Вы находитесь на странице: 1из 10

PHYSICAL REVIEW B 73, 115427 共2006兲

Surface diffusion dewetting of thin solid films: Numerical method and application to Si/ SiO2

E. Dornel,* J-C. Barbé, F. de Crécy, and G. Lacolle

CEA-DRT-LETI - CEA/GRE- 17, rue des Martyrs, 38054 Grenoble cedex 9, France

J. Eymery
Équipe mixte CEA-CNRS-UJF, Nanophysique et Semiconducteurs, CEA/DRFMC/SP2M, 17, rue des Martyrs,
38054 Grenoble cedex 9, France
共Received 18 October 2005; revised manuscript received 30 January 2006; published 27 March 2006兲

A method has been developed to calculate and use a surface chemical potential which is valid in the large
curvature regime for any surface energy function. It is applied to the solid-phase dewetting of a finite film with
an initial rectangular profile and considers the surface diffusion mechanism. For an isotropic surface energy,
the film aspect ratio and the adhesion energy between the film and the substrate are shown to be the main
parameters that quantify the retraction, the breaking time, and the number of agglomerates. Moreover, it is
found that mild surface energy anisotropy with an energy minimum in the horizontal plane postpones the mass
detachment. Simple models of the ␥-plots for the surface energy illustrate the influence of cusp points on the
retraction profiles. Finally, the smooth and facetted experimental surfaces, that are observed in the Si/ SiO2
system after 900 ° C annealing under H2, are explained by a quite small anisotropy of the ␥-plot.

DOI: 10.1103/PhysRevB.73.115427 PACS number共s兲: 68.35.⫺p, 68.55.Jk, 68.37.⫺d, 81.16.Rf

I. INTRODUCTION sion anisotropy has not been seen to be sensitive in two

Thin solid films are basic building blocks in microelec- dimensions 共this paper’s framework兲, contrary to the surface
tronics and optoelectronics. However, due to the shrinking of energy anisotropy 共see hereafter兲. Moreover, a 1 nm Si film
the layer thickness imposed by technology requirements, the 共not voluntarily strained兲 on SiO2 has a residual biaxial stress
surface to volume ratio is drastically increased and this leads of less than 100 MPa grazing incidence x-ray diffraction
to capillarity instabilities. To ensure device reliability, these 共GIXRD兲 measurements have been performed to confirm this
morphological instabilities of stacked layers must be con- point兲. Considering the Grinfeld-Azaro-Tiller instability, this
stress state is not destabilizing for continuous films thinner
trolled during the elaboration process to avoid an agglomera-
than 15 nm.19 In consequence, this paper focuses on the ef-
tion phenomenon that occurs well below the melting tem-
fect of the anisotropy of the surface energy neglecting the
perature. For example, it has been observed that during a strain energy. This paper will address the general problem of
classical thermal annealing at 850 ° C 共⬃2 / 3 of the Si melt- the two-dimensional 共2D兲 retraction of a finite anisotropic
ing temperature兲, a 10.5 nm silicon on insulator 共SOI兲 film film from a rigid substrate. An interesting case to study is an
can be fully agglomerated in a few minutes.1 It has been initial sharp step profile as shown in Fig. 1共a兲. This profile,
reported that the thinner the Si film is, the quicker the ag- representing for example a patterned SOI substrate, exhibits
glomeration occurs.1 More generally, the thermal stability of two singular points: the upper corner and the contact point
silicon1–10 and metals11–13 on amorphous substrates has been with the substrate. It is generally calculated15,18 and con-
intensively studied. firmed by our experiments 共see hereafter Fig. 11兲 that the
In order to prevent the morphological evolution of the film retracts with a thickened edge followed by a valley as
thin films, the influence of the main physical ingredients shown in Fig. 1共b兲. The film may pinch off after a breakup
must be identified. It is well known that an unstrained nearly time called tc. Phenomenology and dynamics differ with re-
plane infinite solid surface flattens due to capillarity effects,14 gard to the aspect ratio of the initial patterned film and the
whereas the boundaries of thin patterns must be considered contact angle. In this paper, a discrete calculation of the sur-
as destabilizing zones because they do not usually minimize face chemical potential is developed to take large curvature
the surface energy.15 At a temperature between half and two morphologies and mass conservation into account. This will
thirds of the melting temperature of a crystalline material, it be applied to the solid-state dewetting phenomenon to under-
is generally assumed that surface diffusion dominates the stand the influences of the contact angle and aspect ratio.
other transport processes of matter, namely viscous flow, This approach can also be used to simulate other physical
evaporation/condensation, and volume diffusion.14,16,17 problems such as hole growth or thermal grooving. It allows
Wong has studied the surface morphology evolution driven taking the anisotropy of the physical properties that signifi-
by surface diffusion in the case of semi-infinite, isotropic, cantly alter the dynamics of the shape evolution20 into ac-
and unstrained film using a numerical approach.15,18 count, as well as the final equilibrium shape known as the
For typical annealing temperatures in the 650 ° C – 900 ° C Wulff’s construction.21
range, it has been shown that specific directions are favored The following section presents the physical basis of the
during the retraction of a Si film from a SiO2 substrate.3–6 problem. The numerical method used to solve the evolution
This indicates that the anisotropy of surface energy and/or equations is described in Sec. III, applied to the dewetting of
diffusion needs to be taken into account. The surface diffu- a finite solid film with different aspect ratios and compared

1098-0121/2006/73共11兲/115427共10兲/$23.00 115427-1 ©2006 The American Physical Society


FIG. 1. Schematics of the dewetting of a thin

film of thickness H and length 2L. Only half of
the film is drawn due to symmetry consideration.
The contact angle between the film and the sub-
strate is described by ␣. The initial edge 共a兲 re-
tracts 共see x0兲 and the profile forms a peak fol-
lowed by a valley 共b兲. The film evolves until
peeling off in two 共c兲 or three 共d兲 agglomerates,
or it rearranges in single agglomerate 共e兲.

