-
III-V and III-Nitride engineered heterostructures: wafer
bonding, ion slicing and more
O. Moutanabbir a, S. Christiansen a, S. Senz a, R. Scholz a, M.
Petzold b,
and U. Gösele a
a Max Planck Institute of Microstructure Physics, Weinberg 2,
Halle (Saale), D 06120 Germany
b Fraunhofer Institute for Mechanics of Materials, Halle,
Germany
Wafer bonding in combination with ion slicing emerges as a
flexible technology by which bulk quality thin layers can be
transferred onto different host materials. In the first part of
this paper a successful application of this process to transfer
4-inch InP layer onto Si wafer is demonstrated. The use of an SiO2
interlayer makes the fabricated heterostructures compatible with
high temperature processes. In the second section we address the
applicability of this process for 2-inch freestanding GaN wafers
which exhibit a strong post-implantation bowing making any bonding
exceedingly difficult. We describe the origin of this bow
enhancement and present a novel strategy to manipulate it. Based on
our approach a successful bonding of 2-inch free-standing GaN onto
sapphire handle wafers is achieved. Finally, by using a variety of
experimental techniques, we explore the atomic processes and
structural transformations involved in H ion-induced GaN splitting.
A plausible mechanistic picture is presented.
Introduction
The first demonstration in the mid 90’s of silicon thin layer
transfer using ion-cut
drew a great deal of attention (1). This process presents an
effective approach for the integration of bulk quality thin layers
onto different substrates achieving a wide variety of
heterostructures frequently unattainable by epitaxy. Moreover, the
concept of repeated transfer from the same wafer makes the process
economically very attractive. Figure 1 illustrates, very
schematically, the scenario leading to the subsurface layer
transfer. Ion slicing consists of hydrogen (H) and/or helium (He)
ion implantation into a donor wafer before bonding it to a handle
wafer. The physical and chemical interactions of the implanted
species with radiation damage and their thermal evolution act as an
atomic scalpel leading to the formation of extended internal
surfaces and to the complete exfoliation of the top layer of the
implanted wafer. This phenomenon can be easily controlled by
adjusting the implantation conditions. However, the most critical
step in ion-slicing has to do with wafer bonding. In fact,
achieving high quality and thermally stable bonding requires
atomically flat surfaces and the absence of any irregularities in
the wafers to be bonded such as the long range waviness and the bow
(2). In addition, the surfaces have to be adequately activated
before bringing them into contact. This surface engineering step
depends sensitively on the nature of the materials to be bonded and
on the subsequent thermal treatments.
ECS Transactions, 16 (8) 251-262 (2008)10.1149/1.2982876 © The
Electrochemical Society
251Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
In this paper, we focus on ion-slicing of 4-inch InP and 2-inch
freestanding (fs-) GaN wafers. In the first section, we will
demonstrate how epitaxy-compatible InP-on-Si substrates can be
achieved. In the second section, we will address different issues
involved in the case of 2-inch fs-GaN. Finally, by using a variety
of experimental techniques, we explore the atomic processes and
structural transformations involved in hydrogen ion-induced GaN
splitting. Understanding these fundamental aspects is vital in
order to control and optimize the ion-cut process.
Figure 1: Schematic illustration of the two steps involved in
the ion-cutting process.
4-inch InP-on-Si heterostructures for high temperature
processing
InP-on-Si is a very attractive heterostructure making
conceivable the monolithic
integration of electronic and optoelectronic devices in the same
platform and the realization of all Si-based optical communications
(3). In addition, these heterostructures can also be used as
templates for the fabrication of cost-effective and high
performance solar cells (4). However, due to the high lattice
mismatch between InP and Si (8.1%), direct heteroepitaxy leads to
large threading and misfit dislocation densities in the grown
layers and deterioration of the optical and electronic properties.
Wafer bonding, in combination with ion-slicing, provides an
economically attractive approach to fabricate bulk quality
InP-on-Si substrates. However, due to their rather different
thermal expansion coefficients (Si ~ 2.6×10-6 K-1 and InP ~
4.6×10-6 K-1), the realization and use of such heterogeneous
substrates face thermal stress problems which usually lead to
debonding before splitting.
