direction depends strongly on mi-crostructure: for amorphous TiO2, JL is below 1.10-6 A/cm2 and increases for crystalline layers to JL >1.10-1 A/cm2.

After electroforming only amorphous TiO2 oxide exhibits stable and reproducible re-sistance switching IV characteristic with ROFF/RON ≈500. The left side of figure 7 shows current-voltage and resistance-voltage curves after forming. The right panel shows read-ings after every tenth cycle of ROFF and RON at +0.2 V during 2000 switching cycles.

Layers with fractions of crystalline TiO2 or fully crystallized to anatase show either an immediate irreversible hard breakdown or fading resistance hysteresis loops where the ratio decreases after few cycles ending up in a permanent low resistance state.

130

4 Summary

The atomic layer deposition of TiO2 and SrTiO3 from Sr(iPr3Cp)2 and TDMAT precur-sors together with H2O as oxygen source has been demonstrated. For the deposition of pure TiO2 crystallization of anatase is observed for Ts > 180 °C. A critical minimum layer thick-ness dc is necessary to obtain crystalline layers. This dc decreases from 19 nm at Ts = 200 °C to 4 nm at Ts = 300 °C. Using the layer by layer approach, ternary ox-ides with a wide range of composition can be produced in the system SrO-TiO2, including the stoichiometric SrTiO3. At a deposition temperature of 320 °C all layers of ternary composition are amorphous By adjusting the layer composition, the refractive index and permittivity, optical gap and leakage current can be controlled between the values of amor-phous TiO2, SrTiO3, and SrOH.

Metal-insulator-metal capacitor structures with an TiN bottom electrode and Au top electrode have been fabricated with 15 nm of either TiO2 or SrxTiyOz insulator layers. MIM structures with amorphous ternary layers are characterized by leakage current densities de-creasing from 1 × 10-6 A/cm2 to 5 × 10-8 A/cm2 for increasing Ti concentration. No switching of electrical resistance could be induced in

Figure 7: Resistance switching characteristic of TiN-TiO2-Au capacitor stack measured on a 100 x100 µm2 contact. The TiO2 layer is 15 nm thick and amorphous. The left side shows current-voltage and resistance-voltage loops observed after an initial forming step. The right graph shows constant levels of ON and OFF state resistance measured during 2000 cycling steps.

this highly insulating layers. MIM structures with amorphous TiO2 insulator layers show leakage current densities of 1 × 10-6 A/cm2, which are increasing to the order of 10-1 A/cm2 for crystalline TiO2. Amorphous titania ALD insulator layers show stable resistance switch-ing behaviour with ROFF/RON ≈500.

Acknowledgement This work was performed within the Cluster of Excellence “Functional Structure Design new high performance Materials via Atom-ic Design and Defect Engineering” (ADDE) that is financially supported by the European Union Regional Development Funds and by the Ministry of Science and Art of Saxony (SMWK). Part of this work was performed within the Virtual Institute MEMRIOX (“Memory Effects in Resistive Ion-beam Modified Oxides”, VH-VI-442). Wafer pro-cessing, including cleaning, oxidation and lithography have been carried out in the cen-tral clean room laboratory of the TU Berga-kademie Freiberg.

131

1. B. O’Regan, M. Grätzel, Nature 353 (1991), 7372. S. Ito, T. N. Murakami, P. Comte, P. Liska, C. Grätzel, M. Nazeeriddin, M. Grätzel, Thin

Solid Films 516 (2008) 46133. A. Fujishima, X. Zhang, D. Tryk, Surf. Sci. Rep. 63 (2008) 5154. M. N. Chong, B. Jin, C. W. K. Cho, C. Saint, Water Res. 4 (2010) 29975. S. K. Kim, S. W. Lee, J. H. Han, B. Lee, S. Han, C. S. Hwang, Adv. Funct. Mater. 20 (2010)

2989 6. M. Popovici, M.-S. Kim, K. Tomida, J. Swerts, H. Tielens, A. Moussa, O. Richard,

H. Bender, A. Franquet, T. Conard, L. Altimime, S. Van Elshocht, J. A. Kittl, Microelectron. Eng. 88 (2011) 1517

7. N. Aslam, V. Longo, W. Keuning, F. Roozeboom, W. M. M. Kessels, R. Waser, S. Hofmann-Eifert, Phys. Stat. Solid. A 211 (2014) 389

8. E. Langereis, R. Roijmans, F. Roozeboom, M.C:M: Van de Sanden, W.M.M. Kessels, J. Electrochem. Soc. 158 (2011) G34

9. H. Y. Jeong, J.Y. Lee, S.-Y. Choi, Adv. Funct. Mater. 20 (2010) 391210. B. J. Choi, D. S. Jeong, S. K. Kim, C. Rohde, S. Choi, J. H. Oh, H. J. Kim, C. S. Hwang,

K. Szot, R. Waser, B. Reichenberg, S. Tiedke, J. Appl. Phys. 98 (2005) 033715 11. K. M. Kim, D. S. Jeong, C. S. Hwang, Nanotechnology 22 (2011) 254002 12. X. B. Yan, K. Li, J. Yin, Y. D. Xia, H. X. Guo, L. Chen, Z. G. Liu, Electrochem.

Solid State Lett. 13 (2010) H8713. R. Dittmann, R. Muenstermann, I. Krug, D. Park, T. Menke, J. Mayer, A. Besmehn,

F. Kronast, C.M. Schneider, R. Waser, Proc. IEEE 100 (2012) 197914. C. Detavernier, J. Dendooven, S. P. Sree, K. F. Ludiwig, J. A. Martens, Chem. Soc. Rev. 40

(2011) 524215. V. V. Hoang, H. Zung, N. H. B. Trong, Eur. Phys. J. D 44 (2007) 51516. H. Zhang, B. Chen, J. F. Banfield, G. A. Waychunas, Phys. Rev. B. 78 (2008) 214106 17. Y. A. Abramov, V. G. Tsirel’son, V. E. Zavodnik, S. A. Ivanov, I. D. Brown, Acta Crystallo-

graphica B: 51 (1995) 942, ICSD No. 8087318. K. van Benthem, R.H. French, C. Elsässer, J. Appl. Phys. 90 (2001) 615619. W. Kern, J. Electrochem. Soc. 137 (1990) 1887 20. B. Abendroth, T. Moebus, S. Rentrop, R. Strohmeyer, M. Vinnichenko, T. Weling,

H. Stöcker, D.C. Meyer, Thin Solid Films 545 (2013) 17621. S. Rentrop, T. Moebus, B. Abendroth, R. Strohmeyer, A. Schmid, T. Weling, J. Hanzig,

F. Hanzig, H. Stöcker, D. C. Meyer, Thin Solid Films 550 (2014) 5322. S. Rentrop, B. Abendroth, J. Walter, J. Rensberg, R. Strohmeyer, H. Stöcker, C. Ronning,

S. Gemming, D. C. Meyer, accepted for publication in Thin Solid Films (2015)23. E.D. Palik, Handbook of optical constants of solids, Academic Press, Orlando, Florida,

198524. H. Fujiwara, Spectroscopic Ellipsometry, Principles and Applications, John Wiley & Sons

Inc, Chichester, U.K., 200725. G. Lim, D. Kim, Thin Solid Films 498 (2006) 25426. Q. Xie, YL. Jian, C. Detavernier, D. Deduytsche, R.L. Van Meirhaeghe, G.P. Ru, B.Z. Li,

X.P. Qu, J. Appl. Phys. 102 (2007) 08352127. Q. Xie, J. Musschoot, D. Deduytsche R.L. Meirhaeghe, Ch. Detavernier, S. Van den Ber-

ghe, Y.L. Jiang, G.-P- Ru, B.Z. Ru, B.Z. Li, X.P. Qu, J. Electrochem. Soc. 155 (2008) H68828. T.E. Weinreich, M. Winterer, S. Seifried, H. Hahn, H. Fuess, Ultramicroscopy 81 (2000)

263-270, ICSD No. 92363

References

132

29. S. E. Rassmussen, Powder Diffraction 18 (2003) 281-284, ICSD 10946930. E.T. Norton, C. Amato-Wierda, Chem. Mater. 13 (2001) 465531. M. Boumann, F.J. Zarea, Electrochem. Soc. 158 (2011) D52432. P. Malet, G. Munuera, J. Chem. Soc. Faraday Trans. 1 (1989) 415733. J. Aarik, A. Aidla, T. Uustare, V. Sammelselg, J Cryst. Growth 148 (1995) 26834. N. G. Kubala, P. C. Rowlette, C. A. Wolden, Electrochem. Solid State Lett. 12 (2009) H25935. M. Rose, J. Niinistö, P. Michalowski, L- Gerlich, L. Wilde, I. Endler, J.W. Bartha, J. Phys.

Chem. C 113 (2009) 2182536. J. Chorn-Cherng, T. Goto, T. Hirai, J. Alloy. Comp. 190 (1993) 197, ICSD 60422037. I. Djerdj, A.M. Tonejc, Journal of Alloys and Compounds 413 (2006) 159-174, ICSD

15460538. A. L. Patterson, Phys. Rev. 56 (1939) 97839. R. L. Penn, J. F. Banfield, Am. Miner. 84 (1999) 87140. L. X Cao, E. Sozontov and J. Zegenhagen, Phys. Status Solidi A 181 (2000) 387–40441. Pradyot Patnaik, Handbook of Inorganic Chemicals, McGraw-Hill, New York, 2002,

888–88942. E. T. Carlson, J. Res. Natl. Bur. Stand., 53 (1954) 37143. Y.-W. Chung, W. B. Weissbard, Phys. Rev. B 20 (1979) 345644. A. Kerber, E. Cartier, R. Degraeve, Ph. Roussel, L. Pantisano, T. Kauerauf, G. Groesene-

ken, H. E. Maes, U. Schwalke, IEEE Trans. Elec. Dev. 50 (2003) 126145. B. Claflin, M. Binger, G. Lucovsky, J. Vac. Sci. Technol. A 16 (1998) 175746. G. S. Lujan, T. Schram, L. Pantisano, J. C. Hooker, S. Kubicek, E. Rohr, J. Schuhmacher,

O. Kilpela, H. Sprey, S. De Gendt, K. De Meyer, Proc. ESSDERC, Firenze, Italy, 2002, 58347. M. Uda, A. Nakamura, T. Yamamoto, Y. Fujimoto, J. Electron. Spectrosc. Relat. Phenom.

88-91 (1998) 643

133

Synthesis and characterization of Ge nanocrystals embedded in high-k materials for alternative non-volatile memory devices David Lehninger 1, Peter Seidel 1, Maximilian Geyer 1, Frank Schneider 1, Alexander Schmid 1, Volker Klemm 2, David Rafaja 2 and Johannes Heitmann 1.

1 TU Bergakademie Freiberg, Institute of Applied Physics, D-09596 Freiberg, Germany. 2 TU Bergakademie Freiberg, Institute of Materials Science, D-09596 Freiberg, Germany.

134

AbstractIn a first study, superlattices consisting of alternating Ge1.6ZrO2 and pure ZrO2 layers were sput-tered on Si wafers, which were either covered by a naturally grown oxide or by a 1.3 nm thick Si3N4 film. After thermal treatment at different temperatures, the crystallization of Ge and ZrO2 was studied by using transmission electron microscopy. A thin layer of SiO2 leads to the growth of elongated crystalline Ge areas with insufficient control of shape, size-, and spatial distri-bution, whereas a thin layer of Si3N4 results in a parallel growth direction of ZrO2 and Ge. A multilayer structure is obtained with thin Ge films between thicker layers of ZrO2. In both cases Ge and ZrO2 start to crystallize simultaneously at about 660 °C. In a second study, ZrO2 was replaced by TaZrOx. The higher crystallization temperature of TaZrOx leads to the formation of spherical shaped Ge-nanocrystals with a small variation of size, areal density-, and depth distribution in amorphous TaZrOx. Thus, TaZrOx is a promising matrix for Ge-nanocrystal based non-volatile memories. In this study, the charge storage characteristics of the Ge-nano-crystals embedded in a TaZrOx were investigated by capacitance-voltage measurements using metal-insulator-semiconductor structures. Samples with one layer Ge-nanocrystals exhibit a counterclockwise hysteresis loop of the flatband voltage shift, whereas reference samples with-out Ge-nanocrystals do not show any hysteresis loop. A second layer of Ge-nanocrystals does not produce a second flat stair in the memory window characteristics. The optimum annealing temperature for achieving the widest hysteresis loop was found to be between 700 and 750 °C.

Introduction Semiconductor nanocrystals (NCs) embed-ded in a dielectric matrix are of great interest for a broad range of applications. They are considered as absorbers for third generation solar cells [1], as absorbers for the excitation of rare earth elements [2] and as light emitters for Si-based photonics [3]. In this work, another promising application of NCs is addressed that is their potential use as charge storage nodes for non-volatile memories.

The basic element of conventional non-volatile memories is the floating-gate (FG) transistor as proposed by Kahng and Sze in 1967, where a continuous FG acts as charge storage node (Fig. 1a) [4]. It suffers from the fact that a single defect in the bottom tunneling oxide discharges the whole FG leading to a loss of information. Beside the charge trapping memories [5], in which the conductive FG was replaced by a nonconductive layer of silicon nitride or oxynitride, the memory cells based on charge storage in discrete NCs [6] (Fig. 1b) are an interesting approach to overcome this issue. NC based FG transistors are more resistant to charge losses and can be produced with thinner tunneling oxides therefor. This

in turn offers the chance for lower program/erase voltages, shorter program/erase times and better endurance [7, 8]. The operation voltage can be further tuned by combining a SiO2 tunneling oxide with a high-k control oxide, such as TaZrOx. A further advantage of NC based FG devices is the coulomb blocking effect, which was demonstrated for Si-NCs embedded in SiO2 [9]. This effect increases the programming voltage that is needed to write additional charges after a certain amount of charge is stored in the NCs. The resulting different states can potentially be used for multi-level cells. A suitable alternative material for the production of NCs is Ge. Ge-NCs are compatible with current CMOS technologies. In comparison with Si-NCs, the Ge-NCs possess a lower crystallization temperature and a negative conduction band offset with respect to the Si substrate. The latter is a promising NCs property allowing faster programming speeds and longer data retention [10]. However, since the bit density of traditional flash memories is increased by shrinking the FG area, which is below 20 x 20 nm² in current technologies, only a very small number of NCs can replace the

135

FG. Thus, a process-related fluctuation in the number of NCs per memory cell will lead to a highly undesired fluctuation of the device’s threshold voltage. Consequently, the NC-based approach is not promising for conventional two-dimensional devices. However, the future trend goes towards three-dimensional memory devices with a circular-shaped FG in order to overcome the natural limitations of the horizontal device scaling [11]. The circular FG provides a much larger

area for the integration of NCs and thus could open the way for NCs to three-dimensional memories.

In this work, Ge-NCs are formed in ZrO2 and TaZrOx by phase separation of co-sputtered Ge-ZrO2 and Ge-TaZrOx layers, respectively. This approach is more promising for three-dimensional flash devices than other approaches such as the widely used implantation of Ge ions [10, 12].

Fig. 1: Schematic structure of a (a) conventional non-volatile memory with a continuous floating gate and a (b) nanocrystal based floating gate memory.

ExperimentalA. Phase separation and crystallization in the System Ge-Zr-O

To study the crystallization of Ge in ZrO2, superlattices consisting of 10 alternating Ge1.6ZrO2- and ZrO2-layers were deposited on p-type Si wafers, which were either covered by a naturally grown oxide (Fig. 2a) or by a 1.3 nm thick Si3N4 film (Fig. 2b). The Ge1.6ZrO2 layers were deposited by co-sputtering of ZrO2 and Ge using a collinear arrangement of the targets. During the deposition of pure ZrO2, the shutter of the Ge target was closed. The superlattices deposited on the wafers with natural grown oxide (Fig. 2a) were additionally covered by a 20 nm thick SiO2 film, in order to suppress unintended oxidation of Ge (by the residual oxygen in the Ar atmosphere) and the out-diffusion of GeOx

during annealing. These stacks were annealed at 660 °C for 15 min in Ar atmosphere to form Ge-NCs. The superlattices deposited on the wafers covered with Si3N4 (Fig. 2b) were not additionally covered with a SiO2 film. Thus these stacks were annealed in vacuum to suppress unintended oxidation of Ge and out-diffusion of GeOx. The annealing temperatures ranged from 300 to 800 °C. The size, areal density and spatial distribution of the Ge clusters were analyzed on cross-section specimens using a JEOL 2200 FS transmission electron microscope (TEM) with a spherical aberration corrector located in the illumination system.

136

B. Formation and non-volatile memory characteristics of Ge-NCs in TaZrOX

To study the crystallization phenomena and charge trapping characteristics of the Ge-NCs embedded in TaZrOx, p-type silicon wafers (100) with a resistivity of 2.75 Ohm cm were (i) subjected to a standard RCA cleaning procedure, (ii) dipped in diluted hydrofluoric solution, and (iii) immediately transferred to an oxidation chamber to form the thermally grown tunneling oxide with a thickness of 5 nm. Afterwards, storage layers with a thickness of 6 nm were deposited by co-sputtering Ge and TaZrOx targets. The number of storage layers varied between zero (reference sample) and two (two-layer sample). In case of the two-layer sample, the storage layers were separated by a Ge-free TaZrOx layer with the thickness of 2 nm. The Ge concentration was varied by using different rf power densities. Subsequently, the blocking oxide was deposited by sputtering pure TaZrOx with a thickness of 16 nm (single-layer and two-layer sample) or 21 nm (reference sample without Ge-NCs) in

order to keep the overall dielectric thickness comparable. Finally, Al2O3 capping layers with the thickness of 10 nm were deposited on all samples to suppress unintended oxidation and out-diffusion of GeOx during annealing. The aluminum oxide was produced in an atomic layer deposition (ALD) process by using water and trimethylaluminum (TMA) as precursors. The samples were annealed for 30 seconds at temperatures between 525 and 800 °C in N2 atmosphere using a rapid thermal processing (RTP) tool to form the Ge-NCs. For electrical studies Ti/Al-gate electrodes were deposited by means of electron beam evaporation applying a shadow mask to form metal-insulator-semiconductor (MIS) capacitors (Fig 3). The capacitance-voltage (C-V) measurements were carried out by using an Agilent E4980A semiconductor parameter analyzer at an ac frequency of 100 kHz.

Fig. 2: Schematical drawing of ZrO2/Ge-ZrO2-supperlattices deposited on p-type Si wafers covered with (a) natural grown oxide and (b) 1.3 nm Si3N4.

137

Fig. 3: MIS capacitors for the electrical characterization of the Ge-NCs: (a) reference sample without Ge-NCs (b) single-layer sample (c) two-layer sample.

Fig. 4 shows TEM images of the stack seen at Fig. 2a after annealing at 660 °C. The darker areas of the left panel are attributed to ZrO2 layers, whereas the brighter layers mainly consist of Ge. The contrast at the interfaces is enhanced by the Fresnel defocus. The ex-pected number of layers present in the mul-tilayer stack confirms that no layer has been desorbed and no inter-diffusion between the layers has occurred during the annealing pro-

cess. The high-resolution TEM (HRTEM) image on the right hand side of Fig. 4 gives a deeper insight into the interface between the first deposited superlattice and the Si sub-strate. After annealing at 660 °C both Ge and ZrO2 are crystallized. Elongated, non spheri-cal Ge-NCs with insufficient control of shape, size, and spatial distribution were created by Ge agglomeration.

Fig. 4: TEM image of a superlattice consisting of alternating [Ge1.6ZrO2]10 and ZrO2 layers annealed at 660 °C (left panel). HRTEM image of the interface region showing Si substrate, native SiO2 and the first part of the Ge/ZrO2 stack (right panel).

Results and DiscussionA. Phase separation and crystallization in the System Ge-Zr-O

138

Fig. 5 shows a HRTEM image of a part the superlattice deposited on the Si3N4 wafer (Fig. 2b) after annealing at 650 °C. Two of the inserts in Figure 7 present the Fast Fourier Transformations (FFT) of the HRTEM from the regions labelled A and B. Region A represents a continuous Ge layer, whereas the regions B stands for nanocrystalline tetragonal ZrO2 grains within the closed ZrO2 layer. The third inset shows the lateral distribution of Ge and Zr in the multilayer stack. The composition profile was measured by using the energy dispersive x-ray spectroscopy (EDX) perpendicular to the

individual layers over a distance of 70 nm and demonstrates that Ge and ZrO2 segregated and formed a ZrO2-Ge superlattice at this temperature. Furthermore, it was found that the Si3N4 layer covering the Si substrate influences the crystallization process of both ZrO2 and Ge strongly. Crystallization of ZrO2 starts at the Si3N4 layer that also determines the orientation of the growing ZrO2 and Ge films resulting in parallel growth directions [110]ZrO || [112]Ge. A multilayer structure is obtained with thin Ge films between thicker layers of ZrO2.

To identify the crystallization temperatures of Ge and ZrO2, GAXRD patterns have been recorded after annealing the stacks grown on Si3N4 at temperatures between 300 and 800 °C [13]. While the GAXRD patterns of the 300 °C and 600 °C samples do not show any sharp diffraction lines, pronounced reflections appear after annealing at 650 °C that can be attributed to both ZrO2 and Ge. The Ge reflections become more intense and narrower with increasing temperatures above 650 °C.

Summarizing the above, it can be said that pure ZrO2 is not a promising material to synthesize spherically shaped Ge-NCs in amorphous blocking oxide, as requested for non-volatile memories. Using a thin SiO2 layer covering the Si substrate leads to the growth of elongated crystalline areas of Ge with insufficient control of shape, size-, and spatial distribution. A thin Si3N4 layer covering the Si substrate results in a parallel growth direction of ZrO2 and Ge. Thus, a multilayer structure is obtained with thin Ge films between thicker ZrO2 layers. However, both cases result in a crystallized matrix which is undesired for memory applications.

Fig. 5: TEM image of a superlattice consisting of alterna-ting [Ge1.6ZrO2]10 and ZrO2 layers annealed at 660 °C (left panel). HRTEM image of the interface region showing Si substrate, native SiO2 and the first part of the Ge/ZrO2 stack (right panel).

2

139

A well-established way to increase the crystallization temperature of a material is to dope the material with foreign atoms of a non-miscible species, such as tantalum.

Fig. 6 shows HRTEM images of the samples schematically drawn in Fig. 3b and 3c after annealing at 700 °C. It can be seen that the TaZrOx blocking oxide stays amorphous, even after thermal treatment at temperatures above 660 °C. During the annealing, thermal energy leads to a phase separation of the co-sputtered storage layers, whereby spherically shaped Ge-NCs are formed within amorphous TaZrOx. The depth distribution of the Ge-NCs can be controlled by the spatial position of the storage layers. The average size of the Ge-NCs is 7 nm in diameter for both samples in Fig. 6 and can be controlled by the thickness of the storage layers. The spatial density of the Ge-NCs was estimated to be about 9 x 1011 cm-2 for the single-layer sample (Fig. 6a) and 4 x 1011 cm-2 for the two-layer sample (Fig. 6b).

It can be controlled by the concentration of Ge within the storage layer. The high Ge concentration in the single-layer sample leads to small distances between adjacent NCs. Some NCs in Fig. 6a are in electrical contact, which may cause an undesired lateral charge transport. Some NCs of the single-layer sample grow into the SiO2 tunneling oxide resulting in an undesired variation of the tunneling oxide thickness (see also Fig. 6a). This indicates that the SiO2 tunneling oxide does not act as a sufficient barrier to prevent Ge diffusion in this material system [14]. Both storage layers of the two-layer sample were sputtered with a lower concentration of Ge compared to the single-layer sample, which results in the above mentioned smaller NC areal density. Hence, these NCs are mostly well isolated from each other and tend less to grow into the tunneling oxide.

Fig. 6: HRTEM image of the (a) single-layer sample and (b) two-layer sample after annealing at 700 °C.

B. Formation and non-volatile memory characteristics of Ge-NCs in TaZrOX

140

Fig. 7 shows a normalized C-V curve of the single-layer MIS capacitors as seen in Fig. 3b after annealing at 700 °C. Negative bias voltages shift the C-V characteristics to the left side (which is in agreement with the trapping of holes), whereas positive bias voltages shift the C-V characteristics to the right side (which is in agreement with the detrapping of holes and / or the trapping of electrons). The number of stored holes is much higher than the number of stored electrons, as revealed by the higher shift to negative flatband voltages than to positive flatband voltages. Charge injection occurs predominantly from the tunneling oxide, as indicated by the counterclockwise hysteresis loop. The width of the resulting hysteresis loop (also called memory window) is proportional to the number of stored

charges. The largest memory window of about 5 V was measured by sweeping the voltage from -7 V to +7 V and back to its initial value. Compared to the review of Chang et al [15]. who compared the memory window of many different material combinations, and to the results of Kanoun et al. (2 V) [16], Das et al. (0.55 V) [17], and Beyer et al. (3.3 V) [18] who studied the memory characteristics of Ge-NCs embedded in SiO2, the memory window reported here is remarkably wide. The reference sample without Ge-NCs (Fig. 3a) does not show any memory window (cf. Fig. 7). This indicates that the charges in the Ge-NCs containing samples are confined in the valence band (holes) or conduction band (electrons) of the NCs and / or trapped at their interfaces to the surrounding material.

Fig. 7: Normalized C-V curves of the single-layer sample and the reference sample. Samples with Ge-NCs show a counterclockwise hysteresis, whereas samples without Ge do not show any hysteresis.

Fig. 8 demonstrates the dependence of the memory window size on the bias voltage sweep for samples annealed at 700 °C. Reference samples without Ge-NCs do not show any memory window for all tested bias voltage sweeps. In contrast, samples with Ge-NCs exhibit memory windows that get wider for larger voltage sweeps. The fitted slope for the two-layer sample is 0.79 (fitted from 2 V to 6 V) if the hysteresis loop was run through fast and 0.70 (fitted from 3 V to 8 V) if the hysteresis loop was run through slowly. The fitted slope for the single-layer sample is 0.99

(fitted from 2 V to 7 V), which is very close to the ideal slope of 1. The smaller slope of the two-layer sample can be explained by the effectively thicker tunnel oxide as shown by HRTEM in Fig 6b. Saturation occurs after sweeping the voltage beyond ±7 V (single-layer sample) and ±6 V (two-layer sample, fast) or ±8 V (two-layer sample, slow). Higher voltages result in distorted C-V curves and decreasing memory windows, which is probably due to the high leakage current that discharges the NCs across the gate contact. The second layer Ge-NCs does not create

141

Fig. 8: Size of the memory win-dow versus the bias voltage sweep for the single-layer sample, the two-layer sample (two different sweeping speeds) and the refe-rence sample without Ge-NCs.

Fig. 9 illustrates the absolute value of the flatband voltage as a function of the programming time for different negative bias voltages applied to the single-layer sample after annealing at 700 °C. Within the first 50 - 100 s of programming, the flatband voltage shifts towards higher absolute values, which means that the number of stored holes increases with increasing time. After that time saturation occurs. Saturation could be caused by (i) trapped holes that produce a repulsive electric field (coulomb barrier) for additional

holes [20] and / or by (ii) absence of free trapping centers. Programming at higher bias voltages causes saturation at higher flatband voltages. The higher bias voltage (i) leads to a higher energy of holes, which enables them to overcome the repulsive field and/or (ii) activates energetically deeper trapping centers by causing a stronger band bending. Hence, a higher bias voltage results in a higher maximum number of stored holes. The different states for different bias voltage could potentially be used for multi-level cells.

a second flat stair in the memory window characteristics of the two-layer sample, as reported for samples with two layers of Si-NCs embedded in SiO2 [19]. Taking into account that the two-layer sample with the overall largest number of NCs stored less charges than the single-layer sample, it is likely to assume that charges are only stored in the first layer, which is placed directly above the tunneling oxide. There are two possible explanations for this: (i) Charges stored in the first layer form a coulomb barrier that shields the second layer from the substrate. To overcome this barrier such high bias voltages are required that appreciable leakage

currents set in, which discharge the Ge-NCs in this system. (ii) The NCs of the first layer share an interface with the tunneling oxide (SiO2) and the control oxide (TaZrOx), whereas the NCs of the second layer share an interface only with the control oxide. While the interface to the tunneling oxide may provide an appreciable number of defects for charge trapping, the interface to the blocking oxide is better passivated. Once it is accepted that charges can only be written in the first layer, the smaller maximum of stored charges in the two-layer sample (Fig. 8) can be explained by the lower NC area density compared to the single-layer sample (as shown by TEM in Fig 6).

142

Fig. 9: Absolute values of the flat-band voltage as a function of the programming time for different negative bias voltages applied to the single-layer sample.

Fig. 10: (a) Memory window size versus annealing temperature. For each C-V curve, the bias voltage was swept from -5 V to 5 V and back. (b-d) Typical C-V curves for the corresponding temperature range.

143

Fig. 10a illustrates how different annealing temperatures influence the size of the memory window. For each annealing temperature, the memory window size was determind by sweeping the bias voltage from -5 V to 5 V and back to its initial value. Even though annealing at temperatures lower than 700 °C leads to the formation of remarkable memory window sizes, corresponding C-V curves (Fig. 10b) look distorded. The latter indicates a high number of defects at the interface between the tunneling oxide and the first deposited layer. The largest memory windows can be achieved after annealing between 700 and 750 °C. Corresponding C-V curves (Fig. 10c) look smooth and seem to be free of defects. Temperatures above 750 °C lead to a successive reduction of the memory window size. After annealing at 800 °C, the memory window vanishes completely. However, corresponding C-V curves (Fig. 10d) look still smooth and without remarkable interface defects.

A cross-sectional TEM image (Fig. 11) was taken of the sample annealed at 800 °C in order to investigate the decreasing memory window after annealing at temperatures above 750 °C. The Al2O3 oxygen diffusion barriere degrades completely during the annealing step, which leads to an unhindered diffusion of residual oxygen. Once the oxygen reaches the Ge clusters, Ge forms GeOx that diffuse to the surface and desorb during annealing. This process starts at about 750 °C. Higher temperatures lead to a stronger degradation and thus to an accelerated out diffusion of GeOx. Consequently, the amount of stored charges decreases for increasing temperatures. Furthermore, the TaZrOx control oxide crystallizes after annealing at 800°C, which leads to poorer leackage current characteristics. Thus, the charges stored in the NCs could be discharged across the control oxide more easily than after annealing at temperatures below the crystallization point. Hence, the optimum annealing temperature range is between 700 and 750 °C.

Fig. 11: Cross-sectional TEM image of the MIS capacitor with one-layer Ge-NCs after annealing at 800 °C for 30 s in N2 atmosphere.

144

Conclusions

Ge-NCs having spherical shape and well-defined size can be formed with good controllable area density and depth distribution in amorphous TaZrOx by thermal annealing of co-sputtered Ge-TaZrOx layers. Samples with Ge-NCs exhibit a counterclockwise hysteresis loop (memory window) with a maximum size of 5 V, which is remarkably wide compared to the memory windows reported for Ge-NCs embedded in

References

1. M.A. Green, Mat. Sci. Eng. B 74, 118 (2000).2. P.G. Kik and A. Polman, J. Appl. Phys. 88, 1992 (2000).3. L. Dal Negro, L. Pavesi, G. Pucker, G. Franzò, and F. Priolo, Opt. Mat. 17, 41 (2001).4. K. Kahng and S. Sze, IEEE Trans. Electron Devices 14, 629 (1967).5. P. Chen, IEEE Trans. Electron Devices 24, 584 (1977).6. S. Tiwari, F. Rana, H. Hanafi, A. Hartstein, E.F. Crabbé, and K. Chan, Appl. Phys. Lett. 68, 1377 (1996).7. H. Hanafi, S. Tiwari, and I. Khan, Fast and long retention-time nano-crystal memory (1996.), IEEE Trans. Electron Devices 43, 1553.8. S. Tiwari, F. Rana, K. Chan, L. Shi, and H. Hanafi, Appl. Phys. Lett. 69, 1232 (1996).9. T.Z. Lu, M. Alexe, R. Scholz, V. Talelaev, and M. Zacharias, Appl. Phys. Lett. 87, 202110 (2005).10. V. Beyer, J. von Borany, and M. Klimenkov, J. Appl. Phys. 101, 94507 (2007).11. SungJin Whang and KiHong Lee DaeGyu, IEDM10-668, 29.7.1–29.7.4 (2010).12. B. Park, S. Choi, H.-R. Lee, K. Cho, and S. Kim, Solid State Communications 143, 550 (2007).13. S. Haas, F. Schneider, C. Himcinschi, V. Klemm, G. Schreiber, J. von Borany, and J. Heitmann, J. Appl. Phys. 113, 44303 (2013).14. H.G. Chew, W.K. Choi, Y.L. Foo, F. Zheng, W.K. Chim, Z.J. Voon, K.C. Seow, E.A. Fitzgerald, and D.M.Y. Lai, Nanotechnology 17, 1964 (2006).15. T.-C. Chang, Materials Today 14, 608 (2011).16. M. Kanoun, A. Souifi, T. Baron, and F. Mazen, Appl. Phys. Lett. 84, 5079 (2004).17. S. Das, R. Aluguri, S. Manna, R. Singha, A. Dhar, L. Pavesi, and S. Ray, Nanoscale Res Lett 7, 143 (2012).18. R. Beyer and J. von Borany, J. Appl. Phys. 105, 64513 (2009).19. T.Z. Lu, M. Alexe, R. Scholz, V. Talelaev, and M. Zacharias, Appl. Phys. Lett. 87, 202110 (2005).20. M. Kanoun, C. Busseret, A. Poncet, A. Souifi, T. Baron, and E. Gautier, Solid-State Electronics 50, 1310 (2006).

SiO2 and other material systems. Below this maximum, the memory window widens as the bias voltage gets larger. The fitted slope of the memory window versus bias voltage characteristics is 0.99 for the single-layer sample, which is very close to the ideal slope of 1. Thus, MIS structures comprising Ge-NCs in amorphous high-k TaZrOx show promising non-volatile memory characteristics.

145

Novel molecular materials for information storage - Synthesis, electronic properties and electrode design M. Mazik 1, E. Weber 1, N. Seidel 1, S. Förster 1, E. Kroke 2, J. Wagler 2, A. Kämpfe 2, J. Kortus 3, T. Hahn 3, S. Liebing 3, Y. Joseph 4, R. Dittrich 4

1 Institute of Organic Chemistry, TU Bergakademie Freiberg

2 Institute of Inorganic Chemistry, TU Bergakademie Freiberg

3 Institute of Theoretical Physics, TU Bergakademie Freiberg

4 Institute of Electronic und Sensor Materials, TU Bergakademie Freiberg

146

AbstractFollowing various strategies which include considerations on electrochemistry and intramolecular isomerism, new compounds based on azulenes, anthraquinones and silicon pyrrolyl complexes were developed for potential application in fields of information storage. The aimed compounds were molecularly designed according to necessary prerequisites for switching functionalities as well as for reliable electrode interaction and intramolecular electron transport. Characterizations regarding structural, spectroscopic and electrochemical properties allow new insights into molecular switching effects which were also studied intensively by means of quantum chemical calculations. Adsorption phenomena of the compounds under investigation on gold surfaces were studied with various sulfur anchoring groups in order to enable electrode contacting of single molecule switches.

Introduction As predicted by G. Moore in 1965 [1], tran-sistor density in microelectronic systems has evolved rapidly in order to meet the constant-ly growing requirements of modern electro-nic devices. Nevertheless, miniaturization and performance enhancement of the so far applied silicon-based technology are about to reach physical and economic limits. There-fore, alternative approaches are necessary to keep pace with the fast development in the field of microelectronics. In this respect, molecular electronics is an area of research that has gained enormously in significance during the last years since it offers good pro-spects to overcome the above difficulties. Here, single or small groups of molecules are used in a device-based structure serving as the fundamental unit for the electronic com-ponent including molecular rectifiers, swit-ches, transistors or wires [2, 3]. In the present article we report on the development of new chemical materials having fair potential to be applied as switchable function and thus for storage of information in molecular electro-nics technology.

Schematic design of a switchable single-mole-cular deviceWithout doubt, one of the main features of a molecular electronic device is the ability to switch. For electronic application bistable switchable molecules allowing the control of the current flow by some external stimulus are of interest [4]. This can be current-, field- or light-induced. Known typical examples fol-lowing this mode of operation are the light-induced photochromic switch consisting of a dithienylethene molecule or the redox-switch based on a 9,10-anthraquinone moiety [5]. In order to use the switchable unit in a practicable functional way, it has to be con-nected by conducting wires (spacers) to span the gap between two electrodes and adhesive groups (anchor groups) contacting the elec-trodes, mostly with a surface modified by gold. This leads to a general composition for a switchable single-molecule device as illustra-ted in Fig. 1.

Fig. 1: Schematic structure of a switchable single molecule between two gold elec-trodes integrated in an electric circuit.

147

In the present case, an azulene moiety, specifically derivatized 9,10-anthraquinones or hypercoordinated dipyrromethene or acylpyrrole silicon complexes serve as potentially switchable units connected to ethynylene wire-type spacers and thiophenol and thiophene terminal anchor groups for contacting to gold electrodes. The following chapters report synthesis and characterization of the new switching units. In addition, evaluation of the measurement results supported by theoretical methods and results of model studies regarding the electric transport and switching operations as well as the assembly behavior of the new compounds are described in more detail.

Synthesis and experimental characterization of the new molecular structures

A switching mechanism may be either of charging/redox or geometrical/conformatio-nal nature giving rise to the different switching units that have been considered in our studies and thus incorporated into respective molecular structures. There are, on the one hand, the redox-active azulene and 9,10-anthraquinone type molecules and, on the other hand, the conformationally influenced dipyrromethene silicon complexes being now discussed regarding their synthesis and properties in this sequence.

Azulene and quinone based target compounds

(a) General. Although azulene (I, Fig. 2a) is known for a long time [6] and has been thoroughly investigated due to its distinct dipolar character (I’ 0.8 Debye) [7], remarkably it has not yet been tested for switching purpose in molecular electronics. As specified in Fig. 2a, azulene may uptake an electron (reduction) to form a radical anion of cyclopentadienyl type (IA) but may also lose an electron (oxidation) to turn into a radical cation of tropylium type (IB), both being aromatic substructures. It is assumed that these charged species differ significantly in conductivity from uncharged azulene. Passing two reduction steps (Fig. 2b), 9,10-anthraquinone (II) can gradually uptake two electrons to be converted via a radical anionic semiquinone structure (IIA) into a 9,10-anthrahydroquinone dianion (IIB). Here, the switching effect bases on the fact that the linearly conjugated dihydroquinone species (IIB) and to a certain degree also the semiquinone (IIA) have a lower electrical resistance compared to the cross-conjugated quinone (II) [8].

Fig. 2: Redox states of azulene I and 9,10-anthraquinone II. 9-10-anthraquinone II (linear and cross-conjugation specified in red) [9,10].

148

Hence, in principle both these functional units, azulene and 9,10-anthraquinone, show three switching states to be potentially exploited. However, it should be possible to read out the information depending on the redox state of azulene and 9,10-anthraquinone including corresponding derivatives as well. Appropriate structures following the construc-tion principle described in Section 2 and schematized in Fig. 1 are specified in Fig. 3, giving representative examples of azulene and

9,10-anthraquinone derived targed molecules (1-7) out of a series of related compounds. They feature structures showing modified connection positions between switching unit and linker moiety (e.g. 1 vs. 2 or 3 vs. 4), different length of the linker element (e.g. 1 vs. 5), varied anchor groups (e.g. 2 vs. 6) or additional substitution of the molecule (e.g. 2 vs. 7). The obtained structural diversity may not only allow different modes of contact to the electrodes but also to investigate effects exerted on the switching behavior.

(b) Synthesis. Preparation of the target compounds require rather elaborate synthetic routes including up to seven individual steps. An example representative of the derivatives of azulene, i.e. preparation of compound 1, is shown in Fig. 4. In the initial step of this synthesis, the desired basic azulene structure is established following a ring anellation method according to T. Nozoe [11]. Subsequently, azulene is gradually halogenated using at first

an electrophilic aromatic substitution and followed by a Sandmeyer analogous reaction. The dihalogenated azulene derivative and a thiophene substituted ethyne, which is separately prepared via another two steps, are then subjected to a palladium-catalyzed cross-coupling reaction (Sonogashira-Hagihara reaction) [12] to yield 1 (78 %) as a brownish-red solid.

Fig. 3: Formula structures of azulene (1,2,5-7) and 9,10-anthraquinone (3,4) derived target compounds.

149

Fig. 4: Synthesis of the target compound 1.

A similar coupling strategy is applied to generate the 9,10-anthraquinone derived target compounds as exemplary illustrated for the synthesis of 3 in Fig. 5. Here, in the initial step 1,4-diamino-9,10-anthraquinone is brominated using a Sandmeyer-type reaction.

(c) Structure analysis. Structures of the target compounds correspond to data obtained from standard analytical methods (NMR, IR, MS) and have also been corroborated in a great number of successfully performed single-crystal X-ray analyses. Those, in particular provide detailed insight with reference

Fig. 5: Synthesis of the target compound 3.

to the conformation and arrangement of the molecules in solid state as well as the influence of connectivity pattern and additional substituents on the molecular and packing structure [13, 14]. A representative example is given with the crystal structure of compound 6 in Fig. 6. The structure is

150

dominated by π···π-stacking interaction of the azulene units along the crystallographic b-axis. Adjacent azulene moieties are mutually shifted. Therefore the typical azulene···azulene interaction [15] between the negatively polarized five-membered ring and

the positively polarized seven-membered ring is observed, being a prevailing contact mode in the series of structures. In agreement with the packing of 6, C-H…π contacts [16] are also found in most of the respective structures.

(d) Electronic behavior. Characterization of the target compounds in terms of their electronic properties is a key point of this work. A first impression of the electronic properties of the individual compounds can be obtained by UV/Vis spectroscopy. For azulene and its derivatives, three adsorption zones are generally perceptible corresponding to π→π* transitions (Fig. 7a) [17]. Actually the S0→S1 transition is forbidden and therefore has a relatively low intensity. Nevertheless, this S0→S1 transition for the parent azulene I, being found in the area of 500 – 700 nm, is the reason of the intense blue color of this compound. The S0→S1 transition in compounds 1 and 2 are shifted hypsochromic (1) and bathochromic (2), respectively, in comparison to structure I. This points to a significant influence of the substituent on the HOMO and LUMO levels. The other two transitions are shifted toward higher wavelength.

Fig. 6: Packing structure of compound 6 viewed along the crystallographic b axis. Dotted lines represent C-H…π contacts. Yellow balls indicate sulfur atoms.

For the anthraquinones, intense π→π* bands in the UV and in the near-UV visible range can be observed. It is noticeable, that the position of the side groups of the two anthraquinones 3 and 4, whose spectra are shown in Fig. 7b, have a significant influence on the electronic structure. Compound 3 shows a single peak (255 nm) while for compound 4 a splitting of the π→π* transition occurs (269, 301, 355 and 376 nm). In addition to the π→π* transitions, the weaker n→π* bands are of relevance as well. These transitions (primarily for the C=O group) extend into the visible region and are therefore responsible for the color of the derivatives of 9,10-anthraquinone.

151

With the help of cyclic voltammetry (-2 V to +2 V), redox processes in solution were investigated. In line with the exposition given in Fig. 2, it is shown that the azulene and 9,10-anthraquinone type compounds possess different switching states, even if not all of them showed reversible behavior (Fig. 8). For compound 1, several oxidation and one reduction process are observed. Polymerisation reactions may be responsible for the irreversible nature of the oxidation processes. For the reduction peak, a back sweep can be observed having an amplitude differing from the forward wave. Taking into

Fig. 7: (a)Absorption spectra of selected azulene derivatives (1,2), the parent azulen I and (b) 9,10-anthraquinone compounds (3,4) in CH2Cl2.

consideration that the difference between the reduction peak and oxidation peak depending on the scan rate dU/dt varies only slightly from around 70 mV (25 mV/s) to 90 mV (200 mV/s), the reduction can be classified as quasi-reversible. The anthraquinon 4 shows two reduction peaks. It is transformed via semiquinone (-1.19 V) into the hydroquinone dianion (-1,74 V). Both the waveform and the difference of the peak potentials suggest a reversible process. The two one-electron transfer steps in the cyclic voltammogram of compound 4 indicate an EE mechanism which is typical for 9,10-anthraquinones [18].

152

(e) Self-assembly property. To investigate the influence of the interface between molecules and metal electrode which is crucial for the electronic properties [19], the assembly behavior of the compounds is examined. With reference to the azulene derivatives, self-assembled monolayers were achieved by placing a gold coated surface into the solution of a selected target compound. They were characterized using the contact angle method and X-ray photoelectron spectroscopy. Two facts indicate that the used anchor groups (tert-butylthiophenyl, thiophene) allow chemisorption of the target molecules on gold substrates. The contact angle between water and the azulene coated substrates, which were rinsed before use, are 15° higher than between water and the pure gold surface. And apart from that, on surfaces, two different

Fig. 8: Recorded cyclic voltammograms of the compounds 1 (25 mV/s) and 4 (50 mV/s).

sulfur species could be identified by XPS measurements indicating a gold attached sulfur-atom (Fig. 9 a) [20].

The contact experiments with the 9,10-anthraquinones lead to slightly different results (increase of 12°-14° compared to water). On the basis of photoelectron spectrum (Fig. 9 b), only one sulfur species can be detected. This leads to the assumption that the molecules lay flat on the gold surface due to the good interaction of the aromatic system with the surface. The poor assembly behavior could also be observed in relation to experiments with gold nanoparticles. Nevertheless, the contact angle for compound 3 indicates a hydrophobization of the surface by coating [21].

153

Silicon dipyrromethene complexes

(a) General. Reports from the literature and ongoing own research have already demonstrated a certain flexibility that goes along with silicon coordination [22], i.e. variable possibilities of the ligand system to bind to the central atom. These examples of intramolecular changes that finally lead to the (co-)existence of coordination isomers encouraged us to investigate new silicon complexes aiming at geometrically switchable molecules. Hence, molecules of this type should not differ in their atomic

Fig. 9: XPS detailed spectra of the sulfur signals from a self-assembled monolayer of the azulene compounds (a) 2 and (b) 3.

composition but in their atomic positioning. Accordingly, we have chosen a bidentate 2-acylpyrrole ligand system initially inspired by dipyrromethenes (Fig. 10) equipped with a nitrogen and an oxygen donor atom, which are known to be good donor atoms with respect to silicon [23]. As a bidentate mono-anionic ligand, this system combines rigidity for pronounced coordination and flexibility for intramolecular switching.

154

(b) Synthesis and structure analysis. Ligands 8a-d readily undergo complexation reactions with SiCl4 which, to our surprise, even proceed in the absence of a sacrificial base. Upon the elimination of HCl, complexes of the type 9a-d are formed (Fig. 11). They could be isolated as crystalline solids (except for 9c)

Fig. 10: Molecular structures of compounds 8a-d in the solid state used as ligands for silicon complexation.

and therefore were characterized by means of X-ray diffraction. As those structures show relative similarity, solely the structure of 9a as a representative example is given in Fig. 12 (left). It shows hexacoordination on the silicon centre with an all-trans arrangement of the donor atoms involved in complexation.

Fig. 11: Synthetic approach towards complexes 9a-d.

155

In the case of 8a, we also observed the forma-tion of compound 9a starting from HSiCl3 and H2SiCl2, respectively. These reactions include unexpected hydrogen deliberation steps. However, compound 9a served exceedingly well as a starting material for further complex syntheses. For example, upon halide exchange with ZnF2, the corresponding silicon-difluoro-complex was formed. As before, in the solid state, this complex (Fig. 12) also

Fig. 12: Molecular structures of compounds 9a (left) and 10 (right) in the solid state determined by X-ray single crystal diffraction (top). Synthetic path from 9a to 10 (bottom).

shows hexacoordination at the silicon atom but in a different arrangement of the donor atoms with respect to each other (N,N-trans). The complex 11 which was obtained via a slightly different synthetic pathway, including the use of a supporting base, revealed in the solid state a different structure (Fig. 13) compared to 9a and 10. Here an O,O-trans arrangement of the donor atoms relating to one another is found.

Fig. 13: Molecular structure of complex 11

156

(c) Experimental characterisation and theoretical study. At this point, we applied quantum chemical methods for a deeper comprehension of the profound energetic circumstances considering different coor-dination modi in various octahedral silicon complexes derived from ligand 8a. Maintaining the donor situation for complexes of the general type L2SiR2 wherein L is the O,N-bidentate mono anion of 8a and R is a monodentate mono anionic substi-tuent, e.g. Cl, F or Ph, there are five possible

arrangements in the coordination sphere as outlined schematically in Fig. 14. Calculations are in full agreement with the structures found for 9a and 10 confirming them as the most stable ones amongst other possible arrangements. However, for 9a the N,N-trans arrangement, which is the donor situation of compound 10, is predicted to be the second stable lying 10.9 (MPW1PW91) and 8.8 kJ/mol (MP2), respectively, higher than the actually found molecular structure.

Fig. 14: General donor arrangements in an octahedral (Si-) complex featuring ligands as described in the text above.

For compound 11, calculations indicate the all-cis arrangement as the theoretically favoured one, which differs from the arrangement found in the solid state. The O,O-trans donor situation present in compound 11 lies according to the calculations slightly (8.8 and 8.4 kJ/mol, respectively) above the predicted all-cis arranged molecule. These differences seem most convenient for promising isomerisation reactions which could lead to molecularly switchable devices [24].

Pursuing further, ligand exchange reaction of 9a by application of trimethylthiocyanate gave rise to the formation of the corresponding thiocyanato-complex 12, which approved to self-assemble on a gold-surface [25]. Fig. 15 shows the determination of contact angles of water and diiodomethane on a clean (A & B) and a coated (C & D) surface by the sessile drop method. It clearly indicates the modification of the Au surface by compound 12, making this compound accessible for electronic contacting.

157

Theoretical modeling of the material systems

The main goal of our theoretical modeling is to achieve a fundamental understanding of relevant aspects to realize an electronic device based on molecular materials. This includes tuning of the intrinsic molecule properties as well as studying the structural, electronic and magnetic properties of the interface between the molecules and potential electrodes.

Calculation of the ground state properties

All first-principle density functional theory (DFT) calculations on molecular structures are carried out using the NRLMOL package which is an all-electron implementation of DFT. NRLMOL combines large Gaussian orbital basis sets, numerically precise variational integration and an analytic solution of Poisson‘s equation in order to accurately determine the self-consistent potentials, secular matrix, total energies and Hellmann-Feynman-Pulay forces [26-30]. The GGA/PBE [31] functional was used to describe the exchange-correlation effects of the electrons. To estimate the stability of the individual molecules, a geometry optimization was carried out. Then, the dependency of various

Fig. 15: Droplets of water and diiodomethane on a clean (A & B) and on a coated (C & D) Au-surface.

electronic characteristics including the nature and position of the linker groups and the influence of additional functional groups on the central ring system were examined. The knowledge of the energy levels and symmetry of the orbitals is necessary to understand the transport properties of these molecular materials properly. The following general conclusions can be drawn from the calculations:

1. With reference to the azulene compounds, the different studied linkers and their locations show that the HOMO-LUMO gap decreases with only a small variation between the values. However, electron affinity and ionization potential show a larger variation (e.g. about 20-30 % in case of electron affinity) which offers interesting possibilities to tune transport parameters. To a certain degree, this is also true for the anthraquinone derivatives. Fig. 16 shows the calculated energy level diagrams together with plots of the HOMO and LUMO orbitals of the azulene 2 and the anthraquinone 3.

158

2. With the help of additional substituents on the central ring system, both the HOMO-LUMO gaps and the position of the Fermi energy can be tailored so that these parameters allow adaptation to a large number of possible contact materials (band gap engineering)[32]. Furthermore, detailed considerations of symmetry and spatial arrangement of the molecular electronic states near the Fermi energy have been carried out. Longer substituents seem to decrease the gap more than shorter substituents. The addition of any substituent does not change the delocalization of the HOMO level across the whole molecule as shown in Fig. 16, while the LUMO in all cases remains localized and similar to the LUMO of azulene, since the LUMO of azulene does not have any significant electron density at the substitution sites.

3. Raman spectroscopy is a very versatile and widely used method to characterize materials. In molecular materials it can identify the fingerprints of single molecules or it can be used to identify a solid material by its characteristic phonon modes. The theoretical determination of Raman spectra from first-principles calculations is also highly desirable, since it can be used to associate Raman bands to the microscopic structure. By calculation, the assignment of measured Raman signals of various target compounds to the vibrational states was possible [13].

Fig. 16: Calculated HOMO (lower molecular orbital plot) and LUMO (upper plot) states and energy levels with respect to vaccum level in comparison for similar azulene and anthraquinone derivatives.

159

Calculation of the transport properties Calculation of the transport properties were done in order to make a possible prediction of U-I curves for devices with an electrode-molecule-electrode arrangement. In a first step, simplified models with geometrically idealized contacts were created. Using the method of Non-Equilibrium-Green-Functions (NEGF) coupled with DFT it becomes possible to calculate the U-I characteristics [33, 34]. The NEGF approach as implemented in QuantumWise [35-37] has been used here. QuantumWise is a commercial implementation of the TranSIESTA-code. It allows one to calculate exactly the electronic properties of molecular junctions by using a localized basis set combined with the pseudopotential approach to achieve computational efficiency suitable for handling large devices at the DFT level of theory.

Fig. 17: Orbitals of compound 4 (left) and 4B (right) near the Fermi level which are responsible for the charge transport through a respective molecular device.

Respective calculations carried out with 9,10-anthraquinone derivatives suggest a switching effect for this type of molecules. The molecular systems, especially regarding the different available types of linker units, were investigated in order to optimize the contact properties to gold and other contact materials (such as copper, silver, cobalt, silicon) [38].

The insight in the electronic structure from first-principles allows for the identification of the relevant transport orbitals (see e.g the plots in Fig. 17 for anthraquinone) and it becomes possible to discuss the influence of different anchor groups on the transport properties [39].

160

The knowledge of the transport orbitals as shown for anthraquinone in Fig. 17 and their interaction with the contacts allows the calculation of transmission function, which contains the probability of an electron passing through the molecular device. In Fig. 18 the calculated transmission function through the model device with 9,10-anthraquinone derivative 4 (black line) and 9,10-anthrahydroquinone derivative 4B

(red line) with junction to the gold electrode is illustrated. The electronic transmission around the Fermi energy of the molecules 4 and 4B differ significantly. This is also reflected on the values of the calculated conductance G in a 0.1 V bias window around the Fermi energy. The computed ratio for G(4B)/G(4) is roughly 4, implicating two readable (on and off) switching states [40].

Fig. 18: Transmission spectra and transport orbitals for anthraquinone 4 and anthrahydroquinone 4B (formula structures of 4 and 4B are given in Fig. 17).

The current through the junction can easily calculated from the transmission function

were the electronic chemical potentials μL/R

are connected to the applied bias voltage via V = (μL−μR)/e (e elementary charge). Therefore, the current is calculated by integrating the self-consistent transmission function within the bias-dependent energy window spanned by the chemical potentials of the contacts. Fig. 19 shows the resulting I-V-curve for anthraquinone and

anthrahydroquinone. The two molecules have similar electronic structure so that one system shows about 200 % higher conductance at an operating voltage of 0.85 V. This could be used to build a molecular switch, because the main difference is due to a different amount of charge in these two molecules. As shown before by using voltammetric measurements, anthraquinone can be reversibly charged. Such charge transfer of two electrons would change the electronic structure of anthraquinone to the electronic structure of anthrahydroquinone, i.e. switching between low and high conductivity states.

161

Spintronics is an emerging new field which couples molecular electronics with magnetism. [41] In our theoretical investigations, we also investigated the influence of magnetic contacts on transport properties. For example, a tunnel magneto-resistance (TMR) of about 11% was calculated for magnetic contacts in

Fig. 19: Current-Voltage curve of anthraquinone 4 and anthrahydroquinone 4B. Anthrahydroquinone 4B can be seen from the point of view of eletronic structure as a charged version of anthraquinone due to charge transfer in a molecular device.

connection with a switching element based on anthraquinone, which would be in principle already sufficient for application as read head in magnetic recording. The TMR in this device originates from the magnetic contacts, however, the molecule is needed to carry the current.

162

Conclusion

In this project, a large number of switchable molecules based on redox-active azulene and 9,10-anthraquinone units as well as geometric variable silicon-complexes could be synthesized. The structures of the target compounds were elucidated and their electronic properties were analyzed.

The HOMO and LUMO energies of the azulene and anthraquinone derivatives were approximately determined from UV/Vis- and cyclic voltammetry data as well as from DFT calculations. From these determined values, the following general conclusions can be drawn: it has been found that variation of the lateral arms, in contrast to the alteration of the link pattern, does not cause major changes in the position of the HOMO and LUMO levels. The HOMO and LUMO levels are in the vicinity of the fermi level of gold. This should have a positive effect for the electrical properties of the interface between electrode and molecule. Our DFT calculations for these molecular systems have led to a better understanding of the underlying electronic structure and transport mechanisms, which enabled us to predict key properties of molecular devices based on azulene and anthraquinone structures.

Furthermore, promising structurally variable silicon complex molecules that are capable of surface interacting with gold electrodes could be isolated. Quantum chemical calculations indicate moderate energy differences between various coordination arrangements of equally composed molecules which, therefore, possess ideal prerequisites for intramolecular switching.

Accordingly, the investigated compounds appear to be suitable as potential switches in the context of molecular electronics. However, more detailed experiments like scanning tunneling microscopy (STM)[41], or the mechanically controlled break junction (MCBJ) technique[42] are required in order to gain insights into the behavior of the compounds when being located between two electrodes. In future, it also remains to be determined how this test assembly affects the stability of the switching states and whether reversible switching is possible under these conditions.

163

References

1. G. E. Moore, Electronics 1965, 38, 114.2. R. M. Metzger (ed.), Unimolecular and Supramolecular Electronics I, Springer, Berlin,

2012.3. M. C. Petty, Molecular Electronics, John Wiley & Sons, New York, 2007.4. R. M. Metzger, J. Mater. Chem. 2008, 18, 4364.5. S. J. van der Molen, P. Liljeroth, J. Phys. Condens. Matter 2010, 22, 133001.6. S. Piesse, C. R. hebd. Séanc. Acad. Sci. Ser. B France 1863, 57, 1016.7. R. D. Nelson, D. R. Lide, A. A. Maryott, NSRDS-NBS 1967, 10, 49.8. E. H. van Dijk, D. J. T. Myles, M. H. van der Veen, J. C. Hummelen, Org. Lett. 2006, 8,

2333. 9. D. Ajloo, B. Yoonesi, A. Soleymanpour, Int. J. Electrochem. Sci. 2010, 5, 459.10. S. Hünig, B. Ort, Liebigs Ann. Chem. 1984, 12, 1959.11. T. Nozoe, K. Takase, M. Kato, T. Nogi, Tetrahedron 1971, 27, 6023.12. K. Sonogashira, Y. Tohda, N. Hagihara, Tetrahedron Lett. 1975, 50, 4467.13. S. Förster, W. Seichter, R. Kuhnert, E. Weber, J. Mol. Struct. 2014, 1075, 63.14. N. Seidel, K. Sandig, W. Seichter, E. Weber, Z. Krist. 2013, 228, 669.15. Y. A. Milkheev, L. N. Gunsera, Y. A. Ershov, Russ. J. Phys. Chem. 2012, 86, 85.16. M. Nishio, Y. Umezawa, K. Honda, S. Tsuboyama, H. Suezawa, CrystEngComm. 2009, 11,

1757.17. E. Heilbronner, Non-Benzenoid, Aromatic Compounds, Interscience Publishers, New

York, 1959, 171.18. G. A. Gruver, T. Kuwana, J. Electroanal. Chem. Inter. Electrochem. 1972, 36, 85.19. M. Knupfer, H. Peisert, Electronic Properties of Interfaces between Model Organic

Semiconductors and Metals in Physics of Organic Semiconductors (ed. W. Brütting), Wiley-VCH, Weinheim, Germany, 2005.

20. N. Seidel, Redoxaktive chinoide Molekülstrukturen als potenzielle Einzelmolekülschalter für die molekulare Elektronik – Entwurf, Synthese und strukturelle Charakterisierung, PhD Thesis, TU Bergakademie Freiberg, 2014.

21. S. Förster, Synthese von modular aufgebauten Azulenverbindungen und deren Charakterisierung als potentielle Schaltelemente für die molekulare Elektronik, PhD Thesis, TU Bergakademie Freiberg, 2014.

22. O. Seiler, R. Bertermann, N. Buggisch, C. Burschka, M. Penka, D. Tebbe, R. Tacke, Z. Anorg. Allg. Chem. 2003, 629, 1403.

23. a) I. Kalikhman, S. Krivonos, L. Lameyer, D. Stalke, D. Kost, Organometallics 2001, 20, 1053; b) J. Wagler, U. Boehme, E. Brendler, G. Roewer, Organometallics 2005, 24, 1348; c) A. Kämpfe, E. Kroke, J. Wagler, Eur. J. Inorg. Chem. 2009, 1027.

24. A. Kämpfe, E. Brendler, E. Kroke, J. Wagler, Chem. Eur. J. 2014, 20, 9409.25. A. Kämpfe, E. Kroke, J. Wagler, Organometallics, 2014, 33, 112. 26. K. A. Jackson and M. R. Pederson, Phys. Rev. B, 1990, 42, 3276. 27. M. R. Pederson and K. A. Jackson, Phys. Rev. B, 1990, 41, 7453.28. M. R. Pederson and K. A. Jackson, Phys. Rev. B, 1991, 43, 7312.29. M.R. Pederson, D. V. Porezag, J. Kortus, and D. C. Patton. Physica.status solidi (b), 2000,

217, 173.30. D. Porezag and M. R. Pederson, Phys. Rev. A, 1990, 60, 2840..31. J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Let. 1996, 77, 3865.

164

32. S.Liebing, Electronic Properties of azulene derivatives - a DFT-study, Bachelor Thesis, TU Bergakademie Freiberg, 2010.

33. M. Strange, C. Roostard, Phys. Rev. B 2011, 83, 115108.34. S. Datta, Electronic Transport in Mesoscopic Systems, Cambridge Studies, 2009.35. Atomistix ToolKit version 12.2.1, Quantum Wise A/S (www. quantumwise.com).36. M. Brandbyge, J.-L. Mozos, P. Ordejon, J. Taylor and K. Stokbro, Phys. Rev. B: Condens.

Matter Mater. Phys. 2002, 65, 165401.37. J. M. Soler, E. Artacho, J. D. Gale, A. Garcia, J. Junquera, P. Ordejon and D. Sanchez-

Portal, J. Phys.: Condens. Matter 2002, 14, 2745.38. C. Lurz, Berechnung der Transporteigenschaften von Benzendithiol gekoppelt an

verschiedene metallische Elektroden. Bachelor Thesis TU Bergakademie Freiberg, 2011.39. S.Liebing, Quantum transport through molecules: influence of contact groups, Master

Thesis, TU Bergakademie Freiberg, 2013.40. N. Seidel, T. Hahn, S. Liebing, W. Seichter, J. Kortus, E. Weber, New J. Chem. 2013, 37,

601.41. G. Binnig, H. Rohrer, Ch Gerber, E. Weibel, Appl. Phys. Lett., 1982, 40, 178.42. M. A. Reed, C. Zhou, C. J. Muller, T. P. Burgin, J. M. Tour, Science, 1997, 278, 252.

165

solution b 1 in CH3CN

(1 mM)

solution a CH3CN / H2O

(? i = 0.5)

solution c [Cu(NH3)4]SO4

in H2O/NH3 (1 mM)

t = 0 s t = 120 s t = 180 s t = 240 s

t = 300 s t = 360 s t = 420 s t = 480 s

Development of an electrically con-ductive coordination polymer based transducer for sensor applications

M. Günthel 1, J. Hübscher 2, F. Katzsch 2, R. Dittrich 3, Y. Joseph 3, E. Weber 2 and F. Mertens 1

1 Institute of Physical Chemistry, TU Bergakademie Freiberg, Leipziger Str. 29, 09599 Freiberg/Sachsen 2 Institute of Organic Chemistry, TU Bergakademie Freiberg, Leipziger Str. 29, 09599 Freiberg/Sachsen 3 Institute of Electronic and Sensor Materials, TU Bergakademie Freiberg, Leipziger Str. 29, 09599 Freiberg/Sachsen

166

AbstractSubject of this article is the development of an electrically conductive coordination polymer and its integration into an electronic system that can be used as a resistive gas sensor. Appro-priately configured linker molecule based on o-acylphenol (1) as well as a special ethynylene bridged adhesion molecule featuring terminal thiol functions (2) were synthesized by using a convergent synthesis path. Compound 2 was used to form a self-assembled monolayer on a gold surface which was characterized by XPS. Performing a thin layer deposition, two gold electrodes, prior functionalized with 2, were then bridged with the organometallic compound Cu(1) that is composed of the linker molecule 1 and copper(II). The coating procedures were monitored by means of a quartz crystal sensor and the quality of each layer was inspected by XPS and SEM. Eventually, the electronic properties of the sensitive layer obtained were inves-tigated. The response to water as a potential analyte reveals a sufficiently large change in the electrical resistance that the layer may be used in sensing application.

Introduction Within the cooperation of the joint project “Functional structure design of new high-per-formance materials via Atomic Design and Defect Engineering” (ADDE) special orga-no-metallic hybrid materials were developed to be used as sensitive media for resistive gas sensors. The schematic construction of such a sensor is depicted in Figure 1.

The detection of gaseous compounds is of great importance in many research areas, e.g. for the control of pollutant emissions, in sa-fety engineering, and in process monitoring [1]. Depending on the requirements which must be met by the sensor, different measu-ring principles can be used. The interest in an analyte detection principle based on resistive measurement is rooted in the ease and the

SAM - molecule

linker molecule

metal center

electrically conductivecoordination polymer

interdigitatedgold structure

Figure 1: Schematic structure of the proposed gas sensor based on an electrically conductive coordination polymer.

167

robustness of the method and the expected fast response of the electrical conductivity of the sensing material. Especially metal and semi-metal oxides are established as sensiti-ve materials in this context [2]. However, the use of metal-organic composite networks, so-called metal-organic frame works (MOFs), could result in significant advantages compa-red to the conventional oxide materials. For one, this compound class is usually described by high internal surface areas, which opens up the possibility of optimizing the process of gas adsorption [3]. In addition, MOFs are charac-terized by a modular structure, which allows to vary the components they are comprised of in ways to intentionally influence and predict some of the properties of these coordination networks. For example, the exact adjustment of the pore volume is possible by the inten-tional variation of the organic components used [4]. In respect to sensor systems, the adjustment of the pore size can be used to implement size selection of molecules in the sensing layer. Regarding the construction of specific pore properties, one can attempt to generate coordinative defects purposefully in order to create attachment points for poten-tial analytes. In addition, the high degree of variability in yet related crystal structures of these composite materials was shown to be of value for the fabrication of specialized MOF layers using step-wise deposition methods [5, 6]. Another key point that arises from the modular design principle for the development of MOFs is the ability to control the materials properties by intentional design of the organic linker molecules, a feature that can be exploi-ted to implement some electric conductivity, at least of semiconductive nature, in the coor-dination networks. In this respect, the linker molecules to be used should possess at least two functionalities, i.e. being capable of co-ordinating metal centers and providing the linkage between these centers via a conjuga-ted π-electron system into which the metal centers are integrated by bonds of π- or π-like character. The attachment of such metal-orga-nic composites to the electrodes can then be implemented by so-called self-assembled mo-nolayers (SAMs), also being electrically con-

ductive. For this application, a second type of molecule possessing an adhesive group having high affinity to the electrode surface must be synthesized.

Results and Discussion

Synthesis of organic building blocks

The molecules to be synthesized in the con-text of this work must meet the following re-quirements:

1. Possessing the ability to self-assemble on a gold surface (SAM formation)

2. Being suitable for coordinating and bridging certain metal centers

3. Facilitating electrical conductivity

The ability to self-assembly on gold first re-quires a gold-affine functionality. Functional units which contain sulfur are widely used in this respect. In addition to the affinity to the gold substrate, the molecules must also be capable of interacting with neighboring mole-cules in order to contribute to the stabilizati-on of the SAM. One approach to generate this functionality is the use of α-mercapto functio-nalized azines (e.g. pyrimidines), since they open up the possibility to establish π inter-actions (CH···π, π···π, etc.) with neighboring molecules. Electrical conductivity may result from the completely conjugated π-electron system. Depending on the type of metal cen-ters, there are numerous ways of implemen-tation. The option selected in this work, is based on the use of β-diketones in the enol form. For the reasons already mentioned, it is also advantageous to embed the correspon-ding functional group in an aromatic system, which was put into practice by the use of o-acylphenols, which has the additional effect that the keto-enol tautomerism of the two keto groups is mostly prevented and therefore no energy for the transformation of the mole-cule to the required tautomer must be spent. Still missing is a suitable spacer unit to con-nect at least two of these functional aromatic structures with each other, in an electrically conductive way. The ethynylene group is well

168

suited to provide this function. Firstly, it pro-vides a rigid backbone which does not allow displacement of the aromatic subunits. Se-condly, it features two π-orbitals perpendicu-lar to each other so that the system becomes rotationally invariant around the axis formed by this group and thus ensuring the conjuga-tion between the two aromatic heads groups. The resulting building blocks applicable as linker and adhesion molecules are shown in Scheme 1.

