Citation for this paper:
Gough, T. & Illner, R. (1999). Modeling Crystallization Dynamics when the Avrami Model Fails. VLSI Design, (9)4, 377-383. http://dx.doi.org/10.1155/1999/38517
UVicSPACE: Research & Learning Repository
_____________________________________________________________
Faculty of Science
Faculty Publications
_____________________________________________________________
Modeling Crystallization Dynamics when the Avrami Model Fails Terry Gough & Reinhard Illner
1999
Copyright © 1999 Hindawi Publishing Corporation. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This article was originally published at: http://dx.doi.org/10.1155/1999/38517
Reprints available directly from the publisher Photocopying permitted by license only
the Gordon and BreachScience
Publishers imprint. Printed in Malaysia.
Modeling Crystallization Dynamics
when the Avrami Model Fails
TERRY GOUGH andREINHARD ILLNERb’*
DepartmentofChemistry,
Univbersity
ofVictoria,P.O. Box3065, Victoria,B.C. V8W3V6, Canada,e-mail:gought@uvvm.uvic.ca; DepartmentofMathematicsand Statistics, UniversityofVictoria, P.O.Box3045, Victoria,B.C. V8W3P4,Canada
(Received13August1997,"Infinalform1 December1998)
Recentexperimentsonthe formation of crystallineCO2fromanewlydiscovered binary
phase consisting ofCO2andC2H2at90 Kfail to beadequatelysimulatedbyAvrami
equations. Thepurpose ofthis noteis to develop analternative to the Avramimodel whichcanmake accuratepredictionsfor these experiments. Thenewmodeluses empiri-cal approximations to the distribution densities of the volumes of three-dimensional
VoronoicellsdefinedbyPoisson-generated crystallization kernels(nuclei). Insideeach
Voronoicell,thegrowthofthecrystalisassumedtobelinear in diameter(i.e.,cubic in
volume)untilthe cellisfilledbythe CO2crystalsand the C2H2(thoughtofas awaste product).The cumulativegrowthcurve iscomputed by averagingthese individualgrowth
curves withrespect to thedistributiondensityofthevolumes of the Voronoi cells.
Agree-mentwiththe experimentsisexcellent.
Keywords: Crystallization,Avramimodel, binary phase
1. INTRODUCTION
Crystal growth is of obvious significance in semiconductor design. The classical approach to
model the formation of crystals is based on a
probabilistic argument and known as the Avrami
model [4]. We review the Avrami model in Sec-tion 2 but state here that the Avrami equations predicting conversion to crystal as a fraction of
total availablevolume are
( t)
e-gt(1,1)
whereKandn are parameters.
The purpose ofthis note is to offer an
alterna-tive totheAvramimodelbasedonstatistical
prop-ertiesofVoronoidiagrams.Itisstraightforwardto see
(and well-known)
that the Voronoi diagrams*Correspondingauthor,e-mail:rillner@math.uvic.caz
378 T. GOUGHANDR.ILLNER
associated with the crystallization kernels
(impu-rities referred to as nuclei) are important for the crystallization process: Crystals will form as
spherical globules and grow linearly with time in
diameter, hence cubic in volume, around the impurities, until these globules impinge into each
other; the latter occurs exactly at the interfaces
defining the Voronoi diagram, because these
interfaces are defined as being equidistant to the
two closest nuclei. See Figure 1.
One characteristic of Avrami equations is that they predict that regardlessofwhatnis,thelength of time required to convert the first half of the
reactant to product is shorter than the time
required for the second half of the conversion. Onecanargue that the slowing down occurswhen
expanding reaction fronts collide atthe boundary
separating adjacentVoronoicells. After such
colli-sions, the surface at which conversion occurs no
longer growsasthecube of time.
The experimentwhichinspiredusto look foran alternative totheAvramimodelwastheformation
and disintegrationof a metastablecrystalline
bina-ryphase
CO2.C2H2
undersuitable conditions: Mix-tures of these two gases are expanded through anozzle onto a zinc selenide window which is
mounted in the beam of an FTIR spectrometer andmaintained at 90K. The pulsedeposits afilm
approximately 200 molecules thick. The obtained
spectra depend markedly onthe conditions ofthe expansion.Ingeneral, absorptionscharacteristicof crystalline
CO2
and crystallineC2H2
are observed together with new features assigned to a mixedphase. For 1:1 mole fractions the absorptions assigned to the purephases are absent, and it was
found that the intensityratioof thenewfeaturesis
independent of the mole fraction of
CO2.
