HAL Id: jpa-00246957
https://hal.archives-ouvertes.fr/jpa-00246957
Submitted on 1 Jan 1994
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.
Growth roughness and instabilities due to the Schwoebel effect : a one-dimensional model
Ilya Elkinani, Jacques Villain
To cite this version:
Ilya Elkinani, Jacques Villain. Growth roughness and instabilities due to the Schwoebel ef- fect : a one-dimensional model. Journal de Physique I, EDP Sciences, 1994, 4 (6), pp.949-973.
�10.1051/jp1:1994238�. �jpa-00246957�
Classification Phy.n<..i Ahsn.a<.li
68.95 05.20D
Growth roughness and instabilities due to the Schwoebel
effect
: aone.dimensional model
Ilya
Elkinani andJacques
VillainCEA, Département de Recherche Fondamentale sur la Matière Condensée, SPSMS. MDN, CENG, 17 rue des Martyrs, F-38054 Grenoble Cedex 9, France
(Re<.eii'e(13 Jamiaiy 1994. a<.<.epled /4 Fehriiari 1994)
Abstract, A very simple, one-dimensional model (« Zeno model ») of
crystal
growth bymolecular beam epitaxy is studied numerically. The essentiel feature of the Zeno model is that ii takes into account the asymetry of the sticking coefficient of adatoms to steps (Schwoebel effect),
i-e-, the
diffusing
atoms stick preferably to the Upper ledge. In contrast with other, moremicroscopic
descriptions, the Zeno model takes the diffusion of adatoms into account through adeterminiUic diffusion
equation,
so that the computing lime isgreatly
reduced and asystematic
investigation of the effect of the different parameters ispossible. Deep
cracks form even for a weak Schwoebel effect, but they form after a time which is very long if the Schwoebel effect is very weak. In certain cases, the roughness increases proportionally to time, in agreement with expenments on silicon and with other calculations. In the absence of Schwoebel effect, surface defects are healed dunnggrowth.
1. The Schwoebel effect and its consequences.
In the
growth
of acrystal by
Molecular BeamEpitaxy (MBE),
thefreshly deposited
atoms(adatom.ç) which diffuse on the surface (as shown
by Fig.
l in the case of ahigh
symmetryFig.
l. Schematic representation of the MBE growth of ahigh
symmetry surface.Deposited
adatoms diffuse to a step or nucleate a new tenace.surface)
do net stick with the sameprobabihty
to theedge
of Upper and lower terraces. The importance of thispeculiarity
was firstemphasized by
Schwoebel[1]
who showed that this mechanism(hereafter
called Schw,oeveleffect)
generates aninstability
in theevaporation
of a vicinal surface. Morerecently,
several authors[2-10] pointed
ont otherimportant
conse-quences of the Schwoebel effect. The present work is much related to that of
Eaglesham
and Gilmer[8]
and ofZhang
et ai.[9].
Before
going further,
it is of interest to recall the mechanism of the Schwoebel effect.Generally,
adatoms stickpreferentially
to upper terraces, because there is apotential
barrer which makes it difficult for them to go down to a lower terrace(Fig. 2a).
The reason isbasically
that, in order to go downstairs, an adatom must eitherpush
another one(Fig. 2b)
orgo
through
an uncomfortableposition (Fig. 2c)
where it has too fewneighbours
and toohigh
anenergy. The
exchange
mechanism offigure
2b(analysed by
Hansen et ai.Il1]
in the case ofordinary
diffusion on ahigh
symmetrysurface)
is often believed to be more favourable intypical
menais than that offigure
2c,investigated by
Bourdin et ai.il 2]. Experimentally,
the Schwoebel barrier has been evaluated to be about 2 000 K forW(110) Il 3].
a
b
c
~> D
d
Fig. 2. The Schwoebel effect. a) Potential seen by an adatom, b) The
exchange
mechanism suggestedby
Hansen et ai. in the case of adatoms diffusion without steps. cl Mechanism treated by Bourdin et ai.d) Probability of
jump
per unit time.The
subject
to be addressed in the present work is thegrowth
of a surface ofhigh
symmetry below ilsroughening
transition. It is clear[4]
that, for such a surface, a strong Schwoebel effectproduces
aninstability.
Indeed, on any growing terrace, adatoms land from the beam and, sincethey
cannot go downstairs,they
areobliged
to nucleate a new terrace on whichanother terrace will be nucleated, etc., so that the surface
is covered
by
towers rather than flan.Even a weak Schwoebel effect may be
expected
toproduce
aninstabihty.