Scheme 1: Desired molecules for the development of both the coordination polymers Cu(1) and the SAM structures.

The bridged β-diketone 1 has been synthe-sized performing a multi-step reaction se-quence using 4-bromophenol as base material. As shown in Scheme 2, first 4-bromophenol was reacted with acetic anhydride to yield 7 followed by a Fries reaction [7] resulting in 5-bromo-2-hydroxyacetophenone (6).

Scheme 2: Systematic build-up strategy for the synthesis of linker molecule 1.

169

The following cross-coupling reaction was carried out with trimethylsilylacetylene (TMSA) under Sonogashira-Hagihara cou-pling conditions [8] to give 5. The cleav-age of the protecting group was then car-ried out using potassium hydroxide ending

up with 5-ethynyl-2-hydroxyacetophenone (4). Additional cross-coupling of the bro-mide 6 with 4 yielded the target molecule 1.

Scheme 3: Synthetic scheme for the creation of SAM molecule 2.

The synthesis of the SAM molecule 2 turned out to be a little more complex (see Scheme 3).

As mentioned above, the respective mole-cules shall be composed of specially linked six membered aza-heterocycles. In this respect, pyrimidine is the hetero-aromatic of choice. Initially, the hydroxy functionalized mole-cule 13 has to be synthesized. To do so, it is necessary to construct the heterocycle by a condensation reaction before the iodination procedure can be carried out [9]. In the step to be followed, the hydroxy function in 13 is substituted using phosphorous oxychloride yielding 12. This molecule possessing two dif-ferent halogene substituents can now be used to perform a nucleophilic substitution with

sodium tert-butylmercaptide to form 11. To insert the ethynylene spacer element a cou-pling procedure was used which is similar to the one described for the linker molecule 1. Initially, the palladium catalyzed cross cou-pling is performed using TMSA yielding 10. This reaction is then followed by a cleavage of the protecting group under alkaline condi-tions to give 9 while the final step is another coupling reaction with the iodine containing molecule 11. Thus, the symmetric ethynylene bridged azine 8 is created. To form the desired adhesion molecule 2, a final reaction step is required to remove the terminal tert-butyl substituents. This kind of cleavage has been conducted using boron tribromide as lewis acid.

170

5 10 15 20 25 30 35 40 45 50

Inte

nsity

/ a.

u.

2θ in °

1a

Cu(1a)

[Cu(NH3)4]SO4

O

OO

O

Cu

n

Coordination polymers

The ethynylene bridged linker molecule 1 was utilized to build the coordination poly-mer Cu(1) via a solvothermal synthesis at 100 °C analog to Pfeiffer’s method [10] with [Cu(NH3)4]SO4 as precursor complex. A 1:1 ratio between copper(II) ions and linker mol-ecules in these material was concluded from elemental analysis data. By means of powder X-ray diffraction (PXRD) the progress of the

Cu(1) formation was monitored (see Fig-ure 2). The figure shows the PXRD data of the synthesized metal-organic compound in comparison to the diffractograms of the crys-tallized reactants.

Figure 2: PXRD diffractograms of 1, Cu(1), and the precursor complex [Cu(NH3)4]SO4.

The completion of the conversion of the start-ing materials is marked by a loss of the cor-responding reflections. The XRD spectra of Cu(1) shows both broad and sharp reflections which may be explained by the presence of two different solid phases. Reflections of the reactants are not found after the event of co-ordinative polymerization.

With the aid of infrared spectroscopy, one can also track the reaction progress based on the changes in the characteristic vibrational bands. Therefore, the IR spectra of the pure linker molecule 1 was recorded and compa-red with those of the copper containing com-pound Cu(1). The C=O stretching vibration of the acetyl groups is usually located within the range between 1700 and 1600 cm-1. This band can be observed in the IR spectra of the pure linker molecule 1 (see Figure 3).

171

Figure 3: Infrared spectra of the metal-organic compound Cu(1) in comparison with the one of the linker molecule 1.

In the corresponding metal complex Cu(1) splitting and shifting effects of the stretching bands to higher energies are observed. A pos-sible interpretation for this observation may be that the phenolic C-O single bond may have obtained partial double bond character during the complexation process, which also could explain the emergence of two C=O va-lence vibrations. However, the loss of the ori-ginal vibration band is definitely an indication for the conversion of the starting material.

Functionalization of the Transducers Surface

After synthesizing the functional units for the gas sensor system, i.e. specially designed surfactant molecules and coordination po-lymers promising useful properties, the next step is the integration of these items into a transducer. Inter-digital gold electrodes coa-ted on an insulating silicon dioxide substrate that was initially processed by a lithographic procedure [11] serve as the base for the trans-ducers to be fabricated. Ahead of the coating procedures, the electrode surface of the whole transducer has been treated with an etching solution (5 minutes) in order to functionalize the silica parts with terminal hydroxy groups [12] .

172

Figure 4: XP spectra of the transducers gold surface which was initially treated with a 0.1 mm solution of 2 in acetoni-trile for 24 h (left) and then exposed to the vapor of 3 for 12 h during the CVD process at the equilibrium vapor pressure of 3 at T = 100 °C (right).

Subsequently, a treatment of the transdu-cer with a 0.1 mm solution of 2 dissolved in acetonitrile (purity 99.99 %, < 10 ppm wa-ter) was applied for 12 hours. Using pho-toelectron spectroscopy the success of the coating process was verified (Figure 4, left).

Table 1 shows the experimental data of the respective XPS measurements (the signals of the gold substrate were used for the binding energy calibration) [13].

173

signal species binding energy[eV]

FWHM[eV]

intensity [at%](theoret. value)

Au4f7/2 Au0 84.00 0.6468 –

S2p3/2 S–Au 162.20 1.0308 4.5 (6)

S2p3/2 S–R 163.62 0.8479 6.0 (6)

S2s S–Au 226.48 1.7143 –

S2s S–R 228.14 1.8787 –

C1s C–C 284.86 1.0011 28.3 (25)

C1s C–N 285.69 1.0801 28.3 (25)

C1s C–NNS 286.71 1.4593 14.1 (13)

N1s C–N–C 399.12 0.9201 18.8 (25)

Table 1: XPS data of 2 deposited as SAM on the gold surface.

Indicated by two doublets for the S2p elec-trons, the existence of two different sulfur signals can be located in the XP spectra of the gold surface which was coated with 2. This observation can be explained by the formation of a strong Au–S bond between the surface and one of the two sulfur atoms of 2. The corresponding signal is detected at lower binding energy than the one attributed to the hydrogen bound sulfur atom [14]. A ratio of 1:1 for the two sulfur species should be observed since each molecule contains ex-actly two sulfur atoms. Therefore, the 1:1.35 ratio (AuS / SR) may be explained by a cer-tain amount (15-20 %) of 2 that may also be attached to the surface or even bound to the SAM. The results show that 2 can be used to

form self-assembled monolayers on gold sur-faces. However, due to its terminal functional-ities it is not useful to attempt building SAMs on silicon dioxide surfaces using 2 (Figure 5, left). As one can see, the surface contains neither sulfur nor nitrogen, which indicates the absence of 2. To achieve a regular coating of the surface between the gold electrodes, 3-mercaptopropylethoxysilane (3) was used to functionalize the OH terminated silicon dioxide surface with thiol groups during a CVD process. The XP spectra on the right hand side in Figure 4 show the characteristics of the electrode surfaces after the CVD, which indicate that no damage happens to the once formed SAMs during this procedure.

174

Figure 5: XP spectra of the transducers SiO2 surface which was initially treated with a 0.1 mm solution of 2 in ace-tonitrile for 24 h (left) and then exposed to the vapor of 3 for 12 h during the CVD process at the equilibrium vapor pressure of 3 at T = 100 °C (right).

In contrast, Figure 5 (right) illustrates a suc-cessful coating of the silicon dioxide surface with 3. The most obvious evidence of the binding event is the occurrence of a signal at 228.17 eV (FWHM 2.0229 eV) which is in the

characteristic range for a 2s signal of a sul-fur atom in 3. Table 2 contains the XPS data concerning the CVD process (the Ar2p3/2 and Si2s signals of the substrate were used for cali-bration) [13, 15].

175

Table 2: XPS data of 3 deposited as SAM on the SiO2 surface.

signal speciesbinding energy

[eV]FWHM

[eV]intensity [at%](theoret. value)

Ar2p3/2 Ar0 241.79 0.9715 –

Si2s SiO2 153.73 2.8754 –

S2s C–S–H 228.17 2.0229 4.0 (10)

C1s C–C 284.54 1.0335 10.7 (10)

C1s C–Si 285.13 1.2371 42.6 (40)

C1s C–O, C–S 285.70 2.4741 42.7 (40)

Au

SiO2

= Ar-SH

= Alk-SH= Si-O-Si

= Ar-S-Au

Si

Au

Ti Ti

The resulting system (Figure 6) represents a possible basis for the controlled assembly of coordination polymers required for the objec-tive to build resistive gas sensors.

Figure 6: Scheme of the functionalized transducer surface after the coating with the SAMs. The gold electrode is covered with layer of 2 and the silicon dioxide substrate with a layer of 3.

176

c

a

bperistalticpump

Deposition of Cu(1a) on the functionalized transducer surface

In order to achieve a continuous coverage of the type of surfaces discussed with the co-ordination polymer Cu(1), a layer-by-layer technique was performed in a customized flow cell growing the coordination network in one direction by alternating the deposition of the copper ions and the linker molecules. The deposition was monitored using a mi-crobalance quartz crystal, which was put in series in a flow cell with the transducer cell

(Figure 7). An ultra-high vacuum sputter-ing process was used to clean the surface of the quartz crystal sensors. A 1 mm solution of 1 in acetonitrile and a 1 mm solution of tetraamminecopper(II) sulfate were alternate-ly pumped through the flow cell using a peri-staltic pump. Every reaction step was followed by a purging sequence utilizing a mixture of acetonitrile and water (1:1).

Figure 7: Setup for the automated coating procedure. The different solutions were pumped in cycles of 480 s through the whole system (each solution for 120 s): rinsing (a), [Cu(NH3)4]SO4 (1 mm in H2O/NH3) (b) and 1 (1 mm in CH3CN) (c). A represents the transducer cell and B the cell imbedding the quartz crystal sensor. Due to geometrical differences both flow cells vary in their flow characteristics.

The software COMSOL Multiphysics was ap-plied to simulate the fluid dynamics during the event of coating. This investigation be-came necessary because a shift of the quartz crystal sensor frequency to higher or lower values depending on the solution used was measured. Assuming that effect this is caused by the different fluidic properties of the solu-tions a, b, and c, calculations were performed using finite element analysis to understand the process better and to find improved cell geometries. A graph displaying the spatial concentration distributions of the reaction solutions at various stages of the coating pro-cess is shown in Figure 8.

By means of the two different quartz crys-tal sensor types, it was possible to study the growth of the coordination polymer Cu(1) on both the gold and the SiO2 surface (Figure 9). A tendency of the layer thickness to grow twice as fast on the SiO2 substrate as on the gold one was indicated by the respective dif-ferences in the mass increase on the sensors.

177

solution b 1 in CH3CN

(1 mM)

solution a CH3CN / H2O

(? i = 0.5)

solution c [Cu(NH3)4]SO4

in H2O/NH3 (1 mM)

t = 0 s t = 120 s t = 180 s t = 240 s

t = 300 s t = 360 s t = 420 s t = 480 s

Figure 8: COMSOL simulation of the concentration profile during the coating procedure conducted in the transducer containing flow cell. The different colors illustrate the concentration of the reactants on the transducer surface during one cycle of the coating procedure. In the middle of the flow cell, where the flow characteristics can be considered to be laminar, Cu(1) is grown on the inter-digital structure.

Figure 9: The absolute change of the vibration frequency of a Cu(1) coated quartz sensor is shown for the two differ-ent substrates within 24 coating cycles by using the corresponding solutions a–c. The solutions were changed every 2 minutes, a process that can directly be observed in the signal of the quartz crystal sensors. The black curve shows the result for the deposition of Cu(1) on a QSX 301 type sensor (Au) previously coated with a SAM of 2. The grey curve shows the respective result for the QSX 303 type sensor (SiO2) previously coated with a SAM of 3.

178

a c

b d

e

f

2 µm 2 µm 2 µm

2 µm 2 µm 2 µm

Figure 10 shows the results of different scan-ning electron microscopy (SEM) measure-ments that were carried out to investigate the transducers surface before and after deposi-tion of Cu(1). A comparison of the uncoated and the homogenously SAM coated substrates emphasizes the importance of a SAM previ-

ously grown on the SiO2 surface for the suc-cessful coating of the respective surface with Cu(1). Obviously, the presence of a suitable SAM on SiO2 was essential in this experiment to grow the coordination polymer across the gap between the electrodes and onto the SiO2 substrate.

Figure 10: SEM images of the pure gold surface (a), the pure SiO2 surface (b), the coordination polymer Cu(1) coated onto the gold surface (c) and onto the SiO2 surface (d) after the corresponding SAMs were formed on these substrates. The formation of the two SAMs is essential because a missing SAM on the SiO2 substrate leads to the absence of Cu(1) on the surface (e). The desired result of the coating procedure is shown in image (f).

The success of the coating of the transducer surface with Cu(1) was verified by photo-electron spectroscopy (Figure 11). The ob-tained spectra are in good agreement with those of the bulk material [16]. The cor-responding XPS data are listed in Table 3. The prominent triplet structure of the Cu2p3/2 signals are caused by hybridization effects [16-18]. The separation of the Cu2p photoelec-tron lines was determined to be ∆ = 19.9 eV. Furthermore, a strong hybridization between the molecular orbitals of metal and linker is

indicated by π → π* charge transfer transitions observable in the spectral regions specific for C1s and O1s electrons. As a consequence, the extension of the conjugated π-electron system from one linker molecule beyond the metal center to another linker molecule can be pos-tulated [19,20]. The electronic communica-tion between the aromatic systems of different linker molecules of Cu(1) across the copper ions will also be a key element for the occur-rence of a significant electric conductivity in the material.

179

Figure 11: XP spectra of Cu(1) deposited on a transducer that was initially treated with a 0.1 mM solution of 2 in acetonitrile for 24 h and then exposed to the vapor of 3 for 12 h during the CVD process at the equilibrium vapor pressure of 3 at T = 100 °C.

signal speciesbinding energy

[eV]FWHM

[eV]intensity [at%](theoret. value)

Cu2p3/2 CuII 934.52 1.2737 2.1 (4)

O1s Cu–O–C 531.69 1.8917 18.7 (18)

C1s C–C aromat. 284.60 1.2577 17.6 (17)

C1s C–C aliphat. 285.88 1.2577 8.8 (9)

C1s C–O, C=O 286.68 2.8290 52.8 (52)

Table 3: XPS data of Cu(1) coated as thin layer on a transducer.

180

Electric characterization of the sensoric layer

The major goal of this project was the devel-opment of a sensitive layer applicable in gas sensors. Therefore, all of the coating experi-ments aimed at the connection between the potential gas sensitive material and an elec-trode surface. Resistive measurements were subsequently carried out to characterize the developed system under different conditions. The electrical conductivity of the material was determined at atmospheres containing differ-ent amounts of water in order to examine the dependency of this property to water adsorp-tion (see Figure 12). A significant electrical conductivity that clearly depends on the water content of the surrounding environment can be credited to the layer. In order to eliminate the presumption of water being necessary to facilitate electrical conductivity, the measure-ments were also carried out in vacuo using a carefully annealed sample.

The current-voltage relationship determined for the system shows nonlinear behavior.

Hence, we only used the linear low voltage region for the determination of the electri-cal resistance, which resulted in a value of R = 70 MΩ from which the specific resis-tance of ρ ≤ 3.7·105 Ωm is derived taking into account an effective electrode area of A = 5·10-8 m². The calculated specific resis-tance, which corresponds to a specific con-ductivity of σ ≥ 2.7·10-8 Scm1, indicates se-miconducting behavior. These values match those of other semiconducting coordination polymers that were already reported in the literature [21]. Beside the documentation of a significant conductivity, Figure 12 also em-phasizes the sensitivity of the coated transdu-cers towards water. An explanation for this capacity may arise from a combination of the porosity indicated by a BET surface area of ABET = 59 m²g-1 and the presumption that the copper(II) ions in Cu(1) feature open metal sites which allow a chemical bonding of water molecules through the oxygen atoms.

Figure 12: Current-voltage characteristics of Cu(1) coated onto a transducer covered with SAMs of 2 and 3 formed on the gold and the silicon dioxide substrates, respectively. All measurements were performed at T = 25 °C, the pressure in the vacuum chamber was p ≤ 10-6 mbar. Water was removed by keeping the sample at T = 100 °C for 12 h. The gap between the electrodes is 10 µm and the height of the gold electrodes is 100 nm.

181

Conclusion The subject of this article is the development of specific organic and metal-organic building blocks and their use in the design of a sensiti-ve layer for gas sensors. To achieve this goal, special o-acylphenol based linker molecules and α-mercaptoazin based adhesion molecu-les were initially synthesized in good yields by multi-step reaction sequences. Potential electrical conductivity of these molecules was addressed by connecting the terminal func-tionalities via ethynyl units to gain fully con-jugated π-electron systems.

Via a solvothermal reaction between the lin-ker molecule 1 and [Cu(NH3)4]SO4, coordi-nation polymer Cu(1) was synthesized. The reaction product was characterized by PXRD and by means of IR spectroscopy.

After the development of the different func-tional materials, these components were inte-grated into a sensor assembly. To achieve this goal, a special transducer being composed of interdigitated gold electrodes deposited on SiO2 was provided with different SAM struc-tures. The transducer surfaces were selectively coated with SAMs, the gold one with a SAM consisting of molecule 2 and the SiO2 one with a SAM consisting of 3-mercaptopropy-lethoxysilane (3). The success of the coating processes was verified using X-ray photoelec-tron spectroscopy (XPS).

After the organic surface-functionalization of the transducer the coordination polymer Cu(1) has successively been deposited in terms of a layer-by-layer growth procedu-re onto the functionalized electrode system. For this purpose, the solutions of the starting materials (linker molecule 1 and [Cu(NH3)4]SO4) were alternately passed over the transdu-cer using a peristaltic pump. The flow dyna-mics was simulated for better understanding by utilizing the COMSOL software and coa-ting steps were reviewed with a quartz micro-balance, which was connected in series to the

transducer cell during the coating procedu-re. Subsequently, the success in bridging the electrodes with the coordination polymer was demonstrated by XPS and scanning electron microscopy (SEM).

Finally, the electric properties of the potenti-al sensor system were examined. In this con-text, the current-voltage characteristics were measured at different water partial pressures. A specific conductivity of σ ≥ 2.7·10-8 Scm-1 determined in vacuo, was directly measured after a thermal treatment that was conducted to remove any coordinated water. In addition, certain sensitivity towards water exposure was then observed, which may be due to the nanocrystalline nature of the coordination polymer and the free coordination sites of the incorporated metal ions.

With these two properties and in conjunction with a sufficient degree of electrical conducti-vity, the developed system provides the neces-sary properties for the application as a sensor. However, detailed studies on the sensory acti-vity in general and in respect to various analy-tes still need to be conducted.

Experimental procedures

The exact procedures to form the coordina-tion polymer Cu(1) as well as the synthesis of the ethynylene bridged β-diketone 1 and the respective precursor molecules 4, 5, 6 and 7 can be found in [16]. For the synthesis of the heterocyclic compounds 8, 9, 10, 11, 12, 13 and 14 see [22]. The synthetic procedure for the adhesive molecule 2 is described in [11].

182

Literature

1. D. K. Aswal, S. K. Gupta, Science and technology of chemiresistor gas sensors: Nova Sci-ence Publishers, New York, 2007.

2. C. Pijolat in Chemical Sensors and Biosensors (Hrsg.: R. Lalauze), John Wiley & Sons, Inc, Hoboken, NJ, USA, 2012, 93.

3. H. K. Chae, D. Y. Siberio-Pérez, J. Kim, Y. Go, M. Eddaoudi, A. J. Matzger, M. O’Keeffe, O. M. Yaghi, Nature 2004, 427, 523.

4. L. Liu, K. Konstas, M. R. Hill, S. G. Telfer, J. Am. Chem. Soc. 2013, 135, 17731.5. O. Shekhah, H. Wang, S. Kowarik, F. Schreiber, M. Paulus, M. Tolan, C. Sternemann, F.

Evers, D. Zacher, R. A. Fischer, C. Woell, J. Am. Chem. Soc. 2007, 129, 15118.6. D. Zacher, O. Shekhah, C. Woell, R. A. Fischer, Chem. Soc. Rev. 2009, 38, 1418.7. S. G. Davies, B. E. Mobbs, J. Chem. Soc., Perkin Trans. 1 1987, 2597 – 2604.8. K. Sonogashira, N. Hagihara, Tetrahedron Lett. 1975, 50, 4467 – 4470.9. C. Pérez-Balado, D. Ormerod, W. Aelterman, N. Mertens, Org. Process Res. Dev. 2007,

11, 237.10. P. Pfeiffer, S. Golther, O. Angern, Ber. Dtsch. Chem. Ges. 1927, 60, 305 – 313.11. M. Günthel, J. Hübscher, R. Dittrich, E. Weber, Y. Joseph, F. Mertens, J. Polym. Sci., Part

B: Polym. Phys. 2014 (submitted).12. P. Uhlmann, N. Houbenov, N. Brenner, K. Grundke, S. Burkert, M. Stamm, Langmuir

2007, 23, 57.13. B. V. Crist, Surf. Sci. Spectra 1992, 1, 376.14. C. Vericat, M. E. Vela, R. C. Salvarezza, Phys. Chem. Chem. Phys. 2005, 7, 3258.15. T. Bekkay, E. Sacher, A. Yelon, Surf. Sci. 1989, 217, L377.16. J. Hübscher, M. Günthel, R. Rosin, W. Seichter, F. Mertens, E. Weber, Z. Naturforsch.

B2013, 68, 214.17. K. Okada, J. Kawai, Kotani. A., Phys. Rev. B 1993, 48, 10733.18. J. S. H. Q. Perera, D. C. Frost, C. A. McDowell, J. Chem. Phys. 1980, 72, 5151.19. D. D. Sarma, A. Chainani, Phys. Rev. B 1990, 41, 6688.20. K. Okada, A. Kotani, J. Phys. Soc. Jpn. 2006, 75, 123703.21. Y. K. Kobayashi, B. Jacobs, M. D. Allendorf, J. R. Long, Chem. Mater. 2010, 22, 4120.22. J. Hübscher, W. Seichter, T. Gruber, J. Kortus, E. Weber, J. Heterocycl. Chem. 2014 (pub-

lished online, DOI: 10.1002_jhet.2122).

183

On the thermal stability of na-noscaled Cr/ta-C multilayers

U. Ratayski 1, Ch. Schimpf 1, T. Schucknecht 1, U. Mühle 1,2,C. Baehtz 3, M. Leonhardt 4, H.-J. Scheibe 4, D. Rafaja 1

1 Institute of Materials Science, TU Bergakademie Freiberg, Gustav-Zeuner Str. 5, 09599 Freiberg, Germany2 Fraunhofer Institute for Nondestructive Testing IZFP, Maria-Reiche-Straße 2, 01109 Dresden, Germany3 Helmholz-Zentrum Dresden Rossendorf, Bautzener Landstr. 400, 01328 Dresden, Germany4 Fraunhofer Institute for Materials and Beam Technology IWS, Winterbergstr. 28, 01277 Dresden, Germany

184

185

AbstractThe influence of the initial microstructure in the as-deposited Cr/ta-C multilayer coatings on the thermal stability of the tetrahedrally bonded carbon (ta-C) layers and on the changes of the multilayer morphology with increasing annealing temperature was investigated. The multilayer structure was prepared by the combined DC arc/laser arc technology. The DC arc evaporation was employed for the deposition of the Cr layers, the laser arc technology for the deposition of the ta-C layers. The laser arc process was performed at three different energies of the carbon ions (Ec) in the range between ~25 eV and 500 eV in order to modify the degree of intermix-ing of species at the interfaces. In situ and ex situ small-angle and wide-angle X-ray scattering (SAXS, WAXS) and high resolution transmission electron microscopy (HRTEM) were used for the characterization of the microstructure evolution of the Cr/ta-C multilayer coatings. The in situ experiments were performed after each annealing step up to annealing temperature of 600 °C. SAXS revealed the changes in the density and thickness of the individual layers in the multilayer stack and in the interface morphology and correlation of the interface roughness. The density of the ta-C layers decreased with increasing annealing temperature, which can be explained by the graphitization of the ta-C layers. The grade of the graphitization depends on the sp³/sp² ratio in the as-deposited multilayers. In samples with originally higher sp³ fraction, ta-C graphitizes at higher temperatures than in samples with a lower sp³ fraction. The thermal treatment led to a significant smoothening of the Cr/ta-C interfaces in the multilayers with a wide intermixing zone and facilitated the formation of metastable fcc-CrC via interdiffusion of C and Cr.Keyword: DLC; multilayer; cathodic arc evaporation; Cr/ta-C; synchrotron; in situ

experiments; SAXS; TEM

Introduction

Diamond-like carbon (DLC) coatings are es-tablished as tribological and wear resistant protective coatings in abrasive environment, e.g., as coatings for piston rings or pins or for inserts in diesel injection devices, etc. [1-3]. The properties of DLC films varies from gra-phitic-like to diamond-like, thus correspond-ingly DLC’s show a wide range of properties [4]. The environmental constraints for the automotive industry are mainly responded by the reduction of friction and wear, which requires the development of more efficient, advanced wear protective coatings, for what DLC coatings are prominent candidates. However, a limiting factor for the application of DLC’s is the adhesion to the substrate [5, 6].

DLC’s are divided into three groups [6, 7]: (i) highly hydrogenated amorphous carbon (polymeric a-C:H), (ii) hydrogenated amor-phous carbon and (iii) non-hydrogenated amorphous carbon (a-C, ta-C). These three groups are further classified according to the ratio of sp³ (diamond) and sp² (graphit-

ic) bonds [6, 7]. However, the amount of sp³ bonds alone is not sufficient to describe the various properties of amorphous carbon coat-ings, e.g. tetrahedrally bonded diamond-like carbon (ta-C) films exhibit high hardness and low coefficient of friction, while polymeric amorphous carbon films (a-C:H) are ductile and exhibit a relatively low hardness. Both DLC’s containing a high amount of sp³ bonds. Therefore, the differentiation of DLC’s by the mechanical and physical properties is more practical.

The formation of ta-C coatings requires high compressive stress in order to stabilize sp³ bonds, which can be achieved by the ion bom-bardment [5]. Thus, the enhancement of the adhesion of the ta-C coatings is required to prevent the delamination [8]. A possible strat-egy for the improvement of the adhesion are the surface pretreatment of the substrate by metal ion bombardment and the deposition of a buffer layer between the substrate and the DLC film consisting of transition metals such

as Ti, Cr, Nb or V [2] as well as the deposition of multilayer coatings. Among the transition metals, chromium is a prominent candidate due to the formation of three stable carbides at moderate annealing temperatures. The chromium carbides with narrow homogene-ity ranges are assumed to act as efficient barri-ers for further diffusion [9]. Furthermore, the environmental temperature plays a crucial role for the usability of ta-C coatings, because their application temperature is limited to about 300 °C due to the graphitization of the sp³ bonds [6].

In our first contribution the effect of the car-bon ion energy (EC) on the microstructure of Cr/ta-C multilayers [10] were investigated ex situ. It was shown that the carbon ion energy used for the deposition of the ta-C layer chang-es the amount of sp³ bonds in the carbon lay-ers, which reached the maximum at medium Ec = 200 eV. A further increase of the carbon ion energy led to the decrease of the tetrahe-drally bonded carbon fraction. Furthermore, both the morphology and the quality of the Cr/ta-C interfaces were affected significantly by the energy of the carbon ions. In order to be able to describe the kinetics of the phase transitions and the related microstructure changes in the Cr/ta-C system in more details, three Cr/ta-C multilayer coatings were pre-pared in analogy to [10] by using a combined DC arc/pulsed arc technique and investigat-ed in situ during annealing up to 600 °C in vacuum. The microstructure evolution caused by the thermal treatment was investigated by glancing angle X-ray diffraction (GAXRD), small angle X-ray scattering (SAXS), X-ray reflectivity (XRR) and transmission electron microscopy (TEM) with high resolution.

Fig. 1: Scheme of the Sulzer Metaplas MZR324 PVD coater equipped with three DC arc sources (1), with a laser arc module with rotating graphite cathodes (2) and with a planetary sample holder in the vacuum chamber (3).

Film preparation

The deposition of the Cr/ta-C multilayer coat-ings was performed in the industrial PVD coater Sulzer Metaplas MZR324. The cham-ber was equipped with three DC arc cathodes made from Cr (diameter 63 mm) and a laser

Experimental Details

controlled pulsed arc module (LAM) with a cylindrical graphite cathode (length 400 mm and a diameter of 160 mm), as schematically illustrated in Fig. 1. No additional gases were required for the coating preparation, thus the base pressure was kept unchanged at 10-3 Pa during the whole deposition process. No ad-ditional substrate heating was applied. There-fore, the substrate temperature did not exceed 100 °C in order to ensure the formation of amorphous diamond-like carbon layers con-taining a high amount of sp³ bonds. The Cr/ta-C films were deposited onto polished (001) oriented Si wafers, which were placed on the outer circle of the planetary sample holder.

186

In order to avoid the increase of the surface roughness of the Si substrates, no ion-assisted substrate cleaning was performed prior to the deposition. The periodic motif of Cr and ta-C layers was built up through the rotation of the sample holder and the subsequent switching of the DC arc and LAM sources. The nomi-nal thickness of individual layers was set to ~ 10 nm; it was controlled by the deposition time for the Cr layers and the number of im-pulses for the ta-C layers in synchronization to the rotation speed of the sample holder. The intended total layer thickness of the mul-tilayer stacks was below 120 nm in order to be able to resolve Kiessig oscillations in the XRR curves. Therefore, the periodic motif (Cr/ta-C) was repeated 6 times in the coat-ings. In order to investigate solely the effect of the carbon ion energy on the microstructure evolution during the thermal treatment, the process parameters for the Cr layers were kept constant for the three multilayer films and the substrate bias was set to -50 V. In contrast, different peak currents and pulse lengths were adjusted during the deposition of ta-C layers in subsequent deposition runs in order to ob-tain three different mean carbon ion energies (Ec) of about 25 eV (DC1), 200 eV (DC2) and 500 eV (DC3).

Microstructure characterization and thermal treatment

The microstructure evolution during the an-nealing of the Cr/ta-C multilayer coatings was analyzed by in situ synchrotron experiments at the Rossendorf beamline (ROBL) BM20 at the European Synchrotron Radiation Facili-ty (ESRF) in Grenoble. The diffractometer at ROBL was equipped with a vacuum chamber (~6x10-5 mbar) with a heating stage installed that enabled in situ experiments within a tem-perature range between room temperature and 600 °C. The annealing of the Cr/ta-C layers was performed by a gradual increase of the annealing temperature from room tem-perature to 600 °C in steps of 100 K. The sam-ples were annealed for 1 h at each annealing temperature. After each temperature step, the Cr/ta-C multilayers were cooled down to

100 °C for the synchrotron measurements. The photon energy for the synchrotron exper-iments was set to 11.5 keV (λ = 0.107812 nm). The diffraction patterns were recorded using a 0D Mythen detector. By means of the in situ synchrotron X-ray reflectivity (XRR) exper-iments, the microstructure was described in terms of the changes of the mean mass den-sity ρ and the layer thickness t with increas-ing temperature. Furthermore, the crystalline phases were revealed by in situ glancing angle X-ray diffraction (GAXRD) experiments. The interface correlation in the multilayers was obtained from the measurement of the res-onant diffuse scattered intensity in the small angle region by laboratory ex situ SAXS ex-periments and from the HRTEM images.

The SAXS measurements were performed on the laboratory diffractometer D8 Advanced (BRUKER AXS), which was equipped with a sealed X-ray tube with a copper anode (λ = 0.15418 nm) and a Goebel mirror in the primary beam in order to assure a parallel pri-mary beam with a divergence of approximate-ly 150 arc sec. The primary beam was reduces in size by using a slit with a width of 0.1 mm in order to prevent the irradiation of the sam-ple holder. In the secondary beam, two slits with a width of 0.1 mm were used. A small divergence of the primary beam is necessary to achieve a high resolution in the reciprocal space. The SAXS measurements were per-formed as a series of ω – scans composed to reciprocal space maps (RSM).

Transmission electron microscopy (TEM) was performed in the high resolution (HR-TEM) mode of the analytical transmission electron microscope JEM 2020 FS from JEOL, which was equipped with an in column filter and a spherical aberration corrector (Cs) lo-cated in the primary beam. The TEM operat-ed at the acceleration voltage of 200 keV. TEM investigations yielded additional information of the morphology of the multilayers and the interface quality. The TEM investigations were performed on the Cr/ta-C multilayers in the as-deposited state as well as ex situ after the synchrotron annealing experiments at 600 °C for 1h.

187

Microstructure of the as-deposited Cr/ta-C multilayer coatings

The periodicity of the Cr/ta-C motif in the multilayers under study was confirmed by XRR measurements (see Fig. 2) and TEM images (see Fig. 3). The XRR curves contain the multilayer features such as Bragg maxima from the periodic motif and the Kiessig oscil-lations related to the total multilayer thick-ness. In order to compare the SAXS measure-ments at different wavelength the diffraction angle θ was converted into the diffraction vector

Due to a variation of the individual layer thickness in the Cr/ta-C multilayers Bragg maxima weaken with increasing diffraction vector qZ. The Bragg maxima of DC1 and DC2 are broader than for sample DC3, which is caused by the variation of the thickness of the periodic motif. The thickness variations of the periodic motif seen by XRR measure-

Results ments were caused either by a chemical gra-dient at the interfaces or by the sample holder rotation during the deposition process. The position of the sample holder could not be controlled to guarantee the same initial posi-tion when switching from Cr cathode to the graphite cathode. Furthermore, the decay of the Bragg maxima was more pronounced with increasing carbon ion energy (see sam-ples DC1 and DC3 in Fig. 2) indicating the loss of the well pronounced periodicity of the multilayer stacks. As seen in Fig. 2 the mea-sured intensities of the XRR curves decreased faster with increasing diffraction vector qZ for the Cr/ta-C multilayer DC3. The decline of the XRR curves is proportional to the sur-face roughness [11, 12]. The high carbon ion energy used for the preparation of the mul-tilayer sample DC3 caused a significant in-crease of the surface roughness.

TEM imaging as shown in Fig. 3 confirmed the gain of the surface and interface rough-ness of sample DC3 (Fig. 3b). The trend of the effect of the carbon ion energy on the inter-

Fig. 2: X-ray reflectivity curves measured for Cr/ta-C multilayercoatings deposited at variouscarbon ion energies of approxmately 25 eV (DC1), 200 eV (DC2)and 500 eV (DC3).

188

Fig. 3: Bright field TEM images of the Cr/ta-C multilayer coatings deposited at EC ~25 eV (a) and at EC ~500 eV (b).

face roughness was already shown in [10]. The Cr layers showed a columnar growth, where-as the ta-C layers were amorphous as seen by the extinction of the diffraction contrasts in the bright field TEM images. The amorphous nature of the ta-C layers was confirmed by the Fast Fourier Transformation (FFT) of the carbon layers of the marked area in the ta-C layer of sample DC1 (see Fig. 4b) and sample DC3 (Fig. 5b). The FFT only showed a diffuse halo without any intensity maxima. The phase analysis performed by GAXRD revealed the body centered cubic (bcc) Cr as the only crys-talline phase. These results are in agreement with our previous published results [10]. The FFT of the Cr layers deposited at a low carbon ion energy (Fig. 4c) confirmed the crystallin-ity of Cr and revealed a <011> orientation of bcc-Cr in the growth direction. On contrary, the FFT of the Cr layer deposited at high car-bon ion energy of ~500 eV (Fig. 5c) showed an intensity ring. The absence of the intensity maxima indicated the loss of the crystalline nature of the Cr layers due to the high car-bon ion energy DC1. These results are in good agreement with our previous contribution on the Cr/ta-C multilayer coatings [10]. Further-more, TEM imaging shows well a different quality of the interfaces between the Cr and the ta-C layers for sample DC1 (see Fig. 4a) and DC3 (see Fig. 5a).

The refinement of the XRR curves yielded information about the thickness of the indi-vidual layers t, the mean thickness of the Cr and ta-C layers, the density of the individual layers ρ and on the interface roughness . The XRR curves were calculated by using the recursive optical Parratt formalism [13] and the computational routine from [14, 15] based on the Fullerton approach [12]. The XRR analysis revealed that the mean den-sity of the ta-C layers (ρC) increases from Ec~25 eV to ~200 eV from 2.8 to 3.0 g/cm³ and declines if the carbon ion energy is fur-ther increased to about 500 eV (see Fig. 6a). The lowest carbon density was obtained for the Cr/ta-C coating deposited at the highest carbon ion energy (Ec~500 eV). A similar trend was found for the mean density of the Cr layers (ρCr) in the multilayer stacks. The mean density of the Cr layers was in all three Cr/ta-C multilayers higher than the theoreti-cal density of Cr with 7.2 g/cm³. The higher mean density of the Cr layers can be caused either by the incorporation of C in the Cr lay-ers, by a high interface roughness between the Cr and the ta-C layer and/or the generation of a diffuse interface. The large error margin of the determined densities of the Cr and the ta-C layers references the correlation of the interface roughness and of the inhomogene-ity of the layers on the scattering potential.

189

The used computational model treats the in-terface roughness statistically and these effects yield similar effects on the scattering potential of the multilayer stack. Therefore, the used re-finement method cannot distinguish between surface roughness and the variation of the electron density of the layers caused by in-corporations or diffuse interfaces. Therefore, the large error margin can be understood as artifact of the scattering potential. The refined density and layer thickness of the Cr/ta-C multilayer are graphically summarized in Fig. 6c) and Fig. 6d), where it is shown that the mean layer thickness of the Cr and the ta-C layers decreased with increasing carbon ion energy. The individual layer thickness of the ta-C and the Cr layers decreased from about 10 nm to 6.5 nm (see Fig. 6c) and 12 nm

to 8 nm (Fig. 6d) for sample DC1 to DC3, respectively. The increase of the carbon ion energy rise the probability of resputtering of the carbon at higher energies, which explains more pronounced decrease of the ta-C layer thickness. If the kinetic energy of the car-bon ions is sufficient high the carbon can be ejected from the growing film by breaking the carbon bonds of the surface layers. Further-more, at higher kinetic energies of the carbon can be incorporated in the Cr interface by ion implantation.

190

Fig. 4: HRTEM image of the Cr/ta-C multilayer coatings deposited with a low carbon ion energy of ~25 eV (a) and the FFT of the marked area of the ta-C (b) and the Cr (c) layer. The arrow marks the growth direction.

Fig. 5: HRTEM image of the Cr/ta-C multilayer coatings deposited with a high carbon ion energy of ~500 eV (a) and the FFT of the marked area of the ta-C (b) and the Cr (c) layer. The arrow marks the growth direction

The influence of the carbon ion energy on the interface quality was investigated by re-ciprocal space maps (RSM) in the small an-gle region (Fig. 7) and was confirmed by TEM investigation (see Fig. 3). The Cr/ta-C multilayers deposited at a carbon ion ener-gy of about 25 eV exhibited a well-defined interface between the individual layers (see Fig. 3a, Fig. 4a), whereas the interfaces of the Cr/ta-C multilayer coating deposited at a carbon ion energy of ~500 eV were strong-ly disturbed (Fig. 3b, Fig. 5a) indicating an enhanced defect density at the interfaces due to the ion bombardment. The influence of the carbon ion energy was further revealed by RSM measurements exemplarily for sam-ple DC1 (Fig. 7a), DC2 (Fig. 7b) and DC3 (Fig. 7c). The RSM of sample DC1 (Fig. 7b), which was deposited at Ec ~25 eV for the car-bon ions, exhibited the typical features of the resonant diffuse scattered intensity of a mul-tilayer coating such as the specular reflect-

ed intensity along qx = 0, well pronounced Yoneda wings and resonant diffuse scattering bananas parallel to the qx axis. The increase of the intensity of the resonant diffuse scatter-ing (so called Holý bananas) is caused by the partial correlation of the interface roughness profiles in the multilayer stack [16] and the position of the Holý bananas intersect the Bragg peaks of the specular reflected intensity. Furthermore, dynamical scattering effects are visible for sample DC1 such as the Bragg-like resonant lines, which intersect the Bragg peaks of the specular reflected intensity and process parallel to the Yoneda wings. The Bragg-like lines occur if the diffraction con-dition of the incident and outgoing waves are fulfilled. Furthermore, the intersection of the Bragg-like lines yields maxima of the diffuse scattered intensity, which are called Bragg-like resonant peaks. The appearance, the form and the intensity of these dynamical effects de-pends on the interface correlation, whereas

Fig. 6: Summary of the mean mass density of the C (a) and the Cr layers (b) in the as deposited Cr/ta-C multilayercoatings.

191

the dynamical effects occur independently of the actual interface correlation function. The RSM of the Cr/ta-C multilayer coating DC3 deposited at Ec ~500 eV (Fig. 7c) showed the absence of the Holý bananas and a continu-ous decrease of the diffuse resonant scattered intensity for increasing qz values. The absence of the Holý bananas indicates the loss of the vertical interface roughness correlation in the multilayer stack, which was also seen by TEM investigations (see Fig. 3b). However, the Yoneda wings can be observed in the sample DC3. The width of the Holý bananas in the qz

direction can be understood as a measure of the replication of the interface profile in the multilayer stack. Therefore, the degradation of the diffuse resonant scattered intensity of the Holý bananas and their broadening are caused by the loss of the vertical interface cor-relation in the Cr/ta-C multilayers caused by the high defect density at the interfaces.

Fig. 7: Reciprocal space maps (RSM) ofthe Cr/ta-C multilayer coatings in theas-deposited state. The RSM are sortedby increasing carbon ion energy fromthe top to the bottom with the order of (a)25 eV (DC1), (b) 200 eV (DC2) and (c)500 eV (DC3).

Microstructure evolution of Cr/ta-C multi-layers during the thermal treatment

The Cr/ta-C multilayers were annealed up to an annealing temperature of 600 °C for 1h per temperature step. The thermal treatment yielded information about the microstructure evolution of the multilayer stacks, especially on the stability of the diamond-like ta-C lay-

ers and on the interdiffusion processes at the Cr/ta-C interfaces. From the XRR analysis, the evolution of the mean densities of the Cr and the ta-C layers in dependence on the an-nealing temperatures are presented in Fig. 8a. The XRR refinement revealed the decrease of

192

the ta-C density in all three multilayer stacks above an annealing temperature of 300 °C (Fig. 8a). Above annealing temperatures of 400 °C, the decrease of the ta-C density was decelerated in all three layers and the density loss achieved saturation above the theoretical density of graphite. It is assumed that the de-crease of the mean density of the ta-C layers is caused by the transformation of the sp³ bonds to the sp² bonds, which can be understood as graphitization of the ta-C layers. However, the refined ta-C layer densities after the annealing at 600 °C varied in the three multilayers be-tween 2.39±0.2 g/cm³ (DC1), 2.49±0.1 g/cm³ (DC2) and 2.3±0.1 g/cm³ (DC3). The high-est density of carbon after the annealing was observed in sample DC2, which was depos-ited at the carbon ion energy of about 200 eV. The Cr/ta-C multilayer coating DC2 also ex-hibited the highest density of carbon in the as-deposited state. The lowest C density after the thermal treatment was observed for sam-ple DC3, which was deposited at the highest carbon ion energy during the coating pro-cess. These results indicate that the expected highest amount of the sp³ bonds in the as-de-posited ta-C layers correlates with the higher density obtained from the XRR analysis. In our first report on the Cr/ta-C layers [10], the amount of sp³ and sp² bonds in DLC coatings was estimated from the intensity ratio of the π* and σ* peaks [17] in the electron energy

loss spectra, which were measured locally in a transmission electron microscope. The com-parison of the densities of the carbon layers with the amount of the sp³ bonds revealed al-most linear relationship between these quan-tities. In this work, the above relationship was used to predict higher amount of sp² bonds in the ta-C layers from their lower density. It was concluded that the amount of sp³ bonds the after thermal treatment depends on the amount of sp³ bonds in the as-deposited ta-C layers. More sp³ bonds in the as-deposited layers lead to more sp³ bonds in the annealed layers.

Also for the Cr layers, a decrease of the mean density was observed (Fig. 8b). In the tem-perature range between 200 to 300 °C, the Cr densities were observed to decrease only slightly in the Cr/ta-C multilayers. Howev-er, in contrary to the ta-C layer a significant decrease of the mean Cr layer densities were observed at annealing temperatures above 300 °C, where the Cr densities decreased to approximately 6.5 ± 0.3 g/cm³ in all three Cr/ta-C multilayer stacks. After the thermal treatment, the determined Cr layer density is lower than the theoretical densities of pure Cr of 7.2 g/cm³ and the three thermodynam-ically described chromium carbides Cr23C6, Cr6C3 and Cr3C2 having the tabulated densi-ties of 6.95, 6.88 and 6.6 g/cm³, respectively.

193

Fig. 8: Densities evolution in the ta-C(a) and the Cr layers (b) depending on the annealing temperatures and the energy of the carbon ions 25 eV (DC1 - black), 200 eV (DC2 - blue) and 500 eV (DC3 - red). The gray planes represent the theoretical density of graphite with 2.26 g/cm³ (a), Cr and the metastable CrC (b). The theoretical density of diamond is 3.51 g/cm³, which is not shown in (a)

However, no significant changes of the Cr and the ta-C layers were observed by XRR refine-ment due to the thermal treatment of the Cr/ta-C multilayer coatings. The formation of the metastable CrC by interdiffusion would led to the change of the thickness ratio of the Cr and ta-C layers therefore it can be assumed that the formation of the metastable CrC was caused by incorporated C in the Cr layers due to ion implantation during the deposition process [19]. The TEM investigation showed the reduction of the ta-C layer thickness. The thickness reduction could not be observed by XRR analysis because of the chemical gradi-ent in the as-deposited samples.

As it can be seen in Fig. 9, the interface morphology of the Cr/ta-C multilayer DC3 changed significantly after the heat treatment. In the as-deposited Cr/ta-C coating (Fig. 3b and Fig. 5a), the interfaces were blurred, whereas the interfaces in the annealed mul-tilayer stack appeared sharp and smoothed (Fig. 9). The smoothening of the interfaces was also observed by the RSM measurements (Fig. 10). The measured resonant diffuse scat-tered intensity of the Cr/ta-C multilayers were enhanced, whereas the specular reflected in-tensity for sample DC1 and DC2 (Fig. 10a,b)

Fig. 9: HRTEM image of the Cr/ta-Cmultilayer coating deposited with a highcarbon ion energy of ~500 eV after thethermal treatment at 600 °C for 1h. Theinset represents the FFT of the markedarea of the Cr layer. The arrow marks thegrowth direction.

However, the lower Cr density might be ex-plained by the formation of the metastable CrC phase in the Cr layers. The formation of the metastable face centered cubic (fcc) CrC phase was reported by Bewilogua et al. [18] as well as Wang et al. in 1993 [19]. However, the existence of the metastable CrC phase is discussed due to the discrepancy to the em-pirical Hägg’s rule, which limits the formation of the solid solutions of metalloid – metal by the radius ratio rc/rMe < 0.59. The stabilization of the metastable fcc-CrC phase was reported to be caused by the high ion dose during the ion implantation of C ions [19]. The theoret-ical density of the fcc-CrC phase is tabulated with 6.49 g/cm³ as calculated for the NaCl structure type and the cubic lattice parame-ter of a = 4.03 Å, which is in good agreement with the determined Cr densities for all three Cr/ta-C multilayer coatings. The existence of the metastable fcc-CrC phase was supported by the TEM investigations of sample DC3 af-ter the thermal treatment at 600 °C (see Fig. 9). The TEM investigation showed that the ta-C layers are amorphous, whereas the ther-mal treatment of the Cr/ta-C multilayers im-proved the crystalline nature of the Cr layers (Fig. 9) and the FFT of the marked area (inset Fig. 9) could be indexed as the fcc-CrC phase.

194

disappeared in the RSM measurements after the thermal annealing at 600 °C (Fig. 7a vs. Fig. 10a). The GAXRD analysis showed ad-ditional peaks of chromium silicides above annealing temperature of 400 °C and TEM investigations of the annealed multilayers revealed the formation of the chromium sil-icides at the Si/Cr interfaces. The formed reaction zone exhibited a rough, corrugated interface between the chromium silicides and the Si substrate, which was responsible for the disappearance of the specular reflected inten-sity in the RSM. However, the Holý bananas (maxima of the resonant diffuse scattering) were still present in the RSM measurements, which revealed that the multilayer structure of the remaining Cr/ta-C multilayer with cor-related interface roughness was still present after the thermal annealing. For sample DC3 (Fig. 10c), local intensity maxima of the dif-fuse scattered intensity were observed at the related Bragg peaks. In the measured RMS of the as-deposited multilayers (Fig. 7c), the

intensity of the resonant diffuse scattering decreased continuously with increasing dif-fraction vector qz with no local maxima of the diffuse scattered intensity, which was caused by the blurred and corrugated Cr/ta-C inter-faces with no vertical correlation in the as-de-posited state. The intensity and the width of the Holý bananas are a measure of the vertical correlation of the interfaces [16]. For samples DC1 and DC2, the increase of the width of the Holý bananas indicated the decrease of the vertical correlation of the interface rough-ness profile. In the RSM of sample DC3, the Holý bananas are weak but still visible indi-cating the sharpening of the Cr/ta-C interfac-es. The sharpening of the interfaces and the improvement of the vertical correlation of the interface roughness profile were confirmed by TEM and HRTEM investigations as seen in Fig. 9.

Fig. 10: Reciprocal space maps (RSM) of the Cr/ta-C multilayer coatings after a thermal treatment up to 600 °C. The RMS are sorted according to increasing carbon ion energy Ec applied during the deposition process from the top to the bottom with the order of (a) 25 eV (DC1), (b) 200 eV (DC2) to (c) 500 eV (DC3).

195

Discussion

The carbon ion energy affected the density of the ta-C layers, which is directly proportion-al to the sp³/sp² ratio in the ta-C layers. The sp³/sp² ratio reached a maximum at the me-dium carbon ion energy. A further increase of the carbon ion energy led to the graphiti-zation of the ta-C layers. This trend was ob-served already in our first report on the Cr/ta-C multilayers [10]. The presented results of the XRR refinements showed the decrease of the mean carbon density with increasing annealing temperature. In all three Cr/ta-C multilayers, the density reached saturation above the annealing temperature of 500 °C. The mean carbon density of the ta-C layers remained higher in the multilayer coatings with a higher sp³/sp² ratio in the as-deposited state. These results indicated that the amount of sp³ bonds in the as-deposited multilayers plays a crucial role for the graphitization of the ta-C layers. Furthermore, the saturation of the transformation from sp³ to sp² bonds was shifted to higher annealing temperatures (see Fig. 8a) with increasing mean carbon density. Sample DC3 reached the saturation of the density loss at the annealing temperature of 300 °C, whereas the decline of the carbon density in DC2 was shifted to the annealing temperatures above 500 °C. Therefore, we as-sume that the initial amount of the sp³ bonds in the as-deposited state is the limiting factor the sp³ to sp² conversion in the Cr/ta-C mul-tilayer coatings. The formation of sp³ bonds in DLC coatings is caused by the subplanta-tion of the carbon ions but the sp³/sp² ratio decreased at high carbon ion energies [4]. The decrease of the fraction of the sp³ bonds with increasing carbon ion energy can be explained by the thermal spike model [4, 20], which de-scribes the thermal activated relaxation of the sp³ bonds and the formation of the stable sp² bonds during the deposition. Furthermore, the amount of sp³ bonds in ta-C coatings is directly correlated to the intrinsic compres-sive stress in these coatings [4]. The increase of the sp³/sp² ratio increases the compressive stress in DLC coatings [21]. The chemical bonding of the ta-C coating is shifted above the stability criterion for sp³ and sp² known

as Berman-Simon line [21]. We assume that due to the increase of the sp³/sp² ratio in the investigated Cr/ta-C multilayers, the thermal activation barrier for the transformation from sp³ to sp² bonds is shifted to higher values, be-cause the higher compressive stresses in the ta-C layers stabilized the sp³ bonds. There-fore, the thermal activation barrier for the graphitization of the sp3 bond is enhanced. Furthermore, the transformation of sp³ to sp² bonds is assumed to be decelerated due to the high compressive stress in the ta-C layers.

As already discussed in our first contribution [10], the interface quality in the Cr/ta-C mul-tilayers is strongly affected by the carbon ion energy, which was applied for the deposition of the ta-C layers. The increase of the carbon ion energy led to an increasing impact of the carbon ions in the Cr layers, which caused an increasing intermixing zone at the interfaces between Cr and C and the incorporation of C in the Cr layer. The intermixing zone can be explained by the subplantation model [4]. The penetration depth of the carbon ions can be explained as a function of the carbon ion en-ergy with a certain penetration threshold (Ep). The penetration threshold is proportional to the difference of the surface binding energy and the displacement energy of the atoms in the coating [4, 22]. Therefore, at a certain ion energy the carbon ions are able to penetrate the surface layer [23]. With increasing carbon ion energy the penetration depth of the carbon ions increases. However, the required energy for the penetration is rather small compared to the total carbon ion energy. The excess en-ergy is used for the displacement of the car-bon atoms and is also converted to thermal energy. The influence of the energy fraction transformed to thermal energy is described in the model of thermal spikes [4, 20]. The thermal energy enables the diffusion of the atoms in the growing films additionally to the intermixing due to subplantation. Therefore, the combination of an increased penetration depth due to the increased carbon ion ener-gy and the thermal activated diffusion in the growing film can be assumed as the driving

196

factors for the formation of the intermixing zone at the Cr/ta-C interfaces and the incor-poration of C in the Cr layers.

The thermal treatment showed the sharpen-ing of the Cr/ta-C interfaces in the multilayer stacks. The XRR and HRTEM investigations showed that the density of the Cr layers de-creased significantly and the formation of the metastable CrC is assumed at higher an-nealing temperatures in the Cr/ta-C multi-layers. According to our results, the thermal treatment initiated the interdiffusion process at the Cr/ta-C interfaces. The decrease of the mean carbon density indicated that no diffusion of Cr in the ta-C layer occurred. A decrease of the Cr layer density occurred at annealing temperature above 300 °C indicat-ing the diffusion of the carbon atom into the Cr layer of the recombination of the Cr layers and the incorporated C atoms. Therefore, the diffusion of Cr can be neglected. In contradic-tion to the thermodynamic prediction for the Cr-C system [24], the formation of the three stable chromium carbides was not observed. If the three stable chromium carbides would have formed at the interfaces, they would oc-cur as very thin layers, which could not be ob-served by XRR. However, the decrease of the Cr density indicated the formation of the metastable fcc-CrC carbide.

According to [18], we assume that the ion bombardment during the deposition of the Cr/ta-C multilayers and the formation of the high compressive stress are the driving factors for the stabilization of the metastable fcc-CrC phase during the thermal treatment. The metastable fcc-CrC phase contains the highest carbon amount of the chromium car-bides and it is likely formed at the Cr/ta-C in-terfaces. Due to the high sp³/sp² ratio in our Cr/ta-C multilayers, the compressive stresses were assumed to be high as well, which can shift the stability criteria for the metastable fcc-CrC phase. According to Bewilogua et al. [18], the metastable fcc-CrC phase trans-forms to the stable Cr3C2 during the thermal treatment, which could not be observed in the Cr/ta-C multilayer coatings. The phase tran-

sition from the metastable CrC to the stable Cr3C2 phase is connected with an increase of the mass density and with an increase of the molar volume. Therefore, the compres-sive stresses would further increase due to the phase transformation of the metastable CrC phase. The transformation of sp³ to sp² bonds in the ta-C layers is linked to the de-crease of the mass density and to an increase of the occupied molar volume so an increase of the compressive stress. Assuming the in-terdiffusion of carbon from the ta-C into the Cr layer to form CrC, the thickness of the Cr layer should increase by 40 % at a total deple-tion of the graphitic layer and a depletion of 70 % assuming a pure diamond layer. How-ever, such significant changes of the layer thicknesses during to the thermal treatment were not observed by XRR; neither for the ta-C layers nor for the Cr layers. Therefore, it can be assumed that the formation of the CrC phase in Cr layer due to the thermal treatment is rather caused by the ion implantation of C in the Cr layer during the deposition than by interdiffusion processes during the annealing from the ta-C layers into the Cr layers. How-ever as seen by HRTEM (compare Fig. 5 and Fig. 9), the ta-C layers in the multilayer stack after the thermal treatment appeared thin-ner than in the as-deposited state and their interfaces to the Cr layers are sharper. The carbon ion bombardment produced an inter-mixed zone in the as-deposited state as seen for DC3 (Fig. 5), which appeared in the XRR curves either as variation of the layer thick-ness or as additional interface roughness. Due to the blurred interfaces in the as-deposited state the decrease of the layer thickness could not be observed by XRR measurements. The sharpening of the interfaces after the ther-mal treatment of the multilayer stacks can be explained by diffusion processes in the orig-inally blurred interfaces. Due to the narrow homogeneity ranges, the gradual change of the carbon concentration has to be reshaped in order to achieve the concentration in the individual chromium carbide phases.

197

Conclusion

In the as-deposited Cr/ta-C multilayer coat-ings, the sp³ fraction in the ta-C layers was shown to modify the thermal stability of the ta-C layer and the amount of preserved sp³ bonds significantly. The increase of the sp³ fraction in the as-deposited state shifted the thermal activated transformation from sp³ to sp² to higher annealing temperatures due to the decelerated transformation of sp³ to sp² bonds due to a higher sp³ fraction. A model was proposed to explain the influence of the sp³ bonds and the compressive stress to ex-plain the density loss saturation depending on the sp³/sp² ratio in the as-deposited coatings. The multilayer structure of the Cr/ta-C coat-ings was still preserved after the annealing for 1h at 600 °C and a smoothening of the blurred interfaces were observed. We proposed the in-terdiffusion of carbon atoms due to the ther-mal treatment to explain the smoothening of the interfaces in the Cr/ta-C multilayers. Furthermore, the formation of metastable fcc-CrC is assumed.

Acknowledgement

This work was performed within the Clus-ter of Excellence “Structure Design of Nov-el High-Performance Materials via Atomic Design and Defect Engineering (ADDE)” that is financially supported by the Euro-pean Union (European Found for Regional Development) and by the Ministry of Sci-ence and Art of Saxony (SMWK).

198

References

1. S.C. Tung, H. Gao, Wear 255/7-12 (2003) 1276.2. L. Nobili, L. Magagnin, Transactions of Nonferrous Metals Society of China 19/4

(2009) 810.3. K. Bewilogua, D. Hofmann, Surf. Coat. Technol. 242/0 (2014) 214.4. J. Robertson, Materials Science and Engineering: R: Reports 37/4-6 (2002) 129.5. D.R. McKenzie, D. Muller, B.A. Pailthorpe, Phys. Rev. Lett. 67/6 (1991) 773.6. S. Neuville, A. Matthews, Thin Solid Films 515/17 (2007) 6619.7. VDI2840, Germany, 2005.8. Y. Pauleau, in: C. Donnet, A. Erdemir (Eds.), Tribology of Diamond-Like Carbon

Films, Springer US, 2008, p. 102.9. W. Mayr, W. Lengauer, P. Ettmayer, D. Rafaja, J. Bauer, M. Bohn, Defect and Diffusi-

on Forum 143-147 (1997) 569.10. U. Ratayski, D. Rafaja, V. Klemm, U. Mühle, M. Leonhardt, H.-J. Scheibe, Surf. Coat-

Technol. 206/7 (2011) 1753.11. D. de Boer, Phys. Rev. B: Condens. Matter 49/9 (1994) 5817.12. E. Fullerton, I. Schuller, H. Vanderstraeten, Y. Bruynseraede, Phys. Rev. B: Condens.

Mater 45/16 (1992) 9292.13. L.G. Parratt, Phys. Rev. 95/2 (1954) 359.14. D. Rafaja, V. Valvoda, J. Kub, K. Temst, M.J. Van Bael, Y. Bruynseraede, Phys. Rev.

Condens. Matter 61/23 (2000) 16144.15. D. Rafaja, H. Fuess, D. Šimek, L. Zdeborová, J. Phys.: Condens. Matter 14 (2002)

10021.16. U. Pietsch, V. Holý, T. Baumbach, High-Resolution X-ray Scattering, Springer Verlag,

1999.17. D. Galvan, Y.T. Pei, J.T.M. De Hosson, A. Cavaleiro, Surf. Coat. Technol. 200/1–4

(2005) 739.18. K. Bewilogua, H.J. Heinitz, B. Rau, S. Schulze, Thin Solid Films 167/1-2 (1988) 233.19. J. Wang, X. Chen, N. Yang, Z. Fang, Appl. Phys. A 56/4 (1993) 307.20. H. Hofsäss, H. Feldermann, R. Merk, M. Sebastian, C. Ronning, Appl. Phys. A 66/2

(1998) 153.21. D.R. McKenzie, Journal of Vacuum Science & Technology B 11/5 (1993) 1928.22. J. Robertson, Diamond Relat. Mater. 2/5-7 (1993) 984.23. J. Robertson, Diamond Relat. Mater. 14/3-7 (2005) 942.24. W. Mayr, W. Lengauer, P. Ettmayer, D. Rafaja, J. Bauer, M. Bohn, Journal of Phase

Equilibria 20/1 (1999) 35.

199

Defect engineering in Ti-Al-N based coatings via energetic particle bombardment during cathodic arc evaporation

Christina Wüstefeld 1, Mykhaylo Motylenko 1, David Rafaja 1, Claude Michotte 2, Christoph Czettl 3

1 Institute of Materials Science, Technische Universität Bergakademie Freiberg, D-09599 Freiberg, Germany

2 CERATIZIT Luxembourg S.à.r.l., L-8201 Mamer, Luxembourg3 CERATIZIT Austria GmbH, A-6600 Reutte, Austria

200

AbstractThis study describes the impact of the bias voltage during reactive cathodic arc evaporation (CAE) on the microstructure development in Ti-Al-N based coatings and their thermal stabil-ity. For this purpose, Ti-Al-N monolayers and Ti-Al-N / Al-Ti-(Ru)-N multilayers were depos-ited from powder metallurgical Ti100-xAlx (x = 40, 50, 60, 67) targets and Ti50Al50 / Ti33-yAl67Ruy (y = 0, 1, 5) targets, respectively. In order to vary the energy of the impinging ions, the CAE dep-osition was done at different bias voltages (UB) between -20 V and -120 V. The microstructure of the coatings was characterized with the aid of glancing angle X-ray diffraction (GAXRD) in the as-deposited and annealed state. GAXRD was done ex situ as well as in situ during anneal-ing up to 950 °C. The microstructure investigated by X-ray diffraction was described in terms of the phase composition, the stress-free lattice parameter, crystallite size and macroscopic lat-tice strain of the face centred cubic (fcc) (Ti,Al)N phase. Furthermore, analytical transmission electron microscopy was performed in conjunction with electron energy loss spectroscopy and X-ray energy dispersive spectroscopy. The microstructure properties were correlated with the hardness of the samples. It was found that the microstructure of the Ti-Al-N based coatings can be designed by the applied bias voltage during the CAE deposition. With increasing bias voltage, higher concentration fluctuations of titanium and aluminium within the fcc-(Ti,Al)N phase and a rising defect density were observed. The high defect density and the higher con-centration fluctuations were accompanied by the formation of minor segregations of Al-rich fcc- (Al,Ti)N at high bias voltages. At high UB the formation of wurtzitic AlN in the Al-rich coatings was apparently retarded in the as-deposited coatings but at elevated temperatures the decomposi-tion was accelerated.

1. Introduction

Ti-Al-N coatings were presented already in 1985/1986 as industrially promising protec-tive coatings [1, 2]. Especially their oxidation resistance [2, 3] and hardness at elevated tem-peratures [4, 5] make them attractive as pro-tective coatings for high speed drills [6] and cutting tools. Thus, already in 1989 Ti-Al-N coated cutting tools were produced for the commercial market [7]. In the following time, further improvements of the coating proper-ties and their thermal stability were aspired by the addition of doping elements like Si [8, 9], B [10, 11], Nb [10], Ta [11], Hf [10], Ru [12, 13] etc. to Ti-Al-N based coatings or by the deposition of a multilayer architecture [11, 14].

The Ti-Al-N based hard coatings are main-ly produced by physical vapour deposition methods (PVD), which facilitate the for-mation of the metastable face centred cubic Ti1-xAlxN phase. The highest experimentally observed solubility limit of Al in fcc-Ti1-xAlxN was x = 0.67 [15, 16], which agrees with ab

initio calculations [17, 18] that revealed the same value. If the metastable fcc-Ti1-xAlxN phase is exposed to high temperatures, it un-dergoes spinodal decomposition and forms Ti-rich and Al-rich fcc-(Ti,Al)N domains, as it was observed by Hörling et al. [16] and Mayrhofer et al. [4] in the temperature range of ~860 - 900 °C. This decomposition step is accompanied by an age hardening effect [4]. At higher temperatures, the authors of Refs. [4, 16] observed the transformation of fcc-AlN into its thermodynamically stable wurt- zite (w) modification. This decomposition step was frequently reported to take place above 1000 °C [4, 14, 16, 19-24]. Furthermore, the formation of w-AlN was considered for a long time to be detrimental for the coating properties like hardness [25-27] and adhesion [5, 28]. In the recent time, it could be shown that w-AlN can form during annealing al-ready in the temperature range of 850 - 900 °C [29-35] and thus coexists with fcc Al-rich (Al,Ti)N. This implies that also a direct trans-formation of fcc-(Ti,Al)N to w-(Al,Ti)N is

201

possible [36]. A hardness increase after an-nealing at 900 °C and the presence of minor amounts of w-AlN that were reported in Refs. [30, 35] support our idea [8, 37-39] that the presence of w-AlN is not generally harmful with regard to the hardness. In particular, the hardness is enhanced through the partial coherent interfaces between wurtzite and fcc phase [8]. The age hardening effect reported for the Ti-Al-N based coatings at ~ 900 °C e.g. in Refs. [4, 16, 30, 35] can be activated during the metal cutting. It could be shown that the temperature at the cutting edge of Ti0.6Al0.4N coated WC-Co cutting inserts was 850 - 900 °C during dry cutting of carbon steel [40].

The properties of Ti-Al-N-based coatings are determined by the microstructure fea-tures like phase composition, crystallite size and lattice strain. The microstructure can be adjusted on the one hand by the Al concen-tration and on the other hand by the depo-sition parameters. The most used techniques for the deposition of the hard coatings based on Ti-Al-N are cathodic arc evaporation and magnetron sputtering (MS). Both deposition techniques are usually carried out in reactive mode, where aluminium and titanium are transferred from the solid phase of the tar-get to the vapour phase. Nitrogen is provided as reactive gas. The generated plasma differs substantially for both methods. In case of MS, energetic ions of an inert gas, usually argon, are produced in a glow discharge and eject at-oms of the target material due to the momen-tum transfer. The fraction of the ionized metal atoms is very low in MS. In contrast to MS, 50 to 100 % of the evaporated metallic spe-cies are ionized [41] in the CAE process as a result of the evaporation of the target material by a high energy arc [42]. The CAE process represents an energetic deposition. If a neg-ative bias voltage is applied to the substrate with respect to the plasma potential, the ions are accelerated towards the growing film. The energy of the ion impact can be modified by the bias voltage which influences the surface mobility of the deposited species.

This study addresses the microstructure for-mation of Ti-Al-N based coatings that were deposited by cathodic arc evaporation. On the one hand, the influence of the Al content and the bias voltage on the microstructure and thermal stability of Ti1-xAlxN monolay-ers were investigated. On the other hand, the effect of the Ru addition and the bias voltage on the microstructure and thermal stability of Ti-Al-N / Al-Ti-Ru-N multilayers were ana-lysed.

2. Experimental

The Ti-Al-N monolayers [39] and Ti-Al-N / Al-Ti-(Ru)-N multilayers [43] were produced in an industrial CAE facility of the Balzers RCS type [44]. The deposition chamber is equipped with six arc sources that are in-stalled at the circumference of the chamber. On two arc sources, pure titanium targets were mounted that were used for the deposi-tion of a ~200 nm thick TiN adhesion layer, which was deposited prior the actual coating. The deposition of the actual coating was done from mixed Ti-Al-(Ru) targets that were in-stalled on the other four arc sources. All tar-gets were produced by Plansee CM using a powder metallurgical route [45]. In case of the Ti-Al-N monolayer coatings, four composi-tions series were deposited via four different target compositions: (I) Ti60Al40, (II) Ti50Al50, (III) Ti40Al60 and (IV) Ti33Al67. In order to modify the ion impact during the deposition process, each composition series was made at four different bias voltages of -20 V, -40 V, -80 V and -120 V. In this way, a matrix of 4 × 4 samples was produced.

