PHYSICAL REVIEW B VOLUME 62, NUMBER 12 15 SEPTEMBER 2000-II

Molecular-dynamics simulation of growth of nanocrystals in an amorphous matrix

J. K. Bording* and J. Taftø
Department of Physics, University of Oslo, P.O. Box 1048, Blindern, 0316 Oslo, Norway
共Received 9 December 1999; revised manuscript received 17 May 2000兲
Using germanium as an example we study the growth of crystalline nuclei by molecular dynamics. Starting
with crystalline nuclei of different sizes embedded in an amorphous matrix we follow the evolution of the
system at the atomic level. At a temperature about halfway between absolute zero and the melting temperature,
we observe that crystallites of diameter larger than 2.0 nm grow, while smaller crystals disappear.

I. INTRODUCTION we will in this paper present details about the molecular-

dynamics simulations we have undertaken. We calculate the
Classical theory of nucleation and growth of crystallites in spatial variations in energy, and we present results for the
a supercooled or amorphous matrix relies on the concept of local structure of the crystalline/amorphous boundary as well
surface tension, or Gibbs free energy to be more precise. The as the evolution of the embedded nuclei.
interface is, in this phenomenological theory, a mathematical
concept that is not easily accessible for experimental obser- Classical nucleation theory
vations. Contrary to the interface energy between a liquid We will now briefly review the classical theory of nucle-
and a smooth surface of a bulk solid, it is extremely difficult ation that we later will use to put the result of our simula-
to measure the interface free energy between a crystalline tions into perspective. According to this theory the additional
nucleus and an amorphous matrix. Furthermore, while the free energy related to the formation of a crystalline nucleus
atomic arrangement at a planar interface between two solid with radius r in an amorphous matrix can be written as
phases can be studied in great detail with a state-of-the-art
high-resolution transmission electron microscope, this chal- 4
lenge awaits realization for the curved interface around a ⌬G total ⫽ ␲ r 3 ⌬G phase ⫹4 ␲ r 2 ␥ . 共1兲
3
small nucleus. The dynamic nature, in addition to compli-
cated geometry, not only calls for high spatial resolution and Here ⌬G phase is the difference in free energy between the
sensitivity, but also for high temporal resolution. nucleus phase and the matrix phase. This is a driving force
Nucleation and growth can be addressed by molecular- for crystallization when the crystalline nucleus phase has
dynamics simulations using powerful computers. Several in- lower free energy than the amorphous matrix phase. ␥ is the
vestigators have used the simple Lennard-Jones pair poten- surface tension, or additional free energy per unit area asso-
tial to study nucleation and growth.1–3 In order to simulate ciated with the boundary between the two different phases.
most systems accurately, more elaborate potentials involving This energy is positive and will act as a barrier to the forma-
many-body interactions are needed. For the semiconductors tion and growth of the nucleus. If ⌬G phase is negative we see
silicon and germanium, accurate interatomic potentials are from Eq. 共1兲 that there exists a critical size
available in the literature.4–6 One of these, the Stillinger and
Weber potential,6 has previously been used to study the ki- ⫺2 ␥
netics of dissolution of silicon crystals.7 We use the potential r *⫽ . 共2兲
⌬G phase
of Tersoff4,5 and study the early stage of crystallization of
germanium at high supercooling. The reason for choosing For nuclei smaller than r * , a reduction in size will also
germanium as the material to simulate is that we8 have al- give a reduction in the total free energy. Such nuclei, often
ready shown by molecular-dynamics simulation with the called embryos, will reduce in size and vanish. However, for
Tersoff potential that the simulated structure of amorphous nuclei larger than r * , the size must increase to reduce the
germanium is in excellent agreement with high-quality dif- total free energy. r * cannot often, however, be calculated
fraction experiments.9 A penalty for using this accurate, but from Eq. 共2兲 because of the difficulty of obtaining reliable
unfortunately rather complex potential is the increasing com- values for ␥ .
putation time. Thus even with the most powerful computers
we are able to scan only a very small volume in space-time, II. SIMULATION PROCEDURE
making it extremely unlikely to observe the creation of a
nucleus that grows to the critical size. We circumvent this The amorphous phase prior to the insertion of the nuclei
problem by embedding crystallites of different sizes in the was a cubic simulation cell with 4096 atoms and periodic
amorphous matrix and study their development with time. boundary conditions. Its structure was a typical continuous
For crystallites smaller than the critical size we mostly ob- random network, with a radial distribution function in very
serve annihilation, which is the time reversal of the ‘‘fluc- good agreement with diffraction experiments on amorphous
tuation’’ causing the creation of a nucleus. germanium.9 Details about the preparation and structure of
After a brief introduction to classical theory of nucleation the amorphous material were reported in a previous paper.8

