You are on page 1of 12

surface s c i e n c e

ELSEVIER Surface Science 375 (1997) 129-140

Diffusion on strained surfaces


M . S c h r o e d e r *, D . E . W o l f 1
Theoretische Physik, FB 10, Gerhard-Mercator-Universitat, 47048 Duisburg, Germany
Received 15 October 1996;accepted for publication 23 October 1996

Abstract

The change of diffusion kinetics when elastic fields are present is discussed for diffusion on (001) surfaces of simple cubic, fcc and
bcc lattices. All particles interact pairwise with a Lennard-Jones potential. The simple cubic lattice was stabilized by an anisotropic
prefactor. It is found that genericallycompressivestrain enhances diffusion whereas tensile strain increases the activation barrier. An
approximately linear dependence of the barrier in a wide range of misfits is found. In heteroepitaxy,diffusion on top of large clusters
is inhomogeneous and anisotropic. The kinetics close to edges and centers of islands are remarkably different. In many cases changes
of binding energies are small compared to those of saddle point energies. Thermodynamic arguments (minimization of free energy)
are not appropriate to describe diffusion on strained surfaces in these cases. © 1997 Elsevier Science B.V. All rights reserved.

Keywords: Models of non-equilibrium phenomena; Models of surface kinetics; Semi-empiricalmodels and model calculations; Surface
diffusion; Surface relaxation and reconstruction; Surface stress

I. Introduction can be reached locally at moderate flux, the whole


system is generally far away from its minimal
Growth of high-quality thin films and multilayer free energy.
systems has attracted increasing interest during Classical rate equation treatments and computer
recent years. Experimentally it is a great challenge simulations have been used to understand growth
to optimize the growth of both hetero- and homo- up to a few monolayers [ 1 - 7 ] . These treatments
epitaxial layers. The adjustment of growth parame- took into account the basic microscopic processes
ters like temperature, flux, substrate preparation, of MBE such as deposition, diffusion, island forma-
growth interruption and heating is, however, tion and growth, but have not considered strain.
mainly guided by the enormous empirical back- Elasticity can influence the surface morphology in
ground. In addition to the technical relevance for three ways: by interactions between steps a m o n g
semiconductor devices, epitaxial growth is also an themselves or with adatoms [ 8 - 1 3 ] , by modifica-
interesting field for theoretical research. The depos- tion of the sticking probability at edges of islands
ition flux prevents the system from reaching its [ 1 4 - 1 7 ] and by strain-dependent surface diffusion
equilibrium morphology. Although equilibrium constants 1-18-20]. A rich variety of microscopic
and mesoscopic approaches has been used to take
* Corresponding author. Fax: +49 203 379 2816;
strain into account in the description of growth
.'-mail: schroed@comphys.uni-duisburg.de
1 Present address: HLRZ c/o Forschungszentrum J01ich, and diffusion. H o w strain affects the kinetics of
D-52425 Jiilich, Germany. pair-bonded systems was addressed with con-

)039-6028/97/$17.00 Copyright © 1997 Elsevier Science B.V. All rights reserved


PII S0039-6028 (96)01250-2
130 M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140

