1: %% LyX 1.4.3 created this file. For more info, see http://www.lyx.org/.
2: %% Do not edit unless you really know what you are doing.
3: \documentclass[english,prb, manuscript]{revtex4}
4: \usepackage{times}
5: \usepackage[T1]{fontenc}
6: \usepackage[latin1]{inputenc}
7: \usepackage{verbatim}
8: \usepackage{amsmath}
9: \usepackage{color}
10: \usepackage{graphicx}
11:
12: \makeatletter
13:
14: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LyX specific LaTeX commands.
15: \newcommand{\noun}[1]{\textsc{#1}}
16: %% The greyedout annotation environment
17: \newenvironment{lyxgreyedout}{\textcolor[gray]{0.8}\bgroup}{\egroup}
18:
19: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% User specified LaTeX commands.
20: %\usepackage{cite}
21: %\usepackage{hyperref}
22:
23: \usepackage{babel}
24: \makeatother
25: \begin{document}
26:
27:
28: \newcommand{\sgrad}{\nabla_{S}}
29:
30:
31: \newcommand{\degree}{^{\circ}}
32:
33:
34: \newcommand{\dif}[1]{\vec{#1}-\vec{#1}'}
35:
36:
37: \newcommand{\dd}[2]{\delta^{#2}\left(#1-#1'\right)}
38:
39:
40: \newcommand{\vd}[2]{\delta^{#2}\left(\vec{#1}-\vec{#1'}\right)}
41:
42:
43: \newcommand{\ds}[1]{\delta\left(#1-#1'\right)}
44:
45:
46: \newcommand{\grad}{\vec{\nabla}}
47:
48:
49: \newcommand{\phik}[3]{e^{#3i#1\cdot#2}}
50:
51:
52: \newcommand{\intn}[4]{\int_{#1}^{#2}d^{n}\vec{#3}\left(#4\right)}
53:
54:
55: \newcommand{\height}{\mathcal{H}}
56:
57:
58: \newcommand{\h}{\height}
59:
60:
61: \renewcommand{\vec}{\mathbf}
62:
63:
64:
65:
66: \title{Anisotropy and Order of Epitaxial Self-Assembled Quantum Dots}
67:
68:
69: \author{Lawrence H. Friedman}
70:
71:
72: \affiliation{Dept. of Engineering Science and Mechanics, Pennsylvania State University,
73: 212 Earth and Engineering Science Building, University Park, Pennsylvania
74: 16802}
75:
76:
77: \date{\today}
78:
79: \begin{abstract}
80: Epitaxial self-assembled quantum dots (SAQDs) represent an important
81: step in the advancement of semiconductor fabrication at the nanoscale
82: that will allow breakthroughs in electronics and optoelectronics.
83: In these applications, order is a key factor. Here, the role of crystal
84: anisotropy in promoting order during early stages of SAQD formation
85: is studied through a linear analysis of a commonly used surface evolution
86: model. Elastic anisotropy is used a specific example. It is found
87: that there are two relevant and predictable correlation lengths. One
88: of them is related to crystal anisotropy and is crucial for determining
89: SAQD order. Furthermore, if a wetting potential is included in the
90: model, it is found that SAQD order is enhanced when the deposited
91: film is allowed to evolve at heights near the critical surface height
92: for three-dimensional film growth.
93: \end{abstract}
94:
95: \pacs{81.07.Ta, 81.15.Aa, 81.16.Dn, 68.65.Hb}
96:
97: \maketitle
98: Epitaxial self-assembled quantum dots (SAQDs) represent an important
99: step in the advancement of semiconductor fabrication at the nanoscale
100: that will allow breakthroughs in electronics and optoelectronics.
101: SAQDs are the result of linearly unstable film growth in strained
102: heteroepitaxial systems such as $\mbox{Si}_{x}$$\mbox{Ge}_{1-x}$/Si
103: and $\mbox{In}_{x}\mbox{Ga}_{1-x}\mbox{As}$/GaAs and other systems.
104: SAQDs have great potential for electronic and optoelectronic applications.
105: In these applications, order is a key factor. There are two types
106: of order, spatial and size. Spatial order refers to the regularity
107: of SAQD dot placement, and it is necessary for nano-circuitry applications.
108: Size order refers to the uniformity of SAQD size which determines
109: the voltage and/or energy level quantization of SAQDs. It has been
110: observed that crystal anisotropy can have a beneficial effect on SAQD
111: order.~\cite{Liu:2003kx} Here, the role of crystal anisotropy in
112: promoting order during early stages of SAQD formation is studied through
113: a linear analysis of a commonly used surface evolution model.~\cite{Spencer:1993vt,Liu:2003kx,Zhang:2003tg,Tekalign:2004jh,Friedman:2006bc}
114:
115: The linear analysis addresses the initial stages of SAQD formation
116: when the nominally flat film surface becomes unstable and transitions
117: to three-dimensional growth. This early stage of SAQD growth determines
118: the initial seeds of order or disorder in an SAQD array, and can be
119: analyzed analytically. A dispersion relation as in~\cite{Spencer:1993vt,Golovin:2003ms}
120: is used, and two predictable correlation lengths that grow as the
121: square-root of time (Eqs.~\ref{eq:lcor} and~\ref{eq:ancor}) are
122: found. One of the correlation lengths results from crystal anisotropy.
123: This length plays a limiting role in the initial order of SAQD arrays,
124: and it is shown that anisotropy is crucial for creating a lattice-like
125: structure that is most technologically useful. This method of analysis
126: can be extended for use in any {}``nucleationless'' model of SAQD
127: growth, although here, the specific instance of elastic anisotropy
128: is treated as elastic anisotropy is the most easily estimated. Anisotropy
129: of surface energy may also effect SAQD order, and a similar analysis
130: would result.
