ANTIMONIDE-BASED III-V CMOS TECHNOLOGY A DISSERTATION SUBMITTED TO THE DEPARTMENT OF ELECTIRCAL ENGINEERING AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY Ze Yuan August 2013
164
Embed
ANTIMONIDE-BASED III-V CMOS TECHNOLOGY A DISSERTATION ... Yuan Thesis.pdf · antimonide-based iii-v cmos technology a dissertation submitted to the department of electircal engineering
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
ANTIMONIDE-BASED III-V CMOS TECHNOLOGY
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF ELECTIRCAL ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Ze Yuan
August 2013
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/yh428qq2909
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Krishna Saraswat, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
James Harris
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
James Plummer
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost for Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
ABSTRACT
As scaling of silicon CMOS becomes increasingly difficult, use of alternate
materials with high carrier mobilities is being explored extensively. Despite
burgeoning interests in III-‐Vs for realizing high performance transistors at low
power, III-‐V MOSFETs have been plagued by
(a) Poor PMOS performance, which makes realizing CMOS in a single
material system difficult;
(b) Low density of states (DOS) for electrons;
(c) Poor surface passivation resulting in high interface trap density;
(d) Poor dopant activation in the source-‐drain (S/D);
(e) High contact resistance;
(f) Hetero-‐integration of III-‐V materials on silicon substrates.
This work will present novel solutions to overcome these problems. The design
is based on the 0.61-‐0.62nm lattice constant system with InGaSb as the channel
material because of its advantages in terms of band engineering and high
mobility/offsets for both electrons and holes. The goal is to achieve high
electron/hole mobility in the same channel material for N-‐ and P-‐channel MOS
devices through the optimization of surface passivation, stoichiometry,
heterostructure design and novel contact/interface engineering. Hetero-‐integration
on a silicon substrate is another key challenge in enabling III-‐V CMOS. The hetero-‐
integration scheme for III-‐V CMOS transistors on silicon using the rapid-‐melt-‐
growth (RMG) technique is demonstrated.
ACKNOWLEDGMENTS
This journey would not have been possible without the help, encouragement,
and contribution of several individuals, to whom I am greatly indebted.
First of all, I would like to express my sincere gratitude to my research advisor
Prof. Krishna Saraswat for his rigorous support and guidance throughout the course
of my graduate studies and research. He has been undoubtedly one of the most kind
and helpful people I have ever interacted with at Stanford. He once told me that a
professor’s job is to not to tell students what to do, but to be surprised by the
students’ ideas and to find money for them. Over the years, he offered great degree
of freedom and the perfect blend of encouragement and guidance. It has been his
trust that made me work hard. His broad intelligence and unique insight to
semiconductor technology have been a constant source of inspiration to me. All
these energy and passion that I gained, at the end, even surprised myself, from
which I realized the beauty of the way he trains students.
I am deeply indebted to my co-‐advisor Prof. James Plummer, who has been an
excellent role model and mentor. I have learned a lot from his vast knowledge, wide
industry experience and strong technical expertise. More than knowledge, what
Prof. Plummer has taught me is how to behave professionally. In spite of his
extremely busy schedule, Prof. Plummer has always made time to give me very
thoughtful and constructive criticism on my research.
I am especially grateful to Prof. H. –S. Philip Wong, who has always been there no
matter what kinds of help I needed. I am also thankful to Prof. James Harris, Prof.
Boris Murmann for their valuable suggestions on the research and for their precious
time and efforts as members of my reading and defense committee.
I felt very fortunate to have worked with Dr. Aneesh Nainani, who has been a
mentor or even “co-‐advisor” to me in a lot of ways. I really cherish the friendship we
have built over the years and I am sure we will have opportunities to work together
once again some time in the future.
I would like to thank Dr. Brian Bennett, Brad Boos, and Dr. Mario Ancona from
Naval Research Laboratory. The experimental work on heterostructure design
would not have been possible without their support on MBE growth. They
generously shared their expertise in antimonide-‐based compound semiconductors
and provided valuable feedback and advice to my research. I would also like to
thank Dr. Albert Wang, Dr. Man-‐Hoi Wong, Dr. Wei-‐E Wang from SEMATECH, and
Prof. Ravi Droopad from the University of Texas at Dallas for the support on
material growth for the project on rapid-‐melt-‐growth.
During my graduate studies at Stanford, I have been fortunate to gain valuable
experience from internships. I would like to thank Dr. Don D’Avanzo, Dr. Thomas
Low, Dr. Barry Wu for my internship experience at Agilent Technologies, Inc., Dr.
Robert Visser, Dr. Bhaskar Kumar for giving me the opportunity to work at Applied
Materials Inc. and Dr. Jing Wang for hiring me at SuVolta, Inc.
My experimental research work relied heavily on the use of the Stanford
Nanofabrication Facility. I really appreciate the great effort of all the SNF staff on
training lab users and maintaining equipment.
I consider myself lucky to have worked with a lot of talented people in and out of
Stanford, from whom I have learnt so much. I would like to acknowledge the
Saraswat group, the Plummer group, the Wong group, the Harris group, the
McIntyre group and the Nishi group for sharing their knowledge and insights with
me. Specifically, I would like to thank Dr. Blanka Magyari-‐Kope, Dr. Peter Griffin, Dr.
Tejas Krishnamohan, Dr. Ximeng Guan, Dr. Yuan Zhang, Hai Wei, Dr. Zihong Liu, Dr.
Yijie Huo, Yi Wu, Dr. Jiale Liang, Dr. Jenny Hu, Chien-‐Yu Chen, Archana Kumar, Dave
Sukhdeo, Shuang Li, Yangsen Kang, Liang Dong, Robert Chen, Dr. Hai Lin, Dr.
Xiaoqing Xu, Rakesh Jeyasingh, Aryan Hazeghi, Dr. Jaesoo Ahn, Dr. Cynthia Ginestra,
Dr. Sun Yun, Shimeng Yu, Liang Zhao and Henry Hony-‐Yu Chen, etc.; all the past and
present group members, Dr. Shyam Raghunathan, Dr. Kyung Hoae Koo, Dr. Gunhan
Ersosum, Dr. Donghyum Kim, Dr. Duygu Kuzum, Dr. Yeul Na, Dr. Arunanshu Roy, Dr.
Crystal Kenney, Dr. Bin Yang, Woo-‐Shik Jung, Suyog Gupta, Donguk Nam, Ju Hyung
Nam, Gautam Shine, Dr. Jason Lin, Ashish Pal, Shashank Gupta, Raisul Islam, Dr. Shu-‐
Lu Chen, Dr. Serene Koh, Xue Bai, etc.
No words would ever do justice to express my deepest thanks and gratitude to
my girlfriend, Christina Yu Zhao, my friends and my family. I will be eternally
grateful to them for being with me at all times. Their continuous love, sacrifice,
support and encouragement have allowed me to pursue my ambitions. I dedicate
this work to them.
TABLE OF CONTENTS
List of tables ..................................................................................................................... iii
List of figures ................................................................................................................... iv
[34] R. Oxland, et al., “An ultralow-‐resistance ultrashallow metallic source/drain
contact scheme for III-‐V NMOS,” Electron Device Letters, vol. 33, no. 4 pp. 501-‐503,
Apr. 2012.
[35] M. V. Fischetti, et al., “Simulation of electron transport in high-‐mobility
MOSFETs: density of states bottleneck and source starvation,” in Proc. IEDM 2010,
Dec. 2010.
[36] A. Rahman, G. Klimeck, M. Lundstrom, “Novel channel materials for ballistic
nanoscale MOSFETs-‐bandstructure effects,” in Proc. IEDM 2005, Dec. 2005.
20
[37] R. Pillarisetty, et al., “High mobility strained germanium quantum well field
effect transistor as the p-‐channel device option for low power (Vcc=0.5V) III-‐V
CMOS architecture,” in Proc. IEDM 2010, Dec. 2010.
[38] L. Witters, et al., “Dual-‐channel technology with cap-‐free single metal gate for
high performance CMOS in gate-‐first and gate-‐last integration,” in Proc. IEDM 2011,
Dec. 2011.
[39] R. Zhang, T. Noriyuki, P.-‐C. Huang, M. Takenaka, S. Takagi, “1-‐nm thick EOT thin
mobility Ge n-‐ and p-‐MOSFETs with ultrathin GeOx/Ge MOS interfaces fabricated by
plasma post oxidation,” in Proc. IEDM 2011, Dec. 2011.
[40] M. Kobayashi, et al., “Uniaxial stress engineering for high-‐performance Ge
NMOSFETs,” IEEE Transactions on Electron Devices, vol. 57, no. 5, 2010.
[41] S. Gupta, et al., “Towards high mobility GeSn channel nMOSFETs: Improved
surface passivation using novel ozone oxidation method,” in Proc. IEDM 2012, Dec.
2012.
[42] M. Yokoyama, et al., “CMOS integration of InGaAs nMOSFETs and Ge pMOSFETs
with self-‐align Ni-‐based metal S/D using direct wafer bonding,” in Proc. VLSIT 2011,
Jun. 2011.
[43] D. Lin, et al., “Enabling the high-‐performance InGaAs/Ge CMOS: a common gate
stack solution,” in Proc. IEDM 2009, Dec. 2009.
[44] B. R. Bennett, R. Magno, J. B. Boos, W. Kruppa, M. G. Ancona, “Antimonide-‐based
compound semiconductors for electronic devices: A review,” Solid-‐State Electronics,
vol. 49, pp. 1875-‐1895, 2005.
[45] B. R. Bennett, M . G. Ancona, J. B. Boos, B. V. Shanabrook, “Mobility enhancement
in strained p-‐InGaSb quantum wells,” Appl. Phys. Lett., vol. 91, p. 042104, 2007.
21
[46] L. Xia, et al., “Performance enhancement of p-‐channel InGaAs quantum-‐well
FETs by superposition of process-‐induced uniaxial strain and epitaxially-‐grown
biaxial strain,” in Proc. IEDM 2011, Dec. 2011.
[47] M. Yokoyama, et al., “Extremely-‐thin-‐body InGaAs-‐on-‐insulator MOSFETs on Si
fabricated by direct wafer bonding,” in Proc. IEDM 2010, Dec. 2010.
[48] H. Ki, et al., “Ultrathin compound semiconductor on insulator layers for high-‐
performance nanoscale transistors,” Nature, vol. 468, pp. 286-‐289, 2010.
[49] N. Waldron, et al., “Integration of InGaAs channel nMOS devices on 200mm Si
wafers using the aspect-‐ratio-‐trapping technique,” ECS Trans., vol. 45, no. 4, pp. 115-‐
128, 2012.
CHAPTER 2: SURFACE PASSVIATION IN ANTIMONIDE-‐BASED
METAL-‐OXIDE-‐SEMICONDUCTOR DEVICES
High interface trap density (Dit) at high-‐k dielectric/III-‐V interface is considered
one of the main challenges for antimonide-‐based III-‐V metal-‐oxide-‐semiconductor
field-‐effect-‐transistors (MOSFETs). In this chapter, we have investigated the effects
of oxidant dosing on GaSb (001) surface prior to atomic layer deposition (ALD). It is
shown that in situ H2O pre-‐pulsing leads to the formation of GaOx and SbOx and
increase in the oxygen content at the interface, identified by X-‐ray photoelectron
spectroscopy (XPS) analysis. Such interfacial layer leads to a reduction in Dit, and a
Dit of 5×1011cm-‐2 near the valence band edge of GaSb is achieved.
In the second part, we have investigated the suppression of interface state
response using band engineering in III-‐V quantum well MOSFETs and
experimentally verified the concept in the antimonide materials system using a
gate-‐stack consisting of Al2O3/GaSb/InAlSb. It is shown that if the thickness of the
interfacial layer of GaSb is scaled down to a few monolayers, the effective bandgap
of the interfacial layer increases dramatically due to quantum confinement, which
leads to the suppression of interface-‐trap response.
2.1 EFFECTS OF OXIDANT DOSING ON GASB (001) PRIOR TO ATOMIC LAYER
DEPOSITION OF AL2O3
Antimonide-‐based compound semiconductors have recently emerged as
potential candidate for replacement of silicon in future high-‐performance, low-‐
power complementary metal-‐oxide-‐semiconductor (CMOS) technologies, due to its
excellent electron and hole transport properties [1-‐6]. Both n-‐ and p-‐channel
devices of high performance have been demonstrated [2-‐9]. However, the highly
23
reactive nature of the surface of anitmonide compound has led to main challenges in
achieving good passivation with high-‐k dielectrics and minimizing the interface trap
density, which are critical for maintaining high carrier mobility.
Atomic layer deposition of high-‐k dielectrics has been widely used to passivate
surfaces of antimonide compounds. Natively grown III-‐V oxides are of complex
structure and composition, thus pin the Fermi-‐level at the interface. Wet chemical
cleans [8-‐11] and in situ hydrogen plasma exposure [12] prior to atomic layer
deposition (ALD) are adopted to clean such GaOx and SbOx layers of poor quality. On
the other hand, silicon-‐silicon dioxide, germanium-‐germanium dioxide interfaces
are proven to give high-‐quality dielectric/semiconductor interfaces [13, 14].
Theoretical calculation shows high interfacial oxygen content at high-‐k
dielectric/GaSb interface can eliminate any gap states [15]. Experimentally, SbOx
grown during atomic layer deposition is found to be important for the modulation of
the Fermi-‐level at the interface [10], which is achieved by low processing thermal
budget and plasma-‐enhanced atomic layer deposition (PEALD) of dielectrics. These
suggest the formation of high oxygen content interfacial oxide layer in a controllable
way is the key to good modulation of charge on GaSb.
In this section, we report an in situ surface treatment process for GaSb surface
prior to ALD deposition of Al2O3. Pre-‐ALD oxidant exposure at low temperature is
used to form high-‐quality interfacial oxide layer. Change in chemical bonding due to
pre-‐ALD oxidant exposure of the substrate surface is investigated by X-‐ray
photoelectron spectroscopy (XPS). It is found that high oxygen content at the
interface could be achieved by careful control over initial nucleation steps. The
spectroscopy results are correlated with frequency dispersion of MOS capacitors,
conductance measurement and extracted interface trap density (Dit).
24
After degreasing with solvent, both n-‐ (Te doped, 5×1017cm-‐3) and p-‐type
(undoped, ~1×1017cm-‐3 due to point defects) GaSb bulk substrates were cleaned
using 1:1 diluted HCl for 5 min to remove native GaOx and SbOx, blown dried with N2
gas, and then immediately transferred into ALD reactor. Subsequently, the wafers
were heated up to 400°C to further remove residual oxide by thermal desorption
[6].
Prior to ALD of Al2O3 using trimethyl-‐aluminum (TMA) and H2O vapor as
precursors, the samples were subjected to different pre-‐ALD pulsing treatments.
Poor quality of the high-‐k/GaSb interface has been partly attributed to the presence
of Sb-‐Sb dimers that create gap states [10]. Initial dosing of TMA is expected to form
dimethyl-‐aluminum (DMA) by dissociative chemisorption, which bonds to Sb
surface atoms, similar to the case of arsenide compounds [16,17]. Full coverage of
DMA is thus desired to minimize Sb-‐Sb metallic bonding states. This was achieved
by pre-‐pulsing TMA for 10 cycles at 200°C. Such temperature was chosen as it has
been shown that the optimal temperature for ALD of Al2O3 on GaSb is below 200°C
[9,10]. Pre-‐pulsing of H2O for 10 cycles at 100°C, was performed to terminate the
remaining dangling bonds with hydroxyl groups and lead to the formation of GaOx
and SbOx. Low substrate temperature was chosen to facilitate such chemical
reaction and prevent deep oxidation [17].
Border traps in Al2O3 near the dielectric/semiconductor interface could lead to
tunneling of carriers into defect states, thus degrades the device performance
[18,19]. Therefore, after this step, the temperature was raised back to 270°C for the
subsequent deposition of Al2O3 in order to guarantee its good quality. 8 nm / 12 nm
Al2O3 was grown by alternatively pulsing of TMA and H2O for 80 / 120 cycles. The
thickness was measured by ellipsometer. 100 nm W was sputtered in a different
chamber. W was chosen due to its good thermal stability, thus allowing post-‐
metallization annealing at high temperature. Post-‐metallization annealing was done
25
at 350°C in forming gas (5% H2/95% N2) to remove fixed charge in the Al2O3 and
further improve its quality [8]. Circular MOS capacitor structures with diameter of
50 μm were patterned by standard lithographic techniques and W was etched by
SF6-‐O2-‐based plasma dry-‐etch. Backside substrate contacts were made by e-‐beam
evaporation of 100 nm Au. Control samples went through the same process except
for H2O pre-‐pulsing.