0163-1829/2000/62共12兲/8098共6兲/$15.00 PRB 62 8098 ©2000 The American Physical Society

PRB 62 MOLECULAR-DYNAMICS SIMULATION OF GROWTH OF . . . 8099

Spherical crystalline nuclei were embedded in this matrix by

replacing the atoms inside a sphere in the center of the simu-
lation cell with a perfect spherical crystal with the diamond
structure. Prior to insertion, both the amorphous matrix and
the crystal were heated to 2000 K* at zero external pressure
using molecular dynamics. Notice the asterisk on the tem-
perature unit. We use this symbol to distinguish the simula-
tion temperature scale from the Kelvin scale 共see Ref. 8 for
further details兲. This is done because germanium simulated
with the Tersoff4,5 potential has a melting temperature of
3200 K* , much higher than real germanium.
We made 15 simulation cells with varying nuclei size.
Because of the varying nuclei size and local density in the
amorphous matrix, the number of atoms in the inserted
spherical crystallite was not necessarily exactly the same as
the number of atoms taken out. Thus the new simulation cell FIG. 1. Energy profile at 500, 2000, and 5000 time steps. The
contained a total number of atoms that could deviate slightly radius of the inserted crystalline nucleus in this case is 1.5 nm, and
from the original 4096. The cubic computation cells had the system was kept at 2000 K* .
sides of about 4.5 nm, large enough to prohibit the crystal-
line nucleus to interact with itself. III. RESULTS
The insertion of the spherical crystal produced an inter-
face with many unfortunate configurations and thus a high A. Internal energy across the interface
internal potential energy. This energy was initially removed In Fig. 1 we have plotted the energy profile at three dif-
by relaxing the system with a steepest-descent method into ferent times at the initial stage of the simulation. With the
the nearest local energy minimum. This was done to avoid energy profile we mean the average single atomic energy ␾ i
the sudden release of large amounts of energy, which would as a function of the distance from the center of the nucleus.
cause the entire system to melt. The new simulation cells, The insertion of the spherical crystal resulted in atoms in the
with an amorphous matrix and a crystalline nucleus, were boundary region with very unfavorable configurations, and
then simulated at 2000 K* with zero external pressure in a thus high energy. After the steepest-descent relaxation most
molecular-dynamics simulation. This is above 21 T m , where
of this energy was removed, but the atoms still had an el-
T m is about 3200 K* using the Tersoff potential, so the at-
evated energy compared to the adjacent bulk phases. This is
oms have significant mobility. At the same time, the tem-
seen by the peak, or barrier, in the energy profile of Fig. 1
perature is low enough to give a critical size so small that it
right at the position of the original boundary. After some
can be probed in a simulation. The systems were initially
annealing at 2000 K* the barrier reduced its height, and
kept at this temperature and pressure for 5000 time steps in
eventually after about 5000 time steps, or 10 ps, it was com-
order to reach steady state, and then another 350 000 time
pletely removed. Although the peak did not last very long, it
steps to study the evolution of the nuclei. A simulation like
lasted long enough to indicate that the rearrangement in the
this, at this temperature, is effectively an annealing. Each
boundary region was more than just a minute movement of
time step is 2.0⫻10⫺15 s, giving a total simulation time for
some of the atoms. The rearrangement in the crystalline
each of the configurations of 0.7 ns. All simulations were
phase was rather gradual, while in the amorphous phase it
done with constant number of atoms, pressure, and tempera-
was more varying. Some atoms moved as far as 0.1 nm,
ture (N PT) using the velocity Verlet algorithm. while others were nearly stationary.
As for most analytical potentials, the Tersoff potential4,5 Notice from Fig. 1 that the potential energy of both the
calculates the potential energy of an ensemble of atoms as a crystalline and the amorphous phase away from the bound-
sum over N individual single atomic energies: ary is constant and equals the energy of the respective bulk
phases. The barrier close to the boundary vanished after
N about 5000 time steps, and thus there is no indication of a
⌽⫽ 兺 ␾i .
i⫽1
共3兲
barrier in the internal energy. However, the barrier may still
be present in the free energy. In order to obtain the free
energy we need to know the entropy.

Although it is only the total potential energy that is a well-

defined quantity, we will in this paper utilize the single B. Atomic order across the interface
atomic energies, ␾ i . By doing so we are able to map the In principle, the entropy can be extracted from molecular
energy with large spatial resolution that is particularly inter- dynamics simulations, and Ma,10 for instance, has proposed a
esting in the boundary region between the amorphous and procedure to do so. However, in practice this is a very time-
crystalline phases. Notice that in order to obtain the total consuming procedure, where an enormous number of states
internal energy, the kinetic energy for the ensemble of atoms has to be probed. Lacking a better alternative, we consider
N 1 2
must be added ( 兺 i⫽1 2 m i v i ). This provides, however, no the crystalline order across the interface to assess the entropy
new information in our simulations. as a function of the distance from the center of the nuclei.
8100 J. K. BORDING AND J. TAFTØ PRB 62

FIG. 2. A snap shot of a typical boundary between a spherical

crystal of 3 nm and the amorphous matrix 共a兲. 共b兲 and 共c兲 are inter-
nal potential energy and order profiles, respectively, across this
boundary. The vertical lines in 共b兲 and 共c兲 indicate the position of
the intermediate value of the internal energy and normalized struc-
FIG. 3. Snapshots of an embedded nucleus larger than the criti-
ture factor, respectively. We chose the vertical line in 共c兲 as the
cal size, at 2000 K* . The time interval between the snapshots are
boundary position. The temperature of the system was 2000 K* . 共d兲
0.5 ns. The size of the squares are 4.5 nm⫻4.5 nm.
is, as explained in the text, an estimate for the free energy across the
boundary at two arbitrary temperatures.
The normalized structure factor is plotted in Fig. 2共c兲. The
More precisely, we calculate the average structure factor normalization is done by dividing the structure factor from
from the atoms lying in a shell, of thickness 0.1 nm, around the shell of simulated atoms, F(R), with the structure factor
the center of the nuclei: from a shell of perfect crystal, F c (R), at the same tempera-

冏兺 冏
NR ture. Thus, for a shell of perfect crystal the normalized struc-
1 ture factor is unity. Due to strain this does not happen close
F 共 g,R兲 ⫽ e 2 ␲ ig•r j , 共4兲
NR j⫽1 to the interface. For crystals smaller than about 2.0 nm, how-
ever, we see from Fig. 2共c兲 that the entire crystal will to
where R is the radius of the shell, r j is the position of atom
some degree be deformed. If the matrix were polycrystalline
j, N R is the number of atoms in the shell, and g is the recip-
rather than a continuous random network, the stress could
rocal 关111兴 vector. Other choices for g were tested, but no
probably be larger, deforming even larger crystals.
important difference was found.
PRB 62 MOLECULAR-DYNAMICS SIMULATION OF GROWTH OF . . . 8101

FIG. 5. Size development of the 15 nuclei at 2000 K* . This plot

clearly shows the critical diameter to be about 2.0 nm. All the larger
nuclei grew, while those smaller disintegrated.