The Ti-Al-N / Al-Ti-(Ru)-N multilayers were deposited from two Ti50Al50 and two Ti33-yAl67Ruy targets. Three multilayer series with different Ru content were produced by the variation of the Ru concentration in the target, namely: (I) y = 0, (II) y = 1 and (III) y = 5. Each multilayer series (I to III) was de-posited at four different substrate bias voltages of -20 V, -40 V, -60 V and -80 V. The CAE dep-osition of all coatings was done in a nitrogen

202

(1)

atmosphere with a working pressure of 3.2 Pa and at a temperature of 450 °C. A two fold rotation was applied during deposition. As substrates polished cemented carbides con-taining 12 wt.% Co and mixed carbides (hex-WC, fcc-TiC) with SNUN 120412 geometry according to DIN ISO 1832 [46] were used.

Additionally, monolithic Ti-Al-N coatings that were deposited onto cemented carbide inserts were investigated [47]. These coatings were reactively deposited from a Ti40Al60 tar-get at a nitrogen pressure of 5 Pa using CAE. The deposition was done in the π300 appara-tus from PIVOT plc with one vertical rotat-ing target in the middle of the chamber. The deposition temperature was ~500 °C and a three-fold rotation was applied. Within this contribution, the vertically positioned sam-ples that were deposited at three different bias voltages of -20 V, -40 V and -80 V as well as the horizontally positioned samples deposited at UB = -80 V are considered.

The chemical composition of the as-deposit-ed coatings was analysed by wavelength-dis-persive electron probe microanalysis (EPMA / WDS) done on a JXA 8900 RL from Jeol. Using EPMA/WDS, the weight fractions of Ti, Al, O and Ru were determined. Since the nitrogen content could not be measured directly due to the overlap of the spectral line Ll of titanium with the Kα1 line from nitro-gen, the nitrogen content in the coating was calculated from the analytical total assuming that N is the complement to the analysed el-ements. The nitrogen content was checked by glow discharge optical emission spectroscopy (GDOES) done on SPETRU-MAT 750 from Leco. The GAXRD experiments were done on a D8 diffractometer (Bruker AXS) that was equipped with a sealed X-ray tube with copper anode; further details are given in Ref. [39]. The angle of incidence of the primary beam at the sample surface was 3°.

(2)

where the constants , , and arerelated to the single-crystal elasticcompliance constants of cubic materials, fordetails see Ref. [49], and

𝑎𝑎𝜓𝜓ℎ𝑘𝑘𝑘𝑘 = 𝐴𝐴2𝑎𝑎0𝜎𝜎 sin2 𝜓𝜓 + 𝐵𝐵2𝑎𝑎0𝜎𝜎 Γℎ𝑘𝑘𝑘𝑘sin2 𝜓𝜓 + 2𝐵𝐵1𝑎𝑎0𝜎𝜎Γℎ𝑘𝑘𝑘𝑘 + 2𝐴𝐴1𝑎𝑎0𝜎𝜎 + 𝑎𝑎0,

𝑎𝑎𝜓𝜓ℎ𝑘𝑘𝑘𝑘 = 𝐴𝐴2𝑎𝑎0𝜎𝜎 sin2 𝜓𝜓 + 𝐵𝐵2𝑎𝑎0𝜎𝜎 Γℎ𝑘𝑘𝑘𝑘sin2 𝜓𝜓 + 2𝐵𝐵1𝑎𝑎0𝜎𝜎Γℎ𝑘𝑘𝑘𝑘 + 2𝐴𝐴1𝑎𝑎0𝜎𝜎 + 𝑎𝑎0,

𝑎𝑎𝜓𝜓ℎ𝑘𝑘𝑘𝑘 = 𝐴𝐴2𝑎𝑎0𝜎𝜎 sin2 𝜓𝜓 + 𝐵𝐵2𝑎𝑎0𝜎𝜎 Γℎ𝑘𝑘𝑘𝑘sin2 𝜓𝜓 + 2𝐵𝐵1𝑎𝑎0𝜎𝜎Γℎ𝑘𝑘𝑘𝑘 + 2𝐴𝐴1𝑎𝑎0𝜎𝜎 + 𝑎𝑎0,

GAXRD was employed to characterize themicrostructure in terms of phasecomposition as well as stress-free latticeparameter ( ), microstrain ( ),macroscopic lattice strain ( ) and residualstress ( ) of the fcc-(Ti,Al)N phase [32]. Thestress-free lattice parameter of the fcc-(Ti,Al)N phase in the Ti-Al-N monolayerswas determined from the linear dependenceof the lattice parameters on [32],where is the inclination of the diffractionvector from the sample surface normaldirection. The Poisson’s ratio was adoptedfrom Ref. [48], which gives ν = 0.3 for fcc-TiN. In case of the Ti-Al-N / Al-Ti-Ru-Nmultilayers, the vs. plots revealed astrong anisotropy of the elastic latticedeformation of the fcc-(Ti,Al)N phase. Dueto the medium [ ratio of 0.53 ±0.01, the lattice strain was higher in the

direction than in the direction,, which corresponds to an

anisotropy factor .

Due to the strong observed anisotropy of theelastic deformation, first of all thedependence of on and on thediffraction indices was described using Eq.(1) [49]:

203

(3)

(4)

The symbol denotes the out-of-plane lat-tice parameter, which is obtained together with from the linear regression (Eq. (5)). The macroscopic lattice strain can be consid-ered as more reliable than the residual stress, since is determined from the XRD data us-ing just two parameters, the stress-free lattice parameter and the Poisson’s ratio [32].

(5)

The microstructure on the nanoscale was vis-ualized with transmission electron microsco-py (TEM) using a JEM 2010 FEF and a JEM 2200 FS from Jeol for the Ti-Al-N monolay-ers and the Ti-Al-N / Al-Ti-(Ru)-N multi-layers, respectively. The Ti-Al-N monolay-ers were prepared in plan-view orientation; details are given in Ref. [39]. The Ti-Al-N / Al-Ti-(Ru)-N multilayers were prepared as cross-sections using focused ion beam tech-nique. Fast Fourier transformation (FFT) was applied locally to the high resolution (HR) TEM images in order to perform the local phase identification, to determine the grain orientations and to describe the orientation relationship (OR) between fcc-(Ti,Al)N and wurtzite phase. The multilayer cross-sections were additionally investigated in scanning TEM modus in order to obtain information about the fluctuations of the metallic spe-cies across the layered structure using X-ray energy dispersive spectroscopy (EDS) and electron energy loss spectroscopy (EELS) of the Ti L2,3 edge. Additionally, the nitrogen K edge was recorded and analysed [50]. The generation of small electron probes and the operation in scanning modus over extended sample areas was achieved by the microscope JEM 2200 FS that is equipped with a spherical aberration (Cs=0.5) corrector. The EELS and EDS data were analysed using the software DigitalMicrographTM from Gatan.

Instrumented nanoindentation experiments were done on locally polished surfaces using the nanohardness tester from CSM Instru-ments equipped with a diamond Berkovich indenter in order to determine the indenta-tion hardness (HIT) of the coatings. The load of 30 mN was applied and removed with a loading rate of 60 mN/min. The recorded load displacement curves were analysed using the Oliver and Pharr method [51]. The coating thickness ranged between 4 and 5 µm and the penetration depth was less than 10 % of the thickness. The average indentation hardness was determined from at least 15 indentations.

were approximated with the linear depen-dence

is the cubic invariant. Afterwards, the Γ-dependent part of Eq. (1) was subtracted from the measured lattice parameters. Finally, the lattice parameters independent of diffrac-tion indices hkl,

with ν=0.3 [48]. For the calculation of the re-sidual stress a Young’s modulus of 500 GPa was assumed [32]. This assumption could be supported by nanoindentation experiments performed on the Ti-Al-N based coatings that contained fcc-(Ti,Al)N as major phase. These nanoindentation experiments yield-ed an indentation modulus in the range of 450-550 GPa. Additionally to the residual stress, the macroscopic lattice strain ε was determined by using a modification of Eq. (4)

204

In order to investigate thermally activated mi-crostructure changes in the coatings, anneal-ing experiments were done on selected coat-ings. The Ti-Al-N monolayer coatings were annealed in vacuum up to 850 °C, details of the annealing sequence are given in Ref. [33]. During annealing of the Ti-Al-N monolayer coatings, in situ GAXRD experiments were performed using synchrotron radiation, see Ref. [33]. For the thermal treatment of the Ti-Al-N / Al-Ti-(Ru)-N multilayers coatings, the samples were encased in silica glass tubes that were filled with argon gas at a pressure of 500 mbar. The glass tubes were annealed at the oven temperatures of 450 °C, 650 °C, 850 °C and 950 °C for 60 min. It is assumed that the temperature at the coating surface is slight-ly below the oven temperature and cannot be compared directly with the temperatures given for the in situ GAXRD experiments. The GAXRD experiments of the Ti-Al-N/Al-Ti-(Ru)-N multilayers coatings were done ex situ.

3. Results and Discussion

3.1 Chemical composition and phase composition of the samples

The metal ratios in the as-deposited Ti-Al-N monolayer [39] and Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings [43] as measured by EPMA / WDS are given in Table 1. The meas-urement revealed a lower [Al] / Σ[Me] ratio (where Σ[Me] = ([Ti] + [Al] + [Ru])) in the coatings as compared to the average Al to metal ratio of the utilized targets. This ef-fect is in accordance with Refs. [52-54] and is caused by an increased ionisation of titanium and by a higher re-sputtering yield of already deposited aluminium atoms. The reduction of the [Al] / Σ[Me] ratio is more pronounced in the Ti-Al-N / Al-Ti-(Ru)-N multilayers than in the monolayer coatings. In the multilay-ers, the Al to metal ratio is decreased approx. 10 % as compared to the targets. In case of the Ti-Al-N monolayer coatings, the reduction of the [Al] / Σ[Me] ratio increases from ~5 % to ~8 % for the coatings deposited from Ti60Al40 and Ti33Al67 targets, respectively. The applied UB had no effect on the chemical composition. GDOES confirmed these results and showed that the coatings contain 50 at.% of nitrogen.

Table 1: Metal ratios of the Ti-Al-N monolayer and Ti-Al-N / Al-Ti-(Ru)-N multilayers as determined by EPMA / WDS where Σ[Me] = ([Ti] + [Al] + [Ru]).

Coating Series Targetsystem combination

Ti-Al-N monolayers

Ti-Al-N / Al-Ti-(Ru)-N multilayers

4 × Ti60Al40

4 × Ti50Al50

4 × Ti40Al60

4 × Ti33Al67

2× Ti50Al50 & 2× Ti33Al67

2× Ti50Al50 & 2× Ti32Al67Ru1

2× Ti50Al50 & 2× Ti28Al67Ru5

0.62 ± 0.01

0.53 ± 0.01

0.44 ± 0.01

0.38 ± 0.01

0.48 ± 0.01

0.47 ± 0.01

0.45 ± 0.01

0.38 ± 0.01

0.47 ± 0.01

0.56 ± 0.01

0.62 ± 0.01

0.52 ± 0.01

0.53 ± 0.01

0.53 ± 0.01

0

0

0

0

0I

II

III

0.005 ± ≤ 0.001

0.02 ± ≤ 0.001

TiΣ[Me]

AlΣ[Me]

RuΣ[Me]

205

The phase composition of all coatings was concluded from the intensity of the diffrac-tion lines corresponding to the expected crystalline phases w-AlN and fcc-(Ti,Al)N. Each diffraction line in the range of 30° to 150° was fitted by a symmetrical Pearson VII function as shown exemplarily in Fig. 1 for the low-angle part of the GAXRD patterns of the Ti-Al-N monolayer coatings deposited at UB = -40 V and UB = -80 V. Further infor-mation about the phase composition was de-duced from the comparison of the stress-free lattice parameter of the fcc-(Ti,Al)N phase with the expected lattice parameter accord-ing to the Vegard-like dependence given in Eq. (6) [32, 38, 55].

(6)

Fig. 1: Low angle parts of the GAXRD patterns measured for the Ti1-xAlxN monolayer coatings that were deposited onto cemented carbide substrates at a bias voltage of -40 V (a) and -80 V (b). The measured data is shown by black points. The grey lines show the individual diffraction peaks as fitted by Pearson VII functions. The positions of the fcc-TiN, fcc-AlN and w-AlN diffraction lines are labelled at the bottom of the figure. The positions of fcc-TiC and hex-WC coming from the substrate are shown at the top of the figure.

The phase analysis revealed a different phase composition of the coatings that had the same chemical composition but were deposited at low bias voltages (UB = -20 V and UB = -40 V) and at high bias voltages (UB = -60 V, UB = -80 V and UB = -120 V).

The phase composition of the Ti-Al-N mon-olayer coatings is summarized in Fig. 2a. The Ti0.62Al0.38N and Ti0.53Al0.47N coatings depos-ited at UB = -20 V and UB = -40 V contained fcc-(Ti,Al)N as a single phase, since no dif-fraction lines of wurtzitic phase could be observed in the diffraction patterns. The for-mation of traces of wurtzite phase beside the major fcc-(Ti,Al)N phase could be recognized for an Al content of x = 0.56 (see Fig. 1a) at low bias voltages. If the Al content was fur-ther increased to x = 0.62, the wurtzite phase became the major phase at low bias voltages. The shift of the diffraction lines of the wurt-zite phase in the coatings from the expected

206

2θ positions of w-AlN with the lattice param-eter a = 0.311 nm and c = 0.498 nm could be attributed on the one hand to the expansion of the elementary cell of the wurtzite phase indi-cating the incorporation of Ti into the wurt-zite phase. The expansion of the elementary cell of the wurtzite phase was also observed in the monolithic Ti0.36Al0.64N and Ti0.39Al0.61N coatings deposited onto cutting inserts by a different CAE process [47]. On the other hand, the shift of the diffraction lines of the wurtzite phase could also be affected by resid-ual stress, which is expected to be lower in the coatings deposited at low UB than in the coat-ings deposited at high UB.

The application of a high UB during the dep-osition of Ti-Al-N coatings suppressed the formation of the wurtzite phase, because fcc-(Ti,Al)N was the major phase in all Ti-Al-N monolayer coatings that were depos-ited at UB = -80 V and UB = -120 V. At high bias voltages, traces of the wurtzite phase could be recognized in the diffraction pat-terns only for an Al content of x = 0.62 (see Fig. 1b). A similar behaviour could be ob-served in Ti1-xAlxN (0.61 ≤ x ≤ 0.64) coatings

deposited onto cutting inserts by another CAE process done at UB = -20 V, -40 V and -80 V [47]. In that case, the estimated phase fraction of the wurtzite phase decreased from (77 ± 6) mol% for UB = -20 V and UB = -40 V to only (7 ± 3) mol% for UB = -80 V. The re-duction of the volume fraction of w-AlN at high bias voltages was also observed by Pfeiler et al. for Ti-Al-N based coatings con-taining tantalum [56] and vanadium [57]. However in our samples [39], local fluctua-tions of the Al content led to the formation of an Al-rich fcc-(Al,Ti)N phase beside the major fcc-(Ti,Al)N phase in the Ti-Al-N monolayer coatings that were deposited at high bias voltages (-80 V and -120 V). This could be concluded from a slight asymmetry of the fcc-(Ti,Al)N diffraction lines to high diffraction angles. This asymmetry was fitted by another Pearson VII function (see Fig. 1b). The asymmetry is more visible for XRD lines with even diffraction indices than for odd indices [36]. This is attributed to the differ-ent magnitude of the structure factors (and thus different diffracted intensities) for fcc- Ti1-xAlxN with low and high Al contents. For diffraction lines with even and odd indices,

Fig. 2: Influence of the bias voltage and the [Ti]/([Ti]+[Al]) ratio on the phase composition (a) and on the stress-free lattice parameter (b) of the fcc-Ti1-xAlxN phase in the Ti1-xAlxN monolayer coatings. The coloured horizontal lines in (b) indicate the stress-free lattice parameter of the fcc-Ti1-xAlxN phase as calculated from the Vegard-like dependence (Eq.(6)) for the mean composition of the coatings determined by using EPMA / WDS (x = 0.38, 0.47, 0.56 and 0.62). The stress-free lattice parameters were obtained by using the modified sin²ψ method as described in [32].

207

(7)

the structure factors of the fcc-Ti1-xAlxN phase correspond to Eq.(7) and (8), respectively:

(8)

Furthermore, the stress-free lattice parameter of the fcc-(Ti,Al)N phase of the coatings de-posited at high UB was larger than the expected lattice parameter according to the Vegard-like dependence (Eq. (6)) for the respective coat-ing composition (Fig. 2b). This implies that the whole amount of Al present in the coating is not incorporated in the fcc-(Ti,Al)N phase, but some Al is present in an Al-rich phase [32].

In the case of the Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings, all three series contained fcc-(Ti,Al)N as major phase. No distinct dif-fraction lines from w-AlN appeared in the GAXRD patterns (Fig. 3). But the increased intensity between 32° and 35.2° 2θ in the GAXRD pattern measured in a laboratory ex-periment with Cu radiation suggests the for-mation of traces of wurtzitic phase as a second phase in all coatings, as shown exemplarily for the Ti-Al-N / Al-Ti-Ru-N multilayers of series III in Fig. 3a. The presence of traces of wurtzite phase in the coatings is better visible in the GAXRD patterns that were obtained with synchrotron radiation ( =0.10781 nm) at Rossendorf Beamline BM 20 at the Euro-pean Synchrotron Radiation Facility (ESRF) in Grenoble and which are shown in Fig. 3b.

Fig. 3: Low angle parts of the Ti-Al-N / Al-Ti-Ru-N multilayer coatings from series III deposited at different bias voltages that were obtained in a laboratory experiment using Cu radiation (a) and in a synchrotron experiment with a wavelength of λ=0.10781 nm (b). The positions of fcc-TiN, fcc-AlN and w-AlN are shown at the bottom of the figure, whereas the positions of hex-WC and fcc-TiC are given at the top of the figure. The green arrows indicate the increased intensity in the diffraction patterns caused by traces of wurtzite.

where , and are the atomicscattering factors of nitrogen, aluminium andtitanium, respectively. As ,the increasing Al content leads always to adecrease of the diffracted intensity, which ishowever stronger for diffraction lines withodd indices than for diffraction lines witheven indices. For the diffraction lines 200 and220, the ratio isnearly twice as high as compared to thediffraction line 111. For the atomic scatteringfactors calculated after the routine given byWaasmaier et al. [58], the above calculationyielded

, and

. Thus, the asymmetry is more visible for the200 and 220 diffraction lines than for the 111diffraction line.

208

The increased intensity between ~22° and ~25° 2θ as well as between ~38° and ~40.5° 2θ in the GAXRD patterns (Fig. 3b) is attributed to the wurtzite phase. The coatings deposited at high UB (-60 V and -80 V) contain apparently Al-rich fcc- (Al,Ti)N as a third phase, since an asymmetry of the fcc-(Ti,Al)N diffraction lines to high diffraction angles (cf. Fig. 3) was observed like in the case of the Ti-Al-N monolayer coatings. Furthermore, the stress-free lattice parameter of the fcc-(Ti,Al)N phase, which was calculated according to the routine giv-en in Ref. [49], was substantially increased as compared to the coatings deposited at low UB (see Fig. 4).

Fig. 4: Dependence of the stress-free lattice parameter of the fcc-(Ti,Al)N phase on the bias voltages for all three Ti-Al-N / Al-Ti-(Ru)-N multilayer series.

3.2 Influence of the bias voltage and Al content on the microstructure parameters in Ti-Al-N coatings

The phase composition, which was adjused by the bias voltage and by the aluminium con-tent, influences the crystallite size and the macroscopic lattice strain. In order to deter-mine the crystallite size, the dependence of the line broadening on the diffraction vector was analysed, which is shown exemplarily for the Ti0.53Al0.47N coatings deposited at different UB in Fig. 5. The dependence found for the

fcc-(Ti,Al)N phase was similar to that one shown in Ref. [32] and suggested the presence of partially coherent nanocrystallites with small mutual misorientations [8, 59].

The size of the fcc-(Ti,Al)N nanocrystal-lites was determined according to the rou-tine proposed in Ref. [59] and is shown in Fig. 6a. At the bottom of Fig. 6a, the phase composition is displayed, which indicates the correlation of the nanocrystallites with the phase composition of the coatings. The single-phase Ti0.62Al0.38N and Ti0.53Al0.47N coatings that were deposited at low bias volt-ages (-20 V and -40 V) contained the larg-est crystallites having a size of ~ 9 - 11 nm. Substantial smaller crystallites were found in the dual phase coatings. In Ti0.44Al0.56N coatings deposited at UB = -20 V and -40 V, which contained a minor amount of w-AlN as second phase, the size of the nanocrystal-lites was ~ 6 nm. When w-AlN became the major phase in the Ti0.38Al0.62N coatings with the highest Al content that were deposited at low bias voltages, the size of the fcc-(Ti,Al)N nanocrystallites could not be determined, be-cause the diffraction lines from fcc-(Ti,Al)N were too broad and too weak. The dual-phase Ti1-xAlxN coatings with 0.38 ≤ x ≤ 0.62 and deposited at high UB (-80 V and -120 V), which contained Al-rich fcc-(Al,Ti)N as second

209

Fig. 5: Dependence of the integral breadth of XRD lines of the fcc-(Ti,Al)N phase on the modulus of the diffraction vector for the Ti0.53Al0.47N coatings deposited at different UB.

crystalline phase, were characterized by crys-tallites having a size of ~ 3.5 - 5 nm. If w-AlN was formed as third phase in these coatings the size of the crystallites became ~ 3 nm.

The macroscopic lattice strain in the fcc-(Ti,Al)N phase, which was determined ac-cording to the procedure described in Ref. [32], is given in Fig. 6b for all Ti1-xAlxN coat-ings that contained fcc-(Ti,Al)N as major phase. Assuming a Poisson’s ratio of 0.3 and a Young’s modulus of 500 GPa, the macroscopic lattice strain in Fig. 6b was transformed into the residual stress (z axis on the right-hand side) [32].

It can be seen that the bias voltage strong-ly influences the stress state of the coatings. Low tensile stress (< 0.5 GPa) and almost zero residual stress were observed in the Ti1-xAlxN coatings deposited at UB = -20 V. When the bias voltage was increased to -40 V, -80 V and -120 V the coatings were character-ized by a compressive stress state. The stress rose by a factor of ~ 4 for x ≤ 0.56 when the bias voltage was increased from -40 V to -80 V. A further increase of the bias voltage from -80 V to -120 V resulted in a minor stress increase for x ≤ 0.47 and in a saturation of the stress for x ≥ 0.56, respectively. The in-crease of the residual stress with increasing UB to -80 V is partially caused by an increasing

kinetic energy of the bombarding ions giving rise to a higher density of lattice defects, e.g., displaced atoms due to direct and recoil im-plantation of film atoms [60]. At UB = -120 V, partial defect annihilation obviously takes place as indicated by the minor increase or rather saturation of the stress as compared to UB = -80 V. According to Ref. [60], this effect is apparently facilitated by vibration and gen-eration of phonons in the vicinity of the colli-sion cascade as a result of the delivery of en-ergy of the impinging ions to the surface. This leads to short range movements of atoms and relaxation of displaced atoms. Additionally local heating of the deposited coating takes place at a high ion flux leading to an increased annealing rate of the created defects [60]. However, the high residual stress in the coat-ings deposited at high UB (-80 V and -120 V) stabilizes Al-rich fcc-(Al,Ti)N which is in ac-cordance with ab initio calculations [61].

The microstructure formation designed by the bias voltage and by the Al content influ-ences the indentation hardness of the coatings as shown in Fig. 6c. The highest hardness was found in the Ti1-xAlxN coatings with x ≤ 0.56 that were deposited at high UB and contained fcc-(Ti,Al)N and Al-rich fcc-(Al,Ti)N. The high hardness in the range of 33 to 37 GPa is achieved by small nanocrystallites in the range of 3.5 to 5 nm and by a high compressive

210

Fig. 6: Influence of the bias voltage and the [Ti]/([Ti]+[Al]) ratio on the crystallite size of the fcc-(Ti,Al)N phase (a), the macroscopic lattice strain and residual stress of the fcc-(Ti,Al)N phase (b) and the indentation hardness (c). At the bottom of the figures the phase composition is illustrated, the symbols have the same meaning like in Fig. 2a.

stress in the range of 6 to 8 GPa. In the coat-ings with the highest Al content (Ti0.38Al0.62N), the hardness was reduced by 5 and 3 GPa for UB = -80 V and UB = -120 V, respectively, as compared to the Ti1-xAlxN coatings with lower Al content and the same bias voltage in each case. In the coatings with the reduced hard-ness, the smallest crystallites were found, which could facilitate increased grain bound-ary sliding [62] and thus the decrease of the hardness. Another possible explanation for the hardness reduction could be the pres-ence of the wurtzite phase exceeding a critical amount, because the volume fraction of the wurtzite phase in the Ti0.38Al0.62N coating was estimated to be (25 ± 10) vol%. A similar hard-ness evolution with different amount of molar fraction of w-AlN was observed in monolithic Ti1-xAlxN (0.61 ≤ x ≤ 0.64) coatings deposit-ed onto cutting inserts at UB = -80 V [47]. The Ti1-xAlxN coatings that contained a low amount of wurtzite phase namely (7 ± 3) mol% had a higher hardness of (31.1 ± 1) GPa than the coatings containing a high amount of wurtzite phase of (38 ± 4) mol% which had a hard-ness of (27.7 ± 0.9) GPa.

Low compressive stress (< 2 GPa) and large crystallites (~ 9 - 11 nm) in single-phase fcc-(Ti,Al)N coatings, which were deposited at low bias voltages, resulted in low hardness of 25 GPa to 30 GPa. A hardness increase by ~ 3 GPa was observed in the Ti0.44Al0.56N coatings deposited at low bias voltages, which contained a low amount of wurtzite phase (approx. (14 ± 5) vol%) as second crystalline phase. This hardness increase can be attribut-ed to a slight increase of the compressive stress (Fig. 6b) and to the reduction of the crystallite size (Fig. 6a) as compared to the single phase coatings deposited at the same bias voltage. This observed hardness increase emphasiz-es the positive influence of low amounts of wurtzite phase in the Ti-Al-N based coat-ings with regard to their hardness. The posi-tive effect of small amounts of w-AlN on the hardness evolution was also shown for anoth-er Ti-Al-N coating series deposited by CAE [37, 55]. This effect is facilitated by partially coherent interfaces between fcc-(Ti,Al)N and

wurtzite phase, which will be discussed in the next Section (3.3). The hardness dropped to the lowest hardness values ranging between 21 and 23 GPa in the Ti0.38Al0.62N coatings de-posited at low bias voltages, which contained the wurtzite phase as major phase. This effect is caused by the predominance of the softer w-AlN phase.

211

Fig. 7: (a) HRTEM of the Ti0.44Al0.56N monolayer coating deposited at UB = -40 V and local FFTs revealing the orienta-tion of fcc-(Ti,Al)N (yellow marked) and wurtzite phase (orange marked) that follows from the indices of the diffrac-tion spots in the simulated electron diffraction patterns (b).

3.3 Orientation relationships between fcc-(Ti,Al)N and wurtzite phase

Several experimentally observed orientationrelationships between partially coherent fcc-(Ti,Al)N and w-AlN are published inliterature [36, 37, 63, 64]. HRTEM done onthe Ti0.44Al0.56N coating deposited at UB = -40 V and local FFTs (Fig. 7) revealed afurther orientation relationship betweenfcc-(Ti,Al)N and w-AlN. Local FFTsperformed in the designated square areas inFig. 7a revealed fcc- (Ti,Al)N with the

direction being parallel to theincident electron beam and the wurtzitephase with the direction beingparallel to the incident electron beam. TheFFTs can be indexed with the simulatedelectron diffraction patterns given inFig. 7b. The OR between both phases ischaracterized by the planes beingparallel to the planes and the

direction being parallel to thedirection (see Fig. 8).

parallel to the planes and thedirection being parallel to the

direction (see Fig. 8).

Obviously this OR represents a preferredone, because it was found in two furthersamples namely in the Ti-Al-N / Al-Ti-Ru-Nmultilayer deposited at UB = -80 V that wasthermally treated at 950 °C and in the virginTi-Al-N / Al-Ti-N multilayer coating depo-sited at UB = -40 V [65]. This OR is charac-terized by a moderate misfit of ~ 3.6 % alongthe and direction (seeFig. 8) regarding the misfit between the Tiand Al atoms. But the misfit along the

and direction is ~ 16 %and huge. It is probably compen-sated bymicrostructure defects.

Obviously this OR represents a preferredone, because it was found in two furthersamples namely in the Ti-Al-N / Al-Ti-Ru-Nmultilayer deposited at UB = -80 V that wasthermally treated at 950 °C and in the virginTi-Al-N / Al-Ti-N multilayer coating depo-sited at UB = -40 V [65]. This OR ischaracterized by a moderate misfit of ~ 3.6 %along the and direction(see Fig. 8) regarding the misfit between theTi and Al atoms. But the misfit along the

and direction is ~ 16 % andhuge. It is probably compensated by micro-structure defects.

212

Obviously this OR represents a preferredone, because it was found in two furthersamples namely in the Ti-Al-N / Al-Ti-Ru-N multilayer deposited at UB = -80 V thatwas thermally treated at 950 °C and in thevirgin Ti-Al-N / Al-Ti-N multilayer coatingdeposited at UB = -40 V [65]. This OR ischaracterized by a moderate misfit of~ 3.6 % along the anddirection (see Fig. 8) regarding the misfitbetween the Ti and Al atoms. But the misfitalong the and direction is~ 16 % and huge. It is probablycompensated by microstructure defects.We sought to study the interface along the

and direction in moredetails by the deposition of a w-AlN filmwith a thickness of ~110 nm on top of a

orientated TiN single crystallinelayer using magnetron sputtering [65]. Thisfcc-TiN single crystalline layer was obtainedby heteroepitaxial growth on top of a

orientated MgO substrate. Furtherdetails and illustrations of the experimentaiming the heteroepitaxial growth of w-AlNon top of fcc-TiN using magnetronsputtering as deposition technology aregiven in Ref. [65]. After deposition a TEMcross-section sample was prepared with the

direction being parallel to thenormal of the TEM foil. The analysis of theORs using selected area electron diffraction(SAED) revealed two major types of AlNcolumns with a further OR [65], which wasnot published before. The observed SAEDpattern of the interface fcc-TiN / w-AlN wassimilar to the simulated one in Fig. 9. Theresults of the analysis can be summarizedbriefly as follows. Two w-AlN column typeswere found, which alternate every < 10 nmalong the flat TiN interface being parallel to

. The OR of the first w-AlN columnis characterized by and

(see red spots in Fig. 9).The second w-AlN column is created by therotation of the first w-AlN crystal (w1) by180° around the normal of the interfaceso that the direction is parallel tothe direction and theplane is parallel to the plane (seeblue spots in Fig. 9). The above mentionedOR that was observed in the Ti-Al-N basedcoatings can be suspected as a thirdorientation which appears only in a smallvolume fraction in this experiment [65].This illustrates that the material is veryflexible by the creation of partially coherentinterfaces offering a variety of orientationrelationships which could be beneficial forthe hardness development.

Obviously this OR represents a preferredone, because it was found in two furthersamples namely in the Ti-Al-N / Al-Ti-Ru-N multilayer deposited at UB = -80 V thatwas thermally treated at 950 °C and in thevirgin Ti-Al-N / Al-Ti-N multilayer coatingdeposited at UB = -40 V [65]. This OR ischaracterized by a moderate misfit of~ 3.6 % along the anddirection (see Fig. 8) regarding the misfitbetween the Ti and Al atoms. But the misfitalong the and direction is~ 16 % and huge. It is probablycompensated by microstructure defects.We sought to study the interface along the

and direction in moredetails by the deposition of a w-AlN filmwith a thickness of ~110 nm on top of a

orientated TiN single crystallinelayer using magnetron sputtering [65]. Thisfcc-TiN single crystalline layer was obtainedby heteroepitaxial growth on top of a

orientated MgO substrate. Furtherdetails and illustrations of the experimentaiming the heteroepitaxial growth of w-AlNon top of fcc-TiN using magnetronsputtering as deposition technology aregiven in Ref. [65]. After deposition a TEMcross-section sample was prepared with the

direction being parallel to thenormal of the TEM foil. The analysis of theORs using selected area electron diffraction(SAED) revealed two major types of AlNcolumns with a further OR [65], which wasnot published before. The observed SAEDpattern of the interface fcc-TiN / w-AlN wassimilar to the simulated one in Fig. 9. Theresults of the analysis can be summarizedbriefly as follows. Two w-AlN column typeswere found, which alternate every < 10 nmalong the flat TiN interface being parallel to

. The OR of the first w-AlN columnis characterized by and

(see red spots in Fig. 9).The second w-AlN column is created by therotation of the first w-AlN crystal (w1) by180° around the normal of the interfaceso that the direction is parallel tothe direction and theplane is parallel to the plane (seeblue spots in Fig. 9). The above mentionedOR that was observed in the Ti-Al-N basedcoatings can be suspected as a thirdorientation which appears only in a smallvolume fraction in this experiment [65].This illustrates that the material is veryflexible by the creation of partially coherentinterfaces offering a variety of orientationrelationships which could be beneficial forthe hardness development.

Fig. 9: Simulated electron diffractionpatterns of an OR found at theinterface of fcc-TiN (white spots)and w-AlN with two differentorientations of w-AlN (w1 = red,w2 = blue). The ,and directions areperpendicular to the plane of thepaper. The black broken line is || tothe interface and the arrowcorresponds to the normal of theinterface. The open squarescorrespond to forbidden reflectionsin the wurtzite crystallites.

Fig. 8: Schematic presentation of fcc-TiN andw-AlN in the mutual ORand . The directionsand are facing downwards. Theinteratomic distances within certain latticeplanes in fcc-TiN and w-AlN are marked bygrey dashed lines. The elementary cells arehighlighted by the broken blue lines. The Alatoms are plotted by orange, the Ti atoms byyellow and the N atoms by green spheres.Spheres that are hollow hatched representatoms lying below the plane of the paper.

213

3.4 Influence of the bias voltage and the Ru content on the microstructure parameters in Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings

coatings can be suspected as a thirdorientation which appears only in a smallvolume fraction in this experiment [65].This illustrates that the material is veryflexible by the creation of partially coherentinterfaces offering a variety of orientationrelationships which could be beneficial forthe hardness development.

Another OR being important in Ti-Al-Nbased coatings is characterized by

and that isexplained in detail in Refs. [36, 55]. Thisorientation offers a good match of bothcrystal structures with a low misfit along the

and direction and alongthe and direction. Themisfit between the w-AlN phase(a = 0.311 nm, c = 0.498 nm) and the fcc-(Ti,Al)N phase depends on the latticeparameter of the fcc-(Ti,Al)N phase.Considering the interatomic distance of themetal atoms in the anddirections, it ranges between 3.6 % for fcc-TiN with a = 0.424 nm and 7 % for fcc-AlNwith a = 0.410 nm. In case of theand directions, the misfit rangesbetween 1.6 % and 4.9 %. With regard to theabove mentioned OR and to in situdiffraction experiments during annealing ofTi1-xAlxN coatings, it is suggested in Ref.[36] that a phase transformation of fcc-(Ti,Al)N to the wurtzite phase is facilitatedby the formation of stacking faults.

The stacking faults are located on theplanes in order to transform the

fcc phase to the wurtzite phase. The faultedfcc crystal is seen by XRD as a faultedwurtzite structure with the stacking faultson the planes. As a result of thefaulting, the diffraction lines 100w, 110w and200w are the most pronounced ones in thediffraction pattern, whereas the diffractionlines 101w, 102w, 103w and 202w are verybroadened due to microstructure defectslike stacking faults and the correspondingpartial dislocations [36]. The increasedlattice parameters of the wurtzite phaseindicated the incorporation of Ti into w-(Al,Ti)N, which in turn facilitated theformation of the stacking faults. This kindof phase transition would preserve thecoherence of the adjacent wurtzite and fccphase..

coatings can be suspected as a thirdorientation which appears only in a smallvolume fraction in this experiment [65].This illustrates that the material is veryflexible by the creation of partially coherentinterfaces offering a variety of orientationrelationships which could be beneficial forthe hardness development.

Another OR being important in Ti-Al-Nbased coatings is characterized by

and that isexplained in detail in Refs. [36, 55]. Thisorientation offers a good match of bothcrystal structures with a low misfit along the

and direction and alongthe and direction. Themisfit between the w-AlN phase(a = 0.311 nm, c = 0.498 nm) and the fcc-(Ti,Al)N phase depends on the latticeparameter of the fcc-(Ti,Al)N phase.Considering the interatomic distance of themetal atoms in the anddirections, it ranges between 3.6 % for fcc-TiN with a = 0.424 nm and 7 % for fcc-AlNwith a = 0.410 nm. In case of theand directions, the misfit rangesbetween 1.6 % and 4.9 %. With regard to theabove mentioned OR and to in situdiffraction experiments during annealing ofTi1-xAlxN coatings, it is suggested in Ref.[36] that a phase transformation of fcc-(Ti,Al)N to the wurtzite phase is facilitatedby the formation of stacking faults.

The stacking faults are located on theplanes in order to transform the

fcc phase to the wurtzite phase. The faultedfcc crystal is seen by XRD as a faultedwurtzite structure with the stacking faultson the planes. As a result of thefaulting, the diffraction lines 100w, 110w and200w are the most pronounced ones in thediffraction pattern, whereas the diffractionlines 101w, 102w, 103w and 202w are verybroadened due to microstructure defectslike stacking faults and the correspondingpartial dislocations [36]. The increasedlattice parameters of the wurtzite phaseindicated the incorporation of Ti into w-(Al,Ti)N, which in turn facilitated theformation of the stacking faults. This kindof phase transition would preserve thecoherence of the adjacent wurtzite and fccphase..

Information about the multilayerarchitecture of the Ti-Al-N / Al-Ti-(Ru)-Nmultilayers were obtained by analyticalscanning (S-) TEM. The layeredarchitecture of the coatings could bevisualized by dark field (DF) STEM imagesas shown in Fig. 10a and Fig. 10b for theRu-free Ti-Al-N / Al-Ti-N multilayer ofseries I and the Ti-Al-N / Al-Ti-Ru-Nmultilayer of series III, respectively. Thelength of the periodic motif estimated fromthe DF STEM images is ~ 21 nm for the Ti-Al-N / Al-Ti-N multilayer of series Ideposited at UB = -40 V and ~ 29 nm for theTi-Al-N / Al-Ti-Ru-N multilayer of seriesIII deposited at UB = -40 V.

The contrasts in the DF STEM micrographsreflect the changes in the mean atomicweight across the layers. In case of aconstant TEM foil thickness, the regionswith a high mean atomic weight appearbrighter than the areas with a lower meanatomic weight. The contrasts differ slightlyfor the coatings with Ru doping and withoutRu addition. Thin bright layers labelled as“A” are visible in the Ti-Al-N / Al-Ti-Ru-Nmultilayer (Fig. 10b) and indicate a highmean atomic weight. These layers were notvisible in the Ru-free multilayer (Fig. 10a).The X-ray spectroscopy (EDS) on the Ti Kα,Al Kα and Ru Kα lines, done in STEMacross the layer stacks with a step size of1 nm, gave information about the variationof the metal ratios (see Fig. 10c).Additionally, the measured intensity of theRu Kα line is given as blue dotted line inFig. 10c. The concentration profile given inFig. 10c revealed the gradual change in theconcentration of the metal species, which isattributed to the 2-fold rotation during CAEdeposition. The EDS analysis also showedthat the layers “A” are attributed to anenrichment of Ru. In these layers also thehighest Al ratio of ~ 0.65 was found,because Ru was deposited from the Al-richtargets. It is assumed that Ru is present in itsmetallic form because the depositiontemperature of 450 °C was much higherthan the decomposition temperature ofRuN that is approx. 100 °C [66]. The highest[Ti] / Σ[Me] ratio is located in the layerslabelled as “B” and approaches 0.56. EELSanalysis of the Ti L2,3 edge (see Fig. 10d)confirmed the fluctuation of the Ticoncentration observed by EDS.Furthermore, EELS analysis of the N K edgeverified the constant amount of nitrogen(Fig. 10d) across the individual layers.Further information about the EELSanalysis in Ti-Al-N based coatings is givenin Ref. [50].

In the Ti-Al-N / Al-Ti-Ru-N multilayer ofthe series III deposited at UB = -40 V, themedium [Ti] / Σ[Me], [Al] / Σ[Me] and[Ru] / Σ[Me] ratios averaged over threeperiodic layer stacks were (0.44 ± 0.01),(0.55 ± 0.01) and (0.01 ± 0.005),respectively. These ratios are in goodagreement with the metal ratios determinedby EPMA / WDS (see Tab. 1).

Since GAXRD indicated traces of wurtzitephase in all coatings (see Section 3.1), theformation of w-AlN in the Al-rich layerswould be possible. The highest Al ratio inthese layers is 0.65, which is above thecritical Al ratio for the formation of the fccsingle phase coatings that was found insimilarly deposited Ti1-xAlxN monolayercoatings (cf. Fig. 2a). Furthermore, this Alconcentration is close to the highestexperimentally observed solubility limit ofAl (x = 0.67) in fcc-Ti1-xAlxN [15, 16].However, HRTEM in combination withFFT across the multilayer stack revealedthat the fcc crystal structure of (Ti,Al)N isbasically maintained over the Ti-rich andAl-rich layers. Although phases with smallvolume fractions of w-AlN are difficult tobe identified by TEM, wurtzite phase in theTi-Al-N / Al-Ti-Ru-N multilayers wasfound at the column boundaries.Furthermore, the wurtzite phase waspartially coherent to the fcc-phase with thesame OR that is shown in Fig. 8, namely:

|| and || .

The macroscopic residual stress of the fcc-(Ti,Al)N phase was determined accordingto the routine proposed in Ref. [49] and isshown in Fig. 11a. For the calculation of theresidual stress, a Young’s modulus of500 GPa and Poisson’s ratio of 0.3 wereassumed. Additionally, the macroscopiclattice strain was calculated. The stressanalysis revealed that all coatings arecharacterized by a compressive stress state.The addition of Ru had no influence on thestress evolution, because the stress of allthree coating series, which were depositedat the same bias voltage, differed onlywithin the expected experimental error. Themacroscopic residual stress as well as thelattice strain is primarily influenced by thebias voltage applied during the CAEdeposition. At low bias voltage of UB = -20 V, the lowest compressive stress of~ 1 GPa was observed. The compressivestress increased nearly linearly to ~ 6.5 GPawith increasing bias voltage from -20 V to -60 V, which can be attributed to anincreasing kinetic energy of the bombardingions giving rise to a higher density of latticedefects [60]. The compressive stressincreased only slightly and saturated at~ 7 GPa, when the bias voltage wasincreased from -60 V to -80 V. The sameeffect was observed in the Ti1-xAlxNmonolayers, when the bias voltage wasincreased from -80 V to -120 V. Thesaturation of the residual stress at thehighest bias voltages was attributed to localheating of the deposited coating as a resultof intensive ion bombardment leading topartial defect annihilation e.g. relaxation ofdisplaced atoms (see Section 3.2)..

coatings can be suspected as a thirdorientation which appears only in a smallvolume fraction in this experiment [65].This illustrates that the material is veryflexible by the creation of partially coherentinterfaces offering a variety of orientationrelationships which could be beneficial forthe hardness development.

Another OR being important in Ti-Al-Nbased coatings is characterized by

and that isexplained in detail in Refs. [36, 55]. Thisorientation offers a good match of bothcrystal structures with a low misfit along the

and direction and alongthe and direction. Themisfit between the w-AlN phase(a = 0.311 nm, c = 0.498 nm) and the fcc-(Ti,Al)N phase depends on the latticeparameter of the fcc-(Ti,Al)N phase.Considering the interatomic distance of themetal atoms in the anddirections, it ranges between 3.6 % for fcc-TiN with a = 0.424 nm and 7 % for fcc-AlNwith a = 0.410 nm. In case of theand directions, the misfit rangesbetween 1.6 % and 4.9 %. With regard to theabove mentioned OR and to in situdiffraction experiments during annealing ofTi1-xAlxN coatings, it is suggested in Ref.[36] that a phase transformation of fcc-(Ti,Al)N to the wurtzite phase is facilitatedby the formation of stacking faults.

The stacking faults are located on theplanes in order to transform the

fcc phase to the wurtzite phase. The faultedfcc crystal is seen by XRD as a faultedwurtzite structure with the stacking faultson the planes. As a result of thefaulting, the diffraction lines 100w, 110w and200w are the most pronounced ones in thediffraction pattern, whereas the diffractionlines 101w, 102w, 103w and 202w are verybroadened due to microstructure defectslike stacking faults and the correspondingpartial dislocations [36]. The increasedlattice parameters of the wurtzite phaseindicated the incorporation of Ti into w-(Al,Ti)N, which in turn facilitated theformation of the stacking faults. This kindof phase transition would preserve thecoherence of the adjacent wurtzite and fccphase..

Information about the multilayer architec-ture of the Ti-Al-N / Al-Ti-(Ru)-N multi-layers were obtained by analytical scanning(S-) TEM. The layered architecture of thecoatings could be visualized by dark field(DF) STEM images as shown in Fig. 10aand Fig. 10b for the Ru-free Ti-Al-N / Al-Ti-N multilayer of series I and theTi-Al-N / Al-Ti-Ru-N multilayer of seriesIII, respectively. The length of the periodicmotif estimated from the DF STEM imagesis ~ 21 nm for the Ti-Al-N / Al-Ti-Nmultilayer of series I deposited at UB = -40 V and ~ 29 nm for the Ti-Al-N / Al-Ti-Ru-N multilayer of series III depo-sited at UB = -40 V.

214

Fig. 10: DF STEM images of the Ti-Al-N / Al-Ti-(Ru)-N multilayers of the coating series I (a) and III (b) deposited at UB = -40 V. The white line in (b) shows the position of the analytical TEM line scan. The results of EDS and EELS analysis are given in (c) and (d), respectively. The thin blue dotted line in (c) shows the intensity of the Ru Kα line.

Information about the multilayerarchitecture of the Ti-Al-N / Al-Ti-(Ru)-Nmultilayers were obtained by analyticalscanning (S-) TEM. The layeredarchitecture of the coatings could bevisualized by dark field (DF) STEM imagesas shown in Fig. 10a and Fig. 10b for theRu-free Ti-Al-N / Al-Ti-N multilayer ofseries I and the Ti-Al-N / Al-Ti-Ru-Nmultilayer of series III, respectively. Thelength of the periodic motif estimated fromthe DF STEM images is ~ 21 nm for the Ti-Al-N / Al-Ti-N multilayer of series Ideposited at UB = -40 V and ~ 29 nm for theTi-Al-N / Al-Ti-Ru-N multilayer of seriesIII deposited at UB = -40 V.

The contrasts in the DF STEM micrographsreflect the changes in the mean atomicweight across the layers. In case of aconstant TEM foil thickness, the regionswith a high mean atomic weight appearbrighter than the areas with a lower meanatomic weight. The contrasts differ slightlyfor the coatings with Ru doping and withoutRu addition. Thin bright layers labelled as“A” are visible in the Ti-Al-N / Al-Ti-Ru-Nmultilayer (Fig. 10b) and indicate a highmean atomic weight. These layers were notvisible in the Ru-free multilayer (Fig. 10a).The X-ray spectroscopy (EDS) on the Ti Kα,Al Kα and Ru Kα lines, done in STEMacross the layer stacks with a step size of1 nm, gave information about the variationof the metal ratios (see Fig. 10c).Additionally, the measured intensity of theRu Kα line is given as blue dotted line inFig. 10c. The concentration profile given inFig. 10c revealed the gradual change in theconcentration of the metal species, which isattributed to the 2-fold rotation during CAEdeposition. The EDS analysis also showedthat the layers “A” are attributed to anenrichment of Ru. In these layers also thehighest Al ratio of ~ 0.65 was found,because Ru was deposited from the Al-richtargets. It is assumed that Ru is present in itsmetallic form because the depositiontemperature of 450 °C was much higherthan the decomposition temperature ofRuN that is approx. 100 °C [66]. The highest[Ti] / Σ[Me] ratio is located in the layerslabelled as “B” and approaches 0.56. EELSanalysis of the Ti L2,3 edge (see Fig. 10d)confirmed the fluctuation of the Ticoncentration observed by EDS.Furthermore, EELS analysis of the N K edgeverified the constant amount of nitrogen(Fig. 10d) across the individual layers.Further information about the EELSanalysis in Ti-Al-N based coatings is givenin Ref. [50].

In the Ti-Al-N / Al-Ti-Ru-N multilayer ofthe series III deposited at UB = -40 V, themedium [Ti] / Σ[Me], [Al] / Σ[Me] and[Ru] / Σ[Me] ratios averaged over threeperiodic layer stacks were (0.44 ± 0.01),(0.55 ± 0.01) and (0.01 ± 0.005),respectively. These ratios are in goodagreement with the metal ratios determinedby EPMA / WDS (see Tab. 1).

Since GAXRD indicated traces of wurtzitephase in all coatings (see Section 3.1), theformation of w-AlN in the Al-rich layerswould be possible. The highest Al ratio inthese layers is 0.65, which is above thecritical Al ratio for the formation of the fccsingle phase coatings that was found insimilarly deposited Ti1-xAlxN monolayercoatings (cf. Fig. 2a). Furthermore, this Alconcentration is close to the highestexperimentally observed solubility limit ofAl (x = 0.67) in fcc-Ti1-xAlxN [15, 16].However, HRTEM in combination withFFT across the multilayer stack revealedthat the fcc crystal structure of (Ti,Al)N isbasically maintained over the Ti-rich andAl-rich layers. Although phases with smallvolume fractions of w-AlN are difficult tobe identified by TEM, wurtzite phase in theTi-Al-N / Al-Ti-Ru-N multilayers wasfound at the column boundaries.Furthermore, the wurtzite phase waspartially coherent to the fcc-phase with thesame OR that is shown in Fig. 8, namely:

|| and || .

The macroscopic residual stress of the fcc-(Ti,Al)N phase was determined accordingto the routine proposed in Ref. [49] and isshown in Fig. 11a. For the calculation of theresidual stress, a Young’s modulus of500 GPa and Poisson’s ratio of 0.3 wereassumed. Additionally, the macroscopiclattice strain was calculated. The stressanalysis revealed that all coatings arecharacterized by a compressive stress state.The addition of Ru had no influence on thestress evolution, because the stress of allthree coating series, which were depositedat the same bias voltage, differed onlywithin the expected experimental error. Themacroscopic residual stress as well as thelattice strain is primarily influenced by thebias voltage applied during the CAEdeposition. At low bias voltage of UB = -20 V, the lowest compressive stress of~ 1 GPa was observed. The compressivestress increased nearly linearly to ~ 6.5 GPawith increasing bias voltage from -20 V to -60 V, which can be attributed to anincreasing kinetic energy of the bombardingions giving rise to a higher density of latticedefects [60]. The compressive stressincreased only slightly and saturated at~ 7 GPa, when the bias voltage wasincreased from -60 V to -80 V. The sameeffect was observed in the Ti1-xAlxNmonolayers, when the bias voltage wasincreased from -80 V to -120 V. Thesaturation of the residual stress at thehighest bias voltages was attributed to localheating of the deposited coating as a resultof intensive ion bombardment leading topartial defect annihilation e.g. relaxation ofdisplaced atoms (see Section 3.2)..

FFT across the multilayer stack revealed thatthe fcc crystal structure of (Ti,Al)N isbasically maintained over the Ti-rich andAl-rich layers. Although phases with smallvolume fractions of w-AlN are difficult to beidentified by TEM, wurtzite phase in theTi-Al-N / Al-Ti-Ru-N multilayers was foundat the column boundaries. Furthermore, thewurtzite phase was partially coherent to thefcc-phase with the same OR that is shown inFig. 8, namely: || and

|| .

215

Fig. 11: Influence of the bias voltage and the Ru addition on the residual stress as well as macroscopic lattice strain (a) and indentation hardness (b).

The macroscopic residual stress of the fcc-(Ti,Al)N phase was determined according tothe routine proposed in Ref. [49] and isshown in Fig. 11a. For the calculation of theresidual stress, a Young’s modulus of500 GPa and Poisson’s ratio of 0.3 wereassumed. Additionally, the macroscopiclattice strain was calculated. The stressanalysis revealed that all coatings arecharacterized by a compressive stress state.The addition of Ru had no influence on thestress evolution, because the stress of allthree coating series, which were deposited atthe same bias voltage, differed only withinthe expected experimental error. Themacroscopic residual stress as well as thelattice strain is primarily influenced by thebias voltage applied during the CAEdeposition. At low bias voltage of UB = -20 V,the lowest compressive stress of ~ 1 GPa wasobserved. The compressive stress increasednearly linearly to ~ 6.5 GPa with increasingbias voltage from -20 V to -60 V, which canbe attributed to an increasing kinetic energyof the bombarding ions giving rise to ahigher density of lattice defects [60].

The compressive stress increased onlyslightly and saturated at ~ 7 GPa, when thebias voltage was increased from -60 V to-80 V. The same effect was observed in theTi1-xAlxN monolayers, when the bias voltagewas increased from -80 V to -120 V. Thesaturation of the residual stress at the highestbias voltages was attributed to local heatingof the deposited coating as a result ofintensive ion bombardment leading topartial defect annihilation e.g. relaxation ofdisplaced atoms (see Section 3.2).

The indentation hardness of the Ti-Al-N/ Al-Ti-(Ru)-N multilayers is given in Fig. 11b.The hardness evolution vs. UB can beclassified into two groups. The first groupcomprises the coatings deposited at low biasvoltages (-20 V and -40 V). Their hardnesslies in the range between 30 and 32 GPa. Thesecond group contains the coatingsdeposited at high bias voltage (-60 V and-80 V). This group is characterized by thehighest hardness values lying between 33 and35 GPa. The increasing hardness with risingbias voltage indicates the contribution of thehigher compressive stress to the hardnessenhancement in the coatings. It seems thatthe Ru addition had no remarkable effect onthe hardness. The indentation hardnessdiffers for different Ru additions within therange of errors.

The macroscopic residual stress of the fcc-(Ti,Al)N phase was determined according tothe routine proposed in Ref. [49] and isshown in Fig. 11a. For the calculation of theresidual stress, a Young’s modulus of500 GPa and Poisson’s ratio of 0.3 wereassumed. Additionally, the macroscopiclattice strain was calculated. The stressanalysis revealed that all coatings arecharacterized by a compressive stress state.The addition of Ru had no influence on thestress evolution, because the stress of allthree coating series, which were deposited atthe same bias voltage, differed only withinthe expected experimental error. Themacroscopic residual stress as well as thelattice strain is primarily influenced by thebias voltage applied during the CAEdeposition. At low bias voltage of UB = -20 V,the lowest compressive stress of ~ 1 GPa wasobserved. The compressive stress increasednearly linearly to ~ 6.5 GPa with increasingbias voltage from -20 V to -60 V, which canbe attributed to an increasing kinetic energyof the bombarding ions giving rise to ahigher density of lattice defects [60].

The compressive stress increased onlyslightly and saturated at ~ 7 GPa, when thebias voltage was increased from -60 V to-80 V. The same effect was observed in theTi1-xAlxN monolayers, when the bias voltagewas increased from -80 V to -120 V. Thesaturation of the residual stress at the highestbias voltages was attributed to local heatingof the deposited coating as a result ofintensive ion bombardment leading topartial defect annihilation e.g. relaxation ofdisplaced atoms (see Section 3.2).

The indentation hardness of the Ti-Al-N/ Al-Ti-(Ru)-N multilayers is given in Fig. 11b.The hardness evolution vs. UB can beclassified into two groups. The first groupcomprises the coatings deposited at low biasvoltages (-20 V and -40 V). Their hardnesslies in the range between 30 and 32 GPa. Thesecond group contains the coatingsdeposited at high bias voltage (-60 V and-80 V). This group is characterized by thehighest hardness values lying between 33 and35 GPa. The increasing hardness with risingbias voltage indicates the contribution of thehigher compressive stress to the hardnessenhancement in the coatings. It seems thatthe Ru addition had no remarkable effect onthe hardness. The indentation hardnessdiffers for different Ru additions within therange of errors.

216

Fig. 12: Summary of the phase composition in the as-deposited coatings (a) and after annealing at 850 °C as estimated from Rietveld analysis (b). The lines are guide for the eyes.

3.5 Thermal stability of Ti-Al-N based coatings

The thermal induced microstructure chang-es were studied by in situ high temperature GAXRD experiments at the Rossendorf Beamline BM 20 at the ESRF and are de-scribed in detail in terms of the stress-free lattice parameter, microstrain, macroscopic lattice strain and phase composition dur-ing annealing of Ti1-xAlxN coatings in Refs. [32, 33, 36]. The results of these analyses re-vealed that the bias voltage applied during the deposition influences also the thermal stability of the coatings. During the annealing at 850 °C, the Ti1-xAlxN coatings with 0.38 ≤ x ≤ 0.56 decomposed into Al deplet-ed fcc-(Ti,Al)N, fcc-AlN and wurtzite phase. It was found that the bias voltage controls the relative amount of fcc-AlN and wurtzite phase, whereas the Al content determines the amount of fcc-(Ti,Al)N. The effect of the bias voltage and the chemical composition on the development of the phase composition of the Ti1-xAlxN coatings in course of the thermal treatment up to 850 °C is presented in a com-pact form in Fig. 12. The changes are illus-

trated by the phase composition in the as-de-posited state (Fig. 12a) and after annealing at 850 °C (Fig. 12b) for three different coating compositions (x = 0.38, 0.47, 0.56) and three bias voltages (UB = -40 V, -80 V and -120 V).

As discussed in Section 3.1, the bias volt-age influences the phase composition in the as-deposited (ad) state. At high UB (-80 V and -120 V), the samples contained fcc-(Ti,Al)N and Al-rich fcc-(Al,Ti)N, whereas the sam-ples deposited at low UB (-40 V) contained fcc-(Ti,Al)N as single-phase for x ≤ 0.47 (Fig. 12a). During annealing at 850 °C, the formation of Al-rich fcc-(Al,Ti)N and wurt-zite phase were observed simultaneously in the coatings. However, the relative amount of wurtzite phase increased with increasing bias voltage (see Fig. 12b). The Ti0.44Al0.56N coating deposited at UB = -40 V deviated from this trend, since wurtzite phase was present al-ready in the as-deposited sample. Hence, only traces of fcc-AlN formed during annealing. The increase of the wurtzite phase fraction and the decrease of fcc-AlN with increasing UB after annealing at 850 °C imply an accel-erated decomposition process in the coatings

217

Fig. 13: Microstructure evolution of the Ti-Al-N / Al-Ti-(Ru)-N multilayers of the three coating series (I = circle, II = triangle, III = square) deposited at UB = -40 V (left column) and UB = -80 V (right column) in course of the thermal treatment showing the stress-free lattice parameter (a,b), macroscopic lattice strain (c,d) and indentation hardness (e,f). The filled and open symbols in (a-d) represent the microstructure parameters of the fcc-(Ti,Al)N and the nearly Al-free fcc-(Ti,Al)N, respectively.

deposited at high bias voltages. It was sug-gested that a retarded reduction of the mac-roscopic lattice strain during annealing of the coatings deposited at UB = -120 V accelerates the formation of w-AlN [33].

A similar influence of the bias voltage on the development of the phase composition was observed in the Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings. Each multilayer coating was annealed consequently at 450 °C, 650 °C, 850 °C and 950 °C for 60 min in argon at-mosphere. After each annealing step, ex situ GAXRD was done and the diffraction pat-terns were analysed with respect to the stress-free lattice parameter and macroscopic lattice strain ε of the fcc phases. Additionally, the indentation hardness was measured after each annealing step. The changes in course of the thermal treatment are shown exemplarily for the coatings with three different Ru addi-tions that were deposited at UB = -40 V and

UB = -80 V in Fig. 13. Comparing the trends of the evolution of (Figs. 13a and 13b), ε (Figs. 13c and 13d) and indentation hardness (Figs. 13e and 13f) in the coatings that were deposited at the same bias voltage, no distinct influence of the Ru addition on their develop-ment during annealing in argon atmosphere could be revealed. Nevertheless, a beneficial effect of the Ru concentration on the thermal stability or hardness is possible during the ap-plication of the Ti-Al-N / Al-Ti-(Ru)-N mul-tilayer coatings in cutting processes as report-ed in [13]. The reason is that the conditions of the annealing experiment differed from the conditions prevailing during cutting. During cutting, the coatings are exposed to high tem-peratures for less than 30 min [13], but the de-composition of metastable phases takes place simultaneously with the oxidation of the sam-ples since cutting is usually done in air.

218

However, the influence of the bias voltage on the thermal stability of the Ti-Al-N / Al-Ti-(Ru)-N multilayer during annealing in argon atmosphere was evident and is shown for the coatings deposited at bias voltages of -40 V and -80 V because both bias voltages represent the general changes taking place in the coatings deposited at low and at high UB, respectively. The first obvious microstruc-ture changes became apparent after annealing at 650 °C. The stress-free lattice parameter and the macroscopic lattice strain of the fcc-(Ti,Al)N decreased. This can be attributed to defect annihilation and the incorporation of Al into fcc-(Ti,Al)N. In the coatings depos-ited at UB = -80 V, the relative reduction of ε and was more pronounced than in the coating deposited at UB = -40 V. For exam-ple after annealing at 650 °C, of the mul-tilayer coating of series III deposited at UB = -80 V decreased to 0.4178 nm (see Fig. 13b). According to the Vegard-like dependence from Eq. (6), this corresponds to an Al con-tent of x = 0.45. However, after annealing at 650 °C, and ε are still higher in the coatings deposited at UB = -80 V than in the coatings deposited at UB = -40 V (see Figs. 13a to 13d).

In all coatings, the onset of the decomposi-tion was observed after annealing at 850 °C. The Ti-rich and Al-rich fcc-(Ti,Al)N formed and the fraction of wurtzite phase increased slightly. These three phases existed beside the major fcc-(Ti,Al)N phase. This coexistence of the phases is exemplarily shown for the Ti-Al-N / Al-Ti-Ru-N multilayer coatings of series III deposited at UB = -40 V and UB = -80 V in Fig. 14. For the coating de-posited at UB = -40 V, the stress-free lattice parameter of the Ti-rich fcc-(Ti,Al)N phase (see open symbols in Fig. 13a) indicated the incorporation of x = 0.16 aluminium in the Ti-rich fcc-(Ti,Al)N. The Ti-rich (Ti,Al)N was under compressive stress (see open sym-bols in Fig. 13c). In the coatings deposited at high bias voltages, the wurtzite phase became more apparent as compared to the coatings deposited at low bias voltages after annealing at 850 °C (cf. Fig. 14).

After annealing at 950 °C, the fcc-(Ti,Al)N phase decomposed nearly completely into wurtzite phase, Ti-rich fcc-(Ti,Al)N and Al-rich fcc-(Al,Ti)N (see Fig.14). The micro-structure analysis revealed that the decom-position process after annealing at 950 °C proceeded further in the coatings, which were deposited at high UB, than in the coat-ings deposited at low UB. This was concluded from (i) of the Ti-rich fcc-(Ti,Al)N phase, (ii) of the Al-rich fcc-(Al,Ti)N phase, (iii) the intensity of the diffraction lines of the wurtzite phase in the GAXRD patterns (Fig. 14) and (iv) the hardness reduction. The indi-vidual items can be explained with the aid of Figs. 13 and 14 as follows.

(i) of the Ti-rich fcc-(Ti,Al)N phase is higher for the coatings deposited at UB = -80 V than for UB = -40 V (see open symbols in Figs. 13a and13b), which indicates a lower Al incorporation in fcc-(Ti,Al)N at high UB than at low UB. For instance, the Al incorpo-ration in the Ti-rich fcc-(Ti,Al)N phase of the multilayer coatings of series III is x = 0.15 for UB = -40 V and x = 0.09 for UB = -80 V, as cal-culated by using Eq. (6).

(ii) of the Al-rich fcc-(Al,Ti)N phase is lo-wer for coatings deposited at UB = -80 V than for UB = -40 V (see filled symbols in Figs. 13a and 13b), which signifies a lower Ti concen-tration in the Al-rich fcc phase at high UB. Both observations denote that the separation of titanium nitride and aluminium nitride progressed to a further extent in the coatings, which were deposited at high bias voltages.

After annealing at 950 °C, the fcc-(Ti,Al)Nphase decomposed nearly completely intowurtzite phase, Ti-rich fcc-(Ti,Al)N and Al-rich fcc-(Al,Ti)N (see Fig.14). Themicrostructure analysis revealed that thedecomposition process after annealing at950 °C proceeded further in the coatings,which were deposited at high UB, than inthe coatings deposited at low UB. This wasconcluded from (i) of the Ti-rich fcc-(Ti,Al)N phase, (ii) of the Al-rich fcc-(Al,Ti)N phase, (iii) the intensity of thediffraction lines of the wurtzite phase in theGAXRD patterns (Fig. 14) and (iv) thehardness reduction. The individual itemscan be explained with the aid of Figs. 13 and14 as follows.

(i) of the Ti-rich fcc-(Ti,Al)N phase ishigher for the coatings deposited at UB = -80 V than for UB = -40 V (see open symbolsin Figs. 13a and13b), which indicates alower Al incorporation in fcc-(Ti,Al)N athigh UB than at low UB. For instance, the Alincorporation in the Ti-rich fcc-(Ti,Al)Nphase of the multilayer coatings of series IIIis x = 0.15 for UB = -40 V and x = 0.09 forUB = -80 V, as calculated by using Eq. (6).

(ii) of the Al-rich fcc-(Al,Ti)N phase islower for coatings deposited at UB = -80 Vthan for UB = -40 V (see filled symbols inFigs. 13a and 13b), which signifies a lowerTi concentration in the Al-rich fcc phase athigh UB. Both observations denote that theseparation of titanium nitride andaluminium nitride progressed to a furtherextent in the coatings, which were depositedat high bias voltages.(iii) The diffraction lines of the wurtzitephase after annealing of the sampledeposited at UB = -80 V possess higherintensity than the diffraction lines observedin the GAXRD pattern for the thermallytreated sample deposited at UB = -40 V (cf.Fig. 14).

(iv) The indentation hardness of themultilayer coatings deposited at UB = -80 Vdropped from approx. 33 GPa to approx.27 GPa (Fig. 13f), whereas after annealing at950 °C the hardness of the coatingsdeposited at UB = -40 V decreased onlyslightly below the hardness level of the as-deposited state, namely to approx. 30 GPa(Fig. 13e). The obvious hardness reductionafter 950 °C in the coatings deposited atUB = -80 V can be attributed to a higherfraction of the wurtzite phase than in thecoatings deposited at UB = -40 V, which wasindicated also by the higher intensity of thediffraction lines of the wurtzite phase. Thehigher fraction of the wurtzite phase inannealed Ti-Al-N / Al-Ti-(Ru)-N multilayercoatings deposited at high UB coincides withthe results found for the Ti1-xAlxN coatings(see first part of Section 3.5) and confirmsthe effect of the bias voltage on the thermalstability of Ti-Al-N based coatings.

The microstructure analysis of the Ti-Al-Nmonolayer and Ti-Al-N / Al-Ti-(Ru)-Nmultilayer coatings produced by CAEelucidated the role of the energetic particlebombardment during CAE deposition. Itcould be shown that the bias voltage is avaluable deposition parameter in order todesign the microstructure, properties andthermal stability of Ti-Al-N based coatings.

For the Ti-Al-N monolayer coatings, it wasshown that the phase composition, the sizeof the fcc-(Ti,Al)N nanocrystallites as wellas the macroscopic lattice strain andconsequently the hardness of the coatingscan be adjusted by the Al content and thebias voltage. Low bias voltages resulted in auniform distribution of Ti and Al, whereashigh bias voltages caused local fluctuationsof Al and Ti in (Ti,Al)N and a rising defectdensity that facilitated the formation of anAl-rich fcc-(Al,Ti)N phase in monolayerand multilayer coatings at high UB (-60 V, -80 V and -120 V). Furthermore, a high UBlimited the volume fraction of wurtzitephase in the as-deposited Ti1-xAlxN coatingsfor x ≥ 0.56 as compared to low UB. It couldbe shown that the presence of low amountsof wurtzite phase (≤ 15 vol%) is notgenerally detrimental for the hardness. Onthe contrary, partially coherent interfacesbetween wurtzite and fcc phase arebeneficial for the coatings’ hardness. Severalorientation relationships between wurtziteand fcc phase offering partial coherentinterfaces could be identified by local FFTsof HRTEM images.