Figure 2.1 Angle resolved XPS spectra (tilted at 25°) comparing the concentration of (a) GaOx and (b) SbOx in samples with and without H2O pre-‐pulsing following TMA pre-‐pulsing. Increase in GaOx/Ga and SbOx/Sb ratios can be observed in sample with
H2O pre-‐pulsing.
26
Capacitance-‐voltage (C-‐V) measurements were acquired at a variety of
temperatures at frequencies ranging from 1kHz to 1MHz. For all the measurements,
the gate voltage was swept across the range-‐of-‐interest while the back contact was
grounded. The effects of H2O pre-‐pulsing on the chemical composition at the
interface were investigated using X-‐ray photoelectron spectroscopy (XPS)
measurements. Samples for XPS measurements were fabricated following identical
ALD procedure, except with only 15 cycles of TMA and H2O pulsing so that only
~1.5nm Al2O3 were deposited.
Figure 2.2 The ratio of integrated peak area between GaOx and Ga, as well as SbOx and Sb in XPS spectra after annealing at different temperature for 1 min indicates
GaOx and SbOx are thermally stable at temperature up to 400°C.
Figure 2.1 plots the angle-‐resolved XPS analysis of the Ga 3d and Sb 4d peaks
comparing samples with and without H2O pre-‐pulsing treatment. For both Ga and Sb
peaks, sample with H2O pre-‐pulsing shows much higher ratios between GaOx/Ga
and SbOx/Sb peaks. This indicates higher Ga-‐oxide and Sb-‐oxide contents at the
interface. Since the binding energy of elemental Sb is similar to that of Sb when
bonded to GaSb, the higher ratio of SbOx/Sb could also indicate the suppression of
27
elemental Sb by H2O pre-‐pulsing. This suggests the effectiveness of H2O pre-‐pulsing
in the formation of high oxygen-‐content GaOx and SbOx, which is important towards
reducing gap states [15]. This also proves that H2O pre-‐pulsing at low temperature
(100°C) is an efficient and reliable method to form SbOx, since SbOx desorption
reaction takes place significantly at temperature above 200°C [20,21].
Figure 2.3 high-‐frequency (100 kHz) capacitance voltage characteristics of W/12nm Al2O3/n-‐GaSb gate stack before and after forming gas annealing at 350°C for 1min.
It has been reported that SbOx has poor thermal stability. SbOx can react with
the GaSb surface forming elemental Sb and GaOx [10,20,21]. On the other hand,
forming gas anneal (FGA) at 350°C could remove the fixed charge effect in Al2O3 by
passivating its defects [8,18]. It is critical to passivate defects in the Al2O3 layer by
post-‐metallization anneal while preserving a high-‐oxygen content
oxide/semiconductor interface. Figure 2.2 plots the ratio of integrated peak area
between GaOx/Ga and SbOx/Sb in XPS spectra after annealing at different
temperature for 1 min. It is found that annealing at temperature up to 400°C does
28
not lead to significant change of the ratio between GaOx and Ga, as well as SbOx and
Sb, indicating GaOx and SbOx remain stable with such thermal budget.
Figure 2.4 Frequency dependence of peak accumulation capacitances shows the impact of border traps near the dielectric/semiconductor interface. Higher deposition temperature and post-‐metallization annealing lead to reduction in
frequency dispersion.
The high-‐frequency (100 kHz) capacitance-‐voltage characteristics of W/12nm
Al2O3/n-‐GaSb gate stack before and after FGA are plotted in Figure 2.3. It is shown
that flat-‐band voltage is negatively shifted and the hysteresis has been reduced from
~0.3V to ~0.05V, indicating FGA efficiently remove the fixed charge in the bulk of
the oxide and/or at the oxide semiconductor interface without de-‐composition of
SbOx. Peak capacitance in accumulation region increases slightly because of the
densification of Al2O3 or/and recovery from plasma damage during the sputtering of
W gate metal.
29
Figure 2.4 plots the accumulation capacitances at different frequencies to show
the impact of border traps, as it is known that uniform frequency dispersions in the
accumulation region can be attributed to traps in oxide near the interface [19]. It is
shown that such frequency dispersions are 4.7%/dec and 2.8%/dec on the
capacitors with Al2O3 grown at 200°C and 270°C respectively, which justifies that
higher growth temperature is important for achieving high-‐quality oxide near the
dielectric/semiconductor interface. The frequency dispersion is further suppressed
to 1.5%/dec by post-‐metallization annealing at 350°C.
To see the effect of H2O pre-‐pulsing, capacitance-‐voltage characteristics are
compared between samples with and without the surface treatment in Figure 2.5
and 2.6 for n-‐type and p-‐type substrate respectively. Less frequency dispersion and
better gate modulation (e.g. higher Cmax/Cmin ratio) can be observed for samples
with oxidant dosing for different measurement temperatures. At room temperature,
frequency dispersions in depletion region have been reduced from 5.0%/dec to
4.1%/dec and 1.3%/dec to 1.0%/dec for n-‐ and p-‐type substrates respectively.
30
Figure 2.5 Capacitance voltage characteristics of W/8nm Al2O3/GaSb gate stack as function of frequency on n-‐type substrate with and without H2O pre-‐pulsing.
31
Figure 2.6 Capacitance voltage characteristics of W/8nm Al2O3/GaSb gate stack as
function of frequency on p-‐type substrate with and without H2O pre-‐pulsing.
Figure 2.7 Cross-‐sectional HR-‐TEM image of the gate stack consists of ~7.3nm Al2O3 and W as gate metal. No clear sign of interfacial layer confirm the scalability of the
approach.
32
The thickness of the interfacial layer is confirmed with the cross-sectional HR-TEM
image of the gate stack as in Figure 2.7, which shows no clear sign of GaOx/SbOx layer.
This proves that the formation of such oxide layer is on sub-nm level, thus does not put
much challenge on further scaling of EOT.
log1
0(Fr
eque
ncy
(Hz)
)
Gp/Atq (cm−2eV−1)
−1 0 1 23
3.5
4
4.5
5
5.5
0
2
4
6
8
10x 1011
170K
a)
VG (V)
log1
0(Fr
eque
ncy
(Hz)
)
−1 −0.5 0 0.5 1 1.5 23
3.5
4
4.5
5
5.5
0.5
1
1.5
2
x 1012
300K
33
Figure 2.8 Conduction plots (Gp/ωAq) of W/8nm Al2O3/GaSb gate stack with H2O pre-‐pulsing showing the trace of Fermi-‐level movements on a) n-‐type and b) p-‐type
substrates. Measurement temperatures are indicated in the figure.
Figure 2.8 show the conductance contour (Gp/ωAq) map of n-‐ and p-‐type
samples with H2O pre-‐pulsing as a function of gate voltage and small-‐signal
frequency at different temperatures, respectively. The trajectory of the peak value of
conduction corresponds to the modulation of the Fermi level at the interface within
the bandgap of GaSb. Peaks of conductance in the accumulation and inversion
regions can be attributed to series-‐resistance or/and border traps, and the
generation of minority carriers, respectively. Interface trap density can thus be
extracted in the depletion regions for different measurement temperature.
Figure 2.9 compares the extracted Dit values for samples with and without H2O
pre-‐pulsing. The Fermi stabilization energy is 0.1eV from the valence band (VB) of
GaSb [10]. Therefore, an asymmetric distribution is expected with more Dit towards
VG (V)
log1
0(Fr
eque
ncy
(Hz)
)
Gp/Atq (cm−2eV−1)
−1 0 1 23
3.5
4
4.5
5
5.5
6
2
4
6
8
10
12
14x 1011
b)
170K
34
the conduction band (CB) edge. Reduction of Dit (>30%) by H2O pre-‐pulsing can be
observed across the bandgap of GaSb, especially for energy levels near the
conduction band edge. Near the valence band edge, a Dit value of 5×1011cm-‐2 is
obtained, which is favored for antimonide-‐based p-‐channel devices.
Figure 2.9 Comparison of extracted interface trap density (Dit) between samples with and without H2O pre-‐pulsing. H2O pre-‐pulsing efficiently reduces Dit especially
at energy levels towards the conduction band edge.
In summary, we have studied the formation of high-‐oxygen-‐content GaOx and
SbOx interfacial layer on GaSb (001) surface by oxidant dosing at low temperature
prior to atomic layer deposition. The suppression of elemental Sb and formation of
GaOx and SbOx lead to less dispersion in capacitance-‐voltage characteristics and
better gate modulation. Reduction of Dit and a Dit value of 5×1011cm-‐2 near the
valence band edge are achieved by the proposed method, which makes antimonide-‐
based compound material a good candidate for ultralow power p-‐channel devices.
35
2.2 AMELIORATION OF INTERFACE STATE RESPONSE USING BAND ENGINEERING
IN III-‐V QUANTUM WELL METAL-‐OXIDE-‐SEMICONDUCTOR FIELD-‐EFFECT
TRANSISTORS
High mobility III-‐V semiconductors have attracted increasing interest for
replacing silicon as the channel material in complementary metal-‐oxide-‐
semiconductor (CMOS) applications due to their excellent transport properties.
Field-‐effect electron/hole mobilities exceeding 6,000cm2/Vs [22,23] and
1,000cm2/Vs [24-‐26] at sheet charge densities near 1012/cm2 have been
experimentally demonstrated with InGaAs and InGaSb as the channel material
respectively. However, a main challenge remains in achieving good passivation on
III-‐V materials and minimizing the impact of interfacial states which degrade the
subthreshold slope (SS) and mobility in MOSFETs fabricated using these materials.
Common solutions have included chemical clean [27,28], insertion of Si/Ge
interfacial layer [29,30], sulfide treatment [31,32] and in-‐situ passivation [33,34].
While great progress has been made in reducing the interface trap density (Dit)
recently, the Dit numbers achieved using approaches hitherto are nowhere close to
the values obtained on silicon/SiO2 system, potentially due to interfacial dangling
bonds and dimer pairs [35,36]. In this work, we propose to ameliorate the effect of
interface states on MOSFET performance through the use of band engineering with
an ultrathin interfacial layer. We demonstrate the concept on the antimonide system
with InGaSb/InAlSb as the channel/barrier material. Ultrathin GaSb, on which good
passivation can be achieved [8] as described in the previous section, is inserted
between the barrier layer (InAlSb) and high-‐k dielectric (Al2O3), forming a quantum
well of Al2O3/GaSb/InAlSb. It is shown that the effective bandgap of the interfacial
GaSb/InAlSb layers rises dramatically with scaled thickness of GaSb due to the
quantum confinement effect. This strong confinement effect in the proposed gate
36
stack not only guarantees there is no parallel conduction path through the thin GaSb
layer, but also results in fewer carriers available near the interface to interact with
the trap states. The concept is verified by the suppression of the response to
interface-‐traps measured using the conductance method, well-‐behaved capacitance
characteristics and excellent subthreshold swing of fabricated MOSFETs.
Heterostructure stacks making use of a wide-‐bandgap barrier layer on top of a
narrow-‐bandgap high-‐mobility channel material have commonly been used in high
electron mobility transistors (HEMTs) in III-‐V materials with Schottky-‐gates. For
InGaAs and InGaSb, aluminum (Al)-‐containing InAlAs and InAlSb provide band
offsets that can confine carriers in the narrow bandgap channel in a heterostructure
design [22,25,26]. For InGaAs another option for the barrier layer has been InP
which is lattice matched to In0.53Ga0.47As, even though it has lower offset for
electrons as compared to InAlAs [22].
More recently, people have applied these technique to MOSFETs as well [22,23],
however their options for the barrier layer have been limited to InP, due to the
difficulty of achieving a quality interface between high-‐k and Al-‐containing III-‐V
semiconductors. Passivation of Al-‐containing III-‐V alloys has proven difficult mainly
due to interfacial AlOx, which was found hard to remove using chemical cleaning /
thermal annealing [11,27]. Improved carrier mobility has been observed with the
use of an InP barrier layer and has been attributed to the reduction in surface
roughness scattering and reduction in interaction with dielectric traps as the
carriers in the channel are kept way from the dielectric interface [22,23]. It has been
proven by both experiments and modeling that the advantages due to the addition
of such a barrier layer outweigh the penalty in equivalent oxide thickness due to
addition of the barrier layer; good SS and short channel effects can still be
maintained in scaled devices [37,38].
37
Figure 2.10 Band diagram and Fermi-‐level movement during MOSFET operation for a) high-‐k dielectric/bulk GaSb; b) high-‐k dielectric/thick GaSb/InAlSb/InGaSb; c) high-‐k dielectric/ultra-‐thin GaSb/InAlSb/InGaSb. In both (a) and (b), electrons (ρe)
and holes (ρh) at the interface could interact with interface defects. Quantum confinement effect becomes more dominant as the thickness of interfacial GaSb
layer scales down as in (c). A dramatic increase of subband energies results in fewer carriers near the interface. Therefore, the interaction between interface traps and
carriers is suppressed.
Figure 2.10 describes the concept behind our approach; it shows the Dit
distributions, band diagrams and Fermi-‐level movements during device operation
for a) high-‐k dielectric/bulk GaSb; b) high-‐k dielectric/thick GaSb/InAlSb/InGaSb; c)
high-‐k dielectric/ultra-‐thin GaSb/InAlSb/InGaSb. For these devices, to effectively
switch the device, the Fermi-‐level has to move from valence band edge to mid-‐gap of
the channel for p-‐channel devices, and from conduction band edge to mid-‐gap of the
channel for n-‐channel. In either case, for device a) the inversion carriers (ρe, ρh)
right at the interface will interact with (charge and discharge) the defect states,
which results in both the loss of carriers density and mobility degradation. Adding
an Al-‐containing wide bandgap cap layer shields the impact of the interfacial defects
on the carriers’ mobility. To achieve a sharp and high-‐quality high-‐k interface,
38
termination of InAlSb with GaSb has been proposed as a solution [11,39]. For GaSb,
good surface passivation has been achieved and a typical U-‐shaped Dit distribution is
observed with higher Dit level expected towards the conduction band edge [38].
Since the physical interfaces between the high-‐k dielectric/semiconductor
throughout the three cases are the same, we can assume they give similar Dit
distribution within the energy gap. When such an interfacial layer of GaSb is
inserted, a type-‐I heterostructure is formed between the high-‐k
dielectric/GaSb/InAlSb. When the thickness of the interfacial GaSb layer is larger
than a few nanometers, quantum confinement effects in such quantum wells are
marginal. As the Fermi-‐level moves during the switching of the devices, it leads to
accumulation of carriers (ρe, ρh) in the GaSb layer, thus causing a parallel conducting
path and equivalent surface channel operation. As the thickness of the interfacial
layer scales down to a few mono-‐layers, the quantum confinement effect becomes
dominant enough so that both the lowest electron and hole subband energies are far
away from the band edges. Therefore, the accumulation of carriers in the interfacial
layer can be eliminated. Meanwhile, due to the absence of carriers right at the
interface, the interaction between carriers and interface defects can be thus
potentially suppressed.
To simulate this quantum confinement effect, a tight-‐binding (TB) approach was
used to model the electronic band structure of high-‐k dielectric/GaSb/InAlSb
heterostructure [40]. Tight-‐binding parameters for ternary compound are
calculated following virtual-‐crystal approximation (VCA) and fitted to bulk band
parameters [41]. On-‐site parameters are shifted to incorporate band offset between
semiconductors. The surface orientation was set as (001), and the semiconductor
interface was considered as hydrogen passivated to eliminate all gap states.
Therefore, an infinite potential barrier was assumed at the high-‐k/GaSb interface
[42]. Hamiltonian is constructed based on the supercell approach by selecting one
representative atom in each atom layer [43].
39
Figure 2.11 Dependence of the effective bandgap (Egeff) of the GaSb/InAlSb heterostructure on the thickness of the interfacial GaSb layer (TGaSb) in terms of
monolayers. When TGaSb is smaller than a few monolayers, the effective bandgap of GaSb/InAlSb approaches that of InAlSb alone.
Figure 2.11 shows the dependence of the simulated effective bandgap (Egeff) of
the heterostructure on the thickness of the interfacial GaSb layer (TGaSb). Egeff
increases dramatically with decreasing number of the layers of Ga/Sb atoms.
Subsequently, as the thickness decreases down to a few monolayers of atoms, the
effective band gap almost approaches that of InAlSb alone. This indicates that as
long as the thickness of the interfacial GaSb is kept below a few monolayers, the
stack is partially equivalent to having just wide bandgap InAlSb alone in terms of
band energies. The benefit is that the physical interface between the high-‐k
dielectric and the semiconductor is now between the high-‐k and GaSb, thus
problems involving AlOx are avoided.
Experimentally, the proposed heterostructure was realized using molecular
beam expitaxy (MBE). The specific stack consisted of two monolayers (0.6nm) of
40
GaSb/In0.2Al0.8Sb (3nm for p-‐channel, 10nm for n-‐channel)/In0.2Ga0.8Sb (7nm for p-‐
channel, 10nm for n-‐channel) with ~1um Al0.8Ga0.2Sb buffer layer/GaAs. The
unintentionally doped stack was p-‐type due to the charged native defect states [4].
