1: \documentstyle[prl,aps,twocolumn]{revtex}
2: %\documentstyle[preprint,aps]{revtex}
3: \input epsf
4: \begin{document}
5:
6: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname@twocolumnfalse\endcsname
7:
8: \title{Competing roughening mechanisms in strained heteroepitaxy: a
9: fast kinetic Monte Carlo study
10: }
11:
12: \author{Chi-Hang Lam$^{1}$, Chun-Kin Lee$^1$, and Leonard M. Sander$^2$}
13:
14: \address{
15: $^1$Department of Applied Physics, Hong Kong Polytechnic University,
16: Hung Hom, Hong Kong\\ $^2$Michigan Center for Theoretical Physics,
17: Department of Physics, Randall Laboratory,
18: University of Michigan, Ann Arbor, MI 48109-1120, USA\\ }
19: \date{\today}
20: \maketitle
21:
22: \begin{abstract}
23: We study the morphological evolution of strained heteroepitaxial
24: films using kinetic Monte Carlo simulations in two dimensions. A
25: novel Green's function approach, analogous to boundary integral
26: methods, is used to calculate elastic energies efficiently. We
27: observe island formation at low lattice misfit and high temperature
28: that is consistent with the Asaro-Tiller-Grinfeld instability
29: theory. At high misfit and low temperature, islands or pits form
30: according to the nucleation theory of Tersoff and LeGoues.
31: \end{abstract}
32:
33: \pacs{PACS numbers: 68.65.-k, 68.65.Hb, 81.16.Dn, 81.16.Rf}
34:
35: ]
36:
37: Coherent three-dimensional (3D) islands in strained heteroepitaxial films
38: are of great interest because they can self-assemble as quantum dots
39: for possible advanced
40: optoelectronic applications \cite{Shchukin,Politi}. They are observed
41: in a variety of film-substrate combinations including
42: Ge/Si, InAs/GaAs, InAs/InP, etc. In these systems, island formation
43: follows the Stranski-Krastanov mode. Initially, two-dimensional (2D)
44: layer-by-layer growth leads to a flat wetting layer under stress.
45: Beyond a threshold film thickness, 3D islands emerge
46: on top of the wetting layer partially relieving the stress.
47: %
48: The precise island formation mechanism is currently under intensive
49: debate. According to the nucleation theory of Tersoff and
50: LeGoues, the growth of stable islands requires overcoming an energy
51: barrier associated with a critical island size \cite{Tersoff_LeGoues}.
52: However, experiments reveal gradual development of ripples
53: \cite{Floro} or pre-pyramids \cite{Vailionis} at the initial stage of
54: island formation. These observations are more consistent with the
55: Asaro-Tiller-Grinfeld (ATG) linear instability theory, which predicts
56: that morphological perturbations at sufficiently long wavelengths grow
57: spontaneously and steadily \cite{Asaro,Srolovitz}. It was originally
58: proposed for smooth surfaces, but extensions to faceted ones based on
59: non-equilibrium deposition conditions \cite{Tersoff01} or finite
60: vicinality of substrates \cite{Eisenberg} have been suggested.
61:
62: To better understand the roughening mechanism of strained layers, we
63: have performed kinetic Monte Carlo simulations using an atomistic
64: model \cite{Orr,Barabasi,Khor,Meixner}. This approach is
65: computationally very intensive but can reliably account for both
66: lattice discreteness and non-equilibrium conditions. Previous
67: simulations successfully demonstrated island formation in strained
68: layers but only via the nucleation mechanism
69: \cite{Orr,Barabasi,Khor}. In this work we introduce significantly more
70: efficient algorithms. Thus we can explore a much wider range of conditions and
71: observe a rich variety of morphologies in better general agreement
72: with experiments. In particular, by lowering the lattice misfit and
73: raising the temperature, the roughening mechanism crosses over from
74: nucleation to instability controlled.
75:
76: We adopt the 2D ball and spring model of heteroepitaxy
77: defined on a square lattice first studied by Orr et al \cite{Orr} and
78: subsequently by Barab\'{a}si \cite{Barabasi} and Khor and Das Sarma
79: \cite{Khor}. Simulations in 3D limited to submonolayer coverage were
80: performed by Meixner et al \cite{Meixner}. Our model parameters are
81: appropriate to
82: the widely studied Si$_{1-x}$Ge$_x$/Si system. We assume a substrate
83: lattice constant $a_s=2.715$\AA ~ so that $a_s^3$ gives the correct
84: atomic volume in crystalline silicon. The lattice constant $a_f$ of
85: the film material is related to the lattice misfit
86: $\epsilon=(a_f-a_s)/a_f$ which has a compositional dependence
87: $\epsilon=0.04 x$.
