Strain relaxation of metastable SiGe/Si: Investigation with twocomplementary X-ray techniquesE. Kasper, N. Burle, S. Escoubas, J. Werner, M. Oehme et al. Citation: J. Appl. Phys. 111, 063507 (2012); doi: 10.1063/1.3694037 View online: http://dx.doi.org/10.1063/1.3694037 View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v111/i6 Published by the American Institute of Physics. Related ArticlesReflectance anisotropies of compressively strained Si grown on vicinal Si1−xCx(001) Appl. Phys. Lett. 102, 011902 (2013) Dislocation structure in AlN films induced by in situ transmission electron microscope nanoindentation J. Appl. Phys. 112, 093526 (2012) Free volume change of elongated polyethylene films studied using a positron probe microanalyzer Appl. Phys. Lett. 101, 203108 (2012) Strain and defects in Si-doped (Al)GaN epitaxial layers J. Appl. Phys. 112, 093102 (2012) On-chip stress relaxation testing method for freestanding thin film materials Rev. Sci. Instrum. 83, 105004 (2012) Additional information on J. Appl. Phys.Journal Homepage: http://jap.aip.org/ Journal Information: http://jap.aip.org/about/about_the_journal Top downloads: http://jap.aip.org/features/most_downloaded Information for Authors: http://jap.aip.org/authors
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
Strain relaxation of metastable SiGe/Si: Investigation with twocomplementary X-ray techniques
E. Kasper,1,a) N. Burle,2 S. Escoubas,2 J. Werner,1 M. Oehme,1 and K. Lyutovich1
1Institut fur Halbleitertechnik, Universitat Stuttgart, Pfaffenwaldring 47, 70569 Stuttgart, Germany2IM2NP – CNRS 6242, Aix Marseille University, Campus St Jerome, 13397 Marseille cedex 20, France
(Received 9 December 2011; accepted 18 February 2012; published online 20 March 2012)
Metastable and strain relaxed SiGe layers with about 20% Ge content have been grown by
molecular beam epitaxy on Si substrates at 550 �C. The thickness regime of metastability and the
onset of strain relaxation were investigated on dust particle free surfaces obtained by careful
chemical cleaning and epitaxy loading under clean room conditions. Compared to earlier results
true metastable regime without misfit dislocations was obtained up to 140 nm thickness. The onset
of strain relaxation started with heterogeneous nucleation sites of misfit dislocations. X-ray
topography proved to be a unique monitoring tool to observe a low density of single dislocations.
From these results we suggested to define a critical thickness band with lower bound tcl from
dislocation nucleation to an upper bound tco (600 nm in our case) defined by the onset of
considerable strain relaxation. The strain relief was measured by X-ray diffraction (reciprocal
space mapping) and found to be very abrupt (76% strain relaxation at 800 nm thickness). VC 2012American Institute of Physics. [http://dx.doi.org/10.1063/1.3694037]
I. INTRODUCTION
SiGe (silicon germanium) is lattice mismatched against
the Si substrate commonly used in microelectronics. The lat-
tice mismatch f is nearly linear with Ge content x (Vegard’s1
law: f ¼ 0.0417x). Exactly, a slight parabolic bow (Dismukes
law) is proven for both bulks SiGe (Ref. 2) and epitaxial
SiGe.3
f ¼ 0:036678xþ 0:005032x2: (1)
With x ¼ 0.2, the Vegard’s law gives aSiGe¼ 0.54765 nm and
f ¼ 0.835%, as Dismukes law leads to aSiGe¼ 0.54720 nm and
f ¼ 0.754%.
The lattice mismatch causes strained layer SiGe growth
(pseudomorphic SiGe) up to a critical thickness tc, Thicker
layers start to relax the strain by the introduction of misfit
dislocation networks at the interface. By kinetic reasons
(nucleation of dislocations at surface sites) the interface dis-
location network is connected to the surface by threading
dislocations. These threading dislocations that penetrate
devices in the SiGe layer prohibited a broad use of relaxed
SiGe layers in microelectronics whereas pseudomorphic
SiGe is now in commercial use in all high-end silicon cir-
cuits except memories. SiGe heterobipolar transistor (HBT)
holds speed records (transit frequency of 500 GHz (Refs. 4
and 5)), and it is used in microwave circuits.6 Since the 90
nm node logic complementary metal oxide silicon (CMOS)
circuits use local SiGe source/drain to boost p-channel per-
formance7 by strain.
We study SiGe layers with about 20% Ge content
because at higher mismatch values a third misfit accommo-
dation mechanism set in: Strong corrugation of the surface
or even islanding. The critical thickness tc under thermal
equilibrium is well understood both from an energetic point
of view (Van der Merwe approach8 and from force consider-
ations Blakeslee-Matthews approach9). Many differing equi-
librium curves have their origin in assumptions about
geometry and Burgers vector of the network, about core
energies and outer cutoff radii.
A typical tc value for 0.8% mismatch of SiGe layers is
about 10 nm. Rather high growth temperatures (900 �C) are
necessary to reach the equilibrium.10 At lower temperatures
(750 �C�550 �C) the onset of strain relaxation is delayed to
higher tc value that was impressively shown by a investiga-
tion of People/Bean.11 The pseudomorphic thickness range
between equilibrium curve and the 550 �C curve is often
named metastable regime12 although models of sluggish dis-
location movement favor an explanation as kinetically con-
trolled plastic deformation of the semiconductor on the
threshold from a deformable material at high temperatures to
a brittle material at room temperature. Even the exact value
of the critical thickness seems to depend on the accuracy
of the methods to measure strain and dislocation density.