tinuous space Monte-Carlo methods [21-23] and What is the energy landscape seen by an adatom
molecular dynamics simulations [24-26], focus- diffusing on top of a large strained island, i.e. how
sing mainly on the resulting layer growth and the are kinetics changed compared to the flat sub-
occurrence of dislocations. Rate equations and strate? How does the adatom mobility on top of a
lattice Monte-Carlo simulations based on the three-dimensional island depend on its height?
Frenkel-Kontorova model [14-16,27,28] were The outline of this paper is the following.
performed to analyse modifications of the step- Section 2 describes the method used to model
flow growth mode, cluster statistics and three- diffusion on discrete elastic solids. In Section 3
dimensional growth in mismatched systems as well elastic properties, including classical interactions
as in harmonic coupled lattices [ 29]. Furthermore, of the modelled solids, are presented. In Sections
molecular dynamics simulations were applied to 4 and 5 results for free diffusion on strained surfaces
study diffusion on semiconductor surfaces, both in and islands are given. The subject of Section 6 is
homo- and heteroepitaxy [30-33]. Though three layer dependent activation energies. A conclusion
particle potentials stabilizing the diamond struc- is given in Section 7.
ture were used, some trends observed on the
strained surface could be understood in terms of
pair potentials. Diffusion equations [ 12,13,34-38]
based on classical elasticity theory were mainly 2. Method
used to address morphological instabilities. In
these treatments elastic interactions between steps We consider three different crystal structures,
and adatoms, or chemical potentials modified by simple cubic (sc), fcc and bcc, in order to distinguish
the elastic energy density, lead to a diffusion bias. general from lattice-specific strain effects. The
Effective medium theory, e.g. Refs. [20,39], and origin of the strain is unimportant for the phen-
first principle calculations [40] were used to calcu- omena. Hence we need not study an adlayer on a
late diffusion barriers on metallic singular surfaces substrate with a lattice mismatch, where the strain
and close to steps, both in homo- and heteroepitax- is 6a/a, but can study a single-species model with
ial systems. Since both approaches include strain due to external forces.
quantum mechanical details which are computa- We use a slab geometry with one free and one
tionally expensive, smaller systems were calculated frozen (001) surface perpendicular to the z direc-
in which long-range elastic fields are cut off. tion, and periodic boundary conditions in the x
Since the surface morphology is determined by and y directions. There may be islands on the free
kinetics as well as thermodynamics it is necessary surface. The strain is imposed by changing the
to calculate the activation barriers from the micro- system size in the x and y directions for a fixed
scopic diffusion trajectories, i.e. to determine the number of atoms in the x-y plane. With this
binding (B) and saddle point (SP) energies. The constraint all atoms, apart from those in the frozen
aim of this work is to analyse the influence of layer, are allowed to relax in order to minimize
strain on these energies. We focus on free diffusion the total energy, given by the sum of pair potentials
of adatoms on inhomogeneously strained surfaces. specified below. Relaxation is done by a conjugate
Edge diffusion, advacancy diffusion, etc. will be gradient algorithm and is stopped when both an
deferred to a forthcoming publication. We take energy convergence criterion is fullfilled and the
care that the crystal lattices considered for the largest force acting in the system is smaller than a
evaluation of energy barriers are large enough in certain upper bound (10 -5 for fcc and bcc, 10 -3
all three dimensions so that the results of classical for sc). The value of this bound was chosen to
elasticity theory are reproduced. guarantee that for a pair of atoms the equilibrium
In particular we are going to address the distance is reached within an error of 0.001%.
following questions. Is the SP or the binding energy Activation barriers for diffusion to a neighboring
more sensitive to strain? Only the latter one can binding site are now calculated using the conven-
be treated within a chemical potential approach. tional minimal energy path saddle-point method
M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140 131

(see e.g. Refs.[33,41]). The relaxed substrate is perfectly justified; the crystal surface breaks rota-
frozen and an adatom is placed on a binding site. tional invariance anyway and the adatom feels the
While this test atom is moved to a neighboring directional bonds sticking out of the surface along
binding site in small steps, the substrate is either the lattice directions.
kept frozen all the time or it is allowed to relax Each atom interacts with its nearest and second-
due to the influence of the adatom, either only nearest neighbors. For barrier calculations the
locally or globally. The adatom is always relaxed third-nearest neighbor is also taken into account.
in the plane perpendicular to the diffusion path.
Of course, global relaxation is the correct pro-
cedure. However, we are going to show that local
relaxation or even a frozen substrate give surpris-
3. Classical elastic interactions
ingly accurate values for the energy barriers. In
the frozen case only the adatom energies need to
It is well known that linear continuum elasticity
be recorded, otherwise one has to measure the
predicts that two one-dimensional defects, such as
total energy of the whole system and small energy
rows of atoms or vacancies, have a repulsive inter-
differences become susceptible to large relative
action of the form E ~ 1Ix 2, where x is the distance
numerical errors.
between them [8-10]. For periodic boundary con-
All atoms interact pairwise with the same poten-
ditions, the interaction also has to take into
tial. For the fcc structure the conventional
account the periodic images of the defects. The
Lennard-Jones (LJ) potential, VLj(r)=4[(tT/r) 1 2 -
leading terms are E.,. A ( x - 2 + (L-X)--2), where L
(a/r)6], is used. The bcc and sc structures are more
is the lateral dimension of the system [42].
subtle as they are not stable for Lennard-Jones
The question arises of whether this interaction
interactions. The slab geometry of a bcc Lennard-
is reproduced by the discrete systems under con-
Jones crystal can be kept metastable by the peri-
sideration here. Fig. 1 shows the results for the fcc
odic boundary conditions in the x and y directions,
crystal, with two rows of vacancies, oriented along
so that an adatom can trace out the diffusion
the [100] directions, on the surface [43]. The
barrier on the flat free surface. However, islands
lateral system size L was a multiple of the lattice
on top of the surface were unstable with respect to
constant of the crystal at temperature T = 0 and
buckling. The reason why we use the Lennard-
pressure p = 0, which we take as the reference state
Jones potential for the bcc crystal, nevertheless, is
the conceptual advantage that the bcc structure
can be viewed as a deformed fcc lattice (as well
known from martensitic transitions). Rescaling the • i , i , r ' i '

x and y directions of the fcc crystal of lattice


constant 1 by a factor a v ~ and the z direction by -3.850
a produces a bcc crystal with lattice constant a.
In order to stabilize a simple cubic lattice, an
anisotropic LJ potential, ~ -3.86o
LU