131:
132: At later stages when surface fluctuations are large, non-linear dynamics
133: come into play. At this stage, there is a natural tendency of SAQDS
134: to either order or ripen.~\cite{Golovin:2003ms,Liu:2003kx} Ripening
135: systems will tend to have increased disorder as time progresses, while
136: ordering systems will tend to order only slightly due to critical
137: slowing down.~\cite{Wang:2004dd} Also, it is important to note that
138: while an ordering system might be {}``ordered'' when compared to
139: other non-linear phenomena such as convection roles, etc.~\cite{Cross:1993ti},
140: the requirements for technological application are much more stringent.
141: In a numerical investigation, it was found that SAQDs have enhanced
142: order due to crystal anisotropy, but soon become disordered from ripening.~\cite{Liu:2003kx}
143: In any case, an understanding of the order during the initial stages
144: of SAQD growth is essential to further investigation of the final
145: SAQD array order.
146:
147: As in~\cite{Zhang:2003tg,Golovin:2003ms,Liu:2003kx,Spencer:1993vt},
148: a wetting energy is included in the analysis. The wetting potential
149: ensures that growth takes place in the Stranski-Krastanow mode: a
150: 3D unstable growth occurs only after a critical layer thickness is
151: achieved, and a residual wetting layer persists. Although somewhat
152: controversial, the physical origins and consequences of the wetting
153: potential are discussed in~\cite{Beck:2004yq,Spencer:1993vt}. The
154: analysis presented here is quite general, and one can exclude or neglect
155: the effect of the wetting potential by simply setting it to zero.
156: That said, if the wetting potential is real, the present analysis
157: shows that it beneficial to SAQD order to grow near the critical layer
158: thickness.
159:
160: The remainder of this report is organized as follows. First, stochastic
161: initial conditions are discussed. Second, the evolution of a single
162: mode for the isotropic case is discussed. Third, the resulting correlation
163: functions and correlation lengths are derived for the isotropic case.
164: Fourth, the analysis is repeated for the anisotropic case using elastic
165: anisotropy as an example. Finally, a representative numerical example
166: is presented using parameters appropriate for Ge dots grown on Si.
167:
168: To analyze resulting SAQD order, the mathematical model must include
169: stochastic effects. For simplicity, stochastic initial conditions
170: with deterministic time evolution are used. This method of analysis
171: yields a correlation function that is used to characterize SAQD order.
172:
173: In this model, the film height $\h$ is a function of lateral position
174: $\vec{x}$ and time $t$. The film height is treated as an average
175: film height ($\bar{\h}$) with surface fluctuations $h(\vec{x},t)$
176: , \begin{equation}
177: \mathcal{H}=\bar{\mathcal{H}}+h(\vec{x},t).\label{eq:total_height}\end{equation}
178: In this way, $\bar{\h}$ functions as a control parameter\cite{Cross:1993ti}
179: physically signifying the amount of available material per unit area
180: to form SAQDs, and $h(\vec{x},t)$ evolves via surface diffusion giving
181: the the resulting surface profile. Order is then analyzed using using
182: the spatial correlation function, $\left\langle h(\vec{x},t)h(\vec{0},t)\right\rangle $
183: and the corresponding spectrum function $\left\langle h_{\vec{k}}(t)h_{\vec{k}'}(t)^{*}\right\rangle $.
184:
185: An initially flat surface is in unstable equilibrium, and it is necessary
186: to perturb it to produce SAQDs. Therefore, stochastic initial conditions
187: are implemented by letting $h(\vec{x},0)$ in Eq.~\ref{eq:total_height}
188: be random white noise. Specifically, $h(\vec{x},0)$ is assumed to
189: be sampled from a normal distribution such that\begin{equation}
190: \left\langle h(\vec{x},0)\right\rangle =0\mbox{, and }\left\langle h(\vec{x},0)h(\vec{x}',0)\right\rangle =\Delta^{2}\delta^{d}\left(\vec{x}-\vec{x}'\right)\label{eq:xmean}\end{equation}
191: where $\Delta$ is the noise amplitude of dimension {[}length]$^{1+d/2}$,
192: $d$ is the dimension of the surface, and $\delta^{d}(\vec{x})$ is
193: the $d\mbox{-dimensional}$ Dirac Delta function. Much of the following
194: analysis uses the Fourier transform with the convention, $h_{\vec{k}}(t)=(2\pi)^{-d}\int d^{d}\vec{k}\,\exp(-i\vec{k}\cdot\vec{x})h(\vec{x},t).$
195: The mean and two-point correlation functions for $h_{\vec{k}}(0)$
196: are \begin{equation}
197: \left\langle h_{\vec{k}}(0)\right\rangle =0\mbox{, and }\left\langle h_{\vec{k}}(0)h_{\vec{k}'}(0)^{*}\right\rangle =\frac{\Delta^{2}}{(2\pi)^{d}}\delta^{d}\left(\vec{k}-\vec{k}'\right).\label{eq:kmean}\end{equation}
198: %
199: \begin{comment}
200: \begin{enumerate}
201: \item As a consequence of Eqs.~\ref{eq:xmean} and~\ref{eq:kmean}, \[
202: \left\langle h_{k}(t)\right\rangle =\left\langle h(\vec{x},t)\right\rangle =0\]
203: for all $t$, and the two-point correlation functions $\left\langle h(\vec{x},t)h(\vec{x}',t)\right\rangle $
204: and $\left\langle h_{k}(t)h_{k'}(t)^{*}\right\rangle $ become important
205: measures of the average film height profile. These measures are well-suited
206: to characterize the order of the SAQD arrays.