219

Fig. 14: Low-angle parts of GAXRD patterns of the Ti-Al-N / Al-Ti-Ru-N multilayer coatings of series III deposited at UB = -40 V (a) and UB = -80 V (b) after annealing at 850 °C and 950 °C. The measured data are shown as blue points and the whole fit as red lines. The grey lines show the individual diffraction peaks as fitted by Pearson VII functions. The positions of the fcc-TiN, fcc-AlN and w-AlN diffraction lines are indicated at the bottom of the figures. The positions of fcc-TiC and hex-WC coming from the substrate are shown at the top.

4. Conclusions

The microstructure analysis of the Ti-Al-Nmonolayer and Ti-Al-N / Al-Ti-(Ru)-N multi-layer coatings produced by CAE elucidatedthe role of the energetic particle bombar-dment during CAE deposition. It could beshown that the bias voltage is a valuabledeposition parameter in order to design themicrostructure, properties and thermalstability of Ti-Al-N based coatings.

For the Ti-Al-N monolayer coatings, it wasshown that the phase composition, the sizeof the fcc-(Ti,Al)N nanocrystallites as wellas the macroscopic lattice strain andconsequently the hardness of the coatingscan be adjusted by the Al content and thebias voltage. Low bias voltages resulted in auniform distribution of Ti and Al, whereashigh bias voltages caused local fluctuationsof Al and Ti in (Ti,Al)N and a rising defectdensity that facilitated the formation of anAl-rich fcc-(Al,Ti)N phase in monolayerand multilayer coatings at high UB (-60 V,-80 V and -120 V). Furthermore, a high UBlimited the volume fraction of wurtzitephase in the as-deposited Ti1-xAlxN coatingsfor x ≥ 0.56 as compared to low UB. It couldbe shown that the presence of low amounts

After annealing at 950 °C, the fcc-(Ti,Al)Nphase decomposed nearly completely intowurtzite phase, Ti-rich fcc-(Ti,Al)N and Al-rich fcc-(Al,Ti)N (see Fig.14). Themicrostructure analysis revealed that thedecomposition process after annealing at950 °C proceeded further in the coatings,which were deposited at high UB, than inthe coatings deposited at low UB. This wasconcluded from (i) of the Ti-rich fcc-(Ti,Al)N phase, (ii) of the Al-rich fcc-(Al,Ti)N phase, (iii) the intensity of thediffraction lines of the wurtzite phase in theGAXRD patterns (Fig. 14) and (iv) thehardness reduction. The individual itemscan be explained with the aid of Figs. 13 and14 as follows.

(i) of the Ti-rich fcc-(Ti,Al)N phase ishigher for the coatings deposited at UB = -80 V than for UB = -40 V (see open symbolsin Figs. 13a and13b), which indicates alower Al incorporation in fcc-(Ti,Al)N athigh UB than at low UB. For instance, the Alincorporation in the Ti-rich fcc-(Ti,Al)Nphase of the multilayer coatings of series IIIis x = 0.15 for UB = -40 V and x = 0.09 forUB = -80 V, as calculated by using Eq. (6).

(ii) of the Al-rich fcc-(Al,Ti)N phase islower for coatings deposited at UB = -80 Vthan for UB = -40 V (see filled symbols inFigs. 13a and 13b), which signifies a lowerTi concentration in the Al-rich fcc phase athigh UB. Both observations denote that theseparation of titanium nitride andaluminium nitride progressed to a furtherextent in the coatings, which were depositedat high bias voltages.(iii) The diffraction lines of the wurtzitephase after annealing of the sampledeposited at UB = -80 V possess higherintensity than the diffraction lines observedin the GAXRD pattern for the thermallytreated sample deposited at UB = -40 V (cf.Fig. 14).

(iv) The indentation hardness of themultilayer coatings deposited at UB = -80 Vdropped from approx. 33 GPa to approx.27 GPa (Fig. 13f), whereas after annealing at950 °C the hardness of the coatingsdeposited at UB = -40 V decreased onlyslightly below the hardness level of the as-deposited state, namely to approx. 30 GPa(Fig. 13e). The obvious hardness reductionafter 950 °C in the coatings deposited atUB = -80 V can be attributed to a higherfraction of the wurtzite phase than in thecoatings deposited at UB = -40 V, which wasindicated also by the higher intensity of thediffraction lines of the wurtzite phase. Thehigher fraction of the wurtzite phase inannealed Ti-Al-N / Al-Ti-(Ru)-N multilayercoatings deposited at high UB coincides withthe results found for the Ti1-xAlxN coatings(see first part of Section 3.5) and confirmsthe effect of the bias voltage on the thermalstability of Ti-Al-N based coatings.

The microstructure analysis of the Ti-Al-Nmonolayer and Ti-Al-N / Al-Ti-(Ru)-Nmultilayer coatings produced by CAEelucidated the role of the energetic particlebombardment during CAE deposition. Itcould be shown that the bias voltage is avaluable deposition parameter in order todesign the microstructure, properties andthermal stability of Ti-Al-N based coatings.

For the Ti-Al-N monolayer coatings, it wasshown that the phase composition, the sizeof the fcc-(Ti,Al)N nanocrystallites as wellas the macroscopic lattice strain andconsequently the hardness of the coatingscan be adjusted by the Al content and thebias voltage. Low bias voltages resulted in auniform distribution of Ti and Al, whereashigh bias voltages caused local fluctuationsof Al and Ti in (Ti,Al)N and a rising defectdensity that facilitated the formation of anAl-rich fcc-(Al,Ti)N phase in monolayerand multilayer coatings at high UB (-60 V, -80 V and -120 V). Furthermore, a high UBlimited the volume fraction of wurtzitephase in the as-deposited Ti1-xAlxN coatingsfor x ≥ 0.56 as compared to low UB. It couldbe shown that the presence of low amountsof wurtzite phase (≤ 15 vol%) is notgenerally detrimental for the hardness. Onthe contrary, partially coherent interfacesbetween wurtzite and fcc phase arebeneficial for the coatings’ hardness. Severalorientation relationships between wurtziteand fcc phase offering partial coherentinterfaces could be identified by local FFTsof HRTEM images.

220

After annealing at 950 °C, the fcc-(Ti,Al)Nphase decomposed nearly completely intowurtzite phase, Ti-rich fcc-(Ti,Al)N and Al-rich fcc-(Al,Ti)N (see Fig.14). Themicrostructure analysis revealed that thedecomposition process after annealing at950 °C proceeded further in the coatings,which were deposited at high UB, than inthe coatings deposited at low UB. This wasconcluded from (i) of the Ti-rich fcc-(Ti,Al)N phase, (ii) of the Al-rich fcc-(Al,Ti)N phase, (iii) the intensity of thediffraction lines of the wurtzite phase in theGAXRD patterns (Fig. 14) and (iv) thehardness reduction. The individual itemscan be explained with the aid of Figs. 13 and14 as follows.

(i) of the Ti-rich fcc-(Ti,Al)N phase ishigher for the coatings deposited at UB = -80 V than for UB = -40 V (see open symbolsin Figs. 13a and13b), which indicates alower Al incorporation in fcc-(Ti,Al)N athigh UB than at low UB. For instance, the Alincorporation in the Ti-rich fcc-(Ti,Al)Nphase of the multilayer coatings of series IIIis x = 0.15 for UB = -40 V and x = 0.09 forUB = -80 V, as calculated by using Eq. (6).

(ii) of the Al-rich fcc-(Al,Ti)N phase islower for coatings deposited at UB = -80 Vthan for UB = -40 V (see filled symbols inFigs. 13a and 13b), which signifies a lowerTi concentration in the Al-rich fcc phase athigh UB. Both observations denote that theseparation of titanium nitride andaluminium nitride progressed to a furtherextent in the coatings, which were depositedat high bias voltages.(iii) The diffraction lines of the wurtzitephase after annealing of the sampledeposited at UB = -80 V possess higherintensity than the diffraction lines observedin the GAXRD pattern for the thermallytreated sample deposited at UB = -40 V (cf.Fig. 14).

(iv) The indentation hardness of themultilayer coatings deposited at UB = -80 Vdropped from approx. 33 GPa to approx.27 GPa (Fig. 13f), whereas after annealing at950 °C the hardness of the coatingsdeposited at UB = -40 V decreased onlyslightly below the hardness level of the as-deposited state, namely to approx. 30 GPa(Fig. 13e). The obvious hardness reductionafter 950 °C in the coatings deposited atUB = -80 V can be attributed to a higherfraction of the wurtzite phase than in thecoatings deposited at UB = -40 V, which wasindicated also by the higher intensity of thediffraction lines of the wurtzite phase. Thehigher fraction of the wurtzite phase inannealed Ti-Al-N / Al-Ti-(Ru)-N multilayercoatings deposited at high UB coincides withthe results found for the Ti1-xAlxN coatings(see first part of Section 3.5) and confirmsthe effect of the bias voltage on the thermalstability of Ti-Al-N based coatings.

The microstructure analysis of the Ti-Al-Nmonolayer and Ti-Al-N / Al-Ti-(Ru)-Nmultilayer coatings produced by CAEelucidated the role of the energetic particlebombardment during CAE deposition. Itcould be shown that the bias voltage is avaluable deposition parameter in order todesign the microstructure, properties andthermal stability of Ti-Al-N based coatings.

For the Ti-Al-N monolayer coatings, it wasshown that the phase composition, the sizeof the fcc-(Ti,Al)N nanocrystallites as wellas the macroscopic lattice strain andconsequently the hardness of the coatingscan be adjusted by the Al content and thebias voltage. Low bias voltages resulted in auniform distribution of Ti and Al, whereashigh bias voltages caused local fluctuationsof Al and Ti in (Ti,Al)N and a rising defectdensity that facilitated the formation of anAl-rich fcc-(Al,Ti)N phase in monolayerand multilayer coatings at high UB (-60 V, -80 V and -120 V). Furthermore, a high UBlimited the volume fraction of wurtzitephase in the as-deposited Ti1-xAlxN coatingsfor x ≥ 0.56 as compared to low UB. It couldbe shown that the presence of low amountsof wurtzite phase (≤ 15 vol%) is notgenerally detrimental for the hardness. Onthe contrary, partially coherent interfacesbetween wurtzite and fcc phase arebeneficial for the coatings’ hardness. Severalorientation relationships between wurtziteand fcc phase offering partial coherentinterfaces could be identified by local FFTsof HRTEM images.

The bias voltage applied during thedeposition of the coatings influenced alsotheir thermal stability during annealing,since the decomposition in the Ti-Al-Nmonolayer and Ti-Al-N / Al-Ti-(Ru)-Nmultilayer coatings deposited at low UB wasretarded as compared to the coatingsdeposited at high UB. Furthermore, theannealing of the Ti-Al-N based coatings invacuum and argon atmosphere led to theformation of Ti-rich fcc-(Ti,Al)N as well aswurtzite phase and its coexistence with fcc-AlN in the temperature range of 850 - 950 °C. This observation supports the ideathat a direct transformation of fcc-(Ti,Al)Nto wurtzite phase is possible.

The addition of Ru to the Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings had noremarkable effect on the microstructure inthe as-deposited state and during annealingin argon atmosphere.

According to the results shown above, Ti-Al-N / Al-Ti-(Ru)-N multilayer coatingswith a medium [ ratio of ~ 0.53and Ti-Al-N coatings with an average[ ratio of ~ 0.47 that weredeposited at low UB (e.g. UB = -40 V) can berecommended for the application atelevated temperatures up to 950°C, since theformation of the large volume fractions ofwurtzite phase is retarded and the hardnessis maintained after annealing. The coatingswith similar [ ratios that weredeposited at high UB (e.g. UB = -80 V) couldbe favoured for the application attemperatures below <850°C, since thehardness is high in the as deposited state butthe generation of wurtzite phase exceeding acritical volume fraction is acceleratedduring annealing leading to a hardnessreduction.

This work was performed within the Clusterof Excellence “Structure Design of NovelHigh-Performance Materials via AtomicDesign and Defect Engineering (ADDE)”that is financially supported by theEuropean Union (European Fund ofRegional Development) and by the Ministryof Science and Art of Saxony (SMWK). Weacknowledge Helmholtz-Zentrum Dresden-Rossendorf for providing us with the in-house beam time at ROBL and thankCarsten Baehtz for assistance during themeasurements on the beamline BM20.

The bias voltage applied during thedeposition of the coatings influenced alsotheir thermal stability during annealing,since the decomposition in the Ti-Al-Nmonolayer and Ti-Al-N / Al-Ti-(Ru)-Nmultilayer coatings deposited at low UB wasretarded as compared to the coatingsdeposited at high UB. Furthermore, theannealing of the Ti-Al-N based coatings invacuum and argon atmosphere led to theformation of Ti-rich fcc-(Ti,Al)N as well aswurtzite phase and its coexistence with fcc-AlN in the temperature range of 850 - 950 °C. This observation supports the ideathat a direct transformation of fcc-(Ti,Al)Nto wurtzite phase is possible.

The addition of Ru to the Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings had noremarkable effect on the microstructure inthe as-deposited state and during annealingin argon atmosphere.

According to the results shown above, Ti-Al-N / Al-Ti-(Ru)-N multilayer coatingswith a medium [ ratio of ~ 0.53and Ti-Al-N coatings with an average[ ratio of ~ 0.47 that weredeposited at low UB (e.g. UB = -40 V) can berecommended for the application atelevated temperatures up to 950°C, since theformation of the large volume fractions ofwurtzite phase is retarded and the hardnessis maintained after annealing. The coatingswith similar [ ratios that weredeposited at high UB (e.g. UB = -80 V) couldbe favoured for the application attemperatures below <850°C, since thehardness is high in the as deposited state butthe generation of wurtzite phase exceeding acritical volume fraction is acceleratedduring annealing leading to a hardnessreduction.

This work was performed within the Clusterof Excellence “Structure Design of NovelHigh-Performance Materials via AtomicDesign and Defect Engineering (ADDE)”that is financially supported by theEuropean Union (European Fund ofRegional Development) and by the Ministryof Science and Art of Saxony (SMWK). Weacknowledge Helmholtz-Zentrum Dresden-Rossendorf for providing us with the in-house beam time at ROBL and thankCarsten Baehtz for assistance during themeasurements on the beamline BM20.

Acknowledgement

The bias voltage applied during thedeposition of the coatings influenced alsotheir thermal stability during annealing,since the decomposition in the Ti-Al-Nmonolayer and Ti-Al-N / Al-Ti-(Ru)-Nmultilayer coatings deposited at low UB wasretarded as compared to the coatingsdeposited at high UB. Furthermore, theannealing of the Ti-Al-N based coatings invacuum and argon atmosphere led to theformation of Ti-rich fcc-(Ti,Al)N as well aswurtzite phase and its coexistence with fcc-AlN in the temperature range of 850 - 950 °C. This observation supports the ideathat a direct transformation of fcc-(Ti,Al)Nto wurtzite phase is possible.

The addition of Ru to the Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings had noremarkable effect on the microstructure inthe as-deposited state and during annealingin argon atmosphere.

According to the results shown above, Ti-Al-N / Al-Ti-(Ru)-N multilayer coatingswith a medium [ ratio of ~ 0.53and Ti-Al-N coatings with an average[ ratio of ~ 0.47 that weredeposited at low UB (e.g. UB = -40 V) can berecommended for the application atelevated temperatures up to 950°C, since theformation of the large volume fractions ofwurtzite phase is retarded and the hardnessis maintained after annealing. The coatingswith similar [ ratios that weredeposited at high UB (e.g. UB = -80 V) couldbe favoured for the application attemperatures below <850°C, since thehardness is high in the as deposited state butthe generation of wurtzite phase exceeding acritical volume fraction is acceleratedduring annealing leading to a hardnessreduction.

This work was performed within the Clusterof Excellence “Structure Design of NovelHigh-Performance Materials via AtomicDesign and Defect Engineering (ADDE)”that is financially supported by theEuropean Union (European Fund ofRegional Development) and by the Ministryof Science and Art of Saxony (SMWK). Weacknowledge Helmholtz-Zentrum Dresden-Rossendorf for providing us with the in-house beam time at ROBL and thankCarsten Baehtz for assistance during themeasurements on the beamline BM20.

According to the results shown above,Ti-Al-N / Al-Ti-(Ru)-N multilayer coatingswith a medium [ ratio of ~ 0.53and Ti-Al-N coatings with an average[ ratio of ~ 0.47 that weredeposited at low UB (e.g. UB = -40 V) can berecommended for the application at eleva-ted temperatures up to 950 °C, since theformation of the large volume fractions ofwurtzite phase is retarded and the hardnessis maintained after annealing. The coatingswith similar [ ratios that weredeposited at high UB (e.g. UB = -80 V) couldbe favoured for the application at tempe-ratures below <850 °C, since the hardnessis high in the as deposited state butthe generation of wurtzite phase exceeding a

The addition of Ru to the Ti-Al-N / Al-Ti-(Ru)-N multilayer coatings had no remarkable effect on the microstructure in the as-deposited state and during annealing in argon atmosphere.

221

References

1. O. Knotek, M. Böhmer, T. Leyendecker, Journal of Vacuum Science & Technology A 4 (1986) 2695.2. W.D. Münz, Journal of Vacuum Science & Technology A 4 (1986) 2717.3. D. McIntyre, J.E. Greene, G. Hakansson, J.E. Sundgren, W.D. Münz, Journal of Applied Physics 67

(1990) 1542.4. P.H. Mayrhofer, A. Hörling, L. Karlsson, J. Sjölén, T. Larsson, C. Mitterer, L. Hultman, Applied

Physics Letters 83 (2003) 2049.5. A. Hörling, L. Hultman, M. Odén, J. Sjölén, L. Karlsson, Surface and Coatings Technology 191

(2005) 384.6. O. Knotek, W.D. Münz, T. Leyendecker, Journal of Vacuum Science & Technology A 5 (1987) 2173.7. T. Leyendecker, O. Lemmer, S. Esser, J. Ebberink, Surface and Coatings Technology 48 (1991) 175.8. D. Rafaja, A. Poklad, V. Klemm, G. Schreiber, D. Heger, M. Šíma, M. Dopita, Thin Solid Films 514

(2006) 240.9. S. Veprek, H.D. Männling, M. Jilek, P. Holubar, Materials Science and Engineering: A 366 (2004)

202.10. K. Kutschej, N. Fateh, P.H. Mayrhofer, M. Kathrein, P. Polcik, C. Mitterer, Surface and Coatings

Technology 200 (2005) 113.11. M. Kathrein, C. Michotte, M. Penoy, P. Polcik, C. Mitterer, Surface and Coatings Technology 200

(2005) 1867.12. M. Pohler, C. Michotte, M. Penoy, C. Mitterer, M. Kathrein, Proceedings of the 17th Plansee Semi-

nar, Reutte, 2009, HM 72/1 13. M. Kathrein, Patent Austria, PCT/AT2008/000231, 2008.14. A. Knutsson, M.P. Johansson, L. Karlsson, M. Odén, Journal of Applied Physics 108 (2010) 044312.15. Y. Setsuhara, T. Suzuki, Y. Makino, S. Miyake, T. Sakata, H. Mori, Surface and Coatings Technology

97 (1997) 254.16. A. Hörling, L. Hultman, M. Odén, J. Sjölén, L. Karlsson, Journal of Vacuum Science & Technology

A 20 (2002) 1815.17. R.F. Zhang, S. Veprek, Materials Science and Engineering A 448 (2007) 111.18. P.H. Mayrhofer, D. Music, J.M. Schneider, Journal of Applied Physics 100 (2006) 094906.19. R. Rachbauer, A. Blutmager, D. Holec, P.H. Mayrhofer, Surface and Coatings Technology 206

(2012) 2667.20. H. Riedl, D. Holec, R. Rachbauer, P. Polcik, R. Hollerweger, J. Paulitsch, P.H. Mayrhofer, Surface

and Coatings Technology 235 (2013) 174.21. P.H. Mayrhofer, L. Hultman, J.M. Schneider, P. Staron, H. Clemens, International Journal of Materi-

als Research 98 (2007) 1054.22. M. Odén, L. Rogström, A. Knutsson, M.R. Terner, P. Hedstrom, J. Almer, J. Ilavsky, Applied Physics

Letters 94 (2009) 053114.23. R. Rachbauer, J.J. Gengler, A.A. Voevodin, K. Resch, P.H. Mayrhofer, Acta Materialia 60 (2012)

2091.24. A. Knutsson, J. Ullbrand, L. Rogström, N. Norrby, L.J.S. Johnson, L. Hultman, J. Almer, M.P. Jo-

hansson Jöesaar, B. Jansson, M. Odén, Journal of Applied Physics 113 (2013) 2131518.25. T. Ikeda, H. Satoh, Thin Solid Films 195 (1991) 99.26. M. Zhou, Y. Makino, M. Nose, K. Nogi, Thin Solid Films 339 (1999) 203.27. K. Kutschej, P.H. Mayrhofer, M. Kathrein, P. Polcik, R. Tessadri, C. Mitterer, Surface and Coatings

Technology 200 (2005) 2358.28. P.H. Mayrhofer, C. Mitterer, L. Hultman, H. Clemens, Progress in Materials Science 51 (2006) 1032.29. L. Chen, K.K. Chang, Y. Du, J.R. Li, M.J. Wu, Thin Solid Films 519 (2011) 3762.30. R. Rachbauer, D. Holec, P.H. Mayrhofer, Surface and Coatings Technology 211 (2012) 98.31. N. Schalk, C. Mitterer, J. Keckes, M. Penoy, C. Michotte, Surface and Coatings Technology 209

(2012) 190.32. D. Rafaja, C. Wüstefeld, C. Baehtz, V. Klemm, M. Dopita, M. Motylenko, C. Michotte, M. Kathrein,

Metallurgical and Materials Transactions A 42 (2011) 559.33. C. Wüstefeld, D. Rafaja, M. Dopita, M. Motylenko, C. Baehtz, C. Michotte, M. Kathrein, Surface

and Coatings Technology 206 (2011) 1727.

222

34. J.L. Endrino, C. Arhammar, A. Gutíerrez, R. Gago, D. Horwat, L. Soriano, G. Fox-Rabinovich, D. Martín y Marero, J. Guo, J.E. Rubensson, J. Andersson, Acta Materialia 59 (2011) 6287.

35. M.P. Johansson Joesaar, N. Norrby, J. Ullbrand, R. M’Saoubi, M. Odén, Surface and Coatings Tech-nology 235 (2013) 181.

36. D. Rafaja, C. Wüstefeld, M. Dopita, M. Motylenko, C. Baehtz, C. Michotte, M. Kathrein, Surface and Coatings Technology 257 (2014) 26.

37. D. Rafaja, A. Poklad, V. Klemm, G. Schreiber, D. Heger, M. Šíma, Materials Science and Enginee-ring: A 462 (2007) 279.

38. D. Rafaja, C. Wüstefeld, M. Dopita, V. Klemm, D. Heger, G. Schreiber, M. Šíma, Surface and Coa-tings Technology 203 (2008) 572.

39. C. Wüstefeld, D. Rafaja, V. Klemm, C. Michotte, M. Kathrein, Surface and Coatings Technology 205 (2010) 1345.

40. N. Norrby, M.P. Johansson, R. M’Saoubi, M. Odén, Surface and Coatings Technology 209 (2012) 203.

41. S. PalDey, S.C. Deevi, Materials Science and Engineering: A 342 (2003) 58.42. A. Anders, Cathodic Arcs: From Fractal Spots to Energetic Condensation, Springer, New York,

2008.43. C. Wüstefeld, M. Motylenko, D. Rafaja, D. Heger, C. Michotte, C. Czettl, M. Kathrein, Proceedings

of the 18th Plansee Seminar, Reutte, 2013, HM140/1.44. O. Durand-Drouhin, A.E. Santana, A. Karimi, V.H. Derflinger, A. Schütze, Surface and Coatings

Technology 163-164 (2003) 260.45. G. Korb, Patent US4752335, 1988.46. DIN ISO 1832, 2004.47. D. Rafaja, C. Wüstefeld, D. Heger, M. Šíma, M. Jílek, Proceedings of the 18th Plansee Seminar, Reut-

te, 2013, HM 49/1.48. A.J. Perry, Thin Solid Films 193-194 (1990) 463.49. D. Rafaja, C. Wüstefeld, J. Kutzner, A.P. Ehiasarian, M. Šíma, V. Klemm, D. Heger, J. Kortus, Zeit-

schrift fuer Kristallographie 225 (2010) 599.50. M. Motylenko, R. Dzhafarov, C. Wüstefeld, D. Rafaja, to be published (2015).51. W.C. Oliver, G.M. Pharr, Journal of Materials Research 7 (1992) 1564.52. A.C. Vlasveld, S.G. Harris, E.D. Doyle, D.B. Lewis, W.D. Munz, Surface and Coatings Technology

149 (2002) 217.53. K. Sato, N. Ichimiya, A. Kondo, Y. Tanaka, Surface and Coatings Technology 163-164 (2003) 135.54. M. Ahlgren, H. Blomqvist, Surface and Coatings Technology 200 (2005) 157.55. D. Rafaja, C. Wüstefeld, M. Motylenko, C. Schimpf, T. Barsukova, M.R. Schwarz, E. Kroke, Chemi-

cal Society Reviews 41 (2012) 5081.56. M. Pfeiler, C. Scheu, H. Hutter, J. Schnoller, C. Michotte, C. Mitterer, M. Kathrein, Journal of Vacu-

um Science & Technology A 27 (2009) 554.57. M. Pfeiler, K. Kutschej, M. Penoy, C. Michotte, C. Mitterer, M. Kathrein, Surface and Coatings

Technology 202 (2007) 1050.58. D. Waasmaier, A. Kirfel, Acta Crystallographica A51 (1995) 416.59. D. Rafaja, V. Klemm, G. Schreiber, M. Knapp, R. Kužel, Journal of Applied Crystallography 37

(2004) 613.60. H. Ljungcrantz, L. Hultman, J.E. Sundgren, L. Karlsson, Journal of Applied Physics 78 (1995) 832.61. D. Holec, F. Rovere, P.H. Mayrhofer, P.B. Barna, Scripta Materialia 62 (2010) 349.62. S. Veprek, Journal of Vacuum Science & Technology A 17 (1999) 2401.63. U. Wahlström, L. Hultman, J.E. Sundgren, F. Adibi, I. Petrov, J.E. Greene, Thin Solid Films 235

(1993) 62.64. R. Rachbauer, S. Massl, E. Stergar, D. Holec, D. Kiener, J. Keckes, J. Patscheider, M. Stiefel, H. Leit-

ner, P.H. Mayrhofer, Journal of Applied Physics 110 (2011) 023515.65. C. Wüstefeld, M. Motylenko, D. Rafaja, in: D. Rafaja (Ed.), 65th Research Conference Freiberg, Frei-

berger Forschungsheft, Freiberg, 2014, 26.66. M.G. Moreno-Armenta, J. Diaz, A. Martinez-Ruiz, G. Soto, Journal of Physics and Chemistry of

Solids 68 (2007) 1989.

223

Experimental and numerical assessment of protective coatings deposited by high velocity oxygen fuel flame spraying: Spraying process and thermo-mechanical behavior

S. Roth 1, M. Hoffmann 2, C. Skupsch 1, M. Kuna 1, H. Biermann 2, H. Chaves 1

1 Institute of Mechanics and Fluid Dynamics, Lampadiusstr. 4, 09599, Freiberg, Germany2 Institute of Materials Engineering, Gustav-Zeuner-Str. 5, 09599 Freiberg, Germany

224

AbstractIn order to predict the life-time of HVOF-sprayed coatings, respective compounds are studied at different stages. First, the application of the coating, i.e. the spraying process itself, is exam-ined. An appropriate observation concept is presented that allows for the visualization of single particle deposition during the spraying process. Secondly, fatigue experiments carried out with coated and stand-alone specimens reveal the particular failure mechanisms occurring under cyclic thermo-mechanical service load. Finally, the damage process has been simulated numer-ically by means of a cyclic cohesive law. The presented results demonstrate that the numerical model is able to reproduce fatigue crack growth at the interfaces between the particles deposit-ed as well as the formation of delamination zones in coated compounds.

S. Roth 1, M. Hoffmann 2, C. Skupsch 1, M. Kuna 1, H. Biermann 2, H. Chaves 1

1 Institute of Mechanics and Fluid Dynamics, Lampadiusstr. 4, 09599, Freiberg, Germany2 Institute of Materials Engineering, Gustav-Zeuner-Str. 5, 09599 Freiberg, Germany

2 Processing and characterization of coating

2.1 Observations of the HVOF-spraying process

The characterization of coatings produced by HVOF spraying has been mostly performed a posteriori. The aim of the present project was to obtain more information about the processes themselves occurring during the application of the coating. The first step was to develop a camera capable of resolving both in time as well as in space the elementary process of par-ticle deposition during HVOF spraying. The requirements for this camera are strict as the particles to be observed are small in the order of 50 µm and have a velocity of about 500 m/s. Thus a framing rate of about 40 million pictu-

1 IntroductionHigh Velocity Oxy-Fuel (HVOF) flame spraying is a widely used thermal spraying technique to apply metallic corrosion pro-tection coatings. HVOF coatings exhibit high strength and hardness, low porosity and high corrosion as well as wear resistance [1]. They are typically applied to the surfaces of boilers in heat generators, heat exchangers or waste incineration plants to protect the structural components operating at high temperatures and in corrosive media. In operation, thermal and mechanical loading cause thermo-me-chanical fatigue (TMF) which is amongst others a consequence of the mismatch of the coefficients of thermal expansion. The com-pound mainly consists of three components: firstly the substrate material, e.g. boiler steel, secondly the plastically deformed coating particles composed of Ni-base alloys, and lastly the interfaces between substrate and coating on the one side and between the coat-ing particles on the other side. The resistive properties mentioned substantially depend on the meso- and microstructure of the coat-ing. The same dependency applies for the damage mechanisms inside the coating-sub-strate compounds.

By means of an integrative approach, corre-lations between coating process, mechanical properties and failure behavior were inves-tigated. A concept was developed to observe the elementary process of particle deposition. This technology allows verifying a hypothesis that describes the formation process of coa-tings. Furthermore, the meso- and microsco-

pic composition of the compounds was ana-lyzed. Fatigue experiments of specimens with and without coatings were performed. Here the influence of the coating on the fatigue life was studied and the specific failure mo-des were identified. All these findings have an important impact on the material modelling based on continuum mechanics concepts. A thermo-dynamically consistent cyclic cohe-sive zone model was implemented into finite element code, which has been proven to re-produce the fatigue mechanisms of the coa-tings under thermo-mechanical loading. In the following sections, these steps towards a reliable prediction of the fatigue life of HVOF coatings are presented in more detail.

225

res per second is needed to capture the impact process of a particle. There are only very few cameras on the market available that can cope with the present needs. However taking into account that the flame and the HVOF spray are very hot (>1000 K) and that a microscopic resolution is needed, no commercial product was found. Our camera works according to the principle of Cranz-Schardin, [2]. The ca-mera has no moving parts, which is obvious at these framing rates. The principle is that multiple light beams illuminate the object at slightly different angles. Each light beam ori-ginates at a pulsed light source of sufficient-ly short pulse duration. In our case the pulse lasts for 5 nanoseconds. The light sources pro-duce the light pulses at the required frequen-cy of images (framing rate). On the opposite side of the object the imaging optics guide the light from each light beam onto a correspon-ding CCD camera and produce an image of the object on the chip. The number of images is limited by the means of separating the light beams and by the size of the cameras. In our case four CCD cameras are used. The details of the camera are described in [3].

A serious problem that arises when looking at very small and high speed particles with a rather limited temporal window, in our case 75 ns, i.e. 3 x 25 ns between 4 images, is the problem of triggering. The images should start just prior to a particle impact within the field of view of the camera, 1.12 mm x 0.84 mm, so that the first image gives a picture of the particle just before impact and the fol-lowing images show the development of the impact phenomenon. However, this instant of time is not known beforehand. Furthermore, a peculiarity of the laser used for illumination is that it is a Nd-Yag laser pumped by a flash.

The pump energy has to be delivered to the laser about 100 µs before the laser pulse can be triggered with a Q-switch.

A set of two very fast light barriers was built to overcome this problem. The small measu-ring volume of each of these barriers was achieved by focusing a laser diode with a lens with some magnification. This gave volumes of about 100 µm in diameter and 500 µm in length. Thus a particle that crosses both barri-ers, with a separation of 6 cm has indeed a ve-locity vector that is aligned with the barriers. The projection of the line between the barriers crosses the field of view of the cameras. The sharp analogue peak produced by the passage of a particle through the barrier is converted to a digital TTL signal that is fed to a micro-processor with 50 MHz clock. The micropro-cessor contains a counter, so that the elapsed time between the passage of both barriers can be measured and the velocity could then be calculated. This value is used to calculate the time when the particle would impact and thus to start the flash of the laser 100 µs in advance and then the Q-switch at the appropriate in-stant of time. The technique was tested with fast diesel fuel sprays from a common-rail injector. However, this technique does not work in the case of HVOF sprays. The reason may be that HVOF sprays are not very dense sprays, see Fig. 1. It shows the particles illumi-nated by a 5 ns laser pulse. The field of view is 150 mm across the image. This explains that it is very much a matter of luck to be able to catch a particle flying with 600 m/s with a field of view of 1.12 mm. A different approach to the problem has still to be found to be able to make images at the right time and place.

Figure 1: Image of a HVOF spray (5 ns illumination pulse) and flame plus particles (100 ms frame integration)

226

One further field of work was the measure-ment of an average particle surface tempera-ture. Figure 1 shows also the averaged light emission of many particles, since the flame itself is not very luminous in the wavelength range of the camera. However the particles are being heated by the flame but also in the com-pression regions of the Mach cells. Since the emission temperature is the surface tempera-ture of the particle, it can react quickly to the changes in temperature of the gas, whereas the inner temperature takes longer. The light emission of a body will increase with tempe-rature and depends also on the emitted wave-length. Qualitatively Fig. 1 is telling us that the surface temperature of the particles is glo-bally decreasing downstream, because the fla-me is becoming colder with distance from the nozzle but also because heat conduction into the particles reduces the surface temperature. Far from the nozzle the particle temperature leads to a much reduced light emission. Ima-ges as Fig. 1 lead to the idea of measuring the particle surface temperature with a two-color pyrometer based on Planck’s Law. It is basi-cally composed of an imaging lens system, a beam splutter and two CCD cameras with fil-ters in the near infrared. Two wavebands were chosen (800 nm ±40 nm and 950 nm ±40 nm) that give satisfactory results in the tempera-ture range between 1000-2400 °C. Figure 2 shows the results of these measurements for an oxy-acetylene HVOF spray impacting on a body on the right side of the image. Near the nozzle the emission of the flame is evident

and the expected temperatures drop further away from the nozzle. But there is an increase of the particle temperature by about 350 °C on the body where the spray impacts. This is due to the dissipation of kinetic energy of the particles transformed into heat. When they impact they are deformed plastically to form the HVOF coating and they can even melt if the process is too hot. Using energy conser-vation and known particle velocities u≈550 m/s measured by PIV [4], one can calculate the temperature increase caused by particle impact for Inconel particles:

Figure 2: Temperature distribution in a HVOF spray measured with two color 2d-pyrometry

In the presented case the temperature is at the limit applicable for the coating process, since the particles should deform plastically and maybe melt to a small extent. If they melt then splashing will occur, i.e. the molten particles create craters and produce small secondary drops (overspray). The quality of such coating is rough and porous.

One can conclude that two color pyrometry can be a tool for the monitoring and control of the HVOF process.

2

p

1 12

T uC

∆ = , with Cp≈0,41 J/K/g=410 J/K/kg

∆T≈370 °C

227

2.2 FEM simulation

In order to investigate the applicability of commercial finite element programs to sim-ulate the deposition of thermally sprayed coatings, a feasibility study was performed. Within the 2D-models, coating particles are assumed to be of circular shape. Elastic-plas-tic material parameters were used according to Johnson-Cook-Plasticity. The material formulation and the loading parameters are explained detailed in [5]. Deposition of one single particle as well as successive deposition of four particles has been simulated. Due to the large plastic deformation expected, Eu-lerian meshes were used (FE program AB-AQUS). That means that basically numerical techniques of fluid dynamics were applied to this problem of solid mechanics. Both initial conditions of the process were varied, i.e. ho-mogeneous particle temperature and velocity. Some results are depicted in Fig. 3.

As can be seen, the assumed plastic clamp-ing between particles is reproduced. The topology of the highly deformed particles is qualitatively consistent with micrographic observations. The unrealistic deformation of the substrate results from the assumptions concerning the material behavior. More pre-dictive modeling requires detailed informa-tion about the initial particle size distribution, realistic initial conditions of the particles

Figure 3: Simulation of a deposition process of four circular particles: a) initial contour; b) deformed state after depo-sition, colors refer to accumulated plastic strain

a) b)

and a thermodynamical material model that captures additionally phase transition pro-cesses occurring possibly inside the particles. Nevertheless, the splashing mentioned above could be simulated by increasing the initial velocity, see Fig. 4.

Figure 4: Splashing after deposition of one single particle with increased initial velocity

228

3 Isothermal and thermo-mechanical properties Coating-substrate compounds are subjected to temporally changing thermal and mecha-nical loadings under operating conditions in heat exchangers and waste incineration plants. These cyclic loading conditions are called thermo-mechanical fatigue and be-long, due to the damaging effects, to the low cycle fatigue loadings. The aim of this pro-ject was to identify the damage mechanisms during isothermal and thermo-mechanical fatigue of the HVOF-sprayed coating IN-625 applied to the substrate 16Mo3. Former studies on this coating-substrate compound were limited to corrosive investigations, only [6, 7]. Isothermal and thermo-mechanical fatigue tests were carried out to characteri-ze the influence of a HVOF-sprayed coating IN-625 on the fatigue behaviour of the sub-strate material 16Mo3 (1.5415). Therefore, the following three modifications were tested:

• uncoated substrate 16Mo3,• HVOF-sprayed coating-substrate com-

pound IN-625 – 16Mo3,• “stand-alone” HVOF-sprayed coating

material IN-625.

The corrosion resistant coating IN-625 was applied by HVOF spraying. This thermal spray technique is characterized by high gas and particle velocities, good bonding to the substrate and very small fractions of porosity. The nominal chemical composition of the ap-plied alloy is given in Tab. 2.

Alloying Elements [w.%] C Si Mn Mo Cr Fe

Uniaxial fatigue (rod) 0.167 0.343 0.783 0.261 0.202 bal.

Table 1: Chemical composition of the substrate material 16Mo3

Table 2: Nominal chemical composition of IN-625 thermal spray powder

Alloying Elements [w.%] Cr Fe Mo Nb+Ta Ni

IN-625 21.5 2.5 9.0 3.7 bal.

3.1 Material

The substrate material 16Mo3 is a ferritic, low alloyed, heat resistant steel with application temperatures up to 530 °C. This steel is main-ly used in heat exchangers, pipe and vessel constructions and waste incineration plants. In these applications the steel is subjected to corrosive and oxidizing hot gases. Therefore, oxidation resistant coatings are necessary to protect the substrate. The very good high tem-perature oxidation properties of the modified IN-625 alloy are well known and reported in literature. The chemical composition of the substrate material used for uniaxial fatigue specimen (cylindrical rods) is given in Tab. 1.

The coating was applied by HVOF on the sub-strate with a thickness of 200 µm. The dama-ge mechanisms during fatigue loading were investigated at the stand-alone coating ma-terial. Therefore, a coating was applied on a U-shaped steel profile with a thickness of 15 mm (Fig. 5) and the specimens were prepared by wire spark erosion, subsequently.

229

Figure 5: U-shaped steel profile with (marked red) stand-alone coating ma-terial IN-625 and tensile test specimens obtained by wire spark erosion.

3.2 Experimental procedure

Contrary to the isothermal fatigue loading, the temperature of TMF tests is cycling bet-ween two limiting test temperatures. Due to the cycling temperature a thermal strain ( ) term has to be considered in the de-termination of the applied mechanical strain ( ). The occurring thermal strain within the test temperature range was measured before the test und described with a pow-er law =A+TB, where A and B are fit parameters of the applied temperature range and T is the temperature. The con-trol variable is the mechanical strain which was calculated as the difference of the mea-sured total strain and the thermal strain.

The uniaxial thermo-mechanical fatigue tests were conducted within a temperature range of 200 °C to 500 °C at a 100 kN servohydrau-lic test rig equipped with an induction hea-ting. To avoid thermal stresses across the test cross section, the heating and cooling rates of all TMF tests were kept constant at 5 K/s, according to the code-of-practice for TMF tests. In this study the two main temperature - strain relationships “in-phase” (IP) and “out-of-phase” (OP, phase angle 180°) were inves-tigated. Additionally, isothermal fatigue (IF) tests were carried out at the maximum tem-peratures of the TMF tests at the same test rig. IF tests were conducted on specimens from the substrate material, the coating-substra-te compound and the stand-alone coating

material. All fatigue tests were carried out with the same specimen geometry given in Fig. 6. The position of the IN-625 coating is also shown in this figure. The preparation of the substrate gauge length is separated into three steps, first mechanical turning and se-cond electro-chemical polishing with the electrolyte Struers A3 and finally coating with IN-625, with preliminary sand blasting.

3.3 Results and discussion

The uncoated substrate exhibited a conti-nuous cyclic softening behaviour at 500 °C under isothermal low cycle fatigue loading. In contrast, at 200 °C a cyclic secondary har-dening was observed due to dynamic strain ageing. These isothermal test results and the monotonic material behavior at various tem-peratures of the uncoated substrate material 16Mo3 are published in [8]. In the run-up to the TMF tests, isothermal fatigue tests were carried out with the coating substrate com-pound and the stand-alone coating material at 500 °C. At these tests, the stand-alone coating exhibited a nearly constant stress amplitude. Additionally, the formation of tensile mean stresses was found, similar as reported in [9]. The brittle behaviour of the stand-alone coa-ting at 500 °C indicates that fatigue cracks in the coating-substrate compound are expected to initiate in the coating.

εtherm

εmech

εtherm

230

Figure 6: Uniaxial IF and TMF specimen geometry with HVOF coating

Figure 7 a): Cyclic stress response curves of the stand-alone coating IN-625 at 500 °C, b) maximum stress response curves of the isothermal fatigue tests with the coating substrate compound

The result of the isothermal fatigue tests at the coating-substrate compound was that two different damage mechanisms were identified. For loading amplitudes less than or equal to , fatigue crack growth occurred within the coating. In Fig. 7b the cyclic ma-ximum stress response is plotted vs. num-ber of cycles N. In this figure the stress drops (arrows) indicate the fatigue life of the coa-ting. This damage mechanism I is characte-rized by fatigue crack growth in the coating (see Fig. 8a). After the coating has failed, the fatigue crack transfers to the substrate. A ma-croscopic delamination of the coating after fracture was not detected. Another damage mechanism II occurs for total strain amplitu-des .

No stress drop occurred at (see Fig. 7b) because the coating was already bro-ken within the tensile part of the first load cycle. The subsequent damage evolution of this damage mechanism II is characterized by delamination of the coating and oxidation of the substrate. Finally, cracks are initiated into the substrate (see Fig. 8b). Due to the surface roughness, the stand-alone coating material exhibits a longer fatigue life than the coating on the substrate at low strain amplitudes. All results of the isothermal fatigue tests at 500 °C are shown in Fig. 9.

231

Figure 8 a): SEM image of a damage mechanism I fatigue crack initialized in the coating and grown into the substrate, SE contrast, b) damage mechanism II coating failure at high total strain amplitudes.

Figure 9: Total strain vs. fatigue life diagram of the isothermal fatigue test results at 500 °C with corresponding fit curves

The stand-alone coating exhibited complete brittle failure behavior at 200 °C. Consequent-ly, the damage mechanism II was detected for all IF tests at this temperature. The brittle coa-ting broke within the first cycle followed by a macroscopic delamination of the coating during fatigue.

Following the IF tests, mechanical strain-con-trolled TMF tests were carried out as in-phase (IP) and out-of-phase (OP) tests, at the coa-ting-substrate compound and the uncoated substrate at different mechanical strain amp-litudes from 0.0025 to 0.006. Due to tempera-

ture dependent flow stress, the development of characteristic mean stresses was observed at these two phase relationships between tem-perature and mechanical strain. Tensile mean stresses were observed for OP tests and com-pressive mean stresses for IP tests. Due to the-se mean stresses, the OP TMF tests achieved shorter fatigue lives compared to the IP TMF tests. The results of the TMF tests for one me-chanical strain amplitude are plotted in Fig. 10. As expected, a longer fatigue life was ob-served for the IP test compared to the OP test at the uncoated substrate 16Mo3. The two da-mage mechanisms that were characterized at

232

Figure 10: Cyclic stress-response curves of coated and uncoated IP and OP TMF tests, within a temperature range of 200 °C - 500 °C, at a mechanical strain amplitude of

the coated IF test are observed at the coated TMF tests, too. IP loading conditions resulted in a longer TMF life of the coating-substrate compound compared to the uncoated sub-strate, at this mechanical strain amplitude, and is characterized by fatigue crack growth within the coating (see Fig. 10 arrow) and subsequent fatigue crack initiation into the substrate (damage mechanism I). This longer fatigue life is due to the supporting effect of the coating under compressive loading in the hysteresis loop and the oxidation protecting effect at maximum temperature. An OP loa-ding had a detrimental effect on the fatigue life, because of its low ductility at low tem-peratures under tensile loading. The coating broke within the first hysteresis loop and de-laminated in the following cycles as observed at the 200 °C IF tests (damage mechanism II).

4 Simulation of material behavior

In order to simulate fatigue of the coated compounds, a cyclic cohesive zone model was developed. Within the concept of cohesive zo-nes, the entire damage and fracture process of a material is concentrated to a thin cohesive zone while the bulk remains free of damage. Since delamination was observed to occur at the interfaces between the deformed coating particles and between coating and substrate [10], this theory is well applicable to the con-sidered coated structures.

∆εmech/2 = 0.004

233

Figure 11: Cyclic cohesive zone model: dotted – expo-nential monotonic envelope (damage locus, DL), endu-rance locus (EL); solid – ex-ample of cyclic separation (path O – A – O – B – O)

4.1 Cyclic cohesive zone model

Cohesive laws describe the interrelationshipbetween the tractions transferred across thecohesive zone and the respectivedisplacement jump (separation). The localdamage state is characterized by a scalardamage variable D. The behaviour of thecohesive zone in consequence of monotonicloading is fully determined by a cohesivepotential which depends on the magnitudeof the separation vector λ and the damagevariable. Model parameters with respect tothe monotonic properties are the cohesivestrength t0 which is the maximumtransferrable traction, an intrinsic lengthδ0, the normalized fracture energy densityΓ0 and some shape parameters ofthe monotonic traction-separation-curve(monotonic envelope, see Fig. 11). The areabelow this curve equals Γ0. In order todescribe the increase of damage duringcyclic subcritical loading, a damageevolution equation has to be formulated.This relation determines the amount ofdamage accumulation as a result of anapplied external separation increment at aspecific material state indicated by λ and D.A local endurance limit was introduced toguarantee a smooth transition fromreversible to damaged separation behaviour.Here, a two-parametric power-lawapproach was proposed which scales themonotonic envelope. As a result, anendurance locus could be found whichforms a lower bound of damage evolution.Below the endurance locus each materialstate is endurable, while between endurancelocus and monotonic envelope damageaccumulates. A respective load cycleconsisting of a monotonic loading path,unloading towards the origin andsubsequent reloading s depicted in Fig. 11.In order to formulate an appropriatecriterion for fatigue crack growth, thesemodel characteristics are found to bedecisive. Further details concerning thecyclic cohesive zone model are reported in[11].

234

A local endurance limit was introduced to guarantee a smooth transition from revers-ible to damaged separation behavior. Here, a two-parametric power-law approach was proposed which scales the monotonic enve-lope. As a result, an endurance locus could be found which forms a lower bound of damage evolution. Below the endurance locus each material state is endurable, while between en-durance locus and monotonic envelope dam-age accumulates. A respective load cycle consisting of a monotonic loading path, un-loading towards the origin and subsequent reloading is depicted in Fig. 11. In order to formulate an appropriate criterion for fatigue crack growth, these model characteristics are found to be decisive. Further details concern-ing the cyclic cohesive zone model are report-ed in [11].

Figure 13: Normalized fatigue crack growth rate, d(∆a/δ0)/dN , in depen-dence of normalized applied load ran-ge, ∆KI /K0 , and ∆G/(t0 δ0 Γ0 ), respec-tively (R = 0); dotted: straight line of Paris law with exponent m = 6.5

Figure 12 a): Boundary layer mo-del with KI-controlled displacement boundary conditions; b) detailed mesh within the fine meshed region, 0 ≤ x ≤ La

In order to validate the proposed model,fatigue crack growth of a semi-infinitecrack under plane strain and small scaleyielding conditions (boundary layer model,see Fig. 12) has been simulated, see [12] fordetails.

So far, none of the existing cyclic cohesivezone models was able to represent all of thethree stages of fatigue crack growth: near-threshold region, Paris line and staticfailure, see Fig. 13. Thus, fatigue crackgrowth rate curves were generated in anextensive parametric study [11]. Thisinvestigation revealed that the assumedstate dependent endurance limit is wellsuited to adjust the near-threshold region offatigue crack growth. A criterion for fatiguecrack growth was deduced, according tothat a fully developed damage zone isneeded that moves with constant width andcrack growth rate. Furthermore, the cyclicload necessary to establish these steady-state conditions, especially the maximumload level, has to exceed a specific value,which depends (amongst others) on theendurance properties of the cyclic cohesivemodel. This relation gives rise to aparameter identification procedure thatallows the determination of the cycliccohesive parameters my means of the Parisparameters, threshold value and staticfracture toughness.

In order to validate the proposed model,fatigue crack growth of a semi-infinitecrack under plane strain and small scaleyielding conditions (boundary layer model,see Fig. 12) has been simulated, see [12] fordetails.

So far, none of the existing cyclic cohesivezone models was able to represent all of thethree stages of fatigue crack growth: near-threshold region, Paris line and staticfailure, see Fig. 13. Thus, fatigue crackgrowth rate curves were generated in anextensive parametric study [11]. Thisinvestigation revealed that the assumedstate dependent endurance limit is wellsuited to adjust the near-threshold region offatigue crack growth. A criterion for fatiguecrack growth was deduced, according tothat a fully developed damage zone isneeded that moves with constant width andcrack growth rate. Furthermore, the cyclicload necessary to establish these steady-state conditions, especially the maximumload level, has to exceed a specific value,which depends (amongst others) on theendurance properties of the cyclic cohesivemodel. This relation gives rise to aparameter identification procedure thatallows the determination of the cycliccohesive parameters my means of the Parisparameters, threshold value and staticfracture toughness.

235

Figure 14 a): RVE; b) Honeycomb structure of the cohesive zone elements inside the RVE

4.2 RVE of a coating-substrate compound

The qualitative separation and failure beha-vior of the interfaces in sprayed coatings was analyzed in [13] by means of a representative volume element (RVE), that comprises a sub-strate and a coating section. A Voronoi-tessel-lation is used for the geometric modeling of the latter. Thereby, each deformed particle is represented by one convex polyhedron. The flatness of the particles, which results from the spraying process, is incorporated by sca-ling the tessellated structure in the direction perpendicular to the substrate surface. Both substrate and particles were assumed to be linear-elastic. The model was meshed with conventional 3D elements in the bulk and user defined cohesive elements throughout all interfaces. Thus, each interface is treated as a potential damage zone. Figure 14b shows the resulting honeycomb structure of the co-hesive zone elements in the RVE (Fig. 14a). It is worth mentioning that the cohesive zone elements are of zero-thickness type. So, the cohesive zone has neither an initial volume nor a mass.

4.3 Damage evolution inside the RVE due to thermo-mechanical fatigue

The capabilities of the cyclic cohesive zone model to predict TMF of coated structures are demonstrated. Therefore, the RVE of the coating-substrate compound presented above was subjected to a thermal load. The material parameters of the substrate and the coating particles are summarized in Tab. 3. As men-tioned before, linear-elastic material behavior is assumed. For the cohesive zone estimated parameters were chosen. The cyclic parame-ters were adjusted to get significant damage evolution within the first cycles. The loading of the RVE is applied by a homogeneous os-cillating temperature cycling with an ampli-tude of 400 K. Apart from that, mechanical boundary conditions were applied. They pre-vent rigid body motion and establish periodic continuation.

b)a)

236

substrate particels interface

ES=210 GPa

νS=0.3

αthS=12.0E-6 K-1

EP=180 GPa

νP=0.3

αthP=13.0E-6 K-1

t0=50 MPa

δ0=1.0E-4 mm

Γ0=exp(1)

Table 3: Material parameters of the RVE components

237

Due to the slight misfit of the thermalexpansion coefficients ath

i and the Young’smoduli Ei, damage zones develop dependingon the specific interface density and theinterface orientations. Selected plots of thedamage distribution after various numbersof load cycles are depicted in Fig. 15. Inorder to visualize the damage evolution,three different views are shown: the top viewtowards coating, a top view on the interfacebetween coating and substrate, and a cutview through the whole RVE. Figure 15ashows the point of damage initiation afterthree load cycles. After some cycles ofdamage evolution at the particle interfaces,the damage zone reaches the interfacebetween coating and substrate (Fig. 15b).Subsequently, the damage zones growfurther into this interface. When the damagevariable reaches the value D=1 (red) therespective regions can be interpreted ascracks. Such interfacial cracks are illustratedin Fig. 15c.

Qualitatively, these numerical results matchthe experimentally observed fatigue failuremechanisms in the HVOF-coatings. Never-theless, these models do not allow at presenta numerical prediction of the life time of realcoated compounds. There are restrictionsbased on the numerical effort. Furthermore,the identification of the material parametersis still an open issue. But with respect to themechanisms that cause thermo-mechanicalfatigue, numerical studies allow to investigatethe influence of different coating topologies[14]. According to this, high fatigue lifes areachievable by coatings without voids andinclusions but with particle sizes as homoge-neous as possible. In order to prevent crackinitiation at the substrate by fatigue crackscoming from the coating, a particle refine-ment in direction towards the coating isrecommended.

Due to the slight misfit of the thermalexpansion coefficients ath

i and the Young’smoduli Ei, damage zones develop dependingon the specific interface density and theinterface orientations. Selected plots of thedamage distribution after various numbersof load cycles are depicted in Fig. 15. Inorder to visualize the damage evolution,three different views are shown: the top viewtowards coating, a top view on the interfacebetween coating and substrate, and a cutview through the whole RVE. Figure 15ashows the point of damage initiation afterthree load cycles. After some cycles ofdamage evolution at the particle interfaces,the damage zone reaches the interfacebetween coating and substrate (Fig. 15b).Subsequently, the damage zones growfurther into this interface. When the damagevariable reaches the value D=1 (red) therespective regions can be interpreted ascracks. Such interfacial cracks are illustratedin Fig. 15c.

Qualitatively, these numerical results matchthe experimentally observed fatigue failuremechanisms in the HVOF-coatings. Never-theless, these models do not allow at presenta numerical prediction of the life time of realcoated compounds. There are restrictionsbased on the numerical effort. Furthermore,the identification of the material parametersis still an open issue. But with respect to themechanisms that cause thermo-mechanicalfatigue, numerical studies allow to investigatethe influence of different coating topologies[14]. According to this, high fatigue lifes areachievable by coatings without voids andinclusions but with particle sizes as homoge-neous as possible. In order to prevent crackinitiation at the substrate by fatigue crackscoming from the coating, a particle refine-ment in direction towards the coating isrecommended.

Figure 15: Evolution of the damage zone at the RVE interfaces (left top: top view, left bottom: interface between bulk and coating, right: cut view): a) damage initiation, N=3; b) damage initiation at substrate-coating interface, N=7; c) separation due to fatigue crack growth throughout the RVE and by delamination, N=20

b)

a)

c)

238

5 Summary and discussion

Coating-substrate compounds produced by a High Velocity Oxy-Fuel flame spraying process were investigated at different stages. Regarding the spraying process, concepts of high-speed camera systems are presen-ted to observe the deposition process and to measure surface temperatures of the coating particles. Although the measurement princip-le works well for fast diesel fuel sprays, the-re are still problems when applied to HVOF spraying. The temperature measurements show that the very fast and large temperature jump during particle impact due to the adia-batic conversion of kinetic energy will deter-mine the outcome of the splat. This may be the key process typical for HVOF spraying. High kinetic energy is delivered for the ne-cessary plastic deformation and at the same time local energy dissipation heats the mate-rial close and may be over the melting point for a short time interval. The skill of finding the operation parameters for HVOF spray-ing is to approach the melting point but not to superheat the material. Then splashing will start to disturb the deposition process.

Coating-substrate compounds and stand-alo-ne coatings were subjected to thermo-mecha-nical fatigue in an extensive testing program. Fatigue lifes were determined and specific failure mechanisms could be identified. The-reby conventional specimen geometries were used. The verification of these results via com-parison with those obtained by miniaturized testing procedures is part of current research. Numerical methods were developed to simu-late the deposition process as well as the in-terfacial fatigue damage. Throughout the ana-lyses qualitative good results were obtained. Nevertheless, there are still severe restrictions resulting from the numerical effort needed. Furthermore, the identification of material parameters for the small particles and the in-terfaces is still an open issue. In order to over-come the latter problem, special miniaturized test procedures adapted to the specific materi-al behavior have to be developed in the future.

239

1. T. Sidhu, S. Prakash, and R. Agrawal. Studies on the properties of high-velocity oxy-fuel thermal spray coatings for higher temperature applications. Journal of Materials Science, 41(6):805-823, 2005.

2. C. Cranz and H. Schardin. Kinematographie auf ruhendem Film und mit extrem hoher Bildfrequenz. Zeitschrift für Physik, 56(3-4):147-183, 1929.

3. C. Skupsch, H. Chaves, and C. Brücker. Cranz-Schardin camera with a large working distance for the observation of small scale high-speed flows. Review of Scientific Instru-ments, 82(8):83706-6, 2011.

4. H. Chaves, S. Herbst, and C. Skupsch. Measurement of the particle velocity in a HVOF spray with PIV under industrial conditions. Journal of Thermal Spray Technology, 21(5):882-886, 2012.

5. H. Assadi, F. Gärtner, T. Stoltenhoff, and H. Kreye. Bonding mechanism in cold gas spray-ing. Acta Materialia, 51(15):4379-4394, 2003.

6. H.Y. Al-Fadhli, J. Stokes, M.S.J. Hashmi, and B.S. Yilbas. The erosion–corrosion behavi-our of high velocity oxy-fuel (HVOF) thermally sprayed Inconel-625 coatings on different metallic surfaces. Surface and Coatings Technology, 200(20-21):5782-5788, 2006.

7. H.Y. Al-Fadhli, M.S.J. Hashmi, and B.S. Yilbas. HVOF coating of Inconel-625 onto stain-less and carbon steel surfaces: corrosion and bond testing. Journal of Materials Processing Technology, 155-156(1-3):2051-2055, 2004.

8. M. Hoffmann, and H. Biermann. Static and cyclic deformation behavior of the ferritic steel 16Mo3 under monotonic and cyclic loading at high temperatures. Steel Research International, 83(7):631-636, 2012.

9. A. Marucco, and B. Nath. Effects of ordering on the properties of Ni-Cr alloys. Journal of Materials Science, 23(6):2107-2114, 1988.

10. Z. Chen, Z. Huang, Z. Wang, and S. Zhu. Failure behavior of coated nickel-based superal-loy under thermomechanical fatigue. Journal of Materials Science, 44(23):6251-6257, 2009.

11. S. Roth, G. Hütter, and M. Kuna. Simulation of fatigue crack growth with a cyclic cohesive zone model. International Journal of Fracture, 188(1):23-45, 2014.

12. S. Roth, and M. Kuna. Finite element analyses of fatigue crack growth under small scale yielding conditions modelled with a cyclic cohesive zone approach. IN: E. Onate, D. Owen, D. Peric, and B. Suarez (ed.), Computational Plasticity XII – Fundamentals and Applications, COMPLAS XII, CIMNE, Barcelona, 1075-1086, 2013.

13. S. Roth, and M. Kuna. Numerical study on interfacial damage of sprayed coatings due to thermo-mechanical fatigue. IN: E. Onate, D. Owen, D. Peric, and B. Suarez (ed.), Compu-tational Plasticity XI – Fundamentals and Applications, COMPLAS XI, CIMNE, Barcelo-na, 1032-1043, 2011.

14. A. Schramm. Untersuchung des Einflusses der Schichtstruktur auf das Grenzflächenschä-digungsverhalten gespritzter Schichten, Masterarbeit, TU Bergakademie Freiberg, 2014.

References

240

Synthesis, properties and potential applications of rocksalt-type aluminium nitride (rs-AlN)

K. Keller 1, M. R. Schwarz 2, S. Schmerler 3, E. Kroke 2, G. Heide 1, D. Rafaja 4 and J. Kortus 3

1 Institute of Mineralogy, TU Bergakademie Freiberg, Freiberg, Germany

2 Institute of Inorganic Chemistry, TU Bergakademie Freiberg, Freiberg, Germany

3 Institute of Theoretical Physics, TU Bergakademie Freiberg, Freiberg, Germany

4 Institute of Materials Science, TU Bergakademie Freiberg, Freiberg, Germany

242

AbstractAluminum nitride usually occurs in the wurtzite-type structure (w-AlN). Sintered w-AlN ceramics are known for their very high thermal conductivity and other useful properties, pure w-AlN is a semiconductor with a wide band gap of 6.2 eV. This article provides an overview on the less known high-pressure phase of AlN with rocksalt structure (rs-AlN). A comprehensive summary on the high-pressure high-temperature synthesis of rs-AlN with static and dynamic shock wave experiments as well as physical and chemical properties is provided. Besides, emphasis is put on the potential technological and industrial relevance of rs-AlN.

Keywords: high pressure, phase transition, density functional theory, shock wave synthesis, hard material

1 Introduction

Under ambient conditions the hexagonal wurtzite structure is the stable phase of aluminium nitride (w-AlN). Because of its similar structure (four-fold coordination of Al with N atoms and vice versa) the cubic zincblende structure (zb-AlN) has almost the same enthalpy of formation (difference 24.5 meV/atom) [1] as w-AlN. Despite this fact it is formed only under certain conditions (e.g. homo- or hetero-epitactic growth of thin films) [2, 3]. At high pressure AlN transforms to the rocksalt structure (rs-AlN, in literature also fcc-AlN), which can be quenched to ambient conditions as a metastable phase [4]. This more dense phase (+25 %) shows a six-fold (octahedral) coordination of Al with the N atoms and vice versa.

Sintered w-AlN materials are outstanding ceramics with a high heat conductivity (single crystal 319 W/mK, ceramic 100-220 W/mK) [5], high electrical resistance (>1011 Ωm) [6] and good mechanical properties (hardness, bending strength and fracture toughness) [7, 8]. Because of its thermal expansion coefficient, which almost matches that of silicon [9], it can be used as a substrate for microelectronics and as a heat sink for high power applications [10]. Wurtzite-AlN is a wide band gap semiconductor (Eg = 6.2 eV) [11] and therefore an interesting material for optical applications, e.g. blue LEDs, laser diodes and UV detectors [12–14]. Although fine-grained w-AlN powders react readily with moisture, the sintered w-AlN ceramic

is very stable against oxidation, acids, metal melts and salt melts [7, 8]. On an industrial scale, most w-AlN powders are prepared by carbothermal reaction of aluminium oxide in nitrogen and ammonia, respectively [7]. Ceramics are manufactured by pressure- sintering or pressureless by the help of sinter additives (Y2O3) under nitrogen atmosphere at temperatures of 1600-1900 °C [7, 8, 15].

A large number of theoretical investigations on the high-pressure behaviour and phase transitions of the III-V semicondutors (AlN, GaN, InN, BP, AlP, GaAs) has been published, documenting the high importance of these materials (see part 2). It was shown that these materials transform either to rocksalt (NaCl), NiAs or Cmcm structure under pressure [16]. In addition to theoretical works, in the last two decades an increased number of high-pressure experiments was dedicated to the transformation of w-AlN into the rocksalt-type high-pressure form, which support the results of the theoretical studies (see part 3.1). The dependence of the phase transition on the stress state and particle size is described in part 3.2. With a transition pressure around 16 GPa for the solid-solid transformation from w-AlN [17] and a thermodynamic boundary between w-AlN and rs-AlN at around 13 GPa [18], rs-AlN requires two to three times higher pressures than contemporary industrial diamond or c-BN synthesis [19–21] and is yet therefore not in the reach of conventional ‘large volume’ high-pressure technology.

243

Investigations using shock waves have however shown that they provide a unique method to synthesise larger amounts of rs-AlN, making it available for further characterisation (see part 3.3). The role of rs-AlN and its solid solutions with transition metal nitrides MNx (M = Ti, Cr, etc.) in wear resistant ceramic coatings has long been recognized due to their great importance for the metal cutting and tooling industry [22, 23]. These coatings are typically manufactured via vacuum deposition and plasma techniques and a large body of literature exists in this area of research. Therefore these systems are only briefly addressed in part 4. Though w-AlN is used in a wide range of applications due to its unique properties, most properties of the high-pressure phase remain unknown. However, first studies showed promising properties of rs-AlN for the use as high-tech material (see part 5). The results of the theoretical and experimental investigations on rs-AlN will be summarised in part 6 and concluded with an outlook on future work and potential application of this material.

2 Theoretical studies on phase relations and properties

An overview of the high-pressure phases of several III-V semiconductors is given elsewhere [16, 24, 25]. Van Vechten predicted that w-AlN converts to the β-tin structure instead of the rocksalt structure at pressures of 90 GPa [26]. Christensen et al. predicted – besides the rocksalt structure – another high-pressure phase of AlN with NiAs structure at pressure above 30-40 GPa [27]. However, it was shown by experiments and theory that the rocksalt structure is stable up to at least 150 GPa (schock experiment and DAC) [28,29] or higher (200 GPa, DFT calculation) [1].

The calculated equilibrium transition pressure (equal enthalpy of both phases) for the wurtzite to rocksalt transformation in AlN scatters within a range from 4.6 to 26.9 GPa (see Table 1). However, most calculated transition pressures are lower than the experimental values, which are almost all above 18 GPa because of a considerable kinetic barrier (see Table 2).

The phase transition takes place by martensitic mechanism and can be observed also at room temperature. The energy barrier ΔH was calculated to be 144 meV/atom or 128 meV/atom, respectively, at the transition point [36, 43]. To overcome this enthalpy barrier a certain amount of energy at a given pressure, e.g. by a temperature raise, is needed. As the pressure is raised the barrier for the w-AlN → rs-AlN transition sinks, while it rises for the backward rs-AlN → w-AlN transition [36]. A broad two phase region (w-AlN + rs-AlN) was observed, so that the pressure for the complete conversion PT,c is much higher than the onset pressure PT,s of the transition. Therefore the phase transition is very sluggish, which is in contrast to other materials, e.g. w-ZnO → rs-ZnO, where a narrow hysteresis is observed [49].

Complete p–T-phase diagrams of AlN were calculated by Siegel et al. [50], Norrby et al. [51] and by Schmerler and Kortus [52]. Furthermore theoretical predictions for the zincblende → rocksalt transition of AlN were carried out in references [41, 53].

244

Table 1: Theoretical investigations on the w-AlN → rs-AlN phase transition, lattice constant a and bulk modulus B0 of rs-AlN.

a equilibrium phase transition pressure (no hysteresis) b equilibrium lattice constant a of rs-AlN c shock loading (impact) on AlN ceramic d lattice constant at 120 GPa from MD calculation e pressure for completion of phase transition PT,c = 23.5 GPa

method PTa ΔV ab B0 year Ref.