V(r)=
I
0.1+8 x

y2 1 I
o^ i I , i , i ,

"3"87z.0 4.0 6.0 8.0 10.0 12.0


x
Fig. 1. Elastic interaction of two rows of vacancies on a (001)
is used. This potential is not rotationally invariant surface of an fcc lattice of size L=25. The line is a fit with the
but for the purpose of simulating a single adatom analytical form of the interaction obtained from continuum
on a crystal surface with cubic symmetry this is elasticity theory, with the parameter A =0.199 (see text).
132 M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140

for the strain. The expected elastic interaction is monicity of the potential. For shear strain it is
nicely reproduced, indicating that already for mod- valid up to ex~ ~0.3.
erate sizes of the discrete system the continuum
description is applicable. We stress that these inter-
actions, calculated for an elastic halfspace, are only 4. Diffusion on homogeneously strained surfaces
observable if the thickness of the slab was larger
than its lateral dimension. Otherwise the frozen In this section the influence of strain on the
first layer gives rise to an additional boundary diffusion barrier of an adatom on a flat (001)
condition that cuts off the elastic interaction along surface will be discussed. Fig. 2 shows the total
the surface at distances of the order of the slab energy of a simple cubic system as function of the
width. x coordinate of an adatom, which is moved from
The simple cubic system behaves completely one binding site to the next one across the saddle
differently. First of all, on an unstrained slab, no point along the [100] direction. Different values
dependence of the energy on x could be observed. of strain are imposed by varying a in the Lennard-
Second, for a strained slab, the total energy did Jones potential. Thus the system's lateral size in
depend on the distance between the vacancy rows units of the equilibrium lattice constant varies
in a qualitatively similar way as in Fig. 1, although while the number of atoms in the x - y plane
the energy variations were smaller by about a remains fixed.
factor of 0.01. This result did not depend on the For tensile strain (upper curves in Fig. 2) the
slab width. In fact, it remained essentially barrier is increased. Compressive strain (lower
curves) decreases the barrier. This is mainly due to
unchanged even if the two terraces, separated by
a change of the saddle-point energy, whereas the
the missing rows, were sitting directly on top of
binding energies are shifted only very little. In
the frozen layer. This shows that the x-dependence
systems with pair potentials this is a general trend.
of the energy is not caused by an elastic interaction
Suppose that only nearest-neighbor interactions
mediated by the bulk. Instead, the edges of the
have to be taken into account and that the mini-
terraces relax. This relaxation must extend over
several lattice constants, so that the energy gain is
largest if neither of the two terraces is too small.
Obviously the relaxation of the terrace edges
does not lead to significant displacements in the -0.8
underlying slab and so no observable elastic inter-
actions are mediated. This is peculiar to the aniso-
c-
tropic LJ potential [Eq. (1)]. Calculating the ILl -1.1
elastic constants C44 and Cll related to shear
(exz S 0 ) and uniaxial strain (ex~ S 0 ) reveals that
C44 is 80 times smaller than Cll, i.e. the terraces
-1.2
almost "float" on top of the underlying material o.o 1.0
when relaxing laterally. Hence, a two-dimensional Position
Frenkel-Kontorova model should be a more
appropriate theoretical description than contin- Fig. 2. Diffusion barrier of a homogeneously strained (001)
surface of a simple cubic lattice. Middle two curves: unstrained
uum elasticity. system. Upper two curves: tensile strain Exx=Err=0.05. Lower
The evaluation of the elastic energies of a unit two curves: compressive strain Exx=Err= --0.05. Energies are
cell as a function of the strain not only leads to shown for global relaxation (filled symbols) as well as for a
the values of the elastic constants but also allows frozen slab (open symbols). For local relaxation close to the
adatom only, the curves lie slightly above those for global
assessment of the limits of validity of linear elastic- relaxation (within half the symbolsize).Energyis given in units
ity. A harmonic approximation of the energy is of the energy parameter of the LJ potential and the zero of the
only reasonable up to e~x ~ 0.03 due to the anhar- energy scale is arbitrary in this plot.
M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140 133

