Multi-Scale Study of Sintering: A Review

Eugene A. Olevsky
w
Department of Mechanical Engineering, San Diego State University, San Diego, California 92182-1323
Veena Tikare and Terry Garino
Sandia National Laboratories, Albuquerque, New Mexico 87107
An integrated approach, combining the continuum theory of
sintering with a kinetic Monte-Carlo (KMC) model-based me-
sostructure evolution simulation is reviewed. The effective sin-
tering stress and the normalized bulk viscosity are derived from
mesoscale simulations. A KMC model is presented to simulate
microstructural evolution during sintering of complex micro-
structures taking into consideration grain growth, pore migra-
tion, and densication. The results of these simulations are used
to generate sintering stress and normalized bulk viscosity for use
in continuum level simulation of sintering. The advantage of
these simulations is that they can be employed to generate more
accurate constitutive parameters based on most general assump-
tions regarding mesostructure geometry and transport mecha-
nisms of sintering. These constitutive parameters are used as
input data for the continuum simulation of the sintering of pow-
der bilayers. Two types of bilayered structures are considered:
layers of the same particle material but with different initial
porosity, and layers of two different materials. The simulation
results are veried by comparing them with shrinkage and warp-
ing during the sintering of bilayer ZnO powder compacts.
I. Introduction
T
HE science of sintering has been being developed for over
60 years. In the last decade, the continuum theory of sinte-
ring has been rigorously elaborated and incorporated in a nite-
element model,
16
to consider the response of a complex-shaped
component with spatial variation in properties. This model is
now capable of accurately predicting macroscopic characteris-
tics owing to shrinkage during sintering of complex porous
powder compacts. The characteristics simulated by the sintering
model include the macroscopic dimensional changes as well as
the evolution of spatial distribution of stress, strain, and density
during sintering. The capability of this model has been demon-
strated in previous works, which have simulated sintering in
components with density variations,
1
sintering with gravity,
45
etc. While the model has enjoyed much success in simulating
sintering behavior, its ability to predict the exact properties of
specific powder compact during sintering is as accurate as the
constitutive parameters are (such as sintering stress, bulk, and
shear viscosity of the sintering body), that describe that partic-
ular powder compact. Unfortunately, the currently used consti-
tutive parameters are rather approximate as they are obtained
from analytical solutions of highly simplied grain and pore
structures. Usually, these solutions are calculated by considering
the evolution of a representative volume (unit cell) of packed
spherical particles, with pores in between. The need for macro-
scopic constitutive parameters based on an analysis of real me-
sostructures evolution during sintering cannot be overstated, as
this would enable improved process control of sintering com-
ponents.
In light of the above-mentioned, while a review article Theory
of Sintering: From Discrete to Continuum of Olevsky
1
solely
described macroscopic approaches in modeling of sintering, the
present paper is focused on a synergistic combination of both
scales of simulation (mesodiscrete, and macrocontinuum).
In the present paper, the material constitutive properties, in
particular, the effective sintering stress and the bulk viscosity,
are obtained from a mesoscale Monte-Carlo (MC) simulation
of the sintering of a realistic grain structure. The model is of
generic character and can be applied to a broad spectrum of
powder materials, including co-sintered or integrated ceramic
components. We demonstrate the usage of this model by apply-
ing it to a simulation of sintering of bilayered ceramic compos-
ites (an important example for low-temperature co-red ceramic
(LTCC) technologies).
This paper summarizes and reviews a number of results
713
obtained by the authors since 2001 when they began working on
the multi-scale modeling of sintering.
II. Constitutive Parameters Used in Macroscale Modeling
The continuum theory of sintering
1
describes the macroscopic
behavior of a porous body during sintering by relating an ex-
ternal load (corresponding to the stress tensors components s
ij
)
to the strain rate tensors components _ e
ij
by the nonlinear
viscous constitutive relationship
s
ij

sW
W
j_ e
ij
c
1
3
j
_ _
_ ed
ij
_ _
P
L
d
ij
(1)
where W is the equivalent strain rate and s(W) is the equiv-
alent stress; j and c are the normalized shear and bulk vis-
cosities; d
ij
is the Kronecker symbol (d
ij
51 if i 5j and d
ij
50 if
iaj); _ e is the rst invariant of the strain rate tensor, i.e. sum of
tensor diagonal components: _ e _ e
ii
_ e
11
_ e
22
_ e
33
. Physical-
ly, _ e represents the volume change rate of a porous body. The
porosity y is dened as the volume fraction of voids in a porous
body.
Equivalent strain rate W is depends on the invariants of the
strain rate tensor
W
1

1 y
p

j_ g
2
c_ e
2
_
(2)
where _ g is the second invariant of the strain rate tensor deviator
and represents, physically, the shape change rate of a porous
body
J
ournal
J. Am. Ceram. Soc., 89 [6] 19141922 (2006)
DOI: 10.1111/j.1551-2916.2006.01054.x
r 2006 The American Ceramic Society
1914
G. Messingcontributing editor
This work was partially performed at Sandia National Laboratories, a multiprogram
laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the USDOE
under the Contract DE-AC04-94AL-85000 and was supported by the National Science
Foundation, Division of Civil and Mechanical Systems (Grant CMS-030115), Division of
Materials Research (Grant DMR-0313346), and Division of Manufacturing and Industrial
Innovations (Grant DMI-0354857) is gratefully appreciated.
Presented at the 9th International Ceramic Processing Science Symposium, Coral Springs,
FL, Jan. 811, 2006.
w
Author to whom correspondence should be addressed. e-mail: eolevsky@mail.sdsu.edu
Manuscript No. 21345. Received January 7, 2006; approved February 11, 2006.
_ g _ e
ij

1
3
_ ed
ij
_ _
_ e
ij

1
3
_ ed
ij
_ _ _ _
1=2
(3)
It can be expressed in terms of the principal elongation rates
_ e
1
; _ e
2
; _ e
3
:
_ g
1