[GPa] [%] [Å] [GPa]

DFT-LDA 16.6 18.1 221 1991 [30]

DFT-LDA 12.9 22.5 4.032 215.0 1991 [31]

DFT-LDA 12.5 19 270 1993 [27]

DFT-LDA 16.6 19.1 4.06 281 1994 [32]

DFT-LDA 9.2 20.1 3.978 272 2000 [1]

DFT-LDA 8.3 2004 [33]

DFT-LDA 20.91 20.67 4.031 274.87 2005 [34]

DFT-GGA 26.91 20.43 4.052 260.60 2005 [34]

DFT-LDA 9.50 3.98 261 2005 [35]

DFT-LDA 9.8 4.021 2007 [36]

DFT-GGA 12.8 20 2007 [37]

DFT-GGA + MDc 25 19 2008 [38]

Periodic Cluster Model 4.56 2008 [39]

DFT-GGA 15.0 4.068 246 2008 [40]

DFT-GGA 17 3.9820 2008 [41]

DFT-GGA + MD 17.8 3.78d 253 2009 [42]

DFT-LDA 9.85 4.02 266 2009 [43]

DFT-GGA 12 4.068 2010 [44]

DFT-GGA 13.3 4.07 251 2010 [45]

DFT-GGA 13.9e 19.39 2010 [46]

DFT-LDA 12.7 18.9 4.001 282.6 2010 [47]

DFT-LDA 13.5 4.016 268.47 2014 [48]

245

3 Experimental investigation on the phase transition and synthesis

3.1 Formation of rs-AlN under static high-pressure

An overview of the experimental work on rs-AlN is given in Table 2. The first synthesis of rs-AlN in combination with structural investigations of the high-pressure modification of AlN was done in the early 1990’s by Vollstädt et al. [4]. The starting material (w-AlN, <1 μm) was compressed to 16.5 GPa and heated to temperatures of 1400-1600 °C using a multi anvil device driven by a 1000 t hydraulic press (MAP). After decompression and quenching to room temperature the samples were charaterised by XRD showing – besides the w-AlN peaks – additional peaks, which were indexed completely to a rocksalt structure. After this work more attention was paid to the high-pressure properties of the III-V semiconductors. Besides more theoretical work on the wurtzite-rocksalt phase transition and the electronic properties of rs-AlN, rs-GaN and rs-InN, some experimental works were conducted using diamond anvil cells (DAC). The transition was observed visually (by blackening of the sample) [30, 54], by Raman spectroscopy (loss of Raman signal due to forbidden Raman scattering of rs-AlN) [55, 56], by in-situ XRD [28, 54, 57] and by second harmonic generation (SHG) measurements [17, 49]. The bulk modulus B0 of rs-AlN was determined from compression data by fitting a Birch-Murnaghan equation of state (295 ± 17 GPa for microcrystalline rs-AlN) [28]. The B0 of the rocksalt phase is significantly higher than for the wurtzite phase (207.9 ± 6.3GPa [54]). The bulk modulus of nanocrystalline rs-AlN was found to be even higher (up to 359 ± 27 GPa [58]), a phenomenon that has also been observed for other nanocrystalline material [23].

The first experimental p–T-phase diagram was suggested by Bayarjargal et al. using results of laser-heated DAC experiments [17]. The onset and completion of the phase transition were determined by second harmonic generation (SHG), which indicates a change in crystallographic symmetry and hence the structural transition from wurtzite to rocksalt structure. Further work to define the phase boundary was done by Schwarz et al. using a multi anvil press (MAP) [18]. By using an Al–N–H precursor the kinetic barrier (c.f. to section 2) was lowered, so that the phase transition pressure is closer to the true thermodynamic phase boundary. It was found that in the temperature range of 1000-2000 K the phase boundary is between 11-12 GPa (see Figure 1), which is lower than in previous experiments and calculations. Moreover, all experimental and theoretical works consistently show that the clapeyron slope of the boundary is negative so that a pressure not much above 10 GPa may be sufficient to obtain rs-AlN if very high temperatures can be applied.

246

Table 2: Experimental investigations on the phase transition of rs-AlN (static and dynamic).

a PT,s start of phase transition b PT,c completion of phase transition c lattice constant a of rs-AlN at ambient pressure and RT d visual: 18-20 GPa e lattice constant a at high-pressure f at pressure release partly reconversion below 1.3 GPa to w-AlN (phase transition at 25 GPa not complete) g visual (blackening of the sample)

method examination PT,Sa PT,c

b T ΔV ac B0(rs) comments year Ref.

[GPa] [GPa] [°C] [%] [Å] [GPa]

shock Hugoniot 21 22 NaCl-structure supposed 1982 [66]

MAP quenched 16.5 1400-1600 21 4.045 1990 [4]

DAC visual 16-17 RT 1991 [30]

DAC in-situ XRD 22.9d RT 17.9 3.938e 1992 [54]

DAC Raman 16-17 RT 1993 [55]

DAC in-situ XRD 14 ≈20 RT 18.6 4.043 221.0 ± 5.0 no pressure medium 1993 [57]

shock Hugoniot 22 20 1994 [67]

shock Hugoniot 19 RT 1994 [68]

DAC in-situ XRD 20.0 31.4 RT 295 ± 17 1997 [28]

shock Hugoniot 19.2 ≈75 21 304 ± 4 1999 [29]

DAC in-situ XRD 14.5 ≈35 RT 20.5 4.045 359 ± 27 nanoparticle 2004 [63]

DAC in-situ XRD 24.9 45.4 RT 20 4.064 319.2 ± 7.6 nanowire 2006 [69]

DAC Raman 18 >25f RT 2008 [56]

DAC in-situ XRD 21.5 27.8 RT 19.5 3.95e 215.8 ± 8.5 nanoparticle 2008 [64]

DAC in-situ XRD 21.5 27.8 RT 20.0 3.91e 208.8 ± 8.8 nanowire 2008 [64]

DAC Raman 24.4 ≤33.1 RT nanowire 2010 [70]

shock quenched <15 750-2100 4.043-4.046 nanoparticle 2012 [71]

DAC SHG 26 26 RT pressure medium KCl 2012 [17]

DAC SHG 22.5 27.7 RT pressure medium Ne 2012 [17]

DAC in-situ XRD 18.6 29.8 RT 19.5 4.048 291 ± 8 AlN:Sc nanoprism 2013 [72]

DAC in-situ XRD 16.2 26.5 RT 19.3 4.044 283 ± 4 AlN:Y nanoprism 2013 [72]

DAC Raman 20.73 RT dendritic crystal 2013 [73]

DAC in-situ XRD 17.7 33.2 RT 21 4.051 327.3 ± 3.9 AlN:Mg nanowire 2013 [74]

DAC in-situ XRD 15.0 31.0 RT 21 4.054 321.0 ± 4.9 AlN:Co nanowire 2013 [74]

DAC IR & XRS 15g >30 RT 2013 [75]

MAP quenched 11-12 1000-2000 4.042-4.046 AlN2H3-precursor 2014 [18]

247

3.2 Influence of stress state and particle size on transition pressure

As it can be seen from Table 2, the transition pressure for the direct conversion of w-AlN to rs-AlN displays a large scattering. In a recent work the influence of shear stress on the transition pressure was investigated by using two different pressure media in DAC experiments (see Table 3) [17]. The transition pressure under quasi-hydrostatic conditions (24.5 GPa) is 6.5 GPa higher than under non-hydrostatic conditions (18 GPa). Also the mixed-phase region (w-AlN + rs-AlN) for quasi-hydrostatic conditions is only half as large for non-hydrostatic conditions. By this fact some discrepancy between the transition pressures observed by other groups can be explained, e.g. the low transition of 14 GPa observed by Xia et al. [57], which is due to the non-hydrostatic pressure (sample loaded without pressure medium).

Another important factor that determines which phase is stable at a certain condition is – besides pressure and temperature – the crystallite size. For most materials the

reduction of particle size results in a higher transition pressure, e.g. for ZnO, CdSe [59, 60], while in some other systems it leads to a reduction of the transition pressure [61, 62]. For w-AlN a reduced transition pressure was observed for nanocrystalline material [49, 63, 64]. From the experimental data the dependence of the crystallite size on the transition pressure was fitted to PT(D) = 22.99 – 68.92 ∙ 1/D with D crystallite size in nanometer [49]. Though the same trend of decreasing pressure with crystallite size was calculated using the Periodic Cluster Model [39], the correlation with the absolute values of pressure is rather poor. The reason for the size-induced reduction of the transition pressure was discussed by the volumetric expansion of nano-AlN, the higher surface energy and a decrease of Poisson`s ratio [58]. The surface energy and surface stress of AlN in the wurtzite and rocksalt phase were calculated [49, 65]. It was found that the surface energy rather than the surface stress dominates the dependence of the transition

Figure 1: Experimental p–T-phase diagram (modified after [18]). The phase boundary calculated by Siegel et al. [50] does not match experiments, while the phase boundaries calculated by Norrby et al. [51] and Schwarz et al. [18] are in good agreement with the experimental data. Additional data points (high-pressure experiments) are from further publications [4, 17, 57].

248

pressure [49]. It can be concluded that the surface energy of the wurtzite phase is higher than the surface energy of the rocksalt phase and therefore a reduction in particle size leads to a decrease of the transition pressure. It is predicted that AlN particles with the size of 8.5 nm are stable in the rocksalt structure even under ambient conditions [49].

For w-AlN nanowires a higher transition pressure (21.5-24.9 GPa [64, 69, 70]; 33.2 GPa and 31.0 GPa for Mg and Co-doped AlN nanowire, respectively [74]) and also a larger two-phase region was found than for bulk AlN.

a PT,s start of phase transition; b PT,c completion of phase transition

Table 3: Dependence of transition pressure for w-AlN → rs-AlN from crystallite size and stress state.

3.3 Shock wave synthesis

The first high-pressure experiment (above the phase transition pressure) on w-AlN was done in 1982 by Kondo et al. under shock loading [66]. Hugoniot measurements (response of a material to intensive shock loading) of dense sintered w-AlN ceramics were performed using a two-stage light-gas gun. A discontinuity in the Hugoniot curve associated with a volume change of −22 % at 21 ± 1 GPa indicated a phase transition into a denser modification. In analogy to the phase transition of w-ZnS to the rocksalt structure, the same phase transformation was proposed for w-AlN. Further studies on the dynamical behaviour of AlN ceramics verify

the phase transition at similar pressures (19-22 GPa) [29, 67, 68, 76, 77] (see also Table 2). Because the sample cannot be recovered with the experimental set-up used by the authors, no verification of the structure of the high-pressure phase was possible. The quenching of the high-pressure modification of AlN in shock wave recovery experiments using w-AlN powder as starting material was attempted, but failed [78]. Also no rs-AlN could be recovered from shock compression of AlN and AlN–Al2O3 powders, though the maximum pressure was above the phase transition pressure [79].

stress particle size PT PT,sa PT,c

b Ref.

nm GPa GPa GPa

quasi-hydrostatic bulk 24.5 22.5 27.7 [17]

quasi-hydrostatic bulk 22.9 [54]

quasi-hydrostatic 100 21.5 18 28 [49]

quasi-hydrostatic 45 24.65 21.5 27.8 [64]

quasi-hydrostatic 20 14 12 19 [49]

quasi-hydrostatic 10 14.5 ≈35 [63]

non-hydrostatic bulk 18 16 26 [17]

non-hydrostatic 100 18 15 22 [49]

249

The first successful recovery of rs-AlN from shock wave experiments was accomplished by the authors [80]. In an explosivly-driven device with flyer-plate set-up nanocrystalline w-AlN (20-50 nm) was shocked without additives in a pressure range of 15-43 GPa, using the reflection geometry [81] with sample thickness typically ≤2 mm. Multiple reflections of the shock wave on the bottom and top of the sample causes a stepwise compression. Due to the porosity of the w-AlN powder heat is generated during shock loading by a gradual increase which helped to overcome the energy barrier for the transition. Interestingly, the yield of the recovered rs-AlN raised with increasing peak shock pressure reaching a maximum at about 25 GPa, but decreased to zero at pressures >40 GPa. A similar correlation was found by varying the shock temperature (powder porosity) at constant peak pressure. The increase of rs-AlN yield can be explained by the thermal activation of the transition due to higher temperature. However, when the post-shock temperature is too high due to insufficient thermal conduction in the vicinity of the sample, the thermally activated reconversion of the metastable phase to the stable phase w-AlN is the dominating process and prevents the (full) recovery of the rs-AlN formed during the shock [71]. For higher cooling rates the impedance method was used [81], in which the sample is only subjected to a single shock. The AlN was mixed with copper powder, which transmits the pressure and causes high cooling rates on the release path after passage of the schock wave. At medium temperatures (750-1200 °C) almost pure rs-AlN (besides oxygen phases from contamination) was recovered [71]. With the impedance method the cooling was even fast enough to partly recover rs-AlN synthesised from micron-AlN, not only from nano-AlN.

4 Rocksalt-type AlN in multinary and composite materials

Protective hard coatings on the basis of ternary transition metal nitrides with aluminium nitride (M–Al–N with M = Ti, Cr, Zr, Hf, V) have a great industrial importance for the fabrication of wear resistant cutting tools with high hardness combined with good oxidation resistance [82–85]. Because the hardness of rs-AlN (39 GPa) is higher than that for the transition metal nitrides high hardness values for M–Al–N coatings can be obtained. However, the hardness is not only controlled by the intrinsic hardness of the nitrides, but also by specific microstructural features like small crystallite size and crystallite interlocking [23, 85]. Because the coatings are usually deposited at relatively low temperatures on tool substrates using vacuum deposition and plasma techniques, metastable compositions M1-xAlxN with a rocksalt structure, very small crystallite size and high microstrains can form.

Upon heating or even during the deposition process, spinodal decomposition of the M–Al–N with segregation of rs-AlN was observed [86, 87]. Because of the spinodal decomposition the lattice plane spacings in the coatings vary, caused by the local fluctuations of the chemical composition. This local heteroepitaxy of rs-AlN results in microstrains, which lead to the stabilisation of the metastable rs-AlN phase. Further heating leads to the conversion of the metastable rs-AlN to the stable w-AlN, which causes a loss of the coherent MN–AlN interfaces and leads to a massive decrease of the hardness. It has been shown on coating material that had been stripped from the substrate that an increase in pressure stabilises rs-AlN to higher temperatures [51]. Also, in coatings consisting of epitaxial superlattices (MN/AlN), the rocksalt structure of AlN can be stabilised, if the AlN layers are thin enough [88–90].

250

5 Properties of rs-AlN

By calculation of the band structure of rs-AlN at ambient conditions indirect band gap energies Eg of 4.04 eV [31], 4.53 eV [91] and 4.65 eV [32] were predicted. It was shown that the material is transparent from partly UV to the visual range of light [91]. The dielectric constant at high frequencies is calculated to be ε(∞) = 5.03 [92]. Also thermodynamic properties, including thermal expansion coefficient and heat capacity of rs-AlN in dependence of temperature were calculated [52, 93].

The first material properties were reported by Vollstädt et al. in a patent claiming sintered composite bodies of w-AlN/rs-AlN manufactured by MAP experiments, however, a corresponding publication in a peer-reviewed journal confirming the reported values was never published. According to the patent, the bulk material shows high hardness (≤4500 HV), a high electrical resistivity (1014 Ωm) and a thermal conductivity of 250-600 W/mK, which is up to six times higher than for w-AlN-ceramics [103]. Own hardness measurements of rs-AlN with a Knoop indenter, carried out on a sintered sample from MAP experiment, showed hardness of 33 GPa (25 g load) and 28-30 GPa (50 g load), which means that it is twice as hard as w-AlN (≈15 GPa) [18]. The hardness of transition metal nitrides (TiN, ZrN, VN, etc.) has been correlated to the elastic constants (bulk modulus B0, Youngs modulus E and shear modulus G), which were obtained by DFT calculations [102]. A good agreement with experimental data (Vickers hardness) for several materials was found by the description H = 2(G3/K2)0.585 - 3 [104], which gives a hardness of 39 GPa for rs-AlN [102]. A theoretical Vickers hardness was very recently reported to be 39.5 GPa [58].

With respect to experimental indentation hardness, it should be noted that hardness readings always depend on the load and indenter shape, also an increasing hardness with decreasing load is frequently observed. For example, for (super)hard nano-polycrystalline cBN it was found that hardness measurements using a Vickers indenter results in hardness readings that are much too high due to a pronounced elastic recovery, while using the Knoop method ensures accurate hardness values [105]. Taking these facts into account, the calculated and measured hardness values are in good agreement and show that rs-AlN is an ‘almost superhard’ (>40 GPa) material.

It was proven that the metastable rs-AlN is also long-term stable (several months) and no back-transformation at ambient conditions occurred [106]. It was shown that rs-AlN (powder) is also chemically stable. It is resistant against several acids and caustic soda at room temperature for several days. By heating the sample followed by ex-situ phase analysis, the thermal stability at different atmospheres was tested. The thermally activated reconversion of rs-AlN → w-AlN starts at 1200 °C (argon, 10 K/min), 1100 °C (vacuum, 10 K/min) and 1000 °C (vacuum, isotherm) [18].

251

Table 4: Comparison of properties of AlN in wurtzite and rocksalt structure at room temperature and ambient pressure.

a w-AlN/rs-AlN sinterbody, no single crystalb using Knoop indenter methodc Vickers hardnessd theoretical value

w-AlN rs-AlN

band gap Eg [eV] direct6.2 [11]

indirect4.04-4.65 [31, 32, 91]

dielectric constant ε(∞) 4.84 [94] 5.03 [92], 5.13 [95]

thermal conductivity λ [W/mK] 319 [5] ≤600a [4]

thermal expansion αa [K-1] 3.021 ∙ 10-6 [96] 7.7 ∙ 10-6 [52], 10.3 ∙ 10-6 [93]

αc [K-1] 2.209 ∙ 10-6 [96]

heat capacity cP [J/gK] 0.69 [97], 0.74 [98] 0.77 [52]

electrical resistivity ρ [Ωm] >1011 [6] 1014 [4]

bulk modulus B0 [GPa] 207.9 ± 6.3 [54] 295 ± 17 [28]

pressure derivative of B0 B0' 6.3 ± 0.9 [54] 3.5 ± 0.4 [28]

shear modulus G [GPa] 122 [45] 242 [58]

Young's modulus E [GPa] 308 [99] 563 [58]

hardness H [GPa] 12b [100]17.7c [101]

28-33b [18]39c,d [102]

elastic constants [45] C11 [GPa] 376 423

C12 [GPa] 127 167

C13 [GPa] 97

C33 [GPa] 355

C44 [GPa] 112 306

Debye temperature Θ [K] 1028 [40] 1278 [40]

Grüneisen parameter γ 1.802 [40] 1.685 [40]

252

6 Conclusions

In this contribution an overview of theoretical and experimental work on the high-pressure phase of AlN with rocksalt structure is given.

Due to the kinetic barrier and a negative clapeyron-slope of the w-AlN/rs-AlN phase boundary, the transition is favored towards higher temperatures, which can help to bring down the high pressures necessary for static synthesis. However, the temperature for back-transformation (rs-AlN → w-AlN) at ambient pressure (1000-1100 °C) poses a limit for the sample temperature after pressure release, which is of particular importance for the full recovery of rs-AlN from shock wave experiments.

Phase-pure rs-AlN with micron-sized crystals has been obtained via static synthesis. However, because of the relatively high transition pressure (13 GPa) the usage of large volume presses in an industrial scale is restricted. Shock wave synthesis is a promising method to obtain larger amounts of rs-AlN for comprehensive testing and production. However, the synthesis has to be improved to obtain pure rs-AlN powder without w-AlN, secondary phases and low oxygen content, which still remains a challenging task. Another drawback of this method is, that only fine powders are obtained, with properties which may significantly differ from that of bulk rs-AlN due to the large surface area. For many applications sintered compacts of rs-AlN are required – and sintering of nano rs-AlN powders at pressures below the w-AlN/rs-AlN transition pressure remains a so far unsolved and challenging task.

Several materials properties of rs-AlN which were reported in literature and compiled by the authors are superior to the properties of w-AlN. These include thermal conductivity, hardness and elastic moduli. Besides, rs-AlN has shown to be thermally and chemically very stable. These promising results cause in growing interest in rs-AlN from the material point of view and demonstrates the potential of rs-AlN as a (super)hard material and for the use in the microelectronic industry, for example.

Acknowledgement

This work was performed within the Cluster of Excellence “Structure Design of Novel High-Performance Materials via Atomic Design and Defect Engineering (ADDE)” that is financially supported by the European Union (European regional development fund) and by the Ministry of Science and Art of Saxony (SMWK). Further contributions of the results reported herein were obtained in the course of research conducted within the Freiberg High-Pressure Research Centre (FHP), financially supported by the “Dr.-Erich-Krüger-Foundation” and the Priority Programme SPP1236 “Synthesis, ‘in situ’ characterisation and quantum mechanical modeling of Earth Materials, oxides, carbides and nitrides at extremely high pressures and temperatures” of the German Research Foundation (DFG).

253

References

1. J. Serrano, A. Rubio, E. Hernandez, A. Munoz, and A. Mujica. “Theoretical study of the relative stability of structural phases in group-III nitrides at high pressures”. Phys. Rev. B 62 (2000), pp. 16612–16623.

2. I. Petrov, E. Mojab, R. C. Powell, J. E. Greene, L. Hultman, and J.-E. Sundgren. “Synthesis of metastable epitaxial zinc-blende-structure AlN by solid-state reaction”. Appl. Phys. Lett. 60 (1992), pp. 2491–2493.

3. M. P. Thompson, G. W. Auner, T. S. Zheleva, K. A. Jones, S. J. Simko, and J. N. Hilfiker. “Deposition factors and band gap of zinc-blende AlN”. J. Appl. Phys. 89 (2001), pp. 3331–3336.

4. H. Vollstädt, E. Ito, M. Akaishi, S. Akimoto, and O. Fukunaga. “High Pressure Synthesis of Rocksalt Type of AlN”. Proc. Japan Acad., Ser. B 66 (1990), pp. 7–9.

5. G. A. Slack, R. A. Tanzilli, R. O. Pohl, and J. W. Vandersande. “The intrinsic thermal conductivity of AlN”. J. Phys. Chem. Solids 48 (1987), pp. 641–647.

6. A. W. Weimer. Carbide, Nitride and Boride Materials Synthesis and Processing. Ed. by A. W. Weimer. Chapmann & Hall, 1997.

7. B. Mussler and D. Brunner. Aluminiumnitridkeramik. In: Technische Keramische Werkstoffe (DKG). J. Kriegesmann, 2001. pp. 1–33.

8. P. Boch, J. C. Glandus, J. Jarrige, J. P. Lecompte, and J. Mexmain. “Sintering, oxidation and mechanical properties of hot pressed aluminium nitride”. Ceram. Int. 8 (1982), pp. 34–40.

9. G. A. Slack and S. F. Bartram. “Thermal expansion of some diamondlike crystals”. J. Appl. Phys. 46 (1975), pp. 89–98.

10. D. G. Brunner. “Ceramics in power electronics”. In: CFI-Ceramic Forum International. Vol. 87. Goller Verlag GmbH, 2010, pp. 25–27.

11. H. Yamashita, K. Fukui, S. Misawa, and S. Yoshida. “Optical properties of AlN epitaxial thin films in the vacuum ultraviolet region”. J. Appl. Phys. 50 (1979), pp. 896–898.

12. O. Ambacher. “Growth and applications of Group III-nitrides”. J. Phys. D: Appl. Phys. 31 (1998), pp. 2653–2710.

13. P. Kung and M. Razeghi. “III-Nitride wide bandgap semiconducters: a survey of the current status and future trends of the material and device technology”. Opto-Electron. Rev. 8 (2000), pp. 201–239.

14. M. Razeghi, A. Yasan, R. McClintock, K. Mayes, D. Shiell, S. R. Darvish, and P. Kung. “Review of III-nitride optoelectronic materials for light emission and detection”. Phys. Status Solidi C 1 (2004), pp. 141–148.

15. M. L. Panchula and J. Y. Ying. “Nanocrystalline Aluminum Nitride: II, Sintering and Properties”. J. Am. Ceram. Soc. 86 (2003), pp. 1121–1127.

16. G. J. Ackland. “High-pressure phases of group IV and III-V semiconductors”. Rep. Prog. Phys. 64 (2001), pp. 483–516.

17. L. Bayarjargal and B. Winkler. “High (pressure, temperature) phase diagrams of ZnO and AlN from second harmonic generation measurements”. Appl. Phys. Lett. 100 (2012), p. 021909.

18. M. R. Schwarz, M. Antlauf, S. Schmerler, K. Keller, T. Schlothauer, J. Kortus, G. Heide, and E. Kroke. “Formation and properties of rocksalt-type AlN and implications for high pressure phase relations in the system Si–Al–O–N”. High Pressure Res. 34 (2014), pp. 22–38.

19. J. Haines, J. Leger, and G. Bocquillon. “Synthesis and Design of Superhard Materials”. Annu. Rev. Mater. Sci. 31 (2001), pp. 1–23.

254

20. L. Vel, G. Demazeau, and J. Etourneau. “Cubic boron nitride: synthesis, physicochemical properties and applications”. Mater. Sci. Eng. B 10 (1991), pp. 149–164.

21. B. Pepekin. “Synthesis of diamond: A review”. Russ. J. Phys. Chem. B+ 4 (2010), pp. 769–772.

22. S. PalDey and S. C. Deevi. “Single layer and multilayer wear resistant coatings of (Ti,Al)N: a review”. Materials Science and Engineering: A 342 (2003), pp. 58–79.

23. D. Rafaja, C. Wüstefeld, M. Motylenko, C. Schimpf, T. Barsukova, M. R. Schwarz, and E. Kroke. “Interface phenomena in (super)hard nitride nanocomposites: from coatings to bulk materials”. Chem. Soc. Rev. 41 (2012), pp. 5081–5101.

24. A. Mujica, A. Rubio, A. Muñoz, and R. J. Needs. “High-pressure phases of group-IV, III-V, and II-VI compounds”. Rev. Mod. Phys. 75 (2003), pp. 863–912.

25. R. J. Nelmes and M. I. McMahon. “Structural Transitions in the Group IV, III-V, and II-VI Semiconductors under Pressure”. In: Semiconductors and Semimetals. Ed. by T. Suski and W. Paul. Vol. 54. Elsevier, 1998. pp. 145–246.

26. J. A. Van Vechten. “Quantum Dielectric Theory of Electronegativity in Covalent Systems. III. Pressure-Temperature Phase Diagrams, Heats of Mixing, and Distribution Coefficients”. Phys. Rev. B 7 (1973), pp. 1479–1507.

27. N. E. Christensen and I. Gorczyca. “Calculated structural phase transitions of aluminum nitride under pressure”. Phys. Rev. B 47 (1993), pp. 4307–4314.

28. S. Uehara, T. Masamoto, A. Onodera, M. Ueno, O. Shimomura, and K. Takemura. “Equation of state of the rocksalt phase of III-V nitrides to 72 GPa or higher”. J. Phys. Chem. Solids 58 (1997), pp. 2093–2099.

29. T. Mashimo, M. Uchino, A. Nakamura, T. Kobayashi, E. Takasawa, T. Sekine, Y. Noguchi, H. Hikosaka, K. Fukuoka, and Y. Syono. “Yield properties, phase transition, and equation of state of aluminum nitride (AlN) under shock compression up to 150 GPa”. J. Appl. Phys. 86 (1999), pp. 6710–6716.

30. I. Gorczyca, N. E. Christensen, P. Perlin, I. Grzegory, J. Jun, and M. Bockowski. “High pressure phase transition in aluminium nitride”. Solid State Commun. 79 (1991), pp. 1033–1034.

31. P. E. Van Camp, V. E. Van Doren, and J. T. Devreese. “High-pressure properties of wurtzite and rocksalt-type aluminum nitride”. Phys. Rev. B 44 (1991), pp. 9056–9059.

32. N. E. Christensen and I. Gorczyca. “Optical and structural properties of III-V nitrides under pressure”. Phys. Rev. B 50 (1994), pp. 4397–4415.

33. A. M. Saitta and F. Decremps. “Unifying description of the wurtzite-to-rocksalt phase transition in wide-gap semiconductors: The effect of d electrons on the elastic constants”. Phys. Rev. B 70 (2004), p. 035214.

34. S. Saib and N. Bouarissa. “Structural properties of AlN from first principles calculations”. Eur. Phys. J. B 47 (2005), pp. 379–383.

35. C. C. Silva, H. W. Leite Alves, L. M. R. Scolfaro, and J. R. Leite. “Pressure-induced phase transitions and polytypic structures in III-Nitrides”. Phys. Status Solidi C 2 (2005), pp. 2468–2471.

36. J. Cai and N. Chen. “Microscopic mechanism of the wurtzite-to-rocksalt phase transition of the group-III nitrides from first principles”. Phys. Rev. B 75 (2007), p. 134109.

37. R. Laskowski and N. E. Christensen. “Pressure-induced delocalization-to-localization transition of excitons in AlN”. Phys. Rev. B 75 (2007), p. 201202.

38. P. S. Branicio, R. K. Kalia, A. Nakano, P. Vashishta, F. Shimojo, and J. P. Rino. “Atomistic damage mechanisms during hypervelocity projectile impact on AlN: A large-scale parallel molecular dynamics simulation study”. J. Mech. Phys. Solids 56 (2008), pp. 1955–1988.

39. A. Costales, M. A. Blanco, E. Francisco, C. J. F. Solano, and A. M. Pendas. “Theoretical Simulation of AlN Nanocrystals”. J. Phys. Chem. C 112 (2008), pp. 6667–6676.

255

40. F. Peng, D. Chen, H. Fu, and X. Cheng. “The phase transition and the elastic and thermodynamic properties of AlN: First principles”. Physica B 403 (2008), pp. 4259–4263.

41. Y. C. Cheng, X. L. Wu, J. Zhu, L. L. Xu, S. H. Li, and P. K. Chu. “Optical properties of rocksalt and zinc blende AlN phases: First-principles calculations”. J. Appl. Phys. 103, 073707 (2008), p. 073707.

42. M. Durandurdu. “Pressure-induced phase transition in AlN: An ab initio molecular dynamics study”. J. Alloy Compd. 480 (2009), pp. 917–921.

43. B. Xu. “Theoretical study of pressure-induced phase transitions and thermal properties for maingroup oxides and nitrides”. PhD thesis. Auburn University, 2009.

44. H. Y. Xiao, X. D. Jiang, G. Duan, F. Gao, X. T. Zu, and W. J. Weber. “First-principles calculations of pressure-induced phase transformation in AlN and GaN”. Comp. Mater. Sci. 48 (2010), pp. 768–772.

45. A. J. Wang, S. L. Shang, Y. Du, Y. Kong, L. J. Zhang, L. Chen, D. D. Zhao, and Z. K. Liu. “Structural and elastic properties of cubic and hexagonal TiN and AlN from first-principles calculations”. Comp. Mater. Sci. 48 (2010), pp. 705–709.

46. Y. Cai and R. Xu. “First-principles study of the angle deformation path in the wurtzite-to rocksalt phase transition in aluminum nitride”. Chinese Phys. Lett. 27 (2010), p. 096402.

47. W. Feng, S. Cui, H. Hu, W. Zhao, and Z. Gong. “Structural stability and optical properties of AlN explored by ab initio calculations”. Physica B 405 (2010), pp. 555–558.

48. S. Louhibi-Fasla, H. Achour, K. Kefif, and Y. Ghalem. “First-principles Study of High-pressure Phases of AlN”. Physics Procedia 55 (2014). 8th International Conference on Material Sciences, CSM8-ISM5, pp. 324–328.

49. L. Bayarjargal, L. Wiehl, and B. Winkler. “Influence of grain size, surface energy, and deviatoric stress on the pressure-induced phase transition of ZnO and AlN”. High Pressure Res. 33 (2013), pp. 642–651.

50. A. Siegel, K. Parlinski, and U. D. Wdowik. “Ab initio calculation of structural phase transitions in AlN crystal”. Phys. Rev. B 74 (2006), p. 104116.

51. N. Norrby, H. Lind, G. Parakhonskiy, M. P. Johansson, F. Tasnadi, L. S. Dubrovinsky, N. Dubrovinskaia, I. A. Abrikosov, and M. Oden. “High pressure and high temperature stabilization of cubic AlN in Ti0.60Al0.40N”. J. Appl. Phys. 113 (2013), p. 053515.

52. S. Schmerler and J. Kortus. “Ab initio study of AlN: Anisotropic thermal expansion, phase diagram, and high-temperature rocksalt to wurtzite phase transition”. Phys. Rev. B 89 (2014), p. 064109.

53. A. Srivastava and N. Tyagi. “Pressure induced zincblende to rocksalt phase transition in AlN nanocrystal”. Journal of Physics: Conference Series 377 (2012), p. 012066.

54. M. Ueno, A. Onodera, O. Shimomura, and K. Takemura. “X-ray observation of the structural phase transition of aluminum nitride under high pressure”. Phys. Rev. B 45 (1992), pp. 10123–10126.

55. P. Perlin, A. Polian, and T. Suski. “Raman-scattering studies of aluminum nitride at high pressure”. Phys. Rev. B 47 (1993), pp. 2874–2877.

56. F. J. Manjón, D. Errandonea, A. H. Romero, N. Garro, J. Serrano, and M. Kuball. “Lattice dynamics of wurtzite and rocksalt AlN under high pressure: Effect of compression on the crystal anisotropy of wurtzite-type semiconductors”. Phys. Rev. B 77 (2008), p. 205204.

57. Q. Xia, H. Xia, and A. L. Ruoff. “Pressure-induced rocksalt phase of aluminum nitride: A metastable structure at ambient condition”. J. Appl. Phys. 73 (1993), pp. 8198–8200.

58. A. Wang, S.-Li Shang, M. He, Y. Du, Li Chen, R. Zhang, D. Chen, B. Fan, F. Meng, and Zi-K. Liu. “Temperature-dependent elastic stiffness constants of fcc-based metal nitrides from first-principles calculations”. J. Mater. Sci. 49 (2014), pp. 424–432.

59. J. Z. Jiang, J. S. Olsen, L. Gerward, D. Frost, D. Rubie, and J. Peyronneau. “Structural stability in nanocrystalline ZnO”. Europhys. Lett. 50 (2000), p. 48.

256

60. S. H. Tolbert and A. P. Alivisatos. “Size Dependence of a First Order Solid-Solid Phase Transition: The Wurtzite to Rock Salt Transformation in CdSe Nanocrystals”. Science 265 (1994), pp. 373–376.

61. Z. Wang, S. K. Saxena, V. Pischedda, H. P. Liermann, and C. S. Zha. “In situ x-ray diffraction study of the pressure-induced phase transformation in nanocrystalline CeO2”. Phys. Rev. B 64 (2001), p. 012102.

62. Z. Jiang, J. S. Olsen, L. Gerward, and S. Morup. “Enhanced bulk modulus and reduced transition pressure in γ-Fe2O3 nanocyrstals”. Europhys. Lett. 44 (1998), pp. 620–626.

63. Z. Wang, K. Tait, Y. Zhao, D. Schiferl, C. Zha, H. Uchida, and R. Downs. “Size-Induced Reduction of Transition Pressure and Enhancement of Bulk Modulus of AlN Nanocrystals”. J. Phys. Chem. B 108 (2004), pp. 11506–11508.

64. W. W. Lei, D. Liu, J. Zhang, Q. L. Cui, and G. T. Zou. “Comparative studies of structural transition between AIN nanocrystals and nanowires”. Journal of Physics: Conference Series 121 (2008), p. 162006.

65. D. Holec and P. H. Mayrhofer. “Surface energies of AlN allotropes from first principles”. Scripta Materialia 67 (2012), pp. 760–762.

66. K. Kondo, A. Sawaoka, K. Sato, and M. Ando. “Shock compression and phase transformation of AlN and BP”. AIP Conf. Proc. 78 (1982), pp. 325–329.

67. M. E. Kipp and D. E. Grady. “Shock phase transformation and release properties of aluminum nitride”. J. Phys. IV France 04 (1994), pp. 249–256.

68. A. Nakamura and T. Mashimo. “Shock-induced phase transition of AlN”. AIP Conf. Proc. 309 (1994), pp. 303–305.

69. L. H. Shen, X. F. Li, Y. M. Ma, K. F. Yang, W. W. Lei, Q. L. Cui, and G. T. Zou. “Pressure induced structural transition in AlN nanowires”. Appl. Phys. Lett. 89 (2006), p. 141903.

70. L. Shen, Q. Cui, Y. Ma, Q. Li, Y. Ma, B. Liu, and T. Cui. “Raman Scattering Study of AlN Nanowires under High Pressure”. J. Phys. Chem. C 114 (2010), pp. 8241–8244.

71. K. Keller. “Schockwellensynthese und Charakterisierung von Aluminiumnitrid mit Kochsalzstruktur”. PhD thesis. TU Bergakademie Freiberg, 2013.

72. R. Cong, H. Zhu, X. Wu, C. Ma, G. Yin, X. Xie, and Q. Cui. “Doping Effect on High-Pressure Behaviors of Sc,Y-doped AlN Nanoprisms”. J. Phys. Chem. C 117 (2013), pp. 4304–4308.

73. X. Li, L. Kong, L. Shen, J. Yang, M. Gao, T. Hu, X. Wu, and M. Li. “Synthesis and in situ high pressure Raman spectroscopy study of AlN dendritic crystal”. Mater. Res. Bull. 48 (2013), pp. 3310–3314.

74. Y. Xu, H. Zhu, C. Ma, P. Zhu, R. Cong, X.Wu, W. Gao, and Q. Cui. “Pressure-induced structural phase transition in AlN:Mg and AlN:Co nanowires”. J. Solid State Chem. 202 (2013), pp. 33–37.

75. M. Pravica, N. Bhattacharya, Y. Liu, J. Robinson, W.-S. Au, T. Mizoguchi, Z. Liu, and Y. Xiao. “High pressure infrared and X-ray Raman studies of aluminum nitride”. Phys. Status Solidi B 250 (2013), pp. 726–731.

76. D. P. Dandekar, A. Abbate, and J. Frankel. “Equation of state of aluminum nitride and its shock response”. J. Appl. Phys. 76 (1994), pp. 4077–4085.

77. Z. Rosenberg, N. S. Brar, and S. J. Bless. “Dynamic high-pressure properties of AlN ceramic as determined by flyer plate impact”. J. Appl. Phys. 70 (1991), pp. 167–171.

78. T. Sekine and T. Kobayashi. “Phase transition in ceramics under shock wave compression”. New Diamond Front. Carbon Technol. 13 (2003), pp. 153–160.

79. K. Kondo, S. Soga, E. Rapoport, A. Sawaoka, and M. Araki. “Shock-compaction features and shock-induced chemical reaction in some ceramic powders”. J. Mater. Sci. 21 (1986), pp. 1579–1590.

257

80. K. Keller, T. Schlothauer, M. Schwarz, G. Heide, and E. Kroke. “Shock wave synthesis of aluminium nitride with rocksalt structure”. High Pressure Res. 32 (2012), pp. 23–29.

81. K. Thoma, U. Hornemann, M. Sauer, and E. Schneider. “Shock waves—Phenomenology, experimental, and numerical simulation”. Meteorit. Planet. Sci. 40 (2005), pp. 1283–1298.

82. O. Knotek, M. Böhmer, and T. Leyendecker. “On structure and properties of sputtered Ti and Al based hard compound films”. Vac. Sci. Technol. A 4 (1986), pp. 2695–2700.

83. V. Derflinger, H. Brändle, and H. Zimmermann. “New hard/lubricant coating for dry machining”. Surf. Coat. Technol. 113 (1999), pp. 286–292.

84. S. Veprek and M. Jilek. “Superhard nanocomposite coatings. From basic science toward industrialization”. Pure Appl. Chem 74 (2002), pp. 475–481.

85. P. H. Mayrhofer, C. Mitterer, L. Hultman, and H. Clemens. “Microstructural design of hard coatings”. Prog. Mater. Sci. 51 (2006), pp. 1032–1114.

86. P. H. Mayrhofer, A. Horling, L. Karlsson, J. Sjolen, T. Larsson, C. Mitterer, and L. Hultman. “Self-organized nanostructures in the Ti–Al–N system”. Appl. Phys. Lett. 83 (2003), pp. 2049–2051.

87. D. Rafaja, C. Wüstefeld, C. Baehtz, V. Klemm, M. Dopita, M. Motylenko, C. Michotte, and M. Kathrein. “Effect of Internal Interfaces on Hardness and Thermal Stability of Nanocrystalline Ti0.5Al0.5N Coatings”. Metall. Mater. Trans. A 42 (2011), pp. 559–569.

88. A. Madan, I. W. Kim, S. C. Cheng, P. Yashar, V. P. Dravid, and S. A. Barnett. “Stabilization of Cubic AlN in Epitaxial AlN/TiN Superlattices”. Phys. Rev. Lett. 78 (1997), pp. 1743–1746.

89. C. Stampfl and A. J. Freeman. “Structure and stability of transition metal nitride interfaces from first-principles: AlN/VN, AlN/TiN, and VN/TiN”. Applied Surface Science 258 (2012), pp. 5638–5645.

90. V. Pankov, M. Evstigneev, and R. H. Prince. “Enhanced stability of rocksalt-type AIN phase in AIN/TiN superlattices synthesized by room-temperature pulsed laser deposition”. J. Appl. Phys. 92 (2002), pp. 4255–4260.

91. W. Zhang, X.-R. Chen, L.-C. Cai, and Q.-Q. Gou. “Electronic and Optical Properties of Rock-Salt AlN under High Pressure via First-Principles Analysis”. Comm. Theor. Phys. 50 (2008), pp. 990–994.

92. S. Saib, N. Bouarissa, P. Rodríguez-Hernández, and A. Muñoz. “Structural and dielectric properties of AlN under pressure”. Physica B 403 (2008), pp. 4059–4062.

93. A. Wang, S. Shang, D. Zhao, J. Wang, Li Chen, Y. Du, Zi-K. Liu, T. Xu, and S. Wang. “Structural, phonon and thermodynamic properties of fcc-based metal nitrides from first-principles calculations”. Calphad 37 (2012), pp. 126–131.

94. A. T. Collins, E. C. Lightowlers, and P. J. Dean. “Lattice Vibration Spectra of Aluminum Nitride”. Phys. Rev. 158 (1967), pp. 833–838.

95. Z.-Y. Jiao, S.-H. Ma, and Ji-F. Yang. “A comparison of the electronic and optical properties of zinc-blende, rocksalt and wurtzite AlN: A DFT study”. Solid State Sci. 13 (2011), pp. 331–336.

96. K. Wang and R. R. Reeber. “Thermal expansion of GaN and AIN”. In: MRS Proceedings. Vol. 482. Cambridge Univ Press. 1998, p. 863.

97. S. Krukowski, M. Leszczynski, and S. Porowski. Thermal properties of the group III nitrides. In: Properties, processing and applications of gallium nitride and related semiconductors (EMIS Datareviews Series). Ed. by J. H. Edgar, S. Srite, I. Akasaki, H. Amano, and C. Wetzel. INSPEC, 1999. p. 217.

98. V. I. Koshchenko, Y. K. Grinberg, and A. F. Demidenko. “Thermodynamic Properties of AlN (5-2700K), GaP (5-1500K), and BP (5-800K)”. Inorg. Mater. 20 (1984), p. 1787-1790.

258

99. D. Gerlich, S. L. Dole, and G. A. Slack. “Elastic properties of aluminum nitride”. J. Phys. Chem. Solids 47 (1986), pp. 437–441.

100. C. F. Cline and J. S. Kahn. “Microhardness of Single Crystals of BeO and Other Wurtzite Compounds”. J. Electrochem. Soc. 110 (1963), pp. 773–775.

101. I. Yonenaga, A. Nikolaev, Y. Melnik, and V. Dmitriev. “High-temperature hardness of bulk single-crystal AlN”. In: MRS Proceedings. Vol. 693. Cambridge Univ Press. 2002, p. I10-4.

102. B. D. Fulcher, X. Y. Cui, B. Delley, and C. Stampfl. “Hardness analysis of cubic metal mononitrides from first principles”. Phys. Rev. B 85 (2012), p. 184106.

103. H. Vollstädt and H. Recht. “Verfahren zur Herstellung von kubischem Aluminiumnitrid”. Patent DD000000292903A5. 1991.

104. X.-Q. Chen, H. Niu, D. Li, and Y. Li. “Modeling hardness of polycrystalline materials and bulk metallic glasses”. Intermetallics 19 (2011), pp. 1275–1281.

105. H. Sumiya, Y. Ishida, K. Arimoto, and K. Harano. “Real indentation hardness of nano- polycrystalline cBN synthesized by direct conversion sintering under HPHT”. Diamond Relat. Mater. 48 (2014), pp. 47–51.

106. K. Keller, T. Schlothauer, M. Schwarz, E. Brendler, K. Galonska, G. Heide, and E. Kroke. “Properties of shock-synthesized rocksalt-aluminium nitride”. Processing and Properties of Advanced Ceramics and Composites IV: Ceramic Transactions 234 (2012), pp. 305–311.

259

Influence of multi-pass roll-bonding on the mechanical properties of twin roll cast magnesium sheets

F. Schwarz 1, St. Reichelt 2, L. Krüger1, R. Kawalla 1

1 Institute of Materials Engineering, TU Bergakademie Freiberg2 Institute of Metal Forming, TU Bergakademie Freiberg

260

Introduction

Magnesium and its alloys are attractive con-struction materials for the application in lightweight structures and vehicle applications since a low density is combined with good mechanical strength. Additionally, magne-sium has a very good damping characteris-tic for acoustic and electro-magnetic waves, making it an ideal material for use in the mo-bile electronic devices.

These benefits are quite annihilated due to the hexagonal closed packed (hcp) crystal structure of magnesium. While five inde-pendent slip systems are necessary to afford a major plastic deformation in a material, the hcp-structure only provides two independent basal slip systems at room temperature. In or-der to raise the material ductility and strength, the general approach of grain size refinement is pursued.

Within the ADDE Cluster of Excellence a technology based on the so called Accumu-lative Roll-Bonding (ARB) for magnesium sheets produced by Twin Roll Casting (TRC) [Kaw11] has been investigated. While the TRC process already aims on property en-hancement by utilizing a near net shape pro-duction technology with reduction in pro-cessing time and costs [Kaw11, Ull12], the

Accumulative Roll-Bonding of Mg-Alloys

Principle of Accumulative Roll-Bonding

The basic idea of the ARB process – as it was introduced by Saito et al. [Sai99] – is a cyclic multi-stage roll-bonding process with a pass reduction of 50 % for metallic sheets of the same material. Due to the cyclic roll-bonding and the 50 % pass reduction, a layered sheet is obtained, which macroscopically maintains the initial thickness while the material has

Abstract A cyclic roll-bonding process based on the principle of Accumulative Roll-Bonding (ARB) was used to refine the microstructure of twin roll cast (TRC) AZ31 magnesium sheet. Therefore, a rolling technology was developed, allowing the reproducible roll-bonding of AZ31 sheet up to three roll-bonding cycles with a reduction in thickness of 50 % within each roll-bonding cycle. The influence of processing conditions and initial surface treatment by brushing on the result-ing bonding strength were investigated and altered for an optimized bonding strength for the subsequent material compounds.

Due to the processing by an ARB-like process, a severe refinement in grain size was observed. While one ARB-cycle already leads to a high refinement in grain size with ultra-fine grained ar-eas, further ARB-cycles result in a more homogeneous microstructure with an decreased mean grain size. As a result of the altered microstructure, the mechanical strength of the material is also increased due to the ARB processing with a slight loss in strain to failure for deformation testing at ambient temperatures. In contrast to the testing at room temperature, the formability of the ARBed materials is significantly enhanced while maintaining a comparable strength level with even showing regions of superplastic deformation under tensile loading.

goal of the ARB process is a further benefit of material properties due to grain refinement. Up to now, no research results are available concerning neither ARB of twin roll cast ma-terials nor of extensive ARB of Mg. Previously only small ARB strips of lengths and thick-nesses smaller than 200 mm and 140 mm, respectively, were processed [Cha10], [Val05], [Fat08], [Li11]. The aim of this work is the detailed charac-terization of influencing process parameters on the roll bonding of magnesium alloy AZ31 and the discussion of the achievable micro-structures and mechanical properties.

261

Figure 3: Exemplary ARB sheets of AZ31, roll-bond-ed over three ARB-Steps with WR indicating the rolling direction.

Figure 2: Roll heating system with control unit applied to the two-high rolling mill at the Institute of Metal Forming.

already undergone a severe plastic deforma-tion. Therefore, the ARB-process is classified within the process class of severe plastic de-formation (SPD) along with other forming processes like Equal Channel Angular Press-ing/Extrusion (ECAP/ECAE) or High Pres-sure Torsion (HPT). The schematic principle for the ARB-process is summarized in Figure 1.

The original idea of the ARB process intends a processing at temperatures below the recrys-tallization threshold. Therefore, the strain of each ARB-cycle is accumulated within the material, which leads to a significant grain re-finement - even at temperatures without the occurrence of classic recrystallization pro-cesses. Grain sizes in the region of ultra-fine grained (mean grain diameter below 1 µm) (e.g. [Val06]) or nano grained scale (mean grain diameter below 100 nm) (e.g. [Jam14]) can be achieved with the intention to enhance the materials properties like strength and de-

formation abilities. Due to the poor formabil-ity of magnesium and magnesium alloys, this restriction cannot be fulfilled and therefore the ARB-process has to be altered to higher processing temperatures. Usually, a process-ing temperature in the region of 300 °C and 400 °C is described in literature for an ARB-like processing of Mg-sheet [Cha10, Che06, Fat08, Per04a, Roo10, Wan07, Val05, Zha07]. Nevertheless, even with this derivation from the original proposal and the occurrence of dynamic softening mechanisms, the process is still referred to as an ARB-process for magne-sium in literature. Also ARB-related variants with stacking of three or five sheets per roll-ing pass and a subsequent rolling reduction of 66 % or 80 % for obtainment of the original thickness are presented and often compared with accumulative roll-bonded material in lit-erature [Cha03, Fat08, Per04a, Per05, Val05, Wan07].

Figure 1: Schematic processing procedure for Accumulative Roll-Bonding based on [Sai99].

262

ARB-Technology for TRC Sheets

A key feature for the processing of magne-sium alloy is the management of the tempera-ture regime, since a too low processing tem-perature will lead to failure or fracture due to the poor forming characteristics of magne-sium at low temperaturs discussed previously. For the roll-bonding trials, the contact with the work rolls is a crucial point, since the heat flow from the workpiece to the cooler rolls can result in a serious temperature drop, es-pecially for the surface and surface near areas or even the whole sample for very thin sheets. For a successful realization of an ARB-process for magnesium TRC-sheets, the prevention of such an excessive heat loss is a key aspect. The temperature loss within the rolling gap is mainly influenced by the surface tempera-ture of the work rolls as well as the contact time, which is preset by the rolling speed. As the desired bonding during the roll-bonding process takes certain time for formation, due to this an increase of rolling speed within the capabilities of the used rolling mill is not a solution. Therefore, higher work rolls tem-perature can help to reduce the temperature loss, since a reduction of the temperature gra-dient between workpiece and work rolls also reduces the heat loss due to heat conduction. For the realization of this approach, a roll heating device was installed at the experi-mental two high rolling mill at the Institute of Metal Forming (IMF) at the beginning of the first ADDE research period. The heating de-vice applied to the work rolls and the control device is shown in Figure 2. With the help of the heating device, a pre-heating temperature of 150 °C for the steel rolls surfaces’ is estab-lished while a proper pre-heating time ensure a good through-thickness heating of the rolls, resulting in a time window of 20-30 minutes for rolling trials with nearly constant surface temperature of the rolls. That allows a specific and successful roll bonding for larger series of roll-bonding trials for magnesium strips (Figure 3).

A second crucial point is a proper lubrication for the hot deformation of magnesium, since magnesium sheets tends to show an adhesive behavior to the work rolls. Since this adhesion is already a problem for normal hot rolling of magnesium alloys with respect to a proper surface quality, the problem accumulates for a magnesium roll-bonding process. When adhesive sticking to the upper and lower roll appears, the bonding strength of the material compound can be reduced due to the tensile stress or even fail in the worst case. At the same time, a sufficient lubricant has to fulfill the other demands like reduction of friction and therefore processing force and formation of a suitable, clean surface. A clean surface is important for the ARB of Mg material, since the rolled surface is the basis for the formation of a new bonding within the next ARB-cycle.

A lubricant usually used for hot rolling of alu-minum alloys was applied for the roll-bonding trials as a starting point for the investigation of an ARB-based processing of magnesium TRC sheets. This lubricant inhibits the ad-hesion, but failed to provide a suitable, clean surface or the roll-bonded compound. Espe-cially for a higher processing temperature of 400 °C a dark layer was found on the surface (see Figure 4), which was difficult to remove prior to the next roll-bonding step. Within the following trials two different lubricants were found to fulfill the given tasks: 1. Beruforge 150D produced by Bechem and 2. Concen-trate 3048 produced by Mobil. Both result in a clean surface while preventing adhesion of the magnesium sheets to the work rolls in combi-nation with a reduction of friction. Since both lubricants are used as an emulsion with water, an optimal pre-heating temperature of 120 – 130 °C for the work rolls was determined. A higher pre-heating temperature provides a slight decrease in heat loss during contact but also the occurrence of a higher evaporation of the lubricant. The evaporation reduces the ad-hesion prevention to the rolls and the lubrica-tion, so that a higher pre-heating temperature is not beneficial at all.

Figure 1: Schematic processing procedure for Accumulative Roll-Bonding based on [Sai99].

263

Enhancement of Bonding Strength

The key issue for every roll-bonding process is the generation of a sufficient bonding strength for the material compound, suitable to ensure further handling, manufacturing and testing without failure of the bonding interfaces.

Many papers on the ARB-processing of mag-nesium alloys have been published over the recent years. The main focus of these papers has been the influence of the ARB-process on the material properties and resulting micro-structure. Any information concerning the bonding strength is given neither in terms of an applied surface preparation nor on the overall bonding strength after the roll-bond-ing trials. Based on this lack of information, basic investigations regarding the bonding strength during an ARB-processing for TRC magnesium sheets were carried out. The re-search is based on long-term experiences of the IMF on roll-bonding, especially cold

roll-bonding, of metallic materials. Accord-ing to the research know-how, the resulting bonding strength is mainly dependent on the surface preparation prior to the roll-bonding as well as the utilized processing parameters.

According to the experience of the authors, a surface preparation by degreasing with ac-etone followed by a wire brushing process is the most successful preparation in terms of bonding strength. Therefore, a prototype brushing machine has been designed at the IMF, allowing a reproducible and specifical-ly modifiable brushing process. The feeding velocity, the rotational speed as well as the work force, pressing the sample surface onto the brush, can be individually adjusted within this device. For the quantification of the re-sulting bonding strength, two different test setups were adopted: the tensile shear test and the Chalmer’s test.

Figure 5: Schematic view of the tensile shear sample.

Figure 4: ARB roll-bonded sample with 400 °C initial temperature and 120 °C roll heating temperature after using the initial lubricant for aluminum hot forming.

264

Figure 6: Schematic view of Chalmer’s sample.

The tensile shear test uses standard tensile sample, which is notched from the two op-posite sheet surfaces of the compounds (see Figure 5). The depth of the notches is chosen to be slightly higher than the layer thickness on each side, so that a rectangular area of the bonding surface is cut free. If the tensile shear sample is subjected to a standard room tem-perature tensile test, this area is loaded under shear stress that will result in a failure of the tested bonding area, after reaching a certain level.

For the rotational Chalmer’s test sample (see Figure 6), a ring shaped area of the bonding layer is cut free. Under compressive loading, the bonding area of the Chalmer’s sample is subjected to a tensile stress.

Both sample types are used to investigate the bonding strength under two different types of external load. The bonding strength is quantified by measuring the maximum force Fmax during the test, which is necessary for a failure of the bonding area. The maxi-mum force is related to the tested bonding area Abond, which results in a quantitative value for the shear bonding strength τs ob-tained by the tensile shear test of the ten-sile bonding strength and σc obtained by the Chalmer’s testing method, respectively: (1)

(2)

By relating of these nominal bonding strength values to the initial flow stress kf of the weakest

material being part of the bonding, the specif-ic shear bonding strength τ*s and the specif-ic tensile bonding strength σc

* are calculated. (3)

(4)

The idea behind this is, that the relevant bond-ing strength cannot be higher than the plas-tic flow stress of the weakest material within the bonding, since a failing mechanism will then divert from the bonding interface into the bulk material, resulting also in a failure of the compound. Within certain limit, this also allows the comparison of different bonding strengths, even with different materials being involved.

Influence of processing parameters

For the investigation of the influence of the processing conditions, an AZ31-AZ31 mag-nesium compound from TRC sheets with an initial thickness of 4.6 mm and 100 mm in width was roll-bonded at different roll-ing speeds, rolling temperatures and height reductions. The surface preparation for all samples was a degreasing with acetone and subsequent wire brushing. A feeding speed of 5 mm/s and a work force of 30 N were used for wire brushing with an alternation of the rotational speed of 500 RPM for the lower and 2000 RPM for the upper sheet. After-wards tensile shear samples were cut out of the roll-bonded compounds and tested at room temperature for determination of the nominal shear bonding strength . Tensile tests at room temperature were also carried out for

265

Figure 7: Influence of initial rolling temperature for a rolling speed of 0.2 m/s (left) and rolling speed for a rolling tem-perature of 300 °C (right) on the resulting specifc shear bonding strength for roll-bonding of an AZ31-AZ31 material compound [Rei12].

the determination of the Rp0.2 yield strength for the roll-bonded material. Rp0.2 was used as kf for the determination of the specific shear bonding strength τs*.

The summarized results of the tensile shear tests in Figure 7 show the influence of dif-ferent initial rolling temperatures and differ-ent rolling speeds on the resulting bonding strength over the rolling reduction.

The reachable bonding strength is promoted by an increase in the rolling reduction per pass. Since the rolling reduction is equivalent to a creation of new, bondable surface area, the bonding strength has to increase due to the increase of bonding surface area with in-creasing deformation. At the same time, the stress within the roll gap is increased with increasing deformation, promoting a better bond formation. After a critical, minimum deformation the resulting bonding strength approaches a saturation level, where a fur-ther increase in deformation leads to just very slight increases in bonding strength.

The bond formation benefits and the bond-ing strength increases by higher processing temperatures as well as higher rolling speeds because of temperature rises in the bonding area. Both for the rolling temperature as well as for the rolling speed, a change from the lowest value of 200 °C up to 300 °C and from 0.1 m/s up to 0.2 m/s, respectively, results

in the highest increase in bonding strength, while a further enhancement of both process-ing parameters leads to slight enhancement in bonding strength only. For the investigat-ed range of rolling speed, this is due to only a marginal increase of the finishing tempera-ture after roll-bonding while at the same time the contact time and therefore the time for bond formation is significantly reduced for an increase of rolling speeds to higher values of 0.5 m/s and above. For the rolling tempera-ture, the change of plastic flow stress, and therefore the change of pressure distribution within the rolling gap, is much smaller for the temperature step from 300 °C to 400 °C than for the change from 200 °C. As a result, a me-dium processing speed of 0.2 m/s to 0.5 m/s combined with a medium temperature of 300 °C gives a quite optimal processing win-dow, since a further increase in both param-eters will not be benefitted by a considerably increase in bonding strength.

Influence of surface preparation

In addition to the processing parameters, the surface of the participant materials controls the bond formation. Therefore, a parameter study on the influence of brushing parameters for TRC AZ31 sheets was conducted at the prototype brushing machine of the IMF. An industrial, high strength steel wire brush with 0.15 mm wire diameter was used. The widths of the brushing samples were 100 mm. For

266

Figure 8: Development of roughness value Rq over forward feeding speed for different rotational speeds and working forces for the brushing of AZ31 TRC sheet surfaces [Rei13].

the evaluation of the brushing, the roughness of the surface treated sheets was measured perpendicular to the brushing direction. For measurement of the roughness a HOMMEL Tester T1000 skid-type roughness testing device was used. The root-mean squared roughness Rq was chosen as a representative parameter for the overall surface roughness. Figure 8 shows that a wide variety of different roughness values can be obtained, depending on the brushing conditions.

As a general trend, the roughness increased with an increase of work force, since this leads to an increased contact pressure between the material surface and the brush wires, work-ing the brush more deeply into the surface. At the same time, an increasing rotational speed results in a decrease of the overall mea-sured roughness. This is due to the stiffness

of the brush and the brush wires is gained with increasing rotational speed, decreasing the bending of the single brush wires, letting the brush surface appear more homogeneous at a macroscopic point of view. Additionally, the increased rotational speed also increas-es the impact of the brush onto the surface area, shifting the overall brushing process to a kind of a micro-machining of the surface. The forward feeding speed as third parame-ters results in a decreased roughness with in-creased speed, since a higher velocity results in a shorter amount of time for the brush to work onto a certain surface area. Therefore, a high feeding speed combined with a high ro-tational speed and small work force leads to a small roughness, while the opposite combina-tion of a slow rotational speed in combination with a slow feeding speed and high work force results in a significant increase of resulting surface roughness.

267

Based on these results of the parameter study, a set of brushing parameters was chosen, to in-vestigate the roughness ratio Χ of the bonding partners on the resulting bonding strength. The roughness ratio describes the propor-tion of the two roughness values between the bonding partners. The chosen parameters are summarized in Table 1.

Roll-bonding samples of AZ31-AZ31 TRC sheet material were prepared with the varia-tions of Table 1. The sheets were roll-bonded at a rolling speed of 0.5 m/s and an initial rolling temperature of 300 °C with different height reductions in the rolling pass. Afterwards, samples for tensile shear test and Chalmer’s

Figure 9: Bonding strength results σc* (left) and (right) for hot roll-bonding of AZ31-AZ31 compounds with different

roughness ratio Χ [Rei13].

test were manufactured out of the received material compounds, which were then tested at room temperature to obtain values for the nominal bonding strength. With addition-al room temperature tensile testing the true flow stress kf was determined, for the obtain-ment of the specific shear bonding strength τ*s and specific tensile bonding strength σc

*. The bonding strength was determined by ten-sile shear testing and the Chalmer’s test setup, with the bonding strength values being relat-ed to the initial flow stress to obtain specific bonding strength values for shear and tensile loading of the bonding area. The results of the bonding strength determination are summa-rized in Figure 9.

combination of Brushing parameters Rq [µm] χ [-]

50 N / 500 RPM / 2.7mm/s | 10 N / 500 RPM / 26.7 mm/s 5.95 0.65 9.15

50 N / 500 RPM / 2.7 mm/s | 50 N / 2000 RPM / 26.7 mm/s 5.95 1.99 5.95

10 N / 1000 RPM / 7.4mm/s | 10 N / 1500 RPM / 26.7 mm/s 2.15 1.04 2.07

50 N / 500 RPM / 2.7 mm/s | 50 N / 500 RPM / 2.7 mm/s 5.95 5.95 1.00

50 N / 2000 RPM / 26.7 mm/s | 50 N / 2000 RPM / 26.7 mm/s 1.00 1.00 1.00

Table 1: Chosen combination of brushing parameters for creation of different roughness ratios [Rei13].

268

Microstructure of AZ31 after Accumulative Roll-Bonding

For the first time this work investigates the ARB of twin roll cast magnesium sheets. The initial twin roll cast material is characterized by a heterogeneous microstructure. Light op-tical analyses of the initial TRC sheets before surface preparation and ARB shows a fine, dendritic microstructure (Figure 10). The dendrites grew along the heat transfer direc-tion and are angled to the normal direction due to the relative movement between the cast and the rolls. Furthermore, segregations occur in the middle of the sheet. The reason for cen-terline segregations is the concentration of the last solidified cast with alloying elements and potential impurities. Electron backscattered diffraction (EBSD) analysis reveals a high defect density in the TRC material which is identified by means of misorientation changes within the grains (Figure 10).

Since a heating up to 300 °C for 30 minutes for the direct ARB processing of the TRC-state material does not remarkably influence the initial TRC-state, an adequate heat treatment can lead to a homogenization of the TRC ma-terial [Sch12]. In this work, a heat treatment at 400 °C for 24 h was carried out that pro-vokes a recrystallized microstructure with an average grain size of 22 µm (Figure 11).

Figure 10: TRC AZ31 consists of a fine, dendritic microstructure (left). EBSD analysis detected a high amount of crystal defects (right) due to the high amount of misorientation changes.

For the Chalmer’s testing, all wire brushed samples show a significant increase in bond-ing strength above a non-wire brushed sam-ple, showing the beneficial effect of a proper surface preparation. Within the prepared samples, a roughness ratio of 2.07 shows the best bonding results. For the tensile shear test-ing, the results are shifted, so that the highest roughness ratios or roughness values involved are showing the highest bonding strength values. In comparison, the tensile shear test in general shows a higher specific bonding strength than the Chalmer’s test. This differ-ence can be explained due to the different loading type within these two tests. For the tensile shear test, the mechanical interlock-ing of the roughness peaks is also measured additionally to the originally generated bond-ing strength while for the Chalmer’s test this interlocking effect is not significantly pres-ent. This also explains the increase of bond-ing strength with increasing roughness for the tensile shear strength, since an increase in roughness will result in an increase of the interlocking effect. Due to that reason, it be-comes clear, that a comparison of bonding strength values is a quite difficult and in terms of low bonding strength the Chalmer’s testing is a more suitable testing method compared to the tensile shear testing, at least for the pa-rameter range investigated within this work.

269

One important quality factor of ARB-material is an excellent bonding. Light optical micro-graphs confirm the good bonding strength values. As can be seen exemplarily in Figure 12, a good bonding based on solid-solid weld-ing and mechanical interlocking is achieved and the bonding surface can hardly be identi-fied within the optical microsection.

The most significant microstructural chang-es occur during the first ARB cycle. Figure 12 shows the microstructure of TRC after the first ARB-cycle. The grain size is already significantly reduced through the first ARB-

Figure 12: TRC AZ31 after one ARB cycle shows a) good bonding between the two sheets and b) no completely recrys-tallized microstructure evolves, while remains of initial segregations can still be identified.

cycle. As discussed in section 2.1, the pro-cess temperature for successful roll bonding of AZ31 is 300 °C. In comparison to conven-tional ARB, the microstructure refinement is therefore based on recrystallization processes. The ARB process is not adequate to homoge-nize the TRC microstructure – neither up to three ARB-cycles. Therefore, a quantitative specification of grain refinement is not car-ried out. Besides a few large, elongated grains (> 200 µm), the accumulative roll bonded TRC magnesium alloy contains fine grained bands, including grains smaller than 1 µm.

Figure 11: Recrystallized microstructure of TRC+HT AZ31: optical microsection (left) and EBSD-image (rate of indexing: > 90 %) (right).

270

An extensive grain refinement was also ob-served for the heat treated TRC AZ31 (Fig-ure 13). The average grain size decreases from 22 µm down to 4 µm. With increasing num-ber of ARB-cycles, the average grain size stays nearly constant after the second and the third cycle with 4 µm and 3 µm, respectively. With increasing number of ARB cycles a more ho-mogeneous microstructure occurs [Sch13].

The ARB at 300 °C provokes the activation of <c+a> dislocationship additional to twinning. That results in the evolution of a strong basal texture in both material states (Figure 14). In materials having a basal texture the c-axis of

Figure 13: Microsection of AZ31 TRC+HT after two ARB-cycles at 300 °C.

Figure 14: Evolution of basal texture through ARB for a) TRC and b) TRC+HT AZ31. The 0002-pole figures corre-spond to initial state and one, two and three ARB-cycles, respectively, from left to right.

the crystallites are nearly perpendicular to the rolling direction. In both material states (TRC as well as TRC+HT) the maximum intensities of the 0002 pole figure rise with increasing number of ARB cycles that show that the bas-al texture becomes more pronounced. Mostly magnesium sheets have no perfect basal tex-ture with only one intensity maximum but often a small tilt of the c-axis in rolling direc-tion is detected. The tilt of the c-axis becomes apparent in the two occurring maxima in the 0002 pole figure. An enhanced basal texture was also observed in [Kaw08] after warm roll-ing at 300 °C of TRC AZ31.

271

Figure 15: Flow behavior of TRC+HT AZ31 a) in dependence on ARB cycles and b) in dependence of temperature (compression axis in rolling direction, strain rate of 10-3 s-1).

Figure 16: Microstructure of TRC+HT AZ31 at compressive strain of 2 % (RT, 10-3 s-1).

Figure 17: Microstructure of TRC+HT AZ31 after two ARB cycles at compressive strain of 5 % (RT, 10-3 s-1).

Mechanical properties of AZ31 after Accumulative Roll-Bonding

Sigmoidal flow curves are observed for all investigated material conditions under com-pressive loading at room temperature. Figure 15 a) presents the flow behavior of TRC+HT AZ31 at room temperature. The highest in-crease in flow stress is achieved after one ARB cycle. Whereby, the rise in strength is based on the Hall-Petch effect and the evolving Bas-al texture. Further ARB cycles only lead to a slight further increase in yield stress and ten-sile strength.

The deformation behavior of magnesium is complex due to its hexagonal close packed crystal structure. Due to the limited number of slip systems, the mechanical behavior of AZ31 magnesium alloy is mainly dependent on twinning and basal glide. Frequently, 10-12<10-11> tensile twins, 10-11<10-12> compression twins and 10-11-10-12 sec-ondary twins occur in magnesium. Partic-ularly 10-12<10-11> tensile twins are im-portant under compressive loading at room temperature and cause the sigmoidal flow curve [Bar07], [Kli02], [Kor10], [Jia07].

272

As can be seen in Figure 16 the initial TRC+HT as well as the roll bonded material after two ARB cycles (Figure 17) show signif-icant twinning, which leads to the consider-able strain hardening. The strain hardening rates and flow stresses of the ARB materials are higher than for the initial TRC+HT ma-terial. The reason can be seen in rising acti-vation stresses for mechanical twinning with decreasing grain sizes [Mey01]. Also in AZ magnesium alloys processed through ECAP a reduced twinning was observed due to rising activation stresses with decreasing grain size [Kru11].