Indeed, itimplies
that the current of
diffusmg
adatoms has the same direction as theheight gradient
j
=
K Vz
(1.1)
with a positive coefficient K. Under certain
fairly
realistic circumstances(implying
inparticular
the absence ofevaporation),
the localheight z(r)
satisfies a conservationequation,
à=/ôt
=
div
j,
which has to be modified because of the effect of the beam. Thus thecorresponding equation
is~~
=
Fa div
j (1.2)
where F is the beam
intensity. Comparing (1.1)
and(1.2),
one obtainsÉ=Fa-KV~z.
(1.3)
A Fourier component =j of
z(r)
satisfies, if k # 0, the equationart
,
=
KL- ri
Since K is
positive,
there is aninstability. However,
thisargument
is notfully convincing
because a nonlinear term
proportional
to(grad z)~
should be added to(1.3). Moreover,
the above arguments saynothing
about the actualshape
of the surface.The best theoretical information about the consequences of the Schoebel effect on the
shape
of a
growing
surface has been obtainedby
Monte carlo simulations[5,
8,10].
Experimentally,
direct evidence of the Schwoebel effect has been obtainedby
field ionmicroscopy [2].
Theinstability
observedby Eaglesham
et ai.Il 4]
in thegrowth
of the(001)
surface of silicon has been attnbuted to the Schwoebel effect
[8].
2, The Zeno model,
In the present work we propose an
oversimplified
mortelwhich,
we believe, willusefully complement
the Monte Carlo work. For reasons which will become clear in thefollowing,
this model will be called Zeno model, from thephilosopher
born in the lucaniancity
of Eleas in the sth century B,C. This model wasalready
described in apreliminary
communicationIl 5].
This model is, to a great extent, deterministic. In
particular,
the fluctuations of the beam intensity F in space and time, as well as the fluctuations of the diffusion current around the average value(1.1),
will beneglected.
Theseapproximations
arejustified
if the number ofpartides
in some charactenstic time islarge enough,
moreprecisely
if thetypical
terrace sizeÎ
is muchlarger
than the atomic distance.In
comparison
with Monte Carlosimulations,
such a deterministic model has theadvantage
(in addition to itssimplicity)
to isolate the contribution of the different factors which can beresponsible
for kineticroughness.
Beam fluctuations and current fluctuations are mdeed such factors [4], and it is of interest tostudy
a modeldeprived
of such fluctuations, in order tostudy
the other causes of
roughness
moreprecisely.
One of these causes is the Schwoebel effect,
although
the term «instability
» may be moreappropriate
than «roughness
» todesignate
a deterministic process. However, mexpenmental
data
(microscope
image or diffractionspectrum)
the distinction is not easy to make, since stochasticphenomena
arealways
important.A pi-tort, there may be another source of
roughness, namely
the nucleation of new terraces ofatomic thickness which will form the successive atomic
layers.
Theplace
and time wherenucleation occurs are stochastic to a
large
extent. Canroughness
anse from thisstochasticity
?The answer can
hardly
be obtained from ananalytic
argument becauselong
time,oscillatmg
current correlations result from nucleation
(Fig. 3).
An important result of the present work will be that nucleationstochasticity
atone cannotproduce roughness,
at least in one dimension.~ /
ll~~s
it'
ô~ -
/
Fig. 3. After nucleation of an island (ai the current in the vicinity of the nucleus remains directed
toward the nucleu; dunng a time
z and then in the opposite direction 16) during a time T'. z and T' depend on the distance to the nucleus, but are both of order (though smaller than)
1/(Fa ).
Another feature of the Zeno model is that it is
« + »-dimensional, i-e- the surface is one dimensional. Of course the real world is
(2
+ 1) dimensional. A(1
+1)-dimensional
world may appear as anunacceptable ~implification.
However, it is such ahuge simplification
that it is a necessary first step,just
as mean field is a nece~sary first step in thestudy
ofphase
transitions. Indeed the
dynamics
of a two-dimensional surfacedepends
on many parameters, inparticular
the diffusion constants of atomsalong
a step. It is of interest toinvestigate
a model with few parameters.We have used two versions of the model the continiious version and the disc.rete version.
Bath versions are discrete in the direction z
perpendicular
to thehigh
symmetry surface. In the contimiou.ç version, there is no atomic distance in thehigh
symmetry direction.iparallel
to the averagesurface,
so that terraces may bemfinitely
small. Theadvantage
of this approximationis to reduce the number of parameters.