to experimental profiles in Sec. IV. Different physical as-

sumptions will be considered: mass retraction with isotropic
surface energy 共Sec. IV A兲 as well as mild and severe aniso-
␦N = −
冉 冕冕 S
div共Jជ兲 · ជ

dS · ␦t 共2兲

tropy 共Sec. IV B兲. Finally, discussion and general conclu-

sions are presented in Sec. V. Assuming a constant atomic volume ⍀0, the volume incre-
ment ␦V is simply related to ␦N by: ␦V = ⍀0 ⫻ ␦N.
In order to obtain dimensionless equations, the material
II. PHYSICAL MODEL dependent parameters and the temperature are included in a
constant B
In this paper, surface diffusion will be described by the
surface chemical potential ␮ defined as the derivative of the B= , 共3兲
Gibbs free energy G with respect to the number of atoms at k BT
constant temperature T and stress ␮ = 共⳵G / ⳵N兲T,stress.22 It is where ␥0 is a surface energy reference 共detailed in Sec.
supposed that the system temperature is homogeneous and, IV B兲. The initial film thickness H constitutes a natural
as long as the strain effect is neglected, the derivative of G length scale and time is made dimensionless by the H4 / B
with respect to N can be seen as an infinitesimal increment factor. As the dimensionless space variable is only a function
⳵G over the infinitesimal increment of volume ⳵V : ␮ of H, the profile shape will only depend on the geometry and
= ⍀0共␦G / ␦V兲 where ⍀0 is the atomic volume. not on the material properties included in B. This latter pa-
In the case of unstrained films and by neglecting the sur- rameter will steer the time evolution.
face stress, the Gibbs energy variation can be taken as the The evolution of the initial profile will be determined by
variation of the integral of the surface energy ␥ over the Eqs. 共1兲 and 共2兲 共a fourth-order differential equation兲 and by
surface: ␦G = ␦兰surface␥dS. Considering an infinitesimal pro- four boundary conditions given in the following section.
trusion on a smoothly curved surface, Herring22 demon-
strates that the corresponding increment in Gibbs energy is
equal to 共␥ + ␥⬙兲␬␦V, where ␬ is the surface curvature and III. NUMERICAL METHOD
the prime denotes the derivation over the angle ␪ formed
between the tangent to the surface and a reference plane. As shown in Fig. 2, three consecutive points M i−1, M i,
This leads to the following one dimensional 共1D兲 expression: and M i+1 of the 1D surface are considered. At each point, the
␮ = ⍀0共␥ + ␥⬙兲␬. This latter potential value is only valid at numerical method developed in this paper consists of a se-
quence of three steps that will be detailed hereafter: 共i兲 cal-
low curvature and then cannot be applied to a facetted sur-
culation of the surface potentials ␮i from the differential
face. So, the derivative form of ␮ will be preferred in the
definition; 共ii兲 calculation of local increments of matter ␦Ni
from Eq. 共2兲; and 共iii兲 the displacement of the points.
According to the Nernst-Einstein equation,17 the surface
共i兲 The chemical potential ␮i is calculated considering an
potential gradient will produce a drift of surface atoms with
infinitesimal displacement ␦␩i of M i perpendicular to
a flux
共M i−1 , M i+1兲

Jជ = −冉 冊
D s␯ ជ
k BT
· ⵜ S␮ , 共1兲
␮i = ␮0 + lim
冉 ␦Gi共␦␩i兲
冊, 共4兲

where ␮0 is a constant which can be set to zero without loss

of generality. Considering the notations of Fig. 2 for the sur-
where ⵜ ជ is the surface gradient operator, D is the surface
S s face energies ␥i and the segment lengths bi, li, ho,i, this equa-
diffusion coefficient, ␯ is the number of atoms per unit area, tion yields
and kBT is the thermal energy. For an isotropic surface en-
ergy and an infinite film,14 the above equation shows that the 2
␮i = ␮0 +
surface tends to flatten 共atoms flow from high to low curva-

再冉 冊冎
b1,i + b2,i

ture areas兲.
The infinitesimal increment of matter ␦N is expressed by ␥i−1 ␥i b2,i␥i⬘ b1,i␥i−1