High temperature deformation behavior of magnesium is often reported (e.g. [Kim08], [Wu10], [Jae04], [Pra08], [Yin05], [Jai07]), but up to now, there are no information about the mechanical behavior of ARB magnesium alloys at elevated temperatures. Figure 15 b) shows the influence of temperature on the flow behavior of the initial TRC+HT and after three ARB cycles at quasi-static strain rate. As presented the flow behavior at room tempera-ture is mainly influenced by twinning. Twin-ning is an athermal process, which means that the activation stress for twinning is relatively unsusceptible of temperature. Hence, at high-

er temperatures thermally activated disloca-tion slip becomes more important. Especially the critical resolved shear stress (CRSS) of prismatic and pyramidal slip increases with increasing temperature [Bar03], [Wat09], [AlS08]. Therefore, the amount of mobile dis-locations increases which supports the cross slip and dislocation climb. The involvement of dynamic recovery and recrystallization processes can be seen at the flow curves at temperatures higher than 150 °C that achieve a steady state stress. Due to the activation of additional deformation systems (e.g. <c+a> slip) the flow stresses decrease and the de-formability rises with increasing temperature. The coarse grained initial TRC+HT AZ31 which was deformed at 150 °C shows still a significant strain hardening. In accordance with room temperature observations, twins can be easier activated in the initial TRC+HT material than in the ARB materials. The fine grained 3 ARB AZ31 softens at lower strains. Twinning plays a minor role as can be seen in Figure 18. Figure 18 shows that there is no strong twinning at compressive strain of 3 %. Furthermore, the reduced grain size after ARB leads also to grain boundary sliding and consequently improved formability.

Figure 18: EBSD picture of the microstructure of TRC+HT AZ31 after three ARB cycles at compres-sive strain of 3 % (RT, 10-3 s-1).

Figure 19: 0.2 % yield stress and maximum strain in de-pendence of temperature for the initial materials and af-ter three ARB cycles.

273

Summary

The following conclusions can be drawn on the basis of the above studies:

1. By applying adequate Accumulative Roll-Bonding parameters successful bonding of Twin Roll Cast AZ31 magne-sium alloy was realized up to three cycles. Surface preparation has a significant im-pact on resulting bonding strength.

2. Temperature management is crucial to decrease edge cracking and other materi-al failure during a multi-cycle roll-bond-ing process for magnesium alloys.

3. Twin Roll Casting in combination with a subsequent heat treatment leads to a fine and homogeneous microstructure after Accumulative Roll-Bonding. The application of Accumulative Roll-Bond-ing on initial Twin Roll Cast AZ31 is not sufficient to dissolve the microstructural heterogeneities.

4. The material strength at room tempera-ture rises after roll bonding. At elevated temperatures the maximum formability is significantly improved. Under ten-sile loading superplastic strains were achieved in the Accumulative Roll-Bond-ing materials for testing at 250 °C.

Under tensile loading at room temperature the material strength rises also after ARB for both investigated material conditions. Fig-ure 19 presents the decrease of 0.2 % yield stress with increasing temperature for TRC and TRC+HT. After three ARB cycles (TRC AZ31) the yield stress drops from 307 MPa down to 136 MPa and 28 MPa at RT, 150 °C and 250 °C, respectively. At room tempera-ture the strain at failure is almost unaffected by ARB. Elevated temperatures give rise to strain at failure. In particular the ARB materi-als show a superplastic deformation behavior, e.g. after three ARB cycles the strain at failure was 18 % at room temperature and achieved 300 % at 250 °C.

Acknowledgements

The authors would like to acknowledge the support of Dr. Armin Franke for sample preparation, Gerhard Schreiber for XRD mea-surements and Christiane Ullrich for helpful discussions. This work was performed within the Cluster of Excellence “Structure Design of Novel High-Performance Materials via Atom-ic Design and Defect Engineering (ADDE)” that is financially supported by the European Union and by the Ministry of Science and Art of Saxony.

274

References

[AlS08] T. Al-Samman, G. Gottstein: Dynamic recrystallization during high tem perature deformation of magnesium, Materials Science and Engineering: A, 490 (2008) 411.[Bar03] M. R. Barnett: A Taylor model based description of the proof stress of mag nesium AZ31 during hot working, Metallurgical and Materials Transactions A, 34 (2003) 1799.[Bar07] M. R. Barnett: Twinning and the ductility of magnesium alloys, Part I: Ten sion twins, Materials Science and Engineering: A, 464 (2007) 1.[Cha10] H. Chang, M. Y. Zheng, K. Wu, W. M. Gan, L. B. Tong, H.-G. Brokmeier: Microstructure and mechanical properties of accumulative roll bonded (ARBed) pure magnesium sheet, Materials Science and Engineering A, 527 (2010) 7176.[Che06] M. C. Chen, H. C. Hsieh, W. Wu: The evolution of microstructures and mechanical properties during accumulative roll bonding of Al/Mg c omposite, Journal of Alloys and Compounds, 416 (2006) 169.[Fat08] S. M. Fatemi-Varzaneh, A. Zarei-Hanzaki, M. Haghshenas: Accumulative roll bonding of AZ31 magnesium alloy, International Journal of Modern Physics B, 22, 18-19 (2008) 2833.[Jae04] A. Jäger, P. Lukác, V. Gärtnerová, J. Bohlen, K. U. Kainer: Tensile properties of hot rolled AZ31 Mg alloy sheets at elevated temperatures, Journal of Alloys and Compounds, 378 (2004) 184.[Jai07] A. Jain, S. R. Agnew: Modeling the temperature dependent effect of twinning on the behavior of magnesium alloy AZ31B sheet, Materials Science and Engineering A, 462 (2007) 29.[Jia07] L. Jiang, J. J. Jonas, A. A. Luo, A. K. Sachdev, S. Godet: Influence of 10-12-extension twinning on the flow behavior of AZ31 Mg alloy, Materials Science and Engineering: A, 445 (2007) 302.[Jam14] R. Jamaati, M. R. Toroghinejad: Effect of alloy composition, stacking fault energy, second phase particles, initial thickness, and measurement position on deformation texture development of nanostructured FCC materials fabricated via accumulative roll bonding process, Materials Science and Engineering A, 598 (2014) 77.[Kaw08] R. Kawalla, M. Oswald, C. Schmidt, M. Ullmann, H.-P. Vogt, N. D. Coung: Development of a strip-rolling technology for Mg alloys based on the twin-roll-cast process, In: Proceedings TMS 2008, 137th Annual Meeting & Exhibition, New Orleans, USA (2008) 177.[Kaw11] R. Kawalla, M. Ullmann, Ch. Schmidt, J. Dembinska, H.-P. Vogt: Properties of magnesium strips produced by twin-roll casting and hot rolling, Materials Science Forum, 690 (2011) 21.[Kim08] W. J. Kim, J. D. Park, U. S. Yoon: Superplasticity and superplastic forming of Mg-Al-Zn alloy sheets fabricated by strip casting method, Journal of Alloys and Compounds, 464 (2008) 197.[Kli02] P. Klimanek, A. Pötzsch: Microstructure evolution under compressive plastic deformation of magnesium at different temperatures and strain rates, Materials Science and Engineering: A, 324 (2002) 145.[Kor10] R. Korla, A. H. Chokshi: Strain-rate sensitivity and microstructural evolution in a Mg-Al-Zn alloy, Scripta Materialia, 63 (2010) 913.

275

[Kru11] L. Krüger, F. Schwarz, U. Martin, H.-J. Roven: Strain rate effects on the flow behaviour and microstructure of ECAPed Magnesium-alloys, Materials Science Forum, 690 (2011) 323.[Li11] X. Li, T. Al-Samman, G. Gottstein: Microstructure development and texture evolution of ME20 sheets processed by accumulative roll bonding, Materials Letters, 65 (2011) 1907.[Mey01] M. A. Meyers, O. Vöhringer, V. A. Lubarda: The onset of twinning in metals: a constitutive description, Acta Materialia, 49 (2001) 4025.[Per04] M. T. Pérez-Prado, J. A. del Valle, J. M. Contreras, O. A. Ruano: Microstructural evolution during large strain hot rolling of an AM60 Mg alloy, Scripta Materialia, 50 (2004) 661.[Per04a] M. T. Pérez-Prado, J. A. del Valle, O. A. Ruano: Grain refinement of Mg-Al- Zn alloys via accumulative roll bonding, Scripta Materialia, 51 (2004) 1093.[Per05] M. T. Pérez-Prado, J. A. del Valle, O. A. Ruano: Achieving high strength in commercial Mg cast alloys through large strain rolling, Materials Letters, 59 (2005) 3299.[Pra08] Y. V. R. K. Prasad, K. P. Rao: Processing maps for hot deformation of rolled AZ31 magnesium alloy plate: Anisotropy of hot workability, Materials Sci ence and Engineering A, 487 (2008) 316.[Rei12] St. Reichelt, M. Schmidtchen, R. Kawalla: The problem of proper surface treatment and various influences on resulting bonding strength for the Accumulative Roll-Bonding of Twin Roll Cast AZ31 Mg-sheet, steel research int. (83), Special Edition proc.: Proceedings of the 14th Internation- al Conference on Metal Forming, Krakow (POL) (2012) 867.[Rei13] t. Reichelt, H. Saleh, M. Schmidtchen, R. Kawalla: On the Bonding Strength of Mg-Mg and Mg-Al Material Compounds, Proceedings of the Internation al Conference on Processing & Manufacturing of Advanced Materials THERMEC 2013, Las Vegas (USA) (2013) 455.[Roo10] A. A. Roostaei, A. Zarei-Hanzaki, M. H. Parsa, S. M. Fatemi-Varzaneh: An analysis to plastic deformation behavior of AZ31 alloys during accumulative roll bonding process, Journal of Materials Science, 45 (2010) 4494.[Sai99] Y. Saito, H. Utsunomiya, N. Tsuji, T. Sakai: Novel ultra-high straining process for bulk materials – development of the Accumulative Roll-Bonding (ARB) process, Acta Materialia, 47, 2 (1999) 597.[Sch12] F. Schwarz, K. Lange, L. Krüger: Material behaviour of twin-roll cast AZ31-magnsium sheets over a wide range of strain rates. In: W. J. Poole, K. U. Kainer: Proceedings of the 9th International Conference on Magnesium and their Applications, MG 2012, 8-12 July 2012 Vancouver/Canada, (2012) 587.[Sch13] F. Schwarz, K. Lange, L. Krüger, R. Kawalla, St. Reichelt: Microstructural and mechanical characterization of ARB AZ31, Materials Science Forum, 765 (2013) 403.[Ull12] M. Ullmann, M. Oswald, S. Gorelova, R. Kawalla, H.-P. Vogt: Strip Rolling Technology of Magnesium Alloys. In: 14th International Conference on Metal Forming - Metal Forming 2012 (Krakow/Poland, 2012) p. 855.[Val03] J. A. del Valle, M. T. Pérez-Prado, O. A. Ruano: Texture evolution during large-strain hot rolling of the Mg AZ61 alloy, Materials Science and Engineering A, 355 (2003) 68.

276

[Val05] J. A. del Valle, M. T. Pérez-Prado, O. A. Ruano: Accumulative roll bonding of a Mg-based AZ61 alloy, Materials Science and Engineering A, 410-411 (2005) 353.[Val06] R. Z. Valiev, Y. Estrin, Z. Horita, T. G. Langdon, M. J. Zehetbauer, Y. T. Zhu: Producing bulk ultrafine-grained materials by severe plastic deformation, JOM, 58, 4 (2006) 33.[Wan07] Q. F. Wang, X. P. Xiao, J. Hu, W. W. Hu, X. Q. Zhao, S. J. Zhao: An ultrafine-grained AZ31 magnesium alloy sheet with enhanced superplasticity prepared by accumulative roll bonding, Journal of Iron and Steel Research International, 14, 5 (2007) 167.[Wu10] H. Wu, W. Hsu: Tensile flow behavior of fine-grained AZ31B magnesium alloy thin sheet at elevated temperatures, Journal of Alloys and Compounds 493 (2010) 590.[Wat09] H. Watanabe, K. Ishikawa: Effect of texture on high temperature deformation behavior at high strain rates in a Mg-3Al-1Zn alloy, Materials Science and Engineering: A, 523 (2009) 304.[Yin05] D. L. Yin, K. F. Zhang, G. F. Wang, W. B. Han: Warm deformation behavior of hot-rolled AZ31 Mg alloy, Materials Science and Engineering A 392 (2005) 320.[Zha07] M. Y. Zhan, Y. Y. Li, W. P. Chen, W. D. Chen: Microstructure and mechanical properties of Mg-Al-Zn alloy sheets severely deformed by accumulative roll-bonding, Journal of Materials Science, 42 (2007) 9256.

277

Mg-Al composite wiresE. Knauer 1, 2, J. Freudenberger 1-3, A. Kauffmann 1, 2 and L. Schultz 1, 2

1 IFW Dresden, Helmholtzstr. 20, 01069 Dresden, Germany2 TU Dresden, Institute of Materials Science, 01062 Dresden, Germany3 TU Bergakademie Freiberg, Institute of Materials Science, Gustav-Zeuner-Str. 5, 09599 Freiberg, Germany

278

AbstractAlthough the workability of magnesium is negligible under the conditions of rotary swaging, co-deformation of magnesium within a tube is possible. This process can be operated success-fully up to a logarithmic deformation strain of at least 8, when utilizing an AA6082 tube. The microstructure shows a decreasing grain size with increasing deformation strain, saturating at a grain size well below 5 µm. In addition, a preferential texture with a [1010] ring fibre com-ponent is established during deformation. The evolution of the microstructure determines the mechanical properties of the composite, which is characterised by an ultimate tensile strength of 240 MPa. This corresponds to a specific strength of 104 MPa/(g/cm³).

IntroductionLight-weight materials represent one aspect of Light-weight materials represent one aspect of reducing the use of energy. This is a crucial is-sue for economic and environmental reasons. In consequence, light-weight materials can improve the efficiency of energy conversion, as e.g. in the case of mobile applications it is possible to reduce the energy consumption using structural optimized systems or using materials with outstanding mechanical prop-erties being normalised to the mass density. Classically, the light-weight materials for con-structive applications are based on Ti, Al and Mg. However, the use of Mg and its alloys is limited to cast alloys and a very small number of wrought alloys, as magnesium shows a very low workability at room temperature origi-nating from the insufficient number of active slip and twinning systems. Although there are Mg-based wrought alloys, their workability is comparably low. While the magnesium alloy AZ31 can be only deformed up to logarith-mic strain of 0.2 by rotary swaging, co-de-formation of a composite with approximately 50 vol% is possible. Indeed, it has been shown that the workability of AZ31 is significantly enhanced when being co-deformed in a tita-nium (grade 1) tube. Such a composite allows being cold worked by rotary swaging up to a logarithmic deformation strain of 2.98 [1]. These promising results originating from an Mg-wrought alloy were encouraging and the question was raised if this would also be pos-sible for a composite containing pure magne-sium. With respect to the first results, the Ti tube should be replaced for various reasons. First of all, titanium tubes are hard to obtain

with a constant wall thickness. Secondly, only titanium with the highest purity is suitable for being cold worked to high deformation strains. This is very sensitive and machining already influences the purity in a reasonable manner in the sense that also the workabil-ity is affected. Finally, the total mass density of the composite can additionally be reduced when an aluminium alloy is used. This arti-cle reports on a composite wire composed from a pure magnesium core and an alumin-ium alloy AlSiMgMn EN-AW 6082 tube. In addition, results for multi-filamentary wires as generated from the accumulative swaging and bundling technique [2-3] are reported. When metallic materials are subjected to large deformation strains the microstructure becomes ultra-fine grained or even nanome-tre-scaled [4-7]. When the grain size is within a few nanometres the dominant deformation mechanism changes from dislocation slip or mechanical twinning to grain-boundary slid-ing and grain rotation [8, 9]. Especially for the case of magnesium with its reduced formabil-ity in coarse grained materials, the enhanced plasticity observed for nanometre-sized ma-terials seems challenging. In consequence, there are numerous studies dealing with the plasticity of magnesium [10-12]. The obser-vation of outstanding mechanical properties are commonly linked to the formation of an ultra-fine-grained microstructure which orig-inates from severe plastic deformation (SPD).

279

Common SPD techniques are equal-channel angular pressing (ECAP) [7, 13-15], high- pressure torsion (HPT) [16,17] and accumu-lative rolling and bonding (ARB) [18, 19], which have been applied to magnesium. The present study flattens the path to obtain an ul-tra-fine grained microstructure with an alter-native method, i.e. accumulated swaging and bundling (ASB) [2]. This technique is bene-ficial as it generates a semi-finished product and is in principle not limited in the size of the wire. In addition the deformation of com-posites is homogeneous.

Experimental details

The initial composite was built from a Mg rod with a diameter of 20 mm, which was put into an AlSiMgMn EN-AW 6082 tube. This tube had an outer diameter of 24 mm and a wall thickness of 2 mm. Rotary swaging is used for deformation of this composite with an area reduction per step of 20 %, each. After this composite was deformed to a diameter of 2.8 mm, which corresponds to a logarith-mic deformation strain of 4.3, the wire was cut into 37 pieces. These wires were closed packed with hexagonal symmetry perpen-dicular to the wire axis and put into a further AlSiMgMn EN-AW 6082 tube (same di-mensions as before). This composite was de-formed to a final diameter of 2.8 mm. In prin-ciple this process of co-deformation, cutting, bundling and further co-deformation can be repeated arbitrarily. The microstructure of the composite wires was investigated by means

of scanning electron microscopy (SEM), FEI Helios 600i operating at 20 kV and 11 nA, equipped with an EDAX DigiView system electron backscatter diffraction (EBSD) unit. For metallographic sample preparation, the samples were cut in longitudinal direction and grinded up to 2500 SiC paper using water as lubricant. The last grinding step was done with 4000 SiC paper and Ethanol as lubricant. Finally, samples were polished using a lubri-cant containing 50 nm colloidal SiO2 particles and a solution from H2O, H2O2 and NH4. In addition, tensile tests were performed at room temperature utilizing an electromechani-cal Instron 8502 testing device at a constant strain rate of 3.3x10-4 s-1. Due to the deforma-tion process, special attention was paid to the sample preparation for the tensile tests in or-der to prevent wire testing. In order to obtain a well-defined narrowing of the samples the samples for tensile tests were swaged in their centre only. The length of the sample was 150 mm and the gauge length was 100 mm. The ratio of diameter within and out of the gauge length amounts 0.9. The strain was measured with the help of a mechanical extensometer with a gauge length of 25 mm, being placed in the narrowed region of the sample.

280

Results and discussion

A key result of the present study, which in its simplicity could not be boosted, is that magne-sium is deformable under special conditions like high temperature. This would not have been expected. The origin of the formability is seen in the co-deformation of magnesium within a shell, i.e. by the rod-in-tube confine-ment. Although further investigations e.g. with the help of finite element analysis are required the stress state within the deformation zone can be identified to be the key parameter that eases the formability of the magnesium core. Rotary swaging in principle generates com-pressive stresses within the deformation zone. However, this stress state is not homoge-neously distributed. The swaging jaws open and close during the process and impact of the material at single positions. During clos-ing of the jaws the stress is enhanced and its field spreads from the contact positions where the jaws hit the material at the beginning to the entire cross section. In consequence, the stress field distribution appears more homo-geneous within the centre and there are sev-eral hot spots at the rim of the processed ma-terial, depending on the number of the jaws [20]. If this inhomogeneous distribution of the stress field is the origin of materials fail-ure, it cannot be deformed by means of ro-tary swaging, as e.g. magnesium. If, on the other hand, this material is not subjected to this disadvantageous stress field, forming be-comes possible. Hence, co-deforming a rod-in-tube composite by rotary swaging allows to subjecting both materials to more or less homogeneous places within the stress field. When an easily deformable tube is used the

inhomogeneous stress field at the rim can be endured without failure. The core would face a homogeneous distribution of the stress field with tangential stresses. Both are beneficial for the formability, which especially holds for hardly deformable materials such as e.g. magnesium. When compared to drawing, where the stress field is also homogeneous and shows tangential stresses, rotary swaging is characterised by the lack of tensile stress-es [21]. It has been shown, that even magne-sium can be cold drawn up to a deformation strain of 30 % [22]. However, inhomogene-ities at the rim of the wire sample are the or-igin of material failure during deformation. The microstructure of the samples has been investigated by means of SEM and EBSD. The EBSD maps were taken from the centre of the wire and from the longitudinal area. However, there was no significant difference of micro-structural features observed from the border compared to the centre of the magnesium. Figure 1 shows the orientation mapping ac-cording to the inverse pole figure of the wire axis of the magnesium core as determined from EBSD measurements. In addition the orientation density as determined upon these measurements is shown. Both results are provided for various deformation strains to reveal microstructural changes during defor-mation. The strains are referred to the loga-rithmic deformation strain ε=ln(Ai/A) being determined upon the initial (Ai) and actual (A) cross sectional area.

281

a)

b)

c)

d)

e)

f) Fig. 1: Orientation mapping according to the inverse pole figure of the wire axis (f) of the magnesium core (left) and orientation density plotted in the in-verse pole figure of the wire axis showing the texture (right) after logarithmic deformation strains of a) ε=0.54, b) ε=1.45, c) ε=1.89, d) ε=3.42 and e) ε=8.38, respectively.

282

Fig. 2: Grain size of the magnesium core in dependence of the applied deformation strain. The evaluation of the grain size has been done ac-cording to the minimum deviation angle across the grain boundary and according to the method of weighting as indicated.

The initial state is non-deformed. The corre-sponding microstructure consists of grains with an equiaxed shape and a grain size of above 20 μm. The deformation causes grain re-finement. However, at the beginning of defor-mation coarse grains are surrounded by small ones (Fig. 1b). Similar observations have been found by Young et al. and Biswas et al. at low strains for the Mg alloy AZ31 [23, 24]. They have shown, that a high deformation strain is required to obtain a narrow grain size dis-tribution [23, 24], which has also been found previously for a Ti-Mg(AZ31) composite [1]. With increasing deformation strain, the coarse grains vanish from the microstructure, which becomes homogeneous. This is also re-flected by the texture analysis. The initial tex-ture is characterised by a random orientation of the basal plane with respect to the wire axis and the c-axis being oriented within the cross section. Already a low deformation strain ap-plied by means of rotary swaging causes the [1010] direction to rotate along the direction of the wire axis; this direction is slightly tilted against the wire axis. This cone fibre texture converts into a centred ring fibre texture with increasing strain. The alignment of [1010] par-allel to the wire axis seems to be the final tex-ture of magnesium when deformed by rotary swaging as for large deformation strains this

texture component just sharpens (see right parts of Figs. 1c - 1e). At these large deforma-tion strains the microstructure shows no fur-ther significant change. The grains are equi-axed and their size remains at the same level. The evolution of the grain size of the mag-nesium core during deformation is shown in figure 2. The grain size has been determined upon the EBSD maps. As the grain size dis-tribution remains slightly asymmetric, two different weighting methods were applied to discuss the grain size evolution. The calcu-lated average is influenced by the grain size/area itself. Small grains strongly influence the arithmetic average (da), because of their large quantity. On the other side, the area weighted average (df) is more sensitive to large grains, due to their large area fraction. In addition, the results shown in Fig. 2 were obtained from two different criteria to gain informa-tion about high and low angel grain boundar-ies (HAGB or LAGB). The EBSD criterion is a minimum misorientation angle between two data points, to define a grain boundary and, thus, to distinguish between several grains. The use of both criteria provides the possi-bility to evaluate the influence of HAGB and LAGB.

283

Fig. 3: Engineering stress-strain curves of AA6082/Mg composites for different previously applied degrees of cold work (log. deforma-tion strain as indicated); left: rod-in-tube, right: multi-filamentary wire.

The initial grain size of the magnesium core lies within 20 μm to 70 μm. The arithme-tic mean value generally exhibits lower val-ues compared to the area weighted average (please note the different scaling of the abscis-sae). The trend for all graphs shows a decreas-ing grain size with increasing strain. The area weighted mean diameter is more sensitive to large grains and, therefore, the decrease of the grain size at the beginning of deformation is clearly depicted. On the other hand, the arith-metic mean value does not reflect this situa-tion correctly. Therefore, it should not be used to characterise the microstructure, which becomes obvious, when considering Fig. 1b and the corresponding mean grain sizes de-picted in Fig. 2. When a logarithmic defor-mation strain of 2 has been reached, the grain size (df) remains at a nearly constant level. In addition, the grain size distribution becomes narrower and the difference between the grain sizes as determined from the 3° and 15° criterion yield comparable values. The differ-ence between the determined values is caused by the substructure. Comparing the diagrams of the two different criteria, the graphs of the diagram for the 15° criterion were shifted to higher values. This fact can be explained by the missing sub-grain information, due to the 15° criterion. In case of 3° criterion sub-grains

were considered within the calculation. The saturation of the grain size indicates the oc-currence of a dynamic recrystallization. Dy-namic recrystallization may also occur at low strains, but is not recognized because of the large contribution of coarse grains. In addi-tion, the asymptotic behaviour at high strains proves the saturation grain refinement caused by cold work. Additionally, a result from pre-vious investigations [1] provides a further mechanism of grain refinement and its satu-ration. A magnesium alloy AZ31, which has been deformed in a titanium tube, revealed that twinning is the initial processes of the grain refinement. This process is followed by dynamic recrystallization, dynamic recovery and twinning, providing the reason for the dynamic equilibrium of the grain size during deformation at large strains [1]. This has to be proven for pure magnesium in further studies. Grain refinement and the increase of the dis-location density are two major impacts of cold work. Both cause an increase in strength. As seen from Figs. 1 and 2 the grain size is re-duced in consequence of cold work. Figure 3 shows engineering stress-strain curves of the composites after different deformation strains have been applied.

284

Fig. 4: Mechanical properties of AA6082/Mg composites determined by tensile tests in dependence of the previously applied deformation strain; up to =4.3: single-filamentary and above: multi-filamentary wire.

Due to the large initial cross section and beat large diameters. Tensile tests were performed for sample diameters of 5.4 mm and below. In the case of the rod-in-tube compounds, this corresponds to a logarithmic deformation strain ranging from ε=2.77 to 4.29 and in the case of the bundled, restacked and further co-deformed multi-filamentary wire to ε=7.07 to 8.39. The ultimate tensile strength sUTS as well as the yield strength sy increase with in-creasing amount of cold work. All curves show a large plastic strain of the compounds and the failure of the sample is ductile as seen from the downward slope at largest strains. When the composite has been subjected to very low deformation strains, i.e. up to ε=3.42, the failure mode can be in two steps as shown for ε=2.77 and ε=3.42 in the left picture of Fig. 3. In this condition, the AA6082 shell fails with the magnesium core being still intact. The core fails at larger strains. This behaviour is not observed in a heavily cold worked con-dition since the yield strength of both metals became comparable due to work hardening.Figure 4 summarises the mechanical be-haviour of the as deformed composite wires. As already seen from Fig. 3 the observed plas-tic strain does not follow an obvious tenden-

cy. Indeed there is a large scattering within the experimental values for the plastic strain. Scattering is less pronounced for sUTS and sy because of which it seems justified to showing just one typical curve in Fig. 3, each. The large scatter of the experimental values for the plas-tic strain arises from the clamping as well as from the preparation of the tensile test sam-ples and cannot be reduced. However, it can be seen from Fig. 4 that the plastic strain in tension is getting lower for samples that were subjected to a larger cold deformation strain. The mechanical strength of the samples in-creases with the amount of cold work to which the composite was subjected to be-fore. After a logarithmic deformation strain of ε=8.38 has been applied, the multi-fila-mentary composite shows an ultimate ten-sile strength of 240 MPa. This corresponds to a specific strength of 104 MPa/(g/cm³). In comparison to the rod-in-tube composite being deformed up to ε=4.29 and showing sUTS=160 MPa, the specific strength has not increased so much, as this composite has a specific strength of 80 MPa/(g/cm³). This is attributed to the newly introduced AA6082 tube, whose strain hardening is lower than that of the already deformed internal composites.

285

ε

Hence, a further bundling, re-stacking and co-deformation step would possibly not yield to a significant enhancement of the specif-ic strength. On the other hand, this flattens the path to tailored wires with a high specif-ic strength and at the same time an adapt-able mass density, within certain limits. The Young’s modulus which has been deter-mined for the wire also reflects the compo-sition. Within the experimental scattering of the data, the Young’s modulus is at a constant level for the rod-in-tube or single-filamentary wire, i.e. up to a deformation strain of ε=4.29. After being bundled and stacked into a fur-ther AA6082 tube, the Young’s modulus is at a higher level, accounting for the larger amount of the aluminium alloy within the composite. Its value is at a constant level again.

Summary

Co-deformation of magnesium within a shell from an aluminum alloy AA6082 is possible up to a logarithmic deformation strain of at least 8.38. This degree of deformation has been achieved by deforming a rod-in-tube composite to a single-filamentary wire and further processing this wire by cutting, bun-dling and co-deforming these wires in a newly deformed AA6082 shell in form of a multi-fil-amentary wire. The cold work causes (i) sig-nificant grain refinement yielding a mean grain size below 5 µm as well as (ii) a preferen-tial texture of crystals with [1010] axis being aligned parallel to the wire axis. Within this condition the AA6082/Mg composite shows a specific strength of 104 MPa/(g/cm³).

References

1. E. Knauer, J. Freudenberger, T. Marr, A. Kauffmann, L. Schultz: Grain refinement and deformation mechanisms in room temperature Severe plastic deformed Mg-AZ31, Metals 3 (2013) 283-2972. T. Marr, J. Freudenberger, A. Kauffmann, J. Scharnweber, C.-G. Oertel, W. Skrotzki, U. Siegel, U. Kuehn, J. Eckert, U. Martin, L. Schultz: Damascene light-weight metals, Advanced Engineering Materials 12 (2010) 1191-1197 3. T. Marr, J. Freudenberger, D. Seifert, H. Klauss, J. Romberg, I. Okulov, J. Scharnweber, A. Eschke, C.-G. Oertel, W. Skrotzki, U. Kuehn, J. Eckert, L. Schultz: Ti-Al composite wires with high specific strength, Metals 1 (2011) 79-97 4. C. C. Koch: Bulk Nanostructured Materials; (Eds. M. J. Zehetbauer, Y. T. Zhu), WILEY-VCH Weinheim, 20095. R. Valiev: Nanostructuring of metals by severe plastic deformation for advanced pro perties, Nature Materials 3 (2004) 511-5166. M. Meyers, A. Mishra, D. Benson: Mechanical properties of nanocrystalline materials. Progress in Materials Science 51 (2006) 427-5567. R. Valiev, R. Islamgaliev, I. Alexandrov: Bulk nanostructured materials from severe plastic deformation. Progress in Materials Science 45 (2000) 103-1898. H. Gleiter: Nanocrystalline materials. Progress in Materials Science33 (1989) 223-3159. H. Gleiter: The mechanism of grain boundary migration. Acta Metallurgica 17 (1969) 565-57310. K. Youssef, R. Scattergood, K. Murty, C. C. Koch: Nanocrystalline Al-Mg alloy with ultrahigh strength and good ductility. Scripta Materialia 54 (2006) 251-256

286

11. S. Zhang, Y. Peng, W. Tang, D. Li: The polycrystalline plasticity due to slip and twin ning during magnesium alloy forming, Acta Mechanica 212 (2010) 239-30312. Z. Pu, G.-L. Song, S. Yang, J. C. Outeiro, O. W. Dillon Jr., D. A. Puleo, I. S. Jawahir: Grain refined and basal textured surface produced by burnishing for improved corro sion performance of AZ31B Mg alloy, Corrosion Science 57 (2012) 192-20113. T.G. Langdon: The principles of grain refinement in equal-channel angular pressing. Materials Science and Engineering A 462 (2007) 3–1114. P. Frint, M. Hockauf, D. Dietrich, T. Halle, M. F. X. Wagner, T. Lampke: Influence of strain gradients on the grain refinement during industrial scale ECAP. Materialwissenschaft und Werkstofftechnik 42 (2011) 680–68515. Y. Zhu, J. Huang, J. Gubicza, T. Ungar, Y. Wang, E. Ma, R. Z. Valiev: Nanostructures in Ti processed by severe plastic deformation. Journal of Materials Research 18 (2003) 1908–191716. A. Vorhauer, R. Pippan: On the homogeneity of deformation by high pressure torsion. Scripta Materialia 51 (2004) 921–92517. M. Zehetbauer, J. Kohout, E. Schafler, F. Sachslehner, A. Dubravina: Plastic deforma tion of nickel under high hydrostatic pressure. Journal of Alloys and Compounds 378 (2004) 329–33418. Y. Saito, N. Tsuji, H. Utsunomiya, T. Sakai, R.G. Hong: Ultra-fine grained bulk aluminum produced by accumulative roll-bonding (ARB) process. Scripta Materialia 39 (1998) 1221–122719. H. Watanabe, T. Mukai, K. Ishikawa: Differential speed rolling of an AZ31 magnesi- um alloy and the resulting mechanical properties. Journal of Materials Science 39 (2004) 1477–148020. A. Hensel, T. Spittel: Kraft- und Arbeitsbedarf bildsamer Formgebung, VEB Verlag für Grundstoffindustrie, Leipzig, 197821. J. Eickemeyer, M. Falter, A. Gueth: Shear crack formation in cold drawing of magnesi- um and cobalt wires, International Journal of Applied Mechanics and Engineering 12 (2007) 1173-117922. S. Biswas, S.S. Dhinwal, S. Suwas: Room-temperature equal channel angular extrusion of pure magnesium. Acta Materialia 58 (2010) 3247-326123. J. Young, M. Heiden, Y. Hovanski, D. Field: Microstructural Analysis of Severe Plastic Deformed Twin Roll Cast AZ31. In Proceedings of the Mg2012: 9th International Conference on Magnesium Alloys and their Applications, Vancouver, Canada, July 2012; Volume 9, 1087-1094

287

A unified approach to identify material properties from small punch test experiments

Martin Abendroth

TU Bergakademie FreibergInstitute of Mechanics and Fluid DynamicsLampadiusstraße 4, 09599 Freiberg, Germany

288

Abstract In recent years the small punch test method has become an attractive alternative compared to traditional material testing procedures, especially in cases where only small amounts of ma-terial are available. It has been applied to determine the current and local material state in structural components under operating conditions. A wide range of material properties like elastic, plastic, creep, damage and fracture behavior can be obtained using this technique. But the assessment of the relevant parameters is not as simple as from standard tests, because of the non-uniform stress and deformation state. However, this can be achieved by comparing the experimental SPT results with those obtained by finite element computations of SPT using advanced material models. Then the task is to determine the parameters of the material mo-dels using special optimization techniques. This paper presents an approach to evaluate several small punch experiments simultaneously, taking advantage of neural network approximations, modern optimization strategies and data bases.

Introduction During the last three decades, the small punch test (SPT) has been established as a suitable and versatile miniaturized test meth-od to determine the mechanical behavior of a broad range of materials. In contrast to conventional material testing techniques its great advantage is the small amount of mate-rial required. Therefore, in combination with small specimen sampling techniques the SPT becomes especially attractive, if the actual material state in structural components has to be evaluated after in service operation under-going embrittlement, fatigue or aging. Anoth-er advantage is the opportunity to investigate local material properties as in functionally graded materials or welded joints, composite layers or coatings, where no traditional bulk specimen can be removed. Compared with other miniaturized test specimens (tensile or bending rods) the SPT has the following ad-vantages: i) the stress state is biaxial, which meets the loading conditions in many struc-tural components (vessels, pipelines, plates) ii) the experimental handling is comparable easy and iii) the involved material volume to be tested is relatively large compared to the specimen volume. Therefore, the potential of the SPT has been recognized by many re-searchers, which has been substantiated by a broad field of applications. The main draw-back of the SPT consists in the non-homo-

geneous stress and strain fields within the specimen, avoiding a simple interpretation of measurement in terms of material parameters as this is possible in conventional testing pro-cedures. For that reason, much effort has been spent in the past to find correlations between SPT results and standard material properties as yield strength, ultimate stress, Charpy en-ergy etc. On the other hand, true material pa-rameters in the sense of constitutive laws in continuum and damage mechanics seem to be more general and allow the transferabili-ty to larger specimens, complex stress states and even structural components. Nowadays, the numerical finite element tools are well developed to simulate complicated specimen geometries together with advanced material laws. To exploit the full information coming out from a SPT experiment, a qualified nu-merical analysis can be performed embedded into an optimization algorithm to identify the unknown material parameters. There exist material models which describe material be-havior over a range of loading parameters as stress or temperature or which are sensitive to internal variables as damage or triaxiality. For such materials multiple tests with vary-ing loads, geometries or changing environ-ments are necessary to investigate the effects of interest. Then the parameter identification strategy must consider all experimental tests.

289

The small punch test Fig. 1 shows the principle sketch of the small punch test with the essential geometrical measures. The small punch test is used in dif-ferent sizes and different types. For example the smallest specimens are standard TEM sized specimen [1], others use specimen cut from remaining pieces of Charpy specimens, which are square shaped [2]. There has been a lot of effort to standardize the SPT and its usage [3], but this is still a running process. At least, there is a common sense about the important features of the test. A disk shaped specimen with a diameter D and a thickness t is placed on a circular die with a receiving

hole of diameter d. This receiving die can have a round or straight chamfer of size r. The specimen can be clamped between the receiving die and a downholder. There are also cases where the specimen is not clamped, usually for testing very brittle materials [4] to avoid initial deformations during clamping. The above mentioned European guideline [3] suggests a standard geometry with values of D=8 mm, t=0.5 mm, R=1.25 mm, d=4 mm and r=0.5 mm. The middle of Figure 1 shows a typical load displacement curve (LDC) for a ductile metallic material.

The LDC, the essential experimental outcome of the SPT, can be split up into several parts. Part I is mainly determined by the elastic properties of the material, Part II reflects the transition between the elastic and plastic be-havior, Part III shows the hardening proper-ties up to part IV, where geometrical softening and damage occurs. At the beginning of Part

Figure 1: left) Principal sketch of the small punch test; middle) a typical resulting load deflection curve for a ductile metal; right) axisymmetric simulation of a SPT.

V a crack initiates. During the steep decent in Part V the crack grows circular around the center of the specimen. The remaining force in Part VI is needed to push the punch trough the already cracked specimen. Baik et al. [5] defined the area under the LDC as the small punch fracture energy and found correlations with the values determined from Charpy-V-notch specimen. Suzuki et al. [6, 7] found correlations between the maximum force Fm/t2 and the ultimate tensile strength Rm as well between the value of Fy/t

2 and the yield strength Ry. Furthermore they also related the equivalent fracture strain using

εqf=ln(t0/tf)=β(uf/t0)n with measures of speci-

men thickness tf and deflection uf at fracture. Recently Garcia et al. [8] compared and dis-cussed several alternatives how to determine yield stress and ultimate tensile stress from SPT experiments. The small punch test to-gether with fracture mechanical simulations is used by Cardenas et al. [9] to determine fracture toughness values of ductile steel. Therefore, the specimens where notched using a micro tool creating a notch with a tip radius of 0.1 mm and 30° cut angle. The fracture toughness values can be identified using simulations and evaluating the J- in-tegral at the point where a crack initiates. Deformation measurements of the specimen can be very useful. Egan et al. [10, 11] sug-gest optical methods to access the deforma-tion profile of the small punch specimen. The measured profiles are valuable information for material parameter identification methods.

290

Figure 2: Typical results for the different types of the SPT (CDR, CF, CD) for a visco-plastic metallic material

The loading can be a constant displacement rate (CDR) of the punch, a constant force (CF) applied to the punch or an initial cons-tant displacement (CD) of the punch followed by a holding (relaxation) time. The experi-mental results of the test are usually the punch displacement and/or the specimen deflection and the punch force. In case of time depen-dent material behavior these values are stored together with the time after starting the test. The typical results for the different types of the SPT are shown in Fig. 2 considering an elastic, visco-plastic material as the most metals are. The CDR-SPT can be performed at different punch velocities (or deflection rates) u. If the punch speed is increased the curve is shifted to higher forces due to the strain rate sensiti-vity and the onset of damage happens at smal-ler deflections [13]. The CF-SPT is used to determine the material creep behavior at dif-ferent loads (stresses). As in tensile creep tests we distinguish three parts of the curve, which are related to primary, secondary and tertiary creep. The primary part of the curve is also in-fluenced by some initial plastic deformation at rather high strain rates. The tertiary part is in-fluenced by a localization of deformation and increasing creep damage. For increasing test forces a higher mean specimen deflection will

(1)

(2)

(3)

be observed together with decreasing failure times [14]. The CD-SPT is not as common as the two other types of the test. But neverthel-ess it could be a rather fast test to determine visco-plastic material behavior. The test starts with a predefined deflection, which is applied in a short time, followed by a longer relaxation period where the deflection is kept constant. The main result is the decreasing part of the curve, which depends on the visco-plastic (creep) material properties. The main advan-tage of this test type is that very small creep rates can be achieved in a rather short time. All tests can be performed at different tempe-ratures or other test conditions. In general one can interpret the experimental results as a set of three functions, one for each test type:

where u denotes the specimen deflection or punch displacement, the deflection rate, F the punch force, t the time and pi a set of i material parameters. The argument list of these functions could be extended by the test conditions like test temperature and geome-try parameters of the SPT. In a more general mathematical framework the three functions represent boundary value problems which are solved using the finite element method (FEM).

u

Ideally would be a three-dimensional strain measuring method, which also accounts for rigid body motions of the specimen. There exist grating methods as mentioned in [12], but the accuracy of those methods is most often not sufficient, especially the distinction between rigid body motions and the overlay-ing small strains is problematic.

CDR( ) ( , , )iF u f u u p=

CF( ) ( , , )iu t f t F p=

CD( ) ( , , )iF t f t u p=.

291

Numerical simulations of the SPT This sections concentrates on aspects which come along with the finite element analysis of the SPT. A general guideline for finite element modeling states: Keep the model as simple as possible, but as detailed as necessary to cap-ture the relevant effects or phenomena with the required accuracy. Since the geometry of the SPT is axisymmetric a two dimensional (axisymmetric) model can be used. But this implies that also the material properties used for the specimen must fulfill this symmetry condition. As long as the material under con-sideration is isotropic or transversal isotropic (isotropic plane perpendicular to the SPT symmetry axis) this condition is fulfilled. As soon as local damage or fracture become re-levant, a three dimensional model should be used. But even then one might consider sym-metries which can be used to simplify the mo-del (half model, quarter model, or an angu-lar section). The computation time depends mainly on the number of degrees of freedom

The model size can be drastically reduced if parts of the model can be treated as rigid bodies. For the SPT everything except the specimen could be modeled as rigid bo-dies as the specimen deformation are large against the elastic deformations of the se-tup. If the elastic deformation of the setup is a substantial part of the total deformation they can be considered using a compliance correction term for the specimen deflection.

Figure 3: Elastic deformations of a SPT setup (without specimen) and the corresponding compliance function.

Figure 4: Influence of geometrical details of the setup on the resulting load-deflection curves.

CPU DOFt nα∝within the model and the bandwidth of the stiffness matrix: with 2 3.α ≈ …

α

Figure 3 shows the elastic deformations of a SPT setup. These deformations depend only on the applied force and not on the tested material. That’s why such a simulation needs to be per-formed only once to determine the setup com-pliance function, which is later used to correct the simulated deflection for a certain force.

292

Material models The choice of a mechanical constitutive model depends on the material and the phenomena which are encountered in the mechanical test. Commercial finite element codes like ABAQUS or ANSYS provide a wide range of models for almost every material and loading situation. The choice of an appropriate model is the crucial task, because this choice defines which phenomena can be modeled. Here, we will concentrate on models for (ductile) me-tallic materials. The standard procedure to de-termine parameters of such materials require a number of standard tensile specimen for determining tensile and creep properties and in case of damage or fracture properties also CT- or 4PB-specimen.

Figure 4 illustrates the problem of an incor-rect model geometry. It was observed that the receiving die and the punch tip contour did not have the correct geometry given in the tech-nical drawings. Therefor, simulations where performed comparing the ideal and real geo-metry and significant differences were found in the resulting load displacement curves. In simulations used to identify material parame-ters always the real geometry should be used. One important part of the model is the contact formulation. Here, we use a node to surface algorithm with a constant friction coefficient µ . The normal contact stress nσ depends on the overclosure h between the contact sur-faces. For h c≤ − the normal contact stress

n 0σ = and for h c> − we have:

This realizes a softened contact as we would have it for a surface roughness of c. n0σ defines the normal contact stress at zero distance between the contact surfaces. In figure 5 the influence of friction on the LDC is shown. One can note that friction influences the LDC after a certain deflec-tion. That is the point where relative mo-tion between punch and specimen occurs. Another important step is to perform a so called mesh convergence test to check that the chosen mesh size is fine enough. The mesh is gradually refined until the changes in the re-

Figure 5: Influence of mesh size (with GTN model) and friction on the resulting load-deflection curves.

(4)

sults are below a predefined accuracy value. But here one have to be careful if a material model is used where localization effects can occur as it is the case for local damage mo-dels like the GTN model. Then the mesh size is part of the material model which requires a local length scale for damaged material zo-nes. At the left of figure 5 different meshes are compared. An increasing number of elements in radial direction leads to earlier damage be-cause a bit less energy is needed to damage smaller elements.

0 1 exp 1 1exp(1) 1

nn

h h

c c

σσ = + + −

−

293

Elastic plastic behavior

Ductile metallic materials show an elastic-plastic behavior, where the elastic strains can be considered small against possible plastic strains. Prior to yield, the material response is assumed to be linear elastic. The strain tensor is split into an elastic and a plastic part.

Due to the axisymmetric geometry and loa-ding of the SPT only isotropic material beha-vior can be identified. Furthermore the SPT applies a monotonic loading thus no kinema-tic hardening, only isotropic hardening can be identified from experimental results.

The stresses in terms of the elastic strains are expressed by the multiaxial Hooke’s law, which for the isotropic case reads as:

where E denotes the Young’s modulus, ν Poisson’s ratio and ijδ the Kronecker delta. The plastic strain rate is derived from a flow potential Φ .

The flow rule is given by

,

where qσ denotes the equivalent (von Mises) stress and pl

y q( )σ ε a yield function. The yield function can contain different hardening laws as the Voce law

or the well known Ramberg-Osgood law.

Viscoplastic behavior

For a rate dependent plasticity model we de-fine the creep strain rate as a combination of

where is the pressure dependent volumet-ric (swelling) strain rate and the equivalent deviatoric creep strain rate. ijn defines the direction of the creep strain derived from the equivalent stress potential.

The volumetric and deviatoric strain rates have evolution laws like:

and

depending on the equivalent von Mises stress qσ and the equivalent pressure

.

A simple example is the Norton creep law where and with the ma-terial parameters A and n. A more advanced creep model is a combination of i Norton laws

,

which allows the modelling of multiple creep mechanisms like different diffusion and dis-location mechanisms. The McAulay brackets have the meaning as ( ) / 2x x x= + . This ensures that a creep mechanism is only active above a respective threshold stress Bi.

Damage behavior

In order to simulate plasticity and ductile da-mage the continuum damage model of Gur-son [15, 16] can be used with the extensions of Tvergaard and Needleman [17, 18]. The central part of the model is the yield function

crp 0ε =

0crq 0( )n

Aσε ε=

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

(13)

(14)

(15)

el pl

ij ij ijε ε ε= +

el el

1 1 2ij ij kk ij

E νσ ε ε δ

ν ν= +

+ −

pl

ij

ij

ε λσ

∂Φ=

∂

( )pl

q y q 0σ σ εΦ = − =

( ) ( )pl pl pl

y q 0 1 q 2 q1 exp nσ ε σ σ ε σ ε= + + − −

( )1

pl

pl q

y q 0

0

nεσ ε σ

ε=

cr cr cr

p q

1

3ij ij ijnε ε δ ε= +

ij

ij

nσ

∂Φ=∂

( )cr cr cr

p p q p q, , , , ,h pε σ ε ε θ= …

( )cr cr cr

q p q p q, , , , ,h pε σ ε ε θ= …

13 ij ijp σ δ= −

cr q

q 0

in

i

i i

B

A

σε ε

−= ∑

crp 0ε =

0crq 0( )n

Aσε ε=

294

Figure 6: Top) GGG-40 CDR-SPT finite element simulation dis-playing the damage at failure. Bottom) P91 CF-SPT finite element model displaying the creep strain at failure.

where 3q 2 ij ijS SΣ = denotes the macroscopic

von Mises and 1p 3 iiΣ = Σ the macroscopic hy-

drostatic stress, expressed by the macroscopic deviatoric stresses pij iij jS δ= Σ −Σ . The mate-rial damage *

ff depends on the void volume fraction .

with

1

* 1f qf = . Up to a critical void volume frac-

tion cf the damage is identical with the value of the void volume fraction. Beyond cf where voids coalescence or micro crack initiation is assumed damage evolution is accelerated un-til a final void volume fraction ff is reached where the material fails. The evolution of the equivalent plastic strain of the matrix mate-rial is obtained from the plastic macroscopic strain rate

The evolution of the void volume fraction is combined of two terms

,

where describes the growth of voids based on the law of conversation of mass

and a void nucleation part, which follows a strain controlled relationship

f

plijE

The normal distribution of the nucleation strain has a mean value nε and a standard deviation of ns . nf denotes the volume frac-tion of void nuclei. For detailed information about the implementation into the FE-Code ABAQUS see [19-21].

Fig. 6 shows results from finite element simu-lations of the SPT and Fig. 7 the correspon-ding specimens after testing. The GGG-40 is a ductile cast iron containing spherical graphite inclusions which act as voids. Therefore the GTN damage model is used to simulate the onset of failure. On the right hand side of Figs. 6 and 7 results for a P91 specimen are shown. P91 is a material used for high temperature components under internal pressure, whereas the creep behavior is of great interest. The si-mulations predict the correct locations of fai-lure (GGG-40) and necking (P91) very well.

(16)

(17)

(18)

(19)

(20)

(21)

( ) ( )( )

*q p

1 2pl pl

y q y q

2*

1

32 cosh

2

1 0

q f q

q f

σ ε σ ε

Σ ΣΦ = +

− + =

( )

c

** f c

c c c f

f c

*

f f

if

if

if

f f f

f ff f f f f f f

f f

f f f

≤

−= + − < <

−

≥

( )

pl

pl pl

0 0q

d .1

t ij ij

q q

Et

fε ε

σ

Σ= +

−∫

gr nuclf f f= +

gr nuclf f f= +

( ) pl

gr 1 kkf f E= −

pl

plq nnnucl q

nn

1exp

22

ff

ss

ε εε

π

−= −

295

quality or exactness of the simulations for each test. Each pair of experiment and corre-sponding simulation gets its own error value, which is multiplied by a certain weight wi. Those weights represent the importance of an experiment or the confidence the user has for this single experiment. For the CDR-SPT the error is defined as the sum of integrals of the normalized difference of the punch force from all CDR-SPT simulations and experi-ments in a predefined interval .

The error for the CF-SPT is the sum of the integrals of the normalized difference of the-punch displacement from all CF-SPT simu-lations and experiments in a predefined time interval plus an error which ex- presses the normalized diffrence beween the times of failure for all the simulations and ex-periments.

Figure 7: Left) GGG-40 CDR-SPT specimen just after failure. Right) P91 CF-SPT specimen after creep testing [14]

The above described model belongs to the class of local damage models. It is well known that the results of these models are mesh de-pendent. A damage zone usually localizes within an one element thick band or plane. To avoid this one can use non-local damage models, see [22–24] for details. Such models usually introduce a characteristic length as an additional parameter which can be related to the spacing of voids or the width of localized damage bands. Using these models element sizes smaller than the characteristic length are required, which can lead to larger mo-dels than those using a local damage model and solving additional field equations might be necessary, which requires the use of non-standard solvers and/or elements within the finite element codes.

Parameter identificationAs shown in the previous section the material model contains a set of parameters pi which are to be determined from experimental re-sults. Especially for the creep parameters more than one CF-SPT experiment at diffe-rent loads is necessary. The general way to find parameter sets is to fit the model to the experimental results, which could be a set of different tests (CDR, CF and CD). Mathema-tically it is an optimization process where the difference between experiments and simula-tions needs to be minimized. The value to be minimized is an error, which measures the

min maxi iu u …

(22)

min maxi it t …

(23)

CDRmax

min

2CDR sim expCDR

max min exp1

( ) ( )di

i

n ui i i i i

iui i i i

w F u F ue u

u u F=

−=

−

∑ ∫CDR

max

min

2CDR sim expCDR

max min exp1

( ) ( )di

i

n ui i i i i

iui i i i

w F u F ue u

u u F=

−=

−

∑ ∫

CFmax

min

2sim expCF CF

max min exp1

2sim exp

f f

exp

f

1 ( ) ( )di

i

n ti i i i

i iti i i i

i i

i

u t u te w t

t t u

t t

t

=

−=

−

−+

∑ ∫

296

297

Figure 8: Scheme for the identification process of material parameters using the SPT and neural networks as an approximation for finite element computations.

The error for the CD-SPT is defined as the sum of the integrals of the normalized dif-ference of the punch force from all CD-SPT simulations and experiments in a predefined time interval .

As normalizing values the observed mean va-lues of the punch force or the punch dis-placement are used. Finally, the three er-rors are summed up and divided by the sum of the weights for all experiments.

The minimization of this error is done within an optimization loop as shown in figure 8. One may notice that the finite element com-putations are not a direct part of the optimi-zation loop.

Instead finite element computations are done in advance using parameters which are varied in reasonable bounds. These computed re-sults are used to train neural networks, which represent an approximation of the finite ele-ment simulation. The quality of the neural network approximations can be measured by comparing predictions of the networks with simulation results that had not been part of the training. Each test type (CDR, CF or CD) requires separate neural networks. For creep tests two networks are used, one predicting the failure time and a second predicting the deflection over time. All the networks used here have the structure of feed forward neural networks as shown in figure 9. They consist of at least three layers of neurons. The two first (left) layers have neurons with sigmoid functions, the last (right) layer has neurons with linear functions. All the layers are fully forward connected, which means that each neuron of one layer is connected with each neuron of the subsequent layer. The number of neurons for the first (input) layer is similar to the number of arguments of the function which the network should approximate.

min maxi it t …

(24)

expiu

expiF

(25)

CDmax

min

2CD sim expCD

max min exp1

( ) ( )di

i

nt

i i i i iit

i i i i

w F t F te t

t t F=

−=

−

∑ ∫

CDR CF CD

CDR CF CDR

CDR CF CD

1 1 1

n n n

i i ii i i

e e ee

w w w= = =

+ +=

+ +∑ ∑ ∑

In our case the last (output) layer has only one neuron, representing the return value of the function. The number of neurons in the middle (hidden) layers depends on the com-plexity of the function which the network should approximate. For the networks, which are used to predict creep failure time and SPT creep curves around 10 hidden neurons are sufficient. More complex predictions like for the CDR-SPT under consideration of damage models can need up to 50 neurons in the hid-den layer(s). The training of a neural network is also an optimization process. Details about the training of neural networks can be found in [13, 25].

Figure 10: Scheme of the WEB based parameter identification.

Figure 9: Structure of neural networks used to approximate the finite element simulations.

WEB based parameter identification

In the sections above all necessary parts for a successful identification of material para-meters have been explained. What is missing is an application bringing all the tools to-gether. This application should be accessible to the experimentalists, some of them might not have access to or experience with finite element analysis and optimization tools run-ning on high performance computers (HPC). The most common tool to connect everything with the users (experimentalists) is of course the Internet. This section will explain a struc-ture for a WEB based parameter identification tool.

298

Figure 11: Graphical user interface for the parameter identification procedure

It is build of four blocks (see Fig. 10), which could run on different computers. The arrowsrepresent the flow of information (data, com-mands) through the network. At least one data base is used where experimental and si-mulated results can be stored. This data base also holds the information about the users who want to use the application and the infor-mation about the setups for simulations. An important part is the HPC-Cluster. This is the machine which does all the heavy computati-onal work, i.e. are the simulations of the expe-riments, the training of the neural networks and the optimization processes. The HPC-cluster gets the experimental results from the data base to compare it with its computations. When one thinks of a simulation as an arti-ficial experiment producing similar results it makes sense to store those results in the samedata base and just mark them as simulations. This creates the opportunity to compare newincoming experiments with simulations al-ready done or with already processed expe-

riments and to make predictions or at least suggestions what a good parameter set for a reasonable material model could be. Those two main tools (Data Base and HPC-cluster) are controlled by a central WEB server, which hosts interactive WEB-pages, where external users can login and upload experimental re-sults, running simulations or starting optimi-zation (parameter identification) processes. The whole structure is very flexible. It’s not necessary to have a single date base, instead it is possible to distribute especially the expe-rimental data on different machines, so that certain users can have full control over their data. The software which controls the simu-lations, the construction and training of the neural networks and the interaction between HPC-cluster and the data base(s) is program-med in PYTHON, which is a platform inde-pendent high-level programming language. Figure 11 shows a typical graphical user in-terface for managing all simulations, network training and parameter identification.

299

Application and results This section shows a collection of different applications of the SPT and results, which are obtained using the above explained parameter identification techniques. Hardening parameters for nickel-base super-alloy A thermal sprayed nickel-base super-alloy (modified IN625) is investigated using the high temperature small punch test (HT-SPT). This alloy is used as a corrosion pro-tection coating for components used in heat exchanging devices which can be exposed to a corrosion aggressive environment as in waste incineration plants. The specimens are cut from the substrate using a core drill. The separation from the substrate is done using a diamond blade saw (Strüers Accutom 50). All specimens are finally grinded to a final thickness of 300 m. The test conditions for the HT-CDR-SPT are u=0.5 mm/min at three different test temperatures (773, 973, 1173 K).

At the highest temperature the test environ-ment was either air or argon. Fig. 13 shows four specimen tested at different temperatures and environments. It is obvious that the dam-age mechanism changes from brittle failure at 773 K indicated by the star shaped crack pat-tern over a brittle-ductile transition failure at 973 K to ductile failure at 1173 K, where the final crack has a circular shape. The load deflection curves in Fig. 12 also reflects this behavior. At 773 K the specimen already fails within the elastic region. The load drops at the end of the curve correspond to crack initia-tion events. Finally the specimen breaks into five almost identical pieces. At 973 K the star shaped cracks occur first, followed by plastic bending and further circular crack growth close to the region where the specimen is in contact with the chamfer of the lower die. At 1173 K the load deflection curve is fully devel-oped indicating a pure ductile failure.

To identify the hardening paramters of the alloy for the different temperatures CDR-SPT simulations where done using a Voce harden-ing law (see Eq. 9). The identified parameters are listed in table 1.

Figure 12: Load deflection curves for Nickel-base super alloy SPT. left) Experiments, right) Comparison with FEM simulations using identified hardening parameters.

Table 1: Identified hardening parameter for the nickel-base super alloy

T [K] E [MPa] σ1 [MPa] σ2 [MPa] σ3 [MPa] n

973 33786 139 50.1 53.0 28.21173 15422 6.8 26.1 75.2 29.3

300

Creep behavior of P91

In Fig. 14 CF-SPT experiments (exp) taken from [14] are shown done at a temperature of 873 K at different loads under argon atmos-phere together with simulated results (res) and corresponding neural network predic-tions (nna). The finite element simulations where done using a user creep law consid-ering two Norton laws (see Eq. 15). For the creep parameter identification two different neural networks where used. One just pre-dicting the time of failure depending on the material parameters and a second one just approximates the specimen deflection over a

Figure 13: Nickel-base super alloy SPT specimen testet at a) 773 K, b) 973 K, c) 1173 K, d)1173K at argon.

normalized time, which is taken from the first network. The optimization routine is a SQP algorithm as explained in detail in [13].

At the right side of Fig. 14 a single CF-SPT simulation together with the corresponding neural network approximation is shown to demonstrate the accuracy of the neural net-works. All seven experiments where evaluated together using the approach in section 5. The identified parameters are listed in table 2.

301

Damage and fracture properties of 22NiMoCr37Fig. 15 shows the comparison between experi-mental and simulated CDR-SPT results for re-actor pressure vessel steel 22NiMoCr37. This material (similar to ASTM A508 cl. 2) is used for pressure vessels and pipings in nuclear power plants. Such components are exposed to neutron radiation, which can cause embrit-tlement of the material, whereas the current fracture toughness has to be supervised dur-ing the lifetime of the respective component.

Figure 14: Left) CF-SPT experiments for P91 at 823 K and different loads together with the simulated tests and the corresponding neural network approximations. right) A single simulation with its neural network approximation.

Figure 15: left) CDR-SPT results for 22NiMoCr37 at RT. right) Simulation of a tensile test using the identified parameters for 22NiMoCr37.

Table 2: Identified parameters for P91

Here, the GTN model was used to simulate hardening and ductile damage due to void nucleation, void growth and coalescence. The identified parameters from the SPT were used to simulate a standard tensile (Fig. 15 right) and compact tension tests (CT-25, see Fig. 16 right). It was found that these models can predict other tests quite well, only using ma-terial parameters from SPT evaluations. More detailed information about the training of the neural networks and the modeling of the frac-ture toughness specimen can be found in [12].

A1 B1 n1 A2 B2 n2 µ[MPa] [MPa] [ - ] [MPa] [MPa] [ - ] [ - ]

550 0.445 11.01 456.4 144.3 12.11 0.5

302

ConclusionThis paper presents a general approach to identify material parameters obtained from small punch experiments by means of finite element simulations. A fully parametric mo-del is used to simulate SPT experiments. The model allows geometry variations as well as the choice of an appropriate material mo-del and specifying material parameters. All SPT types can be simulated with regard to different loading conditions. The parameter identification process can take several expe-riments into account, making it possible to characterize even complex material behavior. The graphical user interface which has been developed makes this approach accessible to experimentalists which don’t have own access or experience in finite element analysis. Finite element analysis using advanced constitutive material models allow us to transfer the iden-tified parameters (from SPT) to simulations of standard specimens and the prediction of material behavior under more general situa-tions.

AcknowledgementThe financial support for this project by the EFRE fund of the European Union is grate-fully acknowledged. Furthermore the author thanks Dr. Petr Dymácek from IPM Brno for providing the P91 creep data. Special thanks goes also to the students and co-workers Christin Heinig, Carolin Ranft, Tobias Kaden, Stefan Soltysiak and Wolfgang Kilian who hel-ped to develop the universal SPT-model and some of the optimization tools. Without the staff from the labs Dagmar Schmidt and Kurt Fredersdorf who were preparing countless specimens and running all the experiments this work could not be done. And last but not least the author thanks Prof. Meinhard Kuna for many fruitful scientific discussions.

Figure 16: Simulation of a CT-25 specimen using the identified parameters for 22NiMoCr37, left) force-load line opening, right) crack resistance curve.

Table 3: Identified hardening and damage parameters for 22NiMoCr37

E[GPa]

σ0 [MPa] ε n f0 fc ff fN ε sN q1 q2

199 430 0.00492 6.44 0.002 0.117 0.2 0.05 0.3 0.1 0.846 1.03

303

References

1. J. Kameda and X. Mao. Small-punch and TEM-disc testing techniques and their applica-tion to characterization of radiation damage. Journal of Materials Science 27(4) (1992) 983–989.

2. T. Linse, M. Kuna, J. Schuhknecht et al. Application of the small punch test to irradiated reactor vessel steels in the brittle- ductile transition region. Journal of ASTM International 5(5) (2008).

3. CEN. Workshop Agreement CWA 15627:2006, Small Punch Test method for Metallic Ma-terials. Technical report, Brussels, Belgium (2006).

4. S. Soltysiak, M. Abendroth, M. Kuna et al. Strength of fine grained carbon-bonded alumi-na (Al2O3–C) materials obtained by means of the small punch test. Ceramics Internatio-nal 40(7, Part A) (2014) 9555 – 9561.

5. J.-M. Baik, J. Kameda and O. Buck. Small punch test evaluation of intergranular embrittle-ment of an alloy steel. Scripta Metallurgica 17(12) (1983) 1443 – 1447.

6. M. Suzuki, M. Eto, Y. Nishiyama et al. Estimation of toughness degradation by microhard-ness and small punch tests. ASTM Special Technical Publikation 1204 (1993) 217–227.

7. M. Suzuki, K. Fukaya, Y. Nishiyama et al. Report JAERI-M 92-086. Technical report, De-partment of high temperature engineering, tokai research establishment japan energy re-search institute tokai-mura, naka-gun, ibaraki-ken, Japan Atomic Energy Research Insti-tute (1992).

8. T. García, C. Rodríguez, F. Belzunce et al. Estimation of the mechanical properties of me-tallic materials by means of the small punch test. Journal of Alloys and Compounds 582(0) (2014) 708 – 717.

9. E. Cardenas, F. Belzunce, C. Rodriguez et al. Application of the small punch test to de-termine the fracture toughness of metallic materials. Fatigue & Fracture of Engineering Materials & Structures 35 (2012) 441–450.

10. P. Egan, M. P. Whelan, F. Lakestani et al. Small punch test: An approach to solve the inverse problem by deformation shape and finite element optimization. Computational Materials Science 40(1) (2007) 33 – 39.

11. P. Egan, M. P. Whelan, F. Lakestani et al. Random depth access full-field low-coherence in-terferometry applied to a small punch test. Optics and Lasers in Engineering 45(4) (2007) 523 – 529. Optical Diagnostics and Monitoring: Advanced monitoring techniques and co-herent sources - volume C.

12. M. Abendroth. Identifikation elastoplastischer und schädigungsmechanischer Materialpa-rameter aus dem Small Punch Test. Ph.D. thesis, TU Bergakademie Freiberg (2004).

13. M. Abendroth and M. Kuna. Identification of ductile damage and fracture parameters from the small punch test using neural networks. Engineering Fracture Mechanics 73(6) (2006) 710 – 725.

14. P. Dymácek and K. Milicka. Creep small-punch testing and its numerical simulations.Ma-terials Science and Engineering A510–511 (2009) 444–449.

15. A. Gurson. Continuum theory of ductile rupture by void nucleation and growth: Part I –Yield criteria and flow rules for porous ductile materials. Journal of Engineering Materials and Technology 99 (1977) 2–15.

16. A. Gurson. Porous rigid-plastic materials containing rigid inclusions-yield function, plas-tic potential and void nucleation. Fracture 2 (1977) 357–364.

17. V. Tvergaard. Influence of Voids on Shear Band Instabilities under Plane Strain Conditions. International Journal of Fracture Mechanics 17 (1981) 389–407.

18. V. Tvergaard and A. Needleman. Effects of nonlocal damage in porous plastic solids. Inter-national Journal of Solids and Structures 32(8–9) (1995) 1063–1077.

304

19. U. Mühlich, W. Brocks and T. Siegmund. A user material subroutine of the Gurson-Tveer-gard-Needleman model of porous metal plasticity for rate and temperature dependent har-dening. Technical Note GKSS/WMG/98/1, GKSS-Forschungszentrum Geesthacht (1998).

20. Z. Zhang. On the accuracies of numerical integration algorithms for Gurson-based pressu-re dependent elastoplastic constitutive models. Computer methods in applied mechanics and engineering 121 (1995) 15–28.

21. Z. Zhang. Explicit consistent tangent moduli with a return mapping algorithm for pressure dependent elastoplasticity models. Computer Methods in Applied Mechanics and Engi-neering 121 (1995) 29–44.

22. T. Linse, G. Hütter and M. Kuna. Simulation of crack propagation using a gradient-enri-ched ductile damage model based on dilatational strain. Engineering Fracture Mechanics 95 (2012) 13–28.

23. T. Linse. Quantifizierung des spröd-duktilen Versagensverhaltens von Reaktorstählen mit Hilfe des Small-Punch-Tests und mikromechanischer Schädigungsmodelle. Ph.D. thesis, TU Bergakademie Freiberg, Berichte des Institutes für Mechanik und Fluiddynamik, Heft 9 (2013).

24. F. Reusch, B. Svendsen and D. Klingbeil. A non-local extension of Gurson-based ductile damage modeling. Computational Materials Science 26 (2003) 219–229.

25. M. Abendroth and M. Kuna. Determination of Deformation and Failure Properties of Ductile Materials by Means of the Small Punch Test and Neural Networks. Computational Materials Science 28(3–4) (2003) 633–644.

305

Atomistic modeling of defects in the framework of the modified embedded-atom method

S. Groh

Institute of Mechanics and Fluid Dynamics, TU Bergakademie Freiberg, 09599 Freiberg, Germany

306

AbstractThe full capacity of the modified embedded-atom method (MEAM) is presented in this review. It is demonstrated that once parameterized, MEAM can be applied in atomistic modeling to (i) reveal new mechanisms to be later validated or disputed by experimental observations and (ii) validate or disprove continuum theories. In particular, MEAM was applied to model the interactions between an edge dislocation and an impenetrable particle in Mg. The analysis of the MEAM-based calculations revealed a new mechanism involved during bypass. The net result obtained after the bypass of a particle with diameter between 2 nm and 8 nm by two glide dislocations consists of two prismatic loops on one side of the particle, and a glide dislocation carrying two super jogs and a prismatic loop on the other side of the particle. Quantitatively, the stress required by a first glide dislocation to bypass an impenetrable particle was in good agreement with continuum predictions from the literature. Furthermore, a systematic decrease of the shear stress needed for a second glide dislocation to bypass an impenetrable particle encircled by a non-planar Orowan loop was also found. Such a decrease of the bypassing stress is interpreted as the evidence that the bypass of the second dislocation by cross-slip maneuvers is a stress relief mechanism. MEAM was also applied to model dislocation core properties in Li in view of validating or disproving the continuum theory. A general procedure to parameterize several potentials was given. A qualitative and quantitative agreement between the dislocation half width and the Peierls stress predicted by atomistic calculations in the MEAM framework and continuum predictions was found.