mum of the pair potential is at a distance to. The This can be seen most easily by expanding
adatom will have this distance to all its nearest
neighbors at the binding site. Applying a compres- A E = [ ~ V((I+
"~'sP)=
~-, --'ad~(SP)',
sive strain the adatom will therefore move a little
away from the surface, while for tensile strain it
will get closer to it, always trying to keep the + 1/2 ~ V((l+O(r,(SP) - r ~(SP))) 1
preferred distance r0 to all its nearest neighbors. t,J
Hence the binding energy will not change as long - [(SP)~(B)] (2)
as only nearest-neighbor interactions are taken
into account and the strain is sufficiently weak. to first order in the strain tensor =e [10]. The
Thus the diffusion barrier is mainly determined by summation indices label the atoms in the slab, and
the strain dependence of the saddle-point energy, rad is the position of the adatom. All atomic
where the interaction with next-nearest neighbors positions in the fully relaxed system depend on
is stronger than at the binding site. This is in whether the adatom is at the saddle point or at
contrast to recent effective medium calculations the binding site. If one ignores the extra strain due
for A g ( l l l ) [20], which indicate that in that case to the presence of the adatom, as will be justified
many body contributions to the energy are strong. below, r!sP)=r! B), SO that the second term in Eq.
In addition, the decrease of the barrier for com- (2) cancels with the corresponding one in the
pressive, and its increase for tensile, strain is a second half of the formula. Then the expansion is
general principle (cf. Ref. [20]). It can be under- simply
stood easily by considering the limits of large (SP)- m.~
(B))e.~ ,
AE = const. + (muv (3)
strains. For large compressive strain the substrate
can be viewed as a continuous film. There are no where
more discrete binding sites in this limit and, hence,
diffusion barriers vanish. In the other limit, for muv = ~ O~V(ri-- r,d)(ri)~ (4)
i
large tensile strains, the substrate consists of iso-
lated atoms. Diffusion is equivalent to breaking a denotes the force dipole tensor of the adatom, to
pair of atoms and building a new pair. Hence the be taken at the saddle point or the binding site,
diffusion barrier is equal to the pair-binding energy. respectively. As we argued above, the binding
Fig. 3 shows that the diffusion barrier AE energy changes much less than the saddle point
depends approximately linearly on strain, as also energy, [m(m[<<[m(SP)[, so that the first order of the
observed in Ref. [20]. This is also a general result. expansion does not vanish. This explains why in
general the diffusion barrier depends linearly on
the strain.

//
i i i
The barriers obtained by global relaxation (filled
~)~ 0.50
symbols in Fig. 2) [44] are only slightly smaller
than the ones for diffusion on a frozen surface
C-
(I)
(open symbols). Energy changes due to adatom-
t- 0.40 induced displacements can be ignored for moderate
o values of the strain. However, this is not true for
.>
D 0.30 strains below 0.01 or above 0.09. This estimate
0 results from a comparison between the extra strain
<
induced by the adatom and the original strain ¢ of
0.20 ' ~ , ' , ' ,
-0.10 -0.05 0.00 0.05 0.10 the slab. The elastic contribution to the diffusion
Strain barrier is correctly estimated from a calculation
Fig. 3. Activationbarrier versus strain for diffusionon a (001) with a frozen substrate as long as the extra strain
surface of a simple cubic crystal. A linear dependenceis given close to the adatom is much smaller than e. For
between strain and activationbarrier for up to 8°,/ostrain. large values of ¢ the extra strain is large because
134 M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140

the elastic constants of the crystal become very All energies and barriers are plotted versus a
small. If one is interested in differences of diffusion since the value of the strain depends on the choice
barriers for different strains, instead of the value of the reference state (bcc or fcc), c = l - a / a r o f ,
of a barrier itself, Fig. 2 shows that a calculation where O'ref is the value for the reference state. In
with a frozen substrate is almost as good as the any case the distance between atoms in units of
correct fully relaxed one because the error one the equilibrium lattice constant decreases for
makes is almost the same for the different strains. increasing a, as explained above. Therefore, the
This justifies the approximation made in Eq. (2). activation barrier for diffusion decreases mono-
On bcc or fcc (001) surfaces the trend is the toneously with a, as expected according to the
same in a wide range of strain (Fig. 4). By varying general arguments above. As for the simple cubic
tr in the Lennard-Jones potential the solid can lattice, a linear dependence between strain e and
be shifted from a bcc crystal (trb~=0.793) to barrier is given in a good approximation around
an fcc type (af~=0.899), although the former is O'bcc and O'fc e.
unstable without periodic boundary conditions For the fcc crystal the binding energy of an
(ground state energy per particle Eb~c= adatom depends only weakly on strain, in
-- 12.708 > -- 13.442 = Erc¢). agreement with the general argument above.
However, the bcc crystal obviously violates this
rule (Fig. 4b)" Here the strain dependence of the
2.5 ' i , i , i ,
diffusion barrier is mainly due to the binding
energy. This exception from the rule can easily be
2.0 ~
== understood. In the bcc crystal the atom sitting in
o
the second layer directly under the adatom is so
._~ 1.5
close that its contribution to the binding energy is
1.0 about three times larger (E~b~=0.187) than for the
fcc crystal (Eft=0.062). For larger tensile strain
0.5 i I I this atom gets closer and closer to the adatom and
(a) at a=0.725 would be as close as the nearest
-2.00 r i i neighbors. The simple argument above fails
because a next-nearest neighbor gives a non-
negligible contribution.
--3.00 (
We increased the number of neighbors taken
into account when the energy of the adatom is

f-
--4.00 [] [] binding energy calculated. For the values of a considered there is
LU no qualitative difference to the data shown.
Quantitative changes are about A E / E ~ 8 % for
--5.00 both the saddle-point and the binding energies.