Actually,
it will be seen that the continuum modeldepends only
on asurgie
parameter, which characterizes thestrength
of the Schwoebel effect.A discrete version, where a finite interatomic distance is introduced, will also be used. It will be descnbed in section 6. In the present section the continuous Zeno model will be defined.
Any
state of the surface is charactenzedby
the positions of the steps. The absciwa of the n-thstep from the left will be called i,,. We need to
specify
twopoints:
ai thevelocity
v,, of each step ; b) the nucleation
probability
p~,, on each terrace. The former quantity caneasily
be deduced from the dassicaltheory,
it is shown inAppendix
A that the absolute value of thevelocity
r,, of a step separating an upper tenace of widthÎ,,
and a lower terrace of width Î,( isr,, =
f (f,,
+ g(fi (2. i)
where the
expressions
off
and g will begiven
below, assuming the upper terrace to be on theleft hand side of the step and the lower terrace to be on the
right
hand side, so thatv~, is positive. The opposite case will
easily
be obtainedby
symmetry.The forms of
f
and gdepend
on the type of terrace, which may be either a « top terrace » a« bottom terrace » or a « vicinal terrace »
(Fig. 4).
T
B
Fig.
4. T is a top tenace, B is a bottom terrace. The other tenaces are« vicinal
»
For a top terrace, g does not exist
(since
that terrace cannot be the lowerterrace)
andf
is given on bothsides,
if a is the atomicdistance, by f(Î)
=
FaÎ/2, independently
of theSchwoebel effect. The rote of the Schwoebel effect is
just
to increase the nucleationprobability.
For a bottom terrace,
f
does not exist(since
that terrace cannot be the topmostterrace)
andg is given on both sides
by g(Î)
=
FaÎ/2, independently
of the Schwoebel effect.For a vicinal terrace, it is shown in
Appendix
A thataF ail
j
(2.2)
f(~)"i
ii ~m ~f+fi
'
~z
and
g(Î)=aFÎ-Î _aFÎ Îj
~ i ~
~'
2~
~
Î
+Îj (2.3)
where a is the interatomic
distance,
F is the beam intensity per unit time and unitlength,
andii
is alength
which characterizes the Schwoebel effect. It tutus out to begiven by
ii
=
~~Î (2.4)
where
Dla~
is theprobability
per unit time for an adatom at abscissa x tojump
to aposition
at abscissa (x + 1) or (x1)
if thisjump
does not reqwre achange
ofheight
for the adatom.D'la~
is theprobabihty
per unit time for an adatom tojump
to aneighbouring
site when thisimplies
going to a lower terrace(Fig: 2d).
D is the diffusion constant of an adatom on theideally
flat surface.After
having
given the formulae for the stepvelocities,
we have to calculate the nucleationprobability
per unit time on a terrace. Its detailed formdepends
on the features of the nucleation process. For instance, if we assume that two adatoms which meet form a pair whichis stable and immobile, then the nucleation
probability
per unit time on a top terrace is found(App. A)
to be~~~~~~~ ÎÎÎ ~
~
~~
~~'~~
The nucleation
probability
on a bottom or vicinal terrace is given m theAppendix. They
arenot essential since most of nucleation events take
place
on top terraces.An important quantity is the
typical
size of a top terrace when nudeation takesplace.
It isargued
in theAppendix
that this sizecorresponds
to a value of the aboveprobabihty
of theJOURNAL DE PHYSIQUE T 4 N'6 )UNE 1994
~~
order of Fa
F~ Î) Îj
fi
~ ~Î~
~~~~'~~
We shall call
f~o
the value ofÎ~
m the absence of Schwoebel effect. ThusFÎ)o
m 12 Da
(2.7)
or
ij~i~
+ 6ii)
=
ij~ (2.8)
STEP MOTION IN THE CASE OF MULTIPLE STEPS. The presence of step bunches ares net
modify
the nucleationprobability
nor the characteristiclengths,
but doesmodify
the stepvelocity.
A bunch of p steps may be considered as a facet on which adatoms diffusefreely.
Thepossibibty
of nucleation on such a facet will beneglected.
An adatom which reaches a step bunch from above will be assumed toglide instantaneously
to the bottom of the bunch(Fig. 5).
This
approximation
is notsatisfactory
forlarge
bunches.---l ---
a
b
c
d
Fig.
5. Advance of a multiple step.Consider a bunch of p steps separating an upper terrace of width f and a lower terrace of
width
Î'.