⫻ h0,i + + − . 共5兲
the matter conservation law ᐉi−1 ᐉi ᐉi ᐉi−1


sign of this sum. The two last boundary conditions necessary

to solve the dynamic equations are given by the zero mass
flux through the symmetry axis and the contact line. This
ensures the mass conservation of the system. The flux going
to the symmetry axis at point M n is the sum of the two
symmetric fluxes from M n−1 and the fictive point M n+1. The
flux going to the contact line at point M 0 is reduced to the
flux from M 1.
共iii兲 Each M i point 共i ⫽ 0兲 is displaced by ␦zi perpendicu-
lar to the base of each 共M i−1, M i, M i+1兲 triangle. Using di-
mensionless equations in the 2D case, the matter increment
␦Ni is equivalent to the surface increment, so that ␦zi is the
matter increment ␦Ni divided by the length of the base of the
共M i−1, M i, M i+1兲 triangle 共called b1,i + b2,i in Fig. 2兲: ␦zi
FIG. 2. Sketch for the calculation of the surface chemical po- = ␦Ni / 共b1,i + b2,i兲. The general conservation of matter has
tential according to Eq. 共5兲 共see text兲 at the point M i between M i−1 been checked during the calculation. The displacements of
and M i+1, which is not positioned at the boundary with the sub- M 0 共resp. M n兲 is forced to be parallel to the substrate 共resp.
strate. li−1 共␥i−1兲 and li 共␥i兲 are the length 共surface energies兲 of the to the axis of symmetry兲 to ensure zero mass flux.
segment 关M i−1M i兴 and 关M iM i+1兴. h0,i is the distance between M i The present numerical approach can be adapted to several
and 共M i−1M i+1兲. b1,i and b2,i are respectively the distance between different physical problems by simply changing the four
M i−1 and M i+1 to Pi, the orthogonal projection of M i on 共M i−1M i+1兲. boundary conditions which represent the potentials and
␦Vi and ␦Gi are respectively the increment of volume and energy of fluxes at the boundaries of the system.
the surface after a displacement ␦␩i of M i along 关PiM i兴. The inset In the early part of the simulation of the film retraction,
shows the sketch of the calculation of the surface chemical potential surface displacement close to the edge is quite large com-
at the point M 0 on the contact line. pared to that close to the symmetry axis. Nevertheless, at a
later stage, the instability propagates towards the middle of
It is worth noting that ␮i does not depend directly on the the film. To achieve good accuracy without excess CPU time
second derivative of the surface energy, but on its first de- consumption, the initial mesh is tightened close to the edge
rivative. However, considering an isosceles triangle and made coarse far from it. Then, during simulation, the
共M i−1 , M i , M i+1兲 and the low curvature limit 共h0,i → 0兲, it is mesh is automatically slowly adapted to follow the instabil-
easy to deduce the finite difference form of the Herring for- ity propagation. At each time step, the positions of the M i
mula ␮ = 共␥ + ␥⬙兲␬. points are shifted parallel with 共M i−1, M i+1兲 towards the me-
Only half of the whole system needs to be simulated due dian of 关M i−1, M i+1兴 of a distance proportional to the distance
to symmetry consideration. The chemical potential values on between M i and the median of 关M i−1, M i+1兴. A coefficient of
the axis of symmetry and at the triple point define the first proportionality of 0.01% has been empirically validated.
two boundary conditions. The chemical potential ␮n on the This lateral movement does not change matter quantity, and
axis of symmetry is calculated using Eq. 共5兲 adding a fictive the mesh becomes nearly homogeneous after ⬃2–3 ⫻ 105
point M n+1, symmetric about M n−1. Since the main concern is time steps. This scheme stabilizes the numerical code and
with discontinuous films, the most important place to ob- enables a decrease in the CPU time by about one or two
serve is the point on the contact line between the layer and decades without significant information loss. The compari-
the substrate. As shown in the inset of Fig. 2, the chemical son of the evolution of films simulated with n points and 2n
potential ␮0 at the contact line is calculated following Eq. 共4兲 points validates the numerical accuracy of this method 共n
where ␦␩0 is taken along the substrate/film interface and ␦G0 ranging from 100 to 200兲. In the worst case, the time and
is the sum of the surface and interface energy contributions position where the film pinches off differs by less than 1.5%

再 冎
and 0.5%, respectively. The temporal convergence was ob-
2 ␥0 b␥⬘ tained by decreasing the time step dt until the breakup time tc
␮0 = ␮0 + ⫻ h0,0 − 0 − 共␥sub − ␥sub/film兲 共6兲 differed by less than 0.2% using dt and dt / 2 time steps.
b ᐉ0 ᐉ0
The code has been checked with three time-dependent
Furthermore, this procedure facilitates simulation of film problems described in the literature. First, an initial rectan-
motion with ␥sub − ␥sub/film = ± ␥0 corresponding to null or gular profile in free air converges at large times for any
180° contact angles. This point is difficult to implement with ␥-plot 共isotropic or not兲 to the thermodynamic equilibrium
other numerical methods when the contact angle is directly shape given by the Wulff’s construction where 兰s␥dS is
imposed as a boundary condition. minimum.21 Second, an initially sinusoidal profile, with
共ii兲 The flux of matter given by Eq. 共1兲 is simply calcu- small amplitude compared to its period, conserves its sinu-
lated by a finite-difference method along a segment between soidal shape and decreases its amplitude exponentially with
two consecutive points. For a 1D surface, the matter incre- time in agreement with the analytical solution z共x , t兲
ment ␦Ni attributed to the point M i 共i ⫽ 0 and i ⫽ n兲 is the = A exp关−B共2␲ / ␭兲4t兴sin共2␲x / ␭兲 关where A and ␭ are constant
algebraic sum of the fluxes along the two segments ending with A Ⰶ ␭, and B is given by Eq. 共3兲兴 discussed by
on M i multiplied by the time step dt. The triangle Mullins,14 Lançon,23 and Cahn.24 Third, the retraction of an
共M i−1 , M i , M i+1兲 will then grow or shrink depending on the isotropic semi-infinite film step on a substrate is reproduced