For n-‐channel samples, tellurium (Te) delta-‐doping was added in both top barrier
and buffer layers during MBE growth [25]. To minimize the exposure with
atmosphere, samples were capped with 50nm arsenic in the MBE system then
transferred to the atomic-‐layer-‐deposition reactor. A clean (001) surface was
prepared by heating up the samples up to 400°C under nitrogen purging, in order to
decap the arsenic completely. This decapping process was carefully monitored by
recording chamber pressure. Subsequently, an Al2O3 film was deposited at 300°C
using trimethylaluminium and H2O as precursors. Smoothness on the atomic level of
the interface between the high-‐k dielectric and the semiconductor heterostructure
can also be achieved with in-‐situ techniques [33,34]. After the deposition of Al2O3,
aluminum was deposited using electron-‐beam evaporation, then patterned to form a
gate electrode. The top channel contact was formed by etching Al2O3. Then the
ohmic contacts were formed by evaporation of Ti/Ni for p-‐channel devices and
Au/Ni/Ge/Au for n-‐channel devices. Forming-‐gas anneal at 350°C was performed
for 30 minutes to further improve the high-‐k film and interface quality. Control
samples undergo the same process except the substrate is bulk GaSb with no
heterostructure.
Capacitance and conductance measurement were performed on both the
heterostructure and the control samples. Both p-‐ and n-‐type capacitors built on the
heterostructure show marginal frequency dispersion (<2.4%/dec for p-‐type,
<2.6%/dec for n-‐type) at 200K as shown in Figure 2.12 (a) and (b) respectively.
Since interface properties, and specifically high Dit level, are one of the major causes
of frequency dispersion in C-‐V measurements, this absence of significant dispersion
indicates the minimal impact of Dit. A few mono-‐layers of interfacial GaSb can be
clearly identified in HR-‐TEM image in Figure 2.12 (c).
41
Figure 2.12 Capacitance measured on the heterostructure with both (a) p-‐ and (b) n-‐ doped channel. Marginal dispersions have been observed at 200K, indicating
minimal impact of interface traps on the capacitance. (c) Ultra-‐thin interfacial layer of GaSb can be clearly identified in HR-‐TEM image.
Figure 2.13 Measured conductance response for the heterostructure compared to bulk GaSb as the control sample. Biasing condition is set to give near flat-‐band
conditions for both cases. Nearly one order-‐of-‐magnitude reduction in conduction peaks is observed for the heterostructure compared to control sample.
42
Figure 2.13 plots the conduction response, which is indicative of the interface
state response [44]. We observe an order of magnitude reduction in the magnitude
of conduction peaks from the sample with the heterostructure in this work
compared to the control sample of bulk GaSb. The gate biasing is set to give near
flat-‐band conditions for both cases. Since the physical interfaces in the two cases are
the same, this demonstrates the suppression of interaction between carriers and the
interface traps due to the effect of band engineering. The result is in qualitative
agreement with the modeling analysis from tight binding, as the interface trap
response is proportional to the number of carriers available at the interface, which
in turn depends exponentially on the effective bandgap.
Figure 2.14 Transfer characteristics of both (a) p-‐ and (b) n-‐channel MOSFETs built on the heterostructure in this work, showing sharp switching behavior.
Subthreshold swings are 33mV/dec and 61mV/dec for n-‐ and p-‐channel devices respectively, close to the thermal limit of ln10 kT/q V/dec.
43
I-‐V measurement on the MOSFETs fabricated on the proposed heterostructure
give sharp switching (low SS) for both p-‐ and n-‐channel devices as shown in Fig.
2.14 (a) and (b) respectively, close to the thermal limit of 30mV/dec at 150K and
24mV/dec at 120K. These measurements on C-‐V, conductance response and
MOSFET characteristics indicate this technique is successful in ameliorating the
response of interface states in III-‐V semiconductors and can be effectively applied in
a MOSFET fabrication process flow. While we have demonstrated this technique on
the ~6.1Å system with an InGaSb channel, it is applicable on an InGaAs system with
InAlAs barrier and InGaAs/InAs cap also and in general possible on any
heterostructure system having the band lineup as shown in Fig 2.10 (c).
Summary
In conclusion, to efficiently passivate III-‐V semiconductors and minimize the
impact of interface traps on carriers in a MOS structure, we proposed a
heterostructure that consists of an Al-‐containing barrier layer and an ultra-‐thin Al-‐
free interfacial layer. We experimentally demonstrated that the heterostructure
enables good physical quality of the high-‐k dielectric/semiconductor interface and
suppresses the impact of interface traps by band engineering.
44
References
[1] B. R. Bennett, R. Magno, J. B. Boos, W. Kruppa, and M. G. Ancona, “Antimonide-‐
based compound semiconductors for electronic devices: A review,” Solid-‐State
Electron., vol. 49, pp. 1875-‐1895, Nov. 2005.
[2] G. Dewey, M. K. Hudait, K. Lee, R. Pillarisetty, W. Rachmady, M. Radosavljevic,
T. Rakshit, and R. Chau, "Carrier Transport in High-‐Mobility III–V Quantum-‐Well
Transistors and Performance Impact for High-‐Speed Low-‐Power Logic
Applications," IEEE Electron Device Lett., vol. 29, pp. 1094-‐1097, Oct. 2008.
[3] A. Ali, H. Madan, A. Agrawal, I. Ramirez, R. Misra, J. B. Boos, B. R. Bennett, J.
Lindemuth, and S. Datta, “Enhancement-‐Mode Antimonide Quantum-‐Well MOSFETs
With High Electron Mobility and Gigahertz Small-‐Signal Switching Performance,”
IEEE Electron Device Lett., vol. 32, pp. 1689-‐1691, Dec. 2011.
[4] A. Ali, H. Madan, M. J. Barth, M. J. Hollander, J. B. Boos, B. R. Bennett, and S.
Datta, “Antimonide NMOSFET with source side injection velocity of 2.7×107 cm/s for
low power high performance logic applications,” in proc. VLSIT 2012, pp. 181-‐182,
Jun. 2012.
[5] Z. Yuan, A. Nainani, A. Kumar, X. Guan, B. R. Bennett, J. B. Boos, M. G. Ancona,
and K. C. Saraswat, in proc. VLSIT 2012, pp. 185-‐186, Jun. 2012.
[6] A. Nainani, Z. Yuan, T. Krishnamohan, B. R. Bennett, J. B. Boos, M. Reason, M.
Ancona, Y. Nishi, and K. C. Saraswat, “Device quality Sb-‐based compound
semiconductor surface: A comparative study of chemical cleaning,” J. Appl. Phys., vol.
110, 014503, Jun. 2011.
45
[7] J. Nah, H. Fang, C. Wang, K. Takei, M. H. Lee, E. Plis, S. Krishna, and A. Javey,
“III–V Complementary Metal–Oxide–Semiconductor Electronics on Silicon
Substrates,” Nano Lett., vol. 12, pp. 3592-‐2595, Jun. 2012.
[8] A. Nainani, T. Iriwaswa, Z. Yuan, B. R. Bennett, J. B. Boos, Y. Nishi, and K. C.
Saraswat, “Optimization of the Al2O3 GaSb Interface and a High-‐Mobility GaSb
pMOSFET,” IEEE Trans. Electron Devices, vol. 58, pp. 3407-‐3415, Oct. 2011.
[9] M. Xu, R. S. Wang, and P. D. Ye, “GaSb Inversion-‐Mode PMOSFETs With
Atomic-‐Layer-‐Deposited Al2O3 as Gate Dielectric,” IEEE Electron Device Lett., vol. 32,
pp. 883-‐885, Jul. 2011.
[10] A. Ali, H. S. Madan, A. P. Kirk, D. A. Zhao, D. A. Mourey, M. K. Hudait, R. M.
Wallace, T. N. Jackson, B. R. Bennett, J. B. Boos, and S. Datta, “Fermi level unpinning
of GaSb (100) using plasma enhanced atomic layer deposition of Al2O3,” Appl. Phys.
Lett., vol. 97, 143502, Oct. 2010.
[11] A. Nainani, Y. Sun, T. Irisawa, Z. Yuan, M. Kobayashi, P. Pianetta, B. R.
Bennett, J. B. Boos, and K. C. Saraswat, “InxGa1-‐xSb channel p-‐metal-‐oxide-‐
semiconductor field effect transistors: Effect of strain and heterostructure design,” J.
Appl. Phys., vol. 109, 114908, Jul. 2011.
[12] L. B. Ruppalt, E. R. Cleveland, J. G. Champlain, S. M. Prokes, J. B. Boos, D. Park,
and B. R. Bennett, “Atomic layer deposition of Al2O3 on GaSb using in situ hydrogen
transistors with buried Al2O3 layers and sulfur passivation: Reduction in carrier
scattering at the bottom interface,” Appl. Phys. Lett., vol. 96, p. 142106, 2010.
[33] Y. C. Chang, C. Merckling, J. Penaud, C. Y. Lu, W.-‐E. Wang, J. Dekoster, M.
Meuris, M. Caymax, M. Heyns, J. Kwo and M. Hong, “Effective reduction of interfacial
traps in Al2O3/GaAs(001) gate stacks using surface engineering and thermal
annealing,” Appl. Phys. Lett., vol. 97, p. 112901, Apr. 2010.
[34] C. Merckling, X. Sun, A. Alian, G. Brammertz, V. V. Afanas’ev, T. Y. Hoffmann,
M. Heyns, M. Caymax and J. Dekoster, “GaSb molecular beam epitaxial growth on p-‐
InP(001) and passivation with in situ deposited Al2O3 gate oxide,” J. Appl. Phys., vol.
109, p. 073719, Apr. 2011.
49
[35] W. Wang, K. Xiong, R. M. Wallace and K. Cho, “Impact of Interfacial Oxygen
Content on Bonding, Stability, Band Offsets, and Interface States of
GaAs:HfO2 Interfaces,” J. Phys. Chem., C, vol. 114, p. 22610, 2010.
[36] J. Robertson, “Model of interface states at III-‐V oxide interfaces,”Appl. Phys.
Lett., vol. 94, p. 152104, Apr. 2009.
[37] M. Radosavljevic, B. C.-‐Kung, S. Corcoran, G. Dewey, M.K. Hudait, J.M.
Fastenau, J. Kavalioros, W. K. Liu, D. Lubyshev, M. Metz, K. Millard, N. Mukherjee, W.
Rachmady, U. Shah, R. Chau, “Non-‐planar, multi-‐gate InGaAs quantum well field
effect transistors with high-‐K gate dielectric and ultra-‐scaled gate-‐to-‐drain/gate-‐to-‐
source separation for low power logic applications,”, in Proc. IEDM 2009, p. 319 Dec.
2009.
[38] A. Nainani, Z. Yuan, T. Krishnamohan and K. C. Saraswat, “Optimal design of
III–V heterostructure MOSFETs,” in Proc. SISPAD 2010, pp. 103-‐106, Sept. 2010.
[39] A. Ali, H. Madan, A. Agrawal, I. Ramirez, R. Misra, J. B. Boos, B. R. Bennett, J.
Lindemuth and S. Datta, “Enhancement-‐Mode Antimonide Quantum-‐Well MOSFETs
With High Electron Mobility and Gigahertz Small-‐Signal Switching Performance,”
IEEE Electron Device Lett., vol. 32, p. 1689, 2011.
[40] J.-‐M. Jancu, R. Scholz, F. Beltram and F. Bassani, “Empirical spds* tight-‐
binding calculation for cubic semiconductors: General method and material
parameters,” Phys. Rev. B, vol. 57, p. 6493, 1998.
[41] I. Vurgaftman and J. R. Meyer, “Band parameters for III–V compound
semiconductors and their alloys,” J. Appl. Phys., vol. 89, p. 5815, 2001.
50
[42] S. Lee, F. Oyafuso, P. von Allmen and G. Klimeck, “Boundary conditions for
the electronic structure of finite-‐extent embedded semiconductor nanostructures,”
Phys. Rev. B, vol. 69, p. 045216, 2004.
[43] X. Guan and Z. Yu, Nanotechnology, “Atomistic Approach to Thickness-‐
Dependent Bandstructure Calculation of InSb UTB,” IEEE Transcations on
Nanotechnology, vol. 6, p. 101, 2007.
[44] K. Martens, C. O. Chui, G. Brammertz, B. De Jaeger, D. Kuzum, M. Meuris, M.
Heyns, T. Krishnamohan, K. Saraswat, H. E. Maes and G. Groeseneken, “On the
Correct Extraction of Interface Trap Density of MOS Devices With High-‐Mobility
Semiconductor Substrates,” Electron Devices, IEEE Transcations on, vol. 55, p. 547
2008.
CHAPTER 3: INGASB: SINGLE-‐CHANNEL SOLUTION FOR III-‐V CMOS
There has been an upsurge of interest in the possibility of a low-‐power, high-‐
performance CMOS based on III-‐V materials. For such a technology to be realized,
comparable high performance from both n-‐ and p-‐channel devices is needed for
complementary logic. The 6.1-‐6.2Å lattice constant system with InGaSb as the single
channel has significant advantages in terms of high mobility, high sheet charge
density, and large band offsets for both electrons and holes. We demonstrate, for the
first time in antimonide systems, n-‐ and p-‐channel MOSFETs using a single channel
material. Electron and hole mobilities greater than >4000 and 900cm2/V-‐sec have
been achieved, respectively. The InGaSb NMOSFETs are the first of their kind, and
when combined with PMOSFETS in the same material, form an attractive approach
for next generation high speed, low power CMOS circuits.
3.1 TIGHT-‐BINDING STUDY OF Γ-‐L BANDSTRUCTURE ENGINEERING FOR
BALLISTIC INGASB NMOSFETS
nMOSFETs based on III-‐V materials have the highest mobility/injection velocity
(νinj), the major concern, however, is the degradation of device performance due to
low density-‐of-‐states (DOS) (low effective mass of carriers) and excessive spillover
of the charge from Γ-‐ to L-‐valley at high sheet charge [1-‐2]. As we increase the In%
in InGaAs / InGaSb channel nMOSFETs: νinj and separation between and the Γ-‐ to L-‐
valley (ΔΓ-‐L) increases while DOS decreases. In this section we study this tradeoff in
InXGa1-‐XSb, which is a promising candidate for replacing silicon as the channel
52
material of future high-‐speed, low-‐power CMOS devices, showing high mobility for
both electrons [3] and holes [4].
Figure 3.1 Structure of InGaSb double-‐gate MOSFET with (100) orientation. Atom arrangement under VCA.
For performance evaluation of ultrathin-‐body (UTB) double-‐gate devices, the use
of bulk effective masses may not be adequate [6]. The E-‐k relations for InXGa1-‐XSb
UTB MOSFET (Fig. 3.1) is calculated using sp3d5s* atomistic tight-‐binding (TB)
model coupled with Poisson’s equation. The effect of varying the In % on DOS,
electron population /separation among Γ-‐, L-‐, and X-‐valleys is studied
systematically. E-‐k band diagram for UTB MOSFET with GaSb and InSb channel is
plotted in Fig. 3.2(a). Amongst different valleys of electrons, Γ-‐valley has lower
effective mass, thus lower DOS and higher νinj. For GaSb (low In% InXGa1-‐XSb) ΔΓ-‐L
may not be sufficient; L-‐valley can also be populated, which has higher DOS but
lower νinj. Increasing In% in the compound brings up L-‐ and X-‐valleys, reduces the
energy for Γ-‐valley, meanwhile reducing the effective mass of Γ-‐valley. However,
high In% makes it difficult to achieve high electron sheet charge density (NS) from Γ-‐
53
valley. From drive current perspective, high νinj and NS have to be achieved
simultaneously, which requires engineering of DOS, νinj and electrons population in
different valleys [3]. This trade-‐off between DOS and νinj is summarized in Fig.
3.2(b).
Figure 3.2 (a) Band diagrams of GaSb, In0.5Ga0.5Sb and InSb; (b) summary of the bandstructure effects by varying In% in InGaSb and their relationships.
TB parameters for ternary InXGa1-‐XSb are calculated following virtual-‐crystal
approximation (VCA) incorporating compositional disorder effect and fitted to bulk
band gap of ternary compound [7-‐9]. 1D Poisson’s equation perpendicular to
channel direction is coupled with TB Hamiltonian by Hartree-‐Fock potential in the
gate stack. Dangling bonds at interface are passivated by hydrogen termination of
hybridized orbitals to eliminate all the states within band gap [10]. A ballistic
54
transport model is adopted to assess transport of electrons [11]. νinj is determined
from full band structure with non-‐parabolic E-‐k relationship considered for all
valleys.
Figure 3.3 Calculated bandstructure from TB for InGaSb at sheet charge density ~3×1012cm-‐2 with a body thickness of 4nm.