The transfer of 2-inch and 3-inch films onto Si wafers was
already demonstrated (5-7).
Extrapolating this concept to 4-inch InP wafers is
technologically highly relevant. However, with a thickness of more
than 600 µm, the interfacial thermal stress will be much more
pronounced in this case. Innovative approaches are needed to
circumvent this problem. A first successful attempt to transfer
4-inch InP layers was achieved by Singh et al. (8). In that work
the InP wafers were coated with a 150-nm-thick spin-on-glass (SOG)
layer before bonding to thermally oxidized Si(001) handle wafers.
Using a SOG intermediate layer leads to a very high surface energy
which helps in avoiding thermal stress-induced debonding (9).
However, the advantage of SOG is diminished (if not canceled) by
the fact that the fabricated heterostructures cannot sustain high
temperature
ECS Transactions, 16 (8) 251-262 (2008)
252Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
in subsequent epitaxy and device fabrication. SiO2 grown by
plasma-enhanced chemical vapor deposition (PECVD) would be the
appropriate interlayer for high temperature processing of InP-on-Si
substrates. However, due to relatively lower surface energy in the
case of hydrophilic bonding, thermal stress problems will have a
greater impact. Here, we propose an effective preannealing
procedure to overcome these problems and achieve a successful InP
layer transfer.
In this work, 4-inch semi-insulting (100) InP wafers (Freiberger
Compound Materials
GmbH) were used. A 150-nm-thick SiO2 layer was deposited on InP
by PECVD. To prevent undesired out-gassing from the PECVD oxide
layer during subsequent heat treatments of the bonded wafer pairs,
the wafers were annealed at 850 oC in N2 atmosphere after
SiO2-layer deposition. Oxide-deposited wafers were subsequently
mirror-polished using chemo-mechanical polishing (CMP). The surface
RMS roughness was in the order of 0.2 nm/10×10 µm2 after CMP,
ensuring the short range flatness required for direct bonding. The
wafers were then subject to He+ ion implantation at -15 oC at 100
keV with a fluence of 5×1016 cm-2. The implanted wafers were bonded
at room temperature to thermally oxidized Si(001) handle wafers
using a Süss Microtec CL200 cleaner. The bonding process was
carried out in a class 1 cleanroom. The quality of the bonded
interface was confirmed using infrared transmission imaging
set-up.
Figure 2. XTEM micrograph of InP layer transferred onto Si wafer
using a PECVD
grown SiO2 interlayer. In order to strengthen the bonding, the
bonded pairs were annealed at 150 oC for 24
hours. Unlike InP/SOG/Si pairs (8), however, annealing at 200 oC
necessary to initiate the splitting process leads to the cracking
and breakage of the bonded wafers before the layer transfer occurs.
As mentioned above, this is a consequence of the thermal mismatch
and the relatively weak surface energy in the hydrophilic bonding.
Fortunately enough, our systematic studies of the influence of
pressure on bonding stability revealed that the thermal stress can
be easily circumvented by applying a pressure of 2 MPa during
annealing at 200 oC. After annealing for 30 min, the pressure was
released and InP splitting takes place. Figure 2 displays a
cross-sectional transmission electron microscope (XTEM) image of
the as-split InP thin layer onto Si. The transferred layer is
nearly 550-nm-thick. The top 220 nm is heavily damaged. Atomic
force microscopy (AFM) analysis shows that the surface is very
rough [Fig. 3 (a)]. Obviously, any growth on such substrates will
lead to poor quality device structures. Therefore, it is necessary
to remove the residual defective layer. For this reason, the
as-split InP-on-Si wafers were subject to a chemical mechanical
polishing (CMP). To optimize this process, we explored several
slurries and polishing parameters. The optimal procedure consists
of using a mixture of syton and water (1:20) and applying a
pressure of 3 kPa to the wafer jig. After polishing, the obtained
surface RMS roughness was in the order of 0.5 nm/20×20 µm2 [Fig.
3(b)].