Asbefore,
we callf(Î)dt
the number of atomssticking
from above in timedt and g
(Î')
dt the number of atomssticking
from below.f(Î
and g(Î')
are still givenby (2.2)
and(2.3)
for vicinal terraces, and areequal
toFaÎ/2
in other cases. The atoms commg fromabove and from below distribute themselves in different ways : an atom
coming
from above will stick above theprevious
one and an atomcoming
from the same level will stick aside theprevious one. Therefore the
f (Î )
dt atomscoming
from above willdisplace
the bunchposition by
the amountv ~
dt
=
f
~~ ~~(2.9a)
while the
g(Î')
dt atoms commg from the same level(Fig. 5b)
willdisplace
the lower step of the bunch atoneby
the amountv~~
dt =g(Î')
dt(2.9b)
Thus,
the bunch tends to separate into an isolated step at the bottom and a bunch of~/J
1)
steps. But this separation willreally
takeplace (Fig. 5c) only
if that bunch does notcatch the bottom step
again
due to thef(Î)dt
atomscoming
fromabove,
therefore ifvj ~ <
v~~
orf(Î)
< ~p
1) g(Î').
In the opposite case(Fig. 5d),
the step bunch does not dissociate and moves at thevelocity
~
f(f)
+g(f')
(2.io)
p
3. Growth without Schwoebel effect.
3.1 NO SCHWOEBEL EFFECT, SURFACE INITIALLY FLAT. We have first
performed
simu-lations without Schwoebel
effect, starting
from a flat surface. Theonly
random process was the nudeation of new terraces, with theprobability given
in theAppendix.
A
typical profile
is shown mfigure
6. The surface thickness isessentially
reduced to two or three levels(Fig. 7a),
but a fewhigher bumps (or deeper holes) occasionnally
appear. If the time unit is thelayer filhng time,
thelargest
terraces have theheight
t at mteger time t, and there are small,freshly
nucleated tenaces atheight (t
+1),
and small terraces atheight (Î 1)
which are on the verge ofdisappearing.
On the otherhand,
if the time ishalf-integer,
ail terraces have
heights (Î
+1/2)
and(Î 1/2)
and areapproximately equal.
2002
2001
2000 z
1999
199s
~~~~.0
5.0 10.0 15.0 20.0 25.0
x
Fig.
6.Typical
surfaceshape
at integer coverage in the absence of Schwoebel effect for anmitially
flat surface. The vertical direction is the height in monolayers ; the horizontal direction is the distance, using Î~~ as length unit.
1200 500
iooo
800
N(ZJ 600 N(1)
400
200 100
0 0
1997 1998 1999 2000 2001 2002 2003 Ô-Ô 0.4 0.8 12 1.6 2.0 2.4 2.8 3.2
z i
a) b)
Fig.
7. Growth of anmitially
flat surface without Schwoebel effect. a)Histogram showing
the proportion of the surface atheights
1, and + at integer time t (there are aise two tenaces at height( + 2 and one at1 2 which cannot be seen on the histogram), b) Terrace size distribution after filling 10 000 layers without Schwoebel effect.
The conclusion is that random nucleation does not lead to
roughness.
The size distribution of the terraces at an integer time is shown in
figure
7b.3.2 HEALING OF DEFECTS FOR SAWTOOTH PROFILES OF SHORT WAVELENGTH. lt is weÎÎ
known
expenmentally
that defects such asbumps
orvalleys
are healedby
MBEgrowth.
Our mortel doesreproduce
this property. Forsimplicity,
we restncted ourstudy
to sawtoothprofiles.
We have studied
numerically
thegrowth
of acrystal
limitedby
a surfaceexhibiting
aperiodic
succession ofstraight
facets(Fig. 8).
We haveimposed
a determmistic nucleation rule,namely,
a new terrace is nucleated at the centre of each top terrace when its size reaches the valueÎ~
= Î~~. Nucleation on terraces other than top terraces never occurs.Figure
8 shows1=0 toe20
t=40 -t=80
120
100
~---
---;80
Z 60 .~
,
40 .'
20
0
0.00 0.20 0.40 0.60 0.80 1.00 1.20
x
Fig.
8.Healing
of a saw-toothprofile
without Schwoebel effect. Only one penod isrepresented
at increasing limes (0, 20, 40 and 60 layers). Nucleation is determmistic.the successive states of the
profile. They correspond
to aperiod
L=
1.2 Î~~ and to an initial
height h(0)
=
100.