O i i i
-6"00.72 0.82 0.92 1.02 5. Diffusion on inhomogeneously strained surfaces
(7
(b)
In this section we investigate the adatom diffu-
Fig. 4. (a) Activation energy for diffusion on homogeneously sion on an island of atomic height on top of the
strained bcc/fcc (001) surfaces. The trend of increasing/ strained slab considered so far. As the island edges
decreasing barrier for tensile/compressivestrain is similar to relax, the strain along the surface is no longer
the simple cubic system. (b) For the bcc crystal the main shift constant but inhomogeneous. We show that bind-
is due to changes of binding energies that are affected by
changes of crystal plane distances according to the Poisson ing and saddle-point energies then also depend on
ratio. Mainly saddle-point energies are changed in the fcc position. Diffusion close to the edge is different
system. from that near the center of the island. This can
M. Schroeder, D.E. Wolf~Surface Science375 (1997) 129-140 135

be understood qualitatively by applying our previ- order in the difference of adatom density and
ous results for homogeneously strained layers binding energy, one obtains
locally, which is not so surprising as the main u(B) ~(B.)
Pm - - Pn Pm -[- Pn ~ m - - L~(n)
energy contributions, proportional to the strain of jmn ~'~JDmn - - "~ ~ n n - - , (6)
the slab, had a local origin, whereas the long-range a 2 a
terms were much weaker (order 10 - 6 , cf. Fig. 2). where Dmn=a2x/Wm~nWn_.mand #m=flD~ are
Whereas the changes of binding and saddle- the local adatom diffusion constant and mobility,
point energies can be predicted by our previous respectively. The first term in Eq. (6) is Fick's law,
results, as will be shown below, their inhomo- the second the drift term. Hence the two cases
geneity has new kinetic implications, which will be shown in Fig. 5, although having the same activa-
discussed now. In general, binding as well as tion barriers for diffusion, imply quite different
saddle-point energies will depend on position but kinetics. Inhomogeneous activation barriers, for
have completely different physical consequences. example, lead to inhomogeneous nucleation of
In order to illustrate this, we compare the two islands and inhomogeneous binding energies
limiting cases illustrated in Fig. 5. Either only the induce a drift. Only the strain dependence of the
saddle-point or only the binding energy depends energy at the binding sites has been dealt with in
on position, in such a way that the respective most of the previous publications (see e.g. Refs.
activation barriers are locally the same. [12,13,17,34-38]). It should be stressed, however,
In the first case (Fig. 5a) diffusion currents on that a strain dependence of the activation barriers
the surface are given by Fick's law, whereas in the also has morphological consequences.
second (Fig. 5b) there is an additional drift towards Fig. 6 shows the energy landscape for diffusion
lower binding energies. This can be easily derived on an island in the simple cubic system both for
by evaluating the current Jm from site m to the 7% tensile and 6% compressive strain. AE changes
adjacent site n, by 28% (21%) for compressive (tensile) strain from
the center to the edge. The shape of the landscape
jmn=a(pmWm~n--pnWn~m), (5)
can be understood by referring to Fig. 3. In both
where a2pmis the probability of finding an adatom cases the island edge is able to relax. The lattice
at sitem and Wm_.n=Vexp~--pt mn -- E~))
"°'E(SP) constant at edges is closer to the one of the
denotes the transition rate from site m to n. unstrained system. Close to the center the surface
Assuming that the attempt frequency v does not is still highly strained. The lattice constant there is
depend on position, and expanding Jm to first close to the one imposed by the substrate. As a
consequence of this inhomogeneous relaxation,
diffusion is faster in the middle of the island for
compressive strain (negative misfit) compared to
the island edge. This leads to a reduction of the
nucleation rate of islands at the center and because
of the increased diffusion length adatoms try more
often to escape from the terrace. This slows down
their accumulation and hence the nucleation of
new islands. Tensile strain (positive misfit), how-
ever, slows diffusion down in the middle of the
island. Therefore adatoms can escape from the
terrace less easily and the nucleation probability
of an island on top of the island is enhanced. For
Position the cubic lattice essentially only the SP energies
Fig. 5. The limiting scenarios of diffusion barrier changes. (a) vary across the island. Hence, no significant drift
Only the saddle-point energies change. (b) Only the binding current arises, which could spoil the simple argu-
energies change. ment by transporting adatoms towards the island
136 M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140

i i i

-0.80

t-
-1.00
LU

- 1.20

-1.40 I i I
200.0 205.0 210.0 215.0 220.0
Position

Fig. 6. Diffusion on top of a simple cubic island for tensile (solid line) and compressive strain (dashed line). Diffusion is faster in the
center (close to the edge) of the island for compressive (tensile) strain.