Keyword

Dislocations; bypass mechanism; dislocation core properties; Peierls stress; Peierls-Nabarro model.

1 Introduction

As described by Binder (1991), computatio-nal methods are complementary to analytical theory and experimental studies. As such, computational methods can be used to pre-dict mechanisms (Groh, 2014) later confir-med by experimental observations, to com-plement experimental observations in view of analyzing an individual mechanism in more details (Groh et al., 2009; Bhatia et al., 2014), or to learn about the relation between a me-chanism and the mechanical response of the material (Groh, et al., 2010; Karewar et al., in preparation). With regards to analytical theo-ry, computational methods can be used to ve-rify the validity of an analytical theory (Alam and Groh, 2015) or to calibrate parameters needed by the analytical models (Groh et al. 2009a, and 2009b.) As example, in the theo-ry of dislocation nucleation from a crack tip proposed by Rice (1992), using a Peierls con-

cept it was demonstrated that the dislocation nucleation is driven by the square root of the unstable stacking fault energy. However, the characterization of the unstable stacking fault is not yet possible using experimental devices (Swygenhoven et al. 2004), and therefore to confirm or disprove Rice’s theory, one strategy would be to use computational methods, and in particular atomistic models.

In crystalline materials, the strength and ductility are governed by the properties of structural defects, such as point defects, dis-locations, surfaces and/or grain boundaries, cracks, inclusions and/or void. Atomistic me-thods can be applied to gain knowledge on the individual properties of these defects and to quantify the interaction between at least two of the previously mentioned defects (Liu and Groh, 2014). As example, once the dislocation

307

core properties are established at the atomic scale (Yasi et al., 2009; Groh et al., 2009c), scale-bridging strategies may be applied to transfer information from the lower length scale up to the engineering scale in order to predict the inelastic behavior of the material (Groh and Zbib, 2009a; Amodeo et al., 2011; Barton et al., 2011; Liu and Groh, 2013).

Large-scale atomistic simulations based on semi-empirical force field are useful to (i) re-veal new mechanisms to be later confirmed by experiments, (ii) validate or disprove the continuum theory, and (iii) identify material parameters significant for the development of continuum theories. Recent studies of the au-thor are used in the present work to exemplify the ability of the modified embedded-atom method in (i) revealing new mechanisms and (ii) validating or disproving a continuum theory. Unlike the first-principles method, atomistic calculations in the framework of the modified embedded-atom method rely on the parameterization of the model. The main equations of the modified embedded-atom method (MEAM) are reviewed in Sec-tion 2 with a focus on describing the model parameters. In Section 3, the accuracy and transferability of a parameterization of the MEAM potential for biocompatible materials using first-principles and experimental data is presented. Section 4 is dedicated to the iden-tification of a new mechanism for dislocation bypassing a particle, while the validation of a continuum theory using atomistic simulati-ons in the MEAM framework is reviewed in Section 5. Concluding remarks are given in Section 6.

2 Interatomic potential

The modified embedded-atom method pro-posed by Baskes (1992) was successfully ap-plied to metals and covalent materials (Baskes and Johnson, 1994; Ravelo and Baskes, 1997; Kuo and Clancy, 2005). The original frame-work, in which interactions only up to the first nearest neighbor were considered, was improved by Lee et al. (2001) to take into ac-count the interactions up to the second nea-

rest neighbors. The second nearest neighbor formulation of the MEAM (denoted as 2NN MEAM) model was then successfully applied to model single elements (Kim et al., 2006; Lee, 2007; Groh and Alam, 2015), binary systems (Kim et al, 2012; Jelinek et al., 2012; Groh, 2015), and ternary systems (Gao et al., 2013). A library of the available MEAM po-tentials for single elements and alloys is given in Lee et al. (2010). Some of these potentials were recently used to perform large-scale ato-mistic simulations to (i) identify void growth mechanisms (Groh et al., 2010) and analyze the interaction between crack and nanocavi-ties (Liu and Groh, 2014), (ii) reveal the na-ture of the screw <a> dislocation core struc-ture in Ti (Ghazisaeidi and Trinkle, 2012) and <c+a> dislocation in Ti (Naik, 2015), (iii) understand the nature of the ductile to brittle transition in the MEAM framework (Ko and Lee, 2014), and (iv) identify the mechanism of dislocation bypass in Mg-Al alloys (Liao et al., 2014; Groh, 2014; Arasu, 2015). For sim-plicity, both abbreviations MEAM and 2NN MEAM refer to the second nearest neighbors modified embedded-atom method.

In the 2NN MEAM, the total energy E of a system of atoms is approximated as the sum of the atomic energies, i.e.:

where φ(rij) is a pair potential. The embedding function is taken as:

where A is an adjustable parameter, Ec is the cohesive energy, and is a scaling parameter. Unlike the original embedded-atom method, in which the electron density is assumed to be spherically symmetric, MEAM assumes that the background electron density at a specific site is a function of angle dependent partial electron densities,

(3)

(1)

(2)

308

and

where t(h) are adjustable parameters, is the spherically symmetric partial electron densi-ty, and are the angular dependent partial electron density functions. The electron den-sities functions are given with an exponential decay of magnitude β(h), which are adjustable parameters of the model.

Therefore, for a single element, the 2NN MEAM framework involves 16 indepen-dent parameters. The parameters of the set are used to calibrate the equation of state, those of the set t(1), t(2), t(3), β(0), β(1), β(2) and β(3) are required to calibra-te the partial electron density, the set A is used to scale the embedding function, and the set Cmin, Cmax, rc, and dr controls the angular screening and the radial smoothing. These 16 parameters are calibrated in such a way that the model reproduces well-known mechani-cal, thermal, and physical material properties.

3 Parameterization of the force field for bio-materials in the MEAM framework

Magnesium (Mg) and its alloys became high-ly attractive due to their high strength, light-weight, good recyclability capability, biocom-patibility, biodegradability properties, and the presence of large amounts of magnesium in natural resources at low cost. This new family of alloys is expected to replace more traditio-nal materials such as stainless steel, aluminum alloys, and even titanium alloys in various fields ranging from automotive and aerospace (Mordike et al., 2001) to orthopedic (Li et al., 2008) industries. However, to be fully integra-ted as new technological materials, a few limi-tations of magnesium and its alloys must be mitigated. Mg alloys have a high anisotropy in plastic deformation resulting in poor form-ability and limited ductility at room tempera-ture, in addition to poor corrosion resistance. It has been demonstrated that addition of

alloying elements such as Al, Ca, Li, Zn, and Zr can improve the formability of Mg-based alloys by enhancing the activity of the <c+a> dislocation (Agnew et al., 2001), and can in-crease the resistance to corrosion of the mag-nesium alloys.

As first-principles calculations provide the most reliable information on atomic-scale materials properties, a large number of nume-rical studies using density functional theory (DFT) were carried out to (i) identify the ef-fect of Ca on the microscopic properties of di-lute Mg-Ca alloys (Chino et al., 2011; Zhang et al. 2014; Shang et al., 2014; Moitra et al., 2014) and (ii) determine new possible ordered Mg-Ca alloys with different concentrations of Ca (Yu et al., 2009; Zhou and Gong, 2012; Zhang et al. 2012; Mao et al., 2014). However, due to the limited number of atoms that can be con-sidered in density functional theory-based calculations, large-scale simulations are re-quired to predict/model the inelastic behavior of the alloys (either ordered or diluted alloys) and to identify mechanisms occurring during mechanical deformation. Important to note here is that the interatomic potential should be able to correctly reproduce various fun-damental physical, mechanical, and thermal properties of the relevant alloys. Thus, as a first step in the understanding of the inelas-tic behavior of new Mg-Ca compounds, one must correlate an interatomic functional form with experimental and ab-initio data to repro-duce mechanical, thermal, and physical pro-perties of the material. To develop the MgCa pair in the MEAM framework, one needs a MEAM potential for Mg and another MEAM potential for Ca. While Groh (2015) conside-red the existing Mg-MEAM potential of Kim et al. (2009) for the development of the MgCa pair, the validation and transferability of the Ca-MEAM potential parameterized by Groh is reviewed in Section 3.1. In Section 3.2, the results of the parameterization of the MgCa-MEAM potential of Groh (2015) are reviewed. The choice of developing the MgCa-MEAM potential with the Mg-MEAM potential of Kim et al. (2009) is driven by the possibili-ty of developing ternary potentials MgAlCa,

(4)

309

and MgLiCa using the existing MgAl-MEAM (Kim et al., 2009), and MgLi-MEAM (Kim et al., 2012) potentials, respectively.

3.1 Validation and transferability of the Ca-MEAM potential of Groh (2015)

The Ca properties reproduced and predicted with the Ca-MEAM potential are presented in Table 1. Lattice properties, elastic cons-tants, structural energy transitions between fcc and bcc, and between fcc and hcp crystal structures, as well as the vacancy formation energy were used to correlate the Ca-MEAM parameters. All these quantities are well re-produced in comparison to the published data obtained by density functional theory, experi-mental studies, or from the parameterization of a Ca embedded-atom method (EAM) po-tential proposed by Sheng et al. (2011) using the energy landscape methodology. As vali-dation and transferability of the Ca-MEAM potential, surface energies and planar defects (generalized stacking fault energy curves) were calculated at zero temperature. A good agreement of these quantities was found with published data (see Table 1).

Furthermore, dislocation core properties were predicted using the parameterized Ca-MEAM potential. Figure 1 shows the core structure of an edge dislocation in Ca after a minimization of the potential energy using a conjugate gra-dient relaxation algorithm. As expected in ac-cordance with the dislocation theory, the core of the edge dislocation is dissociated into two Shockley partials bounding an intrinsic sta-cking fault I2. A distance close to 11.2 nm se-parates the two Shockley partials. The genera-lized stacking fault energy curve determined using the methodology described in Jelinek et al. (2012) predicts an intrinsic stacking fault energy, , of 9.25 mJ·m−2 (see Table 1). The corresponding dissociation length compared well with the dissociation length of 13.1 nm predicted by linear isotropic elasticity (Hirth and Lothe, 1992).

The motion of an edge dislocation in static conditions was generated by applying a ri-gid displacement on the simulation cell’s top region in the direction of the Burgers vector. The application of shear deformation indu-ced elastic deformation of the crystal until a critical stress was reached, at which point the dislocation started to move. The stress was

Figure 1: Core structure of an edge dislocation optimized using the Ca-MEAM potential. Atoms shown in green and red have a local crystal structure of fcc and hcp, respectively (from Groh, 2015).

310

Table 1: Calculated physical properties of Ca using the present MEAM potential in comparison with experimental data and an EAM potential. (*: Quantities used to parameterize the potential).

[1] Andersson et al., 1994. [2] Stassis et al., 1983. [3] Heiroth et al., 1966. [4] Hearn et al., 1996. [5] Anderson et al., 1990.

Experiment or

ab-initio

EAM

(Sheng et al., 2011)

MEAM

(This work)

Lattice properties

a0 (Angstrom) 5.5884[1] 5.5884 5.58*

Ec (eV/atom) 1.84[1] 1.84 1.84*

Elastic constants

C11 (GPa) 27.6 [2] - 22.8 [3] 28.0 28.5*

C12 (GPa) 18.2 [2] - 16.0 [3] 18.0 17.5*

C44 (GPa) 16.3 [2] - 14.0 [3] 17.0 16.8*

Other structures ∆E (eV/atom)

fcc - hcp -0.004 - 0.012 [4] 0.003 0.004*

fcc - bcc 0.013 - 0.024 [4] 0.009 0.020*

fcc - sc – 0.395 0.486

Vacancy

Ef (eV) 1.12[4] 0.95 0.9*

Planar defects

γSF (mJ/m2) – 8.0 9.25

γSF (mJ/m2) – 51 91

Edge dislocation

Peierls stress (MPa) – – 8

Drag coefficient (Pa.s.K1) – – 5.0x10-8

Surfaces Average[4]

γS (110) (mJ/m2) 490 477 470

γS (100) (mJ/m2) 490 426 392

γS (111) (mJ/m2) 490 377 363

Temperature

Melting (K) 1115[5] 983 1055±5

311

calculated using the engineering definition, i.e., by adding the internal force per unit area on the simulation cell’s top region. The calcu-lations were performed using the simulation cell dimensions of 47.1 nm, 13.5 nm, and 2.0 nm along the Burgers vector, normal to the slip plane and dislocation line directions, res-pectively. Rigid displacements corresponding to strain increments of = 10−4 were applied on the top region of the cell. After each incre-ment, the crystallite was allowed to relax until the potential energy was at minimum. This sequence of steps was repeated until the total strain reached 0.4 %. The minimum potential energy was assumed to be reached when one of the following stopping criteria of the conju-gate gradient relaxation algorithm was satis-fied: (i) the change in energy was lower than 10−14 eV or (ii) the mean gradient per atom was lower than 10−14 eV/Å. Figure 2 shows the predicted strain–stress behavior. It was ob-served that the stress increased linearly with strain before the dislocation started to move, a point indicated by the change in the slope of the stress–strain curve. The initial slope cor-responds to the shear modulus, and it has a value of 16.1 GPa, which is in good agreement with C44 = 16.8 GPa as reported in Table 1. Once the Peierls stress was reached, the dislo-cation motion dissipates all the elastic energy.

As a consequence, the stress oscillates around a constant value, the Peierls stress. A Peierls stress around 8 MPa was predicted using the Ca-MEAM potential. It should be noted that the value of the Peierls stress is directly rela-ted to the magnitude of the unstable stacking fault energy predicted by the Ca-MEAM po-tential ( = 91 mJ·m2 as reported in Table 1). Due to unavailable ab-initio data, the unstable stacking fault was not included in the charac-terization of the potential, and it is, therefore, a prediction of the model.

3.2 Parameterization, validation, and trans-ferability of a MgCa-MEAM potential

The parameterization of the pair MgCa in the MEAM framework was performed by reproducing structural parameters, heat of formation, elastic constants, and transition energy between different phases of the com-pound MgCa with a reference structure B2. The MEAM potential was then used to pre-dict heat of formation and elastic constants by changing the Ca concentration in the MgCa compound. The MEAM-based predictions are reported in Tables 2 and 3 and compa-red with first-principles data. Apart from a discrepancy in MgCa2, one can verify that the MEAM-based predictions are in good agree-ment with the first-principles data.

Figure 2. Strain-stress behavior obtained at 0K that models the motion of edge dislocations in Ca (from Groh, 2015)

312

Furthermore, an analysis of the mechanical stability of the different phases was performed. While a cubic crystal is mechanically stable when C11 > 0; C11

2 > C122; and C44 > 0 (Cij being

the conventional elastic constants in Voigt

notation), a hexagonal crystal is mechanically stable when C11>0; C11

2 > C122; C33(C11+C12) > 2

C133; and C11C33 > C13

3. As reported in Table 3, one can conclude that all the elastic constants predicted in the 2NN MEAM framework lead

[1] Zhou and Gong, 2012. [2] Zhang et al., 2012. [3] Yu et al., 2009.

Table 2: Structural properties and bulk modulus of several ordered Mg-Ca phases. a is the lattice constant, ∆E is the structural energy difference, ∆H is the heat of formation, and B is the bulk modulus. (*: Quantities used to parameterize the potential).

Phase Structure Space group Type a

(Å)

c/a ∆E

(eV/atom)

∆H

(kJ/mol)

Mg3Ca DO3 Fm3m MEAM 7.494 0 -5.61 29.87

DFT [1] 7.48 -5.51 29.57

Mg2Ca C14 MEAM 6.22 1.631 0 -15.56 30.5

DFT[1] 6.23 1.620 0 -12.38 30.57

DFT[2] -12.02 30.30

DFT[3] 6.23 1.619 -12.06 31.06

C15 MEAM 8.80 – 0.0015 -15.41 30.89

DFT[1] 8.79 – 0.0038 -12.01 30.82

DFT[2] – -11.73 29.80

DFT[3] 8.79 – 0.003 -11.77 30.81

C36 MEAM 6.22 3.264 0.0007 -15.49 30.87

DFT[2] -11.97 30.20

DFT[3] 6.20 3.255 0.004 -11.67 32.14

MgCa* B2* MEAM 3.96* – – -8.87* 25.16*

DFT[1] 3.96 – – -8.91 25.26

MgCa2 C14 MEAM Not stable –

DFT[2] 29.82 14.70

C15 MEAM – 39.96 14.79

DFT[2] – 30.82 14.30

C36 MEAM 36.70 14.25

DFT[2] 30.30 14.60

MgCa3 L12 Pm3m MEAM 5.34 -0.54 22.12

DFT[1] 5.26 -1.46 20.18

B (GPa)

313

Table 3: Elastic constants of single crystals. All values are in GPa. (*: Quantities used to parameterize the potential).

[1] Zhou and Gong, 2012. [2] Mao et al., 2014. [3] Yu et al., 2009.

to mechanically stable structures. Again, such predictions made with the MEAM potentials are in agreement with first-principles data (Yu et al., 2009; Zhou and Gong, 2012; Mao et al., 2014).

The MgCa-MEAM potential was used to pre-dict the change of unstable and stable stacking fault energy on the basal and second order <c+a> pyramidal slip planes as a function of the Ca concentration in MgCa solid soluti-

on. It was found the that addition of Ca was altering the plastic anisotropy of pure Mg by increasing the lattice resistance to dislocation motion on the basal slip plane, while decre-asing the lattice resistance on the second or-der <c+a> pyramidal slip plane. For further details, the reader is referred to the work of Groh (2015), where the parameterization of the Ca-MEAM and MgCa-MEAM potentials is given.

4 Atomistic models as a predictive tool for revealing new mechanisms

As stated earlier in the introduction, atomis-tic models are considered useful for revealing new mechanisms that can latter be validated or disputed using experiments. In this sec-

tion, an overview of a new dislocation bypass mechanism revealed by atomistic modeling is presented. More details on this study can be found in Groh (2014).

Phase Structure Space group Type C11 C12 C13 C33 C44 Stability

Mg3Ca DO3 Fm3m MEAM 38.77 25.84 – – 24.77 yes

DFT[1] 37.77 25.47 – – 47.81

Mg2Ca C14 MEAM 54.41 20.85 18.83 55.03 14.35 yes

DFT[1] 58.41 17.22 15.45 62.05 17.52

DFT[2] 51.43 14.73 14.73 58.51 14.32

DFT[3] 69.04 17.07 13.80 65.9 17.95

C15 MEAM 49.63 22.07 – – 18.85 yes

DFT[1] 49.56 19.23 – – 23.38

DFT[3] 59.52 15.45 – – 22.03

C36 MEAM 54.38 20.80 18.77 55.67 14.81 yes

DFT[3] 59.66 16.45 19.42 59.39 20.31

MgCa* B2 MEAM 39.74* 18.59* – – 21.61* yes

DFT[1] 32.28 21.75 – – 24.94

MgCa3 L12 Pm3m MEAM 30.70 18.27 – – 16.78 yes

DFT[1] 25.10 17.72 – – 17.41

314

In magnesium single crystal deformed under large shear strain, Hirsch and Lally (1965) experimentally observed via transmission electron microscopy (TEM) analysis the pre-sence of a large number of prismatic loops ge-nerated in front of unidentified particles. The formation of the rows of prismatic loops was explained based on the bypassing mechanism proposed by Hirsch (1957), and later referred to as the Hirsch mechanism. The Hirsch me-chanism involves a series of cross-slip opera-tions that lead to the formation of a prismatic loop on one side of the precipitate, and two super-jogs carried away by the glide dislo-cation on the other side of the precipitate. In Cu-based alloys, Humphreys and Hirsch (1970) hypothesized a variant of the Hirsch mechanism. The mechanism variant invol-ves a pre-existing Orowan loop encircling the particle. Under the influence of the stress field of a glide dislocation, the Orowan loop transforms into two prismatic loops located on each side of the precipitate. After recombi-nation of the glide dislocation with the newly formed prismatic loops, the net result obtai-ned within this mechanism is that after the passage of two dislocations, a prismatic and an Orowan loop are left behind. The forma-tion of Orowan loops and their involvement in the bypassing mechanism is a valid hypo-thesis assuming that climb and pipe diffusion are two possible explanations for the disap-pearance of the Orowan loops between strain testing at low temperature and TEM analysis at room temperature (Hazzledine and Hirsch 1974). In addition, Vivas et al. (1997) observed the formation of dislocation loops induced by cross-slip as well as a few Orowan loops using the transmission electron microscopy in-situ straining technique in aluminum alloys. By means of full three-dimensional dislocation dynamics simulations based on the level set method, Xiang et al. (2003, 2004, and 2006) examined the bypassing mechanisms of par-ticles of different natures (penetrable and im-penetrable, with and without misfit). These authors reported classical and new bypassing mechanisms.

It should be noted, however, that none of the new mechanisms reported by Xiang et al. (2003, 2004, and 2006) have yet been confir-med experimentally or by means of atomistic calculations. Although the dislocation-par-ticle interaction has already been intensively studied at the atomistic scale (see Bacon et al. 2009 for a literature survey), very limited work is dedicated to the study of dislocation–impe-netrable particles, and to the author’s know-ledge, the works of Hatano (2006) and Provil-le and Bakó (2010) are the only ones available in the literature. Hatano (2006) revealed the dynamics of the Hirsch mechanism at finite temperature using molecular dynamics in fcc copper crystal modeled in the framework of the embedded-atom method. He concluded that the bypassing mechanism (either Oro-wan or Hirsch) depends on the nature of the boundary conditions: the Orowan bypassing mechanism being recovered when the shear is symmetrically applied with respect to the middle of the specimen, whereas the Hirsch mechanism being recovered when the sym-metry is broken by fixing the bottom of the specimen and applying a constant velocity on the top layers. Proville and Bakó (2010) reported certain calculations where a second glide dislocation interacted with a spherical precipitate encircled by an Orowan loop re-sulting from the bypass of the precipitate by a first glide dislocation. Their calculations were performed in fcc nickel reinforced by Ni3Al nanophases with L12 crystal structure modeled in the framework of the embedded-atom method. Although Proville and Bakó (2010) observed a decrease of the bypassing stress by a small percentage and two jogs were reported in the second passing dislocation, the nature of the dislocation left around the particle was not reported, and the authors did not discuss the bypassing mechanism.

We simulated by means of molecular statics (MS) calculations (Plimpton, 1995) the pro-cess of a dislocation bypassing impenetrable particles in the model of hexagonal close pa-cked crystal in the framework of the (modi-fied-) embedded atom method (Sun et al., 2006; Kim et al., 2009; Jelinek et al., 2012).

315

Figure 3 shows the stress-strain behavior cal-culated for four edge dislocations of different lengths interacting with a particle of constant diameter. From the analysis of Figure 3, it is noteworthy that the bypassing stress for the second dislocation is systematically lower than the bypassing stress for the first dislo-cation. Such a fact implies that a mechanism that does not involve pilling up of Orowan loops around the particle was involved during the deformation.

Filtering the atomic potential energy revealed the reaction that occurs when the second glide dislocation interacted with the particle encircled by a non-planar Orowan loop. In the snapshots reported in Figure 4(a, c, and e), only the atoms participating in the disloca-tion cores and in the stacking faults are visib-le. For simplicity, the corresponding reactions were sketched, and reported in Figure 4(b, d, and f). Once the second glide dislocation is in the vicinity of the particle, the image force re-sulting from the mismatch in shear modulus between the particle and the matrix acts as a

repulsive force on the second glide dislocation in addition to the repulsive force acting be-tween the glide dislocation and the non-pla-nar Orowan loop (see Figure 4(a-b)). As shown in Figure 3 the second glide dislocation was blocked at the particle between 1.2 % and 1.75 % of shear strain, resulting in a linear increase of the shear stress with respect to the shear strain. Under the action of the ap-plied load, the second dislocation bowed out around the particle encircled by the non-pla-nar Orowan loop. At 1.75 % of shear strain, a second stress peak was reached (see Figure 3). With the help of the stress field carried by the second glide dislocation, the screw parts of the non-planar Orowan loop (segments AB and CD in Figure 4(b)) glided symmetrically along two parallel prismatic planes until the obstacle stopped blocking the movement in the basal plane. Once the movement was al-lowed in the basal plane located at one of the poles of the particle, cross-slip of the screw dis-locations moving in the prismatic planes oc-curred by transformation into a lower energy core structure, i.e., by dissociating in the basal plane.

Figure 3: Representative stress-strain behavior obtained by MS calculations that model the dislocation-particle inter-action. The particle diameter and the dislocation length were constant and varied, respectively (from Groh, 2014).

316

Figure 4: Snapshots extracted from the simulation (a, c, and e) and their corresponding sketches (b, d, and f) illus-trating the bypassing mechanism by an edge dislocation of a particle encircled by a non-planar Orowan loop. The simulation snapshots were created by filtering the potential energy and retaining only the atoms participating to the dislocation cores and stacking faults. (a and b) the particle and pre-existing Orowan loop repulse the glide dislocation. (c and d) the screw components AB and CD double cross-slip to form the prismatic loops GOMN and IJKL. (e and f) the screw components of the segments EF and GH double cross-slip to form the prismatic loop RGEP and a doubly jogged glide dislocation (from Groh, 2014).

(f)(e)

(d)(c)

(b)(a)

317

The screw parts moving in the basal plane an-nihilated each other, and two prismatic loops referred as IJKL and GOMN in Figure 4(d) were created on both sides of the particle. In the alternative bypassing mechanism repor-ted by Humphreys and Hirsch (1970), the dislocation segments EF, LI, HG, and GN re-presented in Figure 4(d) recombined to form an Orowan loop around the particle. Unlike what was hypothesized by Humphreys and Hirsch (1970), our calculations revealed that the screw part of the dislocation curves EF and GH represented in Figure 4(d) cross-slip-ped into the prismatic plane before cross-slip-ping back into a basal plane once the particle was not an obstacle against the movement in the basal plane. Once the operation of double cross-slip occurred, annihilation of the screw components was possible. Such a series of cross-slips lead to the formation of the second prismatic loop on the right hand side of the particle (represented as EPRG in Figure 4(f)) and to the formation of a prismatic loop inter-acting with a doubly jogged glide dislocation on the left hand side of the particle (see Figure 4(f)).

It should be noted that the bypassing mecha-nism revealed by means of molecular statics calculations for particle diameters between 2 nm and 8 nm, and for inter-particle distan-ces between 47 nm and 53 nm, appeared to be independent of the nature of the interatomic potential (Sun et al., 2006; Kim et al., 2009; Je-linek et al., 2012). Based on the proposed me-chanisms, new discretization rules for discre-te dislocation dynamics simulation have been proposed. Moreover, the effect of the stress-relief mechanism proposed by Groh (2014) on the macroscopic deformation of particle hardened Mg-alloys will also be investigated.

Finally, from a qualitative point of view, a new variant of the Hirsch mechanism was revealed by MS calculations. The mechanism by which two consecutive glide dislocations of edge character bypass an array of impenetrable particles can be summarized as follows:

• A first glide dislocation bypasses an im-penetrable particle according to the Oro-wan mechanism leading to a non-planar Orowan loop and a glide dislocation.

• The second glide dislocation triggers double cross-slip of the screw part of the non-planar Orowan loop leading to the formation of two prismatic loops on each side of the precipitate. The screw component of the glide dislocation per-forms a double cross-slip in the opposite direction leading to the formation of a new prismatic loop and a doubly jogged glide dislocation carrying away a prisma-tic loop formed by the bypass of the first glide dislocation.

5 Validation of continuum theory using atomistic method

In addition to enabling the identification of new mechanisms, atomistic models can also be used to validate or disprove continuum theories. Several attempts were made in this direction. Zhou et al. (1994) performed ato-mistic calculations using a force law derived from the universal binding-energy relation to evaluate the dislocation core structure. In ag-reement with a Peierls-Nabarro (PN) model, their data revealed the dependence of the dis-location core structure on the Peierls stress. However, their data were overestimated by nearly four orders of magnitude compared to the predictions obtained by a PN model. Lu et al. (2000a) examined the relation between the dislocation core properties (energetics, core width, and Peierls stress) and the dislo-cation character in Al using a semidiscrete variational Peierls-Nabarro model (Lu et al., 2000b). These authors used the generalized stacking fault energy curves obtained either in the framework of the density functional theory (DFT) or in the embedded-atom me-thod (Ercolessi and Adams, 1994) to model the interfacial restoring stress in the PN mo-del. A perfect match was found between their predictions of the Peierls stress for the screw and 60° dislocation and the data reported by Bulatov et al. (1999) by direct atomistic calcu-lation using the same EAM model.

318

To demonstrate that MEAM can be used to validate or disprove continuum theories, four potentials were parameterized to reproduce identical physical, thermal, and mechanical properties, with the exception of the unstable stacking fault that is varied around a reference value predicted by density functional theory. Dislocation core properties are then calcula-ted by direct atomistic calculations, and com-pared to the dislocation-based continuum predictions. As a representative material, li-thium (Li) was considered owing to its high interest in industrial applications, especially as an alloying element in Mg alloys (Uraka-mi and Fine, 1971; Moitra et al., 2014; Zhang et al., 2014). The procedure to parameterize four Li potentials in the MEAM framework is reviewed in Section 5.1. While these four Li-MEAM potentials were used to predict dislo-cation core properties, a comparison between atomistic and continuum predictions is pre-sented in Section 5.2 with the objective of va-lidating or disproving the continuum model. In this study, the 1D Peierls-Nabarro (PN) model reviewed by Bulatov and Cai (2006) was considered.

5.1 Parameterization of the four Li-MEAM potentials

In the MEAM framework, the effect of each parameter on individual properties is com-plex, and unless very specific, it is impossible to relate one property to a single parameter alone. The effects of certain parameters are, however, confined to only a few properties and, therefore, the evaluation of the parame-

ters can be done systematically. Starting with the Li-MEAM of Kim et al. (2012), a paramet-ric study was performed to identify the corre-lation between the MEAM parameters and the physical and mechanical properties. Knowing the relation between the MEAM parameters and the Li properties forms the starting point of parameterizing four Li-MEAM potentials that fulfill the two following conditions:

i. The elastic constants (C11, C12, and C44), the structural energy transition between fcc and bcc crystal structures (∆Efcc-bcc) and between bcc and hcp crystal structures (∆Ebcc-hcp), and the bulk modulus (B) are unchanged using the four potentials and reproduce the experimental data.

ii. The magnitude of the unstable sta-cking fault, U, on the plane (110) along the direction-<111> is varied around the density functional theory-based prediction.

Using these two conditions completed by the relation between MEAM parameters and physical properties obtained from the para-metric study, the MEAM parameters were adjusted as follows. In a first step, ß0 was va-ried to change the magnitude of the unstable sacking fault energy (U). As a consequence, the elastic constants (C11, C12, and C44), the transition energies between bcc and fcc and between bcc and hcp crystal structures (∆Efcc-bcc and ∆Ebcc-hcp), the bulk modulus (B), and the vacancy formation energy (Evf) were

Table 4: Summarized procedure to parameterize four Li-MEAM potentials.

Step MEAM parameters Controlled properties Uncontrolled properties

1 β0 U C11, C12, C44, ∆Efcc-bcc,∆Ebcc-hcp, Evf, B

2 α B C11, C12, C44, ∆Efcc-bcc,∆Ebcc-hcp, Evf

3 A ∆Efcc-bcc C11, C12, C44,∆Ebcc-hcp, Evf

4 t3 ∆Ebcc-hcp C11, C12, C44, Evf

5 t2 and β2 C44 C11, C12, Evf

6 Cmin and Cmax C11, C12 Evf

319

varied. Adjusting the bulk modulus (B) to the experimental data was performed by scaling the parameter α according to (Bexperimental / Bold)

1/2αold where αold and Bold are the unscaled values of α and of the cor-responding bulk modulus B, respectively. A consequence of rescaling the parameter α is the alteration of all other physical and mechanical properties in MEAM. Howe-ver, modifying the elastic constants (C11 and C12), the energy transition between bcc and hcp crystal structures (∆Ebcc-hcp), the vacancy formation energy (Evf), and the energy tran-sition between bcc and fcc crystal structures (∆Efcc-bcc) was corrected to reproduce the DFT value by scaling the embedding function pa-rameter A. Controlling the structural energy transition between bcc and hcp crystal struc-tures (∆Ebcc-hcp) by varying the t3 parameter re-sulted in increase/decrease of the vacancy for-mation energy (Evf). The shear elastic constant (C44) was correlated to the experimental data by varying t2 and ß2. The above procedure was repeated by iterating over Cmin and Cmax until the elastic constants (C11 and C12) and discon-tinuities on the (110) generalized stacking fault energy along the direction-<111> were removed. As the vacancy formation energy is not expected to affect the dislocation be-havior under static conditions, this property was not correlated to experimental values.

However, as the MEAM parameter t1 was not used in the above procedure, t1 can still be used as an adjustment parameter to charac-terize the vacancy formation energy without affecting the elastic constants, the bulk mo-dulus, the transition energies, or the unstab-le stacking fault energy. Table 4 summarizes the 6 steps to follow for parameterizing the MEAM potentials for the same element while fulfilling the two conditions presented above. As a result of the parameterization of the four Li-MEAM potentials, apart from the mag-nitude of the unstable stacking fault energy, all other physical and mechanical properties included in the materials database and used to correlate the MEAM parameters are re-produced within an error of maximum 0.7% compared to each other. In addition, a good agreement of those properties was found in comparison to the published data obtained via the density functional theory, experimen-tal data, or from the parameterization of Li potentials performed by others. Moreover, the magnitude of the unstable stacking fault varies between 4.76 meV/A2 and 6.15 meV/A2 as expected by the procedure described above. As a validation and transferability of the Li-MEAM potentials, the point defect, the surface energies, the elastic constants in the fcc crystal structure, the structural ener-gy transition between bcc and simple cubic

Figure 5: Relative horizontal displacement versus horizon-tal position obtained for an isolated dislocation modeled with the four Li-MEAM potentials reported in Tables 3 and 4 (from Alam and Groh, 2015).

320

crystal structures, and the derivative of the bulk modulus with respect to the pressure were calculated. It was found that these pro-perties predicted using MEAM are in good agreement with the experimental data. The elastic constant obtained for both the bcc and the fcc crystal structures leads to mechanical stability of the structures. Finally, the stability of the four Li-MEAM models was tested un-der dynamics conditions, and the coefficient of volumetric thermal expansion, the heat capacity at constant volume and the melting temperature are all in good agreement with experimental data. For further details on the parameterization of the four Li potentials in the MEAM framework, the reader is referred to the work of Alam and Groh (2015).

5.2 Properties of dislocation core in bcc lithium: Atomistic vs. continuum

To quantitatively analyze the dislocation core structure obtained by energy minimization in the MEAM framework, the disregistery pro-file across the cut plane, u(x), was extracted from the atomic structure. Defining u+ and u- as the relative displacements of the dislo-cation body compared to the perfect crystal above and below the slip plane, respectively, the disregistery can be calculated from the atomic location using the equation:

(5)

(6)

(7)

Figure 6: Evolution of the dis-location half width as a func-tion of the unstable stacking fault energy. The dislocation half width is derived (i) by correlating the atomistic data with Eq. 6 (strategy 1), and (ii) from the distance for which the disregistery is included between 0.25b and 0.75b (strategy 2).

The normalized disregistery, u(xi)/b, as a func-tion of the atomic locations obtained with the four Li-MEAM potentials is plotted in Figure 5. One can observe that independently of the Li-MEAM potentials, the evolution of the dis-registery given in Figure 5 is the signature of a narrow core structure.

The PN model provides an analytical solution for the disregistery, from which both the dis-location core structure and the Peierls stress can be derived. In 1D, Peierls proposed an analytical solution for the disregistery func-tion in the form:

where ξ, and b are the dislocation half width and the magnitude of the Burgers vector, res-pectively. Furthermore, if a sinusoid function of magnitude U and period 2π/b is considered to describe the variation of the generalized stacking fault energy curve, one can recover that the dislocation half width, ξ, and the ma-gnitude of the unstable stacking fault energy, U, are linked to each other according to the equation:

321

The dislocation half width, ξ, can be extrac-ted from the atomistic disregistery obtained with the four Li-MEAM potentials and com-pared to the analytical evolution given by Eq. 7. In the study of Alam and Groh (2015), two strategies were employed to identify the dislocation half width from the atomic disre-gistery. The first strategy was to identify the dislocation half width by correlating the ato-mic disregistery plotted in Figure 5 with the analytical solution given by Eq. 6. The second strategy was to define the dislocation width as the distance for which the disregistery is

predictions obtained from Eq. 7. However, when derived from the first strategy, the dis-location half width is systematically undere-stimated by about 25 % in comparison to the prediction obtained using Eq. 7. Such a diffe-rence is explained by the fact that the analyti-cal solution given by Eq. 6 was obtained in 1D, while atomic locations were allowed to relax in 3D during the minimization of the disloca-tion core using atomistic modeling.

The Peierls stress can be estimated from the atomistic method by increasing the external load until the dislocation starts to move. By performing such numerical experiments, it was observed that the stress increased line-arly with strain before the dislocation started

Figure 7: Peierls stress evo-lution as a function of the unstable stacking fault ener-gy. The dislocation half width was derived (i) by correlating the atomistic data with Eq. 6 (strategy 1), and (ii) from the distance for which the dis-registery is included between 0.25b and 0.75b (strategy 2).

included between 0.25b and 0.75b (Lu et al., 2000a). The evolution of the dislocation half width, ξ, as a function of the magnitude of the unstable stacking fault energy, U, is plotted in Figure 6. As a general trend, the decrease of the dislocation half width with increasing the magnitude of the unstable stacking fault energy predicted using Eq. 7 is reproduced in-dependently of the strategy applied to extract the dislocation half width from the atomistic disregistery. Furthermore, when derived from the second strategy, the dislocation half width is in good quantitative agreement with the

to move, a point indicated by the change in the slope of the stress–strain curve. The ini-tial slopes correspond to the shear moduli, and have values of 4.46 GPa, 4.47 Gpa, 4.46 Gpa, and 4.45 Gpa obtained with Li1, Li2, Li3, and Li4, respectively. These values are in good agreement with the shear moduli calculated using a simulation cell without dislocation and having the same orientations. Once the Peierls stress was reached, the dislocation motion dissipated the elastic energy applied to the system resulting in an oscillatory stress behavior around a constant value. Peierls stresses of 4.2x10-4 µ, 7.2x10-4 µ, 2.8x10-3 µ, and 4.3x10-3 µ are predicted with Li1, Li2, Li3, and Li4, respectively, where µ is the shear mo-dulus. A sensitivity of the simulation cell’s di-

322

mension on the Peierls stress was performed, and the above values are expected to be size independent.

From the review of the PN model by Joós and Duesberry (1997), the Peierls stress is expo-nentially proportional to the dislocation half width according to the equation:

Although approximate, it is admitted that the prediction of the Peierls stress using Eq. 8 is in agreement with experimental and atomis-tic calculations (Joós and Duesberry, 1997). Figure 7 shows the evolution of the Peierls stress as a function of the unstable stacking fault energy obtained by atomistic calcula-tions with the four Li-MEAM potentials, in comparison to the prediction obtained using Eq. 8. The Peierls stress predicted by Eq. 8 and plotted in Figure 7 was obtained with the dis-location half width derived from the two stra-tegies described above and from Eq. 7. The shear moduli and the Poisson coefficient for the four Li-MEAM potentials were characte-rized from direct atomistic calculations. As a general trend, independently of the method applied to identify the dislocation half width, the increase of the Peierls stress while incre-asing the magnitude of the unstable stacking fault energy is recovered. Moreover, except for the Peierls stress prediction obtained with the dislocation half width characterized using the first strategy, the atomistic predictions are quantitatively in good agreement with all the other predictions.

As a general result of this study, we demons-trated that the parameters of the MEAM framework can be adjusted in such a way that allows testing of a continuum theory. The four Li-MEAM potentials were used to predict the (111) generalized stacking fault energy curves along the direction-<112> in fcc Li. Although not shown in this review pa-per, while the four Li-MEAM potential pre-dict the same magnitude of the intrinsic sta-cking fault, I2, they do not predict the same

magnitude of the unstable stacking fault on the plane (111) along the direction-<112>. Therefore, as the four Li-MEAM predict dif-ferent energy barriers, us - s, where us and s are the magnitudes of the unstable and stable stacking fault energy on the plane (111) along the direction-<112> and elastic constant for fcc-Li, the four Li-MEAM potentials can be used to initiate numerical validation or in-validation of the Rice theory for dislocation nucleation from a crack tip.

6 Conclusion

In this manuscript, the full capability of the MEAM framework was presented. It was de-monstrated that once parameterized, MEAM can be applied in atomistic modeling (i) to reveal new mechanisms later to be validated or disputed by experimental observations and (ii) to validate or disprove continuum theories. In particular, MEAM was applied to model the interaction between an edge dis-location and an impenetrable particle in Mg. The analysis of the calculations revealed a new mechanism involved during the bypass. The net result obtained after the bypass of a partic-le with diameter between 2 nm and 8 nm by two glide dislocations is two prismatic loops on one side of the particle, and a glide dislo-cation carrying two super jogs and a prismatic loop on the other side of the particle. Quan-titatively, the stress required by a first glide dislocation to bypass an impenetrable particle was in good agreement with continuum pre-dictions from the literature. Furthermore, a systematic decrease of the shear stress needed for a second glide dislocation to bypass an im-penetrable particle encircled by a non-planar Orowan loop was found. Such a decrease of the bypassing stress is interpreted as the evi-dence that the bypass of the second disloca-tion by cross-slip maneuvers is a stress relief mechanism. Moreover, MEAM was applied to model dislocation core properties in Li in view of validating or disproving the continu-um theory prediction. A general procedure to parameterize several potentials was proposed.

(8)

γ γ γ γ

323

References

1. Agnew, S.R., Yoo, M.H., and Tomé, C.N., 2001. Application of texture simulation to un-derstand mechanical behavior of Mg and solid solution alloys containing Li and Y. Acta Mater. 49 4277-4289.

2. Alam, M., and Groh, S., Dislocation modeling in bcc lithium: A comparison between ato-mistic and continuum predictions. J. Physics and Chemistry of Solids 82 1-9.

3. Amodeo, J., Carrez, Ph., Devincre, D., and Cordier, P., 2011. Multiscale modeling of MgO plasticity. Acta mater. 50 2291-2301.

4. Anderson, M.S., Swenson, C.A., and Peterson, D.T., 1990. Experimental equations of state for calcium, strontium, and barium metals to 20kbar from 4 to 295 K. Phys. Rev. B 41 3329.

5. Andersson, K.M., Johnston, R.L., and Murrell, J.N., 1994. Empirical potential-energy func-tion for calcium solids and cluster. Phys. Rev. B 49 3089.

6. Arasu, N.P., Atomistic modeling of dislocation/precipitate interaction in magnesium al-loys, Master’s thesis, TU Bergakademie Freiberg, February 2015.

7. Bacon, D.J., Osetsky, Y.N., and Rodney, D., 2009. Dislocation-obstacle interaction at the atomic level. Dislocation in Solids (Elsevier, 2009).

8. Barton, N.R., Bernier, J.V., Becker, R., et al., 2011. A multiscale strength model for extreme loading conditions. Journal of Applied Physics 109 073501.

9. Baskes, M., 1992. Modified embedded-atom potentials for cubic materials and impurities. Phys. Rev. B 46 2727 – 2742.

10. Baskes, M., and Johnson, R.A., 1994. Modified embedded-atom potentials for HCP metals. Modelling Simul. Mater. Sci. Eng. 2 147-163.

11. Bhatia, M.A., Groh, S., and Solani, K.N., 2014. Atomic-scale investigation of point defects and hydrogen-solute atmospheres on the edge dislocation mobility in alpha iron. Journal of Applied Physics 116 064302.

12. Binder, K., 1991. The Monte Carlo method in Condensed matter physics, Volume 71 of Topics in Applied Physics, Springer-Verlag.

13. Bulatov V.V., and Cai, W., 2006. Computer simulations of dislocations, Oxford: Oxford University Press.

14. Bulatov, V.V., Richmond, O., and Glazov, M.V., 1999. An atomistic dislocation mechanism of pressure-dependent plastic flow in aluminum. Acta Metall. 47 3507.

A qualitative and quantitative agreement bet-ween the dislocation half width and the Pei-erls stress predicted by atomistic calculations in the MEAM framework and continuum prediction was found.

Acknowledgements

The study was performed within the Clus-ter of Excellence “Structure Design of Novel High Performance Materials via Atomic De-sign and Defect Engineering (ADDE)” that is financially supported by the European Union and by the Ministry of Science and Art of Sa-xony (SMWK).

324

15. Chino, Y., Ueda, T., Kado, M., and Mabuchi M., 2011. Solid solution softening mechanisms in Mg-Ca alloy. Materials Transactions 52 1840-1843.

16. Ercolessi, F., and Adams, J. B., 1994. Interatomic potential from first-principles calculati-ons: The force-matching method. Europhys. Lett. 26 583.

17. Gao, H., Otero-de-la-Roza, A., Aouadi, S.M., Johnson, E.R., and Martini, A., 2013. An em-pirical model for silver tantalate. Modelling Simul. Mater. Sci. Eng. 21 055002.

18. Ghazisaeidi, M., and Trinkle, D. R., 2012. Core structure of a screw dislocation in Ti from density functional theory and classical potentials. Acta Mater. 60 1287-1292.

19. Groh, S., 2014. Transformation of shear loop into prismatic loops during bypass of an array of impenetrable particles by edge dislocation. Mater. Sci. Eng. A 618 29-36.

20. Groh, S., 2015. Mechanical, thermal, and physical properties of Mg-Ca compounds in the framework of the modified embedded-atom method. Journal of the Mechanical Behavior of Biomedical Materials 42 88-99.

21. Groh, S., and Alam,. M., 2015. Fracture behavior of lithium single crystal in the framework of (semi-)empirical force field derived from first-principles. Modelling Simul. Mater. Sci. Eng. 23 045008.

22. Groh, S., Marin, E.B., and Horstemeyer, H.F., 2010. Nanoscale void growth in magnesium: A molecular dynamics study. Int. J. Applied Mechanics 2 191-205.

23. Groh, S., Marin, E.B., Horstemeyer, M.F., and Bamman, D.J., 2009c.Dislocation motion in magnesium: a study by molecular statics and molecular dynamics Modelling Simul. Mater. Sci. Eng. 17 075009.

24. Groh, S., Marin, E.B., Horstemeyer, M.F., and Zbib, D.J., 2009b. Multiscale modeling of the plasticity in an aluminum single crystal. Int. J. Plasticity 25 1456-1473.

25. Groh, S., and Zbib, H.M., 2009a. Advances in discrete dislocations dynamics and multisca-le modeling. Journal of Eng. Mater. Technology 131 041209-1.

26. Hatano, T., 2006. Dynamics of a dislocation bypassing an impenetrable precipitate: The Hirsch mechanism revisited. Phys. Rev. B 74 020102(R).

27. Hazzledine, P.M., and Hirsch, P.B., 1974. A coplanar Orowan loops model for dispersion hardening. Philos. Mag. 30 1331-1351.

28. Hearn, J.E., Johnson, R.L., Leoni, S.L., and Murrel, J.N., 1996. Global potentials for calcium and strontium solids. J. Chem. Soc., Faraday Trans. 92 425.

29. Heiroth, M., Buchenau, U., Schober, H.R., and Evers, J., 1986. Lattice dynamics of fcc and bcc calcium. Phys. Rev. B 34 6681.

30. Hirsch, P.B., 1957. The interpretation of the slip pattern in terms of dislocation movements. J. Inst. Met. 86 13-14.

31. Hirsch, P.B., and Lally, J.S., 1965. The deformation of magnesium single crystal. Philos. Mag. 12 595-648.

32. Hirth, J., Lothe, J., 1992. Theory of Dislocations, second ed. Wiley, New York. 33. Humphreys, F.J., and Hirsch, B., 1970. The deformation of single crystal of copper and

copper-zinc alloys containing alumina particles II. Microstructure and dislocation-particle interaction. Proc. Roy. Soc. Lond. A 318 73-92.

34. Jelinek, B., Groh, S., Horstemeyer, et al., 2012. Modified embedded-atom method potential for Al, Si, Mg, Cu, and Fe alloys, Phys. Rev. B 85 245102.

35. Joós, B., and Duesberry, M.S., 1997. The Peierls stress of dislocation: An analytical formula. Phys. Rev. Lett. 78 266-269.

36. Karewar, S.V., Martinez, E., Groh, S., Caro, A., and Srinivasan, S.G, 2015. Atomistic simula-tions of deformation of nanocrystalline Mg-Li alloys, Proceedings of the TMS conference, Orlando (USA).

37. Kim, Y.-M., Jung, I.-H., and Lee, B.-J., 2012. Atomistic modeling of pure Li and Mg–Li system. Modelling Simul. Mater. Sci. Eng. 20 035005.

325

38. Kim, Y.M., Kim, N.J., and Lee, B.J., 2009. Atomistic modeling of pure Mg and Mg-Al sys-tems. Calphad 33 650-657.

39. Kim, Y.-M., Lee, B.-J., and Baskes, M.I., 2006. Modified embedded-atom method interato-mic potentials for Ti and Zr. Phys. Rev. B 74 014101.

40. Ko, W.-S., and Lee, B.-J., 2014. Origin of unrealistic blunting during atomistic fracture simulations based on MEAM potentials . Philosophical Magazine 94 1745–1753.

41. Kuo, C.-L., and Clancy, P., 2005. Development of atomistic MEAM potentials for silicon-oxygen-gold ternary system. Modelling Simul. Mater. Sci. Eng. 13 1309–1329.

42. Lee, B.-J., 2007. A modified embedded-atom method interatomic potential for silicon. Calphad 31 95–104.

43. Lee, B.J., Baskes, M.I., Kim, H., and Cho Y.K, 2001. Second nearest-neighbor modified embedded atom method potentials for bcc transition metals. Phys. Rev. B 64 184102.

44. Lee, B.-J., Ko, W.-S., Kim, H.-K., and Kim, E.-H., 2010. The modified embedded-atom method interatomic potentials and recent progress in atomistic simulations. Calphad 34 510-522.

45. Li, Z., Gu, X., Lou, S., and Zheng, Y., 2008. The development of binary Mg-Ca alloys for use as biodegradable materials within bone. Biomaterials 29 1329-1344.

46. Liao, M., Li, B., and Horstemeyer, M.F., 2014. Interaction between basal slip and a Mg17Al12 precipitate in magnesium. Metallurgical and Materials Transactions A 45A 3661-3665.

47. Liu, T., and Groh, S., 2013. Influence of the anisotropy in lattice friction on the macro-scopic behavior of magnesium single crystal: A study by discrete dislocation dynamics. Computational Materials Science 79, 772-780.

48. Liu, T., and Groh, S., 2014. Atomistic modeling of the crack-void interaction in α-Fe. Mat. Sci. Eng. A 609 255-265.

49. Lu ,G., Kioussis, N., Bulatov, V.V, and Kaxiras, E., 2000a. Generalized-stacking-fault energy surface and dislocation properties of aluminum. Phys. Rev. B. 62 3099.

50. Lu, G., Kioussis, N., Bulatov, V.V, and Kaxiras, E., 2000b. The Peierls-Nabarro model revi-sited. Phil. Mag. Lett. 80 675-682.

51. Mao, P., Yu, B., Liu, Z., Wang, F., and Ju, Y., 2014. Mechanical, electronic and thermody-namic properties of Mg2Ca Laves phase under high pressure. Computational Materials Science 88 61-70.

52. Moitra, M., Kim, S.-G., and Horstemeyer, M.F., 2014. Solute effect on the <a+c> disloca-tion nucleation mechanism in magnesium. Acta Mater. 75 106-112.

53. Mordike, B.L., and Ebert, T., 2001. Magnesium: Properties – applications – potential. Ma-ter. Sci. Eng. 302 37-45.

54. Naik., V.S., Atomistic modeling of <c+a> dislocations in Ti, Master thesis, TU Bergakade-mie Freiberg, March 2015.

55. Plimpton, S.J., 1995. Fast parallel algorithms for short-range molecular dynamics. J. Com-put. Phys. 117 1-19.

56. Proville, L., and Bakó, B., 2010. Dislocation depinning from ordered nanphases in a model fcc crystal: From cutting mechanism to Orowan looping. Acta Mater. 58 5565–5571.

57. Ravelo, R., and Baskes, M., 1997. Equilibrium and thermodynamics properties of grey, white, and liquid Tin. Phys. Rev. Lett. 79 2482 – 2485.

58. Rice, J., 1992. Dislocation nucleation from a crack top: An analysis based on the Peierls concept. J. Mech. Phys. Solids 40 239-271.

59. Shang, S.L., Wang, W.Y., Zhou, B.C., et al., 2014. Generalized stacking fault energy, ideal strength and twinnability of dilute Mg-based alloys: A first-principles study of shear defor-mation. Acta Mater. 67 168-180.

60. Sheng, H.W., Kramer, M.J., Cadien, A., Fujita, T., and Chen, M.W., 2011. Highly optimized embedded-atom-method potential for fourteen fcc metals. Phys. Rev. B 83 134118.

326

61. Stassis, C., Zaretsky, J., Misemer, D.K., Skriver, H.L., Harmon B.N., and Nicklow, R.M., 1983. Lattice dynamics of fcc Ca. Phys. Rev. B 27 3303.

62. Sun, D.Y., Mendelev, M.I., Becker, C.A., et al., 2006. Crystal-melt interfacial free energy in hcp metals: A molecular study of Mg. Phys. Rev. B 74 024116.

63. Swygenhoven van, H., Derlet, M.M., and Frøseth, A.G., 2004. Stacking fault energies and slip in nanocrystalline metals. Nature Materials 3, 399-403.

64. Urakami, A., and Fine, M.E., 1971. Yield drop effects for basal and prismatic slip in Mg-Li alloy single crystals. Acta Metall. 19 887-894.

65. Vivas, M., Lours, P., Levaillant, et al., 1997. Determination of precipitate strength in alumi-nium alloy 6056-T6 from transmission electron microscopy in situ straining data. Philos. Mag. A 76 921-931.

66. Xian, Y., Cheng, L.-T., Srolovitz, D.J., and Weinan, E., 2003. A level set method for disloca-tion dynamics. Acta Mater. 51 5499-5518.

67. Xian, Y., and Srolovitz, D.J., 2006. Dislocation climb effects on particle bypass mechanisms. Philos. Mag. 86 3937-3957.

68. Xian, Y., Srolovitz, D.J., Cheng, L.-T., and Weinan, E., 2004. Level set simulation of disloca-tion-particle bypass mechanism. Acta Mater. 52 1745-1760.

69. Yasi, J.A., Nogaret, T., and Trinkle, D.R., et al. 2009. Basal and prism dislocation cores in magnesium: comparison of first-principles and embedded-atom-potential methods pre-dictions. Modelling Simul. Mater. Sci. Eng. 17 055012.

70. Yu, W.-Y., Wang, N., Xiao, X.-B., et al., 2009. First-principles investigation of the binary AB2 type Laves phase in Mg-Al-Ca alloy: Electronic structure and elastic properties. Solid State Sci. 11 1400-1407.

71. Zhang, J., Dou, D., and Dong, H., 2014. Intrinsic ductility of Mg-based binary alloys: A first-principles study, Scripta Mater. 89 13-16.

72. Zhang, H., Shang, S.-L., Wang, Y., et al., 2012. Thermodynamic properties of Laves phases in the Mg-Al-Ca system at finite temperature location core-core interaction and Peierls stress in a model hexagonal lattice. Phys. Rev. B. 49 6451-6456.

73. Zhou, P., and Gong, H.R., 2012. Phase stability, mechanical properties, and electronic structure of an Mg-Ca system. Journal of the Mechanical behavior of biomedical materials 8 154-164.

327

Thermodynamic investigations in the ternary Al-Ti-Cr system

Mario J. Kriegel 1, Olga Fabrichnaya 1, Dietrich Heger 1, David Chmelik 1, David Rafaja 1, Hans J. Seifert 2

1 Institute of Materials Science, TU Bergakademie Freiberg, Germany2 Institute for Applied Materials (IAM-AWP), Karlsruhe Institute of Technology, Germany

328

Abstract

1 Introduction

The partial vertical section at a constant Ti content of 33 at.% was constructed based on themetallographic investigations of three alloy compositions 53Al-33Ti-14Cr (#1), 47Al-33Ti-20Cr (#2) and 43Al-33Ti-24Cr (#3) annealed at 1293 and 1423 K. The investigationscombined X-ray diffraction, scanning electron microscopy, energy-dispersive X-rayspectroscopy and differential scanning calorimetry. The observations showed that theinvariant reaction γ-TiAl + β ↔ τ + C14 takes place at 1354 K, that a continuous series ofsolid solutions in the β phase exists already at 1423 K, and that the homogeneity rangeof the β phase extends with increasing temperature up to the invariant reactionU4: L + γ-TiAl ↔ τ + β at 1570 K.

Based on the phase equilibrium data obtained at 1293 and 1423 K, the respective partialisothermal sections and the reaction plane of the transition reaction γ-TiAl + β ↔ τ + C14were constructed. The assessed phase compositions of the γ-TiAl, β, τ and C14 at thereaction temperature and the averaged chemical compositions of the measured alloys wereused to calculate the amount of phases taking part in the reaction and to extrapolate thereaction enthalpy of ∆Hm(Uinv) = 5211 J/mol. It was found that the precision of thedetermined enthalpy value strongly depends on the accuracy of the determined reactionplane, on the accuracy of the chemical analysis and on the accuracy of the DSCmeasurements.

Titanium aluminides are promising candi-date materials for high temperature appli-cations because of their low density, high-temperature strength as well as a good creepand oxidation resistance [1-5]. Additions of athird alloying element, such as chromium,can further improve the low temperatureductility [6] and oxidation resistance [7].Therefore, the ternary Al-Ti-Cr system is ofgreat importance for industrial applications.Various experimental investigations wereperformed in order to clarify the phaseequilibria in this system. Complete iso-thermal sections in the Al-Ti-Cr system havebeen presented by Jewett et al. [8] at 1073 Kand 1273 K and by Xu et al. [9] at 1323 K.Chen et al. [10] constructed a completesection of the Al-Ti-Cr system consisting of apartial isothermal section at 873 K for thesolid state equilibria for Al > 75 at.% and apartial isothermal section at 1173 K showingthe solid state phase equilibria forAl < 75 at.%. The inconsistencies within theexperimental data at 1473 K were clarified byKriegel et al. [11], who also presented thecorresponding isothermal section. Experi-mental investigations in the ternary Al-Ti-Crsystem involving the liquid phase wereperformed by several authors [10, 12-19].The most comprehensive work was perfor-med by Kriegel et al. [15], who presentedsolidus and liquidus surface projections aswell as a reaction scheme connecting theliquid phase with the solid-state equilibria.Vertical sections in the Al-Ti-Cr system wereshown by Ichimaru et al. [13], Mabuchi et al.[16], Rusnyak et al. [18], Busch [20] andTagunova [21]. A partial vertical section inthe Ti-rich corner of the ternary system wasshown for constant Cr contents of 5 ma.% atthe temperatures between 773 and 1373 K byBusch [20]. Tagunova [1] constructed threevertical sections based on equilibriumstudies. Two partial temperature-compo-sition sections were shown for a constantratio [Al]/[Cr] = 1 for 0 till 16 at.% Al, andfor a constant Cr content of 3.5 at.% and 0 to28 at.% Al. Another vertical section wasconstructed at a constant Ti content of83 at.% [21].Two vertical sections ε-TiAl3 -Al7Cr andAl -TiCr3 were presented by Rusnyak et al.[18]. These vertical sections show extended

Titanium aluminides are promising candi-date materials for high temperature appli-cations because of their low density, high-temperature strength as well as a good creepand oxidation resistance [1-5]. Additions of athird alloying element, such as chromium,can further improve the low temperatureductility [6] and oxidation resistance [7].Therefore, the ternary Al-Ti-Cr system is ofgreat importance for industrial applications.Various experimental investigations wereperformed in order to clarify the phaseequilibria in this system. Complete iso-thermal sections in the Al-Ti-Cr system havebeen presented by Jewett et al. [8] at 1073 Kand 1273 K and by Xu et al. [9] at 1323 K.Chen et al. [10] constructed a completesection of the Al-Ti-Cr system consisting of apartial isothermal section at 873 K for thesolid state equilibria for Al > 75 at.% and apartial isothermal section at 1173 K showingthe solid state phase equilibria forAl < 75 at.%. The inconsistencies within theexperimental data at 1473 K were clarified byKriegel et al. [11], who also presented thecorresponding isothermal section. Experi-mental investigations in the ternary Al-Ti-Crsystem involving the liquid phase wereperformed by several authors [10, 12-19].The most comprehensive work was perfor-med by Kriegel et al. [15], who presentedsolidus and liquidus surface projections aswell as a reaction scheme connecting theliquid phase with the solid-state equilibria.Vertical sections in the Al-Ti-Cr system wereshown by Ichimaru et al. [13], Mabuchi et al.[16], Rusnyak et al. [18], Busch [20] andTagunova [21]. A partial vertical section inthe Ti-rich corner of the ternary system wasshown for constant Cr contents of 5 ma.% atthe temperatures between 773 and 1373 K byBusch [20]. Tagunova [1] constructed threevertical sections based on equilibriumstudies. Two partial temperature-compo-sition sections were shown for a constantratio [Al]/[Cr] = 1 for 0 till 16 at.% Al, andfor a constant Cr content of 3.5 at.% and 0 to28 at.% Al. Another vertical section wasconstructed at a constant Ti content of83 at.% [21].Two vertical sections ε-TiAl3 -Al7Cr andAl -TiCr3 were presented by Rusnyak et al.[18]. These vertical sections show extended

329

Two vertical sections ε-TiAl3 - Al7Cr andAl - TiCr3 were presented by Rusnyak et al.[18]. These vertical sections show extended

Figure 1: Vertical section for constant Ti content of 25 at.% according to Ichimaru et al. [13] and Mabuchi et al. [16].

primary crystallization fields of the binaryAl-Cr phases which leads to a reducedε-TiAl3 primary crystallization phase field.This is in contradiction with the experi-mental work of Chen et al. [10].

Ichimaru et al. [13] and Mabuchi et al. [16]presented the partial vertical sections forTi contents of 25 at.% and for Cr contentsbetween 5 and 30 at.% (see Fig. 1). However,the interpretations of the obtained results aredifferent. In the vertical section constructedby Mabuchi et al. [16], the ternary τ phase isin equilibrium with the C14-Laves phasebelow the solidus line, and the determinedliquidus temperatures are slightly higherthan the temperatures presented in the workof Ichimaru et al. [13]. In the isoplethalsection of Ichimaru et al. [13] the three-phase equilibrium liquid + C14 + τ is indi-cated. This disagrees with the work ofKriegel et al. [11, 15], who showed that theC14 Laves is not in equilibrium with theliquid phase. Instead, the β phase primarycrystallizes from the melt and forms thethree-phase field liquid + β + τ on cooling,which should be present in the verticalsection at 25 at.% Ti.

The precipitation behavior due to extendedhomogeneity ranges of the τ and β phaseswas discussed by Kriegel et al. [11] for analloy with the nominal composition54Al-34Ti-12Cr. It was shown that thecompo-sition of the β phase at 1473 K islocated inside the quadrangle of thetransition reaction γ-TiAl + β ↔ τ + C14.Therefore, the γ-TiAl, τ and C14 Lavesphases can precipitate within the β grainseven during water quenching. In addition,since the composition of the τ phase detectedin the investigations at 1473 K was found tobe located in the γ-TiAl + τ two-phase regionat lower temperatures, precipitates of theγ-TiAl phase could be found in the τ grainsof the quenched samples.

In the present work the invariant reactionγ-TiAl + β ↔ τ + C14 and the vertical sectionat a constant Ti content of 33 at.% have been

primary crystallization fields of the binaryAl-Cr phases which leads to a reducedε-TiAl3 primary crystallization phase field.This is in contradiction with the experi-mental work of Chen et al. [10].

Ichimaru et al. [13] and Mabuchi et al. [16]presented the partial vertical sections forTi contents of 25 at.% and for Cr contentsbetween 5 and 30 at.% (see Fig. 1). However,the interpretations of the obtained results aredifferent. In the vertical section constructedby Mabuchi et al. [16], the ternary τ phase isin equilibrium with the C14-Laves phasebelow the solidus line, and the determinedliquidus temperatures are slightly higherthan the temperatures presented in the workof Ichimaru et al. [13]. In the isoplethalsection of Ichimaru et al. [13] the three-phase equilibrium liquid + C14 + τ is indi-cated. This disagrees with the work ofKriegel et al. [11, 15], who showed that theC14 Laves is not in equilibrium with theliquid phase. Instead, the β phase primarycrystallizes from the melt and forms thethree-phase field liquid + β + τ on cooling,which should be present in the verticalsection at 25 at.% Ti.

The precipitation behavior due to extendedhomogeneity ranges of the τ and β phaseswas discussed by Kriegel et al. [11] for analloy with the nominal composition54Al-34Ti-12Cr. It was shown that thecompo-sition of the β phase at 1473 K islocated inside the quadrangle of thetransition reaction γ-TiAl + β ↔ τ + C14.Therefore, the γ-TiAl, τ and C14 Lavesphases can precipitate within the β grainseven during water quenching. In addition,since the composition of the τ phase detectedin the investigations at 1473 K was found tobe located in the γ-TiAl + τ two-phase regionat lower temperatures, precipitates of theγ-TiAl phase could be found in the τ grainsof the quenched samples.

In the present work the invariant reactionγ-TiAl + β ↔ τ + C14 and the vertical sectionat a constant Ti content of 33 at.% have been

The precipitation behavior due to extendedhomogeneity ranges of the τ and β phaseswas discussed by Kriegel et al. [11] for analloy with the nominal composition54Al-34Ti-12Cr. It was shown that thecomposition of the β phase at 1473 K islocated inside the quadrangle of thetransition reaction γ-TiAl + β ↔ τ + C14.Therefore, the γ-TiAl, τ and C14 Lavesphases can precipitate within the β grainseven during water quenching. In addition,since the composition of the τ phase detectedin the investigations at 1473 K was found tobe located in the γ-TiAl + τ two-phase regionat lower temperatures, precipitates of theγ-TiAl phase could be found in the τ grainsof the quenched samples.

primary crystallization fields of the binaryAl-Cr phases which leads to a reducedε-TiAl3 primary crystallization phase field.This is in contradiction with the experi-mental work of Chen et al. [10].

Ichimaru et al. [13] and Mabuchi et al. [16]presented the partial vertical sections forTi contents of 25 at.% and for Cr contentsbetween 5 and 30 at.% (see Fig. 1). However,the interpretations of the obtained results aredifferent. In the vertical section constructedby Mabuchi et al. [16], the ternary τ phase isin equilibrium with the C14-Laves phasebelow the solidus line, and the determinedliquidus temperatures are slightly higherthan the temperatures presented in the workof Ichimaru et al. [13]. In the isoplethalsection of Ichimaru et al. [13] the three-phase equilibrium liquid + C14 + τ is indi-cated. This disagrees with the work ofKriegel et al. [11, 15], who showed that theC14 Laves is not in equilibrium with theliquid phase. Instead, the β phase primarycrystallizes from the melt and forms thethree-phase field liquid + β + τ on cooling,which should be present in the verticalsection at 25 at.% Ti.

The precipitation behavior due to extendedhomogeneity ranges of the τ and β phaseswas discussed by Kriegel et al. [11] for analloy with the nominal composition54Al-34Ti-12Cr. It was shown that thecompo-sition of the β phase at 1473 K islocated inside the quadrangle of thetransition reaction γ-TiAl + β ↔ τ + C14.Therefore, the γ-TiAl, τ and C14 Lavesphases can precipitate within the β grainseven during water quenching. In addition,since the composition of the τ phase detectedin the investigations at 1473 K was found tobe located in the γ-TiAl + τ two-phase regionat lower temperatures, precipitates of theγ-TiAl phase could be found in the τ grainsof the quenched samples.

In the present work the invariant reactionγ-TiAl + β ↔ τ + C14 and the vertical sectionat a constant Ti content of 33 at.% have been

330

Ichimaru et al. [13] and Mabuchi et al. [16]presented the partial vertical sections for Ticontents of 25 at.% and for Cr contentsbetween 5 and 30 at.% (see Fig. 1). However,the interpretations of the obtained results aredifferent. In the vertical section constructedby Mabuchi et al. [16], the ternary τ phase isin equilibrium with the C14-Laves phasebelow the solidus line, and the determinedliquidus temperatures are slightly higher thanthe temperatures presented in the work ofIchimaru et al. [13]. In the isoplethal sectionof Ichimaru et al. [13] the three-phaseequilibrium liquid + C14 + τ is indicated. Thisdisagrees with the work of Kriegel et al.[11, 15], who showed that the C14 Laves isnot in equilibrium with the liquid phase.Instead, the β phase primary crystallizes fromthe melt and forms the three-phase fieldliquid + β + τ on cooling, which should bepresent in the vertical section at 25 at.% Ti.