Eventually,
a flat surfacecorresponding
tofigure
6 is reached. In the intermediate states(Fig. 8)
the surface consists of flat terraces at its top, which are boundedby
a steep
edge
made of severalcoalescing
atomic steps andseparated by
cracks. We will now tryto understand this structure in the case when the initial terrace width is shorter than
Î~.
This is theexperimentally interesting
case, becauseÎ~
isgenerally
verylarge
in MBE.The variation of the interface thickness
(height
difference between topmost and lowestpoint)
is
displayed
for a sawtoothprofile
infigure
9a in atypical
case. The initial decreasecorresponds
to theflattening
of the top of theprofile,
when the width Î~ of the topmost terrace is smaller than Î~~. Then there is no nucleation at the top and, since bottom terraces areannihilated,
the thickness decreases. WhenÎ~
become of the order ofÎ~o,
the thickness decreases but veryslowly.
120
so ioo
80 60
Ah Ah 60
40
40
20
0 0
0.0 20.0 40.0 60.0 80.0 100.0 0.0 8000.0 16000.0 24000.0 32000.0
Time Tlme
a) b)
Fig.
9. Variation of theheight
difference (« interface thickness ») between thehighest
and the lowest point of the surface as a function of time. There is no Schwoebel effect. In case (a), theperiod
isL=1.2i~~
and the initial height difference is 100 atomic layers. In case (b), the penod is~0 i~~ and the initial height difference is 100. The time unit is (1/Fa).
The evolution shown in
figure
8 iseasily
understood if the initial tenace widthis much shorter than
i~.
Indeed(cf. Fig. loi
the topmost tenace of eachperiod readily
takes a widthi~ w1. Therefore, according
to(2.2)
and(2.3),
and becauseii
=
0 without Schwoebel
effect,
the topmost steps have avelocity
of orderFai~/2,
which is muchlanger
than thevelocity Fai12
of the other steps. Therefore, the topmost steps coalesce with the next steps and form a bunch of two steps. This bunch will thenproceed
with a reducedvelocity.
The reduction factor isapproximately
2 because the progression of the topmost steps ismainly
due to atomsfalling
from the upper tenaces, which are m identicalquantity
with or withoutbunching,
and thisquantity
must be shared between twosteps.
Thus additionalbunching
willfollow,
and abunch of 3 steps, 4 steps, and so on, will form. Thus the
shape
offigure
8 emerges.The
healmg
time of a sawtoothprofile
withperiod
2 L~ 2
i~
iseasily
calculated. Indeed there is no nucleation until the surface is healed. Therefore thehealmg
time r~~~j isjust
the timenecessary to fill each
valley.
For a sawtoothprofile,
thevalleys
have aheight
h=
Lli
andan
area Lh/2
=
L~/21,
andare filled after a time
~~~~~
2
Îa~
~~1
_ a
Fal -
_ i
i
-
j
i b
i i i i i i i i i
C
Fig.
10. Growth of the top ofthe surface. Ifthe widthi~
of the topmost tenace is muchlarger
than the width f of the lower terraces, the upper steps form bunches.We have aise studied the case L
~
i~.
It tums eut to besurprisingly compbcated.
Theprofile
shows a step bunch near the top as
before,
but the new feature is that wavesdevelop
below the bunch(Fig.
ii).
Thewavelength
increases with time, so that the number ofbumps
decrease beforeeventually vanishing.
As a consequence, the surface thickness has also anoscillating
character
(Fig. 9b).
It decreasesabruptly
when abump disappears
from theprofile.
350 600
(a) (b)
330
310 560
Z
290 540
270 520
(c) (d)
s30 ioso
sio ioôo
z
790 1040
770 1020
750
0.00 4.00 8.00 12.00 16.00 u.00 4.00 8.00 12.00 16.00 20.00
x x
Fig. Il- Profile observed at successive limes (a) t =250, b) t=500, c) t=750 and d)
t 000) when heahng a sawtooth
profile.
The period is equal to 20 f~~. There is no Schwoebel effect and the nucleation process is determmistic. The units are the same as mfigure
6.We have net been able to undestand the
origm
of thisinstability.
We firstthought
of apossible
artefact of timediscretization,
but we have checked that the time step was shortenough
to avoid any artefact.Possibly
thenonlinearity responsible
for theinstability
arises from collisions between steps, but we have not studied these collisions.Healmg
of cracks has aise beenreproduced
in the case of a weak Schwoebel effect,again
usmg a determmistic nucleation mechanism.
However,
these results are not reliable because it will be seen in the next sections that in the case of tandemnucleation,
cracks may beproduced
even m the case of a very small Schwoebel effect.