88: %
89: Nearest and next nearest neighboring atoms are directly connected by
90: elastic springs with force constants $k_N=13.85eV/a_s^2$ and
91: $k_{NN}=k_N/2$ respectively. This choice gives the correct modulus
92: $c_{11}$ of silicon and a shear modulus constant along tangential and
93: diagonal directions, despite a slight anisotropy in the Young's
94: modulus. The elastic couplings of adatoms with the rest of the system
95: are weak and are completely neglected for better computational
96: efficiency. Solid-on-solid conditions and atomic steps limited to at
97: most two atoms high are assumed. Every topmost atom in the film can
98: hop to a random topmost site $s$ columns away where $s=\pm 1$, $\pm
99: 2$, ... or $\pm s_{max}$ with equal probability.
100: %
101: Previous simulations allowed only nearest neighbor hopping (i.e.
102: $s_{max}=1$) \cite{Orr,Barabasi,Khor,Meixner}. To speed up the
103: simulations, we put $s_{max}=8$ or 20 respectively for $x > 0.6$ or $x
104: \le 0.6$. These hopping ranges are much shorter than the dimensions
105: of the relevant structures (islands or pits) on the films
106: %
107: and we have checked that decreasing $s_{max}$ does not alter our
108: results. The hopping rate $\Gamma_m$ of a topmost atom $m$ follows an
109: Arrhenius form:
110: \begin{equation}
111: \label{rate}
112: \Gamma_m =
113: {R_0}\exp \left[ -\frac{n_m \gamma
114: - \Delta E_m - E_0}{k_{B}T}\right]
115: \end{equation}
116: Here, $n_m$ is the number of nearest and next nearest neighbors of
117: atom $m$. We choose a bond strength $\gamma=0.4eV$ which will be
118: explained later. The energy $\Delta E_m$ is the difference in the
119: strain energy $E_s$ of the whole lattice at mechanical equilibrium
120: when the site is occupied versus unoccupied. Finally, we put
121: %$E_0=-(0.67eV-3\gamma)$
122: $E_0=0.53$eV and $R_0=2D_0/(\sigma_s a_s)^2$ with $D_0=5.2\times
123: 10^{13}\mbox{\AA}^2 s^{-1}$ and
124: $\sigma_s^2 = \frac{1}{6}(s_{max}+1)(2s_{max}+1)$.
125: This gives the appropriate adatom diffusion coefficient for silicon (100)
126: \cite{Savage}. Our model follows detailed balance.
127:
128: The simulations involve intensive computations resulting solely from
129: the long-range nature of elastic interactions. Practically all the CPU
130: time is spent on the repeated calculations of $E_s$ which is needed to
131: find $\Delta E_m$ and hence $\Gamma_m$ in Eq. (\ref{rate}). The
132: elastic problem is formulated as follows. First, a flat film is
133: homogeneously strained \cite{Politi}. This provides a convenient
134: reference position with displacement $\vec{u}_i=0$ for every atom $i$.
135: In general, the elastic force on atom $i$ by a directly connected
136: neighbor $j$ is $\vec{f}_{ij} = - {\bf K}_{ij} ( \vec{u}_i -
137: \vec{u}_j ) + \vec{b}_{ij} $ after linearization where the $2\times 2$
138: symmetric matrix ${\bf K}_{ij}=k_{ij} \hat{n}_{ij}
139: \hat{n}_{ij}^t $ is the modulus tensor and
140: $\vec{b}_{ij} = (l^0_{ij}-l_{ij}) {\bf K}_{ij} \hat{n}_{ij}$
141: arises from the homogeneous stress in flat films. The spring constant
142: $k_{ij}$ equals either $k_N$ or $k_{NN}$ for tangential or diagonal
143: connection respectively. The unit column vector $\hat{n}_{ij}$ points
144: from the unstrained lattice position of atom $j$ towards that of atom
145: $i$ and $t$ denotes transpose. Furthermore, $l^0_{ij}$ and $l_{ij}$
146: are respectively the natural and homogeneously strained spring lengths
147: which follow easily from $a_s$ and $\epsilon$. Mechanical
148: equilibrium requires $\sum_j \vec{f}_{ij}=0$ for each atom $i$. This
149: leads to a large set of equations coupling the $\vec{u}_i$ of $all$
150: of the atoms.