Helmholtz free energy A⫽U⫺TS. Let us for the sake of

indicating the effect of this shift assume that

S⫽k 1⫺冋 F 共 R兲
册
F c 共 R兲
, 共5兲

where k is a constant. In Fig. 2共d兲 we have calculated the

Helmholtz free energy for two different values of k at a cer-
tain temperature T, which effectively corresponds to two dif-
ferent values of temperature. We note from Fig. 2共d兲 that an
energy barrier is formed and that the height of this barrier
depends on k. The area of this barrier represents the bound-
ary energy ␥ . We also see from Fig. 2共d兲 that with the largest
k, corresponding to the largest temperature, the free energy
of the amorphous matrix is only slightly higher than the free
energy of the crystalline nucleus. This is what we expect at
temperatures slightly below the melting temperature. Exactly
at the melting temperature the energy difference should van-
ish.
FIG. 4. Snapshots of an embedded nucleus slightly smaller than The relationship we have proposed between the entropy
the critical size, at 2000 K* . The time interval between the snap- and the structure factor is certainly only qualitative. Still we
shots is 0.3 ns. The size of the squares are 4.5 nm⫻4.5 nm. are confident that the shift between the vertical lines in Fig.
2共b兲 and 2共c兲 is the origin of the boundary energy ␥ . To
C. Free energy on an atomic scale
make this point it is sufficient to assume that the entropy
By comparing Figs. 2共b兲 and 2共c兲 we note that the order increases monotonically with decreasing structure factor.
as expressed by the normalized structure factor extends far- Figures 2–4 indicate an interface that is sometimes re-
ther away from the center of the nucleus than does the low ferred to as atomically diffuse. Because silicon and germa-
internal energy of the crystalline nucleus. As indicated by the nium have strong covalent bonding, these materials usually
vertical lines in Figs. 2共b兲 and 2共c兲 this shift is about 0.3 nm. exhibit relatively sharp, or facetted, surfaces between a solid
The line in Fig. 2共b兲 is drawn at the position where the av- and liquid phase.11 Due to the smallness of the nuclei, how-
erage internal energy is halfway between the crystalline and ever, we find a diffuse interface with no apparent facets on
amorphous phase 关 U⫽ 21 (U c ⫹U a ) 兴 , and the line in Fig. 2共c兲 the crystals. This effect, which might lead to a surface ten-
is drawn at the position where the normalized structure factor sion depending on the size of the crystal, is observed in other
is one-half 关 F(R)/F c (R)⫽ 21 兴 . This shift may be the origin simulated systems2,7 as well as a high-resolution electron mi-
of the interface free energy. We illustrate this by considering croscopy experiment.12
8102 J. K. BORDING AND J. TAFTØ PRB 62

D. Evolution of the nuclei

Uttormark and Thompson7 discussed different criteria to
distinguish crystalline and liquid regions. They utilized local
configuration parameters such as coordination numbers,
which work well when comparing liquid and crystalline
semiconductors that have distinctly different local order.
However, in our case where we need to distinguish between
amorphous and crystalline semiconductors these criteria are
less useful. In our situation we have the choice of, for in-
stance, using the vertical line in Figs. 2共b兲 and 2共c兲, or the
maximum of the curve in Fig. 2共d兲. These are shifted by 0.3
nm, but for our purpose of monitoring the growth of the
nuclei this is not crucial. We choose to use the vertical line in
Fig. 2共c兲.
In Fig. 5 we have plotted the size of the 15 nuclei as a
function of time. From this figure we can determine the criti-
cal size. All nuclei larger than about 2.0 nm in diameter
grew, while those smaller disappeared. Notice, however, the
one nucleus at 2.0 nm lingering for a long period of time.
This is due to the low driving force for crystals close to the
critical size. Because the small nuclei had fewer atoms, and FIG. 6. Velocity of the crystalline/amorphous boundary as a
because some of them moved slightly off center, the size function of the nuclei size simulated at 2000 K* . At about 2.0 nm
determination of the small nuclei was not as accurate as it the velocity is zero, while for large sizes the velocity approaches
was for the larger nuclei. The relative fluctuations in the about 0.5 m/s.
diameter for these nuclei were also larger, causing the data
for small crystals to be somewhat more noisy. process by molecular-dynamics simulations. We have shown
Figures 3 and 4 show explicitly that a nucleus larger than
that molecular-dynamics simulation is a powerful method to
the critical size grows upon annealing, while a smaller
gain insight into the rearrangement taking place at the atomic
nucleus disappears. This is in agreement with classical nucle-
level when the interface propagates. The profile of the inter-
ation theory. The atoms close to the boundary slowly change
their configuration, joining either the central crystalline nal energy across the interface is rather easily accessible us-
nucleus or the surrounding amorphous matrix. In the case of ing molecular-dynamics simulations for systems like germa-
a subcritical nucleus, the entire nucleus will eventually be nium and silicon where reliable interatomic potentials are
assimilated in the amorphous phase. If the nucleus is larger available. However, when it comes to the free energy we
than the critical size, however, it will grow to fill the entire encounter problems because of the difficulties in extracting
simulation cell. the entropy. We were in this study able to infer the presence
From Fig. 5 we were able to calculate the velocity of the of a barrier in the free energy associated with the interface
boundary as a function of the nucleus size. After averaging tension, by realizing from the simulation that the outer
over all 15 nuclei, we plotted this curve in Fig. 6. From this atomic layer of the crystal was heavily strained. In this layer
figure we see a transition from negative growth to positive the entropy is low relative to the amorphous phase even
growth at about 2.0 nm. For larger nuclei the velocity seems though the internal energy is large relative to the crystal.
to reach a maximum at about 0.5 m/s, while for increasingly The fact that relevant thermodynamic properties such as
smaller nuclei the negative velocity increases. The entropy, and thus free energy, are not easily revealed from
asymptotic velocity is in fairly good agreement with these simulations makes it difficult to compare our results
experiments13 and other simulations on the velocity of this directly with the classical macroscopic theory of nucleation.
boundary.14 It is, however, important to realize that because The simulations rather serve as a microscope with high spa-
of the extremely small surface of the embedded nuclei, the tial and temporal resolution to follow the evolution of the
growth mechanisms are somewhat different from single system at the atomic level. Based on this study of the evolu-
plane growth. All planes, and thus both lateral and kink tion of the system we determine a critical size of the nuclei
growth, are in a way present, each with a different growth that are in good agreement with what has been derived from
rate. We observed no twin boundaries, vacancies, or self- various experiments.12,15–17
interstitials in the grown crystals.