207: \end{enumerate}
208:
209: \end{comment}
210: {}
211:
212: The deterministic evolution of a single Fourier component is determined
213: by surface diffusion with a diffusion potential $\mu(\vec{x},t)$.\cite{Zhang:2003tg,Spencer:1993vt}
214: This model is phenomenological in nature, but contains the essential
215: elements of SAQD formation. Thus, it is an adequate, but not overly
216: complex starting point for the investigation of the effects of crystal
217: anisotropy. Furthermore, models of this nature can be derived from
218: atomic scale simulations.~\cite{Haselwandter:2006hy} At any instant
219: in time, the growing film is described by the curve $\mathcal{H}(\vec{x},t)$.
220: Using Eq.~\ref{eq:total_height} to decompose the film height, \begin{eqnarray*}
221: \partial_{t}h(\vec{x},t) & = & \vec{\nabla}\cdot(\mathcal{D}\vec{\nabla}\mu(\vec{x},t;\bar{\h}));\\
222: d\bar{\h}/dt & = & Q,\end{eqnarray*}
223: where $\mu$ depends on $\bar{\h}$, and $Q$ is the flux of new
224: material onto the surface.
225:
226: %
227: \begin{comment}
228: The surface grows with a velocity normal to the surface,\begin{equation}
229: v_{n}=n_{z}\frac{d\mathcal{H}}{dt}=\sgrad\cdot\mathcal{D}\sgrad\mu+n_{z}Q\label{eq:surfvel}\end{equation}
230: where $n_{z}$ is the vertical $z$-component of the surface normal
231: $\hat{n}$, $\sgrad$ is the surface gradient, $\sgrad\cdot$ is the
232: surface divergence, $\mathcal{D}$ is the surface diffusivity, The
233: diffusivity, \begin{equation}
234: \mathcal{D}=\frac{D_{S}c_{S}}{k_{b}T}\exp\left[-\frac{\Delta E}{k_{b}T}\right],\label{eq:Diffusivity}\end{equation}
235: where $D_{S}$ is the intrinsic surface diffusivity, $c_{S}$ is the
236: mobile surface atoms per unit area, and $\Delta E$ is the surface
237: diffusion activation energy.
238: \end{comment}
239: {}
240:
241: The appropriate diffusion potential $\mu$ must produce Stranski-Krastanow
242: growth. Thus, it must incorporate the elastic strain energy density
243: $\omega$ that destabilizes a planar surface, the surface energy density
244: $\gamma$ that stabilizes planar growth and a wetting potential $W(\mathcal{H})$
245: that ensures substrate wetting. The %
246: \begin{comment}
247: wetting energy is an areal energy density. If it is taken to be a
248: density of the growing film surface as in~\cite[citation needed]{cn},
249: the resulting diffusion potential is \begin{equation}
250: \mu=\Omega\left(\omega-\kappa\left(\gamma+W(\h)\right)+n_{z}W'(\h)\right)\label{eq:cp}\end{equation}
251: where $\Omega$ is the atomic volume, and $\kappa=\textrm{Tr}(\sgrad\hat{n})$
252: is the total surface curvature. The strain energy $\omega$ is found
253: using isotropic linear plane strain elasticity. In general, $\omega$
254: is a function of $x$, and it is a non-local functional of the entire
255: surface profile $\mathcal{H}(x)$. Alternatively, if th
256: \end{comment}
257: {} simplest form that gives the appropriate behavior is
258:
259: \begin{equation}
260: \mu=\Omega\left(\omega-\kappa\gamma+n_{z}W'(\h)\right)\label{eq:cp2}\end{equation}
261: similar to~\cite{Liu:2003kx,Golovin:2003ms,Tekalign:2004jh,Friedman:2006bc}
262: where $\Omega$ is the atomic volume, $\kappa$ is the total surface
263: curvature, and $n_{z}$ is the vertical component of the unit surface
264: normal $\hat{n}$. The strain energy density $\omega$ is found using
265: isotropic linear plane strain elasticity. In general, $\omega$ is
266: a function of $x$, and it is a non-local functional of the entire
267: surface profile $\mathcal{H}(x)$.%
268: \begin{comment}
269: The main difference between the two potentials (Eqs.~\ref{eq:cp}
270: and~\ref{eq:cp2}) is that in Eq.~\ref{eq:cp}, the wetting potential
271: effectively adds an additional surface energy density. However, both
272: forms will effectively produce Stranski-Krastanow growth.
273: \end{comment}
274: {}
275:
276: Consider first, the the one-dimensional and two-dimensional isotropic
277: cases. Following~\cite{Golovin:2003ms,Spencer:1993vt,Tekalign:2004jh},
278: the surface diffusion potential, (Eq.~\ref{eq:cp2}) is expanded
279: to first order in the film height fluctuation $h$. The elastic energy
280: $\omega$ is calculated using linear isotropic elasticity. It is a
281: non-local function of $h(\vec{x})$; thus, it is useful to work with
282: the Fourier transform. In the isotropic case, the linearized diffusion
283: potential (Eq.~\ref{eq:chem_pot}) depends only on the wave vector
284: magnitude $k=\left\Vert \vec{k}\right\Vert $ . Thus, \begin{eqnarray}
285: \mu_{\vec{k}} & = & f(k,\bar{\height})h_{\vec{k}},\mbox{ with}\label{eq:chem_pot}\\
286: f(k,\bar{\height}) & = & \Omega\left(-2M(1+\nu)\epsilon_{0}^{2}k+\gamma k^{2}+W''(\bar{\h})\right).\label{eq:fiso}\end{eqnarray}
287: %
288: \begin{comment}
289: where, $a=\Omega M\epsilon_{0}^{2}$, $b=\Omega\gamma$, and $c=\Omega W''(\bar{\h})$.~\cite{Spencer:1993vt}{[}\noun{and
290: recent work}]
291: \end{comment}
292: {}%
293: \begin{comment}
294: for the case of Eq.~\ref{eq:cp}
295: \end{comment}
296: {}%
297: \begin{comment}
298: \begin{enumerate}
299: \item \[
300: f(k,\bar{\height})=-\Omega\mathcal{E}_{0}k+\Omega(\gamma+W(\bar{\height}))k^{2}+\Omega W''(\bar{\height})\]
301: if Eq.~\ref{eq:cp} is used. Approximating that the film and substrate
302: have similar elastic stiffnesses, $\mathcal{E}_{0}=M\epsilon_{0}^{2}$
303: where $M$ is the biaxial modulus and $\epsilon_{0}$ is the film-substrate
304: mismatch strain~\cite{Spencer:1993vt}{[}\noun{and recent work}].
305: \end{enumerate}
306:
307: \end{comment}
308: {}%
309: \begin{comment}
310: Alternatively, if Eq.~\ref{eq:cp2} is used, $b$ no longer depends
311: on the wetting potential $b=\Omega\gamma$.
312: \end{comment}
313: {}%
314: \begin{comment}
315: \begin{enumerate}
316: \item \[
317: f(k,\bar{\height})=-\Omega\mathcal{E}_{0}k+\Omega\gamma k^{2}+\Omega W''(\bar{\height})\]
318: Both of these forms can be simply expressed as\begin{equation}
319: f(k,\bar{\height})=-ak+bk^{2}+c\label{eq:fiso}\end{equation}
320: where $b$ and $c$ depend on $\bar{\height}$ in the first case,
321: and only $c$ depends on $\bar{\height}$ in the second case.
322: \end{enumerate}
323:
324: \end{comment}
325: {}In the anisotropic case, there will also be a dependence on the wave
326: vector direction $\theta_{\vec{k}}$.
327:
328: %
329: \begin{comment}
330: The surface diffusion equation (\ref{eq:surfvel}) can also be linearized
331: and stated in reciprocal space giving\begin{eqnarray*}
332: \frac{d\bar{\h}}{dt} & = & Q\\
333: \frac{dh_{k}}{dt} & = & -\mathcal{D}k^{2}\mu_{k}=-\mathcal{D}k^{2}f(\vec{k},\bar{\h})h_{k}\end{eqnarray*}
334:
335: \end{comment}
336: {}The time dependence of the film height has a simple solution if there
337: is no additional flux of material ($Q=0$, and $\bar{\h}$ is constant).\begin{eqnarray}
338: h_{\vec{k}}(t) & = & h_{\vec{k}}(0)e^{\sigma_{k}t}\mbox{, with}\label{eq:simple_time}\\
339: \sigma_{k} & = & -\mathcal{D}k^{2}f(k,\bar{\h}),\label{eq:dispersion}\end{eqnarray}
340: where $\sigma_{k}$ is the \emph{dispersion relation} and depends
341: only on the wavevector magnitude $k$. %
342: \begin{comment}
343: %
344: \begin{lyxgreyedout}
345: The statistical measures of $h_{k}(t)$ are easily found using Eq.~\ref{eq:kmean}.
346: \begin{eqnarray*}
347: \left\langle h_{k}(t)\right\rangle & = & 0;\\
348: \left\langle h_{k}(t)h_{k'}(t)^{*}\right\rangle & = & \left\langle h_{k}(0)h_{k'}(0)^{*}\right\rangle e^{(\sigma_{k}+\sigma_{k'})t}=\frac{\Delta^{2}}{(2\pi)^{d}}\delta^{d}\left(\vec{k}-\vec{k}'\right)e^{2\sigma_{k}t}.\end{eqnarray*}
349:
350: \end{lyxgreyedout}
351:
352: \end{comment}
353: {} Modes with positive $\sigma_{k}$ are grow unstably, while modes
354: with negative values of $\sigma_{k}$ decay.
355:
356: The important features of $\sigma_{k}$ are most easily recognized
357: using a characteristic wavenumber is $k_{c}=2M(1+\nu)\epsilon_{0}^{2}/\gamma$
358: and a characteristic time $t_{c}=(\mathcal{D}\Omega\gamma k_{c}^{4})^{-1}$.~\cite{Spencer:1993vt,Golovin:2003ms}
359:
360: \begin{equation}
361: \sigma_{k}=t_{c}^{-1}\alpha^{2}(\alpha-\alpha^{2}-\beta).\label{eq:dp}\end{equation}
362: where the shorthand $\alpha=k/k_{c}$ and $\beta=W''(\bar{\h})/(\gamma k_{c}^{2})$
363: is used.
364:
365: The dispersion relation (Eq.~\ref{eq:dp}) has a peak at $k_{0}=\alpha_{0}k_{c}$
366: where \[
367: \alpha_{0}=\frac{1}{8}\left(3+\sqrt{9-32\beta}\right).\]
368: Expanding $\sigma_{k}$ about $k_{0}$, \begin{eqnarray}
369: \sigma_{k}\approx\sigma_{0}-\frac{1}{2}\sigma_{2}\left(k-k_{0}\right)^{2}\mbox{ where }\nonumber \\
370: \sigma_{0}=\frac{1}{4t_{c}}\alpha_{0}^{2}\left(\alpha_{0}-2\beta\right),\quad\sigma_{2}=k_{c}^{-2}t_{c}^{-1}(3\alpha_{0}-4\beta).\label{eq:G2}\end{eqnarray}
371: Thus,\begin{equation}
372: h_{\vec{k}}(t)\approx h_{\vec{k}}(0)e^{\sigma_{0}t-\frac{1}{2}\sigma_{2}t\left(k-k_{0}\right)^{2}}.\label{eq:TEapprox}\end{equation}
373: %
374: \begin{comment}
375: \begin{eqnarray}
376: h_{k}(t) & = & h_{k}(0)e^{\tau G_{0}}\exp\left[-\frac{\tau}{2}G_{2}\left(\alpha-\alpha_{0}\right)^{2}+\dots\right]\label{eq:TEapprox}\\
377: & = & h_{k}(0)e^{G_{0}t/t_{c}}\exp\left[-\frac{t}{2t_{c}k_{c^{2}}}G_{2}\left(k-k_{0}\right)^{2}+\dots\right]\nonumber \end{eqnarray}
378:
379: \end{comment}
380: {}
381:
382: Now, the statistical correlation functions and correlation lengths
383: that characterize order are derived. Using Eqs.~\ref{eq:simple_time}
384: and~\ref{eq:TEapprox} along with the stochastic initial conditions
385: (Eqs.~\ref{eq:xmean} and~\ref{eq:kmean}) , the mean value of $h_{\vec{k}}(t)$
386: is \[
387: \left\langle h_{\vec{k}}(t)\right\rangle =\left\langle h_{\vec{k}}(0)\right\rangle e^{\sigma_{k}t}=0,\]
388: so that the mean surface perturbation is simply $0$ for all time
389: and all $\vec{k}$. However, the mean-square surface perturbations
390: as characterize by the second-order correlation function~\cite{Zwanzig:2001zf}
391: can be large,\begin{eqnarray}
392: \left\langle h_{\vec{k}}(t)h_{\vec{k}'}(t)^{*}\right\rangle & = & \left\langle h_{\vec{k}}(0)h_{\vec{k}'}(0)^{*}\right\rangle e^{(\sigma_{k}+\sigma_{k'})t}\nonumber \\
393: & = & \frac{\Delta^{2}}{(2\pi)^{d}}\vd{k}{d}e^{2\sigma_{k}t},\label{eq:k-cor}\end{eqnarray}
394: using Eq.~\ref{eq:kmean}. The real space correlation function can
395: be found by taking the inverse Fourier transform of Eq.~\ref{eq:k-cor},
396:
397: \begin{eqnarray}
398: \left\langle h(\vec{x},t)h(\vec{x}',t)^{*}\right\rangle & = & \int d^{d}\vec{k}\int d^{d}\vec{k}'\, e^{i\vec{k}\cdot\vec{x}-i\vec{k}'\cdot\vec{x}'}\left\langle h_{\vec{k}}(t)h_{\vec{k}'}(t)^{*}\right\rangle \nonumber \\
399: & = & \frac{\Delta^{2}}{(2\pi)^{d}}\int d^{d}\vec{k}\, e^{i\vec{k}\cdot\left(\vec{x}-\vec{x}'\right)}e^{2\sigma_{k}t}\label{eq:hcor}\end{eqnarray}
400: where integration over $\vec{k}'$ is simple due to the $\vd{k}{d}$
401: in Eq.~\ref{eq:k-cor}.
402:
403: Using Eq.~\ref{eq:G2}, \[
404: e^{2\sigma_{k}t}\approx e^{2\sigma_{0}t-\frac{1}{2}(2\sigma_{2}t)(k-k_{0}^{2})}\]
405: which is peaked at $k=k_{0}$. This form suggests that the real-space
406: correlation function is periodic with a gaussian envelope that has
407: a standard deviation of\begin{equation}
408: L_{cor}=\sqrt{2\sigma_{2}t}\label{eq:lcor}\end{equation}
409: $L_{cor}$ is the \emph{correlation length} of the film-height profile,
410: and characterizes the degree of order of the SAQD array. For example,
411: in one dimension ($d=1)$,\begin{eqnarray*}
412: & & \left\langle h(x,t)h(0,t)^{*}\right\rangle \dots\\
413: & & \dots=\frac{\Delta^{2}}{2\pi}\sum_{\pm}\int_{0}^{\infty}dk\, e^{2\sigma_{0}t-\frac{1}{2}L_{cor}^{2}(k-k_{0})^{2}\pm ikx}\\
414: & & \dots\approx\frac{2\Delta^{2}}{(2\pi L_{cor}^{2})^{1/2}}e^{2\sigma_{0}t-\frac{1}{2}(x/L_{cor})^{2}}\cos(k_{0}x)\end{eqnarray*}
415: an approximation that is valid if $k_{c}L_{cor}\gg1$. Thus, $L_{cor}$
416: gives the length scale over which dots will be ordered, and this scale
417: grows as $t^{1/2}$. As $t\rightarrow\infty$, \[
418: \left\langle h(x,t)h(0,t)^{*}\right\rangle =\frac{2\Delta^{2}}{(2\pi L_{cor}^{2})^{1/2}}e^{2\sigma_{0}t}\cos(k_{0}x),\]
419: and the entire array should be perfectly ordered.
420:
421: The situation in two-dimensions, however, is less friendly. As $t\rightarrow\infty$,
422: $L_{cor}\rightarrow\infty$, and $e^{-\frac{1}{2}L_{cor}^{2}(k-k_{0})}\approx(2\pi/L_{cor}^{2})^{1/2}\delta(k-k_{0})$
423: \begin{eqnarray*}
424: & & \left\langle h(\vec{x},t)h(\vec{0},t)^{*}\right\rangle \dots\\
425: & & \dots=\frac{\Delta^{2}}{(2\pi)^{3/2}L_{cor}}\int d^{2}\vec{k}\, e^{2\sigma_{0}t+i\vec{k}\cdot\vec{x}}\delta(k-k_{0})\\
426: & & \dots=\frac{\Delta^{2}k_{0}}{(2\pi L_{cor}^{2})^{1/2}}e^{2\sigma_{0}t}J_{0}(k_{0}\parallel\vec{x}\parallel)\end{eqnarray*}
427: Thus, the two-dimensional isotropic case has statistical order at
428: large times, but does not yield SAQD lattices as does the one-dimensional
429: case (see Fig.~\ref{cap:DensityPlots}a and~b).
430:
431: Now, consider the effects of crystal anisotropy. For example, let
432: the elastic energy term $\omega$ have $N$-fold symmetry while the
433: other terms are assumed isotropic. Then, the growth rate depends on
434: the both the wave vector magnitude $k$ (and thus on $\alpha=k/k_{c}$)
435: and direction $\theta_{\vec{k}}$ so that, $\sigma_{k}\rightarrow\sigma_{\vec{k}}$.
436: Naturally, the anisotropic elastic energy term in the growth rate
437: $\sigma_{\vec{k}}$ depends on the specific anisotropic elastic constants,
438: but the general qualitative effect of elastic anisotropy on SAQD growth
439: kinetics can be investigated without incorporating a detailed elastic
440: calculation at this time. Thus, the present work provides motivation
441: for more detailed calculation. A reasonable way to estimate how the
442: elastic energy term varies with direction ($\theta_{\vec{k}}$) is
443: to assume a low order harmonic form with the proper rotational symmetry.
444: The simplest such guess is\begin{eqnarray*}
445: \sigma_{\vec{k}} & = & t_{c}^{-1}\alpha^{2}\left(\alpha\left(1-\epsilon\sin^{2}\left(N\theta_{\vec{k}}/2\right)\right)-\alpha^{2}-\beta\right),\end{eqnarray*}
446: where $\epsilon$ parameterizes the importance of the directional-dependence.
447: More precise calculations using anisotropic elastic constants of real
448: materials (similar to ref.~\cite{Obayashi:1998fk}) will be given
449: in future work.
450:
451: $\sigma_{\vec{k}}$ has peaks at $N$ wave vectors, \[
452: \vec{k}_{n}=k_{0}\left(\left(\cos\theta_{n}\right)\hat{e}_{x}+\left(\sin\theta_{n}\right)\hat{e}_{y}\right)\]
453: with $\theta_{n}=2\pi(n-1)/N$. Around each peak, $\vec{k}$ can be
454: decomposed in the direction parallel ($k_{\parallel})$ and perpendicular
455: ($k_{\perp}$) to $\vec{k}_{n}$. Using this decomposition and expanding
456: $e^{2\sigma_{\vec{k}}t}$ about each peak,\begin{eqnarray}
457: & & e^{2\sigma_{\vec{k}}t}\approx\sum_{n=1}^{N}\exp\left(2\sigma_{n}t\right);\label{eq:expansion}\\
458: & & 2\sigma_{n}t=2\sigma_{0}t-\frac{1}{2}L_{\parallel}^{2}(k_{\parallel}-\alpha_{0}k_{c})^{2}-\frac{1}{2}L_{\perp}^{2}k_{\perp}^{2};\label{eq:tauGn}\\
459: & & L_{\parallel}=\sqrt{2\sigma_{2}t},\mbox{ and }L_{\perp}=\sqrt{(N^{2}\alpha_{0}\epsilon t_{c}^{-1}k_{c}^{-2})t}.\label{eq:ancor}\end{eqnarray}
460: Note that $L_{\parallel}$ is the same as $L_{cor}$ for the isotropic
461: case. Eq.~\ref{eq:expansion} is valid if $k_{c}L_{\parallel}\gg1,\mbox{ and }k_{c}L_{\perp}\gg1$.
462: Using, Eq.~\ref{eq:hcor}, but noting that $\sigma_{\vec{k}}$ now
463: depends on both the magnitude and direction of $\vec{k}$, along with
464: Eqs.~\ref{eq:expansion}, and~\ref{eq:tauGn},
465:
466: \begin{eqnarray}
467: & & \left\langle h(\vec{x},t)h(\vec{0},t)^{*}\right\rangle =\frac{\Delta^{2}}{(2\pi)^{2}}\frac{2\pi}{L_{\parallel}L_{\perp}}\dots\nonumber \\
468: & & \dots\sum_{n=1}^{N/2}e^{2\sigma_{0}t-\frac{1}{2}\left(L_{\parallel}^{-2}x_{\parallel}^{2}+L_{\perp}^{-2}x_{\perp}^{2}\right)}2\cos\left(k_{0}x_{\parallel}\right),\label{eq:hcorA}\end{eqnarray}
469: where $x_{\parallel}=(\cos\theta_{n})x+(\sin\theta_{n})y$, and $x_{\perp}=(-\sin\theta_{n})x+(\cos\theta_{n})y$.
470: Thus, the same tendency to long-range order as for the one-dimensional
471: case is present (see Fig.~\ref{cap:DensityPlots}c and~d).
472:
473: As a numerical example, consider Ge grown on Si. Both the isotropic
474: approximation and an estimated elastically anisotropic case (with
475: $N=4$, and $\epsilon=0.1$) are treated. Neglecting the difference
476: in elastic properties of the Si substrate, $E_{\text{Ge}}=1.361\times10^{12}\mbox{ dyne/cm}^{2}$,
477: $\nu_{\text{Ge}}=0.198$, $\epsilon_{0}=-0.0418$, $\Omega=2.27\times10^{-23}\mbox{cm}^{3}$,
478: $\gamma=1927\mbox{ erg/cm}^{2}$, and $W(\h)=4.75\times10^{-6}/\h\mbox{ erg/cm}^{3}$.
479: The resulting biaxial modulus is $M=1.697\times10^{12}\text{ dyne/cm}^{2}$,
480: characteristic wavenumber is $k_{c}=0.369\text{ nm}^{-1}$ and critical
481: film height is $\bar{\h}_{c}=1.132\text{ nm}\approx4\mbox{ ML}$.
482: If the film is grown to a thickness of $\bar{\h}=\h_{c}+0.25\mbox{ ML}\approx1.203\text{ nm}$,
483: and then allowed to evolve, $\beta=0.2078$, $\alpha_{0}=0.5664$,
484: $k_{0}=0.209\text{ nm}^{-1}$, $\sigma_{0}=0.01206t_{c}^{-1}$, $\sigma_{2}=0.867k_{c}^{-2}t_{c}^{-1}$,
485: $L_{\parallel}=0.746k_{0}^{-1}(t/t_{c})^{1/2}$, and $L_{\perp}=0.539k_{0}^{-1}(t/t_{c})^{1/2}$.
486: Note that the unspecified diffusivity $\mathcal{D}$ has been absorbed
487: into $t_{c}$. The film will stay in the linear regime as long as
488: the surface fluctuations stay small. For this purpose, let {}``small''
489: mean $1\mbox{ ML}=2.83\times10^{-8}\mbox{cm}$. Once the fluctuations
490: become {}``large'', individual dots will begin to form, and a nonlinear
491: analysis becomes necessary. It is useful to know the correlation lengths
492: at this time.
493:
494: To find the correlation lengths, one must choose the initial height
495: fluctuation intensity $\Delta$ and then calculate the time for fluctuations
496: to become {}``large''. The initial fluctuation intensity is somewhat
497: arbitrary, but $\Delta=8.02\times10^{-16}\mbox{ cm}^{2}$ gives an
498: average fluctuation of $1\mbox{ ML}$ over a patch $1\mbox{ ML}^{2}$
499: and seems appropriate. Next Eqs.~\ref{eq:ancor} and~\ref{eq:hcorA}
500: are used to find $t$ for which the r.m.s. fluctuations become large
501: $h_{rms}=\left\langle |h(0,t)|^{2}\right\rangle ^{1/2}=a_{0}$. There
502: are two solutions, $t/t_{c}=5.53\times10^{-3}$, and $t/t_{c}=471$.
503: The first solution is an artifact of the white noise initial conditions
504: and occurs during an initial shrinking of the surface height variance;
505: thus, the second solution is taken. At $t/t_{c}=471$, the correlation
506: lengths are found using Eq.~\ref{eq:ancor}, $L_{cor}=L_{\parallel}=77.5\mbox{ nm}$,
507: and $L_{\perp}=56.0\mbox{ nm}$. The smaller correlation length gives
508: $k_{0}L_{\perp}/\pi=3.73$, so a patch of about 4 dots across is expected
509: to be reasonably well ordered.
510:
511: %
512: \begin{figure}
513: \includegraphics[width=1.75in]{1a.eps}\includegraphics[width=1.75in]{1b.eps}
514:
515: \includegraphics[width=1.75in]{1c.eps}\includegraphics[width=1.75in]{1d.eps}
516:
517:
518: \caption{\emph{\label{cap:DensityPlots}Density plots of surface profiles
519: and correlation functions at} $t/t_{c}=471.$ The scale is in nm,
520: and the physical parameters are listed in the text. Plots (a) and
521: (b) are a sample surface profile and correlation function respectively
522: for an elastically isotropic material. Plots (c) and (d) are a surface
523: profile and correlation function for an elastically anisotropic material
524: with 4-fold symmetry. Surface profile plots saturate black at a minimum
525: of $\h=\bar{\h}$ for clarity. White dots indicate surface peaks.
526: Correlation function plots span $\pm$ the maximum value / 2.}
527: \end{figure}
528: A numerical simulation of linear size $l=24(2\pi/k_{0})=722\text{ nm}$
529: can be easily performed. The discrete initial conditions $h_{\vec{k}}$(0)
530: are taken from a normal distribution of zero mean and variance $\left\langle h_{\vec{k}}(0)h_{\vec{k}'}^{*}(0)\right\rangle _{discrete}=(\Delta^{2}/l^{2})\delta_{\vec{k}\vec{k}'}$,
531: where $\delta_{\vec{k}\vec{k}'}$ is the Kronecker delta, and each
532: vector component of $\vec{k}$ takes discrete values $2\pi m/l$ with
533: $m$ an integer. These components then evolve via Eq.~\ref{eq:simple_time}.
534: The results of an isotropic and anisotropic simulation along with
535: the corresponding real-space correlation functions are shown in Fig.~\ref{cap:DensityPlots}.