FIG. 4. Surface profile simulated for an isotropic surface energy

FIG. 3. Relative difference 共␣ − ␣y兲 / ␣y between the contact
共see Sec. IV for the numerical procedure兲. The initial state is shown
angle ␣ and the equilibrium contact angle ␣y as a function of time
in Fig. 1共a兲 with F = L / H = 500, the equilibrium contact angle is
normalized by the breaking up time tc. Simulations are carried out
160°, and the time is t = 13 000. The insets zoom in on the second
with n, 2n, and 4n points 共n = 75兲, an aspect ratio of F = 150, and
and third peaks. A fit of the profile with the function z共x , t兲
dimensionless adhesion energies corresponding to Young contact
= 1 + A共t兲 exp关−kx / x p共t兲兴cos关2␲x / x p共t兲 + ␸兴 is given with
angle of 45° and 135°. ␣ converges to ␣y at least four decades
A共13 000兲 = 6700, k = 3.87 共close to the value given by Srolovitz
共Ref. 24兲: 2␲ / 冑3 ⬇ 3.63兲, ␸ = 1.1 rd and x p = 48.33 x p共t兲 is found to
before tc.
vary as: x p共t兲 ⬇ 5.4t0.23.
in agreement with the results of Wong15 for the complete
range of contact angle. This point which will be detailed In the second stage, the film smoothes the sharp edge in
later, appears as a particular case of the finite film retraction order to reduce the surface energy in agreement with the
developed in Sec. IV A for isotropic surface energy. More numerical results of Wong15 and the analytical steady state
generally, the surface energy anisotropy effects 共with or solution derived by Srolovitz and Safran in the small slope
without cusp points兲 will be discussed in Sec. IV B. limit 共⳵z / ⳵x Ⰶ 1兲.25 Figure 4 plots an example of the profile
with a solution-like fit calculated by Srolovitz and Safran.
The first valley from the edge corresponds to the mini-
IV. SIMULATION OF THIN FILM RETRACTION mum film thickness and is denoted zmin 共Fig. 1兲. Its evolution
A. Isotropic surface energy
as a function of time for ␣ = 45° and for different aspect
ratios F = L / H is presented in Fig. 5共a兲 共H and L are respec-
As shown in Fig. 1, at the first stage of the simulation, the tively the initial film thickness and half length兲. At large
film profile changes at the contact point to reach the equilib- F 共more than about 500兲, a semi-infinite behavior is
rium contact angle ␣y given by the Young equation: observed.15zmin共t兲 decreases continuously with time and
␥sub − ␥sub/film = ␥0 cos共␣y兲 where ␥sub is the surface energy of reaches the substrate at tc. At this time, mass is detached
the substrate and ␥sub/film is the energy of the substrate/film from the film, and at least 3 agglomerates will be formed
interface. In the anisotropic case, ␥0 is replaced by ␥共␣y兲 and 关Fig. 1共d兲兴. For smaller F, the edge coupling induced by the
a nontrivial solution must be calculated. Note that this equa- surface diffusion mechanism cannot be neglected and zmin共t兲
tion is derived by calculating the variation of the surface free oscillates around the semi-infinite case until it strongly de-
energy caused by a small displacement ␦␩0 of the triple creases leading to the breaking point. Moreover, the total
point26 which corresponds exactly to our numerical method energy of the film, the retraction quantity x0, and the position
for the calculation of surface potential. Figure 3 shows that of the first minimum xzmin oscillate around the semi-infinite
the convergence of ␣ to ␣y is very fast and the local equilib- case with amplitude increasing with time. Two or three ag-
rium state is reached four orders of magnitude before the glomerates can be observed as in Figs. 1共c兲 and 1共d兲. If the
breakup time tc. It has also been checked that the equilibrium film breaks in the middle, 2 agglomerates are formed; if not,
state can be achieved even more quickly with a tightened at least 3 agglomerates are formed. It is worth noting that the
mesh, and that tc is, to a first approximation not significantly breaking time can be largely brought backward and can oc-
influenced by the early nonequilibrium state close to the cur a decade before the semi-infinite case. For even smaller
triple line. In consequence, we can consider that the contact F values 共F ⬍ 100兲, coalescence occurs and the film never
angle equilibrium is quasi-instantaneous compared to general breaks 关Fig. 1共e兲兴.
profile evolution. For the sake of simplicity, we will use the The influence of ␣ on zmin is shown in Fig. 5共b兲 for
contact angle ␣共t → ⬁ 兲 = ␣y in the text as a system variable F = 150. The breaking time tc is decreased with ␣. It indicates
instead of its equivalent in energy 共␥sub − ␥sub/film兲. In Fig. 3, that the coupling between the two film edges appears later
the variation of ␣ − ␣y at later time 共t / tc ⬎ 10−4兲 corresponds for acute angles. A limit between the regimes of weak and
to numerical noise and is mainly determined by the time strong edge coupling can be drawn. To further analyze the
step. dependence of critical breaking time tc on both F and ␣, Fig.


FIG. 6. 共Color online兲 Logarithm of the critical breakup time

tc : ln共tc兲 as the function of ln关1 / sin共␣ / 2兲兴 共␣: contact angle兲 for
different aspect ratio F. The points correspond to numerical
simulations for an isotropic surface energy, and the lines to
empirical adjustments. Two regimes can be identified. The first
one is the region where the edge-coupling effect is negligible.
tc共F , ␣兲 depends only on F, and can be fitted by: ln共tc兲
= 4.092 ln关1 / sin共␣ / 2兲兴 + 9.4225. In the second one, the edge cou-
pling cannot be neglected and tc is depending both on ␣ and F. tc is
relatively well fitted by the general law: ln共tc兲 = C1 / sin共␣ / 2兲 + C2,
where C1 = −0.328 ln共F兲 + 2.23 and C2 = 3.42 ln共F兲 − 7.39.

The fits of the boundaries between the different regions

given in the caption of Fig. 7, allows prediction of the frag-
mentation of the film of a given material only defined by its
FIG. 5. Evolution of the minimum dimensionless film thickness
adhesion energy 共i.e., ␣兲. The above phenomenology of the
zmin 共see Fig. 1兲 vs dimensionless time t for a contact angle of 45°
and for different aspect ratio F. The surface energy is considered to
film retraction will now be considered taking into account
be isotropic. F = 500 is close to the semi-infinite case. When zmin the anisotropy in the surface energy.
reaches 0, the film breaks up, and when zmin goes higher than 1, the
dip is filled up and the matter rearranges in single agglomerate.
zmin共t兲 oscillates around the semi-infinite case with amplitude in-
creasing with time, before the breakup or the coalescence. b兲 Mini-
mum dimensionless film thickness zmin vs dimensionless time t for
an aspect ratio F = 150 and for different contact angle ␣. Edge-
coupling effect and breakup time appear earlier with an increase of
the contact angle.

6 plots ln共tc兲 as a function of ln关1 / sin共␣ / 2兲兴for different F

values. Two clear regimes can be identified on this figure. At
large F and ␣, tc is independent of F and increases as ␣
decreases. In this region, no edge coupling effect is observed
and the behavior is close to the semi-infinite film case. A
Taylor development at small ␣ of the fit of the linear portion
in Fig. 6 共see figure caption兲 gives: tc = 211⫻ 103 ⫻ ␣−4 in
agreement with Wong.15 When decreasing F and/or ␣, tc fol-
lows a law depending on both F and ␣ that can be empiri-
cally fitted by ln共tc兲 = C1 / sin共␣ / 2兲 + C2 where where C1 and FIG. 7. 共Color online兲 Number of agglomerates calculated with
C2 linearly depend on ln共F兲 and are given in the caption of an isotropic surface energy vs 1 / sin共␣ / 2兲 共␣ is the contact angle兲
Fig. 6. and aspect ratio F. The points correspond to the numerical
The number of agglomerates 共1, 2, 3, and more兲 is given simulations. The dashed lines show the boundaries for the
in Fig. 7 in the 关sin共␣ / 2兲−1 , F兴 coordinates. In this represen- formation of 1, 2, and 3 or more agglomerates. They can be
tation, the domain boundaries correspond to straight lines fitted by F = 48.3/ sin共␣ / 2兲 − 4.33 for 1–2 agglomerates and
indicating the increase of the areas with F and sin共␣ / 2兲−1. F = 123.4/ sin共␣ / 2兲 − 24.6 for 2–3 and more agglomerates.


FIG. 8. Ratio of anisotropic to isotropic breaking time

共tc,aniso / tc,iso兲 vs the corresponding isotropic Young angle ␣ and the
aspect ratio F. The anisotropic surface energy follows Eq. 共7兲 with
␰ = 0.013 32 and ␪ = ␲ / 8. For large F and ␣, tc can be several times FIG. 9. 共Color online兲 Number of agglomerates calculated with
higher in the anisotropic case than in the isotropic one. an anisotropic surface energy 共defined by Eq. 共7兲 with ␰ = 0.013 32
and ␪0 = ␲ / 8兲 as a function of 1 / sin共␣ / 2兲 共␣ is the corresponding
isotropic Young angle兲 and aspect ratio F. The points correspond to
B. Anisotropic surface energy the numerical simulations. The boundaries between the numbers of
Surface anisotropy has been considered in two ways agglomerates 共solid lines兲 are shifted compared to the isotropic sur-
depending on the surface stiffness parameter ⌺共␪兲 face energy calculations 共dashed lines taken from Fig. 7兲.
= ␥ + ⳵2␥ / ⳵2␪ and on the existence of cusp points in the
␥-plot curve. The anisotropy is called mild for strictly posi- than in isotropic case, and tc,aniso / tc,iso increases with ␣ and
tive ⌺共␪兲 and severe in the other case.20 For mild anisotropy, F. On the contrary, when the coupling effect cannot be ne-
all plane orientations are stable and no faceting is expected.20 glected, tc,aniso does not vary more than 20% compared to
For severe anisotropy, missing angles exist in the Wulff’s tc,iso.The number of agglomerates vs. ␣ and F for this aniso-
construction. Cusp points in the ␥-plot curve have been in- tropic surface energy is presented in Fig. 9. The boundaries
troduced to fit the experimental results where spontaneous between the numbers of agglomerates 共solid lines兲 are shifted
faceting occurs. For consistency, the anisotropic and isotro- to large F and ␣ compared to the isotropic case 共dashed
pic cases will always be compared with the same nondimen- lines兲. To conclude with a smooth eightfold ␥-plot where the
sional adhesion energy ⌬W* = 共␥sub − ␥sub/film兲 / ␥0 − 1 and we plane parallel to the substrate has a minimum in energy, the
film gains in stability regarding to tc and to the number of
will consider the corresponding isotropic Young angle:
agglomerates. Simulations with ␰ ⬎ 1 / 63 共not shown here兲
␣y = cos−1共⌬W* + 1兲.
have also shown that this gain can be enlarged with severe
1. ␥-plot without a cusp point eightfold ␥-plot, and that tc is postponed even further with an
increase of ␰.
For the sake of simplicity, an eightfold normalized ␥-plot
The dependence of tc on ␪0 with the same eightfold ␥-plot
is assumed
has been checked with ␰ = 0.013 32 and ␪0 = 0, ␲ / 16, ␲ / 8,
␥共␪兲 and 3␲ / 16 corresponding to a ␥-plot rotation. For ␪0 = 0 and
= 1 + ␰ cos关8共␪ − ␪0兲兴. 共7兲 ␲ / 8, the plane parallel to the substrate is a maximum and a
minimum in surface energy, respectively. In the two other
␪ = 0 corresponds to the reference plane 共parallel to the sub- cases, the plane 共O , x , y兲 is neither a minimum nor a
strate兲, ␥0 is the average of the surface energy on ␪, and ␰, ␪0 maximum. The surface profiles are reported in Fig. 10 for
are two adjustable constants. This function is compatible t = 3000, ␣ = 135°, and F = 250 共i.e., without edge coupling兲.
with cubic symmetry,20,27 and yields to the surface stiffness: Table I gives the relative position of the first valley compared
⌺共␪兲 = 1–63␰ cos关8共␪ − ␪0兲兴. So, ␰ ⬍ 1 / 63 leads to a mild an- to the isotropic configuration. It is observed that for
isotropy and ␰ 艌 1 / 63 to a severe one. The experimental ␪0 = ␲ / 8 共0兲, the first valley is shallower 共deeper兲 than the
measurements of Bermont28 for silicon at 1323 K give a isotropic case. For ␪0 = ␲ / 16 共3␲ / 16兲, the position of the
weak anisotropy around ␥max / ␥min − 具␥共␪兲典 = 0.027 corre- minimum thickness moves away 共gets closer兲 from the edge
sponding to a mild anisotropy with ␰ = 0.013 32 关where than the isotropic case. This can be explained as follows. For
具␥共␪兲典 is the average of the surface energy on ␪兴. ␪0 is set to the ␥-plot of Eq. 共7兲, a minimum surface energy occurring at
␲ / 8 to fix an energy minimum at the 共O , x , y兲 plane parallel 共O , x , y兲 for ␪0 = ␲ / 8 corresponds to a maximum surface
to the substrate. Film motion has been simulated with these ␰ stiffness that locally favors a low curvature to maintain a low
and ␪0 values and for different ␣ 共0° to 180°兲 and F 共40 to gradient of ␮. The opposite behavior is obtained for ␪0 = 0.
500兲. Figure 8 shows the ratio of anisotropic to isotropic For ␪0 = ␲ / 16 and 3␲ / 16, the 共O , x , y兲 plane has the same
breaking time 共tc,aniso / tc,iso兲 as a function of 共␣ , F兲. For large energy as in the isotropic case, and zmin is similar. But, the
␣ and F, the critical breakup time tc is several times higher positions of the valleys differ along x. For ␪0 = ␲ / 16, the


FIG. 11. 共Color online兲 1.5 ␮m ⫻ 1.5 ␮m AFM image 共tapping

mode兲 of the edge retraction of a film of Si on SiO2 after a 30 mn
annealing at 950 ° C 共initial thickness 13.8 nm兲. The first peak is
96 nm high and is followed by a dip. This profile is similar to the
retraction simulations with an isotropic surface energy model.