edge or center and thereby also decreasing or compared with the above data, since the vertical
increasing the nucleation rate. corrugation affects all energetic shifts. However,
The fcc system shows a similar behavior though depending on the sign of strain, barriers are shifted
the effect is less pronounced than for the sc system, in agreement with the fcc and sc systems to higher
since AE changes from the center to the edge only or lower values.
by about 13% (Fig. 7). The reason is the stronger We stress that the energy landscapes of Figs. 6,
coupling to the underlying crystal layers, which 7 and 8 are due to the inhomogeneous strain. They
hinders the relaxation of the island edges. are hardly affected (comparable to the line width)
Although both systems are completely different when elastic step-adatom interactions are switched
with respect to coordination numbers and elastic off. This was checked by calculating the barriers
properties, there is a remarkable agreement in the on a frozen island, similar to Section 4. Classical
trends observed for free diffusion on their strained elastic interactions can be neglected for strong
surfaces. In the fcc system again predominantly enough lattice mismatch.
the SP energies are shifted by strain.
On the bcc (001) surface an island is unstable
against a vertical corrugation, indicating that the 6. Layer-dependent diffusion barriers
bcc structure is unstable for Lennard-Jones sys-

jJl/i/!/iji
tems. Nevertheless, barrier calculations were per- If an island on top of a strained slab is several
formed (Fig. 8). The results cannot easily be atomic layers high, the relaxation at its edges is

--2.6

==
tu --3.6

--4.6 i

255.0 260.0 265.0 270.0 275.0


Position

Fig. 7. Diffusion on strained fcc (001) islands. Dashed top curve: compressive strain. Lower curve, energies shifted by - 1: tensile strain.
M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140 137

-, I i , ..

-3.00

-4.00
(1)
(-
klJ

-5.00

i i i

-6"0013.5 523.5 533.5 543.5


Position
Fig. 8. Diffusionon a strained bcc (001) island, ___3% strain. The vertical corrugation, indicating that the bcc structure is unstable
for LJ interactions, causes a modulated energy landscape. For tensile strain barriers increase (dashed line).

height dependent. In continuum elasticity this can diffusion barriers and the strain, the activation
be modelled by the strain field in the island for a barrier, measured, for example, at the center of the
strain boundary condition on one side (correspond- island, approaches its value for the stress-free
ing to the uppermost complete layer of the slab) surface exponentially with increasing height of the
and free boundary conditions (stress equal to zero) island, with the same characteristic length
on all other sides. Above a certain height, propor- (Fig. 10).
tional to the lateral size of the island, the height
dependence of the displacement fields becomes
negligible; the strain tensor approaches a constant.
Correspondingly, the diffusion barriers depend on 7. Conclusion
the height of the island.
This should not be confused with recent experi- We analysed the effects of strain on surface
ments [18] and first principle calculations [19] diffusion barriers for (001) surfaces of three
which showed that activation energies depend on different crystal structures: simple cubic, bcc and
the thickness of a heteroepitaxial film. In contrast fcc. In all cases diffusion is slowed down for tensile
to our island, the film was not limited by edges and enhanced for compressive strain. We claim
[19] or the film thickness was much smaller than that in purely pair-bonded systems this is always
the lateral extent [18]. the case, irrespective of coordination numbers, i.e.
For both simple cubic and fcc systems com- of surface geometry. This is in agreement with
pressed islands of varying thickness were set up on more refined calculations and experiments [ 18-20]
a frozen substrate to analyse height-dependent for special systems. For strains up to + 5 % the
barriers in our model. They had the shape of a relation between activation barrier and strain is
square cylinder. Fig. 9 shows that lateral displace- linear to a very good approximation.
ments, measured at the edges, in a certain layer The similarity of the results for all three crystal
depend on the lateral size L of the islands. structures might seem surprising in view of their
All curves can be fitted with an exponential very different elastic bulk properties, as discussed
~ exp(-z/~), where the characteristic length ~ is in Section 3 and reflected in the different mediation
proportional to the lateral island size (Fig. 9c), as of elastic interactions between surface defects. The
expected. This is analogous to the well-known reason is that activation barriers for surface diffu-
result of continuum elasticity theory, where sion are mostly influenced by the lattice distortion
the displacement field of an infinite half-space, in the neighborhood of the adatom, whereas elastic
subject to a periodic force distribution with wave interactions mediated by the bulk are only small
length L on its surface, decays with an corrections. Also, for moderate physical values of
exponential exp(--z/L) into the bulk. Because of strain it is found that displacements imposed on
the approximately linear relation between the the bulk by the adatom force distribution can be
138 M. Schroeder, D.E. Wolf/Surface Science 375 (1997) 129-140

i , i

0.0

c
-0.2
E
o
^^AAAA^A^AAAAA

"1o
-- 0.4 ~ .××x×xxx

t~
_.1
, ',',',-i-I-
--0.6

--0.8
0.0 10.0 20.0 30.0 40.0
Height
(a)
i

-0.20

e-
Q
E
--0.40

"o

0 i I

--0.60 13 23 33
L

i
-,0.80
0.0 10.0 20.0
Height
(b) (c)

Fig. 9. Lateral displacements at the island edge of an fcc (001) island as a function of island height. Different symbols correspond to
different lateral sizes L of the island, as specified in (c). All curves were fitted with an exponential ~ exp(-z/~). (a) Same as as (b)
for simple cubic island. (c) The characteristic length ~ is proportional to L. Upper line: fcc. Lower line: simple cubic.

neglected as far as activation barriers are case, as a next-nearest neighbor gives an important
concerned. strain-dependent contribution to the binding
In the fcc system, and even more so for the sc energy.
crystal, it is mainly the shift of the saddle-point Diffusion barriers on top of islands, which could
energy which is responsible for the strain depen- be several layers high, were analysed. Because of
dence of the activation barrier. We argued that the height-dependent strain relaxation the barriers
nearest-neighbor pair potentials do not contribute approach the value of the unstrained surface expo-
to the strain dependence of the binding energy. nentially for increasing island height. This is in
Therefore, if the next-nearest-neighbor interaction agreement with continuum elasticity theory.
has a sufficiently weak strain dependence, the Generically, diffusion on top of islands is aniso-
binding energy will be essentially constant. tropic and inhomogeneous. Since island edges are
However, on the bcc (001) surface this is not the able to relax more effectively, the activation barrier
M. Schroeder, D.E. Wolf/Surface Science 375 (1997) 129-140 139

0.38
~J

i/
0.36

1.20
0¢,,.
0
o
t- g 0.34
O

< 1.15
J .-

0.32

0.30 i i i

, , , 0.0 10.0 20.0 30.0 40.0


1. 1 0
-0.0 10.0 20.0 30.0 40.0 Layer index
Layer index
(a) (b)

Fig. 10. Layer-dependent barriers for diffusion on (a) fcc and (b) simple cubic (001) islands. The barriers were calculated at the
island centers.

changes across the island, which is clearly seen for SP and B energies might be affected by strain and
the stable fcc and simple cubic lattices. For tensile (ii) the induced shifts can be as large as those
strain, diffusion is enhanced near the edges com- induced by classical elastic interactions.
pared to the center of the island; for compressive The inhomogeneous kinetics on an island can
strain it is slowed down. This has to be distin- be traced back to position-dependent strain relax-
guished from inhomogeneities of the energies at ation. One gets the correct qualitative picture by
binding sites, which vary only very little in our applying the energy-strain relations obtained for
systems. Such a variation is responsible for a drift the diffusion on uniformly strained surfaces locally.
of adatoms towards stronger binding sites. We This also works in more complicated geometries.
conclude that in the fcc and sc lattices we studied We have checked that one gets the correct physical
the dominant kinetic effect of strain is due to the answer, e.g. for detachment barriers, from steps. It
modification of the saddle-point energies. This is generally believed that detachment barriers are
feature cannot be captured by a strain-dependent always reduced for strained layers. The argument
chemical potential; instead, it corresponds to a is based on the correct idea that the binding energy
strain-dependent surface diffusion constant. of an adatom at the island edge is reduced because
According to Lau and Kohn I-9] the chemical attachment increases the island size and thereby
potential of an adatom, i.e. its binding energy, is the strain energy per atom stored in the island.
roughly changed by an amount of 0.1 eV by elastic However, we find that the strain dependence of
interactions with another adatom two lattice con- the saddle points may overcompensate this effect
stants apart (metallic surface). In all of our systems so that the detachment barrier may increase or
we saw mainly saddle-point changes that were at decrease depending on the sign of the strain 1-45].
least of that order of magnitude for reasonable Although in this paper we have emphasized
strains of e ~ ___0.05. However, the changes were some elementary properties, which can be argued
not due to adatom-induced displacements, i.e. elas- to be typical for diffusion on many strained sur-
tic interactions. Whereas shifts of saddle-point faces, we stress that, in general, aspects of strain
energies are certainly pronounced in purely pair- have to be analysed in a case-by-case manner,
bonded systems it was demonstrated that (i) both especially when more complicated situations like
140 M. Schroeder, D.E. Wolf~ Surface Science 375 (1997) 129-140