IV. SUMMARY ACKNOWLEDGMENT

We have studied the interface between crystalline and Financial support from the Norwegian Research Council
amorphous germanium during the nucleation and growth is gratefully acknowledged.
PRB 62 MOLECULAR-DYNAMICS SIMULATION OF GROWTH OF . . . 8103

*Email: J.K.Bording@fys.uio.no 1985兲.

1 11
P.R. ten Wolde, M.J. Ruiz-Montero, and D. Frenkel, J. Chem. K.A. Jackson, Kinetics of Reaction in Ionic Systems, edited by T.
Phys. 104, 9932 共1996兲. J. Gray and V. D. Fre’chette 共Plenum, New York, 1969兲, p 229.
2
K. Yasuoka and M. Matsumoto, J. Chem. Phys. 109, 8451 共1998兲. 12
J.L. Batstone, Philos. Mag. A 67, 51 共1993兲.
3
L.A. Baez and P. Clancy, J. Chem. Phys. 102, 8138 共1995兲. 13
G. Aichmayr, D. Toet, M. Mulato, P.V. Santos, A. Spangenberg,
4
J. Tersoff, Phys. Rev. B 38, 9902 共1988兲. S. Christiansen, M. Albrecht, and H.P. Strunk, J. Appl. Phys. 85,
5
J. Tersoff, Phys. Rev. B 39, 5566 共1989兲. 4010 共1999兲.
6
F.H. Stillinger and T.A. Weber, Phys. Rev. B 31, 5262 共1985兲. 14
T. Motooka, Mater. Sci. Eng., A 253, 42 共1998兲.
7 15
M.J. Uttormark, M.O. Thompson, and P. Clancy, Phys. Rev. B H. Hofmeister, J. Dutta, and H. Hofmann, Phys. Rev. B 54, 2856
47, 15 717 共1993兲. 共1996兲.
8
J.K. Bording, Phys. Rev. B 共to be published兲. 16
M.H. Brodsky, Light Scattering in Solids, edited by M. Cardona
9
G. Etherington, A.C. Wright, J.T. Wenzel, J.C. Dore, J.T. Clark, 共Springer, Berlin, 1975兲.
and R.N. Sinclair, J. Non-Cryst. Solids 48, 265 共1982兲. 17
R. Bisaro, J. Magariño, Y. Pastol, P. Germain, and K. Zellama,
10
S.K. Ma, Statistical Mechanics 共World Scientific, Singapore, Phys. Rev. B 40, 7655 共1989兲.