Band structures of InXGa1-‐XSb at NS of ~3×1012cm-‐2 for different In %’s are
compared in Fig. 3.3 (TBODY=4nm). Each series of curves in the E-‐k plot represents
each sub-‐band, black and blue for conduction bands and valence bands respectively.
To achieve high νinj at relevant NS, it is important to confine sufficient amount of
electrons in high-‐ νinj Γ-‐valley. The overall population among different valleys is
determined by the energy separations as well as difference in DOS among Γ-‐, L-‐, and
55
X-‐ valleys. For low In % InXGa1-‐XSb, because of quantum confinement (QC) effects,
the energy difference between Γ-‐ and L-‐valleys is marginal, especially under high VG
and thin body thickness, resulting in Fermi level moving into L-‐valley. Band gap (Eg)
and ΔΓ-‐L are shown in Fig. 3.4: with higher In%, ΔΓ-‐L is increased to confine more
electrons in Γ-‐valley counteracting the effects of quantization (Fig. 3.4(a)), which is
more dominant for thin TBODY (Fig. 3.4(b)). At higher NS (/VG), though Eg is lowered
due to Quantum Confinement Stark Effect [12], ΔΓ-‐L stays low (Fig. 3.4(c)). This can
be attributed to Γ-‐valley’s curvature getting blunt (effective mass becomes larger) at
high NS, thus alleviated quantization effect.
Figure 3.4 Calculated bandgap and Γ-‐L energy separation vs. (a) In composition for 4nm body thickness and ~1011 and 1012 cm-‐2 sheet charge density; (b) body
thickness at sheet charge density of 1012 cm-‐2; (c) sheet charge density for 4nm body thickness.
Fig. 3.5 plots the 2-‐D DOS, Γ-‐valley has much lower DOS compared with L-‐valley
which is required to achieve high NS. In-‐rich compounds lead to decrease of DOS at
56
conduction band edge meaning further movement of Fermi level is necessary to
achieve same NS.
Figure 3.5 2D density of states for GaSb and InSb for body thickness of 4nm.
% occupation of electrons in Γ-‐, L-‐valleys in InXGa1-‐XSb is plotted against In %,
TBODY and NS in Fig. 3.6. From ΔΓ-‐L, adequate percentage of charge can be confined in
Γ-‐valley even at high NS and thin TBODY. Fig. 3.7 illustrates (a) sheet charge density as
a function of gate voltage and (b) sub-‐threshold swing comparison. The kink in
charge density indicates the population of electrons into L-‐valley. For higher In
composition, loss in DOS at the band edge costs more gate voltage to obtain sheet
charge density of a reasonable level for device operation. The change in DOS as well
as dielectric constant (from 14.4 GaSb to 16.8 InSb) reflects on the degradation of
sub-‐threshold swing with increasing In %. For device geometry of interest, DOS
determined quantum capacitance has an evident impact on the subthreshold
behavior as subthreshold swing decreases with larger TBODY.
57
Figure 3.6 Fraction f electron occupation in Γ-‐valley in InXGa1-‐XSb vs. (a) In composition for 4nm body thickness and ~1011 and 1012 cm-‐2 sheet charge density; (b) body thickness for 1012 cm-‐2 sheet charge density; (c) sheet charge density for
4nm body thickness.
νinj: ID-‐VG is evaluated by integrating NS with average velocity of electrons along
transport direction at each k point. <100> is set as the transport direction. Under
the ballistic transport model, when Fermi level is below conduction band edge, the
injection velocity stays constant; it increases as Fermi level moves into conduction
band. The average velocity at given gate bias can be calculated by taking the ratio
between the overall current density and sheet charge density [11]. As shown in Fig.
3.8, since electron population is mostly in L-‐valley, the overall injection velocity of
GaSb is low. Ga-‐rich InXGa1-‐XSb has low νinj as most of the electrons are in L-‐valley at
high NS. In-‐rich compound gives high νinj, because of sharper curvature of Γ-‐valley in
calculated E-‐k relation. The high injection velocity is maintained even at high NS as
electrons are properly confined in Γ-‐valley.
58
Figure 3.7 (a) sheet charge density as a function of gate voltage. Gate voltage is adjusted to give 5×107cm-‐2 sheet charge at 0V; (b) subthreshold swing with varing
In composition for body thickness of 4nm and 5nm.
For the optimal overall driving capability, InXGa1-‐XSb can gain from injection
velocity and DOS of L-‐valley simultaneously. In Fig. 3.7 (a), the filling of L-‐valley
gives increase in NS. Without excessive filling of L-‐valley, average injection velocity
maintains high as shown in Fig. 3.8. With these factors combined, Fig. 3.9 shows
proper amount of In percentage (~25%) in InXGa1-‐XSb can give a overall
improvement in drive-‐current. Further increase in In composition leads to
significant loss in NS. The optimal current density is 80% higher than silicon, 50%
higher than GaSb and 160% higher than InSb. Impact of quantum capacitance on
subthreshold swing for In-‐rich compound is avoided for In percentage of ~25% as
well.
59
Figure 3.8 Injection velocity as a function of sheet charge density for different In composition, the drop in injection velocity at high sheet charge density is due to L-‐
valley population.
Figure 3.9 Saturation current as a function of gate voltage; Gate voltage is adjusted to give 15uA/um current density at 0V.
60
In conclusion, by using an sp3d5s* TB model, we explored the energy dispersion
relations of InXGa1-‐XSb UTB device. Ballistic currents for nMOSFETs are calculated
based on full band analysis. It is shown that band structure engineering of InXGa1-‐
XSb allows sufficient electron population in Γ-‐valley to maintain high injection
velocity, meanwhile avoids the degradation of sheet charge density due to quantum
confinement and loss in DOS. Optimal design of InXGa1-‐XSb channel can deliver
~80% improvements in drive current under ballistic limit for low supply voltage
application compared with silicon.
3.2 HETEROSTRUCTURE DESIGN FOR SINGLE-‐CHANNEL III-‐V CMOS
As Moore’s law is approaching the sub-‐20nm regime, III-‐V semiconductors are
considered as promising candidates for potentially replacing silicon as the channel
material in future technology nodes due to their high-‐performance and low-‐power
advantages [12]. Most of the research in the community has been focused on the
excellent electron transport in various III-‐V material systems demonstrated over the
years [13,14]. However, the hole mobility achieved in III-‐V quantum wells has
traditionally lagged behind in comparison to silicon. The requirements on circuit
design for complementary logic suggests that a future III-‐V CMOS technology should
strive for a PMOS that can deliver no worse than one-‐third of the current density of
the NMOS transistor [15]. It has been proposed that the relatively high mobility for
holes in germanium could make it a potential PMOS device to complement the III-‐V
NMOS [16,17]. However, this can result in complications in CMOS circuit design and
manufacturing, such as cross-‐contamination problems associated with co-‐
integration of group IV and III-‐V materials on a common Si platform and different
61
gate stack solutions for Ge and III-‐V. Thus it would be worthwhile to explore if a
single material system can enable both good NMOS and PMOS transistors.
To realize CMOS in a single material system, progress has been made on Ge
NMOS and InGaAs PMOS in hopes of getting both high-‐performance CMOS
transistors in either Ge [18] or InGaAs [19] systems. Meanwhile, antimony (Sb)-‐
based compound semiconductors have the highest electron and hole mobilities
among all III-‐V materials. The electron saturation velocity in InSb is the highest
among all the conventional semiconductor materials. Schottky-‐gate FETs with an
InGaSb n-‐channel have achieved fT of 305GHz at 85nm gate length [20] and fT of
140GHz at 40nm gate length for p-‐channel [21]. Heterostructures with an InGaSb
channel exhibiting a room temperature electron mobility of 25,000cm2/Vs
(21,000cm2/Vs) with a sheet carrier density of 1.5×1012/cm2 (2.8×1012/cm2) have
been achieved [22,23]. Room-‐temperature hole mobility as high as 1,500cm2/Vs in a
strained InxGa1-‐xSb channel at a sheet charge density near 1012/cm2 has also been
demonstrated [24].
In addition to good mobilities for the electrons and holes, InxGa1-‐xSb has high
conduction and valence band offsets with AlxIn1-‐xSb and AlxGa1-‐xSb for flexible
heterostructure design [25]. Despite the larger lattice constants, antimonides are
also found to be potentially more suitable for heterointegration [26]. Thus, InxGa1-‐
xSb has the potential to enable a complementary technology with both high
performance NMOS and PMOS in a single channel material, outperforming silicon
[25-‐27].
In this section, we study the type-‐I heterostructure design of the AlxGa1-‐
xSb/InxGa1-‐xSb system. Carrier transport in the heterostructure is investigated by
gated Hall measurements. High-‐mobility InGaSb n-‐ and p-‐channel MOSFETs using
the same channel material are reported for the first time.
62
Figure 3.10 Schematic of the heterostructure studied in this work.
Fig. 3.10 shows the schematic of the heterostructure design studied in this work.
The stack grown by molecular beam epitaxy on semi-‐insulating GaAs substrate [26]
consists of a 1um thick AlGaSb buffer layer, a 10nm InGaSb channel, which is capped
with an InAlSb top barrier. For p-‐channel devices, the channel was unintentionally
doped p-‐type due to the charged native defect states. For n-‐channel devices, both
the top and bottom barrier layers of AlInSb and AlGaSb were delta-‐doped with Te to
provide carriers to the acess region. InGaSb of low InSb mole fraction was chosen
because of its potential to properly confine charge in Γ-‐valley and achieve optimal
trade-‐off between the density-‐of-‐state and the injection velocity [27].
Cross-‐sectional TEM image of the heterostructure stack is shown in Fig. 3.11.
0.04um AlSb nucleation layer is grown first, which is proven to give better surface
morphology. Misfit dislocations are generated at the interface between antimonide
layers and GaAs substrate due to the lattice mismatch between the substrate and
buffer. It can be observed that most of the dislocations/defects are contained in the
buffer layer, thus it is possible to obtain good crystal quality near the channel layers.
63
Figure 3.11 Cross-‐sectional TEM image of the heterostructure stack grown by
Molecular Beam Epitaxy (MBE). The stack is grown on GaAs substrate. The lattice-‐
mismatch is released by 1μm thick AlGaSb buffer layer.
The type-‐I band alignment formed by AlGaSb/InGaSb heterostructure is shown
in Fig. 3.12. The valence-‐band offset (VBO) of 0.30eV can be determined by
comparing the valence band (VB) spectrum of InGaSb and AlGaSb using the Ga peak
as a reference as in Fig. 3.13(a). Bandgaps of InGaSb (Fig. 3.13(b)), AlGaSb (Fig.
3.13(c)) and GaAs (Fig. 3.13(d)) are extracted from photoluminescence
measurement at 80K to be 1.32eV, 0.70eV and 1.5eV respectively.
The conduction-‐band offset (CBO) between AlGaSb and InGaSb can thus be
calculated to be 0.32eV. The information of band-‐energies is summarized in Fig.
3.11. Therefore, such a heterostructure can provide both decent CBO and VBO, and
thus has the potential to properly confine both electrons and holes.
64
Figure 3.12 Band energies of the type-‐I heterostructure formed by InGaSb/AlGaSb, which offers sufficient CBO/VBO to confine both electrons and holes.
Figure 3.13 (a) VBO is measured using UV-‐XPS by taking the difference between VB spectrum from the channel and the buffer layer. Bandgaps of (b) AlGaSb, (c) InGaSb
and (d) GaAs are measured using photo-‐luminescence (PL) at 80K.
65
The confinement of carriers is studied using a tight-‐binding simulation [7-‐9]. Fig.
3.14 (a) shows the accumulation of both electrons and holes in a double-‐gate
structure with 2nm AlGaSb barrier layer, 8nm-‐thick InGaSb channel and 0.5nm
equivalent-‐oxide-‐thickness (EOT). High sheet charge density and reasonable
switching of devices can be obtained in the heterostructure even with such highly
scaled electrostatics. The percentage of carriers that is confined in the high-‐mobility
channel is plotted in Fig. 3.14 (b). At sheet charge densities (NS) relevant for
MOSFET operation, 80/90% of the electron/hole can be properly confined.
Figure 3.14 Results from tight-‐binding calculations show (a) heterostructure design can provide high NS at given gate voltage. Nearly eight orders-‐of-‐magnitude
switching of sheet charge densities can be obtained within 1V for both carriers; (b) most of the charge (above 80% and 90% of electrons and holes respectively) is confined in the high-‐mobility channel even at high sheet charge density up to
1013cm-‐2.
Device-‐quality surface passivation on the stack can be achieved by covering the
surface with ultra-‐thin layer of GaSb, on which a high-‐k dielectric of Al2O3 was
deposited by atomic-‐layer-‐deposition (ALD) [28]. To study the transport of carriers,
structure for gated Hall measurement was fabricated on the samples. In Fig. 3.15,
the heterostructure show 7X / 2.5X higher electron mobility (μe) and 7X / 4X higher
66
hole mobility (μh) at an NS of 1012cm-‐2\6x1012cm-‐2 for electrons and holes
respectively, in comparison to silicon.
Figure 3.15 a) Electron and (b) hole mobility measured in InGaSb channel as a function of sheet charge density (NS) using gated Hall measurement. Electron and hole mobility of 4000cm2/Vs and 900cm2/Vs can be obtained at sheet density of 1012cm-‐2 respectively, and are 2.5X and 4X as compared with Si universal even at
high NS.
The electron transport can potentially be further enhanced by increasing the In
composition in the channel material, since InSb has the highest electron mobility
amongst all the conventional semiconductors. However, this approach has its
drawbacks. High In% InGaSb leads to low DOS of electrons, thus eventually result in
67
excessive loss in sheet charge density as described in the previous section. Another
factor is the strain effect. Based on current buffer design, which consists of 1um
thick AlGaSb, the channel layer of InGaSb grown on top is compressively strained.
Figure 3.16 X-‐ray diffraction of 004 rocking curve of the heterostructure stack that consists of 1um thick AlGaSb buffer layer, a 10nm InGaSb channel grown on top of
GaAs SI substrate.
Fig. 3.16 shows the XRD [004] rocking curve of the heterostructure stack. Buffer
layer of AlGaSb is fully relaxed on GaAs substrate. The channel layer grown on top
has a larger lattice constant with respect to the buffer layer, thus is compressively
strained. Using In0.2Ga0.8Sb as the channel material, the strain can be estimated to be
0.7%. Compressive strain is expected to split the heavy and light hole bands near the
Γ-‐point. For holes, such splitting of bands results in more holes population in the
light hole band of lower transport mass. Therefore, compressive strain is beneficial
to hole transport. On the other hand, compressive strain raises up the band edge of
68
Γ-‐valley, meanwhile decrease its curvature. Thus the transport mass of Γ-‐valley
increases, and mobility could be lowered.
3.3 SURFACE-‐CHANNEL INGASB CMOS TRANSISTORS
Due to the poor recovery of crystalline quality after implantation associated with
GaSb, it is favorable to adopt a device architecture that allows an implantation-‐free
process for MOSFETs fabrication.
Figure 3.17 Raman spectrum of CZ-‐grown bulk GaSb and GaSb after Be implantation
and consequent annealing.
Fig. 3.17 shows the Raman spectrum of CZ-‐grown bulk GaSb and GaSb after Be
implantation (ion energy of 10keV, dose of 9x1014cm-‐2) and consequent annealing at
350°C for 30mins. The main peaks correspond to LO-‐phonon mode. After the
69
implantation, the peak is shifted to the left and broadened, which is an indication of
poorer crystalline quality due to the incomplete recovery of crystal structure even
after annealing. Therefore, it is desirable to develop an implant-‐free device structure
and process flow for transistors.
S/D designs with low extrinsic resistance are challenging for III-‐Vs, because of
low solubility and poor activation of dopants. Use of metal S/D overcomes these
problems, a slight overlap is introduced between the gate and S/D electrodes to
reduce the access resistance as shown in Fig. 3.18. There is a 2μm overlay region
between the S/D and the gate electrodes with Al2O3 inserted in between.
Figure 3.18 Metal S/D structure, in which a slight overlap of gate and S/D electrodes
is introduced to lower the access resistance. Top surface layers are n-‐type or p-‐type
doped for NMOS or PMOS respectively.
Both PMOS and NMOS are fabricated on the same heterostructure stack except
with the top layer being intrinsically p or n-‐doped respectively. After MBE growth,
the samples are capped with elemental arsenic to protect the surface from oxidation
during the transfer. After solvent clean to degrease the samples, they are transfer
70
into atomic layer deposition machine. The base pressure of the main chamber is
approximately 100mTorr. The samples were subject to heating up to 400°C for
10mins. This guarantees that all the arsenic is fully decapped inside the ALD
chamber and fresh surface of antimonide, free from oxidation is prepared for ALD of
Al2O3. After the growth of gate dielectrics, gate electrodes were deposited and
patterned using lift-‐off process. Contacts to the channel were made after removal of
Al2O3, followed by lift-‐off of the contact metals of Ti/Ni. Well-‐behaved CGC-‐V
characteristics are obtained on both p-‐ and n-‐type samples.