The used methods include X-ray diffraction, Raman-
spectroscopy, defect etching, luminescence, and transmission
electron microscopy (TEM).
Characteristic for the relaxation process is a start with
heterogeneous nucleation centers on the surface. In this ini-
tial phase the strain relaxation is rather small and hard to
quantify even with X-ray diffraction (XRD). Needed are
methods to observe and analyze low density dislocation net-
works, later the networks are dense and the strain relaxation
is best followed with diffraction. In order to follow all stages
of the process from pseudomorphic to relaxed growth, we
used two complementary X-ray methods: X-ray topogra-
phy(XRT) for strained and low dislocated layers and X-ray
diffraction reciprocal space mapping (RSM) for middle and
high dislocated partially and fully relaxed layer.a)Electronic mail: [email protected].
0021-8979/2012/111(6)/063507/10/$30.00 VC 2012 American Institute of Physics111, 063507-1
JOURNAL OF APPLIED PHYSICS 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
The classical work on strain relaxation was done with
laboratory experiments under not very well defined clean
room conditions. Submicron sized dust particles may stick to
the substrate surface and being overgrown by the epitaxial
layer they may act as nucleation centers for defects. We
processed the wafer in a clean room facility (Institute of
Semiconductor Engineering) using a molecular beam epitaxy
(MBE) growth equipment with surface down mounted sub-
strates. This procedure reduces the particle density. We
expect a strong influence on the initial phase of strain relaxa-
tion that we observe without limits given by a single charac-
terization method.
The onset of strain relaxation in the metastable regime
is retarded by dislocation nucleation barriers and by slow
motion of misfit dislocations. It is generally believed that ho-
mogeneous nucleation is energetically unfavorable13 for lat-
tice mismatch values below 2%. Heterogeneous nucleation
of dislocations on surface sites depends on occurrence of sur-
face defects.
II. EXPERIMENTAL
SiGe with lattice mismatch below 1% was grown epitax-
ially on dislocation-free Si substrates. We tried to suppress
the density of surface nucleation sites by careful pretreat-
ment before and during epitaxy. The first stages of misfit dis-
location generation were observed with X-ray topography,
which is the only technique allowing individual subsurface
dislocations to be detected. On the contrary, XRT is limited
to low densities (about 103 cm/cm2). Remarkable strain
relaxation sets in after generation of a high density of dislo-
cations; it was monitored by XRD. This technique allows
also layer thicknesses and Ge contents to be verified.
A. Molecular beam epitaxy (MBE) of SiGe layers on Si
SiGe layers with about 20% atomic percent (x) Ge con-
tent were grown in an MBE equipment14 the loading station
of which was connected to the clean room of the IHT, Stutt-
gart University. The nominal clean room class (class 100)
above the chemical bench and the loading station were
improved to be better than class 10 by a low throughput
cycle. The 100 mm diameter Si substrates (Czochalsky
growth, p-doped, 1–4 X�cm specific resistivity) were
removed immediately before the epitaxy process from the
sealed package, chemically precleaned (RCA etch) to get an
around 1 nm protective surface-oxide14 and loaded into the
wafer magazine of the MBE load lock, and after pump down
under ultrahigh vacuum (UHV) conditions transferred into
the MBE growth chamber. Here, as a second part of the Si
surface cleaning scheme an in situ anneal (900 �C, 5 min)
was performed in UHV vacuum that resulted in thermal dis-
sociation (SiO2Xþ Si ! 2SiO) and removal of the protec-
tive oxide. Epitaxy starts with a 100 nm Si buffer layer
(growth temperature: 600 �C) to cover the small residual
contaminations (<1/100 monolayer C, O, B). After the
buffer layer, the temperature was reduced to 550 �C and the
SiGe layer was grown from Si beam (source: e-gun evapora-
tor) and Ge beam (source: diffusion cell with graphite cruci-
ble). The beam intensities (nominal Ge/Si intensity of 0.2)
were controlled using a mass spectrometer with cross beam
ion source (Balzers). Growth rate was 0.1 nm/ s.
Thickness of the SiGe layers ranged from 50 nm to
800 nm. The thicknesses were monitored in situ with inter-
ferometric reflectometry15 and ex situ by surface stylus mea-
surement of the epitaxy edge. The sample identification
(A50 to A800) describes the process equipment (MBE A
with e-gun and effusion cell sources for Si and Ge, respec-
tively) and the thickness (in nm) of the SiGe layer. The
selected thickness range covers metastable strained layers
and partly relaxed layers.
B. X-ray topography
We used XRT transmission setting with Rigaku rotating
anode X-ray generator and Silver anode (kka¼ 0.057 nm).
Best resolution images are proceeded on silver photographic
emulsion on glass plates.
X-ray Topography (by transmission or reflection) is an
imaging technique based on Bragg’s law
2dhkl � sinh ¼ k (2)
in quite perfect monocrystals, which can be used in synchro-
tron apparatus or with laboratories X-ray sources. (In labora-
tories transmission settings conditions, sample are usually
about one square centimeter and some hundred micron
thick.) Crystal defects like dislocations or precipitates give a
contrast different from the perfect crystal area as far as they
introduce enough deformation (Fig. 1) of the lattice.16,17 On
the other hand, if the deformation is too high the Bragg posi-
tion varies a lot from one point to another so it is no longer
possible to image an entire sample. This technique has been
used for a long time in various materials like metals or
ceramics but it is perfectly suited also for semiconductors
like Si or Ge.18
In the case of epitaxial systems, the layer and the sub-
strate can be imaged either in the same time, or separately,
depending on conditions. Using the diffracting planes
perpendicular to the growth surface implies that layer and
FIG. 1. (Color online) Imaging XRT conditions from diffracting planes in
epitaxial layer (light-colored) and substrate (dark).