InP
SiO2 (PECVD)
Si
Surface
Damage layer
150 nm
ECS Transactions, 16 (8) 251-262 (2008)
253Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
The crystalline quality of InP layer was confirmed by x-ray
diffraction (XRD). Work is underway to grow device quality
heterostructures on our InP-on-Si substrates.
Figure 3. AFM images of InP-on-Si substrates immediately after
splitting (a) and
after polishing (b). Note the difference in X-Y scales. The
corresponding RMS roughness is indicated.
Bonding of H-implanted 2-inch fs-GaN
GaN and related materials hold promise of future solid-state
lightening, advanced
optoelectronics and even high-power radio-frequency electronics
(10). However, the potential of these devices is still limited by
the quality of the grown heterostructures. Presently, GaN used in
device fabrication is epitaxially grown on sapphire despite its
poor lattice and thermal match to GaN. The densities of misfit and
threading dislocations in GaN layers deposited on sapphire range
typically from 108 to 1010 cm−2 whereby the efficiency of GaN
devices is limited. High quality GaN bulk substrates can be
produced by hydride vapor phase epitaxy growth of thick GaN layers
on sapphire and subsequent separation from the sapphire substrate.
The current cost of these freestanding wafers is still so high that
the concept of transfer of many layers from one fs-GaN wafer to
appropriate host substrates by ion-cut is technologically and
economically highly attractive. Here, we address some issues
related to the application of H-ion cutting of 2-inch fs-GaN
wafers.
Achieving high quality wafer bonding presents the most critical
step in the process
(11). In fact, one of the major problems faced in bonding of
2-inch free standing GaN wafers is the strong enhancement of the
bow due to hydrogen implantation (12). In this work, we present a
novel approach to manipulate the implantation induced bowing
phenomenon. By strain engineering at the back-side of the
H-implanted fs-GaN wafer, we achieved a bow reduction in the first
place and consequently high quality direct wafer bonding of the
H-implanted 2-inch fs-GaN wafers to sapphire handle wafers. This is
a crucial step towards fs-GaN thin layer transfer onto foreign
substrates.
Experimental details
~300 µm-thick 2-inch double side polished fs-GaN wafers were
used in this study.
Some wafers were subject to room temperature hydrogen ion
implantation under the optimal conditions of ion-cutting at the
energy of 50 keV with a fluence of 2.6 × 1017 atom/cm2. Elastic
recoil detection (ERD) analysis shows that the H-implantation depth
is ~450 nm with a concentration peak of ~13% around 320 nm under
the selected
RMS = 39.63 nm
RMS = 0.5 nm
(a) (b)
ECS Transactions, 16 (8) 251-262 (2008)
254Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
implantation conditions (13). The bow of the wafers was measured
using a long range DEKTAK 8 stylus profilometer on a length of 4.5
cm with a horizontal resolution of 3 µm. Microstructural
information about H-ion implantation induced damage was obtained
using (XTEM) and (XRD).
For bonding experiments, a 100 nm-thick SiO2 layer was deposited
on fs-GaN by
PECVD. SiO2-deposited fs-GaN wafers were annealed at 850 oC in
N2 atmosphere to
avoid any undesirable outgassing. The polished fs-GaN wafers
with SiO2-layer and sapphire wafers (handle wafers) received a 15
min piranha (H2O2:H2SO4 = 1:3) solution cleaning followed by a
deionized water rinse and N2 gun dry. After these cleaning steps
the surfaces were terminated with hydroxyl (OH-) groups necessary
to initiate the bonding (2).
Post-implantation bowing
During the implantation of energetic H ions several physical and
chemical processes
take place leading to a variety of radiation damage-related
structures including interstitials, vacancies, hydrogen-point
defect complexes, voids, and free hydrogen. XTEM images of
as-implanted fs-GaN wafers show a broad damage band extending over
a 300 nm-thick layer below the surface [Fig. 4(a)].
Figure 4. (a) XTEM image of damage band induced in GaN by H ion
implantation at
50 keV with a fluence of 2.6 × 1017 atom/cm2. (b) X-ray θ/2θ
scans of (0002) fs-GaN after H implantation.
Typical x-ray diffraction data of fs-GaN before and after H
implantation are presented
in Fig. 4(b). We notice that H implantation creates a
significant out-of-plane tensile strain which is detectable as
interference fringes extending from the left side of the (0002) GaN
diffraction peak. This out-of-plane tensile strain is accompanying
an in-plane compressive stress. As it will be addressed below, this
radiation damage-induced in-plane compressive-strained layer has
undesirable consequences on direct bonding of H-implanted fs-GaN
wafers.
Figures 5 display the profilometer measurements at the N face of
the fs-GaN wafers
for three different 2-inch wafers labeled A, B, and C. We note
that all three wafers are dome-shaped with an initial bow (dashed
lines) in the order of 9.9, 21.8, and 23.75 µm for wafers A, B, and
C, respectively.
Damage band
50 nm -4000 -3000 -2000 -1000 0
100101102103104105
Inte
nsity
θ/2θ [arcsec]
out-of-plane tensile strain
HRXRD (0002)(a) (b)
ECS Transactions, 16 (8) 251-262 (2008)
255Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
A spectacular increase in the bow is recorded after
H-implantation under ion-cut conditions [Fig. 5, red/dark lines].
We found that post-implantation bow reaches 38.9, 59.7, and 62 µm
for wafers A, B, and C, respectively. This strong enhancement is a
result of the in-plane compressive strain induced by high dose
H-implantation as evidenced by x-ray analysis [Fig. 4(b)]. The
average stress in the damaged layer can be approximately estimated
from bow measurements (12). By analogy to a heteroepitaxial
compressively strained layer (14), the average stress in the
damaged layer can be estimated using the modified Stoney’s formula
(15):
f
s
tL
btE
2
2
31
∆⎟⎠⎞
⎜⎝⎛
−=
νσ [1]
where E is Young’s modulus of GaN, ν is Poisson’s ratio, ∆b is
the variation of the wafer bow induced by H-implantation, L is half
of the profilometer scan length, ts and tf represent the
thicknesses of the substrate and the damaged layer, respectively.
Using the data from Fig. 5A-5C, the average stress in the implanted
region was found to be in the order of -1.4 GPa.
Figure 5. Bow measurements on three different 2-inch fs-GaN
wafers labeled A, B,
and C: unimplanted (dashed lines), single side (red/dark lines)
and double side H-implanted (green/light lines).
The bowing of H-implanted fs-GaN wafers strongly limits or even
prohibits direct
bonding and, consequently, the application of the ion-cut
process to transfer thin layers of high quality fs-GaN to other
host materials. It is well known that for direct bonding the wafers
should be as flat as possible (2). The tolerable bows depend
sensitively on the material properties and wafer thickness.
Unfortunately, the observed post-implantation bowing is too high to
allow any contacting of the wafers to be paired and bonded. Indeed,
our bonding experiments on as-implanted wafers show that the gap
between the two
0 1 2 3 40
20
40
60 C
Scan distance [cm]
0
20
40
60
Bow
[µm
]
0
20
40
B
A
ECS Transactions, 16 (8) 251-262 (2008)
256Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
surfaces is too large to even permit van der Waals forces to
come into play and to initiate the bonding process [Fig. 6(a)]. It
is also worth mentioning that our bonding tests under applied
forces (e.g. a load in the bonding machine) have led to a
systematic breakage and cracking of the wafers to be bonded and
therefore does not present a real alternative option [Fig. 6 (b)].
The use of an intermediate layer as adhesive to ‘glue’ H-implanted
fs-GaN to the handle wafer can lead to room temperature bonding.
However, this approach is not suitable in our case since the
subsequent steps of our process involve thermal treatments at high
temperatures, e.g. the splitting at ~700 oC and the heteroepitaxy
of device structures on the transferred GaN layers at at least 1050
oC. To the best of our knowledge, the commonly used adhesive layers
cannot stand such high temperatures. Therefore, to avoid any
complication in the process, the bonding via adhesive layers was
not considered in our case. For technological reasons, the main
challenge in fs-GaN layer transfer remains the reduction of the
post-implantation bow in order to meet the direct bonding
criterion.