151: The solution then gives the elastic energy stored in every spring and hence
152: $E_s$.
153:
154: We now introduce a Green's function approach for calculating $E_s$
155: efficiently requiring the explicit consideration of $only$ the surface
156: atoms. It is a lattice analogue of boundary integral methods and is
157: superior to boundary element techniques for our intrinsically discrete
158: problem. We first derive the exact formalism. Figure 1 shows an
159: example of a small lattice of atoms (solid circles). As a
160: mathematical construct, we extend the lattice by adding $ghost$ atoms
161: (open circles) with similar elastic properties. Unphysical couplings
162: are hence introduced but can be exactly cancelled by applying
163: {\it external} forces $\vec{f}^e_j$ and $\vec{f}^e_{j'}$ to every real
164: surface atom $j$ and ghost surface atom $j'$ respectively with
165: \begin{eqnarray}
166: \label{freal2}
167: \vec{f}^e_j &=& \sum_{j'}({\bf K}_{jj'}\vec{u}_j -\vec{b}_{jj'})\\
168: \label{fghost2}
169: \vec{f}^e_{j'} &=& - \sum_{j} {\bf K}_{jj'} \vec{u}_j
170: \end{eqnarray}
171: The summation in Eq. (\ref{freal2}) is over each ghost atom $j'$
172: connected directly to the real atom $j$ and it is analogous in
173: Eq. (\ref{fghost2}).
174: It is easy to see that the real atoms are then exactly decoupled from
175: the ghost atoms which are held precisely at their homogeneously strained
176: positions.
177:
178: \begin{figure}[htp]
179: %{\centering \epsfxsize 0.65\columnwidth \epsfbox{exforce.eps} ~\\}
180: \centerline{\epsfxsize 0.65\columnwidth \epsfbox{exforce.eps}}
181: \vskip 2ex
182: \caption{
183: \label{fig:exforce1}
184: Ghost atoms (open circles) are added on top of the real atoms
185: (solid circles) forming an extended lattice. Unphysical couplings
186: are exactly cancelled by external forces (arrows) applied to
187: the real and ghost surface atoms.
188: }
189: \end{figure}
190:
191: A lattice Green's function is then applied to express the displacement,
192: $\vec{u}_i$, of every real surface atom $i$ under the influence of the
193: external forces:
194: \begin{equation}
195: \label{uife}
196: \vec{u}_i = \sum_j {\bf G}_{ij} \vec{f}^e_j
197: + \sum_{j'} {\bf G}_{ij'} \vec{f}_{j'}^e
198: \end{equation}
199: Note that the Green's function ${\bf G}$ is
200: defined for the extended lattice and is {\it independent of the film
201: morphology.} It can thus be computed numerically once prior to the start
202: of the simulation. It should not be confused with the
203: half-plane Green's function which provides simpler but only
204: approximate results \cite{Tersoff_LeGoues,Meixner}. Combining Eqs.
205: (\ref{freal2}-\ref{uife}), we arrive at the reduced set of equations
206: \begin{equation}
207: \label{ui_new}
208: \vec{u}_i =
209: \sum_{jj'}
210: [({\bf G}_{ij} - {\bf G}_{ij'})
211: {\bf K}_{jj'} \vec{u}_j - {\bf G}_{ij} \vec{b}_{jj'}]
212: \end{equation}
213: coupling only the real surface atoms where the sum is over all pairs
214: of directly connected real and ghost surface atoms $j$ and $j'$
215: respectively. The solution of Eq. (\ref{ui_new}) gives $\vec{u}_i$.
216:
217: The elastic energy $E_s$ can then be calculated directly from
218: \begin{equation}
219: \label{Es}
220: E_s = E_s^{0} - \frac{1}{2}\sum_{jj'} \vec{b}_{jj'} \cdot \vec{u}_j
221: \end{equation}
222: which is derived from a simple consideration of virtual work. Here, the sum is
223: defined similarly to that in Eq. (\ref{ui_new}) and $E_s^{0}$ equals
224: the unrelaxed value of $E_s$ when $\vec{u}_i\equiv 0$ which can be
225: straightforwardly computed. The method summarized in Eqs.