of a 13.8 nm initial film thickness of Si on SiO2 close to the

initial sharp corner, after 30 min annealing at 950 ° C under
FIG. 10. Surface profiles corresponding to a film retraction with H2 atmosphere. This experimental profile is close to the pre-
an anisotropic surface energy without a cusp point. The eightfold diction of the diffusion motion model presented so far. How-
␥-plots and equation are shown in the inset. Four orientations ␪0 of ever, with similar experimental conditions 共13.0 nm film
the minimum in surface energy are drawn and discussed in the text. thickness, 2 min annealing at 900 ° C under H2 atmosphere兲,
The corresponding isotropic Young angle ␣ is 135° and dimension- a facetted profile can also be observed as illustrated in Fig.
less time is 3000. The film aspect ratio is F = 250, i.e., long enough 12. A matter front is observed with well-defined facets and
to avoid the edge-coupling effect. The initial sharp corner 关see Fig. without a dip in the film. This image has been analyzed by
1共a兲兴 was at the abscissa x = 0. use of a polar plot revealing the 兵113其 planes on the four
sides of the top of the matter front whereas the plane 共1̄1̄1兲 is
plane with the orientation ␪ = −␲ / 16 is more stable and the developed close to the contact line. Note that the top of the
valley moves away from the first peak to enlarge this plane peak seems not to be perfectly flat 共see Fig. 12兲. The differ-
orientation. Inversely, for ␪0 = 3␲ / 16, the plane with the ori- ence between the experimental results presented in Figs. 11
entation ␪ = ␲ / 16 is more stable and the valley gets closer to and 12 might be explained by the temperature difference.
the first peak to enlarge this plane. These interpretations stay Minima of the experimental ␥-plot are probably not very
valid for a large range of ␣ 共confirmed from 45° to 165°兲 and sharp.28,29 So, the ␥-plot must be very sensitive to wafer
larger ␰ 共in the mild and severe cases兲. contamination or structural defects. Actually, a fraction of
carbon monolayer on silicium can drastically modify the sur-
2. ␥-plot with cusp points: The case of Si(001) films
face energy anisotropy30 and can explain the highly facetted
It has been seen that with a ␥-plot without a cusp point, profile in Fig. 12.
severe or not, the geometry close to the first valley can be To fit the experimental dewetting profile shown in Fig. 12,
very different whereas the retraction quantity x0 of the con- the ␥-plot needs to be modified to include cusp points at the
tact line is roughly the same 共see Fig. 10兲. Moreover, in
every case, a dip below the initial thickness follows the first
peak. This is in agreement with the retraction of SOI at
950 ° C as seen in Fig. 11. This atomic force microscopy
共AFM兲 picture represents the three-dimensional 共3D兲 profile

TABLE I. zmin and xzmin 共see Fig. 1兲 carried out from the aniso-
tropic cases of Fig. 10, for different ␪0 compared to the isotropic
case. When the surface energy of the 共0 , x , y兲 plane is a minimum
共␪0 = ␲ / 8兲, the hollow is shallower, and inversely for ␪0 = 0. When
␪0 = ␲ / 16, the plane disoriented by −␲ / 16 from 共0 , x , y兲 is a mini-
mum of energy and the peak tends to develop this plane, so the
valley moves away from the first peak compared to the isotropic
case, and inversely for ␪0 = 3␲ / 16.

zmin xzmin
FIG. 12. 共Color online兲 Image: 1 ␮m ⫻ 1 ␮m AFM image 共tap-
␪0 = 0 ⬍ ⬇ ping mode兲 of the Si on SiO2 retraction after a 2 min annealing at
␪0 = ␲ / 16 ⬇ ⬎ 900 ° C 共the initial film thickness is 13.0 nm兲. The peak is facetted
␪0 = ␲ / 8 ⬎ ⬇ and reaches 52 nm high. No dip appears behind the 共113兲 plane.
␪0 = 3␲ / 16 ⬇ ⬍ Curve: Profile along 共0 , x兲. It is similar to the retraction simulations
with an anisotropic surface energy model.


FIG. 13. Schematics of 2D network of atoms placed at the in-

tersection of two dense planes P1 and P2. The disorientation angle
between P1 and P2 is ␪12. The distance between the P1 planes is a FIG. 14. Surface profiles corresponding to a film retraction with
and between the P2 planes is b. The disorientation angle of the an anisotropic surface energy with cusp point. The inset shows the
vicinal surface is ␪ from the reference planes P1. The terrace and ␥-plot with a 12-fold symmetry corresponding to the 11̄0 zone and
step length are respectively p and q. The distance between 2 con- to the 兵001其, 兵113其, 兵111其, and 兵110其 dense planes. ␥共␪兲 is given by
secutive steps is l. Eq. 共10兲 between these dense plane orientations. ␥113 is varied from
0.94 to 1.03 and the other dense plane energies are fixed to 1.00.
angles corresponding to the experimental minimum in en- Note that the dense plane extension on this profile directly depends
ergy, namely the 兵001其, 兵113其, 兵111其, and 兵110其 planes 共the on the surface energy ratio.
共1̄1̄0兲 plane cannot be probed on the AFM image, but 兵110其
planes have been observed on single agglomerates兲. A semi- the 11̄0 zone with cusp points corresponding to the dense
empirical analytic ␥-plot has been developed for this purpose planes 兵001其, 兵113其, 兵111其, and 兵110其 joined by a curve de-
within a simple bond-breaking model where surface passiva- fined by Eq. 共10兲. This ␥-plot results from a very simple
tion, reconstruction, relaxation, and step interactions are not model 共cut valence bonds model兲 but has the advantages that
considered. This can be described by considering two dense the minima can be adjusted to be realistic and the analytic
planes P1, P2 of surface energies ␥ P1, ␥ P2 separated by an form is suitable to simulate the dewetting. Here, the refer-
angle ␪12, and a vicinal plane P with a disorientation angle ␪ ence surface energy is ␥001 关noted ␥0 in Eq. 共3兲兴. To simplify
from P1共0 艋 ␪ 艋 ␪12兲 共see Fig. 13兲. Let l be the distance be- the writing, it was decided to use dimensionless energy and
tween 2 steps, x the length of the terraces, y the step length, to fix ␥001 = ␥110 = ␥111 = 1.00 and to vary ␥113. It is worth
a the distance between 2 atoms on P1, and b the distance noting that the profile time evolution can be simulated for
between two atoms on P2, as schematized in Fig. 13. The null surface stiffness for every orientation whereas it cannot
numbers of atomic bonds broken on P1 along the length p be done using the potential ␮ = 共␥ + ␥⬙兲␬. Figure 14 shows
and on P2 along the length q are, respectively that the film only exhibits dense plane orientations for ␥113 in

冉 冊
the 关0.94, 1.03兴 range. This is in agreement with the experi-
p l sin共␪兲 q l sin共␪兲
= cos共␪兲 − and = . 共8兲 mental profile given in Fig. 12. The dense planes observed in
a a tan共␪12兲 b b sin共␪12兲 the simulation depend on the surface energy ratio and can be
The energy of the P plane of length l is given by directly compared to the Wulff’s construction. From the
usual Wulff’s construction, it can be shown that the four
␥共␪兲l = ␥ P1 p + ␥ P2q. 共9兲 dense plane families are present in the equilibrium shape for
It gives the expression of the surface energy per unit length ␥c1 ⬍ ␥113 ⬍ ␥c2 with
of a plane with disorientation ␪ 苸 关0 , ␪12兴 ␥c1 = 3/冑11 ⬇ 0.9045, 共12兲

␥共␪兲 = ␥ P1 冉 cos共␪兲 −

+ ␥ P2
. 共10兲

␲ 1
+ arccos共冑2/3兲 − arccos共冑2/11兲 冊
For that particular ␥-plot, only dense plane orientations 4 2

冉 冊
␥c2 = ⬇ 1.125.
appear on the Wulff’s construction. It can be demonstrated ␲ 1 冑
cos − arccos共 2/3兲
that this type of ␥-plot is formed of portions of circles going 4 2
through the origin. To the first order, this ␥-plot is equivalent
to the one generally developed for a vicinal surface with a
step-free energy ␤ given by 共where h is the step height兲29 Below ␥c1, the planes 兵001其 are unstable and above ␥c2 the
␤ ␥ P2 ␥ P1 planes 兵113其 do not appear in the Wulff’s construction any-
= − . 共11兲 more. In Fig. 14, all four dense planes are developed for ␥113
h sin共␪12兲 tan共␪12兲
in the 关0.94, 1.03兴 range, in agreement with the stability do-
The ␥-plot used to simulate the experimental dewetting of mains. The relative facet extensions directly depend on the
共001兲 silicon on SiO2 共see the inset of Fig. 14兲 is defined by ratio of the surface energy of the dense planes. For example,


as ␥113 increases, the 共001兲 plane becomes larger in the occur. During the diffusion motion of a finite film, two re-
Wulff’s construction and appears larger on the top of the gimes can be clearly identified. At the beginning, the behav-
peak during dewetting. Moreover, the dip can be drastically ior of the film profile is similar to the semi-infinite geometry.
favored or almost vanished depending on ␥113. As ␥113 in- Then, edge-coupling effects due to surface diffusion may be
creases, the dip is shallower and larger, and the pinch off is observed if the film is short enough. The aspect ratio as well
postponed. A small variation of ␥113 yields large modification as the energy of the contact line largely influences the break-
of the breakup time and phenomenology. For ␥113 = 0.94, the ing time and the phenomenology. An empirical law is pro-
breakup time is comparable to the isotropic case. For posed, predicting the fragmentation of the film defined only
␥113 艌 0.97, it is more than fifteen times larger than the iso- by these two parameters.
tropic case whereas for ␥113 = 0.905, it is more than sixty Experimental measurements have shown that annealing of
times lower. In the range ␥113 苸 关␥c1 , 1.00兴, the breakup time different thin silicon patterns on amorphous silicon oxide
varies over three orders of magnitude. Even for ␥113 ⬎ ␥c2, a within similar experimental conditions 共thermal annealing
very shallow dip is still observed but the 共111兲 plane is de- from 900 ° C to 950 ° C under H2 atmosphere兲 can lead to a
veloped on the right of the peak instead of 共113兲. For smooth or facetted dewetting profile. The experimental fac-
␥113 艌 1.10, ␣ = 90°, and F = 500, the film does not breakup etted profiles have been fitted qualitatively by a ␥-plot with
and rearranges itself in a single agglomerate. For ␥113 ⬍ ␥c1, cusp points and small anisotropy. The simulated and experi-
the peak does not exhibit a 共001兲 plane anymore which is in mental profiles of Figs. 14 and 12 are in qualitative good
agreement with stability domains, and the rest of the film agreement for a ␥-plot following Eq. 共10兲 and plane energies
set around 1.00. For a slightly modified ␥-plot, the phenom-
develops microfacets with 共1̄1̄3兲 and 共113兲 planes. For dif-
enology and the breaking time can be very different. For
ferent ␣, the phenomenology and the breaking time are also
example, changing the surface energy at a cusp point by less
drastically modified by a very small variation of ␥113. In
than 7% can introduce a variation of the breaking time over
consequence, an accurate description of the ␥-plot features is
three orders of magnitude. So, to be able to use this to pre-
needed to predict the facet morphologies. This study also
dict film behavior, a very high accuracy of the ␥-plot de-
indicates that surface tension anisotropy can favor the film
scription is needed. Nevertheless, for a mildly anisotropic
stability from the pinch off point of view. However, the film
␥-plot with ␥001 being a minimum, the pinch off is post-
retraction x0 is comparable 共within 20%兲 to the isotropic case
poned. It can even be annihilated with ␥-plots having cusp
共see Figs. 10 and 14兲. Indeed, the value and speed of retrac-
points. Therefore, from the mass detachment point of view,
tion seem to be mainly governed by the contact line equilib-
the stability domain can be extended thanks to surface en-
rium, and not by anisotropy effects.
ergy anisotropy. At any given adhesion energy, our simula-
tions show that the retraction quantity x0 slightly depends on
V. DISCUSSION AND CONCLUSION the ␥-plot. The contact line retraction speed is not affected
by the surface energy anisotropy which mainly affects the
A numerical method for the calculation of surface chemi-
pinch off phenomenology. It appears from this study that the
cal potential has been developed that goes beyond the usual
best way to decrease the retraction speed of the film is to
surface potential developed by Herring22 which is limited to
increase the adhesion energy. The anisotropy of surface dif-
small curvature and cannot be applied to a severe anisotropy
fusion coefficients has not been discussed in this paper, and it
where faceting is expected. This method accommodates ev-
could be interesting in future work to study the interplay
ery anisotropic feature of the surface energy 共or diffusion兲
between the faceting imposed by the surface energy and the
and can be applied to a large number of physical problems
surface diffusion coefficient anisotropies. Elastic energy
such as hole opening, and thermal grooving as well as to
could also be taken into account to simulate strained films
nonuniform substrate interfaces 共nonplanar geometry and/or
that are currently used in microelectronics. Finally, 3D simu-
nonconstant interface energy兲. The surface energy is de-
lations could be performed with the same numerical method
scribed by a so called ␥-plot where cusp points may occur. In
共including anisotropy and/or strain effects兲 to get better un-
this paper, the retraction of finite thin films defined by their
derstanding of the highly symmetric experimental patterns
aspect ratio has been analyzed considering only the surface
obtained during solid-state thin film retraction.6
diffusion mechanism. The contact line between the film and
the substrate is defined by its energy, and the contact angle ACKNOWLEDGMENTS
results from local equilibrium conditions. This is a natural
result from the simulation and no numerical artifact constrain We wish to acknowledge O. Cueto, D. Jamet, and B.
the simulation in any way. For every ␥-plot, the film is sta- Mathieu for advice in numerical computation. We are grate-
bilized in its equilibrium shape given by the Wulff’s con- ful to F. Lançon and J. Villain for many useful scientific
struction or by the analytic procedure given by Khare.31 discussions. We thank Y. Campidelli and Y. Bogumilowicz
The surface diffusion mechanism produces a minimum of for technical assistance with the AFM instrumentation. Our
the surface thickness below its initial value. As this mini- gratitude is also extended to J-M. Hartmann for the time
mum deepens, the film may break up and mass shedding may spent supporting the annealing experiments.


*Corresponding author. Electronic address: erwan.dornel@cea.fr Mater. 48, 1719 共2000兲.

C. Jahan, O. Faynot, L. Tosti, and J. M. Hartmann, J. Cryst. 16
C. Herring, J. Appl. Phys. 21, 301 共1950兲.
Growth 280, 530 共2005兲. 17
W. W. Mullins, J. Appl. Phys. 28, 333 共1957兲.
2 Y. Ono, M. Nagase, M. Tabe, and Y. Takahashi, Jpn. J. Appl. 18 H. Wong, M. J. Miksis, P. W. Voorhees, and S. H. Davis, Acta

Phys., Part 1 34, 1728 共1995兲. Mater. 45, 2477 共1997兲.

3 R. Nuryadi, Y. Ishikawa, and M. Tabe, Appl. Surf. Sci. 159-160, 19 B. J. Spencer, P. W. Voorhees, and S. H. Davis, J. Appl. Phys. 73,

121 共2000兲. 4955 共1993兲.

R. Nuryadi and Y. Ishikawa, J. Vac. Sci. Technol. B 20, 167 20
W. Zhang and I. Gladwell, Comput. Mater. Sci. 27, 461 共2003兲.
共2002兲. 21
G. Wulff, Z. Kristallogr. 34, 449 共1901兲.
5 Y. Ishikawa, M. Kumezawa, R. Nuryadi, and M. Tabe, Appl. 22 C. Herring, Physics of Powder Metallurgy, edited by Walter E.

Surf. Sci. 190, 11 共2002兲. Kingston 共McGraw-Hill, New York, 1951兲.

6 Y. Ishikawa, Y. Imai, H. Ikeda, and M. Tabe, Appl. Phys. Lett. 23 F. Lançon and J. Villain, in Kinetics of Ordering and Growth at

83, 3162 共2003兲. Surfaces, edited by M. G. Lagally 共Plenum Press, New York,
Y. Wakayama, T. Tagami, and S. Tanaka, J. Appl. Phys. 85, 8493 1990兲.
共1999兲. 24
J. W. Cahn and J. E. Taylor, Acta Metall. Mater. 42, 1045 共1994兲.
8 B. Legrand, V. Agache, J. P. Nys, V. Senez, and D. Stiévenard, 25 D. J. Srolovitz and S. A. Safran, J. Appl. Phys. 60, 255 共1986兲.

Appl. Phys. Lett. 76, 3271 共2000兲. 26 P. Roura and J. Fort, J. Colloid Interface Sci. 272, 420 共2004兲.
9 B. Legrand, V. Agache, T. Mélin, J. P. Nys, V. Senez, and D. 27 G. B. McFadden, S. R. Coriell, and R. F. Sekerka, Acta Mater.

Stiévenard, J. Appl. Phys. 91, 106 共2002兲. 48, 3177 共2000兲.

10 28
M. Czubanowski, C. Tegenkamp, W. Ernst, and H. Pfnür, Appl. J. M. Bermont, J. J. Métois, X. Egéa, and F. Floret, Surf. Sci.
Phys. Lett. 84, 350 共2004兲. 330, 48 共1995兲.
C. M. Kennefick and R. Raj, Acta Metall. 37, 2947 共1989兲. 29
J. M. Bermont, J. J. Métois, J. C. Heyraux, and F. Floret, Surf.
E. Jiran and C. V. Thompson, J. Electron. Mater. 19, 1153 共1990兲. Sci. 416, 430 共1998兲.
13 E. Jiran and C. V. Thompson, Thin Solid Films 208, 23 共1992兲. 30 Private discussion with J. J. Métois and A. Saul.
14 W. W. Mullins, J. Appl. Phys. 30, 77 共1959兲. 31 S. V. Khare, S. Kodambaka, D. D. Johnson, I. Petrov, and J. E.
15 H. Wong, P. W. Voorhees, M. J. Miksis, and S. H. Davis, Acta
Greene, Surf. Sci. 522, 75 共2003兲.