detachment, attachment and diffusion along or [19] Ch. Ratsch, private communication, to be published.
across edges are concerned. [20] H. Brune, K. Bromann, H. R6der, K. Kern, J. Jacobson,
P. Stoltze, K. Jacobson and J. Norskov, Phys. Rev. B 52
(1995) R14380.
[21] B.W. Dodson and P.A. Taylor, Phys. Rev. B 34
(1986) 2112.
Acknowledgements [22] J. Kew, M.R. Wilby and D.D. Vvedensky, J. Cryst.
Growth 127 (1993) 508.
We thank Pavel Smilauer, Jacques Villain and [23] H. Sitter, Thin Solid Films 267 (1995) 37.
Andy Zangwill for illuminating discussions. This [24] M. Schneider, A. Rahman and I.K. Schuller, Phys. Rev.
work was supported by Deutsche Forschungs- Lett. 55 (1985) 604.
[25] S. Das Sarma, S.M. Paik, K.E. Khor and A. Kobayashi,
gemeinschaft through SFB 166. J. Vac. Sci. Technol. B 5 (1987) 1179.
[26] P.A. Taylor and B.W. Dodson, Phys. Rev. B 36
(1987) 1335.
[27] C. Ratsch and A. Zangwill, Surf. Sci. 293 (1993) 123.
References [28] C. Ratsch, A. Zangwill and P. Smilauer, Surf. Sci. 314
(1994) L937.
[1] S. Stoyanov and D. Kashchiev, in: Current Topics in [29] B.G. Orr, D. Kessler, C.W. Snyder and L. Sander,
Material Science, Vol. 7, Ed. E. Kaldis (North-Holland, Europhys. Lett. 19 (1992) 33.
Amsterdam, 1981). [30] S.V. Ghaisas, Surf. Sci. 223 (1989) 441.
[2] J. Villain, A. Pimpinelli and D.E. Wolf, Comments [31] C. Roland and G. Gilmer, Phys. Rev. B 46 (1992) 13428.
Condens. Mater. Phys. 16 (1992) 1. [32] C. Roland and G. Gilmer, Phys. Rev. B 47 (1993) 16286.
I-3] M.C. Bartelt and J.W. Evans, Phys. Rev. B 46 (1992) [33] H. Spjut and D.A. Faux, Surf. Sci. 306 (1994) 233.
12675. [34] D.J. Srolovitz, Acta Metall. 37 (1989) 621.
[4] C. Ratsch, A. Zangwill, P. Smilauer and D.D. Vvedensky, [35] B.J. Spencer, P.W. Voorhees and S.H. Davis, Phys. Rev.
Phys. Rev. Lett. 72 (1994) 3194. Lett. 67 (1991) 3696.
[5] G.S. Bales and D.C. Chrzan, Phys. Rev. B 50 (1994) 6057. [36] B.J. Spencer, S.H. Davis and P.W. Voorhees, Phys. Rev.
[6] C. Ratsch, P. Smilauer, A. Zangwill and D.D. Vvedensky, B 47 (1993) 976.
Surf. Sci. 329 (1995) L599. [37] B.J. Spencer, P.W. Voorhees and S.H. Davis, J. Appl.
[7] M. Schroeder and D.E. Wolf, Phys. Rev. Lett. 74 Phys. 73 (1993) 4955.
(1995) 2062. [38] J. Tersoff, Y.K. Phang, Z. Zhang and M.G. Lagally, Phys.
[8] V.I. Marchenko and A.Ya. Parshin, Sov. Phys. JETP 52 Rev. Lett. 75 (1995) 2730.
(1980) 129. [39] C.-L. Liu and J.B. Adams, Surf. Sci. 294 (1993) 197.
[9] K.H. Lau and W. Kohn, Surf. Sci. 65 (1977) 607. [40] R. Stumpfand M. Schemer, Phys. Rev. Lett. 72 (1994) 254.
[10] J. Villain and A. Pimpinelli, Physique de la Croissance [41] W.K. Rilling, C.M. Gilmore, T.D. Andreadis and J.A.
Christalline, Collection A16a Saclay (l~ditions Eyrolles, Sprague, Can. J. Phys. 68 (1990) 1035.
1995). [42] One-dimensional defects are modelled by rows of vacen-
[11] R. Kern and M. Krohn, Phys. Status Solidi A 116 cies, i.e. as two stripes of different width. Hence, an
(1989) 23. island-island edge interaction +In(r), depending on the
1-12] C. Duport, P. Nozieres and J. Villain, Phys. Rev. Lett. sign of the steps, is not observable; repulsive and
74 (1995) 134. attractive contributions add up to zero.
[13] C. Duport, P. Politi and J. Villain, J. Phys. I France 5 [43] Following elasticity theory for the isotropic solid the
(1995) 1317. prefactor A is given by A ~ [(1--~2)/nY]Lm2, where ~ is
[14] C. Ratsch and A. Zangwill, Appl. Phys. Lett. 63 the Poisson ratio and Y is Young's modulus. We
(1993) 2348. estimated m ~ l [9,12] and calculated 4=0.42 and Y=
[15] C. Ratsch, M.D. Nelson and A. Zangwill, Phys. Rev. B 71, which gives A~0.1.
50 (1994) 14489. [44] Global relaxation requires us to ignore the force acting
[16] C. Ratsch, P. Smilauer, D.D. Vvedensky and A. Zangwill, in parallel to the diffusion direction, otherwise the whole
J. Phys. I France 6 (1996) 575. crystal would move, i.e. the adatom would be relaxed
[ 17] D. Kandel and E. Kaxiras, Phys. Rev. Lett. 75 (1995) 2742. into a binding site again. We used a frozen bottom layer
[18] J.A. Meyer, P. Schmidt and R.J. Behm, Phys. Rev. Lett. to prevent the whole system from moving.
74 (1995) 3864. [45] M. Schroeder and D.E. Wolf, in preparation.

You might also like