Figure 3.19 Well-‐behaved CGC vs VG characteristics are obtained on both a) p and b)
n-‐type InGaSb surface channel devices.
71
For devices, initially 2nm Al2O3 was deposited after the decapping of arsenic.
Source/drain electrodes were deposited and patterned using lift-‐off of 5nm Pt. Al2O3
is then removed by 2% HF dip, then immediately transferred into ALD chamber.
10nm / 25nm Al2O3 were deposited on p-‐ / n-‐channel devices respectively. Gate
electrode is then deposited and an overlap of 2um is intentionally introduced
between the gate and S/D electrodes. Finally, Al2O3 was etched at the S/D pads for
probing.
Transfer and output characteristics of the transistors are shown in Fig. 3.20 and
Fig. 3.21, respectively. On/off ratios of 104 and 100 are achieved for PMOS and
NMOS respectively. SS values are ~160mV/dec and ~500mV/dec correspondingly.
The switching characteristics of the devices is limited by the leakage through the
buffer layers. On-‐current of n-‐channel device is largely limited by the contact
resistance due to the factor that Fermi-‐level is severely pinned near valence band
edge of InGaSb.
Figure 3.20 Transfer characteristics for a) PMOS and b) NMOS with metal S/D.
NMOS ID-‐VG scales linearly with drain voltage due to the high contact resistance.
72
Figure 3.21 Output characteristics for a) PMOS and b) NMOS with metal S/D. NMOS
drain currents are largely limited by the high contact resistance.
Due to the severe Fermi level pinning and high contact resistance for n-‐type
contacts, the overall resistance of n-‐channel devices is dominated by contact
resistance. To improve the n-‐type contact, a TiO2 interfacial layer was inserted to
lower the barrier height for electron injection. In contrast to other MIS schemes for
Fermi-‐level unpinning, TiO2 lowers the barrier height without introducing tunneling
resistance due to the low CBO of TiO2/GaSb. More than 5X improvement in ION is
obtained for NMOS with introduction of TiO2, as shown in Fig. 3.22. The NMOS on-‐
state current remains limited by contact resistance due to the thinness of the
surface channel and consequent depletion of electrons in the contact regions. This
leaves room for further improvement in the device performance of NMOS with MIS
S/D contacts.
73
Figure 3.22 Structure of n-‐channel devices with TiO2 inserted in between the
semiconductor and metal at S/D. 5X increase in on-‐state current for NMOS is
achieved with the insertion of TiO2.
To investigate metal contact on antimonides, we built Schottky diodes on
moderately doped n-‐ and p-‐type GaSb substrates of carrier concentration ~1017cm-‐
3. Native oxides were removed by HCl clean. Metals were deposited by e-‐beam
evaporation and patterned to make top electrode. Backside contact was formed
using blanket evaporation of metal to measure diode I-‐V. Ideally, by Schottky-‐Mott
relation, low workfunction metals such as Al and Ti would give higher Schottky
barrier for holes than for electrons. Strong Fermi-‐level pinning is observed at
metal/GaSb interface. Fig. 3.23 (a), (b) show the J-‐V characteristics of Al/n-‐, p-‐GaSb,
and Ti/n-‐, p-‐GaSb respectively. Both Al and Ti show rectifying behaviors on n-‐GaSb
and ohmic contact on p-‐GaSb. Lower (higher) Schottky barrier height for metal/p-‐
74
(n-‐) GaSb is confirmed with temperature dependence measurement of Al/GaSb
Schottky diode as shown in Fig. 3.24. Due to the intrinsic gap state distribution, the
charge-‐neutral level (ECNL) is found to be near valence band edge as shown in Fig.
3.25. These confirm that antimonide (Sb binary and Sb rich ternary) compound
exhibits Fermi-‐level pinning towards valence band.
Figure 3.23 I-‐V characteristics of (a) Al/GaSb (b) Ti/GaSb contacts on both n-‐type
and p-‐type substrate.
Figure 3.24 Temperature dependence of I-‐V characteristics of Al/GaSb Schottky
diodes.
75
Figure 3.25 Band diagrams of metal contacts to n-‐type and p-‐type GaSb.
Leakage current of the devices is limited by the leakage through buffer layer
beneath the channel. Point defects in antimonide can lead to shallow acceptor-‐type
energy levels within the bandgap, and thus can be p-‐type activated. Because of this,
undoped GaSb is p-‐type doped. For the same reason, point defects in the buffer layer
can contribute to p-‐type conduction. Leakage current could be further reduced if the
defect states in the buffer layer were deactivated. Fig. 3.26 shows the temperature
dependence of the transfer characteristics of the p-‐channel devices. At low
temperatures, the leakage current is greatly reduced. An on/off ratio of 107 is
obtained at 77K. Therefore, in heterostructure device, it is critical to control the
point defects in the buffer layer in order to achieve low leakage.
Figure 3.26 Temperature dependence of the transfer characteristics of p-‐channel
devices shows the significant drop in leakage current at low temperature due to the
deactivation of p-‐type defect states.
76
3.4 BURIED-‐CHANNEL INGASB CMOS TRANSISTORS
Because of the Fermi-‐level pinning near the valence band edge and thin channel
thickness of the surface channel devices, population of electrons in the channel is
difficult to achieve. Fig. 3.27 shows the simulation results of electron profile for
different doping levels (1017cm-‐3, 1018cm-‐3, 1019cm-‐3) in the surface channel.
Figure 3.27 Band diagrams and carrier concentration profile (blue: electrons, red:
holes) of the surface channel device with 10nm InGaSb as the channel and AlGaSb as
the buffer layer. The surface is terminated with 1nm GaSb and the Fermi-‐level is
pinned at 0.1eV from the valence band edge of GaSb. Doping levels are changed from
1017cm-‐3, 1018cm-‐3 to 1019cm-‐3, indicated by the directions of the arrows.
Since Fermi-‐level is pinned at the valence band at the surface, there will be a
depletion region for electrons. By a simple analysis, if the depletion width exceeds
that of the channel thickness, all the electrons are depleted as a result. To get around
this, one solution is to add another top buffer layer above the channel, so that it
0 5 10 15 200.0
0.5
1.0
E - EF (
eV)
z (nm)
10-10
10-9
10-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
1
101 n p
n an
d p
(1018
cm
-3)
increasing ND
77
confines the depletion region in the top buffer layer. In other words, the band
bending in the top layer will allow the relative position of the Fermi-‐level with
respect to the channel material to go further into the conduction band, thus helping
the population of electrons.
Fig. 3.28 shows the simulation results for the modified buried-‐channel device,
where 10nm In0.2Al0.8Sb top buffer layer is added to shield away the surface pinning
effect. 2x1012cm-‐2 modulation doping was added in both the top and bottom barrier
layers. Activation energy of 0.1eV is set to the dopants to simulate the dopant
activation of Te in antimonide.
Figure 3.28 Band diagrams (blue: conduction band, light green: heavy hole band,
dark green: light hole band) and electron concentration (purple: L-‐valley electrons,
red: Γ-‐valley electrons) of the buried channel device with 10nm InAlSb top barrier
layer, 10nm InGaSb as the channel and AlGaSb as the buffer layer. The surface is
terminated with 1nm GaSb and the Fermi-‐level is pinned at 0.1eV from the valence
band edge of GaSb. 2x1012cm-‐2 modulation-‐doping layers are included in both top
and bottom barrier layers.
78
The problem with the buried-‐channel approach is now top buffer layer presents
a barrier for carriers to tunnel through in order to reach to the channel, especially
for electrons. Therefore, alloy contacts have to be used in the source/drain, where
annealing causes the reaction of contact metal with antimonide to form metal alloy.
As a result, the whole top barrier layer is consumed by the reaction and alloyed
contact reach the channel. The schematic of the structure is shown in Fig. 3.29.
Figure 3.29 Schematic of the structure of the buried channel devices. 3nm and 10nm
AlInSb top barrier layer were used for PMOS and NMOS respectively.
To optimize the electrical contact with alloyed metal scheme, we used n-‐type
bulk GaSb of electron concentration of ~1017cm-‐3. The I-‐V characteristics are shown
in Fig. 3.30. Au/Ge/Ni/Au scheme is chosen for n-‐type contact. The theory behind
the alloy contact is complicated, especially with multiple metals involved, since it
leads to complex phases of the metal alloys. Au and Ni form alloy with antimonide
upon annealing. Ge diffuses into the semiconductor and typically serves as donors.
It is found the current density of Au/Ni/Ge/Au contact with n-‐GaSb consistently
increases with increase in annealing temperature and duration, while the surface
morphology of the contact degrades accompanied with lateral diffusion of the metal
alloy. Pd-‐based alloy contact, on the other hand, shows lower current than Au-‐based
79
contact, possibly because the absence of dopant species, even though Pd forms alloy
with antimonide at a lower temperature.
Figure 3.30 I-‐V characteristics of Au-‐ and Pd-‐based alloy contact with n-‐GaSb with
different annealing conditions. Current density increases consistently with higher
annealing temperature and duration.
Auger Electron Spectroscopy (AES) is used to study the depth profile of the alloy
contact to make sure the metal alloy diffuses through the top barrier layer as shown
in Fig. 3.31. 40nm Au/ 12nm Ge/ 12nm Ni/ 700nm Au metal contact is deposited on
the heterostructure stack for buried-‐channel devices and then annealed at 300°C for
5mins. It is found that there is a clear sign of the intermixing of antimonide with the
metal. Antimony atoms have a tendency to diffuse toward the surface. Even though
higher annealing temperature leads to further increase in current density, it also
leads to excessive lateral diffusion of the contact. Therefore, final condition for
annealing of Au-‐based alloy contact is set at 300°C for 5mins. Alloy scheme for p-‐
type contact is similar to n-‐type, except Pd-‐based alloy contact is used.
80
Figure 3.31 AES depth profile of Au/Ni/Ge/Au alloy contact on the heterostructure
for buried-‐channel devices.
Finally, both p-‐ and n-‐channel MOSFETs were fabricated with the same
heterostructure design except for the channel p-‐ or n-‐doped respectively. After the
ALD deposition of Al2O3 as the high-‐k dielectric, aluminum was deposited using e-‐
beam evaporation and patterned to form gate electrode. The interface quality was
further improved by forming-‐gas anneal at 350°C for 30mins. Pd/Pt/Au and
Au/Ni/Ge/Au metal stacks were deposited and then alloyed to form an embedded
direct contact to the p-‐ and n-‐channel quantum well (QW) respectively. Alloyed
contact to p-‐type channel was established during the forming-‐gas anneal at 350°C.
The annealing condition for n-‐type contact was optimized to be at 300°C in a
nitrogen ambient. Transfer characteristics of the fabricated MOSFETs are shown in
Fig. 3.32. Both PMOS and NMOS demonstrate decent switching behavior; on/off
ratio up to 104 and 3x102 was observed for PMOS and NMOS respectively. Output
characteristics (Fig. 3.33) show on-‐current of 5μA/μm, 3.8μA/μm at LG=50μm for
PMOS and NMOS respectively. The NMOS performance is limited by contact and
0 2 4 6 8 10 12 14
Inte
nsity
(a. u
.)
Sputtering time (mins)
Ni Ge Au Sb
81
series resistance, which leaves room for further improvement, and would be
essential if LG were to be scaled down.
Figure 3.32 Transfer characteristics of (a) PMOS and (b) NMOS. On/off ratio up to
104 and 3x102 can be obtained for PMOS and NMOS respectively. NMOS current
scales non-‐linearly with VDS due to high contact resistance.
Figure 3.33 Output characteristics of (a) PMOS and (b) NMOS. Characteristics can be
further improved by reduction of contact and series resistance.
82
This is, to our best knowledge, the first demonstration of a high mobility NMOS
and PMOS in the same channel material with comparable ION, (Table.1) which makes
InGaSb an attractive prospect for realizing complementary logic in III-‐Vs.
Table 1: Comparison of on-‐state current and mobility for different candidate of
CMOS technology
Ion (μA/μm) N/P μ at 1012cm-‐2 (cm2/V/s)
p n p n
This work
LG=50μm
~4
InGaSb
~3.8
InGaSb 0.9
Gated Hall
900 4000
Ref. [16]
LG=50μm
~2.4
Ge
~8
InGaAs 3.3 260 1800
Ref. [30]
LG=1.5μm
~110
Ge
~450
InGaAs 4.1 400 1300
Summary
In Summary, in this section, we discussed the heterostructure design of 6.1-‐6.2Å
lattice system to support InGaSb as the single-‐channel material for III-‐V CMOS. It is
shown that AlGaSb/InGaSb provides favorable band alignment that give good
confinement of both electrons and holes. Excellent transport properties of InGaSb
83
were measured by gated Hall experiments and showed moblities up to 4000cm2/Vs
and 900cm2/Vs for electrons and holes respectively. NMOS and PMOS built on the
heterostructure delivered on-‐current of 3.8μA/μm and 5μA/μm at LG=50μm, which
could be further improved by parasitic engineering. Therefore, InGaSb, as an
attractive channel material, has the potential to support both n-‐ and p-‐channel
devices with a close gap in performance for realizing complementary logic in III-‐Vs.
84
References
[1] S. Takagi, T. Irisawa, T. Tezuka, T. Numata, S. Nakaharai, N. Hirashita, Y.
Moriyama, K. Usuda, E. Toyoda, S. Dissanayake, M. Shichijo, R. Nakane, S. Sugahara,
M. Takenaka, N. Sugiyama, “Carrier-‐transportenhanced channel CMOS for improved
power consumption and performance,” IEEE Trans. Elect. Dev., vol. 55, p. 21, Jan.
2008.
[2] A. Rahman, G. Klimeck, M. S. Lundstrom, “Novel channel materials for ballistic
nanoscale MOSFETs-‐bandstructure effects,” in Proc. IEDM 2005, p. 604, Dec. 2005.
[3] M. Rodwell, W. Frensley, S. Steiger, E. Chagarov, S. Lee, H. Ryu, Y. Tan, G.
Hegde, L. Wang, J. Law, T. Boykin, G. Klimeck, P. Asbeck, A. Kummel, J. N. Schulman,
“III-‐V FET channel designs for high current densities and thin inversion layers,” in
Proc. DRC 2010, p. 149, Jun. 2010.
[4] B. R. Bennett, R. Magno, J. B. Boos, W. Kruppa, M. G. Ancona, “Antimonide-‐
based compound semiconductors for electronic devices: A review,” Solid-‐State
Electronics, vol. 49, p. 1875, Dec. 2005.
[5] A. Nainani, T. Irisawa, Z. Yuan, Y. Sun, T. Krishnamohan, M. Reason, B. R.
Bennett, J. B. Boos, M. G. Ancona, Y. Nishi, K. C. Saraswat, “Development of high-‐k
dielectric for antimonides and a sub 350°C IIIV pMOSFET outperformaning
Germanium,” in Proc. IEDM 2010, p. 138, Dec. 2010.
[6] L. Yang, N. Neophytou, T. Low, G. Klimeck, M. S. Lundstrom, “A tight-‐binding
study of the ballistic injection velocity for ultrathin-‐body SOI MOSFETs,” IEEE Trans.
on Elect. Dev., vol. 55, p. 866, Mar. 2008.
85
[7] J.-‐M. Jancu, R. Scholz, F. Beltram, F. Bassani, “Empirical spds* tightbinding
calculation for cubic semiconductors: General method and material parameters,”
Physical Review B, vol. 57, p. 6493, Mar. 1998.
[8] S. J. Lee, H. S. Chung, K. Nahm, C. K. Kim, “Band structure of ternarycompound
semiconductors using a modified tight-‐binding method,” Physical Review B, vol. 42,
p. 1452, Jul. 1990.
[9] S. Lee, F. Oyafuso, P. von Allmen, G. Klimeck, “Boundary conditions for the
electronic structure of finite-‐extent embedded semiconductor nanostructures,”
Physical Review B, vol. 69, 045216, Jan. 2004.
[10] A. Rahman, G. Jing, S. Datta, M. S. Lundstrom, “Theory of ballistic
nanotransistors,” IEEE Trans. on Elect. Dev., vol. 50, p. 1853, Sept. 2003.
[11] X. Guan, Y. Tan, J. Lu, Y. Wang, Z. Yu, “A self-‐consistent simulation of InSb
double-‐gate MOSFETs using full-‐band tight-‐binding approach,” in Proc. SISPAD 2007,
p. 161, Sept. 2007.
[12] S. Takagi, T. Irisawa, T. Tezuka, T. Numata, S. Nakaharai, N. Hirashita, Y.