4. Evolution of a surface with Schwoebel effect.
4.1 CRACKS. In this section we report simulations with Schwoebel
effect, starting
from aninitially
flat surface. Random nucleation is assumed. Thestrength
of the Schwoebel effect is characterized, as said above,by
the ratio ~=
ijli~~,
note that it is convenient todistinguish i~,
the actual characteristiclength
of nucleation on the topmost tenaces, and i~~ givenby (2.7).
i~
andi~~
are relatedby (2.8).
With a strong Schwoebel effect
(say,
for ~ ~ l cracksdevelop
veryearly. They
do not heal within the time of oufcalculation,
so that theirdepth
isessentially equal
to Fat. This feature isa consequence of the continuum
approximation.
Thedensity
of cracks cannot belanger
thanlli~,
and seems to converge to a value of this order when ~ increases.Actually,
even with afairly
weak Schwoebeleffect,
cracks also appear veryrapidly.
Figure
12 shows atypical profile
obtained forijli~~
= 0.2. The cracks look similar to those m
a strong Schwoebel effect
they
become verydeep
while their width increases veryslowly.
However
they
appearprogressively.
For instance, if ~=
0.20,
the first crackstypically
appearafter
filling
20layers (in
asample
of size 100i~~).
Then the cracks'depth
growsnearly proportional
to time(the
coefficient ofproportionality
is atomic size ofdepth
petdeposited layer).
The
deep
cracks observed infigure
12a areunphysical
because their width is smaller thanthe interatomic distance.
However,
the other features(density
of cracks, crackshape,
roughness)
are the same as for aphysical
model with a finite atomic distance(Fig. 12b).
We shall now try to understand the evolution of cracks. if the crack
profile
may beapproximated by
ananalytic
function z(x)
of the « horizontal » coordinate(parallel
to thehigh
symmetryorientation),
the evolution can be treatedby
thetheory
of Franck[16].
The bottom and the top of the crack requirespecial
attention because, as will beargued,
theprofile
there is700 700
600
soo soo
400 400
Z Z
300 300
200
ioo ioo
o 0
0.00 20.00 40.00 60.00 80.00 100.00 0.00 20.00 40.00 60.00 80.00 100.00
X X
a) b)
Fig.
12.Typical profile
at successive times forijli~~
=
0.20 in the Zeno model with random nucleation. a) Continuum limit. b) Fimte atomic distance, a
=
0.001f~~.
net
analytic.
In the next subsection we consider adeep
crack, without caring about how it formed.4.2 EvoLuTioN OF THE BOTTOM OF A CRACK. We consider a crack with a very
sharp
bottom. It is clear that the
profile
is notanalytic
at the bottom. Asharp
bottomimplies
that thetenace widths are small,
i~
«ii.
Thisimplies
that the Schwoebel effect for these tenaces canbe considered as very strong. This argument is
only
valid in the continuumapproximation,
and has notnecessarily
a counterpart in a reabsticproblem.
Assummg
the Schwoebel effect to beinfinite,
theequations
of evolution for a crackhaving
a symmetry axis at ~r =0 are
(Fig.
13a)where
x,,(t)
denotes theposition
of the n th step at time t.à 'Z
x ~
~
~
2
i~
~---~----/---_
à
x~ z
x
b
1---$
Fig. 13. a) Labelhng the steps in the bottom of a deep crack (a) and at the top of the profile (b).
4.3 THE TRACK OF ZENO OF ELEAS.
Equation (4, la)
isindependent
of the other ones and its solution isxi
(t
= A"1(0 exp(-
Fat(4.2)
This
apparently
harmless resultimplies
that xi(t)
never vanishes.During
the nominallayer filling
time, which is il(Fa
),only
a fraction(1
île)
of the lowestlayer
is filled. After a timen/(Fa),
which isnominally
the time to fill nlayers,
a fractionxi
~
=xj(0)e~" (4.3)
Fa
is still unfilled. This is reminiscent of Zeno's
paradox according
to which Achilles will never catch the turtle if the latter has some initialadvantage,
say km. Indeed when Achilles will have run that kilometer, the turtle will advance, say one meter. Then, m the time necessary for Achilles to mn that meter, the turtle will mn one millimeter. At the n'th step of the argument, the turtle will still have anadvantage equal
tud~ = 10~ ~~ kilometers
(4.4)
The
similarity
between(4.4)
and(4.3) justifies
theterminology
« Zenoinstability
» for the effect described above. There are of course differences between Zeno'soriginal paradox
and its modem versionapplicable
to MBE. First ofail,
ouf derivation and ouf result arestrictly
correct within the continuum model. Zeno's argument was
obviously
wrong. As fat as weunderstand, it
just
reflects theignorance
ofAlgebra
which affected the ancient Greeks. It should be remembered that Zeno's time was verybright
for literature, but the great time of Greek Science was two centuries later On the otherhand,
Zeno'stheory
was formulated in adispute against
thedisciples
ofPythagoras. Presumably [17, 18],
the latter may have known the sum of thegeometrical
series withgeneral
term(4.4) although they
were not able toformulate the calculation m a
convincmg language.
Thesimple algebraic language
which isnowadays taught
in ouf schools was not invented before the end of the sixteenth century[19].
The
general
solution of(4,1)
iseasily
seen to bex,,
(t )
=
~jj' ~~ [~ Y
x,,~
(0
exp Fat(4.
5Thus,
if xi, x~,...,
x~
are known at time0, they
are also known at a later time t.Dur numerical simulations on the model described in section 2 confirm this result : if the Schwoebel effect is strong
enough,
cracks form theirdepth
increasesproportionally
to lime, andthey
do not healduring
the simulation.Then,
how can one understand that, for appropnatevalues of the Schwoebel effect, the crack
ultimately
heals after an initial increase ? To answer,one should worry about the
edges
of the crack.4.4 EVOLUTION OF THE CRACK EDGES. While it is easy, as seen m the previous section, to
study
the evolution of the bottom of acrack,
it is more difficult tofigure
ont the evolution of a crackedge.
Apriori,
twosimple
schemes may beexpected.
The twopossibilities
can be bestdescribed in the case of a
periodic profile
with deterministic nucleation. In contrast with the previoussection,
it is convenient tu label the steps from the top of theprofile (Fig. 13b).
A first
possibility
isthat,
near itsedge,
theprofile
goes to a well-defined hmit. In otherwords,
if one considers for instanceonly
the times t where a new tenace isnucleated,
one has for any given nLifrl
1,,(t
= 1~°(4.6)
Co
Of course, in the continuum model addressed
here,
the crackdepth,
and therefore themaximum value of n goes to
infinity
with time. Therefore the bottom of a crack cannot go to alimit.
The other
possibility
is that thefreshly
nucleated steps merge into a step bunch, as occurred without Schwoebel effect whenhealing
a sawtoothprofile. Then,
theposition
u(t)
of the step brunch on the left of a maximum is anincreasing
function oftime,
except if a bunch appears above a bunch an event which has never been observed. Let h(t
be theheight
of thebunch, H(t)
the crackdepth
and L theperiod. H(t)
is counted from the top of thebunches,
so thath(t) ~H(t).
The numerical calculation suggests that the nucleation rate, as well as the increase of H, is
approximately uniform,
so thatH(t)~F't
where F' may be different from F. The
quantity
of matter which faits mto the crack pet unit time is alsoapproximately uniform,
and therefore it isbigger
thanF"i~,
where F" is anumber. This matter is to be distributed among the h steps of the bunch, so that the width
(L/2 u)
of the crack decreasesby
the amount F"/h in a time unit. Thereforedu
F"i~ F"i~ F"i~
à
h ~ H~F't
Integration yields
F "
i~
"~~ ~
F'
~~
~~'~~
This expression goes to
infinity
with time t. Thisimplies
that the crack widthL/2
u(t)
vanishes after some time. Thus the cracks should heal.The two scenarios described above
(crack reaching
astationary shape
and crackheabng)
arethe
simplest possible
ones. In the absence of Schwoebeleffect,
and also m weak Schwoebeleffect, preexisting
cracks seem to healaccording
to the second scenario.Figure
14gives
anidea of the evolution of a crack for a
fairly
strong Schwoebeleffect, iili~
= 0.5. As shown m
figure
lsa, theslope
at the crackedge regularly
increases as,à.
This isconsistent with the second of the scenarios described above. The crack will
presumably
described above. The crack willpresumably
heal after a verylong
time,although
this has not been observed in the present calculation.5. Weak Schwoebel effect.
5.1 EVOLUTION OF THE PROFILE wiTH TIME.
-Figure16
shows theprofile
obtained forijli~~
= 0.05 at successive times from an
mitially
flatprofile, using
the velocities definedby
(~.l-3),
and the nucleationprobability (2.5).