536: These plots clearly demonstrate the importance of anisotropy in producing
537: long range order.
538:
539: Figs.~and~\ref{cap:DensityPlots}c and~d appear to agree qualitatively
540: with observations of nucleationless growth of Ge$_{x}$Si$_{1-x}$
541: nanostructures on Si~\cite{Berbezier:2003mw,Brunner:2002gf,Gao:1999ve},
542: although mostly with $x<0.5$. Typical observed dot arrangements
543: appear to correspond to lower values of $\beta$ than 0.2078 used
544: for the example as they are quasiperiodic but less ordered than Fig.~\ref{cap:DensityPlots}c.
545: Quantitative reporting of correlation lengths would assist comparison
546: and possibly enable better tuning of phenomenological models to experiments.
547: InP/InGaP nanostructures reported in~\cite{Bortoleto:2003zh} appear
548: similar.
549:
550: From the analysis of the isotropic model, it is clear that long range
551: statistical order (long correlation lengths) require tight distributions
552: in reciprocal space. This long range statistical order is achieved
553: in the large time limit, but this statistical order is insufficient
554: to produce a well-ordered array of SAQDs. This lack of usable order
555: is reflected in the real-space two-point correlation function of the
556: isotropic model. However, in the one-dimensional and two-dimensional
557: anisotropic cases, there is tendency to form a lattice after a long
558: time. In the anisotropic case, there are two correlation lengths that
559: characterize observed SAQD array order. Analytic formulas for these
560: correlation lengths have been given for a model with elastic simplified
561: anisotropy, but the general conclusions and method of analysis should
562: apply to any source of anisotropy. Additionally, from a simple form
563: of the wetting potential it is observed that at film heights just
564: above the critical threshold $\h_{c}$, the correlation lengths can
565: grow quickly while the height fluctuations grow slowly; thus order
566: is enhanced. Availability of measured SAQD correlation lengths would
567: help to improve this analysis and to engineer more ordered quantum
568: dot arrays.
569:
570: \begin{thebibliography}{10}
571:
572: \bibitem{Liu:2003kx}
573: P.~Liu, Y.~W. Zhang, C.~Lu.
574: \newblock {\em Phys. Rev. B}, 67:165414, 2003.
575:
576: \bibitem{Spencer:1993vt}
577: B.~J. Spencer, P.~W. Voorhees, S.~H. Davis.
578: \newblock {\em J. Appl. Phys.}, 73(10):4955--4970, 1993.
579:
580: \bibitem{Zhang:2003tg}
581: Y.W. Zhang, A.F. Bower, P.~Liu.
582: \newblock {\em Thin solid films}, 424:9--14, 2003.
583:
584: \bibitem{Tekalign:2004jh}
585: W.~T. Tekalign, B.~J. Spencer.
586: \newblock {\em J. Appl. Phys.}, 96(10):5505--5512, 2004.
587:
588: \bibitem{Friedman:2006bc}
589: Lawrence~H. Friedman, Jian Xu.
590: \newblock {\em Appl. Phys. Lett.}, 88:093105, 2006.
591:
592: \bibitem{Golovin:2003ms}
593: A.~A. Golovin, S.~H. Davis, P.~W. Voorhees.
594: \newblock {\em Phys. Rev. E}, 68:056203, 2003.
595:
596: \bibitem{Wang:2004dd}
597: Yu~U. Wang, Yongmei~M. Jin, Armen~G. Khachaturyan.
598: \newblock {\em Acta Mater.}, 52:81--92, 2004.
599:
600: \bibitem{Cross:1993ti}
601: M.~C. Cross, P.~C. Hohenberg.
602: \newblock {\em Rev. Mod. Phys.}, 65(3):851--1112, 1993.
603:
604: \bibitem{Beck:2004yq}
605: M.~J. Beck, A.~van~de Walle, M.~Asta.
606: \newblock {\em Phys. Rev. B}, 70(20):205337, Nov 2004.
607:
608: \bibitem{Haselwandter:2006hy}
609: C.~Haselwandter, D~VVedensky.
610: \newblock {\em 2006 APS March Meeting}, 2006.
611:
612: \bibitem{Zwanzig:2001zf}
613: Robert Zwanzig.
614: \newblock {\em Nonequilbrium Statistical Mechanics}.
615: \newblock Oxford University Press, New York, 2001.
616:
617: \bibitem{Obayashi:1998fk}
618: Y.~Obayashi, K.~Shintani.
619: \newblock {\em J. Appl. Phys.}, 84(6):3141, 1998.
620:
621: \bibitem{Berbezier:2003mw}
622: I.~Berbezier, A.~Ronda, F.~Volpi, A.~Portavoce.
623: \newblock {\em Surf. Sci.}, 531:231--243, 2003.
624:
625: \bibitem{Brunner:2002gf}
626: Karl Brunner.
627: \newblock {\em Rep. Prog. Phys.}, 65(1):27--72, 2002.
628:
629: \bibitem{Gao:1999ve}
630: H.~J. Gao, W.~D. Nix.
631: \newblock {\em Annu. Rev. Mater. Sci.}, 29:173--209, 1999.
632:
633: \bibitem{Bortoleto:2003zh}
634: J.~R.~R. Bortoleto, H.~R. Gutierrez, M.~A. Cotta, J.~Bettini, L.~P. Cardoso,
635: M.~M.~G. de~Carvalho.
636: \newblock {\em Appl. Phys. Lett.}, 82(20):3523--3525, 2003.
637:
638: \end{thebibliography}
639:
640: \end{document}
641: