International Journal of Heat and Mass Transfer 100 (2016) 355–361

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

journal homepage: www.elsevier.com/locate/ijhmt

A numerical investigation on the heat conduction in high filler loading

particulate composites
Zhen Tong a, Meng Liu b, Hua Bao a,⇑
a
University of Michigan-Shanghai Jiao Tong University Joint Institute, Shanghai Jiao Tong University, Shanghai 200240, China
b
School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China

a r t i c l e i n f o a b s t r a c t

Article history: Particle-filled composite materials have been widely used as thermal interface materials (TIMs) to reduce
Received 3 February 2016 the thermal contact resistance. For industrial applications, the particle-filled composite usually has high
Accepted 25 April 2016 volume fraction (>50%). However, most of the research on the thermal properties of particle-filled com-
Available online 10 May 2016
posites has been focusing on low volume fraction composites. In this work, the finite element method
(FEM) is adopted to investigate the particle-filled composites with high filler loading. We consider the
Keywords: close-packed simple cubic (SC), face-centered cubic (FCC), and a dual diameter (DD) model with even
Finite element method
a higher volume fraction than the FCC. It is found that with a high volume fraction, small increase in vol-
Composite material
Thermal conductivity
ume fraction can lead to a strong enhancement in the overall thermal conductivity. With a certain filler
High volume fraction loading and thermal conductivity of the matrix, the effective thermal conductivity first dramatically
increases with the thermal conductivity of the filler and then saturates. We show that the effective med-
ium theory based models cannot properly predict the effective thermal conductivity for the close-packed
structures. The percolation theory based on the resistance network agrees surprisingly well with our sim-
ulation results. Through a careful investigation of the effect of proximity between adjacent particles, it is
found that good contact between particles is crucial to the enhancement of the overall thermal conduc-
tivity. We also considered the interface thermal resistance between fillers and matrix and compared the
simulation results with analytical models. Our analysis provides a better understanding on the heat
transfer in the high volume fraction composite materials and is important for the fabrication of high ther-
mal conductivity TIMs.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction Recent experimental investigations of particle filled TIMs

With the increasing power density of electronic device, heat
dissipation has become one of the most critical challenges [1]. TIMs
[2] have been widely used in electronic packaging and cooling sys-
tem to reduce the contact thermal resistance across jointed sur-
faces, such as the contact between microprocessors and heat
sinks [3]. Particle-filled composite materials made of thermal
grease matrix (with typical thermal conductivity of 0.1–0.4 W/
m K [3,4]) and highly conductive ceramic or metallic particles are
the most widely used commercial TIMs, of which the thermal con-
ductivity values are generally in the range of 2–5 W/m K [4].
Although there are practical concerns such as thermal stability,
wetting ability, and viscosity, to further increase the effective ther-
mal conductivity of TIMs is still highly desirable to the electronic
packaging industry.
⇑ Corresponding author. Tel.: +86 21 34205660.
E-mail address: hua.bao@sjtu.edu.cn (H. Bao).
mainly take two different approaches, i.e., increasing the volume
fraction and mixing different types of fillers together. For example,
Yu et al. [5] reported that the composite of hybrid size alumina fil-
lers with the addition of graphene of 1 wt.% filled in silicone matrix
had an effective thermal conductivity of 3.45 W/m K at a filler vol-
ume fraction of 63%. Gao et al. [6] found that the alumina with a
diameter of 75 lm filled silicone matrix composite had an effective
thermal conductivity of 2.25 W/m K at a high volume fraction of
62%. Some commercial TIMs even report thermal conductivity as
high as 17 W/m K [7], although the data have not been verified
by other sources. It can be seen that the reported effective thermal
conductivity of composites with high volume fraction varies signif-
icantly in different studies.
Actually, the effective thermal conductivity of composites is
determined by many factors [8] other than the volume fraction,
such as the particle type and shape, the collective arrangement
of particles, and the interfacial thermal resistance between fillers
and matrix. It is generally difficult to separate the effect of these