As in the case of a strong Schwoebel effect studied in the previous section, there aredeep cracks,
visible mfigures
16c to e.However,
there aresignificant differences,
mparticular
:i)
For coverages lower than 1000layers,
theprofile
isjust
a littlerougher
than without Schwoebel effect. There is nostriking
difference between theprofile shapes
observed afterdeposit
of 500 and 1000layers.
ii) Healmg
of cracks isoccasionnally
observed. For instance, thedeep
crack observed near1= 80 in
figure
16c is nolonger
observed as adeep
crack infigure
16d.However,
a small2050
(a) (b)
iooo
9so 19so
z 900 1900
sso isso
soo
(c) (d)
3000
2950 7950
Z 2900
2850
2800 7800
2750 7750
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00 1.20
x x
Fig.
14. Profile of a crack at coverages 1000 (a), 2000 (b), 3000 (c) and 8000(d) forf~/f~~ = 0.5 in the continuous model. Nucleation is deterrninistic.
iooo.o
100.0
-,-__--
' '
-~
-,-
~
i
~
- .
iooo.o
Time Time
a) b)
Fig. 15.
increase as functions of t'~
depression
is observed at the sameplace,
and ispresumably responsible
for the crack which reappears later at the sameplace.
iii)
For coverages lower than 2000layers,
the bottom of ail cracks movesupward,
mcontradiction with the argument à la Zeno given m section 4.2. As seen m section
4.4,
this canbe understood as an effect of the crack
edges. Figure
17 shows the interface width for asample
99s
~ 990
485 ?Si
48Ù
~~j ~~j 98Ù
iioo
z
(c) (ii
iioo
145Ù
14ÙÙ Z
13so
1300
(ti (o
Ù.ÙÙ lÙ,ÙÙ 4Ù,ÙÙ lÙ,ÙÙ 8Ù.ÙÙ Ù.ÙÙ lÙ.ÙÙ 4Ù,ÙÙ lÙ,ÙÙ 8Ù.lÙ IÙÙ.ÙÙ
x x
Fig.
16. Profile at coverages 500 (a) 1000 (b) 15 000 (c) 2 000 (d) 2 500 (e) and 5 000 (fl monolayers forfj/f~~
= 0.05. Note that the ordinate~ have different scaies in the different pictures. The
units are the same as m
figure
6.of size 100 i~~ for a
particular
run. It showsstraight
pieces ofslope
terminatedby
anabrupt
decrease. This decrease
corresponds
to thehealmg
of thedeepest
crack.At
longer
times, the lifetime of the cracks becomes muchlonger.
For instance, at coverages between 3 300 and 5 000, the interface width is a linear function of time withslope (Fig.
17).This means that the
deepest
crack dues not healduring
this time.5.2 ROUGHNESS. The
roughness
at time t is defined asIL
L 1/2ôh(t)=-
ΰ d£rdV[h(V,t)-h(,1,t)l~) (5.1)
o o
where the
angular
brackets denote an average on many runs. Inpractice,
we have madeonly
one run. The results are shown m
figure
18. It is of interest to compare thisroughness
with theexperimental
observation ofEaglesham
et ai.[14, 8]
that theroughness
mcreaseslinearly
withtime until a critical coverage of a few hundreds of
monolayers,
when the material becomes3000 200
2500
150 2000
" isoo " ioo
iooo
so soo
0
~~~~~ ~ 20000.0 3°°~°'°
0
~ 2000,0 3000.0 4000.0 5°°°'° ~'~
Tlme
0.0 1000.
~~
a) b)
Fig. 17. Interface thickness as a function of time : a) contmuum model with
ij/f~~
= 0.05 ; b) discrete model with f~/f~~ 0.03, and
a/f~~
O.COI. The units are the same as infigure
9.30.0 10.0
25.0
8.0
)
20.0(
6.0i~ 15.0
~
Î )
4.05.0 2.0
0.0 0.0
0.0 10000.0 20000.0 30000.0 0.0 1000.0 2000.0 3000.0 4000.0 5000.0
Time Time
a) b)
25.0
20.0
(
6.0(
lS.0Z z
Îi Î~
4.0 1 10.0
2.0 5.0
0.0
0.0 400.0 800.0 1200.0 1600.o 0.o 1000.0 2000.0 3000.0 4000.0 5000.0
Time Time
c) d)
Fig.
18.Roughness
vs. time forfj/f~~
0.03 (a), 0.05 (b), O.l (c) and 0.2 (b). Curves (b) and (c) correspond to the continuum model, and curves (a) and (d) correspond to an atomic distance a, with a/f~~ 0.005. The coordinates are thesame as m figure 9.