Figure 6. (a) Schematic illustration of the bow-induced gap at
the interface of the
wafers to be bonded. (b) Optical image of H-implanted fs-GaN (a
half of 2-inch wafer) bonded to 2-inch sapphire wafer annealed
under a pressure of 200 kPa. Note that due to radiation damage
fs-GaN wafer exhibit a change in color from transparent to
golden-brown after H implantation.
Bow manipulation by strain engineering
As it is addressed above, the strain at or below the surface
strongly influences the
wafer bow. Similarly, a compressively strained layer at the back
side (the Ga-face) of the H-implanted fs-GaN wafer (implantation on
the N-face of the wafer) can enhance the bow from the Ga face of
the wafer, naturally, thereby reducing the bow at the N-face of the
wafer. The final value of the bow depends directly on the amount of
strain induced from the Ga-face. Our finite element calculations
demonstrate that the bow can be effectively reduced by ~87% when
the strain is locally manipulated at the back side of the implanted
2-inch fs-GaN wafers [Figure 7]. However, this will remain a simple
theoretical approach which can be hardly achieved
experimentally.
Figure 7. Finite element modeling of bow reduction by wafer back
side stress
manipulation.
Bowed wafer
1.0% 0.6% 0.5%
Flat wafer
Implanted zone
Handle Bonded area
Cracks
ECS Transactions, 16 (8) 251-262 (2008)
257Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
A strained layer in the back side of H-implanted wafers can be
induced by mechanical polishing, heteroepitaxy, plasma treatment,
or radiation damage. Ion implantation appears to be the most
evident choice in our case since it can be included without any
additional setup in the ion-cut process. Ga-face H-implantation
under the exact conditions will certainly equibalance the bow
enhancement observed after H-implantation at the N-face and the
final bow on the wafer surface to be bonded can be reduced down to
its initial value. Consequently, direct bonding of these wafers
will sensitively depend on the bow of the original wafers. From
data reported in Fig. 5, we see that unimplanted fs-GaN wafers
exhibit already a large bow which cannot be tolerated in the direct
bonding. Hence, in this specific case, further decrease of the bow
is required in order to meet the bonding conditions. By examining
Eq. 1, we note that the bow variation ∆b scales linearly with the
thickness of the damaged layer tf at a fixed value σ, and vise
versa. This means that producing a larger damage band or a higher
strain can decrease the bow below its initial value. Since the
thickness of the damage band depends on the H-ion energy and strain
value varies linearly with the implanted fluence, the bow can be
manipulated by adjusting the implantation energy and/or fluence. In
this study, we chose to increase slightly the implantation depth
while keeping the same ion fluence of 2.6 × 1017 H/cm2. fs-GaN
wafers were implanted from the back side at about 65 nm deeper than
the first implantation on the N-face. Results of profilometer
measurements after double side implantations are shown in Fig. 5
(green/light lines). Expectedly, all the wafers exhibit a bow
slightly smaller than the initial value.
Bonding of H-implanted 2-inch fs-GaN for potential layer
transfer
Using double-side implantation direct bonding of H-implanted
2-inch fs-GaN wafers
to sapphire wafers has become possible. Fig. 8 displays images
of a bonded pair. This achievement presents a critical step towards
heterointegration of high quality GaN thin layers, which will have
a major impact in the various GaN based device technologies.
Although back-side implantation is found to be a very effective
strategy to achieve
direct bonding, one may think that it will present an additional
costly process step, which may hardly justify the need for
ion-cutting to achieve cost-effective substrates at the industrial
level. We are developing new approaches to overcome this potential
limit (16).
Figure 8. Optical (left) and IR (right) micrographs of
H-implanted fs-GaN bonded to
2-inch sapphire wafer. The few interface bubbles observed can be
caused by several factors including: particles on the bonding
surfaces, localized surface protrusions, and trapped air
pockets.
2-inch
ECS Transactions, 16 (8) 251-262 (2008)
258Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
Mechanistic picture of H ion-induced fs-GaN splitting In this
section, we address the underlying physics of ion-induced thin
layer
exfoliation. The ion-cut process has been successfully applied
to various semiconductor materials such as Ge, InP, and GaAs.