063507-2 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
substrate planes are in continuity with the same reticular dis-
tance, so they diffract with the same Bragg angle and are
imaged together. This is true as far as the layer relaxation is
low enough to resolve each dislocation as individual and
“far” from the others. (Fig. 1) The width resolution on the
imaging plate is about one micron, so the smallest defect that
can be observed must have a deformation field extended
over several microns and the nearest next defect-field should
be farther than some tens microns.
C. X-ray diffraction
High resolution X-ray diffraction is proceeded on a Pan-
Analytical X’Pert 4 circles goniometer. Sample positioning
is in reflection and the anode gets copper target source (kka¼1.541 A). In high resolution setup a monochromator is neces-
sary to select the Ka1 radiation and to get a parallel incident
beam, followed by an analyzer to achieve a better angular
resolution.19,20 For that purpose, we use a two bounces Ge
(220) monochromator associated with a parabolic variable-
step-multilayer mirror between the copper tube and the
monochromator. A three bounces Ge (220) analyzer is
mounted in front of the point detector. In order to extract the
relaxation rate and/or the Ge content in the SiGe layers, two
kinds of measurements can be performed: x�2h scans and
reciprocal space mappings (RSM).
The x�2h scans correspond to a rotation of the sample
with an x angle simultaneously with a rotation of the detec-
tor with a 2h angle. In a first step, x�2h scans are performed
on 004 planes, such symmetrical configuration corresponds
to a scan along qz vertical direction in the reciprocal space;
the position of the SiGe layer Bragg peak position with
respect to the Si substrate one allows to determine the per-
pendicular parameter a?SiGe with a precision of 0.002 A. If the
relaxation rate is known (for example, in the case of fully
strained samples as observed by XRT in Fig. 1) the germa-
nium content can be easily deduced as far as the Poisson
coefficient is determined. The precision on the composition
is limited by the acceptance of the Vegard’s law on Poisson
coefficient more than on the Bragg angle itself.
In order to extract directly axial strain values along the
two directions (in-plane and out-of-plane), asymmetric dif-
fraction21 or reciprocal space maps22 (RSM) are necessary
(see the 224 reflexion in Fig. 2). In the asymmetrical geome-
try, the diffusion vector ~q ¼ ~k � ~k0 is not perpendicular to
the sample surface, thus both vertical and horizontal compo-
nents of the scattering vector qz and qx are available:
qx ¼ qy ¼4pk
sinh sinðh� xÞ (3)
with h the Bragg angle and x the incident angle with respect
to the crystal surface.
We have measured two reflections for a better accuracy:
both 224 planes and 115 planes.
III. DISLOCATION NUCLEATION
The lattice mismatch f between SiGe and the substrate
Si causes a compressive strain e, which for a pseudomorphic
planar film is given by the same amount as the mismatch but
opposite sign: pseudomorphic film
e ¼ �f : (4)
The opposite sign stems from the usual definitions of lattice
mismatch (positive for the larger film constant) and strain
(negative for compressive strain).
The biaxial film strain is correlated with an in plane
stress r whose magnitude depends on the elastic properties.
r ¼ E
1� ve; (5)
where E is the elastic modulus (defined in uniaxial deforma-
tion), � is the Poisson coefficient. For Si � value is 0.278; E
value is 131GPa in h100i direction; for Ge E[100] is equal to
107GPa and � to 0.27.
In biaxial deformation, the shear modulus G is generally
used, with the relation
2Gð1þ vÞ ¼ E: (6)
The shear modulus G is equal to 51GPa for Si and 40GPa in
Ge. The elastic moduli of SiGe are slightly smaller if one
assumes a linear interpolation between Si and Ge. For SiGe
(x ¼ 0.2) we expect a reduction in stiffness by about 5%.
The relevant stress for misfit dislocation glide is the resolved
excess stress rex. This resolved stress is depending on the
geometric characteristics of the glide plane and the Burgers
vector by respect with the main strain direction; it has been
given by Dodson and Tsao23 as
rex ¼ 2Gjej 1þ v
1� vcos k cos u� Gb� cosulnðat=bÞ
4ptð1� vÞ : (7)
The first term is caused by the misfit strain e, the second
term is related to the self energy of dislocations.
FIG. 2. (Color online) Geometry of a 224 reciprocal space map. In the
asymmetrical configuration the scattering vector q (~q ¼ ~k � ~k0 ) has two
components qz and qx.
063507-3 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
In the diamond cubic lattice structures of Si, SiGe, and
Ge the dislocations propagate primarily by glide on (111)
planes inclined to the (001) interface. The geometry of the
interfacial dislocation array is therefore defined by the inter-
section of the inclined glide planes with the interface. The
misfit dislocation lines run along orthogonal [110] directions.
The dislocation properties are defined by b, a, k, u, the
film thickness is t. The Burger’s vector (amount b) of a disloca-
tion in the diamond lattice points in a [1/2]h110i direction
(length b is 0.384 nm in Si), a is a factor that describes the dislo-
cation core energy. Usually selected values of a ranges from 1
to 3. The exact choice is not very critical as long as the thickness
t is much larger than the Burger’s vector length b. Then the
strain energy around the dislocation dominates the total disloca-
tion energy. The angles k and u are defined by Burger’s vector
to dislocation normal in the interface and by glide plane normal
to interface normal, respectively. The values of cos k and cos uare 0.5 and 0.58, respectively, for the 60� dislocation (60� angle
between line and Burger’s vector) on (001) interfaces.