Theseobservations establishthe stoichiometry of thenew
phaseas
CO2.C2H
2(for
moredetails,see[3]).
Itwasthen observed from spectrarecorded over aperiod of5 hours thatthe
CO2.C2H2
decomposesinto
CO2
andC2H2.
Experimental data forfrac-tional conversion versus time, based on spectral intensities,aregiveninFigure2.
The
CO2
formed hasasharpspectrumindicating crystallinity, whereas theC2H2
has broad absorp-tions indicating a more amorphous state. It wasestablished [3] that the reaction is a solid state transformation. The plotin Figure 2 is sigmoidal, suggesting that an Avrami analysis might be ap-propriate.
However,
the decomposition does notshow the characteristic slowing down discussed earlier:Rewriting
(1.1)
as(t)
e-:t"taking logarithms
ln(1
99(t))
Kt and taking logarithms again yieldsIn(-
ln(1
(t)))
In
K+
nIn
t.(1.2)
So the left-handsideof
(1.2)
is a linear function inInt. However,
plottingln(-ln(1-converted
frac-tion)) versusIn yields the curvegiveninFigure3,which is obviouslynot linear.
The curve in Figure 3 indicates that nincreases
smoothly from at the beginning to 4 at its con-clusion.
A
least squaresfit to the overall datapro-vided avalue ofn 3.3.
The basic idea put forward in this paper is to
produce approximations to the crystal growth
curves
(as
afractionof totalvolume)
by averagingthe growth curves associated with the individual Voronoicells withrespectto thestatistical
proper-ties of a Voronoidiagram forwhichthe impurities
aregenerated byaPoissonprocess. Inthe absence
0.8
=
0.6 0 0.4 O o0.2 Decomposition of Carbon dioxide-Acetylene. 0 2 3 Time inHours. Avrami Plot -1.5 -1 -0.5 0 0.5 1.5 In(-In(1-conversion))380 T. GOUGHAND R. ILLNER
ofother information, assuming that the impurities
are Poisson-distributed is natural.
Ourapproachcontains two inherent difficulties:
1. Thegrowthcurveofthepieceofcrystalinside a Voronoi cell depends on the geometry of that cell. The volumegrowthisassumed to becubic with time until aboundaryisreached,thenslow
down because there cannot be any growth
beyond that boundary, and stop completely once all the corners of the cell are filled. The
endphase of the growthwillthereforevaryfrom celltocell.
2. While the statisticalproperties ofVoronoi cells
are awidely investigated topic 1],theexact prob-ability distribution ofthe volumes ofVoronoi
cells associated with Poisson-distributedpoints
isnotknown.Intwo dimensions,thereare alot
of empiricalstudiesandmatches tothesestudies with variousad hoc approximations, like Gam-madistributions, Maxwell distributions or
log-normal distributions. Werefer to[1] for details.
An
empirical numerical study on the volumedistribution of three-dimensional Voronoi cells generated byaPoisson process isgivenin[2]. Wecannotdealwiththesedifficultiesrigorously;
weavoid thembymaking simplifying assumptions.
Asfor 1., we simplyassume that growthproceeds
as kt3untilthe cellis full, i.e., fora cell ofvolume
a > 0the growthcurveisgivenby
g(a,t)
{
kt3’
O<_
<
(a/k)
1/3
a,
<_
Here, k denotes the (linear) growth rate of the radius oftheglobuleforminginsidethe cell.
Thisassumptionisclearly simplistic, butthe error is small for Voronoi cellswhich are regularin the
sense that they are well approximated by spheres
(we
refrain from trying to produce a rigorous approximation criterion).A
mitigating factor in this context is the observation that thecrystal-lizationprocessinvolvestwo products, namely the pure
CO2
crystals,whichgrowfromthe nuclei, andthemoreamorphous C2H2,which will accumulate
attheboundariesandinthecornersoftheVoronoi
cells.
Asfor2.,wetried anumberofdensity distribu-tion funcdistribu-tionsforthe volume of Poisson-generated
Voronoicells.Itwillbe shown below that thereis a choice ofa probability density which reproduces theAvramimodel.
However,
using semi-empirical approximations as suggested in[1]
and[2]
pro-duced much better agreement with ourexperi-mentaldata.
Theplan of this paper is as follows. We briefly
review the Avrami model in Section 2. InSection
3,weestablish ageneral formula forgrowthcurves
by averaging over the curves given in
(1.3)
withrespect to a. Several examples are discussed, and
one of them reproduces the Avrami model. For
other semi-empirical approximations to the Vor-onoi cell volume distribution, the general formula yields integrals which we evaluated by using the
MAPLE symbolic computation program.We
pre-sentthebest matchweobtained in thiswayto adata
setproducedin
[3]
in Section 4.2. REVIEW OFTHE AVRAMI MODEL
A
direct derivation of the Avrami model from probabilistic considerationsworks as follows.Assume
that N nuclei are equidistributed dis-tributed in a(macroscopic) Volume V> 0.Chooseanarbitrary butfixedreference pointPand letXbe thedistance to thenearestnucleus. Xis arandom variable, and
V
(2.1)
so thecumulative distribution function ofXis
FN(X)
P{X
x}
(1
7rx3
1
N
V
In the limit Nee, Voc such that
(N/V)
converges to
F(x)
e-
ax3.
If crystalline globules grow at speed v from each
nucleus, this calculationtranslates directlyintothe probability that crystallizationat time hasreached
the point P:
(t)- P{X < vt}-
1-e-
"3’3.
(2.2)
3. AVERAGED GROWTH CURVES
We now present an alternative approach which contains the Avrami equation
(2.2)
as a specialcase. Suppose thatf(a), a
>_
O, is the density dis-tribution function for the volume of the Voronoicells (in
3D)
associated with Poisson distributed nuclei. We average the individual growth curvesgiven by
(1.3)
with respect tof
and compute amacroscopic, observable growthcurve as
c?(t)
g(a,
t)
f
(a)
da ktL
af
(a)
da+
kt3
fkt
f
(a)
da(3.1)
The upper limit o in the integrals on the right is
usedbecausewe canassumethat the largest
theoreti-cally possible Voronoi cellis ofmacroscopic scale relative to the typical one. The two-dimensional
analogueto
(3.1)
is k(’)
f
af
(a)
daq-kl2Sk
f
(a)
da.dO
Formula
(3.1)
is offered as an alternative to theAvramiequation
(2.2).
Ofcourse,wehavetoknow whatf
is in order to produce a usable equation. Unfortunately, the truef
is not known (but anobject ofstudy, see
[1]);
wediscuss a fewexampleswith ad hocchoicesfor
f.
Thelastexampleisthen matched to experimental datainthe next section.Example 3.1 To get a feeling for the type of
curves which
(3.1)
produces, we assume thatf
isthe equidistribution on an interval [0, A], with A>O,i.e.,
f
(a)
X[0,A](a).
For this f, the integration in
(3.1)
can be done explicitly, and yieldsp
{
k237
k2 t6
for kt<
Afor kt
>
A.These curves already display the correct logistic-typegrowth,andtheyshowthe kind ofasymmetry aboutthe half-lifepoint (i.e., the point where half
ofthe substance has crystallized) which was seen
inthe experiment described in[3].
Example3.2 The equidistribution usedin
Exam-ple3.1 isclearlynot averyintelligent guess for the volumedistribution of Voronoi cells. Letusmake
a more systematic attempt, following the deriva-tion oftheAvrami model in Section 2.
Assume
thatNnuclei are Poisson distributed ina(macroscopic) Volume V> 0.Chooseone ofthese
nucleiarbitrary butfixedandletXbe thedistance to its nearestneighbor. Xisarandom variable,and
so the cumulative distribution function ofXis
FN(X)- P{X
_<
x}-
1-(1
hence the densityfu(X)
isfu(X)
F’
N(X)
4rN-V
4
X3)
N-2 In the limit N-+,
V-+ such that(N/V)-+
A
> 0 (intensity of the Poisson process),fN
382 T. GOUGHAND R.ILLNER
converges to
f
(x)
47r/kx2e
-’xx3
and
FN
toF
(x)
e-}’xx3Weusethesetocompute the densityofthe random
variableS:=volumeof thespherewith radiusX/2; this is the largest sphere which will fit into the Voronoicell centered atthe chosen nucleus.Clearly,
P{S
<_
s)-P
X<_
se
-8"xs.
The associated densityis
g(s)-
8e-.
(3.2)
We note forlater reference thatand hence the expected volume occupied by all suchspheresaroundallnuclei is
(N/SA)
(V/8).
So7/8
ofthe available volume belongs to corners ofVoronoicellsin this sense.
Ifwe use g as obtained in
(3.2)
in the formula(3.1),
we find kt(,)
8i ,0/
ae-8"Xada
-+-
8/kt3
Skl3
e-8"Xada
8A(3.3)
and modulo a normalizing factor this is the
Avramiequation
(2.2)
rediscovered.Example 3.3 The lastexample which we present
uses the density
f(x)-
ctxZe
-/x2,
where a,/3
areparameters; they are linked because
f
has to be aprobability density, so
f
dependson onlyone freeparameter.Wearrive attheformula
k
o(t)
a/
x3e-nX2dx
+
akt3
f
x2e-X2dx.
JO(3.4)
The first integral is easy to compute; the second requires theerrorfunction tobe expressedinclosed form. Symbolicevaluationof
(t)
iseasily donebyusing a symbolic computation package like, e.g., MAPLE.
Weexperimentedwithmany otherdensities.The three given above were chosen for the following
reasons:3.1 isthe simplest example;3.2reproduces the Avrami equation; and 3.2 gave us the best
agreementwithexperiments,asweshowinthefinal section.
4. COMPARISON WITHTHE EXPERIMENT
The density from Example 3.3, with
/3--
1, gave the match to the data points which we present inFigure4.
The agreement with the experimental data is
striking, suggesting that the assumptions under-lying the modeling processare realistic.
Infrared spectroscopy of the reactant and pro-duct phases shows that thereactantphase and the product carbondioxide are both highly crystalline
o:5
while the product acetylene phase is amorphous.
These observations are consistent withthe product consistingofcrystals ofcarbondioxideembeddedin aless rigid acetylenematrix. This meansthat each Voronoi cell willonly be half fullofrigidmaterial
andthis materialmay be freeto move.
Also, we point out that the density
f(x)-cx2e-x2 appearsto be agood approximation for the densitydistributionofthe volumes ofVoronoi
cellsassociated withPoisson-distributednuclei
(see
[2] for anempiricalstudy).Wunderlich
[4]
hasreported experimental calori-metricdata for the crystallization ofacopolymer of ethylene terephthalateand ethylenesebacate. Thissystem resembles the present one in that
crystal-lizationdoesnotproceedto completion andsothe final product consists of crystalline regions
im-bedded inalessrigidamorphousmatrix. Thedata
were subjected to an Avrami analysis and n was
found to be 3.2. Furthermore, the data deviated
from the expectations of the model in the same sense as the present data. We conclude that the
present model would better describe the
crystal-lization of the copolymer, and presumably many
othersystems.
Acknowledgement
This research was supported by grants from the Natural Sciences and Engineering Research Coun-cil ofCanada. The authors are gratefulto Dennis Manke,who produced Figure with his
Voronoi-diagram producingsoftware.
References
[1] Okabe, A., Boots, B. and Sugihara, K. (1992). Spatial
Tesselations: ConceptsandApplications of Voronoi Dia-grams,J. Wiley.
[2] Quine,M. P.andWatson,D.F.(1984).Radial Generation of n-Dimensional Poisson Processes, J. Appl. Prob., 21,
548-557.
[3] Rowat, T. (1997). Stoichiometry and Stability of Binary Phase Crystals formed between Acetylene and Nitrous
Oxide/CarbonDioxide,Ph.D. Thesis,Universityof Victoria.
[4] Wunderlich, B. (1976). Macromolecular Physics, Vol. 2:
Crystal Nucleation, Growth, Annealing, p. 132 ft.
Aca-demicPress.
Authors’Biographies
Terry
Gough was born on October 12, 1939 in Portsmouth, England. He attended thePorts-mouth Grammar School, and the University of
Leicester where he obtained his B.Sc. and Ph.D. The latter degree was obtained under the
super-vision of Professor M.C.R. Symons, ostensibly studyingion solvationandassociationusing
ultra-violet spectroscopy. In 1965he joined the
Depart-ment ofChemistry at the University ofWaterloo to establish anuclear magnetic resonancefacility.In 1976 he began a collaboration with Giacinto
Scoles which lead to the formation ofthe Centre for Molecular Beams and Laser Chemistry at the
University ofWaterloo. In 1989 he moved to the University ofVictoria as Chairman ofChemistry where he is currently studying vibrational over-tones ofmolecular beams and the FTIR spectro-scopy oflarge molecular clusters.
Reinhard Illner was born on
January
2, 1950inWabern,
Germany.
He studied Mathematics atthe Universities ofHeidelberg, Berkeley andBonn and obtained his Ph.D. under the supervision of
Professor J. Frehse in Bonn in 1976. He has held positions at the Universities of
Bonn,
Kaiserslau-tern, Duke University and the University of
Victoria, whereiscurrently full professor andchair of the Mathematicsand Statistics department. His main field of research is the mathematical theory
of Nonequilibrium Statistical Mechanics. He is
co-author of the monograph "The Mathematical