Independently of the material, the interaction of the implanted
species with the radiation damage seems to play the key role in the
splitting process. Apart from Si which was intensively investigated
(17), only few studies were devoted to investigate the atomic
processes involved in the splitting of other semiconductors. GaN
exfoliation induced by H implantation was first reported by
Kucheyev et al. (18). However, detailed studies on the mechanisms
of ion slicing of GaN are still missing. In addition to XTEM, ERD,
and XRD, we used Rutherford backscattering spectrometry in
channeling mode (RBS/C) and positron annihilation spectroscopy
(PAS) to investigate H-defect interactions leading to the splitting
process (13).
0 5 10 15 20 25 30
0.50
0.51
0.52
0.53
0.54
0 87 261 480 765 1127 1484
virgin
Depth [nm]
S-p
aram
eter
Positron energy [keV]
as-implanted
0 100 200 300 400 5000
5
10
15
20
25
30
0.0
0.2
0.4
0.6
0.8
1.0Hydrogen
Ato
mic
Dis
plac
emen
ts [%
]
Depth [nm]
displacement field
H C
once
ntra
tion
[x10
22 H
/cm
3 ]
Figure 9. (left) S parameter depth profile measured before
(triangles) and after
(squares) H implantation. (right) H atom depth profile (circles)
and implantation damage profile (red line) as deduced from ERD and
ion channeling, respectively.
Several atomic processes take place during the implantation of
energetic H ions
generating defects from both sublattices. Fig. 4(a) shows a
broad damage band extending over a 300 nm-thick layer starting
about 200 nm below the surface. No extended defects are observed at
the implanted fluence. High magnification images of the implanted
zone taken under focus (not shown) indicate the presence of a high
density of nanoscopic bright spots of ~1-2 nm diameter. The change
in their contrast during focus variation suggests that they are
void-like structures. We name these nanoscopic voids nanobubbles.
Fig. 9 (left) displays Doppler broadening S parameter depth
profiles measured before and after implantation. In PAS, the line
shape of the γ ray depends on the electron momentum distribution.
In the regions distant from atomic nuclei, such as vacancies or
voids, the line has a distinctly smaller high momentum tail. The S
parameter measures the “sharpness” of the line, so it takes higher
values in vacancies and cavities. In the implanted region an
enhancement of the S parameter of 6.5 % is observed indicating the
presence of open volume defects which support XTEM observations.
Such an increase is too large to be caused by monovacancies;
obviously small vacancy clusters are detected. Figure 9 (right)
displays H and atomic displacement depth profiles in as-implanted
GaN as deduced from ERD and ion channeling analyses, respectively.
The data show that H distribution is peaked at ~320 nm in agreement
with SRIM calculations (19). H concentration at the peak is found
to be ~1.2 × 1022 H/cm3 corresponding to an atomic concentration of
about
ECS Transactions, 16 (8) 251-262 (2008)
259Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
~13% which is ~3 times higher than the concentration needed for
Si exfoliation (17). The atomic displacement field reaches a
maximum of ~27% at a shallower depth around 285 nm. We estimate the
displacements per ion to be ~2.5, much smaller than ~10 calculated
at a lattice temperature of 0 K (i.e., without dynamic annealing)
(19). This suggests that about 75% of Frenkel pairs recombine
during the implantation process. This annihilation rate is
relatively smaller than in the case of Si implanted under ion-cut
conditions (17). Therefore, the dynamic annealing cannot explain
the unusually high fluence required for the splitting of GaN. The
nature of H-defect complexes and their thermal evolution may play
the most critical role than the absolute amount of the surviving
defects.
Figure 10. (left) RBS/C yields as a function of annealing
temperature for GaN
substrates implanted with H at 2.6 × 1017 atom/cm2. (right) XTEM
micrographs of H-implanted GaN annealed at different temperatures:
450 oC, 500 oC, and 600 oC.