In an unrelaxed layer, the strain and therefore the stress
are constant with the layer regardless of thickness. The elas-
tic energy stored in the layer increases with thickness. The
relaxation begins when the stored energy is high enough to
overlay the formation of dislocation line. The extension of
misfit dislocation is only allowed above the equilibrium criti-
cal thickness tc where the excess stress is above zero. The
relative contribution of the self energy (second term in Eq.
(7)) shrinks rapidly with thickness t, then we can assume a
resolved excess shear stress rex ¼ 430 MPa for pseudomor-
phic or weakly relaxed layers.
The pseudomorphic growth and the early stages of dislo-
cation nucleation under the influence of the excess stress
were observed by XRT. The layers with thicknesses (A50,
A100) up to 100 nm proved to be dislocation free. Some-
times, the substrate showed reflections from precipitates
deep in the substrate. These precipitates are probably SiO2
complexes created in oxygen rich regions of the Czochralski
grown silicon substrate during the preepitaxial heat treatment
(900 �C). The XRT image (Fig. 3) is shown here to demon-
strate that substrate precipitates are not connected with heter-
ogeneous dislocation nucleation.
First misfit dislocations are observed with A200 (Fig. 4)
These misfit dislocations belong to heterogeneous nucleation
centers (density below 100/cm2), which generate one or sev-
eral dislocation half loops with a misfit dislocation segment
in the interface and two threading dislocation arms connect-
ing the interface with the surface.
The threading arms in the thin layer (200 nm) cannot be
detected directly with XRT but according to the dislocation
theory, requests that single dislocation cannot end inside the
volume. Stronger contrast from some nucleation centers indi-
cates multiple generations of dislocations from the same source.
In the 400 nm thick sample (A400) multiple generation
is seen frequently (Fig. 5) and the length of the dislocation
segments (several mm) is now high enough to obtain cross-
ings between dislocations of different sources. Such cross-
ings can act as second generation nucleation sites (Frank-
Read source12 as a general multiplication mechanism and
Hagen-Strunk source24 as specific source for strained layers).
We will see later that even the misfit dislocation network in
A400 does not relax the strain considerably. A mean dislocation
spacing is hardly to define in this heterogeneous nucleation
phase but from large area observation we obtain 600 lm and
about 150 lm for 200 nm and 400 nm thickness t, respectively.
FIG. 3. XRT image of a section of sample A100 showing precipitate (prob-
ably SiO2) contrast within the substrate. No dislocations from the 100 nm
thick SiGe layer are visible.
FIG. 4. XRT image of a section of sample A200. Typical straight misfit dis-
location contrast in both h110i directions is visible. The h110i directions are
the intersections of the (001) interface with the (111) glide planes.
FIG. 5. XRT image of a section of sample A400. Multiple dislocation gen-
eration and overlap of dislocations from different sources happens
frequently.
063507-4 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
Larger relaxation is evidenced in the thickest sample
A800. Here, the density of misfit dislocations (Fig. 6) is too
high to be measured by XRT. The “wavy” or “grainy” struc-
ture must not be correlated with the dislocation shape: The
stresses induced by interactions between dislocations are
overlapping all other contrasts and no information can be
extracted about individual dislocations.
The lengths of dislocations from individual sources
(Figs. 4 and 5) let us assess the dislocation velocity under
the given temperature (550 �C) and the given excess
stress (400 MPa), assuming that the dislocations have
been gliding above a thickness of 150 nm. Misfit disloca-
tion length is between 1 and 5 mm for 500 s to 2500 s
gliding time.
A mean dislocation velocity of 2 lm/s is extracted. That
is higher than data from Hull25 and Houghton26 but fits with
an extrapolation of Yamashita data27 extracted from lower
shear stresses. The reason for our higher velocity data could
be the low nucleation density that allows the dislocation arm
to move without interaction with other dislocations.
Strain measurements are performed with XRD x�2hscans and reciprocal space maps (RSM) around Si Bragg
peaks. In Fig. 7 are presented the symmetrical (004) x�2hscans for the fully strained SiGe layers (A100, A200 and
A400). The SiGe peak angular position allows determining
the perpendicular lattice parameter and thus the Ge content,
assuming no relaxation. Values are slightly larger (22.2%,
see Table I) than the nominal 20% one. If the layers are per-
fect with planar interfaces, interference fringes can be
observed, which spacing (in reciprocal space units) is inver-
sely proportional to the SiGe layer thickness t ¼ 2p/Dqz.
This is the case for samples A100, A200, and A400 (Fig. 7);
the calculated thickness values confirm the nominal thick-
nesses within a few percent (Table I).
Figure 8 represents the (224) RSM of sample A200,
with the axis in hkl units. The relation between reciprocal
space vector coordinates and hkl units are the following:
qx ¼2ph
aSi; qy ¼
2pk
aSi; qz ¼
2pl
aSi; (8)
with aSi the silicon lattice parameter (in nm).
On the asymmetrical RSM, the Si and SiGe Bragg peaks
are aligned along the vertical direction that confirms that the
relaxation is quite zero on these samples, as previously
observed by XRT. The vertical position of the SiGe peak,
with respect to the Si one allows to determine the Ge content
in the layer and is in agreement with the one deduced from
the (004) x�2h scan (see Table I).
IV. METASTABLE CRITICAL THICKNESS AND STRAINRELAXATION
A. Brief report on metastability and relaxation rate
Above the critical thickness tc the strain e in the partially
relaxed layer is reduced by the introduction of a misfit dislo-
cation network. This network is best characterized by a mean
distance p between the misfit dislocations lying in the
FIG. 6. (Color online) XRT image of a section of sample A800. The high
density dislocation network cannot be resolved with XRT. But the grainy
structure indicates variable strain levels probably caused by the inhomoge-
neous dislocation nucleation.
FIG. 7. (Color online) (004) x� 2hscans of samples A100, A200, A400,
reported in qz units.
063507-5 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
interface. (Note: The usual dislocation density notation based
on dislocation lengths divided by volume is not appropriate
for an interface network.) On (001) Si the misfit dislocation
lines create an orthogonal network along h110i directions
that are the intersections of the f111g glide planes with the
(001) substrate surface (Fig. 9). The Burger’s vector of the
dislocation is [1/2]h110i, the one lying 60� to the dislocation
direction is preferred. (Note: The Burger’s vector parallel to
the dislocation line gives a screw dislocation that is ineffec-
tive in misfit strain relaxation, it would cause a twist.) For
strain relaxation the edge component of the Burger’s vector
projected to the normal of the dislocation is the active
component
b0 ¼ b� cosk: (9)
Each dislocation displaces the film lattice planes by b’. The
(1/p) dislocations per unit length reduce the compressive
strain (-e) to
ð�eÞ ¼ f � b0
p: (10)
Total strain relaxation (e ¼ 0) is obtained by a relaxation
spacing prel
prel ¼b0
f: (11)
For our experimental set of f ¼ 0.008 we obtain a relaxation
spacing prel ¼ 24 nm when we assume b0 ¼ b/2 ¼ 0.192 nm.
At higher strain values dislocation reactions28 may occur
that deliver sessile Lomer dislocations for which b0 ¼ b.
John Bean’s group discovered that at moderate growth
temperatures (550 �C) the measured critical thickness for
onset of misfit dislocation generation is much higher than the
equilibrium value. Most cited is a fit through the experimen-
tal points given by People/Bean.11 Let us assign the metasta-
ble critical thickness tcp to this fitting curve.
The value of tcp29 for the given lattice mismatch is tcp ¼
150 nm, compared to the equilibrium value of tc ¼ 10 nm.
tcp
b� f 2 ¼ 1
200ln
tcp
b
� �: (12)
It is an ongoing debate (for a review, see Ref. 12) if the met-
astable critical thickness is due to a retarded relaxation pro-
cess where the given value tcp only marks a detectable
dislocation spacing or if it is a real metastable process
defined by a nucleation barrier for dislocation generation.
The degree of strain relaxation r is now defined as
r ¼ f þ ef¼ prel
p: (13)
The degree of relaxation can be extracted from measurement
of strain e or measurement of dislocation spacing p following
Eq. (10).
B. Experimental measurements of relaxation
In the case of 20% Ge layers on Si, the misfit would be
fully relaxed (assuming 60� misfit dislocations) by a squared
dislocation network with an average distance p about 25 nm
between two neighboring dislocations, which is 4�105
TABLE I. Ge contents x (from (004) XRD, (224) and (115) RSM), thick-
nesses tSiGe from XRD fringes (except A800, which gives the nominal value
as fringes are only visible in unrelaxed layers) and relaxation r (from XRT
and RSM).
Sample x Ge % tSiGe (nm) r %
A100 23.0 9362 0 (XRT)
A200 22.9 19562 0.004 (XRT)
A400 21.2 405612 0.016 (XRT)
A800 21.6 800(nom) 7464 (RSM)
FIG. 8. (Color online) 224 RSM of sample A200. The axis are in hkl units.
Si Bragg peak is located at h ¼ k ¼ 2 and l ¼ 4, while the SiGe one appears
at lower l value (strain).
FIG. 9. Geometry of a parallel set of misfit dislocations along [ �110 ]. The
orthogonal dislocation set is not shown to avoid overload of the drawing.
The dislocation spacing is given by the distance p. The Burger’s vector b,
and its projection length b0 on the interface normal to the dislocations are
shown for the (111) glide plane
063507-6 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
dislocation lines/cm in cross section or 8�105 dislocation line
cm/cm2 in plane view. So the highest density that can be
clearly imaged by XRT corresponds to a relaxation rate
below 1%.
The relaxation rate cannot be measured by RSM when it
is lower than a few percent. It can be only estimated from
the observed dislocation densities in XRT images; we found
0% in A100 (Fig. 3), 0.004% (30 cm/cm2) in A200 (Fig. 4),
and 0.016% (130 cm/cm2) in A400 (Fig. 5) assuming b0¼0.192 nm.
On the contrary, the thicker sample A800 exhibits dislo-
cation density higher than some 104 cm/cm2 so we can only
say from XRT that r is at least 10% (Fig. 6). In that case, the
relaxation rate can be precisely measured by RSM (Fig. 10).
The RSM were recorded around the 224 reflection. As
seen previously, an identical in-plane reciprocal vector for
the Si substrate and SiGe film indicates a pseudomorphic
structure (Fig. 8, A200). As expected from the XRT results,
we cannot detect any relaxation up to A400 because disloca-
tion spacings above 25 lm result in less than 0.1% relaxa-
tion. In a relaxed structure the Bragg peak would shift to
smaller reciprocal space value for a tensile strain and to
higher value for a compressive strain. For the sample A800
(Fig. 10), both qx and qz SiGe positions are lower than the Si
Bragg peak, which means that the layer started to relax. We
measure a relaxation rate equal to 76% for this sample (see
Table I).
The whole results extracted from HR-XRD (scans and
RSM) are presented in Table I.
V. DISCUSSION
Let us discuss the results keeping in mind the open ques-
tion if the metastable critical thickness (as given, for
instance, by the People-Bean interpolation, Eq.(12)) marks
the easy visibility of misfit dislocations. We remind the
reader that experiments described in this paper were per-
formed under state of the art clean room conditions for
substrate cleaning and epitaxy loading to avoid particle
contamination.