226: (\ref{ui_new}) and (\ref{Es}) is exact and practical for
227: simulations at moderate scales.
228:
229: We can go further and use a coarse-grained version of our Green's function
230: approach for a further boost on the computational efficiency. Finding
231: the strain energy $\Delta E_m$ of the atom $m$ needed in Eq.
232: (\ref{rate}) requires calculating the strain energy $E_s$ of the whole
233: lattice twice with and without the atom $m$. Certain fine details of
234: the surface far away are obviously unimportant and can be neglected.
235: Specifically, surface atoms are grouped into sets with the $I$th of
236: them denoted by $\Omega_I$. We neglect fluctuations within a set by
237: assuming identical displacement $\vec{u}_i\equiv \vec{u}_I$ for each
238: member $i\in \Omega_I$. Equation (\ref{ui_new}) is then
239: approximated as
240: \begin{equation}
241: \label{ui_coarsened}
242: \vec{u}_{I}
243: = \sum_J \left[
244: \sum_{j\in \Omega_J,j'}
245: ({\bf G}_{Ij} - {\bf G}_{Ij'})
246: {\bf K}_{jj'}
247: \right] \vec{u}_J
248: - \sum_{jj'} {\bf G}_{Ij} \vec{b}_{jj'}
249: \end{equation}
250: where $G_{Ij}=G_{ij}$ with the lattice point $i$ at the centroid of
251: the set $\Omega_I$. Every atom within 3 columns from the atom $m$ is
252: not coarsened and constitutes their own single-membered set. Farther
253: away at $r$ columns from the atom $m$, sets contain atoms in
254: neighborhoods of $2r/3+1$ columns wide, a form motivated by simple
255: error analysis. We have checked numerically that a smaller degree of
256: coarsening leads to no noticeable difference in our results.
257:
258: Surface diffusion can be simulated using the hopping rates in Eq.
259: (\ref{rate}) as $\Delta E_m$ is now readily computable. We adopt an
260: acceptance-rejection algorithm aided by tabulated values of $\Delta
261: E_m$ for $5^8$ sample surface configurations. Details will be
262: explained elsewhere. In our main simulations, the lattice is of 1024
263: atoms wide following periodic boundary conditions. The substrate is
264: 1024 monolayers (ML) thick while the extended lattice on which we
265: compute the Green's function includes also a film of 80ML. Fixed
266: boundary conditions are applied to the top and bottom layers of the
267: extended lattice.
268:
269: We first simulate deposition of pure Ge film (i.e. $x=1$) with misfit
270: $\epsilon=4$\% at temperature $T=600K$. At very high deposition rate
271: $R=80$MLs$^{-1}$ [Fig. 2(a)], we observe layer-by-layer growth. At
272: slower deposition rate $R=8$MLs$^{-1}$ [Fig. 2(b)], the film is
273: initially flat but pits then develop. A detailed examination of the
274: morphological evolution indicates that the pits appear rather
275: independently and suddenly. Once created, they are immediately
276: bounded by side-walls at an energetically favorable 45$^\circ$
277: inclination. These features strongly support the nucleation mechanism
278: for their formation, noting that pits are energetically more favorable
279: than islands \cite{Tersoff_LeGoues}. This pit nucleation process is
280: similar to that in Fig. 1(d) of Ref. \cite{Orr}. At
281: $R=0.8$MLs$^{-1}$ [Fig. 2(c)], islands with 45$^\circ$ side-walls
282: nucleate at very early stage before the film is sufficiently thick
283: for pit formation. The result is analogous to those in Fig 1(a) of
284: Ref. \cite{Orr} and also Refs. \cite{Barabasi,Khor}. Further
285: decreasing $R$ towards realistic values of order 0.01MLs$^{-1}$ leads
286: to similar but more widely separated islands.