http://dx.doi.org/10.1016/j.ijheatmasstransfer.2016.04.092
0017-9310/Ó 2016 Elsevier Ltd. All rights reserved.
356 Z. Tong et al. / International Journal of Heat and Mass Transfer 100 (2016) 355–361
Nomenclature
T1, T2 (K) temperature
kf (W/m K) thermal conductivity of filler
km (W/m K) thermal conductivity of matrix
keff (W/m K) effective thermal conductivity
Q (W) heat flow rate
A (m2) cross-sectional area
DT (K) temperature difference
L (m) length of the cubic model
D (m) diameter of the particle
d (m) gap distance
f volume fraction
DS (m2) contact area of the cushion
S (m2) surface area of the particle
t (m) the half of the thickness of the cushion
factors by purely experimental approach. Therefore, theoretical
understanding is vital to guide the design and fabrication of better
composite material for thermal application. There are two main
categories of analytical or semi-analytical theories. One is the
effective medium theory. The widely used models include the
Maxwell [9] and Bruggeman [10] symmetric model. These simple
models only consider the effect of the volume fraction and are gen-
erally valid for low volume fraction (usually less than 40 vol% [11]).
Another approach is based on the Rayleigh’s mathematical tech-
nique of expressing the temperature field as a multipole expansion
in the composites [12]. This technique works for cubic arrange-
ments (including SC, BCC, FCC) of spherical fillers and gives more
accurate solutions for larger volume fraction. However, the Ray-
leigh’s original treatment was later questioned by Levine et al.
[13] because it just considered low order multipole expansion
coefficients and involved the summation of a non-absolutely con-
vergent series. In order to overcome these difficulties, McPhedren
et al. [14] and McKenzie et al. [15] extended the Rayleigh’s work
to take into account multipoles of arbitrarily high order and over-
come the non-absolutely convergent difficulty for the case of SC
particle arrangement and the case of BCC and FCC, respectively.
Sangani et al. [16] modified the multipole expansion by taking
advantages of the symmetric particles and the periodicity of the
temperature to calculate the effective thermal conductivity for
cubic array of spheres (including SC, BCC, FCC) from dilute to
nearly touching volume fraction. There are also other approaches
that investigate the effective thermal conductivity of composites
for cubic array of spheres, such as the boundary integral method
developed by Zick [17] and the asymptotic expression derived by
Keller [18]. However, these works generally only consider cubic
lattice arrangement of particles and are difficult to apply to volume
fraction higher than the closed-packed FCC structure. To further
include the effect of interface thermal resistance, several exten-
sions of the analytical theories have been developed. For example,
Hasselman and Johnson [19] derived an expression based on Max-
well model that includes the effect of interface thermal resistance.
With the modification of Bruggeman model, Every et al. [20] pro-
posed an expression that includes a dimensionless parameter Biot
number to consider the interfacial thermal resistance between fil-
lers and matrix. Cheng et al. [21] applied the Rayleigh’s method to
calculate the effective thermal conductivity of periodic arrays of
spherical inclusions with interfacial thermal resistance, and an
accurate approximation formula was provided. Some of the models
are summarized in review articles [22,23].
On the other hand, the direct numerical solution of the heat
diffusion equation [24] can obtain detailed information of heat
transfer in composites and theoretically consider any type of
d (m) thickness of the interfacial layer
kc (W/m K) thermal conductivity of the cushion
ks (W/m K) thermal conductivity of the shell
Rb (m2 K/W) interfacial thermal resistance
Rs (m2 K/W) thermal resistance of the shell
Rm (m2 K/W) the matrix intrinsic thermal resistance correspond-
ing to thickness of d
Bi Boit number
f2 geometric parameter
b1 dimensionless parameter
R dimensionless parameter
k ratio of thermal conductivity of fillers to matrix
a dimensionless parameter
d0 (m) critical diameter
arrangement of particles. Numerical simulations were performed
to study different types of fillers and different arrangements of fil-
lers, such as cubic [25], spherical elliptic fillers [26], and fibers [27],
with ordered or random distribution [28]. The lattice Boltzmann
method is also employed to calculate the effective thermal conduc-
tivity of particulate TIMs in squeeze flow [29] and study the heat
transfer in porous media [30]. Most of these studies are focusing
on composites with relatively low filler volume fraction as well.
There are also some methods developed for high volume fraction,
such as the Agari’s [31] semi-empirical model which requires
experimental data to obtain the fitting curve as the function of fil-
ler concentration. Some other methods are based on the assump-
tion that the composites can be treated as resistance networks.
For example, the effective unit cell model (EUCM) [32] uses the
effective medium theory to determine the value of the resistance,
and another model uses the FEM [33] to obtain the value of the
resistance. The percolation model proposed by Devpura et al.
[34] models the composite material as a binary compound of cubic
units with the same size and then uses the resistance network to
obtain the effective thermal conductivity of the composite. It needs
an empirical correction factor to be applied to spherical fillers.
These models for a high volume fraction either need fitting param-
eters, or the assumption that the system can be modeled as a resis-
tance network. In contrast, a direct numerical simulation of the
high filler loading composite materials does not require any
assumptions. It can help to develop a better understanding of the
heat transfer mechanism in the high filler loading composite and
also to validate the numerous models that have been proposed in