One sees immediately that the first dislocations (A200)
appear above the People-Bean curve. This observation con-
firms the thickness area below the People-Bean curve (and
above the equilibrium curve) to be a true metastable region pro-
tected (at 550 �C) from heterogeneous nucleation by an energy
barrier if strong disturbances as surface particles are absent.
Speculations seem valid that moderate misfit dislocation den-
sities observed earlier at the People-Bean curve are connected
to particle contamination under routine laboratory conditions.
The second observation is the slow onset of relaxation
that gives an apparent higher critical thickness from relaxa-
tion measurements (X-ray diffraction, Raman shift belong to
this type of measurement, but also TEM because of the high
magnification method). For a broad range of measurement
methods it seems therefore more appropriate to define a criti-
cal thickness band instead a curve. The lower critical thick-
ness bound is defined by misfit dislocation counting
(extrapolation to zero density), the upper critical thickness
bound is defined by relaxation measurement (extrapolation
to zero relaxation).
On the search for an extrapolation scheme, let us con-
sider first the equilibrium case. Here, the strain e and the mis-
fit dislocation distance p are given by a minimum of the total
energy (per area) Etot, which is composed of the contribu-
tions of the film strain energy Ef and the dislocation net-
works energy En.
Etot ¼ Ef þ En; (14)
where isotropic elastic theory delivers for a biaxial stressed
layer
Ef ¼ C1e2t;
C1 ¼ 2G½ð1þ vÞ=ð1� vÞ� (15)
and dislocation theory,30 we use here for simplicity the edge
dislocation energy, delivers for a single dislocation
Ed ¼ C2½1þ ln ðRa=riÞ�;
C2 ¼ Gb2=½4pð1� vÞ�: (16)
The network energy En is then given by (the factor 2 refers
to the orthogonal network on (001) orientations)
En ¼2
pEd; (17)
The quantities ri and Ra in Eq.(16) are the inner and outer
cutoff radii, respectively.
The inner cutoff radius ri is a fraction of the Burgers
vector length b, its exact value depends on the dislocation
core energy, and this value is related to the value a in Eq. (7).
a ¼ 2eðb=riÞ; (18)
where e is the base of the natural logarithm (2.7). The outer
cutoff radius is about twice the film thickness t at low
FIG. 10. (Color online) 224 RSM of sample A800. The axis are in hkl units.
Si Bragg peak is located at h ¼ k ¼ 2 and l ¼ 4, while the SiGe one appears
at lower l and h values (larger lattice parameters).
063507-7 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
dislocation densities. This value is assumed for the critical
thickness calculation. At higher dislocation densities
(p/2� t) the mutual dislocation interaction drops the outer
cutoff radius to p/2.
Ra ¼ 2t ðregime I; t < p=2Þ;
Ra ¼ p=2 ðregime II; t � p=2Þ: (19)
For more refined calculations a smooth transition between
the two regimes is assumed,31 which is neglected for our
purposes.
The minimum energy condition is found considering the
relation between e and p (Eq. (10))
@Etot
@e¼ 2C1etþ
2
p
@Ed
@eþ Ed
2
b0: (20)
For regime I (Eq.(19)) the solution is straight forward
because Ra does not depend on e, that means @Ed=@e ¼ 0.
This leads for regime I to
et ¼ �ðEd=C1Þb0 ¼ C3½1þ ln ð2t=riÞ�;
C3 ¼ �C2
C1b0¼ �b½8pð1þ vÞ cosk��1; (21)
For regime II the term @Ed=@e reads C21p
b0
ðeþ f Þ2 which to-
gether with Eqs. (10), (16), (20) gives
et ¼ C3 2þ lnb0
2r1ðeþ f Þ
� �� �: (22)
The equilibrium critical thickness tc immediately follows
from Eq. (21) when one sets –e ¼ f and t ¼ tc
f � tc ¼ �C3½1þ ln ð2t=riÞ�; (23)
which is equivalent to the force-considerations (Matthews-
Blackeslee approach) when replacing the inner cutoff radius
ri by Eq. (18) (ri ¼ 2eb/a).
For comparing an experimental group of samples with
some scattering in the Ge content and therefore mismatch f,the consideration of the relaxation degree is more general.
We obtain by combining Eqs. (13), (20), (22) and (23) an
equilibrium relation for the relaxation branch above the criti-
cal thickness (t larger tc).
� etf¼ ð1� rÞt ¼ tc
1þ ln ð2t=riÞ1þ ln ð2tc=riÞ
: (24)
The left side of Eq. (24) is proportional to the force with
which a strained layer bends the substrate (note: Stress
times thickness describes the force from a unit length). In
Fig. 11 the experimental term (1�r)t is given as a function
of the thickness t. We see the linear increase of force with
thickness for pseudomorphic and weakly relaxed (r< 0.02)
layers and with higher thickness (below 800 nm) a sudden
decrease caused by strong relaxation. As an aid for the
eye, a solid line with strong relaxation onset at 600 nm is
drawn.
As the right term in Eq. (24) is only slowly varying with
lnt we alternatively can suggest a double logarithmic presen-
tation with the variables ln(1�r) versus lnt (Eq. (24))
ln ð1� rÞ ¼ �lntþ lntc þ ln1þ ln ð2t=riÞ1þ ln ð2tc=riÞ
: (25)
Figure 12 exhibits log(1�r) versus logt according to the
form of Eq. (25). Again the straight line is an aid for the eye
demonstrating the sudden decrease in strain by the strong
relaxation.