287:
288: \begin{figure}
289: %{\centering \epsfxsize 0.95\columnwidth \epsfbox{graph.depos600.eps} ~\\}
290: \centerline{\epsfxsize 0.95\columnwidth \epsfbox{graph.depos600.eps}}
291: \vskip 2ex
292: \caption{
293: Simulations of deposition of Ge films at $T=600K$ with substrate
294: width and thickness both equal 1024$a_s \simeq$ 2780\AA. The axes
295: are in unit of $a_s$}
296: \end{figure}
297:
298: Figure 3 shows results for deposition at misfit $\epsilon=2$\% with
299: $x=0.5$ at $T=1000K$. Depending on $R$, we observe analogous
300: layer-by-layer growth [Fig. 3(a)], layer-by-layer growth followed by
301: roughening [Fig. 3(b)], and island growth [Fig. 3(c)]. However, the
302: islands in Figs. 3(b)-(c) emerge gradually from ripple-like
303: perturbations with local surface inclinations increasing steadily and
304: relatively synchronously in agreement with experiments
305: \cite{Floro,Tromp,Sutter} and ATG instability theory
306: \cite{Asaro,Srolovitz}. This regime has not been reported in previous
307: atomistic simulations mainly due to inaccuracies in accounting for the
308: long-range parts of the elastic interactions or the rather thin
309: substrates used \cite{Orr,Barabasi,Khor,Meixner}. Instead, it was
310: observed in continuum simulations \cite{Yang}, which however cannot
311: realize the nucleation mechanism.
312:
313: \begin{figure}
314: %{\centering \epsfxsize 0.95\columnwidth \epsfbox{graph.depos1000.eps} ~\\}
315: \centerline{\epsfxsize 0.95\columnwidth \epsfbox{graph.depos1000.eps}}
316: \vskip 2ex
317: \caption{
318: Simulations of deposition of Si$_{0.5}$Ge$_{0.5}$ films at $T=1000K$.
319: }
320: \end{figure}
321:
322: The importance of the lattice misfit in deciding the roughening
323: mechanism is particularly easy to understand. The nucleation of
324: islands or pits occurs at a rate $R_{nucl} \sim \exp(- c \epsilon^{-4})$
325: with $c$ being a constant \cite{Tersoff_LeGoues} and becomes
326: very slow at small $\epsilon$ \cite{Sutter}. The ATG
327: instability with a roughening rate $R_{inst} \sim \epsilon^{8}$
328: \cite{Srolovitz} then dominates.
329: %
330: For Figs. 2(b) and 3(b), we have chosen deposition rates close to
331: the relevant roughening rates. We then observe kinetically limited
332: wetting layers prior to roughening which is characteristic of
333: Stranski-Krastanov growth. However, the threshold thickness depends
334: strongly on $R$ contrary to experimental findings \cite{Politi}. A
335: more realistic model in the future should include
336: other mechanisms such as film-substrate
337: interactions \cite{Tersoff91} or nonlinear elasticity \cite{Eisenberg}
338: which have been argued to give a more stable wetting layer.
339:
340: We have also simulated annealing
341: of initially flat films of 30ML at
342: $T=1000K$.
343: At this high temperature, roughening is mainly due to the ATG
344: instability, and we observe the development of ripples followed by
345: islands. Figure 4 shows the surface profiles after the islands are
346: fully developed. The cusps on the surfaces \cite{Gao} are limited by
347: either the substrate or the local step height limit imposed in our
348: model. We have measured the island size $l$ from power spectra
349: of 11 realizations of similar surfaces. We obtain $l \sim x
350: ^{-1.8}$ in reasonable agreement with $x^{-2}$ from the instability
351: theory \cite{Asaro} but slightly different from $x^{-1}$ from
352: experiments \cite{Tromp,Sutter}. The discrepancy with experiments may
353: be improved if the compositional dependence of film properties such as
354: bond energies and diffusion coefficients are properly considered.
355: The island sizes in general lie within the experimental range due to
356: our choice of the bond strength $\gamma=0.4eV$, which in fact is a
357: reasonable value.
358:
359: \begin{figure}
360: %{\centering \epsfxsize 0.95\columnwidth \epsfbox{graph.depos0.eps} ~\\}
361: \centerline{\epsfxsize 0.95\columnwidth \epsfbox{graph.depos0.eps}}
362: \vskip 2ex
363: \caption{
364: Simulations of annealing of initially flat Si$_{1-x}$Ge$_{x}$ films
365: of 30ML at $T=1000K$ for a period of time $t$. }
366: \end{figure}
367:
368: We have already discussed possible extensions of the model to include
369: film-substrate chemical interactions, nonlinear elasticity, and
370: composition-dependent material properties. To further enhance the
371: morphological resemblance with experiments, one can explore more
372: sophisticated forms of bond energies favoring 2D analogs of (105) and
373: (113) facets \cite{Floro}.
374:
375: We should note that the validity of ATG instability
376: theory in 3D is not clear due to a singular form of the equilibrium
377: surface energy in the presence of facets \cite{Politi}. Critical tests of the
378: theory by 3D simulations are important, and may be feasible using our
379: method.
380: %
381: For example, our 2D simulation leading to Fig. 2(b) involves about
382: $6\times 10^7$ hops
383: %and about $3\times 10^6$ elastic energy calculations
384: and ran for 18 hours on a pentium 2GHz computer.
385: Thus, extensions to 3D should be practical.
386:
387: In conclusion, using accelerated algorithms which properly and
388: efficiently account for long-range elastic interactions, we have
389: simulated deposition and annealing of strained heteroepitaxial layers
390: in 2D.
391: %
392: At low misfit
393: and high temperature, we observe ripples and subsequently gradual
394: island formation in consistent with the ATG instability theory. At
395: high misfit and low temperature, islands or pits are generated via the
396: nucleation pathway.
397: These suggest a non-trivial competition between roughening
398: mechanisms, although reliable quantitative determination of the
399: crossover conditions is beyond the scope of our model.
400: %
401: The ATG instability is the most promising description of island
402: formation in Si$_{1-x}$Ge$_{x}$ films at low, and probably also at
403: high lattice misfit \cite{Floro,Vailionis}. However, the nucleation
404: mechanism applied to high misfit regimes in certain experimental
405: situations has not been ruled out \cite{Kastner}. Thus the
406: competition of mechanisms can be important for interpreting
407: experimental results.
408: %
409: In our simulations, for deposition rates close to the relevant
410: roughening rate, kinetically limited wetting layers develop prior to
411: roughening. At lower but more realistic deposition rates, islands
412: form at an early stage and are more widely separated. This may
413: be related to the great variation in how closely the islands are
414: packed under various conditions in experiments
415: \cite{Floro,Vailionis}.
416:
417: We thank B.G. Orr for helpful comments. CHL is supported by PolyU
418: grant no. 5309/01P.
419:
420: \begin{references}
421:
422: \bibitem{Shchukin}
423: % Spontaneous ordering of nanostructures on crystal surfaces
424: V.A. Shchukin and D. Bimberg,
425: Rev. Mod. Phys. {\bf 71}, 1125 (1999).
426:
427: \bibitem{Politi}
428: P. Politi, G. Grenet, A. Marty, A. Ponchet, and J. Villain,
429: Phys. Rep. {\bf 324}, 271 (2000).
430:
431: %\bibitem{Mo}
432: %Kinetic Pathway in Stranski-Krastanov Growth of Ge on Si(001)
433: %Y.-W. Mo, D.E. Savage, B.S. Swartzentruber, and M.G. Lagally,
434: %Phys. Rev. Lett. {\bf 65}, 1020 (1990).
435:
436: %\bibitem{Pimpinelli}
437: %A. Pimpinelli and J. Villain, Physics of crystal growth (Cambridge
438: %University Press, 1998).
439:
440: \bibitem{Tersoff_LeGoues}
441: % Competing Relaxation mechanisms in strained layers
442: J. Tersoff and F.K. LeGoues, Phys. Rev. Lett. {\bf 72}, 3570 (1994).
443:
444: \bibitem{Floro}
445: %Evolution of coherent islands in Si 1-x Ge x / Si(001)
446: J.A. Floro et al,
447: % E. Chason, L.B. Freund, R.D. Twesten, R.Q. Hwang, and G.A. Lucadamo,
448: Phys. Rev. B {\bf 59}, 1990 (1999).
449:
450: \bibitem{Vailionis}
451: % Pathway for the Strain-Driven Two-Dimensional to Three-Dimensional
452: % Transition during Growth of Ge on Si(001)
453: A. Vailionis et al,
454: % B. Cho, G. Glass, P. Desjardins, David G. Cahill, and J. E. Greene,
455: Phys. Rev. Lett. {\bf 85}, 3672 (2000).
456:
457: \bibitem{Asaro}
458: R.J. Asaro and W.A. Tiller, Metall. Trans {\bf 3}, 1789 (1972);
459: M.A. Grinfeld, J. Nonlinear Sci. {\bf 3}, 35 (1993).