Moriyama, K. Usuda, E. Toyoda, S. Dissanayake, M. Shichijo, R. Nakane, S. Sugahara,
M. Takenaka, and N. Sugiyama, "Carrier-‐transport-‐enhanced channel CMOS for
improved power consumption and performance," IEEE Trans. Electron. Devices, vol.
55, no. 1, p. 21, Jan. 2008.
[13] M. Radosavljevic, G. Dewey, D. Basu, J. Boardman, B. Chu-‐Kung, J. M.
Fastenau, S. Kabehie, J. Kavalieros, V. Le, W. K. Liu, D. Lubyshev, M. Metz, K. Millard,
N. Mukherjee, L. Pan, R. Pillarisetty, W. Rachmady, U. Shah, H. W. Then and R. Chau,
"Electrostatics improvement in 3-‐D tri-‐gate over ultra-‐thin body planar InGaAs
quantum well field effect transistors with high-‐K gate dielectric and scaled gate-‐to-‐
drain/gate-‐to-‐source separation," in Proc. IEDM 2011, pp. 765-‐768, Dec. 2011
86
[14] F. Xue, H. Zhao, Y.-‐T. Chen, Y. Wang, F. Zhou and J. C. Lee, "High-‐k InGaAs
metal-‐oxide-‐semiconductor field-‐effect-‐transistors with various barrier layer
materials," Applied Physics Letters, vol.99, no.3, p.033507, July 2011.
[15] J. A. del Alamo, "Nanometre-‐scale electronics with III-‐V compound
semiconductors", Nature, vol. 479, pp. 317-‐323, 2011.
[16] M. Yokoyama, S. H. Kim, R. Zhang, N. Taoka, Y. Urabe, T. Maeda, H. Takagi, T.
Yasuda, H. Yamada, O. Ichikawa, N. Fukuhara, M. Hata, M. Sugiyama, Y. Nakano, M.
Takenaka and S. Takagi, "CMOS integration of InGaAs nMOSFETs and Ge pMOSFETs
with self-‐align Ni-‐based metal S/D using direct wafer bonding," in Proc. VLSIT 2011,
pp.60-‐61, June 2011.
[17] R. Pillarisetty, B. Chu-‐Kung, S. Corcoran, G. Dewey, J. Kavalieros, H. Kennel,
R. Kotlyar, V. Le, D. Lionberger, M. Metz, N. Mukherjee, J. Nah, W. Rachmady, M.
Radosavljevic, U. Shah, S. Taft, H. Then, N. Zelick and R. Chau, "High mobility strained
germanium quantum well field effect transistor as the p-‐channel device option for
low power (Vcc = 0.5 V) III–V CMOS architecture," in Proc. IEDM 2010, pp. 150-‐153,
Dec. 2010.
[18] A. Toriumi, C. H. Lee, S. K. Wang, T. Tabata, M. Yoshida, D. D. Zhao, T.
Nishimura, K. Kita and K. Nagashio, "Material potential and scalability challenges of
germanium CMOS," in Proc. IEDM 2011, pp.646-‐649, Dec. 2011.
[19] L. Xia, V. Tokranov, S. R. Oktyabrsky and J. A. del Alamo, "Performance
enhancement of p-‐channel InGaAs quantum-‐well FETs by superposition of process-‐
induced uniaxial strain and epitaxially-‐grown biaxial strain," in Proc. IEDM 2011,
pp.315-‐318, Dec. 2011.
[20] S. Datta, T. Ashley, J. Brask, L. Buckle, M. Doczy, M. Emeny, D. Hayes et al.
"85nm gate length enhancement and depletion mode InSb quantum well transistors
87
for ultra high speed and very low power digital logic applications." In Proc. IEDM
2005, pp. 763-‐766, Dec. 2005.
[21] M. Radosavljevic, T. Ashley, A. Andreev, S. D. Coomber, G. Dewey, M. T.
Emeny, M. Fearn et al. "High-‐performance 40nm gate length InSb p-‐channel
compressively strained quantum well field effect transistors for low-‐power (VCC=
0.5 V) logic applications." In Proc. IEDM 2008, Dec. 2008.
[22] L. Desplanque, D. Vignaud, S. Godey, S. Plissard, X. Wallart, P. Liu and H.
Sellier, "Electronic properties of the high electron mobility
Al0.56In0.44Sb/Ga0.5In0.5Sb heterostructure," J. Appl. Phys., vol. 108, no. 4, 043704,
Aug. 2010.
[23] R. Loesch, R. Aldam, L. Kirste and A. Leuther, "Molecular beam epitaxial
growth of metamorphic AlInSb/GaInSb high-‐electron-‐mobility-‐transistor structures
on GaAs substrates for low power and high frequency applications," J. Appl. Phys.,
vol. 109, no. 3, 033706, Feb. 2011.
[24] B. R. Bennett, M. G. Ancona, J. B. Boos and B. V. Shananbrook, "Mobility
P-‐channel MOSFETs with self-‐aligned Ni-‐alloy as S/D were fabricated on both a
bulk n-‐GaSb substrate and a heterostructure stack with 7nm undoped In0.2Ga0.8Sb as
the channel, following similar process as described in [10]. After surface clean, ~8
nm Al2O3 was deposited as the gate dielectric by ALD [8]. 30 nm of W was then
sputtered, followed by a 350°C anneal for 1min in N2/H2 (95%/5%) ambient to
improve the quality of the gate dielectric and recover the plasma damage. Gate
electrodes were patterned using contact lithography, followed by e-‐beam
evaporation of 7nm Ni. Source/drain metal pads were deposited by lift-‐off of 70nm
of Pt. RTA was performed at 300°C for 3mins, followed by etching of unreacted Ni by
HCl.
96
Figure 4.5 Transfer characteristics of devices (LG~5μm) on bulk GaSb substrate and heterostructure stack. Higher on-‐current and lower off-‐current were achieved with
the heterostructure design.
Fig. 4.5 shows the transfer characteristics of devices (LG~5μm) on both the bulk
n-‐GaSb substrate and the heterostructure sample. On the bulk substrate, the ION/IOFF
ratio (~50), as well as subthreshold swing (~250 mV/dec) of the devices, are
largely limited by the ION/IOFF of the Ni-‐GaSb/GaSb Schottky diode. It is shown that
the leakage current is greatly reduced by having a wide-‐bandgap buffer layer
(Al0.8Ga0.2Sb) beneath the channel. A high ION/IOFF of >104 and subthreshold swing of
140mV/dec are achieved with the heterostructure design. The on-‐state current for
the heterostructure design is also higher than that for bulk GaSb design due to the
presence of ~0.7% biaxial compressive strain and confinement of carriers in the
channel [6].
Drain current density scales with 1/LG with channel length ranging from 100μm
to 5μm, indicating low external resistance for the set of devices. Extraction of
external resistance is difficult, considering the channel lengths are comparatively
97
large, thus might result in relative large error in the calculated value using
extrapolation.
Figs. 4.6 and 4.7 show the transfer and output characteristics of the devices
(LG~5μm) with heterostructure design, which shows a peak transconductance of 29
μS/μm at VDS=1V and a decent ION of 42 μA/μm. Further improvement would be
possible with the scaling of the gate dielectric thickness. Fig. 4.8 plots the mobility
extracted using split-‐CV analysis and benchmarks it against the mobility in
unstrained silicon and strained GeSn p-‐channel MOSFETs with a metallic S/D [12].
In the unstrained case, bulk GaSb gives a mobility up to 2 times higher than that of
silicon. Because of the 0.7% biaxial strain, the hole mobility is further improved by 2
times in the heterostructure devices. For antimonide devices, mobility enhancement
drops at high sheet charge density due to the spill-‐over of charge into low-‐mobility
buffer layer and surface roughness scattering [6].
Figure 4.6 Transfer characteristics of the devices with heterostructure design and Ni-
alloy as metal S/D.
98
Figure 4.7 Output characteristics of the devices with heterostructure design and Ni-alloy
as metal S/D.
Figure 4.8 a) The gate to channel capacitance (CGC) measured at 100kHz. b) Extracted effective-‐field mobility using split-‐CV analysis and benchmark against
silicon and strained GeSn.
99
Studies on the formation and electrical properties of Ni-‐antimonide formed by
direct reaction of Ni with antimonide compound demonstrate the Ni-‐antimonide
alloy is a suitable source/drain material in metal source/drain antimonide-‐based p-‐
MOSFETs. A self-‐aligned process for metal S/D MOSFETs using an epitaxial
heterostructure with high hole mobility was developed and demonstrates the
potential of such source/drain design in further channel length scaling of III-‐V p-‐
channel MOSFETs.
4.2 SCHOTTKY BARRIER HEIGHT REDUCTION FOR METAL/N-‐GASB CONTACT BY
INSERTING TIO2 INTERFACIAL LAYER WITH LOW TUNNELING RESISTANCE
Antimonide-‐based III-‐V compound materials have great potential for
applications ranging from next generation complementary metal oxide
semiconductor devices to optoelectronics [13-‐15]. Among the potential candidates
for replacing silicon as channel materials, InSb has theoretically and experimentally
demonstrated high electron mobility [13] and InGaSb shows promising hole
mobility [14]. Band alignments of relevant ternaries and quaternaries are suitable
for optoelectronic devices, e.g. infrared wavelength photo-‐detectors [15]. GaSb
emerges as one of the crucial materials for 6.1Å lattice system based electronic
devices, which has a bandgap of 0.7eV at room temperature, suitable for use in fiber
optic communications systems. For high performance devices in these applications,
Ohmic contact with low contact-‐resistivity is required. However, metal/GaSb
interface exhibits Fermi level pinning near the valence band of GaSb [13,15] which
makes low resistance Ohmic contact formation on n-‐type GaSb difficult. For example
to get low resistance contacts to Sb-‐based HEMTs, alloyed contacts with large
contact area and large source-‐drain separation to prevent shorting by metal
diffusion are employed presently [16-‐18]. Furthermore, poor n-‐type dopant
activation in GaSb limits the tunneling current through the Schottky barrier [15].
100
Therefore, tuning of the electron Schottky barrier height (ΦBN) to make non-‐alloyed
Ohmic contact on lightly doped n-‐type GaSb remains a challenge.
Figure 4.9 Band diagrams of depinning effects (a) by MIGS theory, (b) by interface dipole theory and (c) with low CBO dielectric. In the case of (a) and (b), dielectrics with high CBO introduce high tunneling resistance, while in (c) low Schottky barrier
is achieved with low tunneling resistance due to low CBO.
Large ΦBN can be attributed to Fermi level pinning at metal/n-‐GaSb interface
near the charge neutrality level (ECNL) of the semiconductor, which is found to be
near the valence band for GaSb [19]. By inserting a dielectric between metal and
semiconductor, Fermi level pinning at metal/semiconductor interface can be
101
alleviated [20]. This depinning effect has been explained by metal-‐induced gap
states (MIGS) [21-‐22] (Fig. 4.9(a)) and interfacial dipole effect [23] (Fig. 4.9(b)).
By the MIGS theory, inserting a thin insulator attenuates the metal electron
wavefunction in the insulator prior to penetrating the semiconductor, resulting in
fewer charges to drive the Fermi level towards ECNL. In accordance with interface
dipole theory, by inserting thin dielectrics, electronic dielectric dipole is introduced
between the insulator and semiconductor native oxide, which leads to a barrier
shift. Connelly et al. demonstrated reduction in ΦBN to Si by inserting SiNX between
metal and Si [20]. Other high-‐k dielectrics have also been adopted to tune the
effective barrier height (ΦBNeff) of contacts to Si [24], InGaAs/GaAs [25], as well as to
n-‐type germanium [26]. While the depinning effect is beneficial for reducing
Schottky barrier height, the tunneling barrier introduced by the dielectric layer
leads to additional tunneling resistance, thereby increasing the specific contact
resistivity [20,24-‐26]. For dielectrics such as Al2O3, high tunneling resistance limits
the current level with increase in the thickness of the dielectric (Fig. 4.9(a), (b)). The
optimal thickness for contact resistivity is below 2nm [24-‐25]. Recently, modeling
work suggested using dielectrics with very low conduction band offset (CBO), such
as TiO2, to provide low contact resistivity to n-‐Ge [27], which has similar band
lineup as GaSb and also shows Fermi level pinning at valence band edge. This allows
depinning of the Fermi level without introduction of tunneling resistance (Fig.
4.9(c)).
In this study, we demonstrate band engineering to tune ΦBN without adding
tunneling resistance and achieve increase in current densities for metal/n-‐GaSb
contact by insertion of TiO2. Al/TiO2/n-‐GaSb Schottky diodes with different
thicknesses of TiO2 were fabricated to study the change in ΦBN. Obtained J-‐V
characteristics are compared with control samples with Al2O3 interfacial layers, for
which the current is limited by tunneling resistance. Aluminum (Al) is chosen
102
because of its low workfunction (4.06-‐4.26eV [28]) to better study the Fermi level
movement towards the conduction band of GaSb (electron affinity: 4.06eV [15]).
The band offsets of TiO2/GaSb and Al2O3/GaSb are determined using Synchrotron
Radiation Photoemission Spectroscopy (SRPES).
The Te-‐doped bulk n-‐GaSb substrates grown by the Czochralski (CZ) technique
were chosen to be lightly doped with carrier concentration of ~1017cm-‐3 to put more
emphasis on thermionic injection over thermionic field emission and field-‐emission
current. First, samples were degreased in solvents. Surface cleaning was done by
15% HCl to remove native oxide, directly followed by atomic layer deposition (ALD)
of TiO2 using tetrakis(dimethylamido)titanium (TDMAT) and H2O precursors at
200°C. Control samples of Al2O3 were deposited by ALD at 300°C using
trimethylaluminum (TMA) and H2O as precursors. Al was deposited by e-‐beam
evaporation and patterned as top contact. Backside contact was formed using
blanket evaporation of Al. SRPES is used for the determination of band offsets of
TiO2 and Al2O3 with GaSb [29-‐30]. The SRPES experiments were performed at
beamline 8-‐1 of Stanford Synchrotron Radiation Lightsource, which provides a
tunable range of monochromatic photons up to 160eV, yielding high surface
sensitivity with a minimum electron escape depth around 5 Å. The Ga 3d, Al 2p, Ti
3p and the valence-‐band (VB) spectra can be studied with high accuracy in this low
energy range. By studying the emission from valence electrons, the valence band
offset (VBO) of TiO2 and Al2O3 relative to GaSb can be determined by comparing the
VB spectra after surface cleaning and after deposition of thin TiO2/Al2O3, while
using the Ga 3d peak from the substrate for alignment (Fig. 4.10(a), (b)). The VBO of
TiO2 and Al2O3 on GaSb were measured to be 2.6eV and 3.0eV respectively. The ALD
Al2O3 bandgap was found to be 6.1eV as determined from the energy loss spectrum
of the Al 2p peak shown in Fig. 4.10(c). Due to the low bandgap value of TiO2,
estimation of bandgap from energy loss of Ti 3p peak is difficult. The bandgap of
TiO2 and GaSb are assumed to be within the range of 3.3-‐3.5eV [31] and 0.7eV [15]
103
respectively. CBO can then be calculated by subtracting the bandgap of GaSb and
VBO from the bandgap for both TiO2 and Al2O3. The CBO was thus calculated to be
2.4eV for Al2O3 on GaSb and between 0-‐0.2eV for TiO2 on GaSb. Values of band
offsets are summarized in Fig. 4.10(d).
Figure 4.10 SRPES spectra for (a) TiO2 and GaSb valence bands, (b) Al2O3 and GaSb
valence bands, and (c) Al 2p peak from Al2O3. (d) Summary of band offsets
information for Al2O3 and TiO2 with GaSb. The TiO2/GaSb and Al2O3/GaSb valence
band offsets were measured to be 2.6eV and 3.0eV, respectively. The energy loss
spectrum of the Al 2p peak indicates the bandgap of 6.1eV for Al2O3. Photon energy
used for valence band and Al 2p were 120eV and 160eV respectively.
104
Figure 4.11 J-‐V characteristics of Al/n-‐GaSb Schottky diode with interfacial layer of (a) TiO2 and (b) Al2O3. (a) The current density increases with increase in TiO2
thickness even up to 7.5nm. Contact characteristic changes from rectifying to quasi-‐ohmic. (b) With Al2O3, the current level increases slightly initially and drops significantly for thickness larger than 1nm. The left inset of (a) shows J-‐V characteristics of TiO2 devices for bias voltage from -‐1V to 1V, the rectifying
behavior for Al/n-‐GaSb is clearly visible. The right inset of (a) shows temperature-‐dependent J-‐V measurements for 2nm and 9.5nm TiO2 devices, which show effective
electron barrier heights of 0.14eV and 0.07eV respectively.
105
It has been shown, due to the Fermi-‐level pinning effect, Al/n-‐GaSb Schottky
diodes without the interfacial dielectric layer exhibit rectifying behavior [32-‐33].
ΦBN for Al/n-‐GaSb is measured to be 0.56-‐0.57eV [32], meaning the Fermi level is
pinned towards the GaSb valence band edge at Al/GaSb interface.
Fig. 4.11(a), (b) show the room temperature J-‐V characteristics of the GaSb
Schottky diodes for different thicknesses of the TiO2 and Al2O3 interfacial layer.
Rectifying behavior for Al/n-‐GaSb is clearly visible on the scale plotted in Fig.
4.12(a):left inset. With increasing thicknesses of TiO2, increase in current density
was observed for n-‐GaSb diodes. This trend of increase in current only saturates
when the thickness of TiO2 reaches 7.5nm, indicating tunneling barrier is not a
critical factor limiting electron transport. The J-‐V characteristics change from
rectifying to quasi-‐Ohmic for n-‐type GaSb diodes, indicating a significant reduction
in ΦBN.
Meanwhile, the increase in current is only observed for ultra-‐thin Al2O3 layer
below 1nm (Fig. 4.11(b)), since the tunneling resistance largely limits the current
for thicker Al2O3. The current density for the metal/semiconductor contact structure
consists of the current emission over the Schottky barrier (thermionic emission)
and the current emission through the Schottky barrier (thermionic field emission
and field emission current) [34]. Meanwhile for the metal/dielectric/semiconductor
system it is limited by transport through the barrier of the dielectric layer [20,24-‐
26]. The modeling of current characteristic through this system is a challenging
exercise and there have been some recent papers using Tsu–Esaki model [27] and
non-‐equilibrium green function [25] to model this system. In the absence of a model
that can accurately captures these different components and tunneling currents, we
adopt the thermionic emission model to estimate the reduction in ΦBN with TiO2
insertion through temperature dependent measurements from 77K to 300K [25-‐
26]. The effective Schottky barrier height can be extracted by Richardson plot [25-‐
106
26,35], in which the slope corresponds to -‐qΦBNeff /kB, where q and kB correspond to
electronic charge and Boltzmann constant (Fig. 4.11(a):right inset). ΦBNeff extracted
is not exact due to the presence of dielectric between the metal and semiconductor,
but rather it is a representative of the electrical behavior. By insertion of TiO2 layer,
the effective Schottky barrier height ΦBNeff decreases to 0.14eV and 0.07eV for 2nm
and 9.5nm TiO2 respectively. Increase in dielectric thickness leads to lower
extracted ΦBNeff. Modulation of diode current and the decreasing trend in extracted
barrier height from thermionic emission model verify that the interfacial TiO2 layer
reduces Schottky barrier height.
Fig. 4.12 plots the current density levels at -‐0.1V for the TiO2 and Al2O3 diodes as
a function of dielectric thickness. For Al2O3, the depinning effect leads to the initial
increase in current density. However, high CBO with GaSb (2.4eV) limits the
electron transport as the thickness of Al2O3 increases, adding substantial tunneling
resistance to the contact. The optimal thickness for larger current density has to be
kept ultra-‐thin (~1nm). Meanwhile, electron transport by tunneling through the
dielectric has an exponential dependence on the thickness of the barrier; therefore
the current density is sensitive to the dielectric thickness. This puts challenge to
achieve uniform sub-‐nanometer deposition of the dielectrics on wafer scale in order
to achieve high current density for contacts using high CBO dielectrics, such as
Al2O3. For TiO2, due to the near zero CBO of TiO2 with GaSb (0-‐0.2eV), no excessive
tunneling resistance is introduced by increasing TiO2thickness up to 7.5nm, while
the depinning effect increases current density. As shown in Fig. 4.12, the current
density increases consistently with increase in TiO2 thickness and remains fairly
constant over a large range of dielectric thickness. This proves the use of thicker
dielectric without excessive tunnel resistance such as TiO2 is advantageous.
107
Figure 4.12 Normalized current density of Al/n-‐GaSb Schottky diode with interfacial layer of TiO2 and Al2O3 of different thicknesses biased at -‐0.1V. TiO2 diodes show a substantial increase in current from reduction in ΦBN. Increase in current density with thickness is maintained up to 9.5nm. For Al2O3, after the initial increase, tunneling resistance reduces current density. Current levels are sensitive to the
thickness of Al2O3.
In conclusion, we have demonstrated band-‐engineering of Fermi level depinning
for metal/n-‐GaSb contact with insertion TiO2 interfacial layer. The low CBO leads to
reduction in Schottky barrier height without introduction of excessive tunneling
resistance. The contact characteristics change from rectifying to quasi-‐Ohmic and
more than four orders of magnitude improvement in current density is achieved.
The shift of Schottky barrier height was verified through the modulation of diode
current and temperature dependent measurement. The advantage of using TiO2
108
compared to other dielectrics, e.g. Al2O3 is the low conduction band offset estimated
by SRPES.
109
References
[1] G. Dewey, M. K. Hudait, L. Kangho, R. Pillarisetty, W. Rachmady, M.
Radosavljevic, T. Rakshit, R. Chau, "Carrier Transport in High-‐Mobility III–V
Quantum-‐Well Transistors and Performance Impact for High-‐Speed Low-‐Power
Logic Applications," IEEE Electron Device Lett, vol.29, no.10, pp.1094-‐1097, Oct.
2008.
[2] A. Ali, H. Madan, A. Agrawal, I. Ramirez, R. Misra, J. B. Boos, B. R. Bennett, J.
Lindemuth, S. Datta, "Enhancement-‐Mode Antimonide Quantum-‐Well MOSFETs
With High Electron Mobility and Gigahertz Small-‐Signal Switching
Performance," IEEE Electron Device Lett., vol.32, no.12, pp.1689-‐1691, Dec. 2011.
[3] D.-‐H. Kim, T.-‐W. Kim, R. J. Hill, C. D. Young, Y. K. Chang, C. Hobbs, P. Kirsch, J.
A. del Alamo, R. Jammy, "High-‐Speed E-‐Mode InAs QW MOSFETs
With Al2O3 Insulator for Future RF Applications," IEEE Electron Device Lett., vol.34,
no.2, pp.196-‐198, Feb. 2013.
[4] J. Lin, D. A. Antoniadis, J. A. del Alamo, "Sub-‐30 nm InAs Quantum-‐Well
MOSFETs with self-‐aligned metal contacts and Sub-‐1 nm EOT HfO2 insulator,"
in Proc. IEDM, Dec. 2012, pp. 757-‐760.
[5] M. Xu, R. Wang, P. D. Ye, "GaSb Inversion-‐Mode PMOSFETs With Atomic-‐
Layer-‐Deposited Al2O3 as Gate Dielectric," IEEE Electron Device Lett., vol.32, no.7,
pp.883-‐885, Jul. 2011.
[6] A. Nainani, Z. Yuan, T. Krishnamohan, B. R. Bennett, J. B. Boos, M. Reason, M.
G. Ancona, Y. Nishi, K. C. Saraswat, “InGaSb channel p-‐metal-‐oxide-‐semiconductor
field effect transistors: Effect of strain and heterostructure design,” Journal of
Applied Physics, vol. 110, 014503, Jul., 2011.
110
[7] K. Takei, M. Madsen, H. Fang, R. Kapadia, S. Chuang, H. S. Kim, C.-‐H. Liu, E. Plis,
J. Nah, S. Krishna, Y.-‐L. Chueh, J. Guo, A. Javey, “Nanoscale InGaSb Heterostructure
Membranes on Si Substrates for High Hole Mobility Transistors,” Nano Lett., vol. 12,
pp. 2060-‐2066, Mar. 2012.
[8] A. Nainani, T. Irisawa, Z. Yuan, B. R. Bennett, J. B. Boos, Y. Nishi, K. C. Saraswat,
“Optimization of the Al2O3/GaSb Interface and a High-‐Mobility GaSb pMOSFET,"
IEEE Trans. Electron Device, vol.58, no.10, pp.3407-‐3415, Oct. 2011.
[9] S. H. Kim, M. Yokayama, N. Taoka, R. Nakane, T. Yasuda, O. Ichikawa, N.
Fukuhara, M. Hata, M. Takenaka, S. Takagi, “Sub-‐60 nm deeply-‐scaled channel length
extremely-‐thin body InxGa1−xAs-‐on-‐insulator MOSFETs on Si with Ni-‐InGaAs metal
S/D and MOS interface buffer engineering," in Proc. VLSIT 2012, pp. 177-‐178.
[10] C. B. Zota, S. H. Kim, Y. Asakura, M. Takenaka, S. Takagi, "Self-‐aligned metal
S/D GaSb p-‐MOSFETs using Ni-‐GaSb alloys," in Proc. DRC 2012.
[11] C. B. Zota, S.-‐H. Kim. M. Yokoyama, M. Takenaka, S. Takagi, “Characterization
of Ni-‐GaSb alloys Formed by Direct Reaction of Ni with GaSb,” Applied Physics
Express, vol. 5, 071201, Jul. 2012.
[12] X. Gong, G. Han, F. Bai, S. Su, P. Guo, Y. Yang, R. Cheng, D. Zhang, G. Zhang, C.
Xue, B. Cheng, J. Pan, Z. Zhang, E. S. Tok, D. Antoniadis, Y.-‐C. Yeo, "Germanium–Tin
(GeSn) p-‐Channel MOSFETs Fabricated on (100) and (111) Surface Orientations
With Sub-‐400°C Si2H6 Passivation," IEEE Electron Device Lett., vol.34, no.3, pp.339-‐
341, Mar. 2013.
[13] B.R. Bennett, R. Magno, J.B. Boos, W. Kruppa and M.G. Ancona, ”Antimonide-‐
based compound semiconductors for electronic devices: A review” Solid-‐State Elect.,
vol. 49, pp. 1875-‐1895, Nov. 2005.
111
[14] B. R. Bennett, M. G. Ancona, J. B. Boos and B. V. Shanabrook, “Mobility
[33] F. S. Juang and Y. K. Su, “Electrical properties of Al/n-‐GaSb contact,” Solid-‐
State Elect. vol. 32, p. 661, 1989.
[34] Rideout, V. L. "A review of the theory and technology for ohmic contacts to
group III–V compound semiconductors." Solid-‐State Elect., vol. 18, pp. 541-‐550,
1975.
[35] W. Monch, Electronic properties of semiconductor interfaces, Springer, New
York, 2004
CHAPTER 5: HETERO-‐INTEGRATION OF III-‐V COMPOUND
MATERIALS ON SILICON USING RAPID-‐MELT-‐GROWTH
The integration of III-‐V FinFETs on 300mm silicon wafers is a key technological
challenge due to the large lattice-‐mismatch between III-‐Vs and silicon [1]. Different
requirements of n-‐ and p-‐channel devices pose a challenge to the co-‐integration of
more than one type of material on a silicon platform. In this section, the co-‐
integration of InAs-‐OI NMOS and GaSb-‐OI PMOS on silicon is proposed for its
excellent carrier transport and favorable band-‐lineup. Such hetero-‐integration is
demonstrated on silicon substrate using rapid-‐melt-‐growth technique.
5.1 INTRODUCTION TO RAPID-‐MELT-‐GROWTH
As the research community demonstrates the huge potential of high-‐mobility
alternative channel materials for future high-‐performance, low-‐power logic
applications, one technological challenge remains, which is the monolithic
integration of III-‐V compound semiconductors on a silicon substrate. There are two
conventional methods for integrating high quality single crystal on top of a silicon
substrate: wafer bonding and epitaxial growth. There are many techniques for wafer
bonding. These approaches are usually material-‐specific, meaning that the process
conditions such as the control of temperature and pressure might differ for different
materials [1,2]. In recent years, patterned layer transfer technique has been
developed for a variety of material systems [3,4]. However, the approach cannot be
easily scaled to be compatible with state-‐of-‐art 300mm manufacturing. Compared
115
with wafer bonding which requires two separate wafers, epitaxial growth usually
refers to growing a high quality single crystal layer on a single wafer. Generally, a
relatively thick layer is required to relax the lattice mismatch, followed by several
medium to high temperature annealing cycles to further reduce the density of
defects in the epitaxial layer. The growth of more-‐than 1μm-‐thick buffer layer makes
this approach less favorable for production [5]. Optimization of the buffer layer is
the key to epitaxial growth and the process is usually material-‐specific, making the
integration of more than one type of material difficult.
Besides direct, large-‐area epitaxial growth, there are other approaches to reduce
the buffer layer thickness required to relax the lattice mismatch. For instance,
attempts have been made to develop universal buffer layer to support the growth of
the active device layer on top [6]. Another promising method is called aspect ratio
trapping. This technique uses high aspect ratio sub-‐micron trenches to trap
threading dislocations and reduce the dislocation density of the lattice-‐mismatched
material grown on silicon [7]. The problem with this approach is the incapability of
the control over defect propagation along the trench direction [8].
Rapid-‐Melt-‐Growth (RMG) was first developed and introduced in 2004 [9,10] for
growing single crystal germanium thin films on an amorphous dielectric layer. On a
silicon bulk substrate, amorphous dielectric layer, e.g. SiOx, SiNx, is first deposited as
an isolation layer. Small seeding windows are patterned to expose the underlying
single crystal silicon substrate to serve as the crystalline template. Amorphous or
polycrystalline germanium is deposited uniformly across the wafer and patterned
into stripes with only one end touching silicon in the seed region. The whole
structure is then covered with an oxide layer acting as a micro-‐crucible to hold the
germanium stripe during melting. Typical dielectric materials have melt
temperature higher than that of silicon and can maintain the structural integrity.
Finally, the whole structure is heated up above the melting point of germanium by
116
rapid thermal annealing (RTA) for a short period of time, and then cooled down.
Since the stripe is thermally isolated, heat can only be dissipated through the silicon
seed. As a result, during the cooling process, liquid germanium solidifies from the
seed region; hetero-‐epitaxial growth starts from the crystalline silicon seed, and
then propagates to the other end of the stripe. The same principle has been applied
to III-‐V materials.
A typical structure is shown in Fig. 5.1. One of the advantages of this approach is
the use of the necking mechanism to control the propagation of the defects
generated due to the lattice mismatch between Ge/III-‐Vs and silicon seed, thereby,
confining the defects to the neck region.
Figure 5.1 Top-‐view illustration of the structure for germanium/III-‐V RMG process (left). Cross-‐sectional view of the structure, which is typically used for TEM study of
material properties of the stripe (right).
Despite a variety of experiments showing the robustness of the approach, an
accurate estimation of the defect density of the stripe after RMG is still problematic.
However, large area growth of single crystalline material is difficult with this
Ge /
Ge /
117
method. Therefore, some conventional methods, e.g. XRD, plan-‐view TEM, cannot be
easily adopted to characterize the film quality after RMG. An attempt has been made
to quantify the defect density by aggregating the data from cross-‐sectional TEM of a
number of stripes. The defect density after RMG can be estimated to be ~ 106cm-‐2
[10], which meets the requirement of high-‐performance logic and memory devices.
This demonstrates the potential of achieving low point defect density with RMG
approach. The high crystalline quality of the recrystallized stripes was also
confirmed by electrical measurements. High-‐performance gate all around (GAA)
germanium PMOS transistors were fabricated showing steep subthreshold slope ~
71 mV/dec and 40% higher hole mobility than silicon as in [11].
In 2010, Chen, et al., adopted the same method for III-‐V compound materials for
their hetero-‐integration on silicon [12]. The transformation of amorphous to single
crystal for both GaAs and GaSb on bulk Si substrates was demonstrated. Similar to
the case of germanium, high-‐resolution TEM image and selective area diffraction
patterns were used to study the single-‐crystal films seeded from the Si substrate and
propagated along the patterned stripe on top of the insulator. Besides the
confirmation of the crystallinity of the film, energy-‐dispersive X-‐ray spectroscopy
was applied to investigate the stoichiometry of the compound material after re-‐
solidification. The results show a direct relationship between crystal quality and
atomic composition, which suggests a congruent growth for the III-‐V material
during the solidification process despite the possibility of out gassing. This proves
thatat least for binary III-‐Vs, RMG has this type of “self-‐correction” effect for the
stoichiometry of the recrystallized films.
In comparison to direct wafer bonding [13] and aspect ratio trapping [14], which
are promising techniques for hetero-‐integration of III-‐V on Si, rapid melt growth
(RMG) stands out in overall capabilities across different criteria as shown in Table 3.
118
Table 3: Comparison amongst promising process technologies for hetero-‐integration of III-‐Vs on silicon
Direct Wafer
Bonding
Aspect Ratio
Trapping
Rapid Melt
Growth
Defect control ++ -‐ +
Multiple materials + Buffer
redesign +
300mm compatible -‐ + +
Through-‐put -‐ + +
GAA + -‐ +
Compared with aspect ratio trapping, where defect-‐propagation cannot be
trapped along the length of the trench, RMG efficiently terminates twining, anti-‐
phase defects, etc. by the necking mechanism, an analogous process to the defect
necking from the seed crystal that occurs in Czochralski growth as shown in Fig. 5.2.