The precipitation behavior due to extendedhomogeneity ranges of the τ and β phases wasdiscussed by Kriegel et al. [11] for an alloywith the nominal composition 54Al-34Ti-12Cr. It was shown that the compositionof the β phase at 1473 K is located inside thequadrangle of the transition reactionγ-TiAl + β ↔ τ + C14. Therefore, theγ-TiAl, τ and C14 Laves phases canprecipitate within the β grains even duringwater quenching. In addition, since thecomposition of the τ phase detected in theinvestigations at 1473 K was found to belocated in the γ-TiAl + τ two-phase region atlower temperatures, precipitates of the γ-TiAlphase could be found in the τ grains of thequenched samples.

In the present work the invariant reactionγ-TiAl + β ↔ τ + C14 and the vertical sectionat a constant Ti content of 33 at.% have beeninvestigated in order to confirm theseconclusions from the findings. The obtainedresults can provide both phase equilibria andthermodynamic data, which can be useful forthe development of a reliable thermodynamicdescription of the ternary Al-Ti-Cr system. Inaddition, precipitates of intermetallic phasessuch as the γ-TiAl, τ and C14 phases insidethe β grains of processed alloys can havestrong impact on the mechanical propertiesof these alloys. Therefore, an understandingof the precipitation behavior is essential forthe development of new high temperaturematerials based on titanium aluminides.

The samples with the weight of approximately3 g were prepared by arc melting of high-purity materials (Al: 99.999 %; Ti: 99.995 %;Cr: 99.995 %; Alfa Aesar) in an argon(99.995 %; Ar 5.0) atmosphere. The purematerials were weighed in the respective ratiosand stacked in the melting molds of an electricarc furnace (Edmund Buehler GmbH,Germany) working with a non-consumabletungsten electrode. To ensure a goodhomogeneity of the samples, the resultingingots were turned over using the samplemanipulator and re-melted at least 3 times.After the third melting step the ingots werebroken and re-melted several times. Nosignificant mass change (less than 0.5 % of theoriginal sample mass) before and after meltingin the arc furnace was observed. The heattreatments were performed below theinvariant reaction temperature at 1293 K andabove at 1423 K. For the long term annealingexperiments, the samples were wrapped in aMo foil (99.95 %, Alfa Aesar) and placed intoa quartz capsule together with additional Tisponge that served as an oxygen gettermaterial. The quartz capsule with the sampleswas evacuated, backfilled with Ar (4 x 104 Pa),and annealed at 1293 K for 171 h. The heattreatments at 1423 K were performed in avertical tube furnace under flowing argonatmosphere (10 – 20 ml/min) for 75 h. Afterthe heat treatment, all samples were quenchedinto cold water.

Scanning electron microscopy (SEM) wasapplied to characterize the microstructures ofthe heat-treated alloys. The investigationswere performed on the LEO 1530 GEMINI(Zeiss, Germany) equipped with a fieldemission cathode using the accelerationvoltage of 20 kV and working distances ofapproximately 8 – 10.5 mm. Back-scatteredelectrons were used for imaging themicrostructures in the SEM (SEM/BSE).

The chemical composition of each phase andthe overall composition of the alloys weredetermined using electron probe micro-analysis with wavelength-dispersive X-rayspectroscopy (EPMA/WDS). The chemicalcompositions of individual phases weremeasured locally. The overall chemicalcompositions of the respective alloy wereobtained from the EPMA/WDS measurem-ents that were performed as line-scans over1 mm using the step size of 1 μm. Thus, theoverall chemical composition of the alloysannealed at the respective temperature wasaveraged over 1000 measurement points. Thechemical compositions of the samplesmeasured by EPMA/WDS were comparedwith the nominal alloy compositions, and onlyslight deviations in the concentration ofindividual elements below 1 at.% wereobserved.

The phase identification was performed usingX-ray diffraction (XRD). All XRDmeasurements were performed on powdersamples in the conventional Bragg-Brentanogeometry using Cu-Kα radiation(λ = 0.15418 nm). The diffractometer wasequipped with a curved graphite mono-chromator in the diffracted beam. The fullpowder pattern refinement (Rietveldalgorithm [22, 23], MAUD software [24]) wasemployed to identify the present phases and todetermine their lattice parameters.

In order to determine the transitiontemperatures for the construction of thevertical section at a constant Ti content of33 at.%, thermal analysis experiments werecarried out using the heat flux DSC Pegasus404C (NETZSCH, Germany) with a disc-typemeasuring system. All measurements wereperformed in inert Ar atmosphere (99.999 %;Ar 5.0) using gas flow rates of 50 ml/min. Thesamples were placed in Al2O3 crucibles. Ineach DSC measuring run, the samples wereheated to the annealing temperature of1293 K, kept for 3 h to homogenize andequilibrate, heated to the maximumtemperature of1673 K and afterwards cooleddown to room temperature. Both, the heatingand cooling rates were 10 K/min. The DSC-device was calibrated using the meltingtemperatures of the pure elements In, Sn, Al,Ag, Au, Cu, Ni (certified standard referencematerials from the National Institute ofStandards and Technology, USA).

The calibration of the heat was applied for theinvestigation of the reaction enthalpy for thetransition reaction γ-TiAl + β ↔ τ + C14. Thetemperature of this reaction was found to bebetween 1273 and 1373 K. Therefore, thecalibration was performed only in thistemperature range using the pure elementsAu, Ag and Cu. The correction function wasdetermined using the linear extrapolationmethod for the heating rate of 10 K/min.Following the recommendations of Boettingeret al. [25] to obtain accurate heat fluxinformation, the same heating / cooling rate,gas flow rate, sample / reference crucibles(Pt/Rh crucible including a very thin Al2O3crucible to avoid the contact between thesamples and the Pt/Rh crucible) andtemperature range were employed forcalibration and sample measurements. Theannealed samples were also equilibrated in theDSC device by including a 3 h long isothermalsequence at the respective heat treatmenttemperature before subsequent heating andcooling at rates of 10 K/min. The peaks of themeasured DSC curves were integrated usingthe Proteus software (sigmoidal baselinesubtraction method) delivered by themanufacturer of the DSC device (NETZSCH,Germany).

The samples with the weight of approximately3 g were prepared by arc melting of high-purity materials (Al: 99.999 %; Ti: 99.995 %;Cr: 99.995 %; Alfa Aesar) in an argon(99.995 %; Ar 5.0) atmosphere. The purematerials were weighed in the respective ratiosand stacked in the melting molds of an electricarc furnace (Edmund Buehler GmbH,Germany) working with a non-consumabletungsten electrode. To ensure a goodhomogeneity of the samples, the resultingingots were turned over using the samplemanipulator and re-melted at least 3 times.After the third melting step the ingots werebroken and re-melted several times. Nosignificant mass change (less than 0.5 % of theoriginal sample mass) before and after meltingin the arc furnace was observed. The heattreatments were performed below theinvariant reaction temperature at 1293 K andabove at 1423 K. For the long term annealingexperiments, the samples were wrapped in aMo foil (99.95 %, Alfa Aesar) and placed intoa quartz capsule together with additional Tisponge that served as an oxygen gettermaterial. The quartz capsule with the sampleswas evacuated, backfilled with Ar (4 x 104 Pa),and annealed at 1293 K for 171 h. The heattreatments at 1423 K were performed in avertical tube furnace under flowing argonatmosphere (10 – 20 ml/min) for 75 h. Afterthe heat treatment, all samples were quenchedinto cold water.

Scanning electron microscopy (SEM) wasapplied to characterize the microstructures ofthe heat-treated alloys. The investigationswere performed on the LEO 1530 GEMINI(Zeiss, Germany) equipped with a fieldemission cathode using the accelerationvoltage of 20 kV and working distances ofapproximately 8 – 10.5 mm. Back-scatteredelectrons were used for imaging themicrostructures in the SEM (SEM/BSE).

The chemical composition of each phase andthe overall composition of the alloys weredetermined using electron probe micro-analysis with wavelength-dispersive X-rayspectroscopy (EPMA/WDS). The chemicalcompositions of individual phases weremeasured locally. The overall chemicalcompositions of the respective alloy wereobtained from the EPMA/WDS measurem-ents that were performed as line-scans over1 mm using the step size of 1 μm. Thus, theoverall chemical composition of the alloysannealed at the respective temperature wasaveraged over 1000 measurement points. Thechemical compositions of the samplesmeasured by EPMA/WDS were comparedwith the nominal alloy compositions, and onlyslight deviations in the concentration ofindividual elements below 1 at.% wereobserved.

The phase identification was performed usingX-ray diffraction (XRD). All XRDmeasurements were performed on powdersamples in the conventional Bragg-Brentanogeometry using Cu-Kα radiation(λ = 0.15418 nm). The diffractometer wasequipped with a curved graphite mono-chromator in the diffracted beam. The fullpowder pattern refinement (Rietveldalgorithm [22, 23], MAUD software [24]) wasemployed to identify the present phases and todetermine their lattice parameters.

In order to determine the transitiontemperatures for the construction of thevertical section at a constant Ti content of33 at.%, thermal analysis experiments werecarried out using the heat flux DSC Pegasus404C (NETZSCH, Germany) with a disc-typemeasuring system. All measurements wereperformed in inert Ar atmosphere (99.999 %;Ar 5.0) using gas flow rates of 50 ml/min. Thesamples were placed in Al2O3 crucibles. Ineach DSC measuring run, the samples wereheated to the annealing temperature of1293 K, kept for 3 h to homogenize andequilibrate, heated to the maximumtemperature of1673 K and afterwards cooleddown to room temperature. Both, the heatingand cooling rates were 10 K/min. The DSC-device was calibrated using the meltingtemperatures of the pure elements In, Sn, Al,Ag, Au, Cu, Ni (certified standard referencematerials from the National Institute ofStandards and Technology, USA).

The calibration of the heat was applied for theinvestigation of the reaction enthalpy for thetransition reaction γ-TiAl + β ↔ τ + C14. Thetemperature of this reaction was found to bebetween 1273 and 1373 K. Therefore, thecalibration was performed only in thistemperature range using the pure elementsAu, Ag and Cu. The correction function wasdetermined using the linear extrapolationmethod for the heating rate of 10 K/min.Following the recommendations of Boettingeret al. [25] to obtain accurate heat fluxinformation, the same heating / cooling rate,gas flow rate, sample / reference crucibles(Pt/Rh crucible including a very thin Al2O3crucible to avoid the contact between thesamples and the Pt/Rh crucible) andtemperature range were employed forcalibration and sample measurements. Theannealed samples were also equilibrated in theDSC device by including a 3 h long isothermalsequence at the respective heat treatmenttemperature before subsequent heating andcooling at rates of 10 K/min. The peaks of themeasured DSC curves were integrated usingthe Proteus software (sigmoidal baselinesubtraction method) delivered by themanufacturer of the DSC device (NETZSCH,Germany).

2 Experimental Procedures

The samples with the weight of approximately3 g were prepared by arc melting of high-purity materials (Al: 99.999 %; Ti: 99.995 %;Cr: 99.995 %; Alfa Aesar) in an argon(99.999 %; Ar 5.0) atmosphere. The pure

The samples with the weight of approximately3 g were prepared by arc melting of high-purity materials (Al: 99.999 %; Ti: 99.995 %;Cr: 99.995 %; Alfa Aesar) in an argon(99.995 %; Ar 5.0) atmosphere. The purematerials were weighed in the respective ratiosand stacked in the melting molds of an electricarc furnace (Edmund Buehler GmbH,Germany) working with a non-consumabletungsten electrode. To ensure a goodhomogeneity of the samples, the resultingingots were turned over using the samplemanipulator and re-melted at least 3 times.After the third melting step the ingots werebroken and re-melted several times. Nosignificant mass change (less than 0.5 % of theoriginal sample mass) before and after meltingin the arc furnace was observed. The heattreatments were performed below theinvariant reaction temperature at 1293 K andabove at 1423 K. For the long term annealingexperiments, the samples were wrapped in aMo foil (99.95 %, Alfa Aesar) and placed intoa quartz capsule together with additional Tisponge that served as an oxygen gettermaterial. The quartz capsule with the sampleswas evacuated, backfilled with Ar (4 x 104 Pa),and annealed at 1293 K for 171 h. The heattreatments at 1423 K were performed in avertical tube furnace under flowing argonatmosphere (10 – 20 ml/min) for 75 h. Afterthe heat treatment, all samples were quenchedinto cold water.

Scanning electron microscopy (SEM) wasapplied to characterize the microstructures ofthe heat-treated alloys. The investigationswere performed on the LEO 1530 GEMINI(Zeiss, Germany) equipped with a fieldemission cathode using the accelerationvoltage of 20 kV and working distances ofapproximately 8 – 10.5 mm. Back-scatteredelectrons were used for imaging themicrostructures in the SEM (SEM/BSE).

The chemical composition of each phase andthe overall composition of the alloys weredetermined using electron probe micro-analysis with wavelength-dispersive X-rayspectroscopy (EPMA/WDS). The chemicalcompositions of individual phases weremeasured locally. The overall chemicalcompositions of the respective alloy wereobtained from the EPMA/WDS measurem-ents that were performed as line-scans over1 mm using the step size of 1 μm. Thus, theoverall chemical composition of the alloysannealed at the respective temperature wasaveraged over 1000 measurement points. Thechemical compositions of the samplesmeasured by EPMA/WDS were comparedwith the nominal alloy compositions, and onlyslight deviations in the concentration ofindividual elements below 1 at.% wereobserved.

The phase identification was performed usingX-ray diffraction (XRD). All XRDmeasurements were performed on powdersamples in the conventional Bragg-Brentanogeometry using Cu-Kα radiation(λ = 0.15418 nm). The diffractometer wasequipped with a curved graphite mono-chromator in the diffracted beam. The fullpowder pattern refinement (Rietveldalgorithm [22, 23], MAUD software [24]) wasemployed to identify the present phases and todetermine their lattice parameters.

In order to determine the transitiontemperatures for the construction of thevertical section at a constant Ti content of33 at.%, thermal analysis experiments werecarried out using the heat flux DSC Pegasus404C (NETZSCH, Germany) with a disc-typemeasuring system. All measurements wereperformed in inert Ar atmosphere (99.999 %;Ar 5.0) using gas flow rates of 50 ml/min. Thesamples were placed in Al2O3 crucibles. Ineach DSC measuring run, the samples wereheated to the annealing temperature of1293 K, kept for 3 h to homogenize andequilibrate, heated to the maximumtemperature of1673 K and afterwards cooleddown to room temperature. Both, the heatingand cooling rates were 10 K/min. The DSC-device was calibrated using the meltingtemperatures of the pure elements In, Sn, Al,Ag, Au, Cu, Ni (certified standard referencematerials from the National Institute ofStandards and Technology, USA).

The calibration of the heat was applied for theinvestigation of the reaction enthalpy for thetransition reaction γ-TiAl + β ↔ τ + C14. Thetemperature of this reaction was found to bebetween 1273 and 1373 K. Therefore, thecalibration was performed only in thistemperature range using the pure elementsAu, Ag and Cu. The correction function wasdetermined using the linear extrapolationmethod for the heating rate of 10 K/min.Following the recommendations of Boettingeret al. [25] to obtain accurate heat fluxinformation, the same heating / cooling rate,gas flow rate, sample / reference crucibles(Pt/Rh crucible including a very thin Al2O3crucible to avoid the contact between thesamples and the Pt/Rh crucible) andtemperature range were employed forcalibration and sample measurements. Theannealed samples were also equilibrated in theDSC device by including a 3 h long isothermalsequence at the respective heat treatmenttemperature before subsequent heating andcooling at rates of 10 K/min. The peaks of themeasured DSC curves were integrated usingthe Proteus software (sigmoidal baselinesubtraction method) delivered by themanufacturer of the DSC device (NETZSCH,Germany).

The samples with the weight of approximately3 g were prepared by arc melting of high-purity materials (Al: 99.999 %; Ti: 99.995 %;Cr: 99.995 %; Alfa Aesar) in an argon(99.995 %; Ar 5.0) atmosphere. The purematerials were weighed in the respective ratiosand stacked in the melting molds of an electricarc furnace (Edmund Buehler GmbH,Germany) working with a non-consumabletungsten electrode. To ensure a goodhomogeneity of the samples, the resultingingots were turned over using the samplemanipulator and re-melted at least 3 times.After the third melting step the ingots werebroken and re-melted several times. Nosignificant mass change (less than 0.5 % of theoriginal sample mass) before and after meltingin the arc furnace was observed. The heattreatments were performed below theinvariant reaction temperature at 1293 K andabove at 1423 K. For the long term annealingexperiments, the samples were wrapped in aMo foil (99.95 %, Alfa Aesar) and placed intoa quartz capsule together with additional Tisponge that served as an oxygen gettermaterial. The quartz capsule with the sampleswas evacuated, backfilled with Ar (4 x 104 Pa),and annealed at 1293 K for 171 h. The heattreatments at 1423 K were performed in avertical tube furnace under flowing argonatmosphere (10 – 20 ml/min) for 75 h. Afterthe heat treatment, all samples were quenchedinto cold water.

Scanning electron microscopy (SEM) wasapplied to characterize the microstructures ofthe heat-treated alloys. The investigationswere performed on the LEO 1530 GEMINI(Zeiss, Germany) equipped with a fieldemission cathode using the accelerationvoltage of 20 kV and working distances ofapproximately 8 – 10.5 mm. Back-scatteredelectrons were used for imaging themicrostructures in the SEM (SEM/BSE).

The chemical composition of each phase andthe overall composition of the alloys weredetermined using electron probe micro-analysis with wavelength-dispersive X-rayspectroscopy (EPMA/WDS). The chemicalcompositions of individual phases weremeasured locally. The overall chemicalcompositions of the respective alloy wereobtained from the EPMA/WDS measurem-ents that were performed as line-scans over1 mm using the step size of 1 μm. Thus, theoverall chemical composition of the alloysannealed at the respective temperature wasaveraged over 1000 measurement points. Thechemical compositions of the samplesmeasured by EPMA/WDS were comparedwith the nominal alloy compositions, and onlyslight deviations in the concentration ofindividual elements below 1 at.% wereobserved.

The phase identification was performed usingX-ray diffraction (XRD). All XRDmeasurements were performed on powdersamples in the conventional Bragg-Brentanogeometry using Cu-Kα radiation(λ = 0.15418 nm). The diffractometer wasequipped with a curved graphite mono-chromator in the diffracted beam. The fullpowder pattern refinement (Rietveldalgorithm [22, 23], MAUD software [24]) wasemployed to identify the present phases and todetermine their lattice parameters.

In order to determine the transitiontemperatures for the construction of thevertical section at a constant Ti content of33 at.%, thermal analysis experiments werecarried out using the heat flux DSC Pegasus404C (NETZSCH, Germany) with a disc-typemeasuring system. All measurements wereperformed in inert Ar atmosphere (99.999 %;Ar 5.0) using gas flow rates of 50 ml/min. Thesamples were placed in Al2O3 crucibles. Ineach DSC measuring run, the samples wereheated to the annealing temperature of1293 K, kept for 3 h to homogenize andequilibrate, heated to the maximumtemperature of1673 K and afterwards cooleddown to room temperature. Both, the heatingand cooling rates were 10 K/min. The DSC-device was calibrated using the meltingtemperatures of the pure elements In, Sn, Al,Ag, Au, Cu, Ni (certified standard referencematerials from the National Institute ofStandards and Technology, USA).

The calibration of the heat was applied for theinvestigation of the reaction enthalpy for thetransition reaction γ-TiAl + β ↔ τ + C14. Thetemperature of this reaction was found to bebetween 1273 and 1373 K. Therefore, thecalibration was performed only in thistemperature range using the pure elementsAu, Ag and Cu. The correction function wasdetermined using the linear extrapolationmethod for the heating rate of 10 K/min.Following the recommendations of Boettingeret al. [25] to obtain accurate heat fluxinformation, the same heating / cooling rate,gas flow rate, sample / reference crucibles(Pt/Rh crucible including a very thin Al2O3crucible to avoid the contact between thesamples and the Pt/Rh crucible) andtemperature range were employed forcalibration and sample measurements. Theannealed samples were also equilibrated in theDSC device by including a 3 h long isothermalsequence at the respective heat treatmenttemperature before subsequent heating andcooling at rates of 10 K/min. The peaks of themeasured DSC curves were integrated usingthe Proteus software (sigmoidal baselinesubtraction method) delivered by themanufacturer of the DSC device (NETZSCH,Germany).

The chemical composition of each phase andthe overall composition of the alloys weredetermined using electron probe micro-analysis with wavelength-dispersive X-rayspectroscopy (EPMA/WDS). The chemicalcompositions of individual phases weremeasured locally. The overall chemicalcompositions of the respective alloy wereobtained from the EPMA/WDS measure-ments that were performed as line-scansover 1 mm using the step size of 1 μm. Thus,the overall chemical composition of thealloys annealed at the respective temperaturewas averaged over 1000 measurement points.The chemical compositions of the samplesmeasured by EPMA/WDS were comparedwith the nominal alloy compositions, andonly slight deviations in the concentration ofindividual elements below 1 at.% wereobserved.

331

Phase Spacegroup

Pearsonsymbol

Structuretype

Lattice parametera [nm] c [nm] Reference

β, (Cr) Im m cl2 W 0.28845 [26]

β, βTi Im m cl2 W 0.33112 [28], 1193 K

γ-TiAl P4/mmm tP2 AuCu 0.2832 0.4070 [27]

ζ-Ti2Al5 P4/mmm tP28 Ti2Al5 0.39053 2.91963 [29]

C14 P63/mmc hP12 MgZn2 0.4932 0.8005 [31]

τ (Al2.67Cr0.33Ti) Pm m cP4 AuCu3 0.3958 [30]

Table 1: Crystallographic data of the phases stable in the partial vertical section at a constant Ti content of 33 at.%.

performed in inert Ar atmosphere (99.999 %;Ar 5.0) using gas flow rates of 50 ml/min.The samples were placed in Al2O3 crucibles.In each DSC measuring run, the sampleswere heated to the annealing temperature of1293 K, kept for 3 h to homogenize andequilibrate, heated to the maximumtemperature of1673 K and afterwards cooleddown to room temperature. Both, theheating and cooling rates were 10 K/min.The DSC-device was calibrated using themelting temperatures of the pure elementsIn, Sn, Al, Ag, Au, Cu, Ni (certified standardreference materials from the NationalInstitute of Standards and Technology,USA).

The calibration of the heat was applied forthe investigation of the reaction enthalpy forthe transition reaction γ-TiAl + β ↔ τ + C14.The temperature of this reaction was foundto be between 1273 and 1373 K. Therefore,the calibration was performed only in thistemperature range using the pure elementsAu, Ag and Cu. The correction function wasdetermined using the linear extrapolationmethod for the heating rate of 10 K/min.Following the recommendations ofBoettinger et al. [25] to obtain accurate heatflux information, the same heating / coolingrate, gas flow rate, sample / referencecrucibles (Pt/Rh crucible including a verythin Al2O3 crucible to avoid the contactbetween the samples and the Pt/Rh crucible)and temperature range were employed forcalibration and sample measurements. Thepeaks of the measured DSC curves wereintegrated using the Proteus software(sigmoidal baseline subtraction method)delivered by the manufacturer of the DSCdevice (NETZSCH, Germany).

performed in inert Ar atmosphere (99.999 %;Ar 5.0) using gas flow rates of 50 ml/min.The samples were placed in Al2O3 crucibles.In each DSC measuring run, the sampleswere heated to the annealing temperature of1293 K, kept for 3 h to homogenize andequilibrate, heated to the maximumtemperature of1673 K and afterwards cooleddown to room temperature. Both, theheating and cooling rates were 10 K/min.The DSC-device was calibrated using themelting temperatures of the pure elementsIn, Sn, Al, Ag, Au, Cu, Ni (certified standardreference materials from the NationalInstitute of Standards and Technology,USA).

The calibration of the heat was applied forthe investigation of the reaction enthalpy forthe transition reaction γ-TiAl + β ↔ τ + C14.The temperature of this reaction was foundto be between 1273 and 1373 K. Therefore,the calibration was performed only in thistemperature range using the pure elementsAu, Ag and Cu. The correction function wasdetermined using the linear extrapolationmethod for the heating rate of 10 K/min.Following the recommendations ofBoettinger et al. [25] to obtain accurate heatflux information, the same heating / coolingrate, gas flow rate, sample / referencecrucibles (Pt/Rh crucible including a verythin Al2O3 crucible to avoid the contactbetween the samples and the Pt/Rh crucible)and temperature range were employed forcalibration and sample measurements. Thepeaks of the measured DSC curves wereintegrated using the Proteus software(sigmoidal baseline subtraction method)delivered by the manufacturer of the DSCdevice (NETZSCH, Germany).

performed in inert Ar atmosphere (99.999 %;Ar 5.0) using gas flow rates of 50 ml/min.The samples were placed in Al2O3 crucibles.In each DSC measuring run, the sampleswere heated to the annealing temperature of1293 K, kept for 3 h to homogenize andequilibrate, heated to the maximumtemperature of 1673 K and afterwards cooleddown to room temperature. Both, theheating and cooling rates were 10 K/min.The DSC-device was calibrated using themelting temperatures of the pure elementsIn, Sn, Al, Ag, Au, Cu, Ni (certified standardreference materials from the NationalInstitute of Standards and Technology,USA).

332

The phases occurring in the investigatedpartial vertical section of the phase diagramat a constant Ti content of 33 at.% in theternary Al-Ti-Cr system are listed in Table 1,together with the crystallographic data fromreferences [26-31].

The nominal and measured chemicalcompositions of the alloys (determined fromthe initial mass ratio of the elements andfrom EPMA/WDS, respectively), the presentphases and their lattice parameters (bothobtained from XRD), and the chemicalcompositions of individual phases (revealedby EPMA/WDS) present in the heat treatedsamples are given in Table 2.

Three alloys with the alloy compositions53Al-33Ti-14Cr (#1), 47AL-33Ti-20Cr (#2)and 43Al-33Ti-24Cr (#3) were prepared,annealed and subsequently quenched intowater to determine the reaction enthalpy ofthe transition reaction γ-TiAl + β ↔ τ + C14.Two annealing temperatures (and annealingtimes) were selected: one below the invariantreaction temperature (1293 K / 171 h),another one above the reaction temperature(1423 K / 75 h). The microstructures of theseheat treated alloys are shown together withthe partial isothermal sections constructedfrom the experimental information infigures 2 and 3.

In the SEM/BSE micrographs, the grains ofthe β and C14 phases are bright, whereas theγ-TiAl and τ grains appear dark due todifferences in the chemical compositions ofthe respective phases (elemental contrast).The partial isothermal sections wereconstructed based on the chemicalcompositions of the respective phasesmeasured by EPMA/WDS. Due to the smalldifferences in the chemical composition ofthe γ-TiAl and τ phases, the differentiationbetween the γ-TiAl and τ grains is difficultby means of the SEM/BSE micrographs only.Therefore, the local chemical compositionswere measured via EPMA/WDS, and thecorresponding phases were identified byusing XRD. The grains are larger in samplesat 1423 K than in the samples annealed at1293 K as expected for faster diffusionkinetics at higher annealing temperatures.

At 1293 K, all three alloys are located insidethe three-phase triangle γ-TiAl + τ + C14.Therefore, it was not possible to detectthe remaining three-phase equilibriumβ + τ + C14. At the annealing temperature of1423 K, the alloy 53Al-33Ti-14Cr (#1) wasfound to be located inside the three-phasetriangle γ-TiAl + β +τ. The alloys47Al-33Ti-20Cr (#2) and 43Al-33Ti-24Cr(#3) consisted of a γ-TiAl + β two-phasemicrostructure, which implies that the βcontinuous series of solid solutions from theCr-rich up to the Ti-rich corner of theternary system already exist at 1423 K.Furthermore, the temperature, at which thetwo two-phase equilibria γ-TiAl + β andβ + C14 intersect in order to form two γ-TiAl + C14 + β-Ti and γ-TiAl + C14 + β-Crthree-phase equilibria, has to be between thetransition reaction temperature and 1423 K.

The phases occurring in the investigatedpartial vertical section of the phase diagramat a constant Ti content of 33 at.% in theternary Al-Ti-Cr system are listed in Table 1,together with the crystallographic data fromreferences [26-31].

The nominal and measured chemicalcompositions of the alloys (determined fromthe initial mass ratio of the elements andfrom EPMA/WDS, respectively), the presentphases and their lattice parameters (bothobtained from XRD), and the chemicalcompositions of individual phases (revealedby EPMA/WDS) present in the heat treatedsamples are given in Table 2.

Three alloys with the alloy compositions53Al-33Ti-14Cr (#1), 47AL-33Ti-20Cr (#2)and 43Al-33Ti-24Cr (#3) were prepared,annealed and subsequently quenched intowater to determine the reaction enthalpy ofthe transition reaction γ-TiAl + β ↔ τ + C14.Two annealing temperatures (and annealingtimes) were selected: one below the invariantreaction temperature (1293 K / 171 h),another one above the reaction temperature(1423 K / 75 h). The microstructures of theseheat treated alloys are shown together withthe partial isothermal sections constructedfrom the experimental information infigures 2 and 3.

In the SEM/BSE micrographs, the grains ofthe β and C14 phases are bright, whereas theγ-TiAl and τ grains appear dark due todifferences in the chemical compositions ofthe respective phases (elemental contrast).The partial isothermal sections wereconstructed based on the chemicalcompositions of the respective phasesmeasured by EPMA/WDS. Due to the smalldifferences in the chemical composition ofthe γ-TiAl and τ phases, the differentiationbetween the γ-TiAl and τ grains is difficultby means of the SEM/BSE micrographs only.Therefore, the local chemical compositionswere measured via EPMA/WDS, and thecorresponding phases were identified byusing XRD. The grains are larger in samplesat 1423 K than in the samples annealed at1293 K as expected for faster diffusionkinetics at higher annealing temperatures.

At 1293 K, all three alloys are located insidethe three-phase triangle γ-TiAl + τ + C14.Therefore, it was not possible to detectthe remaining three-phase equilibriumβ + τ + C14. At the annealing temperature of1423 K, the alloy 53Al-33Ti-14Cr (#1) wasfound to be located inside the three-phasetriangle γ-TiAl + β +τ. The alloys47Al-33Ti-20Cr (#2) and 43Al-33Ti-24Cr(#3) consisted of a γ-TiAl + β two-phasemicrostructure, which implies that the βcontinuous series of solid solutions from theCr-rich up to the Ti-rich corner of theternary system already exist at 1423 K.Furthermore, the temperature, at which thetwo two-phase equilibria γ-TiAl + β andβ + C14 intersect in order to form two γ-TiAl + C14 + β-Ti and γ-TiAl + C14 + β-Crthree-phase equilibria, has to be between thetransition reaction temperature and 1423 K.

333

3 Results and Discussion

Nom

inal

com

posi

tion

[at.%

]EP

MA

/WD

S re

sults

[at.%

]La

ttic

e pa

ram

eter

[nm

]EP

MA

/WD

S re

sults

[at.%

]

Allo

yA

lTi

Cr

Al

TiC

rH

eat

trea

tmen

tPh

ases

pres

ent

ac

Al

TiC

r1

5333

1451

.733

.315

.012

93K

/171

hγ-

TiA

l0.

2796

8(6)

0.40

400(

3)55

.334

.99.

8τ

0.39

670(

8)56

.531

.512

.0C

140.

5066

3(9)

0.82

821(

8)39

.233

.127

.7

51.6

33.9

14.5

1423

K/7

5h

γ-Ti

Al

0.27

970(

2)0.

4037

9(1)

54.5

37.3

8.2

β0.

3096

8(9)

41.1

30.5

28.4

τ0.

3970

4(6)

56.4

32.4

11.2

247

3320

46.3

33.3

20.4

1293

K/1

71h

γ-Ti

Al

0.28

010(

3)0.

4041

1(4)

55.3

35.3

9.4

τ0.

3966

6(5)

56.7

31.4

11.9

C14

0.50

667(

4)0.

8282

0(5)

39.6

33.0

27.4

45.5

33.6

20.9

1423

K/7

5h

γ-Ti

Al

0.28

003(

5)0.

4043

5(5)

53.7

38.3

8.0

β0.

3101

5(6)

40.5

31.0

28.5

343

3324

42.6

33.6

23.8

1293

K/1

71h

γ-Ti

Al

0.28

008(

4)0.

4040

0(5)

54.0

36.4

9.6

τ0.

3968

8(7)

56.4

32.4

11.2

C14

0.50

664(

2)0.

8280

5(8)

39.6

33.1

27.3

41.6

33.2

25.2

1423

K/7

5h

γ-Ti

Al

0.28

031(

7)0.

4046

4(4)

53.2

39.6

7.2

β0.

3106

9(4)

40.5

32.7

26.8

Tabl

e 2:

EPM

A/W

DS

and

XRD

resu

lts o

f the

sam

ples

hea

t tre

ated

at 1

293

and

1423

K. Th

e un

cert

aint

y of

the

EPM

A/W

DS

mea

sure

men

ts w

as a

sses

sed

to b

e 0.

1 at

.%.

334

Figure 2: Partial isothermal section at 1293 K showing the three-phase equilibrium γ-TiAl + C14 + τconstructed based on the EPMA/WDS results and microstructures of the alloys 53Al-33Ti-14Cr (#1), 47Al-33Ti-20Cr (#2) and 43Al-33Ti-24Cr (#3).

Figure 3: Partial isothermal section at 1423 K showing the three-phase equilibrium γ-TiAl + β + τ and two γ-TiAl + β two-phase equilibria constructed based on the EPMA/WDS results, and microstructures of the alloys 53Al-33Ti-14Cr (#1), 47Al-33Ti-20Cr (#2) and 43Al-33Ti-24Cr (#3).

335

Based on the DSC measurements (see Fig.4), the reaction temperature of thetransition reaction γ-TiAl + β ↔ τ + C14was determined to be 1354 K, which is 31 Kabove the reaction temperature reported byCupid et al. [32]. The reaction temperatureof the alloy 43Al-33Ti-24Cr (#3) was foundto be shifted toward slightly higher values(as compared to 1354 K). This effect can beexplained by a slower system responsecaused by the system inertia due to theabout half lower sample mass of alloy 43Al-33Ti-24Cr (#3) compared to the alloys53Al-33Ti-14Cr (#1) and 47Al-33Ti-20Cr(#2). The peaks appearing in the hightemperature part of the DSC curves (seeFig. 4) are related to the melting of thealloys. Thus, the first deflection from thebaseline corresponds to the solidus tempe-rature of the respective alloys. The liquidustemperatures can be assigned to the pointof inflection on the measured DSC curvejust before the DTA curve returns tobaseline.

Figure 5 shows the constructed partialisoplethal section at 33 at.% Ti and thesuperimposed DSC measurements of alloy53Al-33Ti-14Cr (#1). The heating curveshown in this figure was composed of thetwo DSC measurements (for the respectiveheat treatment). The first curve wasrecorded between 1293 and 1423 K,whereas the second curve is from 1423 to1650 K. The first DSC peak at 1354 Kcorresponds to the invariant reactionγ-TiAl + β ↔ τ + C14. At temperaturesabove, alloy 53Al-33Ti-14Cr (#1) was foundto be located inside the γ-TiAl + β + τthree-phase triangle (see Fig. 3). In thesecond DSC curve, a further peak at 1570 Kcorresponding to the transition reactionU4: liquid + γ-TiAl ↔ τ + β and theliquidus temperature of 1606 K wasdetected. This invariant reaction and theliquidus temperatures perfectly coincidewith the experimental results alreadyshown in a previous publication devoted tothe investigation of the solidus and liquidusprojections in the ternary Al-Ti-Cr system[15].

The two DSC heating curves for alloy 47Al-33Ti-20Cr (#2) are shown in figure 6 forboth annealing temperatures (curve 1: DSCisotherm at 1293 K, curve 2: isotherm in theDSC at 1423 K). Similarly to alloy 53Al-33Ti-14Cr (#1), the first peak in theheating curve at 1354 K corresponds to theinvariant reaction γ-TiAl + β ↔ τ + C14.The observations at 1423 K show thestability of the γ-TiAl + β two-phasemicrostructure, which imply the occur-rence of an additional thermal effectbetween the reaction temperature and1423 K. However, a vertical jump in theDSC heating curve was observed only at1559 K, which corresponds to theγ-TiAl + β + τ→ γ-TiAl + β transition. Thisindicates that at 1423 K the samples werelocated inside the three-phase triangleγ-TiAl + β + τ and very close to theboundary tie-line γ-TiAl + β. Therefore, theamount of the τ phase is very small andthus not detectable by the microstructuralinvestigation methods used (SEM/EDS,EPMA/WDS and XRD). Another explana-tion for the absence of the τ phase may bethat the variations in the chemicalcomposition due to the inhomogeneities inthe prepared sample lead to differences inthe observed phase equilibria. Sincedifferent pieces of the sample were used forthe EPMA/WDS and DSC investigations,some discrepancies could be detected.

The second peak observed in the heatingcurve 2 of alloy 47Al-33Ti-20Cr (see Fig. 6)is caused by the heat needed to melt thisalloy. The measured solidus temperature of1578 K is slightly higher than thetemperature of the transition reaction U4occurring at 1570 K, which indicates thatthe composition of this alloy does notintersect the plane of the transitionreaction. The temperature of 1629 K wasdetermined to be the liquidus temperaturefor this alloy.

Based on the DSC measurements (see Fig.4), the reaction temperature of thetransition reaction γ-TiAl + β ↔ τ + C14was determined to be 1354 K, which is 31 Kabove the reaction temperature reported byCupid et al. [32]. The reaction temperatureof the alloy 43Al-33Ti-24Cr (#3) was foundto be shifted toward slightly higher values(as compared to 1354 K). This effect can beexplained by a slower system responsecaused by the system inertia due to theabout half lower sample mass of alloy 43Al-33Ti-24Cr (#3) compared to the alloys53Al-33Ti-14Cr (#1) and 47Al-33Ti-20Cr(#2). The peaks appearing in the hightemperature part of the DSC curves (seeFig. 4) are related to the melting of thealloys. Thus, the first deflection from thebaseline corresponds to the solidus tempe-rature of the respective alloys. The liquidustemperatures can be assigned to the pointof inflection on the measured DSC curvejust before the DTA curve returns tobaseline.

Figure 5 shows the constructed partialisoplethal section at 33 at.% Ti and thesuperimposed DSC measurements of alloy53Al-33Ti-14Cr (#1). The heating curveshown in this figure was composed of thetwo DSC measurements (for the respectiveheat treatment). The first curve wasrecorded between 1293 and 1423 K,whereas the second curve is from 1423 to1650 K. The first DSC peak at 1354 Kcorresponds to the invariant reactionγ-TiAl + β ↔ τ + C14. At temperaturesabove, alloy 53Al-33Ti-14Cr (#1) was foundto be located inside the γ-TiAl + β + τthree-phase triangle (see Fig. 3). In thesecond DSC curve, a further peak at 1570 Kcorresponding to the transition reactionU4: liquid + γ-TiAl ↔ τ + β and theliquidus temperature of 1606 K wasdetected. This invariant reaction and theliquidus temperatures perfectly coincidewith the experimental results alreadyshown in a previous publication devoted tothe investigation of the solidus and liquidusprojections in the ternary Al-Ti-Cr system[15].

The two DSC heating curves for alloy 47Al-33Ti-20Cr (#2) are shown in figure 6 forboth annealing temperatures (curve 1: DSCisotherm at 1293 K, curve 2: isotherm in theDSC at 1423 K). Similarly to alloy 53Al-33Ti-14Cr (#1), the first peak in theheating curve at 1354 K corresponds to theinvariant reaction γ-TiAl + β ↔ τ + C14.The observations at 1423 K show thestability of the γ-TiAl + β two-phasemicrostructure, which imply the occur-rence of an additional thermal effectbetween the reaction temperature and1423 K. However, a vertical jump in theDSC heating curve was observed only at1559 K, which corresponds to theγ-TiAl + β + τ→ γ-TiAl + β transition. Thisindicates that at 1423 K the samples werelocated inside the three-phase triangleγ-TiAl + β + τ and very close to theboundary tie-line γ-TiAl + β. Therefore, theamount of the τ phase is very small andthus not detectable by the microstructuralinvestigation methods used (SEM/EDS,EPMA/WDS and XRD). Another explana-tion for the absence of the τ phase may bethat the variations in the chemicalcomposition due to the inhomogeneities inthe prepared sample lead to differences inthe observed phase equilibria. Sincedifferent pieces of the sample were used forthe EPMA/WDS and DSC investigations,some discrepancies could be detected.

The second peak observed in the heatingcurve 2 of alloy 47Al-33Ti-20Cr (see Fig. 6)is caused by the heat needed to melt thisalloy. The measured solidus temperature of1578 K is slightly higher than thetemperature of the transition reaction U4occurring at 1570 K, which indicates thatthe composition of this alloy does notintersect the plane of the transitionreaction. The temperature of 1629 K wasdetermined to be the liquidus temperaturefor this alloy.

The two DSC heating curves for alloy47Al-33Ti-20Cr (#2) are shown in figure 6for both annealing temperatures (curve 1:DSC isotherm at 1293 K, curve 2: isothermin the DSC at 1423 K). Similarly to alloy53Al-33Ti-14Cr (#1), the first peak in theheating curve at 1354 K corresponds to theinvariant reaction γ-TiAl + β↔ τ + C14. Theobservations at 1423 K show the stability ofthe γ-TiAl + β two-phase microstructure,which imply the occurrence of an additionalthermal effect between the reactiontemperature and 1423 K. However, a verticaljump in the DSC heating curve was observedonly at 1559 K, which corresponds to theγ-TiAl + β + τ → γ-TiAl + β transition. Thisindicates that at 1423 K the samples werelocated inside the three-phase triangleγ-TiAl + β + τ and very close to theboundary tie-line γ-TiAl + β. Therefore, theamount of the τ phase is very small and thusnot detectable by the microstructuralinvestigation methods used (SEM/EDS,EPMA/WDS and XRD). Another explana-tion for the absence of the τ phase may bethat the variations in the chemicalcomposition due to the inhomogeneities inthe prepared sample lead to differences inthe observed phase equilibria. Since differentpieces of the sample were used for theEPMA/WDS and DSC investigations, somediscrepancies could be detected.

336

Based on the DSC measurements (see Fig. 4),the reaction temperature of the transitionreaction γ-TiAl + β ↔ τ + C14 wasdetermined to be 1354 K, which is 31 Kabove the reaction temperature reported byCupid et al. [32]. The reaction temperatureof the alloy 43Al-33Ti-24Cr (#3) was foundto be shifted toward slightly higher values(as compared to 1354 K). This effect can beexplained by a slower system responsecaused by the system inertia due to theabout half lower sample mass of alloy43Al-33Ti-24Cr (#3) compared to the alloys53Al-33Ti-14Cr (#1) and 47Al-33Ti-20Cr(#2). The peaks appearing in the hightemperature part of the DSC curves (seeFig. 4) are related to the melting of the alloys.Thus, the first deflection from the baselinecorresponds to the solidus temperature ofthe respective alloys. The liquidustemperatures can be assigned to the point ofinflection on the measured DSC curve justbefore the DTA curve returns to baseline.

Figure 4: DSC heating curves composed of two measurements of the alloys 53Al-33Ti-14Cr, 47Al-33Ti-20Cr and 43Al-33Ti-24Cr annealed at 1293 and 1423 K, respectively. The DSC measurements were performed in Aratmosphere at the heating rates of 10 K/min. The displayed transition temperatures were determined from the onsets or peaks of the 1st derivative of the measured curves.

Figure 5: Constructed partial vertical section in the ternary Al-Ti-Cr system at a constant Ti content of 33 at.%. The green lines represent the DSC heating curve of alloy 53Al-33Ti-14Cr (#1) composed of two DSC measurements of the annealed samples at 1293 and 1423 K (horizontal gray lines).

337

Figure 6: Constructed partial vertical section in the ternary Al-Ti-Cr system at a constant Ti content of 33 at.%. The green lines represent the DSC heating curve of alloy 47Al-33Ti-20Cr (#2) composed of two DSC measurements of the annealed samples at 1293 and 1423 K (horizontal gray lines).

Also for alloy 43Al-33Ti-24Cr (#3), themeasured DSC heating curves were used toconstruct the isopleth at 33 at.% Ti (seeFig. 7). The initial peak at 1354 Kcorresponding to the transition reactionγ-TiAl + β ↔ τ + C14 was detected to possessthe highest heat effect (see Fig. 4). Thisindicates that the intersection point of thistransition reaction should be located closer tothe chemical composition of 43Al-33Ti-24Cr(#3) than to the other alloy compositions. Thevertical shift in the DSC curve at 1491 K(see Fig. 7) indicates a change in the numberof phases because the measured DSC curvereturns to the baseline. In this particular case,the γ-TiAl phase disappears. This alloy wasfound to be located inside the γ-TiAl + βtwo-phase region at 1423 K and inside the βsingle phase region up to 1491 K. This meansthat the homogeneity range of the β phase isextended at higher temperatures. Therefore,the precipitation behavior found in aprevious work about the investigation of theisothermal section at 1473 K [11] wasconfirmed in the present work. Theappearance of the first portion of liquid wasobserved at 1611 K. At 1651 K, alloy43Al-33Ti-24Cr (#3) was completely melted.

Also for alloy 43Al-33Ti-24Cr (#3), themeasured DSC heating curves were used toconstruct the isopleth at 33 at.% Ti (seeFig. 7). The initial peak at 1354 Kcorresponding to the transition reactionγ-TiAl + β ↔ τ + C14 was detected to possessthe highest heat effect (see Fig. 4). Thisindicates that the intersection point of thistransition reaction should be located closer tothe chemical composition of 43Al-33Ti-24Cr(#3) than to the other alloy compositions. Thevertical shift in the DSC curve at 1491 K(see Fig. 7) indicates a change in the numberof phases because the measured DSC curvereturns to the baseline. In this particular case,the γ-TiAl phase disappears. This alloy wasfound to be located inside the γ-TiAl + βtwo-phase region at 1423 K and inside the βsingle phase region up to 1491 K. This meansthat the homogeneity range of the β phase isextended at higher temperatures. Therefore,the precipitation behavior found in aprevious work about the investigation of theisothermal section at 1473 K [11] wasconfirmed in the present work. Theappearance of the first portion of liquid wasobserved at 1611 K. At 1651 K, alloy43Al-33Ti-24Cr (#3) was completely melted.

338

Figure 7: Constructed partial vertical section in the ternary Al-Ti-Cr system at a constant Ti content of 33 at.%. The green lines represent the DSC heating curve of alloy 43Al-33Ti-24Cr (#3) composed of two DSC measurements of the annealed samples at 1293 and 1423 K (horizontal gray lines).

(1)

(2)

The results of the DSC measurements showthat depending on the chemical compositionof the alloys the heat effects differ (seeFig. 8). The reaction enthalpy reaches itsmaximum in the intersection point of thetie-lines γ-TiAl + β and τ + C14. At thispoint, the τ + C14 two-phase microstructuretransforms completely into the γ-TiAl + βtwo-phase microstructure on heating. For allother chemical compositions within theplane of the transition reaction, fractions ofone or two respective phases are in excessand thus do not participate in the invariantreaction. Depending on the alloy positioninside the plane of the transition reaction,the calculated amounts of phases change,which means that only limited phaseamounts are available for the reaction. Thetransition reaction takes place alwaysaccording to the lever rule applied on the tie-lines γ-TiAl + β and τ + C14 with therespective intersection point. Thus, based onthe calculated phase fractions of the alloys(nγ, nβ, nτ and nC14) slightly below and abovereaction temperature T(Uinv) the phaseamounts taking part in the reaction n(Uinv)can be calculated as follows:

inside the plane of the transition reaction,the calculated amounts of phases change,which means that only limited phaseamounts are available for the reaction. Thetransition reaction takes place alwaysaccording to the lever rule applied on the tie-lines γ-TiAl + β and τ + C14 with therespective intersection point. Thus, based onthe calculated phase fractions of the alloys(nγ, nβ, nτ and nC14) slightly below and abovereaction temperature T(Uinv) the phaseamounts taking part in the reaction n(Uinv)can be calculated as shown below whereequation 1 is valid for the cooling andequation 2 for the heating process.

339

Figure 8: DSC heating curves of alloys 53Al-33Ti-14Cr (#1), 47Al-33Ti-20Cr (#2) and 43Al-33Ti-24Cr (#3) heat treated at 1293 K. The chemical compositions of respective alloys shown in the figure were measured using EPMA/WDS. The peaks were integrated using the sigmoidal baseline subtraction method.

340

Using the EPMA/WDS results of theannealing experiments at 1293 and 1423 K,the plane of the transition reaction wasconstructed and the chemical compositionsof respective phases taking part in thereaction were determined (see Fig. 9). Usingthese phase compositions and the chemicalcompositions of the alloys measured in theDSC, the phase amounts taking part in thereaction were calculated and correlated withthe heat effects (see table 3).

Figure 9: Partial isothermal section at 1354 K (temperature of the reaction γ-TiAl + β↔ τ + C14) showing the chemical and phase compositions of the samples investigated by DSC (green and red circles, respectively).

Alloy composition [at.%] Phase composition [mol.%] Reacting phase

amount [mol.%]

Reaction enthalpy[J/mol]Al Ti Cr γ β τ C14

T > T (Uinv)

52.9 33.3 13.8 53.9 19.5 26.6 0 32.6 100252.5 33.5 14.0 59.9 21.7 18.4 0 36.3 113248.4 33.6 18.0 54.0 24.3 0 21.7 40.7 241742.7 33.1 24.2 19.7 16.0 0 64.3 26.7 3250

T < T (Uinv)

52.9 33.3 13.8 40.8 0 41.0 18.2 32.6 100252.5 33.5 14.0 45.3 0 34.5 20.2 36.3 113248.4 33.6 18.0 37.6 0 18.0 44.4 40.7 241742.7 33.1 24.2 9.0 0 11.8 79.2 26.7 3250

Table 3: Experimental results (EPMA/WDS; DSC) of the investigation of the transition reaction γ-TiAl + β↔ τ + C14.

Using the EPMA/WDS results of theannealing experiments at 1293 and 1423 K,the plane of the transition reaction wasconstructed and the chemical compositionsof respective phases taking part in thereaction were determined (see Fig. 9). Usingthese phase compositions and the chemicalcompositions of the alloys measured in theDSC, the phase amounts taking part in thereaction were calculated and correlated withthe heat effects (see table 3).

341

It was found that the amount of the reactingphase n(Uinv) is obtained with largeuncertainty. In order to determine the phasefractions ni below and above the invariantreaction, a linear system of equations withthree unknowns need to be solved for bothtemperatures T > T(Uinv) and T < T(Uinv),where:T > T(Uinv):

T < T(Uinv):

With , , and the chemicalcompositions (i = Al, Ti, Cr) of the γ-TiAl, β,τ and C14 phases, respectively, and ,and the chemical compositions of thesamples. Assuming an experimentaluncertainty of 1 % for the chemicalcompositions of the phases taking part in thereaction (γ-TiAl, β, τ and C14), and anuncertainty of 0.5 % of the overall chemicalcompositions of the measured samples, theerror of the determined reacting phaseamount was calculated using the errorpropagation law. The results show (seeFig 10) that quite large experimentaluncertainties between 20 and 35 % arise andthat the accuracy of the reacting phaseamount strongly depends on the accuracy ofthe determined plane of reaction and themeasured chemical alloy compositions. Thereaction enthalpy was found to be 5211 J/molwhich is in reasonable agreement with thereaction enthalpy of 5653 J/mol calculatedusing the thermodynamic descriptiondeveloped by Cupid et al. [32].

In order to develop thermodynamicdatabases, which enable the extrapolation tohigher-order system, the usage ofexperimental phase equilibria data incombination with thermodynamic data suchas enthalpies, entropies or the heat capacitiesof the respective phases is essential. Eventhough the determined value of the reactionof the transition reaction is large, theuncertainties can be taken into accountduring the optimization process to improveexisting thermodynamic descriptions of theternary Al-Ti-Cr system.

With , , and the chemicalcompositions (i = Al, Ti, Cr) of the γ-TiAl, β,τ and C14 phases, respectively, and ,and the chemical compositions of thesamples. Assuming an experimentaluncertainty of 1 % for the chemicalcompositions of the phases taking part in thereaction (γ-TiAl, β, τ and C14), and anuncertainty of 0.5 % of the overall chemicalcompositions of the measured samples, theerror of the determined reacting phaseamount was calculated using the errorpropagation law. The results show (seeFig 10) that quite large experimentaluncertainties between 20 and 35 % arise andthat the accuracy of the reacting phaseamount strongly depends on the accuracy ofthe determined plane of reaction and themeasured chemical alloy compositions. Thereaction enthalpy was found to be 5211 J/molwhich is in reasonable agreement with thereaction enthalpy of 5653 J/mol calculatedusing the thermodynamic descriptiondeveloped by Cupid et al. [32].

In order to develop thermodynamicdatabases, which enable the extrapolation tohigher-order system, the usage ofexperimental phase equilibria data incombination with thermodynamic data suchas enthalpies, entropies or the heat capacitiesof the respective phases is essential. Eventhough the determined value of the reactionof the transition reaction is large, theuncertainties can be taken into accountduring the optimization process to improveexisting thermodynamic descriptions of theternary Al-Ti-Cr system.

(3)

(4)

It was found that the amount of the reactingphase n(Uinv) is obtained with largeuncertainty. In order to determine the phasefractions ni below and above the invariantreaction, a linear system of equations withthree unknowns need to be solved for bothtemperatures T > T(Uinv) and T < T(Uinv),where:

342

Figure 10: Extrapolation of the reaction enthalpy ∆Hm(Uinv) from the resulting phase fractions taking part in the invariant reaction n(Uinv).

The partial isopleth at a constant Ti contentof 33 at.% was constructed as based on theexperimental results observed for alloys53Al-33Ti-14Cr (#1), 47Al-33Ti-20Cr (#2)and 43Al-33Ti-24Cr (#3), which wereannealed at 1293 and 1423 K. Theobservations showed that the slope of theγ-TiAl + β +τ solvus line is very steep and thatthe γ-TiAl + β + C14 region is very narrow.The narrowness of the γ-TiAl + β + C14phase region indicates that the two-phaseequilibria γ-TiAl + β and β + C14 intersectslightly above the transition reaction. Due tothe extension of the homogeneity range of theβ phase with increasing temperature up to theinvariant reaction U4: L + γ-TiAl ↔ τ + β at1570 K, the precipitation behavior found in aprevious work about the investigation of theisothermal section at 1473 K [11] wasconfirmed.

The transition reaction γ-TiAl + β ↔ τ + C14was investigated for the first time. It wasfound that this transition reaction takes placeat 1354 K, which is in reasonable agreementwith 1323 K predicted by Cupid et al. [32] onthe basis of a thermodynamic assessment. Inaddition, microstructures were shown slightlyabove/below this reaction, the reaction planewas constructed based on EPMA/WDSresults, and an extrapolation method todetermine the reaction enthalpy wasintroduced. The enthalpy of the transitionreaction γ-TiAl + β ↔ τ + C14 wasextrapolated based on the constructedreaction plane and enthalpy measurementsusing DSC. The reaction enthalpy of∆Hm(Uinv) = 5211 J/mol was determined,which is 8 % lower than the value calculatedusing the database from Cupid et al. [32].Thus, the enthalpy of the transition reactionγ-TiAl + β ↔ τ + C14 is in reasonableagreement with Cupid et al. [32] despite alarge uncertainty of the extrapolated reactionenthalpy and although the precision of thedetermined value strongly depends on theaccuracy of the determined composition ofthe reaction plane, the alloy compositionsand on the accuracy of the DSCmeasurements.

The partial isopleth at a constant Ti contentof 33 at.% was constructed as based on theexperimental results observed for alloys53Al-33Ti-14Cr (#1), 47Al-33Ti-20Cr (#2)and 43Al-33Ti-24Cr (#3), which wereannealed at 1293 and 1423 K. Theobservations showed that the slope of theγ-TiAl + β +τ solvus line is very steep and thatthe γ-TiAl + β + C14 region is very narrow.The narrowness of the γ-TiAl + β + C14phase region indicates that the two-phaseequilibria γ-TiAl + β and β + C14 intersectslightly above the transition reaction. Due tothe extension of the homogeneity range of theβ phase with increasing temperature up to theinvariant reaction U4: L + γ-TiAl ↔ τ + β at1570 K, the precipitation behavior found in aprevious work about the investigation of theisothermal section at 1473 K [11] wasconfirmed.

The transition reaction γ-TiAl + β ↔ τ + C14was investigated for the first time. It wasfound that this transition reaction takes placeat 1354 K, which is in reasonable agreementwith 1323 K predicted by Cupid et al. [32] onthe basis of a thermodynamic assessment. Inaddition, microstructures were shown slightlyabove/below this reaction, the reaction planewas constructed based on EPMA/WDSresults, and an extrapolation method todetermine the reaction enthalpy wasintroduced. The enthalpy of the transitionreaction γ-TiAl + β ↔ τ + C14 wasextrapolated based on the constructedreaction plane and enthalpy measurementsusing DSC. The reaction enthalpy of∆Hm(Uinv) = 5211 J/mol was determined,which is 8 % lower than the value calculatedusing the database from Cupid et al. [32].Thus, the enthalpy of the transition reactionγ-TiAl + β ↔ τ + C14 is in reasonableagreement with Cupid et al. [32] despite alarge uncertainty of the extrapolated reactionenthalpy and although the precision of thedetermined value strongly depends on theaccuracy of the determined composition ofthe reaction plane, the alloy compositionsand on the accuracy of the DSCmeasurements.

4 Summary and Conclusion

343

The partial isopleth at a constant Ti contentof 33 at.% was constructed as based on theexperimental results observed for alloys53Al-33Ti-14Cr (#1), 47Al-33Ti-20Cr (#2)and 43Al-33Ti-24Cr (#3), which wereannealed at 1293 and 1423 K. Theobservations showed that the slope of theγ-TiAl + β +τ solvus line is very steep and thatthe γ-TiAl + β + C14 region is very narrow.The narrowness of the γ-TiAl + β + C14phase region indicates that the two-phaseequilibria γ-TiAl + β and β + C14 intersectslightly above the transition reaction. Due tothe extension of the homogeneity range of theβ phase with increasing temperature up to theinvariant reaction U4: L + γ-TiAl ↔ τ + β at1570 K, the precipitation behavior found in aprevious work about the investigation of theisothermal section at 1473 K [11] wasconfirmed.

The transition reaction γ-TiAl + β ↔ τ + C14was investigated for the first time. It wasfound that this transition reaction takes placeat 1354 K, which is in reasonable agreementwith 1323 K predicted by Cupid et al. [32] onthe basis of a thermodynamic assessment. Inaddition, microstructures were shown slightlyabove/below this reaction, the reaction planewas constructed based on EPMA/WDSresults, and an extrapolation method todetermine the reaction enthalpy wasintroduced. The enthalpy of the transitionreaction γ-TiAl + β ↔ τ + C14 wasextrapolated based on the constructedreaction plane and enthalpy measurementsusing DSC. The reaction enthalpy of∆Hm(Uinv) = 5211 J/mol was determined,which is 8 % lower than the value calculatedusing the database from Cupid et al. [32].Thus, the enthalpy of the transition reactionγ-TiAl + β ↔ τ + C14 is in reasonableagreement with Cupid et al. [32] despite alarge uncertainty of the extrapolated reactionenthalpy and although the precision of thedetermined value strongly depends on theaccuracy of the determined composition ofthe reaction plane, the alloy compositionsand on the accuracy of the DSCmeasurements.

This project is a part of the Cluster ofExcellence “Structure Design of Novel High-Performance Materials via Atomic Designand Defect Engineering (ADDE)”, which isfinancially supported by the European Union(European Regional Development Fund) andby the Ministry of Science and Art of Saxony(SMWK).

Acknowledgments

References

1. R.W. Hayes and P.L. Martin, Tension creep of wrought single phase γ-TiAl, Acta Metall. Mater., 1995, 43(7), p. 2761-27722. C.T. Liu, et al., Tensile properties and fracture toughness of TiAl alloys with c ontrolled microstructures, Intermetallics, 1996, 4, p. 429-4403. C. Mercer, J. Lou, and W.O. Soboyejo, An investigation of fatigue crack growth in a cast lamellar Ti-48Al-2Cr-2Nb alloy, Mater. Sci. Eng., A, 2000, 284, p. 235-2454. A. Rahmel, M. Schuetze, and W.J. Quadakkers, Fundamentals of TiAl oxidation - A critical review, Mater. Corros., 1995, 46(5), p. 271-2855. M. Yamaguchi, H. Inui, and K. Ito, High-temperature structural intermetallics, Acta Mater., 2000, 48(1), p. 307-3226. W.O. Soboyejo and C. Mercer, On the Influence of Thermomechanical Processing on the Fracture Behavior of γ-Based Titanium Aluminides. II. Effects of Alloying with Cr and Nb, Mater. Manuf. Processes, 1996, 11(3), p. 431-4487. D.B. Lee, et al., Effect of Cr on the high temperature oxidation of L12-type Al3Ti intermetallics, Mater. Sci. Eng., A, 2000, 290(1-2), p. 1-58. T.J. Jewett, B. Ahrens, and M. Dahms, Stability of TiAl in the Ti-Al-Cr System, Mater. Sci. Eng. A, 1997, 225, p. 29-379. H. Xu, Z. Jin, and R. Wang, Study on the Phase Equilibria of the Al-Cr-Ti System at 1050°C, Scr. Mater., 1997, 37, p. 1469-147310. H. Chen, et al., Reaction Scheme and Liquidus Surface of the Ternary System Aluminum-Chromium-Titanium, Metall. Mater. Trans. A, 2009, 40(12), p. 2980-298611. M.J. Kriegel, et al., Phase equilibria at 1473 K in the ternary Al-Cr-Ti system, J. Alloys Compd., 2013, 550, p. 519-525 12. O.M. Barabash, et al., Formation of periodic microstructures involving L12 phase in eutectic Al-Ti-Cr alloys, Intermetallics, 2003, 11, p. 953-96213. M. Ichimaru, N. Mori, and Y. Miura, Al-Ti-Cr ternary phase diagram and solidifica tion structure of (Al,Cr)3Ti alloys, J. Jpn. Inst. Light Metals, 2001, 51, p. 640-64514. I.I. Kornilov, V.S. Mikheev, and T.S. Chernova, Investigations of Phase Diagram of the Ti-Cr-Al System, Zh. Neorg. Khim., 1958, 3, p. 786-79615. M.J. Kriegel, et al., Constitution of the liquidus and solidus surfaces of the Al–Ti–Cr system, J. Alloys Compd., 2014, 584, p. 438-44616. H. Mabuchi, et al., Microstructure and Mechanical Properties of Ternary L12 Inter metallic Compound in Al-Ti-Cr System, Mater. Trans., JIM, 1997, 38, p. 560-565

344

17. N. Mori and K. Ogi, Solidification-Path and Microstructure of Al-Ti-Cr Alloy Analyzed by Progressive-Type Solidification Equation, Mater. Trans., JIM, 2003, 44(11), p. 2334-233818. Y.I. Rusnyak, E.F. Kazakova, and T.P. Loboda, Phase equilibria in the Al-Cr-Ti system, Met. Sci. Heat Treat., 2009, 51(9-10), p. 433-43519. G. Shao and P. Tsakiropoulos, Solidification Structures of Ti-Al-Cr Alloys, Intermetallics, 1999, 7, p. 579-58720. L.S. Busch, Transformation characteristics of 3% Al - 5% Cr titanium alloys, J. Metals, 1953, 5, p. 146-14921. T.V. Tagunova, The Ti-Cr-Al System, Zh. Neorg. Khim., 1958, 3, p. 815-81922. H.M. Rietveld, Line profiles of neutron powder-diffraction peaks for structure refinement, Acta Cryst., 1967, 22(1), p. 151-15223. H.M. Rietveld, A profile refinement method for nuclear and magnetic structures, J. Appl. Cryst., 1969, 2(2), p. 65-7124. L. Lutterotti, S. Matthies, and H.R. Wenk, MAUD: A friendly Java program for mate rial analysis using diffraction, IUCr: Newsletter of the CPD, 1999, 21, p. 14-1525. W.J. Boettinger, et al., Recommended Practice Guide: DTA & Heat-Flux DSC Measu rement of Alloy Melting and Freezing, NIST Special Publication 960-15, 200626. L.M. Baklanova, et al., Crystal structure of Chromium below Neel temperature, Dokl. Akad. Nauk SSSR, 1977, 237, p. 7927. P. Duwez and J.L. Taylor, Crystal structure of TiAl, Trans. Am. Inst. Min. Metall. Pet. Eng., 1952, 194, p. 70-7128. N. Schmitz-Pranghe and P. Duenner, Gitterstruktur und thermische Ausdehnung der Übergangsmetalle Scandium, Titan, Vanadin und Mangan, Z. MetaIlkd., 1968, 59, p. 377-38229. J.C. Schuster and H. Ipser, Phases and phase relations in the partial system TiAl3-TiAl, Z. MetaIlkd., 1990, 81(6), p. 389-39630. G.J. Sparks, et al., Formation of Cubic L12 Phases from Al3Ti and Al3Zr by Transiti on Metal Substitutions for Al, Mat. Res. Soc. Symp. Proc., 1991, 186, p. 175-18031. V.N. Svechnikov, et al., Three Modifications of TiCr2, Dopov. Akad. Nauk Ukr. RSR, 1970, 32, p. 837-84232. D.M. Cupid, et al., Thermodynamic assessment of the Cr-Ti and first assessment of the Al-Cr-Ti systems, Intermetallics, 2011, 19(8), p. 1222-1235

345

List of authors

Abendroth, Barbara

Abendroth, Martin

Baehtz, Carsten

Barchuk, Mykhailo

Behm, Thomas

Bertau, Martin

Biermann, Horst

Bregolin, Felipe Lipp

Butnitzki, Michael

Chaves, Humberto

Chmelik, David

Czettl, Christoph

Dittrich, Rosemarie1

Fabrichnaya, Olga

Förster, Sebastian

Freudenberger, Jens

Funke, Claudia

Galindo, Vladimir

Gemming, Sibylle

Geyer, Maximilian

Groh, Sebastién

Günthel, Michael

Hahn, Torsten

Hanzig, Florian

Hanzig, Juliane

Heger, Dietrich

Heide, Gerhard

Heitmann, Johannes

Hoffmann, Markus

Hübscher, Jörg

Jachalke, Sven

Joseph, Yvonne

Kämpfe, Alexander

Katzsch, Felix

Kauffmann, Alexander

Kawalla, Rudolf

Keller, Kevin

Klemm, Volker

Knauer, Enrico

Kortus, Jens

Kriegel, Mario J.

Krockert, Katja

Kroke, Edwin

Krüger, Lutz

Kuna, Meinhard

Lehninger, David

Leonhardt, Michael

Liebing, Simon

346

118

288

184

78

26, 42

64

224

6

2

224

328

200

146, 166

328

146

278

26

26

118

134

306

166

146

100

100

328

242

134

224

166

100

146, 166

146

166

278

260

242

134

278

78, 146, 242

328

6

146, 242

260

42, 224

134

184

146

Lukin, Gleb

Mazik, Monika

Mehner, Erik

Mertens, Florian

Meyer, Dirk C.

Michotte, Claude

Möller, Hans Joachim

Motylenko, Mykhaylo

Mühle, Uwe

Münchgesang, Wolfram

Nitzbon, Ivonne

Obst, Andreas

Pätzold, Olf

Prucnal, Slawomir

Rafaja, David

Ratayski, Ulrike

Reichelt, Stephan

Rensberg, Jura

Rentrop, Solveig

Röder, Christian

Ronning, Carsten

Roth, Stephan

Scheibe, Hans-Joachim

Schimpf, Christian

Schmerler, Steve

Schmid, Alexander

Schmid, Ekateriana

Schneider, Frank

Schucknecht, Torsten

Schultz, Ludwig

Schwarz, Friederike

Schwarz, Marcus R.

Seidel, Nadin

Seidel, Peter

Seifert, Hans-Jürgen

Šingliar, Ute

Skorupa, Wolfgang

Skupsch, Christoph

Stelter, Michael

Stöcker, Hartmut

Wagler, Jörg

Weber, Edwin

Würzner, Sindy

Wüstefeld, Christina

Zschornak, Matthias

347

78

146

100

166

100, 118

200

6, 26, 42

200

184

118

64

64

26, 78

6

78, 134, 184, 200, 242, 328

184

260

118

118

78

118

224

184

184

242

134

26

134

184

278

260

242

146

134

328

64

6

224

26, 78

100, 118

146

146, 166

26

200

100