Both diagrams (Figs. 11 and 12) deliver from the extrap-
olation of experimental points in their relaxed branch a value
of the upper bound tco of the metastable critical thickness
band.
The strong onset of relaxation is linked to dislocation
multiplication sources that can work after crossings of dislo-
cations from different first generation heterogeneous sources
occur. Within the metastable critical thickness band, a pre-
sentation following Eq. (24) is useless because of relaxation
degrees much smaller than 1, which lets the factor (1�r) vir-
tually unchanged. For this band regime between lower bound
tcl and upper bound tco of the critical thickness, we use a
simplified version of a general relaxation scheme given by
Dodson-Tsao25 as extrapolation relation for tcl
FIG. 11. The relative force of a strained layer (1�r) � t vs thickness (logt).Increase of the force up to the upper bound tco then sudden decrease at the
relaxation branch. The force causes a curvature of the wafer.
FIG. 12. Experimental (crosses) relative strain values (1�r) vs thickness
(logt). The strain is nearly constant up to an upper bound tco where strong
relaxation sets in.
063507-8 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
dedt¼ Constðe� eequÞ2 ðf þ eþ esÞ: (26)
For the critical thickness band the conditions jej jeequjand jf þ ej < es are valid, with eequ the equilibrium strain
and es a strain value characteristic for the heterogeneous
nucleation site density. The simplification under this condi-
tion reads
dedt¼ Const ðese
2Þ (27)
with the solution
1
fþ 1
e¼ �Const ðt� tclÞes (28)
if relaxation starts at the lower bound tcl. Relaxation r is then
a linear function of the thickness
r ¼ �Const ðt� tclÞ ees: (29)
Figure 13 gives the presentation of r versus t below the upper
bound thickness. A lower bound of about 140 nm is found.
The samples with thicknesses below the lower bound A100,
A50 had zero misfit dislocation densities as demonstrated by
X-ray topography.
In the absence of particle contamination the heterogene-
ous nucleation starts below a thickness of 200 nm which is
similar to the critical thickness predicted by the People-Bean
interpolation formula. Serious strain relaxation (more than
2%) may be found at higher thickness values. We define a
critical thickness band with lower bound given by the onset
of dislocation nucleation and an upper bound defined from
strain relaxation measurements.
For the given experimental MBE conditions:
Ge content about 20% (lattice mismatch f ¼ 0.8%,
growth temperature 550 �C, growth rate 0.1 nm/s, p-substrate
doping ffi 1016=cm3, wafer surface finish from supplier
A) the metastable critical thickness band is given by
tcl¼ 140 nm to tco ¼ 600 nm.
VI. CONCLUSIONS
We investigated the strain relaxation mechanisms of epi-
taxial SiGe layers at 550 �C growth temperature. A rather
low Ge content of 20% was chosen to avoid competition in
compressive strain relaxation mechanism between misfit
dislocation generation and surface roughening via Stranski-
Krastanov growth mode.32 At the selected growth tempera-
ture (550 �C) the strain relaxation in the heterostructure layer
with about 0.8% lattice mismatch is given by misfit disloca-
tion network generation at the substrate-film interface. At
higher growth temperatures the competition from surface
roughening occurs at lower mismatch values as already
shown in the first work on SiGe/Si heterostructure stability.33
The distinct growth temperature of 550 �C was chosen to
allow a comparison with the classical People/Bean11 publica-
tion on non equilibrium critical thickness. In their original
paper11 People/Bean claimed to have found an energy bal-
ance calculation different from v. d. Merwe8 but now the
research community (see Refs. 25 and 31) agrees that the
People/Bean critical thickness curve separates metastable
strained layer growth from partially strain relaxed films. Ear-
lier work (Refs. 25, 28, and 12) considered the curve as a
detection limit of dislocations in thin films. Our approach
differed twofold compared with earlier work:
(i) The epitaxy under modern clean room conditions
avoided dust particle contamination of the wafer surface.
Even very small submicron dust particles act as strong heter-
ogeneous dislocation sources when overgrown by a film.
(ii) Observation of dislocation networks by two comple-
mentary X-ray methods (topography, diffraction) allowed a
rigorous dislocation and strain monitoring on the same sam-
ple series. Especially, X-ray topography is a unique method
to monitor low dislocation densities from heterogeneous
nucleation sites because it combines sensitivity to single dis-
locations with large area observations. Competing methods
suffer from low sensitivity to single dislocations (e.g.,
haze34) or from small area observations (e.g., atomic force
microscopy of slip steps). In principle selective defect etch-
ing can be applied to low density defects but with thin films
the etching depth is restricted, which in practice demands
high magnification observation of small areas.
X-ray diffraction methods are established for strain
relaxation measurements with more than 2% relaxation
degree,35 which covers the high density network side. Simi-
lar restrictions are valid for other strain measurement meth-
ods such as Raman spectroscopy.36 Under these
experimental conditions we found the following:
(i) The Si0.8Ge0.2 films are growing (at 550 �C) misfit
dislocation free up to a critical thickness of 140 nm
(much thicker than the equilibrium thickness of
10 nm, roughly about the People/Bean curve). The
dislocation freedom proves the region between 10 nm
and 140 nm thickness to be a true metastable regime
where a nucleation barrier avoids dislocation genera-
tion if dust particle contamination is deleted.
(ii) Measurable strain relaxation (>2%) is shifted to
thickness values (about 600 nm) considerably higher
than the People/Bean curve. Similar observations are
FIG. 13. Relaxation r vs thickness t. The early stage of heterogeneous
nucleation is represented by a straight line the extrapolation of which to
r¼ 0 gives the lower bound tcl of the critical thickness band.
063507-9 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions
also made with chemical vapor deposition epitaxy35
but at different temperature (650 �C). We assume
therefore that this observation is general for epitaxy
under particle free condition.
(iii) We propose a metastable critical thickness band with
a lower and an upper bound, respectively. The lower
bound tcl (140 nm) marks the onset of heterogeneous
nucleation whereas the upper bound tco (600 nm)
marks the onset of measurable strain relaxation.
We could not identify the nature of the dislocation
nucleation sites in particle free epitaxy. However, X-ray to-
pography demonstrated clearly that oxide precipitations
inside in annealed Czochralski grown wafers are not respon-
sible for the nucleation. We speculate that surface step accu-
mulations (ripples), self assembled Ge content variations or
mechanical surface polishing defects could deliver that
remaining nucleation sites.
The metastable growth regime with fully strained SiGe
layers (pseudomorphic growth) could possibly extended even
more if these nucleation sites may be reduced or the barrier
height may be increased. These future investigations need close
cooperation with wafer manufacturers because on commercial
wafers the surface finishing is not well known to the customer.
1L. Vegard, Z. Phys. (Germany) 5, 169 (1920).2J. P. Dismukes, L. Ekstrom, and R. J. Paff, J. Phys. Chem. 68, 3021
(1964).3E. Kasper, A. Schuh, G. Bauer, B. Hollander, and H. Kibbel, J. Cryst.
Growth 157, 68 (1995).4R. Kritkivasan, Y. Lu, J. D. Cressler, J. -S. Rieh, M. H. Khater, D.
Ahlgren, and G. Freeman, IEEE-EDL 27, 567 (2006).5N. Zerounian, F. Aniel, B. Barbalat, P. Chevalier, and A. Chantre,
Electron. Lett. 43(14), 747 (2007).6E. Kasper, D. Kissinger, P. Russer, and R. Weigel, IEEE Microwave Mag.
10, 528 (2009).
7T. Ghani et al., Tech. Dig. IEDM, 978 (2003); Z. Lou et al., ibid. 489
(2005).8J. H. Van der Merwe, J. Appl. Phys. 34, 113 (1963).9J. W. Matthews and A. E. Blakeslee, J. Cryst. Growth 27, 118 (1974).
10M. L. Green et al., J. Appl. Phys. 69, 745 (1991).11R. People and J. C. Bean, Appl. Phys. Lett. 47, 322 (1985); 48, 229 (1986).12R. Hull, EMIS Data Reviews Series (IEE, London 2000), Vol. 24, p. 21.13R. Hull and J. C. Bean, J. Vac. Sci. Technol. A 7, 2580 (1989).14E. Kasper, M. Bauer, and M. Oehme, Thin Solid Films 321, 148
(1998).15M. Bauer, M. Oehme, and E. Kasper, Mater. Sci. Eng. B 89, 263 (2002).16A. R. Lang, J. Appl. Phys. 29, 597 (1958).17A. Authier; B. Rogers, and A. R. Lang, Philos. Mag. 12. 547 (1965).18B. K. Tanner, X-ray Diffraction Topography (Pergamon, Paris, 1976).19V. Holy, U. Pietsch, T. Baumbach, High Resolution X-ray Scattering from
Thin Films and Multilayers, Springer Tracts in Modern Physics, Vol. 149
(Springer, 1999).20P. F. Fewster, X-ray Scattering from Semiconductors (Imperial College,
2003).21H.-J. Herzog and E. Kasper, J. Cryst. Growth 144, 177(1994).22E. Koppensteiner et al., Appl. Phys. Lett. 62, 1783 (1993).23B. W. Dodson and J. Y. Tsao, Appl. Phys. Lett. 51, 1325 (1987).24W. Hagen and H. Strunk, Appl. Phys. 17, 85 (1978).25R. Hull, J. C. Bean, D. Bahnk, L. J. Peticolas, K. T. Short, and F. C. Unter-
wald, J. Appl. Phys. 70, 2052(1991).26D. C. Houghton, J. Appl. Phys. 70, 2136 (1991).27Y. Yamashita et. al., Philos. Mag. Lett. 67, 165 (1993).28Y. B. Bolkhovitynov, A. S. Deryabin, A. K. Gutakovskii, and L. V.
Sokolov, J. Crystal Growth 312, 3080 (2010).29E. Kasper, Appl. Surf. Sci. 102, 189 (1996).30F. R. N. Nabarro, Theory of Crystal Dislocations (Clarendon, Oxford,
1967).31E. Kasper and H. J. Herzog, Thin Solid Films 44, 357 (1977).32K. Tillmann, H. Trinkaus, and W. Jager, EMIS Data Reviews Series (IEE,
London, 2000), Vol. 24, p. 63.33E. Kasper, H. J. Herzog, and H. Kibbel, Appl. Phys. 8, 199 (1975).34V. Destefanis, J. M. Hartmann, A. Abbadie, A. M. Papon, and T. Billon,
J. Cryst. Growth 311, 1070 (2009).35J.M. Hartmann, A. Abbadie, and S. Favier, J. Appl. Phys. 110, 083529
(2011).36T. S. Perova, J. Wasyluk, K. Lyutovich, E. Kasper, M. Oehme, K. Rode,
and W. Waldron, J. Appl. Phys. 109, 033502 (2011).
063507-10 Kasper et al. J. Appl. Phys. 111, 063507 (2012)
Downloaded 07 Feb 2013 to 195.221.220.19. Redistribution subject to AIP license or copyright; see http://jap.aip.org/about/rights_and_permissions