3
p

_ e
1
_ e
2

2
_ e
2
_ e
3

2
_ e
3
_ e
1

2
_
(4)
s(W) is responsible for the constitutive behavior of a porous
material. If s(W) is given by the linear relationship:
s(W) 52Z
0
W, where Z
0
is the shear viscosity of a fully dense
material, Eq. (1) is transformed to
s
ij
2Z
0
j_ e
ij
c
1
3
j
_ _
_ ed
ij
_ _
P
L
d
ij
(5)
Effective sintering stress P
L
is the product of the local sinte-
ring stress P
Lo
3a=r
0
(a is the surface tension, r
0
is the average
particle radius) and the normalized effective sintering stress

P
L
P
L
P
Lo

P
L

3a
r
0

P
L
(6)
In the constitutive relationship (1), three main parameters P
L
,
j, and c are functions of porosity, which have to be determined
based on the consideration of a poregrain structure evolving
during sintering. This requires a mesoscale analysis. Such an
analysis should be a natural addition to the macroscopic con-
tinuum theory of sintering, supplying an important information
on the status of the macroscopic constitutive parameters.
III. Mesoscale Sintering Simulations: Basic Concepts
For the utilization of the continuum theory of sintering,
1
the
exact values of the material characteristics are necessary, such as
sintering stress and the bulk viscosity. These parameters are not
widely available for most sintered materials. Many of analytical
expressions (usually micromechanics based) for the sintering
constitutive parameters make simplifying geometric assump-
tions to make the problem analytically tractable. In this paper,
we describe the mesoscale simulation of the microstructural ev-
olution of sintering in powder compacts. More accurate expres-
sions for sintering stress and the bulk viscosity necessary for
macroscale simulation of sintering can be obtained from the
details of the direct consideration of the realistic poregrain
structure evolution.
(1) Current Mesoscale Models of Sintering
Early models of sintering were used primarily to predict inter-
particle neck size and shrinkage as functions of diffusion mech-
anisms, diffusion coefcients, interfacial energies, grain size and
other material properties, and geometric measures. Microstruc-
tural evolution during sintering has been studied starting from
the late 1940 when Kuczynski,
14
Kingery and Berg,
15
Exner,
16,17
Johnson and colleagues,
1820
Coble,
21
Swinkels and Ashby,
22
and Nichols and colleagues
2325
considered idealized powder
compacts consisting of two or three spherical particles of equal
size sintered by various diffusion mechanisms. The main accom-
plishment of these early models was in understanding the driving
forces, transport mechanisms, and densication processes for
sintering of crystalline materials. Next, periodical unit cells of
the same geometry were utilized by DeHoff,
26
Bouvard and
McMeeking,
27
Svoboda and Riedel,
28
Riedel et al.,
29
and
Occhionero and colleagues
3032
for the description of sintering.
In these models, each repeating cell consisted of a matrix (grain
substance) and the voids imbedded in it. The intermediate sinter-
ing stage, where the solid and porous phase are interconnected
and the nal sintering stage where pores become isolated, could
be described in more detailed fashion in terms of the shapes of
grains and pores based on these models. In addition, some of
these idealized geometric simulations
2729
have been used to ob-
tain the sintering stress necessary for modeling sintering at the
continuum level.
Besides the above-mentioned attempts to derive analytical
solutions for the sintering constitutive parameters, in the last
15 years, numerical simulations have been used by many to
study sintering. Sintering of nano-particles
33,34
has been studied
by methods of molecular dynamics. These simulations showed
that additional mechanisms such as particle rotation may be
active in nano-systems. Very accurate particle shape evolution
has been modeled by continuum mechanics methods for the
sintering of two,
35
three,
36
and a row
37
of particles. Sintering of
two particles,
38
a row and close-packed particle,
27
and unit cells
of different packing
28
analyzed by continuum thermodynamic
principles have also given more accurate data on the shrinkage
kinetics during sintering. Sintering by surface diffusion in a
multiple particle system was studied by a cellular model
39
and an
MC model
40,41
was used to simulate nal-stage sintering of two
and multiple grains.
All of the above-mentioned numerical simulations have pro-
vided more accurate results for densication and other impor-
tant parameters. However, with the exception of two,
3941
they
are still far from being a true mesoscale simulation of sintering
as only a limited number of sintering particles are considered.
The two mesoscale simulations are related to viscous phase sin-
tering
39
in amorphous materials and to nal stage sintering in
crystalline materials.
41
In the present research, a mesoscale unit
cell including hundreds of particles serves as a basis for a kinetic,
Monte-Carlo (KMC) model used to simulate sintering. With
proper use of this KMC simulation, very few assumptions about
the geometry of the particles and their evolution during sintering
have to be made. Thus, more general thermodynamic (sintering
stress and bulk viscosity) and kinetic data (densication rate,
etc.) for sintering can be obtained from the KMC simulation.
(2) Model Description
A KMC model is used in this study that can simulate (i) coars-
ening of grains by short-range diffusion across grain boundaries,
(ii) pore migration and pore coarsening by surface diffusion, (iii)
vacancy diffusion along grain boundaries, and (iv) vacancy an-
nihilation at the grain boundaries. This model produces images
of the microstructure as a time function, which is related to real
time linearly. These series of microstructural images can be used
to calculate various constitutive parameters, which in our case
are sintering stress and normalized bulk viscosity.
The KMC model was used to simulate 2D microstructural
evolution during isothermal sintering. The model presented here
is limited to consideration of the following processes:
(1) long-range diffusion of material to pores by grain
boundary diffusion and surface diffusion;
(2) grain growth by short-range diffusion of atoms from
one side of the grain boundary to the other;
(3) vacancy annihilation at grain boundaries.
In the model, grain sites populating a square lattice can as-
sume one of Q distinct, degenerate states, where the individual
state is designated by the symbol q and the total number of
states in the system is Q, q
grain
5[1, 2, yQ]. The pore sites can
assume only one state, q
pore
51. Poregrain interfaces exist
between neighboring pore and grain sites and grain boundaries
exist between neighboring grain sites of different states, q.
The sum of all the neighbor interaction energies in the system
is given by
E
1
2

N
i1

8
j1
1 dq
i
; q
j

_ _
(7)
where N is the total number of sites, d is the Kronecker delta
with d(q
i
5q
j
) 51 and d(q
i
aq
j
) 50, q
i
is the state of the grain or
June 2006 Multi-Scale Study of Sintering 1915
pore at site i and q
j
is the state of the nearest neighbor at site j.
There are no pore boundaries and all pores sites coalesce as pore
sites can assume only one state, q
pore
51. Conversely, grain
sites can assume many different states making grain boundaries
possible. Thus, a two-component, two-phase system with uni-
form, isotropic interfacial energies between grains and between
grains and pores is formed. The only energy considered in the
simulation is the interfacial energy and all unlike neighbors con-
tribute one arbitrary unit of energy to the system.
The method developed in previous works
4245
is employed for
the simulation of grain growth. First, a grain site is chosen at
random from the simulation space. Then, a new state q is chosen
at random from the Q possible states in the system. The grain
site is temporarily assigned the new state and the change in en-
ergy is evaluated using Eq. (7). Next the standard Metropolis
algorithm is used to perform the grain growth step based on
Boltzmann statistics. A random number, R, between 0 and 1 is
generated. The transition probability, P, is calculated using
P
exp
DE
kBT
_ _
for DE > 0
1 for DE 0
_
(8)
where k
B
is the Boltzmann constant and T is temperature. If the
RrP, then the grain growth step is accepted, if not, the original
state is restored. The simulation temperature used for grain
growth was k
B
T50, which has been shown to simulate grain
growth well.
44
The total number of pore sites and grain sites is the same after
a pore migration step (pore migration is simulated using con-
served dynamics
41
). A pore site is chosen and next a neighboring
grain site is chosen. The two sites are hypothetically exchanged
with the grain site assuming a new state q where q results in the
minimum energy. This minimum-energy, poregrain exchange
simulates pore migration by surface diffusion.
46
Equation (7) is
used to calculate the change in energy for this exchange and
again the standard Metropolis algorithm is used to carry out the
pore migration step using Eq. (8) to determine the transition
probability. The simulation temperature used for the pore mi-
gration step was k
B
T50.7. This higher temperature was neces-
sary to simulate pore migration and is discussed in another
work.
47
Densication
48
in crystalline solids occurs by vacancy anni-
hilation at the grain boundaries. This process may be visualized
as described by DeHoff
26
as vacancies being painted on the
grain boundary, then an entire monolayer of the vacancies being
annihilated with the center of mass of the adjacent grains mov-
ing toward that grain boundary. The rate-limiting step is for
vacancies to diffuse to and cover the entire grain boundary. In
the MC model a vacancy is dened as a single, isolated pore site
that is not connected to any other pore sites. The algorithm used
for pore annihilation is the following. A pore site is chosen. If it
happens to be a vacancy, an isolated pore site, on a grain
boundary, it is annihilated. The frequency of the annihilation
attempts is adjusted to simulate the diffusion of vacancies to the
entire grain boundary. Annihilation is simulated as follows. A
straight line is drawn from the isolated pore site located on a
grain boundary through the center of mass of the adjacent grain
to the outside boundary of the sintering compact. Next, the iso-
lated pore site and the outside grain site are exchanged with the
grain site assuming the q state of the adjacent grain. This algo-
rithm conserves mass globally, moves the center of mass of the
adjacent grain toward the annihilation site, and annihilates a
vacancy. This algorithm to simulate densication does have the
artifact of moving mass from the outside boundary of the sim-
ulation to the interior. However, this artifact was found to have
a negligible effect on simulation results and is discussed in an-
other paper.
49
Time in the KMC model is measured in units of Monte-Carlo
step (MCS); one MCS corresponds to N attempted changes
where N is the total number of sites in the system. MC time is
linearly proportional to real time. The proportionality constant
of a given material can be found by comparing simulation mi-
crostructural evolution with that of the material (one can com-
pare the densication rates for the real material and obtained
through the mesoscale simulationssee Section IV(1).
The efciency of the above-mentioned approach has already
been demonstrated in our previous publications.
5052
IV. Results of Mesoscale Sintering Simulations
(1) Microstructural Evolution
The algorithm described above was applied to a 2D microstruc-
ture consisting of 30% initial porosity with simulation size of
500 500-squire sites. The starting microstructure had grains of
size d
g
510.6 and pores of size d
p
511.1, where d is the diameter
of a circle of equivalent area. The initial ratio of grain growth to
pore migration to pore annihilation attempts is 10:10:1. Then
the ratio changes to 10:10:n, where n is chosen to ensure the
simulation of vacancy annihilation in proportion to diffusion
along the length of the grain boundary. Thus, curvature driven
grain growth, mass transport by grain boundary diffusion, and
pore annihilation at the grain boundaries are simulated simul-
taneously. Microstructures at various stages during KMC sim-
ulation of sintering are shown in Fig. 1.
As the simulation continues, grains grow from very ne, in-
terconnected microstructure; the number of pores decreases,
pores become increasingly isolated and densication occurs as
pores shrink and disappear. The nal microstructure at time
t 5100 000 MCS, shows an almost fully dense microstructure
with grains that are two orders of magnitude larger in area than
the starting microstructure. These characteristics can be quan-
tied as a function of time and are presented in Figs. 2 and 3.
Figure 2 is a plot of density as a function of simulation time.
Rapid densication occurs early in the simulation when both
grains and pores are of a small size and have highly curved sur-
faces. Figure 3 is a plot of grain size and pore size (normalized
with respect to the cell size) as a function of time. Grains grew
during the entire simulation. This was anticipated and is ob-
Fig. 1. Microstructures of a sintering compact at times 50, 10000, and 30000 MCS. Grains are the white features separated by black grain boundaries.
Pores are drawn in gray.
1916 Journal of the American Ceramic SocietyOlevsky et al. Vol. 89, No. 6
served in most experimental systems. Pore size remained ap-
proximately constant as densication progressed suggesting that
pore grew by coalescence as overall porosity decreased by an-
nihilation. The microstructures shown in Fig. 1 are produced
from a simulation of size 100 100. However, the data shown in
Figs. 2 and 3 are from a simulation of size 500 500. The small-
er scale simulation is shown in Fig. 1, so that the microstructural
features may be seen more clearly.
(2) Sintering Stress
The simulated microstructures can be used to calculate the in-
terfacial-free energy for that microstructure. As a series of
microstructures are generated as a function of time, the interfa-
cial free energy as a function of any time dependent variable can
also be determined. This information can be used to calculate
the sintering pressure and bulk and shear viscosity for contin-
uum scale sintering models. This is a unique ability of these mi-
crostructural evolution simulations.
The continuum mechanics definition of sintering stress
1
is
P
L

qF
qyT
(9)
where F is interfacial-free energy and can be obtained by simply
measuring the pore surface length of the 2D microstructure
shown in Fig. 1. Parameter W is the area of the 2D sintering body
and is inversely proportional to the density r, 1/r. This, too, is
easily obtained from the simulations shown in Fig. 1. Numer-
ically differentiating the pore surface-free energy F with respect
to 1=rgives the sintering stress, which is plotted in Fig. 4 for the
microstructural evolution shown in Fig. 1. The sintering stress
shows a slight increase with increasing density (decreasing po-
rosity). Owing to the uctuating behavior of the free energy,
common in MC calculations, the numerical differentiation has
been applied to the spline approximation of the raw KMC data.
Using regression analysis, an analytical approximation of the
numerical data (shown in Fig. 4) has been obtained. The ana-
lytical approximation used for the normalized effective sintering
stress

P
L
(see Section II, which describes the relationship be-
tween

P
L
and P
L
) is of the form

P
L
a1 y
b
(10)
The porosity y is dened as 1 r=r
T
where r is density and r
T
is
the theoretical density. The unknown parameters a and b are
determined based on the minimum square deviations approach.
Finally, for the effective sintering stress the following analytical
expression is derived

P
L
1:71 y
0:26
(11)
Both numerical and approximate analytical results are plotted in
Fig. 4 and compared with the known expression of Skorohod
53

P
L
1 y
2
(12)
One can see that like the mesoscale simulations, the values of
the effective sintering stress are significantly higher. The differ-
ence between two models becomes smaller with higher relative
density. This is attributed to the different dimensionality of the
two models: the mesoscale simulations are in 2D, whereas,
Skorohods model is based on 3D (stochastic) analysis (Skoro-
hod
53
has derived Eq. (12) based on the consideration of a stoc-
hastic mixture of 3Dspherical voids embedded in a
continuous matrix).
(3) Effective-Normalized Bulk Viscosity
Continuum model of free sintering for a linear-viscous material
(see Eq. (5)), states the hydrostatic stress p and shrinkage rate _ e
are related as
P
L
2Z
0
c_ e (13)
where Z
0
is the shear viscosity of the fully dense substance;
P
L
P
Lo

P
L
, where P
Lo
is the local sintering stress. In accord-
ance with the Skorohods model,
53
P
Lo
3a=r
0
where a is the
surface tension, r
0
is the average particle radius. The condition
of continuity is given by
_
y
1 y
_ e (14)
0
5
10
15
20
25
30
0 20000 40000 60000 80000 100000
Monte - Carlo time steps, MCS
n
o
r
m
a
l
i
z
e
d

g
r
a
i
n

a
n
d

p
o
r
e

s
i
z
e

Grains
Pores
Fig. 3. Grain growth and pore growth kinetics obtained from meso-
scale simulations.
2
1.8
1.6
1.4
1.2
0.8
0.7 0.75 0.8
Potts Model Skorohod Model
0.85 0.95 0.9
0.6
1
Relative Density
N
o
r
m
a
l
i
z
e
d

S
i
n
t
e
r
i
n
g

S
t
r
e
s
s
Fig. 4. Comparison of Skorohod and Potts models for the effective
sintering stress.
0.7
0.75
0.8
0.85
0.9
0.95
1
0 20000 40000 60000 80000 100000
Monte - Carlo time steps, MCS
r
e
l
a
t
i
v
e

d
e
n
s
i
t
y
Fig. 2. Densication kinetics obtained by mesoscale simulations.
June 2006 Multi-Scale Study of Sintering 1917
The effective normalized bulk viscosity c can be determined
based on Eqs. (13) and (14)
c
P
L
1 y
2Z
0
_
y
or
c

P
L
1 y
2 dy=dt
(15)
where t
s
is the specific time of sintering and is related to real
physical time t as
t
s

_
t
0
P
Lo
Z
0
dt (16)
This normalized time representation
1
provides a sintering master
kinetic curve kind of description enabling the independence of
the calculation results with respect to powder material parameters.
Thus, if the effective sintering stress and the sintering kinetics
are known, the normalized bulk viscosity c can be determined.
Besides the effective sintering stress, presented in Fig. 4, the
mesoscale simulations enabled the determination of the sintering
kinetics (see Fig. 2). In order to use the mesoscale kinetic infor-
mation, the KMC simulation time must be related to the one
used in the continuum kinetics relationship (15). Time in the
KMC simulations is given in units of MCS, the number of spin
change attempts. These two time scales are adjusted as follows,
as a result of Eqs. (13) and (21) below
t
s
ln
y
i
y
_ _
4=3
(17)
where y
i
is the initial porosity.
This specific time of sintering is associated with the nal
number of attempts A
f
employed as the time scale in MC sim-
ulations and required to achieve the same porosity y
t
s
$ A
f
(18)
Then, the data in Fig. 2 can be replotted as a function of specific
time of sintering and used to nd the derivative of porosity with
respect to the specific time in Eq. (15). As a result, the effective
normalized bulk viscosity is determined numerically and plotted
as a function of porosity in Fig. 5.
Using regression analysis, a relation of the form given by Eq.
(19) is used to t the numerical data
c
2
3
1 y
c
y
d
(19)
The unknown parameters c and d are determined based on
the minimum square deviations approach. Finally, the following
analytical expression for the effective normalized bulk viscosity
is derived:
c
2
3
1 y
2:23
y
1:12
(20)
The described approach can be rened by employing Eq. (20)
with Eq. (13) to derive the next iteration of expression (17). Then
again Eq. (18) is employed, and so on.
The approximate analytical results are plotted in Fig. 5 and
compared with the known expression of Skorohod
53
c
2
3
1 y
3
y
(21)
One can see that, in accord with the mesoscale simulations,
the values of the effective normalized bulk viscosity are lower.
The difference between two models becomes smaller with higher
relative density. Similar to the deviations in the effective sintering
stress, this should be attributed to the different dimensionality of
the two models: the mesoscale simulations assume a 2D unit cell,
and Skorohod model is based on 3D (stochastic) analysis.
Using the determined analytical expressions (11) and (20) for
the effective sintering stress and the effective normalized bulk
viscosity, and solving Eq. (13), one can determine the free sin-
tering kinetics (the time dependence of relative density or po-
rosity). The obtained kinetics data can be compared with the
KMC data by employing the time transition approach described
above (the correspondence between KMC and physical times
see Eq. (18)).
V. Solution of the Problem of Sintering of Bilayered Ceramic
Composites
(1) Finite-Element Algorithm
Using the conventional nite-element analysis for a linear-vis-
cous porous body, Eq. (5) can be reduced to a set of linear
equations as functions of unknown nodal velocities
_
W
B
T
DB dW
_ _
fV
n
g
_
W
B
T
P
L
f1g dW (22)
where [B] is the matrix correlating the strain rates with the nodal
velocities V
n
; [D] is the matrix correlating the stresses with the
strain rates (matrix of viscosities); W is a macroscopic porous
volume under investigation. The right-hand part of the equation
represents the nodal forces. If material properties are uniformly
distributed in the volume W, the nodal forces associated with the
sintering stress P
L
(the rst term in the right-hand part of
Eq. (22)) will be equal to zero everywhere except of the nodes,
which belong to the external boundary of W. The multiplier
_
y
B
T
DB dW
_ _
in the left-hand part corresponds to the co-
efcients in the set of linear Eq. (22) relative to the unknown
nodal velocities V
n
.
In the following solutions, the effective sintering stress, the
effective normalized bulk viscosity, given by expressions (11)
and (20) are used. The effective normalized shear viscosity is
dened following Skorohod
jy 1 y
2
(23)
After the solution of set (22) with respect to the nodal veloc-
ities, the eld of strain rates is calculated. The new values of
0
0 0.1 0.2 0.3 0.4
20
40
60
80
100
120
Porosity
N
o
r
m
a
l
i
z
e
d

B
u
l
k

V
i
s
c
o
s
i
t
y
Normalized Bulk Modulus
(Skorohod)
Normalized Bulk Modulus (Pott's
Model - Analytical Approximation)
Fig. 5. Dependence of the normalized effective bulk viscosity on porosity.
1918 Journal of the American Ceramic SocietyOlevsky et al. Vol. 89, No. 6
relative densities are calculated using the continuity Eq. (14) for
each nite element. The nite-element calculations are carried
out in terms of the specific (dimensionless) time of sintering t,
given by Eq. (16).
(2) Solution of Problems of Sintering of Bilayer Ceramic
Composites
As it has been mentioned above, the developed model can be
applied to a broad range of possible applications including, in
particular, the processing of multi-layered ceramic composites.
The growth in radio frequency (RF) wireless communication
appliances (cell phones, pagers, PDAs, GPS, etc.) is driving the
development of multilayer ceramic technology to integrate RF
circuitry/components required for these applications. However,
the densication/sintering behavior of composite-layered mate-
rial systems is not well understood and co-re (co-sintering) ca-
pability is largely developed by trial and error experimental
methods. Progress in the integration of these layered material
systems will be greatly enhanced by the development of mate-
rials and process models that can predict the densication/
sintering behaviormuch as modeling has enhanced the devel-
opment of advanced semiconductor integrated devices.
54
When
incorporated in a nite-element code, the continuum theory of
sintering enables the solution of one of the most important
problems of co-ring multilayer ceramic compositesit can
predict warpage (camber formation) caused by differential sin-
tering.
The nite-element algorithm described above is used to sim-
ulate shape and dimensional changes during the co-ring (sin-
tering) of bilayer structures under four different conditions. The
conditions were chosen to address some of the reasons for shape
distortions in bilayered structures during co-sintering: (i) effect
of boundary conditionskinematic constrains imposed on one
of the layers; (ii) difference in layers porosities; (iii) difference in
layers material composition (chemical composition, grain size,
etc.). The inuence of thermal expansion is neglected in the fol-
lowing problems solutions. All the problem solutions shown in
Figs. 69 are obtained assuming a sintering regime correspond-
ing to the specific time of sintering (see Eq. (16)) t
s
52.0.
The rst simulation considered sintering of a monolayer sam-
ple on a rigid substrate. The initial porosity of the sample is
40%. The monolayer is assumed to be restricted from the mo-
tion parallel to the substrate at the interface with the substrate.
The results of the modeling are shown in Fig. 6. One can see the
upper unconstrained area of the sample has densied the most,
leading to a curved upper surface with inclined sides. Further-
more, the densication domain propagates starting from the
upper periphery toward the middle of the bottom. The areas of
lowest density are located at the bottom outside lateral parts
of the specimen. The results represented in Fig. 6 indicate that
boundary kinematic constrains (such as adhesion of one surface
to a rigid substrate) can cause a substantial shape distortion.
The second and third simulations are the sintering of bilayers on a
rigid substrate. The second is the sintering of a bi-porous struc-
ture (initial porosities are 40% for the top layer and 20% for the
bottom layer) with the same chemical composition of the two
layers or the same sintering stress and bulk viscosity for a given
porosity. The third problem is the sintering of a bilayer structure
with layers of the same initial porosity (40%) and different
shrinkage rates (sintering stress ratio is 1:50). One can think of
these bilayers as having different surface energies resulting in
different values of the sintering stress. The results of the mode-
ling are shown in Figs. 7 and 8, respectively. In both cases, the
bilayers warp with the top and side surfaces becoming curved
and the densication front propagates starting from the top pe-
ripheral areas, which shrink rapidly to the highest density. The
bottom peripheral zones have the highest porosity owing to the
imposed constraint from the rigid substrate. In both cases, a
porous core is formed in the top layer. Some difference was ob-
served between the densication front propagation for the two
problems considered. In the case of the bi-porous structure, the
interface boundary serves as a source for the densication wave,
while the bilayer with differences in chemical composition main-
tains a self-similar (steady shape) density (porosity) spatial dis-
tribution in the bottom layer.
The fourth simulation is the free sintering of a bilayer structure
with the uniform initial distribution of porosity (40%) and with
different shrinkage rates for the two layers. The ratio of the ef-
fective sintering stresses is assumed to be 1:50. Figure 9 shows a
3D image of the sintered cylindrical bilayer specimen with its
porosity distribution. Owing to the difference in shrinkage rates,
Fig. 6. Sintering of a porous monolayer on a rigid substrate. Initial
porosity is 40%.
Fig. 7. Sintering of a biporous structure on a rigid substrate. Initial
porosity is 40% in top layer and 20% in bottom layer.
Fig. 8. Sintering of a bilayered structure on a rigid substrate. Initial
porosity in both layers is 40%, sintering stress ratio is 1:50.
June 2006 Multi-Scale Study of Sintering 1919
the top layer contracts faster than the bottom one, causing the
pronounced bending of the specimen.
VI. Model Experiments on Bilayered Zinc Oxide Powders
(1) Experimental Determination of the Constitutive
Parameters of the Powder Material
The solid phase shear viscosity of Aldrich 0.2 mm ZnO was
measured (St. Louis, MO). Cylindrical ZnO samples (cold
pressed under 20 ksi), 3/8 inches in diameter, were heated at
51C/min in a high temperature MTS machine with cyclic loading
schedule (following the approach of Green et al.
55
): nominal
load of 0.5 MPa applied uniaxially for 251C and then removed
for 251C (0.03 MPa applied) starting at 6501C.
From Eq. (5), one can obtain
_ e
11

P
L
6Z
0
c

2c
1
3
j
6Z
0
jc
s
11

c
1
3
j
2c
1
3
j
s
22
s
33

_ _
(24)
Under uniaxial loading, s
22
s
33
0, and Eq. (24) reduces to
_ e
11

s
11
_ e
11
_ e
f
2c
1
3
j
6jc
(25)
where _ e
f
is the linear shrinkage rate for free sintering
_ e
f

P
L
6Z
0
c
(26)
The parameter Z
u
s
11
=_ e
11
_ e
f
is the uniaxial viscosity. It
has been calculated from empirically determined linear shrink-
age for sintering under load _ e
11
and for free sintering _ e
f
at
the same porosity. The measured uniaxial viscosity has been
substituted in Eq. (25) for the determination of the material
shear viscosity Z
0
. Here the expressions (20) and (23) have been
used for the normalized bulk and shear viscosity, respectively.
The shear viscosity Z
0
obtained in this manner is plotted in
Fig. 10 as a function of temperature T. The quadratic trinomial
was tted to the Z
0
(T) data using the mean-square deviations
regression analysis
Z
0
T 51:7
T
750
_ _
2
106:6
T
750
56:4
_ _
10
10
Pa s (27)
The densication curve for free sintering of ZnO heated at a
rate of 51C/min is given in Fig. 8. Substituting (11) and (20) in
(26), as well as considering Eq. (14) and relationships _ e 3_ e
11
,
r51y, and P
L
P
Lo

P
L
2a=r
0

P
L
(see Section II), a dif-
ferential equation describing the shrinkage rate for free sintering
can be derived
_ r
_
y
2:55a
r
0
Z
0
1 r
1:12
r
0:97
(28)
where a is the surface tension and r
0
is the average radius of
particles. Substituting Z
0
(T) from Eq. (27), one can solve Eq.
(28) for a given value of the surface tension a of a ZnO powder.
The value of a has been estimated using the mean-square devi-
ation regression analysis to obtain the best t to the experimen-
tal data shown in Fig. 11 and is
a 1:27 J=m
2
(29)
(2) Experimental Verication of the Model Results on
Distortion of Bilayered Specimens
The model was veried by comparing the dimensional changes
predicted by continuum simulations for the sintering of ZnO
bilayers with those observed in experiment. A bilayer of ZnO
powder compact (0.2 mm, Aldrich) was made by uniaxially
pressing each layer individually and then pressing both layers
together. The top layer was pressed at 5 ksi and had 45% initial
density (pressed at 5 ksi) and the bottom layer, pressed at 20 ksi,
had 57% initial density. Both layers were then pressed together
at 20 ksi and assumed 47% and 57% pre-sintering density, re-
spectively. The bilayer was sintered at a heating rate of 51C/min
up to a temperature of 10001C. Final dimensions were 6.7 mm
Fig. 9. Free sintering of a bilayered structure. Initial porosity for both
layers are 40%, sintering stress ratio is 1:50.
3
2.5
1.5
0.5
0
1
2
600 650
Experimental data Regression analysis
700 750
Temperature (C)
S
o
l
i
d

P
h
a
s
e

V
i
s
c
o
c
i
t
y

(

1
0
E
1
0

P
a

s
)
800 850 900
Fig. 10. Dependence of the shear viscosity on temperature for ZnO
powder.
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 2000 4000 6000
Regression analysis Experimental data
8000 10000 12000 14000
Time, s
R
e
l
a
t
i
v
e

D
e
n
s
i
t
y
Fig. 11. Sintering kinetics for ZnO powder.
1920 Journal of the American Ceramic SocietyOlevsky et al. Vol. 89, No. 6
(long) 2.15 mm (thick) 3.89 mm (deep). The results of the
experiment are shown in Fig. 12, in the form of consecutive im-
ages of the specimens prole.
The sintering of a ZnO bilayer with the same porosities and
dimensions was simulated using the continuum model incorpo-
rated in the nite-element code. The normalized bulk and shear
viscosity and the effective sintering stress were used in accord-
ance with Eqs. (20), (23), and (11), respectively. The shrinkage
rate and surface tension for ZnO calculated from mesoscale
simulations and from experiments and given by Eqs. (27) and
(29), respectively, were utilized in the continuum simulations.
The results of the simulation are compared with experimental
results in Fig. 12. One can see that the modeling results agree
satisfactorily with the experimental data on the sample distor-
tion. The quantitative comparison has been carried out by em-
ploying the shape distortion parameter introduced by Olevsky
et al.
5
The shape distortion has been calculated for both exper-
imental and modeling-based images of the cross-sections of the
bilayered sintered ZnO specimen. The deviation is about 11%.
VII. Conclusions
We have demonstrated the use of a novel method to combine
mesoscale and continuum scale models to simulate sintering in
complex powder compacts with minimal experimental measure-
ments of material parameters. A mesoscale model was used to
simulate microstructural evolution in a ZnO powder compact.
These simulations were used to obtain the normalized effective
sintering stress and the normalized bulk viscosity. Experiments
on ZnO powder compacts helped x these relative (normalized)
parameters to absolute ones by measuring the shear viscosity
and surface tension. These material parameters were used in a
nite-element model based on the continuum theory of sintering
to successfully predict the dimensional changes during sintering
of ZnO bilayers.
References
1
E. A. Olevsky, Theory of Sintering: From Discrete to Continuum. Invited
Review, Mater. Sci. Eng. Rep., 23 [2] 40100 (1998).
2
E. A. Olevsky, H. J. Dudek, and W. A. Kaysser, HIPing Conditions for
Processing of Metal Matrix Composites Using Continuum Theory for Sintering I.
Theoretical Analysis, Acta Metall. Mater., 44, 70713 (1996).
3
E. A. Olevsky, H. J. Dudek, and W. A. Kaysser, HIPing Conditions for
Processing of Metal Matrix Composites Using Continuum Theory for Sintering II.
Application to Fibre Reinforced Titanum Alloys, Acta Metall. Mater., 44, 715
24 (1996).
4
E. A. Olevsky and R. M. German, Effect of Gravity on Dimensional
Change During Sintering, I. Shrinkage Anisotropy, Acta Mater., 48, 115366
(2000).
5
E. A. Olevsky, R. M. German, and A. Upadhyaya, Effect of Gravity on
Dimensional Change During Sintering, II. Shape Distortion, Acta Mater., 48,
116780 (2000).
6
E. A. Olevsky and A. Molinari, Instability of Sintering of Porous Bodies,
Intern. J. Plasticity, 16, 137 (2000).
7
E. A. Olevsky and V. Tikare, Combined MacroMeso Scale Modeling of
Sintering. Part I, Continuum Approach, NATO Sci. Ser., Sub-Ser. III: Comput.
System Sci., 176, 8593 (2001).
EXPERIMENT MODELING
green specimen
sintered specimen
25C
800C
892C
1001C
Fig. 12. Free sintering of a bilayered ZnO powder specimen. Comparison of experimental and modeling results on shape distortion.
June 2006 Multi-Scale Study of Sintering 1921
8
V. Tikare, E. A. Olevsky, and M. V. Braginsky, Combined Macro-Meso Scale
Modeling of Sintering. Part II, Mesoscale Simulations, NATO Sci. Ser., Sub-Ser.
III: Comput. System Sci., 176, 94104 (2001).
9
E. A. Olevsky, A. L. Maximenko, J. H. Arterberry, and V. Tikare, Sintering
of Multilayer Powder Composites: Distortion and Damage Control, Adv. Powder
Metall. Partic. Mater., 5, 15664 (2002).
10
E. Olevsky, Modeling of Sintering: Challenges and Further Development;
pp. 2733 in Proceedings of International Conference On Process Modeling in Pow-
der Metallurgy And Particle Material, Newport-Beach, Edited by A. Lawley and
J. Smugeresky. MPIF, 2002.
11
E. A. Olevsky, B. Kushnarev, A. L. Maximenko, M. Braginsky, and V.
Tikare, Multi-Scale Modeling of Sintering Shrinkage Anisotropy, p. 6 in Sin-
tering 2003, Edited by R. M. German, G. L. Messing, and R. G. Cornwall. Marcel
Dekker, New York, 2003.
12
E. A. Olevsky, A. L. Maximenko, J. Arterberry, and V. Tikare, Multi-Scale
Modeling of Sintering: Application to Laminated Composites, p. 6 in Proceedings
of the Nineth International Conference on Mechanical Behavior of Materials, Ge-
neva, Edited by S. R. Bodner, D. Rittel, and D. Sherman. MBM Soc., Geneva,
Switzerland, 2003.
13
E. A. Olevsky, B. Kushnarev, A. L. Maximenko, and V. Tikare, Hierarchical
Analysis of Sintering Anisotropy, p. 6 in Proceedings of Powder Metallurgy
World Congress, Edited by A. Fernandez. EPMA, Vienna, Austria, 2004.
14
G. C. Kuczynski, Self-Diffusion in Sintering of Metal Particles, Trans. Am.
Inst. Min., 185, 16978 (1949).
15
W. D. Kingery and M. Berg, Study of the Initial Stages of Sintering Solids by
Viscous Flow, EvaporationCondensation, and Self-Diffusion, J. Appl. Phys., 26,
1205 (1955).
16
H. E. Exner, Neck Shape and Limiting Gbd Sd Ratios in Solid-State Sin-
tering, Acta Metall., 35, 587 (1987).
17
H. E. Exner, Principles of Single Phase Sintering, Rev. Powder Metall. Phys.
Ceram., 1, 7 (1979).
18
D. L. Johnson and I. B. Cutler, Diffusion Sintering. I. Initial Stage Sintering
Models and their Application to Shrinkage of Powder Compacts, J. Am. Ceram.
Soc., 46, 541 (1963).
19
D. L. Johnson and T. M. Clarke, Grain Boundary1Volume Diffusion in
Sintering of Silver, Acta Metall., 12, 1173 (1964).
20
D. L. Johnson, New Method of Obtaining Volume Grain-Boundary and
Surface Diffusion Coefcients from Sintering Data, J. Appl. Phys., 40, 192 (1969).
21
R. L. Coble, Initial Sintering of Alumina and Hematite, J. Am. Ceram. Soc.,
41, 55 (1958).
22
F. B. Swinkels and M. F. Ashby, Overview 11A 2nd Report on Sintering
Diagrams, Acta Metall., 29, 259 (1983).
23
F. A. Nichols and W. W. Mullins, Morphological Changes of a Surface of
Revolution Due to Capillarity-Induced Surface Diffusion, J. Appl. Phys., 36,
1826 (1965).
24
F. A. Nichols, Coalescence of 2 Spheres by Surface Diffusion, J. Appl.
Phys., 37 [7] 2805 (1966).
25
R. M. German and J. F. Lathrop, Simulation of Spherical Powder Sintering
by Surface-Diffusion, J. Mater. Sci., 13, 921 (1978).
26
R. T. DeHoff, New Directions for Materials Processing and Microstructural
Control; in Science of Sintering, Edited by D. P. Uskokovic., et al Plenum Press,
New York, 1989.
27
D. Bouvard and R. M. McMeeking, Deformation of Interparticle Necks by
Diffusion-Controlled Creep, J. Am. Ceram. Soc., 79 [3] 666 (1996).
28
J. Svoboda, H. Riedel, and H. Zipse, Equilibrium Pore Surfaces, Sintering
Stresses and Constitutive-Equations for the Intermediate and Late Stages of Sin-
tering. I. Computation of Equilibrium Surfaces, Acta Metall., 42 [2] 435 (1994).
29
H. Riedel, H. Zipse, and J. Svoboda, Equilibrium Pore Surfaces, Sintering
Stresses and Constitutive-Equations for the Intermediate and Late Stages of Sin-
tering. 2. Diffusional Densication and Creep, Acta Metall., 42 [2] 445 (1994).
30
M. A. Occhionero and J. W. Halloran, The Inuence of Green Density Upon
Sintering; pp. 89102 in Sintering and Heterogeneous Catalysis, Edited by G. C.
Kuczynski, A. E. Miller, and G. A. Sargent. Plenum Press, New York, 1984.
31
J. Zhao and M. P. Harmer, Effect of Pore Distribution on Microstructure
Development. I. Matrix Pores, J. Am. Ceram. Soc., 71 [2] 113 (1988).
32
T. Ikegami, Microstructural Development During Intermediate-Stage and
Final-Stage Sintering, Acta Metall., 35, 667 (1987).
33
P. Zeng, S. Zajac, P. C. Clapp, and J. A. Rifkin, Nanoparticle Sintering
Simulations, Mater. Sci. Eng., A252, 301 (1998).
34
K. Tsuruta, A. Omeltchenko, R. K. Kalia, and P. Vashishta, Early Stages of
Sintering of Silicon Nitride Nanoclusters: A Molecular-Dynamics Study on Par-
allel Machines, Eur. Phys. Lett., 33, 441 (1996).
35
W. Zhang and J. H. Scheibel, The Sintering of 2 Particles by Surface and
Grain-Boundary DiffusionA 2-Dimensional Numerical Study, Acta Metall.,
43, 4377 (1995).
36
H. Zhou and J. J. Derby, Three-Dimensional Finite-Element Analysis of
Viscous Sintering, J. Am. Ceram. Soc., 81, 478 (1998).
37
A. Jagota and P. R. Dawson, Micromechanical Modeling of Powder Com-
pacts. I. Unit Problems for Sintering and Traction Induced Deformation, Acta
Metall., 36, 2551 (1988).
38
J. Pan, H. Le, and S. Kucherenko, A Model for the Sintering of Spherical
Particles of Different Sizes by Solid State Diffusion, Acta Metall., 13, 4671 (1998).
39
J. W. Bullard, Digital-Image-Based Models of Two-Dimensional Micro-
structural Evolution by Surface Diffusion and Vapor Transport, J. Appl. Phys.,
81 [1] 15968 (1997).
40
S. Bordere, Original Monte Carlo Methodology Devoted to the Study of
Sintering Processes, J. Am. Ceram. Soc., 85 [7] 184552 (2002).
41
G. N. Hassold, I.-W. Chen, and D. J. Srolovitz, Computer-Simulation of
Final-Stage Sintering. I. Model, Kinetics, and Microstructure, J. Am. Ceram.
Soc., 73 [10] 285764 (1990).
42
M. P. Anderson, D. J. Srolovitz, G. S. Grest, and P. S. Sahni, Computer-
Simulation of Grain-Growth. I. Kinetics, Acta Metall., 32 [5] 78391 (1984).
43
J. Wejchert, D. Weaire, and J. P. Kermode, Monte-Carlo Simulation of
the Evolution of a Two-Dimensional Soap Froth, Philos. Mag., B53, 1524
(1986).
44
E. A. Holm, J. A. Glazier, D. J. Srolovitz, and G. S. Grest, Effects of Lattice
Anisotropy and Temperature on Domain Growth in the 2-Dimensional Potts-
Model, Phys. Rev. A, 43 [6] 26628 (1991).
45
D. J. Srolovitz, G. S. Grest, M. P. Anderson, and A. D. Rollett, Computer-
Simulation of Recrystallization. II. Heterogeneous Nucleation and Growth, Acta
Metall., 36 [8] 211528 (1988).
46
V. Tikare and E. A. Holm, Simulation of Grain Growth and Pore Migration
in a Thermal Gradient, J. Am. Ceram. Soc., 81 [3] 4804 (1998).
47
V. Tikare and J. D. Cawley, Numerical Simulation of Grain Growth in Liq-
uid Phase Sintered MaterialsII. Study of Isotropic Grain Growth, Acta Me-
tall., 46 [4] 134356 (1998).
48
B. H. Alexander and R. W. Balluf, The Mechanism of Sintering of Cop-
per, Acta Metall., 5, 66677 (1957).
49
V. Tikare, M. V. Braginsky, E. A. Olevsky, and R. T. DeHoff, A Combined
Statistical-Stereological Model for Simulation of Sintering; pp. 40541 in Sinte-
ring Science and Technology, Edited by R. German and G. Messing. Marcel
Dekker, New York, 2000.
50
E. A. Olevsky, B. Kushnarev, A. Maximenko, V. Tikare, and M. Braginsky,
Modeling Anisotropic Sintering in Ceramics, Philos. Mag., 85, 212346 (2005).
51
M. Braginsky, V. Tikare, and E. Olevsky, Numerical Simulation of Solid
State Sintering, Int. J. Solids Struct., 42, 62136 (2005).
52
V. Tikare, M. Braginsky, E. Olevsky, and D. L. Johnson, Numerical
Simulation of Anisotropic Shrinkage in a 2D Compact of Elongated Particles,
J. Am. Ceram. Soc., 88 [1] 5965 (2005).
53
V. V. Skorohod, Rheological Basis of the Theory of Sintering. Kiev, Naukova
Dumka, 1972.
54
S. H. Lee, G. L. Messing, and D. J. Green, Warpage Evolution of Screen
Printed Multilayer Ceramics During Co-Firing, Key Eng. Mater., 264268, 3218
(2004).
55
D. J. Green, P. Z. Cai, and G. L. Messing, Residual Stresses in Alumina-
Zirconia Laminates, J. Eur. Ceram. Soc., 19, 251117 (1999). &
1922 Journal of the American Ceramic SocietyOlevsky et al. Vol. 89, No. 6