The thermoevolution of H-induced damage is investigated next. In
figure 10, we
summarize RBS/C and XTEM data obtained after annealing at
different temperatures. We note that RBS/C yield slightly decreases
after heating at 300 oC. This can be attributed to the return of
some interstitial atoms to substitutional sites. RBS/C spectra
recorded after annealing at higher temperatures show unexpected
features. The first is a strong broadening of damage-related peak
observed only at the left side for backscattered energies below
1.28 MeV. This asymmetric broadening appears to be related to H
since it occurs around the region where H reaches its peak
concentration [Fig. 9 (right)]. The second feature in the spectra
is the increase of dechanneling beyond the implanted zone. Finally,
annealing above 300 oC causes also a strong enhancement of
dechanneling near the surface. Note that the increase of the peak
intensity comes simply from the increase of dechanneling
background. Interestingly, the dechanneling level decreases above
450 oC. In XTEM data, we note that annealing up to 450 oC does not
trigger any significant morphological changes in the damage band.
Similarly to the as-implanted sample, the implanted zone remains
decorated with nanobubbles. A small increase in temperature above
450 oC leads to the formation of nanoscopic cracks or platelets
parallel to the surface. Further increase in the annealing
temperature induces large cracks leading to a complete exfoliation
of a ~340 nm-thick layer. Our detailed XTEM data show that
structural transitions from nanobubbles to platelets and from
platelets to microcracks occur within temperature windows as narrow
as 25 and 50 oC, respectively. 450 oC is
50 nm
450 oC
50 nm
500 oC
100 nm
600 oC
0.50 0.75 1.00 1.25 1.50 1.75
500 oC
600 oC
450 oC
300 oC
Random
Nor
mal
ized
Bac
ksca
ttere
d Y
ield
Backscattered Energy [MeV]
Virgin
as-implantedplatelets
microcrack
ECS Transactions, 16 (8) 251-262 (2008)
260Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
identified as the critical temperature at which transformation
processes commence. Interestingly, at this temperature the
dechanneling enhancement in RBS/C yield attains its maximum.
A possible origin of lattice distortion would be trapping of H2
molecules in
nanobubbles leading to the buildup of internal pressure.
However, highly pressurized nanobubbles alone cannot explain the
observed dechanneling since their stress field can hardly reach the
surface. Since damage layer contains also other defect complexes
such as self-interstitial clusters, one can suppose that their
combined influence with hydrogen-induced internal pressure can
increase the in-plane compressive strain causing a strong lattice
distortion. This process will ultimately lead to a weakening of the
atomic bonding. The system attains the criticality around 450 oC.
The dechanneling decreases above this temperature suggestive of a
partial relief of the internal strain following the formation of
platelets parallel to the surface. These platelets define the
fracture paths for the exfoliation. Interestingly, Doppler
broadening measurements for samples annealed at T ≤ 450 oC were
found to be identical to those of the as-implanted state. The
absence of vacancy clustering in this temperature range indicates
that the necessary voids for splitting assemble dynamically during
the implantation process. This behavior differs completely from the
evolution observed in Si where an important increase of void-like
defects was found to precede the exfoliation (20). This remarkable
difference in vacancylike defects thermal behavior observed between
GaN and Si suggests that a general and predictive microscopic model
of ion-cut process has to consider the intrinsic properties of the
material, the nature of H-defect complexes, and point defects
diffusivities. These aspects are still poorly understood for GaN.
Additional systematic experimental studies and calculations would
be highly valuable.
Conclusion
In summary, epitaxy-compatible InP-on-Si substrates were
successfully fabricated by
thin layer transfer from 4-inch InP donor wafer. We also
demonstrated that H-implanted 2-inch fs-GaN wafers can successfully
be bonded to sapphire by the manipulation of wafer bow using back
side implantation. This makes the heterointegration of bulk quality
GaN layers possible by using wafer bonding and ion-slicing. By
using a variety of experimental techniques, we investigated the
critical structural transformations involved in splitting of GaN by
H. We found that vacancy clustering during the implantation process
leads to the assembly of 1-2 nm nanobubbles. Lattice distortion was
observed around the implanted zone in the temperature range 300–450
oC. The distortions result from the internal pressure buildup
inducing a strain field around damage region. Strain relaxes by way
of Ga-N bonds breaking leading to the nucleation of platelet
embryos of extended internal surfaces.
Acknowledgments The authors acknowledge contributions from F.
Süßkraut and R. Krause-Rehberg
(Univ. of Halle-Wittenberg), M. Chicoine (Univ. of Montreal), O.
Seitz and Y. J. Chabal (Univ. of Texas at Dallas). InP wafers were
supplied by Freiberger Compound Materials GmbH. This work has been
partly funded by German Federal Ministry of Education and Research
(CrysGaN Project).
ECS Transactions, 16 (8) 251-262 (2008)
261Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp
-
References
1. M. Bruel, Eletron. Lett. 31, 1201 (1995) 2. Q.-Y. Tong and U.
Gösele, Semiconductor Wafer Bonding: Science and
Technology, The Electrochemical Society Series
(Wiley-Interscience Publication, 1999).
3. A. W. Fang, H. Park, Y. –H. Kuo, R. Jones, O. Cohen, D.
Liang, O. Raday, M. N. Sysak, M. J. Paniccia, and J. E. Bowers,
Mat. Today 10, 28 (2007).
4. J. M. Zahler, K. Tanabe, C. ladous, T. Pinnington, F. D.
Newman, H. A. Atwater, Appl. Phys. Lett. 91, 012108 (2007).
5. Q. –Y. Tong, Y. –L. Chao, L. -J. Huang, and U. Gösele,
Electron. Lett. 35, 341 (1999).
6. E. Jalaguier, B. Aspar, S. Pocas, J. F. Michaud, A. M. Papon,
and M. Bruel, Proc. 11th Int. Conf. On InP and related materials,
p26 1998 (Piscataway, NJ : IEEE).
7. S. Hayashi, D. Bruno, and M. S. Goorsky, Appl. Phys. Lett.
85, 236 (2004). 8. R. Singh, I. Radu, R. Scholz, C. Himcinschi, U.
Gösele, and S. H. Christiansen,
Semicond. Sci. Technol. 21, 1311 (2006). 9. M. Alexe, V. Dragoi,
M. Reiche, and U. Gösele, Electron. Lett. 36, 677 (2000). 10. S.
Nakamura and G. Fasol, The blue Laser Diode: GaN Based Light
Emitters and
Lasers (Springer, Berlin, 1997). 11. O. Moutanabbir, R. Scholz,
S. Christiansen, U. Gösele, M. Chicoine, R. Krause-
Rehberg, and Y.J. Chabal, MRS Spring Meeting: Symposium C, March
24-28, 2008.
12. R. Singh, I. Radu, G. Bruederl, C. Eichler, V. Haerle, U.
Gösele, and S. H. Christiansen, Semicond. Sci. Technol. 22, 418
(2007).
13. O. Moutanabbir, R. Scholz, S. Senz, U. Gösele, M. Chicoine,
F. Schiettekatte, F. Süßkraut, and R. Krause-Rehberg, Appl. Phys.
Lett. 93, 031916 (2008).
14. Y. Suprun-Belevich, F. Cristiano, A. Nejin, P.L.F. Hemment,
and B. Sealy, Nucl. Instrum. Methods Phys. Res. B 140, 91
(1998).
15. J. Chen and I.D. Wolf, Semicond. Sci. Technol. 18, 261
(2003). 16. O. Moutanabbir and U. Gösele, to be published. 17. For
a recent review, see: B. Terreault, phys. stat. sol. (a) 204, 2129
(2007). 18. S.O. Kucheyev, J.S. Williams, C. Jagadish. J. Zou, and
G. Li, J. Appl. Phys. 91,
3928 (2001). 19. J.F. Ziegler, J.B. Biersack, and U. Littmark,
The Stopping and Range of Ions in
Solids (Pergamon, New York, 1985): www.srim.org. 20. O.
Moutanabbir, B. Terreault, M. Chicoine, F. Schiettekatte, and P.J.
Simpson,
Phys. Rev. B 75, 075201 (2007).
ECS Transactions, 16 (8) 251-262 (2008)
262Downloaded 27 Oct 2009 to 192.108.69.177. Redistribution
subject to ECS license or copyright; see
http://www.ecsdl.org/terms_use.jsp