460:
461: \bibitem{Srolovitz}
462: % On the stability of surfaces of stressed solids
463: D.J. Srolovitz, Acta Metall. {\bf 37}, 621 (1989);
464: %\bibitem{Spencer93}
465: %Morphological instability in epitaxially strained dislocation-free solid films:
466: %linear stability theory
467: B. J. Spencer, P.W. Voorhees, and S.H. Davis, J. Appl. Phys. {\bf 73},
468: 4955 (1993).
469:
470:
471: \bibitem{Tersoff01}
472: % Facet Growth under Stress: The limits of strained-layer stability
473: J. Tersoff,
474: Phys. Rev. Lett. {\bf 87}, 156101 (2001).
475:
476: \bibitem{Eisenberg}
477: % Wetting layer thickness and early evolution of epitaxially strained think films
478: %H.R. Eisenberg and D. Kandel, Phys. Rev. Lett. {\bf 85}, 1286 (2000).
479: % Thge origin and properties of the wetting layer and early evolution
480: %of epitaxially strained thin films
481: H.R. Eisenberg and D. Kandel, cond-mat 0201238 (2002).
482:
483: \bibitem{Orr}
484: % A Model for strain-induced roughening and coherent island growth
485: B.G. Orr, D.A. Kessler, C.W. Snyder, and L.M. Sander,
486: Euro. Phys. Lett., {\bf 19}, 33 (1992).
487:
488: \bibitem{Barabasi}
489: %Self-assembled island formation in herteropeitaxial growth
490: A.-L. Barab\'{a}si, Appl. Phys. Lett {\bf 70}, 2565 (1997).
491:
492: \bibitem{Khor}
493: K.E. Khor and S. Das Sarma, Phys. Rev. B {\bf 62}, 16657 (2000).
494:
495: %\bibitem{Ratsch}
496: % Mechanism for coherent island formation during heteroepitaxy
497: %C. Ratsch, P. Smilauer, D.D. Vvedensky, and A. Zangwill,
498: %J. Phys. I France {\bf b}, 575 (1996).
499:
500: \bibitem{Meixner}
501: %Self-assembled quantum dots: crossover from kinetically controlled to thermodynamically limited growth
502: M. Meixner et al, %E. Sch\"{o}ll, V.A. Shchukin, and D. Bimberg,
503: Phys. Rev. Lett. {\bf 87}, 236101 (2001)
504:
505: \bibitem{Savage}
506: D.E. Savage et al,
507: % in Germanium Silicon: Physics and materials,
508: in Semiconductors and Semimetals {\bf 56}, R. Hull and J.C. Bean
509: Ed. (Academic Press 1999).
510:
511: \bibitem{Tromp}
512: % Instability-driven SiGe island growth
513: R.M. Tromp, F.M. Ross, and M.C. Reuter,
514: Phys. Rev. Lett. {\bf 84} 4641 (2000).
515:
516: \bibitem{Sutter}
517: %Nucleationless three-dimensional island formation in low-misfit heteroepitaxy
518: P. Sutter and M.G. Lagally,
519: Phys. Rev. Lett. {\bf 84}, 4637 (2000).
520:
521: \bibitem{Yang}
522: %Cracklike surface instabilities in stresses solids
523: W.H. Yang and D.J. Srolovitz,
524: Phys. Rev. Lett. {\bf 71}, 1593 (1993);
525: %\bibitem{Muller}
526: %model of surface instabilities induced by stress
527: %(Phase field method)
528: J. M\"{u}ller and M. Grant,
529: Phys. Rev. Lett. {\bf 82}, 1736 (1999).
530:
531: \bibitem{Tersoff91}
532: % Stress-induced layer-by-layer growth of Ge on Si(100)
533: J. Tersoff, Phys. Rev. B {\bf 43}, 9377 (1991).
534:
535: \bibitem{Gao}
536: %Surface roughening of heteroepitaxial thin films
537: H. Gao and N.D. William, Annu. Rev. mater. Sci. {\bf 29} 173 (1999).
538:
539: \bibitem{Kastner}
540: %Kinetically self-limiting growth of Ge Islands on Si(001)
541: M. K\"{a}stner and B. Voigl\"{a}nder, Phys. Rev. Lett. 2745 (1999).
542:
543: \end{references}
544:
545:
546:
547: \end{document}
548: