HAL Id: jpa-00247194
https://hal.archives-ouvertes.fr/jpa-00247194
Submitted on 1 Jan 1996
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Current Induced Faceting in Theory and Simulation
Harvey Dobbs, Joachim Krug
To cite this version:
Harvey Dobbs, Joachim Krug. Current Induced Faceting in Theory and Simulation. Journal de
Physique I, EDP Sciences, 1996, 6 (3), pp.413-430. �10.1051/jp1:1996166�. �jpa-00247194�
Current Induced Faceting in Theory and Simulation Harvey Dobbs (*)
and JoachimKrug (**)
Institut für
Festkôrperforschung, Forschungszentrum,
D-52425Jülich, Germany
(Received
8September
1995, received in final form 9 November,accepted
10November1995)
PACS.68.35.Ja Surface and interface
dynamics
and vibrations PACS.68.35.Rh Phase transitions and criticalphenomena
PACS.05.70.Ln Non
equilibrium thermodynamics,
irreversible processesAbstract. We consider the effect that a directed
migration
of adatoms, such as that whicharises due to an
externally applied
electricfield,
bas over themorphology
of acrystal
surface.Applying
linearstability
arguments to a continuum model we find that thestability
of a surface is determmedby
thedepeiidence
of the adatommobility
on the surface orientation. In onedimension,
surfaces may be either stable or unstable. In two dimensionshowever~
we find thata surface of general orientation is always unstable. These results are confirmed
by
computer simulations of solid-on-sohdmodels,
which also show late time coarsemng behaviour in cases of surfaceinstability,
such that the typical domam size is seen to increase as a power law of thesimulation time. Our numerical data demonstrate a
growth
exponent1/4
inone dimension and
1/2
in twodimensions,
which can besupported by
the continuumtheory.
1. Introduction
In 1938 Johnson
[ii, investigating
the surfacemorphology
oftungsten
filaments in bumed-out incandescent
lamps,
made a remarkable observation: the surfacesappeared
smooth if thelamps
had beenoperated
onaltemating
current, butdeveloped
a charactensticpattern
ofripples
under d-c- conditions. Hepointed
out that the effect could bequahtatively explained by
the directedmigration
oftungsten
atoms under the influence of the electricfield,
if it wereassumed that 'the rate
of drift depends
on thetype of underiying surface'.
At about the same
time, Bagnold
[2]developed
his celebratedtheory
ofripple
formationon wmd blown sand.
Despite
thevastly
different scalesinvolved, closely
related mechanisms underlie the twophenomena
observedby
Johnson andBagnold.
In both cases there is a flow of materialalong
a surface. If the flow ratedepends
on the local surfaceslope
in such a way that the flo~v islarger
in theuphill
direction than in the downhilldirection,
then localprotrusions
areamplified
and anpple morphology develops (Fig. 1).
For solid surfaces aslope dependence
of the flow would begenerally expected
because ofcrystalhne anisotropy.
In thecase of sand
ripples
the situation is somewhat more subtle: there the main mass flowalong
the surface is carriedby
creepinggrains,
which receive their motion energy from collisions with'saltating'
grainssuspended
in the air and carriedalong by
thewind;
the momentum transfer(*)
Present address: The BlackettLaboratory,
ImperialCollege,
London SW7 2BZ, UnitedKingdom.
(**)
Author forcorrespondence (e-mail: jkrug©iff079.iff.kfa-juehch.de).
Present address: FachbereichPhysik,
Universitàt GH Essen, D-45117 Essen,Germany
Q
LesÉditions
dePhysique
1996(a) ~b)
~ ~
/~
i
~ ~
i
fl
,
Fig.
1. Basic mechanism of surfaceinstability
due to atangential
mass flow.(a)
If themass current is an
increasing
function ofslope,
then smallheight
fluctuations areamplified
and the surface becomesunstable
(dashed arrow);
in the opposite case 16) the current stabilizes the flatconfiguration.
from the
saltating grains
islarger
on the windward side of aprotrusion
than on the leeside,
and
consequently
theuphill
flow is enhanced [2]. The firstquantitative expression
of the effect(for
solidsurfaces)
wasgiven by Frohberg
and Adam[3],
who showed how thecompetition
between triedestabilizing
surface flow andsmoothening
due to surface diffusiongives
rise to a selectedwavelength
at which theinstability
first appears.Recent interest in current-induced surface instabilities has been
triggered by carefully
con-trolled
experiments
on surfaces vicinal toSi( ii1),
where a direct bulkheating
current was found to give rise to avariety
of stepbunching phenomena [4-8].
The characteristicdependence
of the stepbunching
on current direction(relative
to themiscut) implicated
a mechanism related tosurface eiectromigration [9],
as had beenoriginally proposed by
Johnson[ii.
Inspired by
theseexperiments, Stoyanov [10] developed
a one-dimensionalstep-train
model of current-inducedstep bunching,
in which trieelectromigration
flux induces anasymmetry
in triestep dynamics
but trie motion of adatoms is notexplicitly
treated. This work wassubsequently
extended to indude transverse step
meandering [11]
andadvacancy
motion[12],
and bas been shown toprovide
agood description
of trieexperimentally
observedphenomenology
[8].A certain drawback of
step-dynamical
models is their limitation to vicinal surfaces as we baveemphasized
in triepreceding discussion,
current-induced instabilities occur under muchmore
general conditions, essentially
whenever there is a directed flow of materialalong
a surface.We bave
recently presented
a continuumtheory
of current-inducedfaceting,
which is better suited to describe trie uniuersai features of triephenomenon [13].
Besidesproviding
ageneral
criterion for the
stability
of agiven
surface orientation [3], thetheory predicts
the orientations of the facets that appear after the initial orientation has becomeunstable,
and it allo,vs one to address thesubsequent
coarseningdynamics
of the faceted surface.The continuum
theory
is based on twoorientation-dependent, macroscopic quantities
the surface stiffness and the adatommobility
which have to be obtained from amicroscopic,
statistical mechanical model
[14].
~Ve have thereforesupplemented
our continuumanalysis
with àlonte Carlo simulations of a solid-on-solid
(SOS model,
in which the standard transition rates for surface diffusion [15] are given a directional bios to account for theelectromigration
effect. Once the
correspondence
betweenmacroscopic
andmicrosopic quantities
has been established[14],
the continuumtheory
can be used topredict
themorphological
evolution of the SOS surface.In this paper we
give
a detailed account of the work announced in[13],
and extend itin several directions. In Section
3,
we show how the orientationdependence
of the surface stiffness can be included in the calculation of the selectedfacets,
and illustrate theprocedure
for the standard SOS and the discrete Gaussian models [16] in onedimension;
in Section 4 weanalytically
derive thet~R coarsening
law that~&ras found
numerically
in our one-dimensionalsimulation
[13]; and,
mostimportantly,
Section 5 containsanalytic
and numerical results for two-dimensionalsurfaces, demonstrating
inparticular
that the facet structure coarsens ast~/2
in this case, in accordance with recent
experiments
onSi(1ii) [17].
Section 2 introduces thediscrete and continuum models used in this
work,
and Section 6 offers someconduding
remarks.2. SOS and Continuum Models
2.1. A ONE-DIMENSIONAL SO S MODEL. For
making microscopic
simulations weadopt
the usual SOSdescription
of acrystal
surface in terms ofinteger height
variablesh(1),
=1,. L,
with HamiltonianL
~ = K
~j (h(1+1) h(1)(". (1)
~=i
We consider both the standard n
= SOS model and the n
= 2 Gaussian model. An average surface
slope
ù isimposed by adopting
'helical'boundary
conditionsh(L
+1)
=h(1)
+fiL, although only
with aninteger
value for the vertical offset ùL.During
simulation of surfacediffusion,
an adatom on arandomly
chosen columnattempts
tostep
to arandomly
chosenneighbouring
columnj
=1+1,
so thath(1)
-h(1)
-1 andh(j)
-
h(j)
+1. Moves to theright (j =1+1)
areattempted
withprobability
p,
and moves to the left with
probability
1- p. Anattempted
move is thenaccepted
with aprobability r~j,
known as the transition rate.For
equilibrium simulations,
moves to theright
areattempted
withequal probability
asmoves to the
left,
p =1/2,
and the transition rates mustsatisfy
detailed balance withrespect
to
~,
F~ /Fj~
=exp(-A7i~ (2)
where
A~~j
is thechange
in the value of ~ associated withmoving
apartide
from toj (note
that the
temperature
has been adsorbed into the definition of thecoupling
K in(1) ). Apart
from theserequirements,
thespecific
choice ofdynamics
is unrestricted and inprinciple
does not effect the outcome of the simulation.The influence of
electromigration
can be indudedby considering
moves of adatoms to theright
withprobability
p different from1/2.
Thus for p >1/2
there is a bias for moves to theright-
The choice of transition ratesr~j
is Rot altered.Microscopically,
an adatom on the surface of acurrent-carrying crystal
issubject
to the combined effect of the electrostatic field Edriving
thecurrent,
and the 'wind force' due to the current itself. The total force F on theatom can be wntten as
F = eZE
(3)
which defines the effective valence Z. The bios p is then
simply
related to Fby p/(1 p)
=
exp(Fa/kT)
where a is the distance betweenadjacent
surface sites. ForCu(111)
at roomtemperature
it has been estimated[18]
that Z m-21,
so that for an electric field of1mV/cm
the relative bias introduced into the
hopping
of an adatom(p 1/2)
istypically
of order10~~ 10~~
ForSi(111)
the valence isexpected
to beconsiderably
smaller Z m 0.01 ù-1[8,19]
but with thecorrespondingly higher
fields used in trieexperiments (E
= 7000
mV/cm
in [8]
),
the bios is of the same order ofmagnitude.
With such a smallbias,
surface instabilitiesrequire
tens of seconds orlonger
todevelop,
and simulation over such alength
of time isexpensive. However,
we find thephenomena
causedby
the bios to beindependent
of itsmagnitude
in allrespects
other than the time-scalethey
take todevelop,
and so we consider muchlarger
biases(p
= 0.7 or
0.8)
in our simulations.Finally,
we comment about our choice of transition rates. Fordynamic
simulations of surfacediffusion,
rates chosenaccording
to trie Arrheniusprescription
aregenerally adopted, r~j
=
exp(-2Kn~)
where n~ =0, 1,
or 2 is trie lateral coordination number of trie adatom to be moved from column[20].
These rates are based on theassumption
that 1) surface diffusion isan activated process, and
ii)
the activation barrier isequal
to thebinding
energy at the initial400
t=10° t-io7
t=io6
zoo t=10 ~
B #
t=105
~ ~
t=lo~
° t-io4
-lùÙ t=10~
-200 -200
0 100 200 300 400 500
a) b)
Fig. 2.
Faceting
in the biased one-dimensional n= 1 and n
= 2 SOS models. A
single
reahsation of trie surface issho~vn,
at times measured inattempted
moves per site which are indicated above theprofile. Subsequent profiles
ai-edisplaced by
100 units in the vertical direction forclarity.
For then = 1 model
la),
data for ù= 0, K
= 2.0 and p
= o-7 is sho~vn. For the n
= 2 model
16),
we have taken ù = -o.25, K= I.o and p = 0.8.
site. The latter
clearly
constitutes anoversimplification,
since itimplies
that thehopping
rater~j
isindependent
on the sitej
= 1+ to which the adatom is
moved;
as a consequence, the bios has no effect ~vhatsoever on the surface structure[14].
In our simulations we therefore useMetropolis
rates, in which the barrier energy is identified with the energydijference
between final and initial site,r~j
= min[1, exp(-A~~)] (4)
While these rates are not
necessarily
more realistic thon Arrheniusdynamics, they
reflect thatgenerically hopping
rates shoulddepend
on both the initial and the final environment of the adatom [14] such adependence
appears to be crucial for the effects of interest in this paper.Figures
2a and 2b show the results of simulation of biased n= 1 and n
= 2 models. Both simulations
began
from surfaces with a small uniformtilt, h(1)
=int(ùi).
For the n= 1
model,
the surfacedevelops
into regions of uniformnegative slope (facets), separated by
narrowregions
oflarge positive slope (macrosteps),
which are akin to thestep
bunches observed inexperiments.
As the simulation progresses, the
regions
ofnegative slope
coalesce and growlarger although
their average
slope
remains constant. In contrast themacrosteps
do not growappreciably
in width but do grow inheight.
thus preserving the averageslope
of the surface. For the n= 2
model,
the situation differs in that facets of twodifferent,
finiteslopes
appear- Theslope
of these facets does notchange
as the simulation continues, and there are nomacrosteps.
This
mstability
of a flat surface is observed to occur for all values of p different from1/2, although
the time taken for the facets to form increases as the bias is reduced.Further,
as far as one cantell,
theslope
of the facets isindependent
of the bios as p-
1/2+ Trivially.
the
system
is invariant under simultaneous reversal of the bias p - 1- p and the average tilt ù - -ù, so that we willonly
consider cases p >1/2.
As well as
investigating
the effect of thedegree
ofbias,
we can alsoinvestigate
the effect oftrie average surface
slope
ù. Triepicture
that emerges for trie n = 1model,
is that when facetsform,
trieslope
of trie facets isindependent
of trie average surface tilt.However,
trie surface will form facetsonly
if the averageslope
is intermediate between that of trie facets and that of themacrosteps (+oc).
Surfaces with an averageslope
less than that of the facets are stable andremain unfaceted.
Thus,
the facet orientation appears to coincide with theboundary
between surfaces stable and unstable to current inducedfaceting.
For the n= 2
model,
it is found that a uniform surface of any tilt other thanprecisely
that of one of trie facet orientations isunstable. Trie surface will evolve into facets of two orientations: one with a
slope greater
than and trie other with aslope
less than the average surfaceslope.
2.2. CONTINUUM MODEL. Since bath facets and
macrosteps
appear on a scalelarger
than that of theunderlying lattice,
it ispossible
to describe their formation andgrowth using
acontinuum
theory
for the surface. Thisreplaces
the discrete column labels iby
a continuous coordinate ~, and describes the surface in terms of a continuousheight
variableh(~,t).
Becausewe
only
allow surface diffusion processes to occur, the surface evolvesaccording
to anequation
ofcontinuity
ôth
+ôzJ
" 0
(5)
where
subscripts
denote the differentiationvariable,
andJ(~,t)
is the surface adatom current.This is 'driven'
by
localgradients
of the chemicalpotential
~1, as well as theelectromigration
force
F,
so thatJ =
a(ô~h) [-ô~~l+ Fi (6)
The constant of
proportionality
a is called trie 'adatommobility'
and we bave written inits
dependence
on the local surfaceslope ô~h.
To relate the SOS mortel to the continuumdescription,
we need to evaluate both the chemicalpotential gradient
and themobility
interms of trie
microscopic parameters.
Trieprocedure
forthis,
which we outlinehere,
can befound in a
previous publication [14].
To calculate the chemical
potential
in anon-equilibrium situation,
we imagine an SOS surface with a Hamiltonian as inequation il)
in which the averages of the localheight
variables(h(i))
are
slowly varying
withtime,
so thatiocaiiy
the mortel is inequilibrium.
We then consider an SOS mortel with aslightly
different Hamiltonian for which the averageequilibrium profile
is made to match that of theclose-to-equilibrium
surfaceby adding
aninhomogeneous
chemicalpotential
term to ~~p
=~j ~1(1)h(1). (7)
~
This can be rewritten in terms of the local
slope
variableu(1)
=h(1+1) h(1)
and a local'slope potential' m(1)
which satisfies ~1(1)=
-(m(1) m(1- 1))
~iv
=£mli)U(1). (8)
Since the local
slope
variables are notcoupled by
either 7i or7ip,
it is easy to evaluate their thermal averages(u(1)).
For the n= mortel this can be clone
analytically
~~~~~~
coshll~lllm(il'
~ ~' ~~~For the n
= 2
mortel, explicit
numerical evaluation isquite
easy. To calculate the chemicalpotential,
we now invert trierelationship
betweenm(1)
and(vii)),
and take trie continuumhmit for trie coordinate 1,
~ ~ ~
~~~
~~~~ ÎÎÎ dÎ
d~2 ~~~~
where we
drop
trie bracketsil
for trie averagesu(~)
andh(~).
If a surface is close toequilibrium,
this relation holds
although
with apartial
second derivative. Trie factormultiplying
trie second derivative isjust
trie surface stiffness1,
and it is a function of trie localslope u(~).
Trie calculation of trie adatom
mobility presents
more of achallenge. According
to linear responsetheory, given
aspecific microscopic
model and aspecific
set ofdynamical rules,
triemobility
can inprinciple
be calculated from aknowledge
of the averagejumping
rate andtrie current-current correlation function in
equihbrium [21].
Inpractice,
it isusually
verycomplicated
to calculate triedynamic
correlation function for trie surface currents. However trie average of trieequilibrium jumping
rate can be calculated for one-dimensional models and this forms an upper limit for triemobility
of ~(r~j)
ea+ (11)
It is easy to calculate
a+
for trieMetropohs dynamics
as a function of either trie localslope potential
m or trie localslope
u. For trie n= 1 model
~+
(y~~)=
~°~~ ~
~~P(~~~
~j~~)
2 Siuh K
This
expression
bas a minimum at m = 0(u
= 0 and tends to
1/2
for m- +K
(u
-+oc).
Simulations of trieequihbrium roughening
of a surface indicate thata+
is in fact a fairapproximation
to triemobility [14].
For trie n
= 2 model
only
a numerical evaluation ispossible, although
it is easy to see that botha+
anda are
periodic
functions of trieslope potential
m and trie surfaceslope
u since trie Boltzmannweights
of triestep
variablesu(1)
for trie Hamiltonian 7i +7ip
are invariant up to a factor under transformation of trieslope potential
and trieslope m(1)
-m(1)
+2K~
u(1)
-vii)
+ 1. Hencea(m
+2K)
=
a(m)
and~(m
+2K)
=
u(m)
+ 1.Putting together
trieequation
ofcontinuity (5)
and trie formula for the surface current(6),
we thus arrive at the
required
continuumequation
for the surfaceôth
=-ô~ (a(ô~h)(ô~(1(ô~h)ôjh)
+F)) (13)
In fact we will
study
the time evolution of the localslope potential m(~),
because this removes the stiffness from theequation
andgives
a linear second derivative m the expression for thecurrent. To do this we differentiate
equation (13)
with respect to ~,giving
anequation
for thedynamics
of the localslope u(~),
which can then be convertedusing
trie relation to trieslope potential
ôtm
=1)) ai ja(m) (à]m
+F) j. (14)
3. One-Dimensional
Analysis
With trie continuum
theory
of trie lastSection,
we nowtry
to understand trie simulation results.Our first aim is to look at trie circumstances under which a surface is unstable when a field is
applied,
and we do thisby
using a hnearstability analysis.
Dur second aim is toprovide
a
description
of trie facets andmacrosteps
that appear, inparticular
we wish toexplain why macrosteps
appear in trie n= but not trie n
= 2 model- To do this we
analyse steady
states of trie continuumequations.
3.1. LINEAR STABILITY. For a flat surface of average
slope
ù with a smallperiodic
distur- bancesuperimposed
h(~,t)
= ù~ +hq exp(~oqt) cos(q~) (15)
a
simple
lineansation ofequation (13)
shows that triegrowth
rate ~oq is~dq =
F~q~ a(ù)ilù)q~. l16)
Although
thesign
of the q~ term isalways negative,
that of theleading quadratic
term is determinedby
thedependence
of themobility
a on the averageslope
of the surface [3]. Thus if themobility
is anincreasing
function of theslope,
~oq ispositive
for small values of q and the surface suffers from along wavelength instability,
whereas if a decreases withincreasing slope
the surface islinearly
stable. This isprecisely
the effect illustrated inFigure
1- Forinstance,
in the case of the n= 1 SOS model for which the
mobility a(ù)
has asingle
minimum at ù = 0 and increases to a finite limit for ù -+oc,
surfaces withpositive slope
arelinearly unstable,
and those withnegative slope
arelinearly
stable.This
simple macroscopic
criterion isreproduced by
the one-dimensional Burton-Cabrera- Frank(BCF)
model forstepped
surfaces[10]. Quite generally
thevelocity
of astep edge
isproportional
to the sum of the currentsj+ impinging
on thestep edge
from the terrace below(+)
and above(-).
Thus trieequation
of motion for trieposition
of trie i'tristop
xi has trie formÎ~
=-J+ix~ x~-i) J-(x~+i x~) iii)
where we take the x axis as
pointing
up the step trainii-e-
the surfaceslope
ispositive).
The linearstability
of a uniformstep
train of terrace width 1is then determinedby
thedependence
of the currents on 1: forstability
( (J+(1) -1-(1))
> °.l18)
The term in brackets is minus the function
#(1)
introducedby Stoyanov
in his discussion[10].
In cases without
desorption
of adatoms from terraces, the current is uniform over thelength
of each
terrace,
so thatj+
=-j-
and isequal
to themacroscopic
surface current J. Thus the condition(18)
holds when the surface current is adecreasing
function of the surfaceslope Ill-
For small
driving fields,
the surface current will beproportional
toF,
and a surfacemobility
can be defined
accordingly
which then satisfies the samestability
criterion as in the continuumtheory.
We also remark that an
analysis
ofperiodic perturbations
to a step train shows that it is either stable or unstable for allwavelengths.
The lack ofstability
for smallwavelengths
is due to trie absence ofcapillarity
effects in trie BCF model.Although
this omission is notimportant
fordetermining
whether a surface is unstable tolong wavelength perturbations,
it is seriouswhen one wishes to consider trie structure of any step bunches that
form,
since trie surfacecapillarity
is a consequence of therepulsive
forcesoperating
between the surface steps-So far our discussion bas been limited to l1~lear
stability,
which is not aguarantee
ofstability.
Thermal fluctuations cause local
regions
of a stable surface to become tilted over toward orien- tations that areunstable,
so that furtherinstability
may occur. Hence it ispossible
for surfaces which arelinearly
stable to be in factunstable, although
the converse is never true. As anexample,
the simulations described in Section 2 seem tosuggest
that the actualboundary
be-tween stable and unstable surfaces for the n = 1 model coincides with the selected orientation
of the facets
11
-30°),
which lies well inside theregion
of linearstability (ù
<0)-
A similarsituation may have been observed in
experiments
on surfaces vicinal toSi(001),
where step trains withlarge step spacings (small slopes)
were found to be unstable bath in the'step up'
and the'step
down' current direction[22]
within linearstability analysis,
reversing trie currentdirection should
àlways
transform a stable vicinal surface into an unstable one, and mce versa.To go
beyond
the linearstability,
we now consider'steady
state' solutions toequation (13),
which allows us to calculate the Lacet
slopes,
3.2. STEADY STATES. Because the
groin>th
rate of the facets that form after the initialinstability
in either the n = 1 or 2 models decreases as the facets grow, theirasymptotic shape
can be calculated
by setting
the time derivative inequation (13) equal
to zero- Because of the one-dimensional nature oî theproblem,
it follows from theequation
ofcontinuity
that the s~arface current is a constant J* for all values of x.Using equation (6)
and the formula for the chenlicalpotential,
thequasi-static
surfaceprofile
is then determinedby
a second-orderequation
for trie localslope i~(~)
a(u) () ~i(u)))
+
F~
= J*.(19)
1 X ~
A
simple approximation
would be toignore
theslope dependence
of the surface stiffness§ [13].
However,
such anapproximation
is to be avoided instudying
the formation ofmacrosteps
since in the case of the n
= 1 SOS model the surface stiffness~vanishes for
large
surfaceslopes-
Instead we con rewrite
equation (19)
in terms of the localslope potential m(x)
(j
=
J*laIn~)
F.12°)
In
solving
thisequation
we look foroscillatory solutions,
so that there must be an inflexion point at somerepeated
value m*. This is then related to the surface currentJ*
=
Fa(m*). j21)
We can
explore
the character of solutions to(20) by mabing
ananalogy
with the motion ofa
partide
in apotential
wellV(m),
where ~plays
the role of the tiine variable. In this casev(m)
= F
j~(1 a(m*) la(m'))dm'. (22)
Note that for
oscillatory solutions,
m* must be a minimum of thispotential
so thata'(m*)
> 0.Typical shapes
of thepotential
for trie n= 1 and 2 SOS models are shown in
Figure
3. In bothcases the
mobility
is an even function of trie surfaceslope,
so thatV(m)
is an odd function ofm with a ma~imum at -m*. For the n
= mortel this maximum is unique, whereas for the
n = 2
model,
thepotential
hasperiodically repeated
lraxima, due to theperiodicity
of themobility. Integrating equaiion (20)
gives a first orderequation
for the energy E~Î
~~
~~~~
~' ~~~~There are now two free
parameters,
theslope potential
at theturn1~lg point
m* and the energyV(-m*)
> E >V(m*). However,
there are two conditions on the solution: the averageslope
oî the surface is fixedby
theboundary conditions,
and theperiod
of theoscillatory
motionmust be chosen to
correspond
with thewavelength
L of thestationary
surfaceprofile.
o.oi
o.5
% ~
£ Î
> Ç
~2 -1 0 2 ~'~~~2 -1 0 2
m m
a) b)
Fig. 3. Potentials
V(m)
for trie classical mechanicsanalogues
to the n= 1 ~ud n
= 2 SOS models.
For the
n = 1
model,
thepotential
shown inFigure
3a is calculated usmg theanalytic
upper bounda+(m),
equation(12),
with K= 2.0. The value m*
= 0.89 was chosen to
satisfy
the condition(24), although
thegeneral shape
of thepotential
is the same for all m* > 0. For the n= 2
model, Figure 3b,
thepotential
is calculated from a numerical evaluation of a+ For K=
I.o,
m*= I.à in order to
satisfy (26).
Considering
the case of the n= model
first,
we see that oscillations ofarbitrarily long period develop
when the energy Eapproaches
the maximum value of thepotential V(-m*)
The
partide spends
most of the time in thevicinity
of thispoint,
andonly
a finite time at theopposite
extreme of the motion. Thus trie surfacedevelops
facets ofslope u(-m*), separated
by
narrowregions
ofpositive slope.
As trielength
of trie facetsincreases,
so trie maximumpositive slope
that the surface attains must also increase to preserve the averageslope
of the surface. Thus in thislimit,
as thenegative
extreme of thepartiale
motionapproaches -m*,
thepositive
extremeapproaches
the surfacecoupling K,
so that the surfaceslope diverges
within the
macrosteps (compare
to(9)).
Thus theasymptotic
value of m* is determinedby
the condition
V(-m*)
=
V(K),
and the facet orientation can be calculated from aknowledge
of the orientational
dependence
of themobility
j~ (i a(m*) la(m))
dm= o.
j24)
Note that there is no
dependence
of the facet orientation on themagnitude
of thedriving
forceF,
as confirmedby
the pindependence
of the simulations. We can estimate the facetslope,
using the
analytic
upper bound(12)
for themobility
a+ We find for K= 2.0 that m* m
0.89, corresponding
to a facet orientation of -23° which is rather smaller than the value of about -30° seen in the simulations. The reason for suchdîsagreement
is that the difference between themobility
and the upper bound islargest
for surfaces at smallslopes,
for which themobility
is smallest and which
give
the mostimportant
contribution to theintegral
in(24).
The manner in which the extremes of the
partide
motionapproach
theasymptotic
values-m* and
K, depends
on the average surfaceslope
ù,although
suchoscillatory
solutions canalways
be foundprovided
ù >~t(-m*).
For~t(-m*)
< ù < 0 these solutions coexist with thelinearly
stable uniform solution h= fix
(see
Sec.3.1).
We have no means ofdeciding
which ofthe two solutions will be
truly
stable in the presence offluctuations;
the simulations describedin Section 2.1 indicate that
faceting
occursthroughout
theregion
ofbistability.
Analysis
of theequations
of motion show that for solutionsm(x)
with finiteperiod L,
thereare small finite size corrections to both the facet
slope
and the value of m* of orderL~~,
which are related to the behavior of the stiffness
1
=
dm/d~t
forlarge slopes.
There isalso, by equation (21),
an associated finite size correction to the surface currentJ*,
which is mostconveniently expressed
in terms of theheight
A of themacrosteps
J*lCC) J*l~)
"
~) )~)~~~~~
+°(~~~). 125)
This correction will be very
important
when we discuss the rate ofgrowth
of the facets after their initialdevelopment
in Section 4.For the n
= 2 model the
potential V(m)
isperiodic
as shown inFigure 3b,
with aperiod equal
to twice the SOScoupling K,
and with maxima at m = 12K m* where1 is anyinteger.
The faceted surfaces seen in the simulation
correspond
topartide
motion betweenadjacent maxima,
and the facetslopes
that appear are then 1+~t(-m*).
Thus amajor
difference from the n= model is that a surface of any
slope,
other than thatcorresponding
to one of the facetorientations,
is unstable towardsfaceting.
The value of m* con be calculated from thecondition that the maxima of
V(m)
areequal,
so that/~~ 1 ~~j~
dm = 0.(26)
For low values of the surface
coupling, 1faim)
has asmall,
almost cosinusoidalvariation,
andin this limit m*
=
3K/2. Using
thisapproximation,
we find for K= 2.0 facet
slopes
in thevicinity
of zero tilt ofapproximately
-42° and6°,
ingood correspondence
with the valueswe can estimate from
Figure
2b(-41°
and5°).
Finiteperiod
corrections to theslopes
andcurrents can also be
calculated, and,
in contrast to the n= 1
model, they
areexponentially
small as theperiod
L increases.4. Trie
Coarsening
ProcessDuring
the course of asimulation,
the average facetsize1increases steadily
both for the n= 1
and n
= 2
models,
as can be seenby inspection
ofFigures
2a and 2b- This process can bequantitatively
monitoredby following
the behaviour of the real spaceheight-height
correlation function- In one dimension this oscillates as a function ofdistance,
and a convenient measure of the facet size is theposition
of the first zero(.
InFigures
4a and 4b we show how thislength
scale increases as a function of simulation time t. In each case we also show another measure of the
faceting
process. For the n= 1 model we
plot
inFigure
4a the averageseparation
of the macrostepsd,
which isequal
to the average facetlength
after an initialperiod
in which the macrosteps areforming.
For the n= 2 model we show in
Figure
4b the surface width w, which is alsoproportional
to the average facetlength.
For both the n= 1 and n
= 2
models,
it appears that the facets growaccording
to a power lawm~ t~ with the same exponent s m 0.25.
In this section we
investigate
this behaviour. For the case of the n= 1 model we find that the finite size corrections to the surface current mentioned in the
previous
section areresponsible
for thecoarsening,
so that the evolution of thesystem
after the initial appearance of the facetsis
largely
deterministic. For the n= 2
model,
we find that random fluctuations of the surface current areresponsible
for thecoarsening.
It iscoincidental,
that these different mechanismsgive
the same coarseningexponent.
100 d _
io
le+03 le+04 le+05 le+06 le+07 le+03 le+04 le+05 le+06 le+07
t t
a) b)
Fig.
4.Coarsening
of the one-dimensional SOS models. In each case the data is the average of 4 realisations of a system of size 2048. For the n= 1 model
la)
the simulation wasperformed
withù = o, p = o.7 and K
= 2.o. We show bath the first zero of the
height-height
correlation function(
and the average separation of macrostep
d,
which showsan initial decrease while the macrosteps in the system form before
increasing
as average facetlength
increases. For the n= 2 model 16) we teck ù = -o.25, p = 0.8 and 11'
= 1.o. Data for
(
and the surface width w are shown. In bathfigures
thedashed/dotted
lines haveslope 1/4.
In
studying coarsening
behaviour there are avariety
ofapproaches
that may be taken.A
systematic approach
[23] is to consider astationary, yet
unstable situation with a finite characteristiclength
L and consider thegrowth
rate of a smallperturbation.
Thegrowth
lawcan then be ascertained from the manner in which this rate scales with
increasing
L. Weapply
this method to current induced
faceting, by considering
a smallperturbation ni(x,t)
to thestationary slope potential m(~)
of theprevious section,
andexpanding equation (14)
togive
ou =
1)) lai ja(m)ôl
+Fa'(m) Ii) (27)
Although
thisequation
islinear,
formalanalysis
would be veryheavy. However,
thegrowth
rate for ni can be estimated. All of the terms on the
right
hand side areconstant, except
within theregion
of a macrostep in the n = 1 model or at theboundary
betweenadjacent
facets in the n= 2 model. We discuss each case in turn.
For the n
= 1
mortel,
the factorsa(m)
anda'(m)
remain finite for all ~ as the coarsening progresses, and theonly
term on theright
hand side ofequation (27)
whichchanges appreciably
is the surface stiffness
dm/d~t.
This has a broad minimum in thevicinity
of amacrostep,
whichcon be shown to decrease with
step height
A asA~~. During
facetgrowth
theslope
andslope potential change only
in the immediatevicinity
of themacrosteps,
so that the linear mode nithrough
which coarseningproceeds
is localised in theregions
wheredm/d~t
is small. Itsgrowth
rate is therefore of the order of the value ofdm/d~t
at themacrostep, proportional
toA~~
Since theheight
ofmacrosteps
and the facetlength
grow inproportion
to oneanother,
we deduce that the average facet size
1should
increase with time taccording
to im~t~R
This result can also be
reproduced by
anothersimple argument
which relies on the finite size corrections to the surface current mentioned in the previous section. Consider twoadja-
cent macrosteps of different
height Ai
>A2,
in asystem
withperiodic boundary
conditions.Because the current over the taller
macrostep
isslightly greater
than that over the shorter onet=ixio(
__
00
ÎÎÎÎÎ17
~Î.~~ ..______~~~~~-
"""...,
~~~"jf
0 ....,-i oo
~~""'
"....,,
o 100 200 300
Fig.
5. In a system of two macrosteps, theIarger
macrostep grows at the expense of the smaller.The simulation shown
began
with a macrostep ofheight
100 at site1= 100 and
one of
height
120 at= 200, and we teck p =
0.7,
K= 2.0.
the facet 'downwind' of it will rise and the facet
'upwind'
willfait,
so that the diiference inheight
of the twosteps
becomesexaggerated,
untileventually
the smaller macrostep vanishescompletely.
InFigure
5 we show the result ofsimulating
such asystem
which confirms that thisis indeed what
happens.
For a constant ratioAi /A2
and a fixed averageslope,
it follows from(25)
that the diiference in current between the twomacrosteps
is of the order ofA~~
r-
h~
The
disappearance
of the smallerstep
requires thetransport
of a volume of orderAi
r-
F
Thus the associated time scale isf~,
whichreproduces
the coarsening lawfr- t~R
For the n
= 2
mortel, analysis
of the noiselessequation (14)
shows that thegrowth
rate ofperturbations
vanishesexponentially
as the average facetlength increases,
andanalysis
of the facet currents showsexponentially
small finite size corrections.Thus,
thecoarsening predicted
is not a power
law,
but rather alogarithmic growth f
r- In t[23].
This does not agree with thesimulations,
and we are forced to consider the eifect that statistical fluctuations in the surfacecurrent may have on the
coarsening
behaviour. To do this we add a random termJR(x, t)
tothe deterministic current
J(x,t)
inequation (6).
Such a term ansemainly
from 'shot noise' in the surface current. Since fluctuations atseparate positions
and times areuncorrelated,
wetake
iJRix,t)i
= °iJ~ix, t)J~i~', t'ii
=
J*oit t'iôix x'i.
1281where J* is the uniform surface current of the
steady
state. Theequation
ofcontinuity
then becomes aLangevin equation
for the surface.Although
ananalysis
of thisalong
traditional hnes[24]
would bepossible,
the consequences for thecoarsening
behaviour can be understoodin
purely simple
terms[25].
Consider the eifect of the current fluctuations on a
single
faceted section of the surface.During
a time t there will have been a random net influx of adatoms into trie facet with zeromean but with an rms value
proportional
tot~/~. Thus,
the surface of the facet isdisplaced randomly
up or clown as inFigure
6 and the 'mass' fi4 under the facetrepresented by
the shaded area in thefigure)
will besimilarly
reduced or increased. If such a fluctuation reducesthe mass to zero, then the facet vanishes. In this way, the mass under each facet of the surface