119
Figure 5.2 Schematics of defect analysis of aspect ratio trapping (ART) and rapid melt growth (RMG). The control of defect propagation along the length of the trench is difficult for ART, while RMG can efficiently terminate defect propagation by the
necking mechanism.
5.2 CO-‐INTEGRATION OF III-‐V CMOS TRANSISTORS USING RAPID-‐MELT-‐GROWTH
In future transport-‐enhanced CMOS devices, it is favorable to adopt different
channel materials for n-‐ and p-‐channel devices for optimal performance of each
device. In the previous chapter, we argue that the use of different materials for n-‐
and p-‐channel devices brings challenges to process technology. For instance,
common gate stack solution [15] and source/drain technology [14] are difficult to
achieve. Amongst the different candidates, a combination of InGaAs-‐based NMOS
120
and Ge-‐based PMOS stands out from different perspectives. First of all, the electron
and hole mobility in InGaAs and Ge respectively are the highest amongst all
conventional semiconductors. From source/drain and gate passivation point of
view, the band lineup of In-‐rich InGaAs and Ge is also very favorable. The charge-‐
neutrality level (ECNL) lies in the conduction band of InAs, and is close to the valence
band of Ge [16]. However, the co-‐integration of InGaAs and Ge on silicon substrate is
very challenging. Besides the involvement of multiple epitaxial growths, the cross
contamination issue is critical and inherent with this approach, as both group IV and
III-‐V materials are co-‐integrated on silicon substrate.
To work around this problem, we proposed the use of all III-‐V CMOS transistors.
Because of the superior and unique hole transport properties, antimonides are
identified as promising candidates for p-‐channel material. Electron transport is
found to be excellent in the same material system as well. At the same time, due to
charge-‐neutrality-‐level (ECNL) being close to the valence band, it is very difficult to
achieve low interface defect density towards the conduction band, and low contact
resistivity. Within III-‐Vs, ECNL for InAs and GaSb is located near the
conduction/valence band respectively [16], as shown in Fig. 5.3. Bulk electron/hole
mobility in InAs/GaSb is measured to be ~10,000 / ~1,000 cm2/Vs [17]. However,
the challengeis then to co-‐integrate both arsenide and antimonide on silicon with a
common gate-‐stack and source/drain solution. In this section, we demonstrate the
co-‐integration of both InAs NMOS and GaSb PMOS on silicon using rapid-‐melt-‐
growth. Material characterization of the recrystallized film shows the feasibility of
this approach for both InAs and GaSb and good material quality. InAs-‐OI, GaSb-‐OI
process is further demonstrated.
121
Figure 5.3 InAs NMOS, GaSb PMOS offer advantages in favorable band line-‐up relative to ECNL and excellent electron and hole transport properties.
The band-‐lineup with respect to ECNL leads to low Dit level near valence-‐band
edge of GaSb and low Schottky barrier height for holes indicated by I-‐V dependence
on temperature, both favored for PMOS operation, as shown in previous chapters.
Similar argument also holds for InAs NMOS.
Table 4: Melting temperature and lattice mismatch with respect to silicon of InAs and GaSb
Si InAs GaSb
Melting T (°C) 1412 942 712
Mismatch % 0 11.7 12.5
Since the deterministic parameter in RMG is the melting temperature, the
technique can be easily applied to more than one III-‐V material regardless of their
degrees of lattice mismatch relative to silicon. The lattice constant of InAs and GaSb
are 6.0583Å and 6.0959Å at 300K. This means more than 11% lattice mismatch with
122
silicon. The principle of rapid melt growth guarantees the control over the
propagation of misfit dislocations. Therefore, InAs NMOS and GaSb PMOS can be co-‐
integrated on silicon using RMG.
Table 4 shows the melting temperature and the lattice mismatch against silicon
of InAs and GaSb, respectively. The melting temperature of InAs is 942°C, which is
more than 400°C below that of silicon. The phase diagram of In-‐Si shows that
alloying with In will lower the melting temperature of the alloy [18]. The melting
temperature drops sharply with In ratio higher than 90%. In fact, the same principle
has been applied to achieve melt-‐induced-‐crystallization (MIC) [19], in which the
lowering of melting temperature is used to induce regrowth during annealing to
form crystalline structure. During rapid melt growth, at the interface of III-‐V
materials and Si, such phenomenon is expected to cause the melting of the silicon
seed even though the annealing temperature is well below the melting temperature
of silicon. Once the material is in liquid phase, the diffusivity of silicon and III-‐V
elements could be order-‐of-‐magnitude higher than those when the material is in
solid phase. Even during the short period of time of growth, which is estimated to be
on μs level based on a growth rate of more than 1m/s, the intermixing between
silicon and III-‐Vs could be a critical issue, especially when the annealing time is
constrained to a couple of seconds in RTA system. Therefore, the lower melting
temperature of InAs and GaSb, compared to GaAs (1238°C), InP (1062°C), etc., is
favorable for the suppression of such intermixing effect.
Another effect associated with melting temperature is the high vapor pressure
caused by the evaporation of group V atoms. It is difficult to measure the exact
vapor pressure at high temperature close to the melting temperature of the III-‐V
materials. It can be estimated close to 1000°C, arsenic vapor pressure can be larger
than 1 atm. The strength of the micro crucible material may not be strong enough to
hold the molten material. Thus, for the integrity of the structure, lower melting
123
temperature of the III-‐V materials is also desired. In summary, the combination of
InAs and GaSb stands out in terms of low melting temperature.
Due to lattice mismatch, there are a number of point defects or planar defects
that can be generated, including stacking faults, twins, and domain boundaries.
Different types of defects have different impact on the electrical properties of the
film. Planar defects lead to leakage conduction path. Point defects, on the other
hand, could be electrically active and of charged states, thus act similar to dopant-‐
like inter-‐band energy levels.
RMG process for III-‐Vs is very similar to that for germanium. The III-‐V material is
deposited on top of a dielectric on silicon wafer by MBE or sputtering, patterned and
then encapsulated with LPCVD SiO2. Stripes are heated beyond the melting
temperature using RTA. During the cooling process, epitaxial growth starts from Si
seed and propagates through the III-‐V stripe.
Transmission Electron Microscopy (TEM) is a common approach to study the
defect formation during epitaxial growth. Planar defects can usually be clearly
identified by cross-‐sectional TEM, while a more sophisticated approach has to be
used to study point defects, e.g. plan view TEM. However, constrained by the
geometry of the stripe, which is at most 2-‐5μm in width, a plan view TEM is hard to
achieve due to the difficulty in sample preparation. Other approaches like X-‐ray
diffraction is also difficult for the structure as well. As the stripes are isolated, the
density of the materials may not be high enough to give the signal strength in XRD.
Therefore, TEM is used to study the material quality of the III-‐V materials after rapid
melt growth.
For hetero-‐epitaxy, lattice misfit can be accommodated by inclined stacking fault
defects. A stacking fault occurs with an extra plane of atoms inserted into or the
removal of one plane in the stacking sequence. The typical width of stacking fault in
124
GaSb can be estimated to be ~15nm, which could be clearly identified in TEM [20].
As shown in Fig. 5.4, TEM images of GaSb after RMG show no such defects in the
lateral part of the stripe on large-‐scale. This indicates that the lattice misfit is
released near the interface between GaSb and silicon and not affecting in the lateral
part of the GaSb, where the devices are to be built.
Figure 5.4 TEM images of GaSb stripe after RMG show no visible defects, e.g. threading dislocation, stacking faults, etc.
Twin is one type of planar defect resulting from a change in the stacking
sequence during crystal growth. During crystal growth and there is not a gradual
change in crystallographic orientation, as the twin boundary moves across a region,
the crystal lattice flips, and a discrete increment of deformation is attained,
indicated by twinning features.
125
Figure 5.5 TEM images of GaSb near the Si seed after RMG. Defects, e.g. twinnings, etc., only propagate over short range due to the efficient termination of defects near
the neck region.
At the beginning of the growth, the original growth plane is (001), the surface of
the twinned crystal is the (221) plane, which bring the surface to various (221)
planes or back to (001). As shown in Fig. 5.5, TEM images of GaSb stripe near the
silicon seed window indicate that twinning only appears over a relatively small
range away from the interface between GaSb and Si. This demonstrates the efficient
termination of defect propagation near the neck region.
HR-‐TEM of GaSb/InAs stripes 1μm away from the seed window, as in Fig. 5.6,
shows high-‐quality crystalline structure. The seed silicon wafer has a crystal
orientation of (001). As indicated by the diffraction pattern, the crystal orientation
of GaSb/InAs is preserved from the silicon seed. From the reciprocal distance
between the dots of different crystal planes, it confirms the lattice constants of
GaSb/InAs are close to those of relaxed bulk materials.
126
Figure 5.6 High-‐resolution TEM images of GaSb and InAs ~1μm away from the seed windows. Diffraction-‐patterns indicate single-‐crystallinality. Crystal orientation of Si
seed is preserved.
Figure 5.7 Diffraction patterns taken at different locations along the stripe of a) GaSb, b) InAs 1μm, 2μm, 3μm away from seed confirm the propagation of the
growth as zone axis remains unchanged. The distance from the seed is indicated in the figure.
127
As shown in Fig. 5.7, the diffraction patterns taken along the stripes confirm the
propagation of the growth as the zone axis remains unchanged up to 3μm away
from the silicon seed window. This also indicates that there is no twisting in the
crystal orientation during the growth.
Figure 5.8 Tilted-‐view SEM image showing rough surface and poor coverage around the neck region for InAs thin film grown on SiNx by MOCVD.
Smooth surface morphology is critical in achieving good carrier transport in
MOSFETs devices. Without the involvement of additional process step, e.g. chemical
mechanical planarization (CMP), the surface roughness of the melted materials is
largely determined by the as-‐deposited films. Physical vapor deposition (PVD) of the
initial III-‐V thin films offers advantages over chemical vapor deposition (CVD) from
this perspective. The nucleation process involved in CVD process is hard to initiate
on the amorphous dielectrics, resulting in multiple domains or grains and rough
surface as shown in Fig. 5.8. Another consideration is step coverage at the edge of
the seed window. If the film thickness is to be scaled down, large surface roughness
may result in partial coverage of the step of the dielectric material as shown in the
SEM images in Fig. 5.9. During the melt-‐regrowth, due to the surface tension of the
2μm
128
liquid, the material can break at the neck region, which stops the propagation of the
epitaxial growth. Depending on the deposition approach for the initial thin film,
even though the crystalline quality of the as-‐deposited thin film is not important,
optimization is still necessary to achieve good step coverage and low surface
roughness.
Figure 5.9 Tilted-‐view and cross-‐sectional SEM images of GaSb thin film on dielectric on silicon, showing lack of step coverage around the neck region due to the poor
nucleation during MBE growth of GaSb thin film.
500nm
500nm
129
Figure 5.10 Top-‐view and cross-‐sectional SEM images of GaSb thin film on dielectric on silicon showing smooth surface on dielectric and step coverage.
A low-‐temperature MBE growth is developed to achieve uniform nucleation of
the III-‐V layer on dielectric as shown in Fig. 5.10. For GaSb, growth temperature as
low as 300°C is used to prevent the nucleation issue on dielectric. The lower limit
for growth temperature is set by the stoichiometry of the film. MBE is usually grown
with high ratio between group V / III flows to prevent the desorption of the III-‐V
materials. At a low temperature, this may result in accumulation of group V atoms.
Even though rapid-‐melt-‐growth has the self-‐correction effect [12] to expel the
excessive element toward the end of the stripe, it is still favorable to have the film
stoichiometry correct to begin with. XPS analysis on low-‐temperature MBE grown
2μm
200nm
130
GaSb shows that even at 300°C, the ratio between group V and III atoms can be
preserved reasonably well to 1:1, while achieving a smooth surface.
Figure 5.11 AFM measurement on as-‐deposited film and stripe after rapid-‐melt-‐growth, showing the surface roughness is improved after RMG.
Fig. 5.11 shows the atomic-‐force microscopy measurement on the as-‐deposited
GaSb thin film and the stripe after rapid-‐melt-‐growth process. The surface
roughness after RMG is improved over the as-‐deposited film. The improvement
potentially comes from the wet etching process involved. In principle, the
morphology of the thin film should be preserved during the melting by the micro-‐
crucible, thus there will not be a degradation of surface roughness due to the
melting process.
131
Figure 5.12 Raman spectra of bulk and RMG GaSb show similar FWHM and peak position.
Micro-‐Raman spectroscopy is another available approach to study the materials
quality of the thin film after rapid-‐melt-‐growth. For comparison, CZ-‐grown bulk
GaSb is used as control sample. As shown in Fig. 5.12, Raman spectroscopy shows
similar position and FWHM of LO-‐mode peak compared to bulk sample, indicating
good crystalline quality.
Figure 5.13 SEM image of the TLM structure for GaSb stripe.
132
To study the electrical properties of the thin film, structure for transfer length
method is built on stripes after RMG as shown in Fig. 5.13. The melted thin film of
GaSb shows sheet resistance of 9300Ω/sq, for a film thickness of 120nm.
Considering point defects are electrically active in GaSb, high resistivity number
could be a suggestion of good crystalline quality. Good contacts with ρC of 7.9×10-‐
6Ωcm2 were obtained by alloying with Ni, together with more than 10 times
reduction in sheet resistance.
Figure 5.14 TLM measurements on GaSb stripes give high sheet resistance,
indicating low concentration of defects, which are electrically active; Ni-‐alloy shows
low sheet resistance and ρc.
A common gate stack and source/drain technology was developed as shown in
Fig. 5.15. The RMG process was very similar for both InAs and GaSb except for the
annealing temperature. After obtaining the single-‐crystalline stripes of the III-‐V
materials, gate dielectric of Al2O3 was deposited by ALD. W was deposited by
133
sputtering for better sidewall coverage, and then patterned as gate electrode. Thin
Ni was deposited and annealed to form Ni-‐III-‐V alloys, followed by the etching of
unreacted Ni. Finally, source/drain contact pads were formed by lift-‐off of Ti/Pt.
The process flow is summarized in Fig. 5.16. SEM image of the fabricated device is
shown in Fig. 5.17.
Figure 5.15 InAs NMOS, GaSb PMOS with common device process flow with the same gate stack and self-‐aligned Ni-‐alloy S/D.
Figure 5.16 Process flow for InAs-‐OI NMOS, GaSb-‐OI PMOS transistors with Ni-‐alloy SDs integrated on silicon using RMG.
134
Transfer characteristic of fabricated transistor shows that an on/off ratio of 103
can be achieved as in Fig. 5.18. The on/off ratio is currently limited by the film
thickness. Due to the low band gap of GaSb (~0.7eV), a thinner film is required to
achieve lower off-‐state current. On the other hand, the lower limit of the film
thickness is constraint by the step coverage issue. On-‐current of the device can also
be further improved with smoother surface.
Figure 5.17 SEM image of GaSb PMOS on insulator on silicon integrated with RMG process.
Figure 5.18 Transfer characteristics of GaSb-‐OI PMOS showing a on/off ratio higher 103.
135
Back-‐surface between III-‐Vs and the underlying dielectric may not be well
passivated, causing back-‐side leakage. Besides the scaling of film thickness, another
solution to the problem is through the use of gate-‐all-‐around (GAA) structure, in
which the underlying dielectric could be etched away. Fig. 5.19 shows the side-‐view
SEM image of the stripe after RMG. Undercut of the dielectric beneath was obtained
by wet-‐etch.
Figure 5.19 SEM image of sidewall profile of GaSb stripe after RMG shows that underlying dielectric can potentially be etched away to enable GAA structure.
Figure 5.20 Output characteristics of InAs-‐OI NMOS, GaSb-‐OI PMOS transistors integrated on silicon using RMG.
136
Finally, the output characteristics of GaSb-‐OI PMOS, InAs-‐OI NMOS with self-‐
aligned Ni-‐alloy SD integrated on silicon are shown in Fig. 5.20. It demonstrates the
potential of RMG approach for the co-‐integration of more than one type of material
on silicon for CMOS transistor fabrication.
In summary, long stripes of InAs and GaSb stripes with high crystalline quality
suitable for making SOI-‐FinFET transistors were fabricated on silicon substrates
using RMG. III-‐V-‐O-‐I transistors with ION/IOFF > 103 were demonstrated. RMG
enables a path for transistors with strong electrostatic control and III-‐V CMOS using
multiple materials on silicon substrates.
137
References
[1] U. Gosele, “Semiconductor wafer bonding,” Annual Review of Material Science,
vol. 28 pp. 215-‐241, 1998.
[2] G. Taraschi, “Strained Si, SiGe, and Ge on-‐insulator: Review of wafer bonding