literature.
In this work, a numerical simulation of particle-filled compos-
ites is carried out, focusing on high filler loading composites model.
A few high volume fraction unit cell models are established. On the
premise of neglecting the microscale effect, the FEM is employed to
investigate the effects on the effective thermal conductivity of
these cells. The factors, such as volume fraction, thermal conduc-
tivity of the filler, the proximity effect of closely packed fillers,
and the interfacial thermal resistance between fillers and matrix,
have been analyzed. The results are also compared with different
models and discussed.
2. Model description and simulation details
To model high volume fraction composite materials, ideally one
should build a large simulation domain and have many particles
randomly placed in the domain [35]. However, for the high volume
fraction configurations, many particles will be close to each other.
As will be shown later, the proximity effect between adjacent
 Z. Tong et al. / International Journal of Heat and Mass Transfer 100 (2016) 355–361
particles has a strong effect on the overall thermal conductivity,
and extremely fine mesh has to be used to properly resolve the
gap between adjacent particles. Therefore, to achieve a high vol-
ume fraction, keep a reasonable amount of computational cost,
and preserve the accuracy, we considered ordered arrangement
of close-packed fillers in the composite. Three composite models
have been established, including simple cubic (SC), face-centered
cubic (FCC) and a dual diameter (DD) model by filling the voids
in FCC model with other smaller size particles. The unit cells of
these models are shown in Fig. 1. In the subsequent discussions,
we define D to be the diameter of the particle in the SC or FCC
model, and the diameter of larger particle in the DD model.
In the finite element simulation, a representative cubic unit cell
is used. We take the SC unit cell model shown in Fig. 1(a) as an
example. The upper and lower surface (i.e., ABCD and EFGH) are
set to be constant temperatures T1 and T2, respectively. The other
four surfaces are assumed to be adiabatic. For such a model, the
effective thermal conductivity of the unit cell can be calculated by
QL
keff ¼ ; ð1Þ
ADT
where keff is the effective thermal conductivity, Q is the heat flow
rate, A is the cross-sectional area of the heat flow, DT is the temper-
ature difference between upper and lower surface (T 1  T 2 ), and L is
the height of the simulation domain. The tetrahedral element is
used to mesh the model, and then the steady state heat diffusion
equation is solved by FEM. The heat flow rate Q, across the upper
and lower surface, is calculated by integrating the heat flux on
the corresponding surface.
3. Results and discussion
3.1. Convergence test
Unlike the low volume fraction case, for close-packed struc-
tures, the fillers are treated as spheres that are tangential to each
other. However, an exact tangential structure has infinitely small
feature size, which cannot be resolved with a practical mesh den-
sity. A small gap d always exists between neighboring particles, as
shown in the inset of Fig. 2(a). In our simulations, the smallest ele-
ment has an edge length of 105 of the side length of the cube and
the number of elements varies from 3  105 to 2  106 depending
on the particular model. To find out the smallest gap that our mesh
can resolve, we tested different gap distances and compared our
FEM results with the semi-analytical data provided by Sangani
et al. [16]. Since the data for FCC and infinite particle thermal con-
ductivity were available in these references [16,36], this case was
chosen to validate our simulation results. The thermal conductivity
ratio of the particle to the matrix is set to be 107 in our simulations
to approximate the infinite thermal conductivity. From Fig. 2(a),
Fig. 1. Composite unit cell models. (a) SC composite unit cell model, the km represents the matrix, and the kf represents the filler, the same as below. (b) FCC composite unit
cell model. (c) DD composite unit cell model. The smaller particles have the same size.
357
we see that our result agrees well with the Sangani’s result when
the ratio of gap to the diameter (d/D) is larger than 0.001. It starts
to deviate from the semi-analytical value when d/D is smaller than
0.001. This indicates that our mesh can properly resolve the gap as
long as d/D is larger than 0.001. Therefore, in all the subsequent
simulations, the models are built so that d/D is 0.001. For the DD
case, the gaps between any two large particles are kept the same
value (d/D = 0.001). It should be pointed out that when the gap
between the spheres is vanishing, the effective thermal conductiv-
ity of the composite material increases significantly, as shown in
Fig. 2(b). Such a conclusion is true as long as the particles have
much larger thermal conductivities than the matrix material
[14,15], which is usually the case for TIMs. Therefore, one should
take extra caution while performing FEM simulation for those
nearly touching spheres. Very careful convergence tests have to
be carried out before producing any useful data. This is also partly
the reason why we are not able to build a much larger domain and
perform simulations of pseudorandom configurations with a large
number of particles.
3.2. Effect of volume fraction and thermal conductivity of fillers
The effective thermal conductivity values of composites are cal-
culated for different models, as shown in Fig. 3. It should be noted
that d is 0.001D for SC and FCC models, which gives 52.20% and
73.83% volume fraction, respectively. For the DD model, gap dis-
tance between large spheres is 0.001D, while the gap distance
between small sphere and large sphere may vary (so is the volume
fraction). The 79.05% volume fraction case corresponds to the case
in which the gap between a small sphere and a large sphere is
0.001D. In this calculation, we consider how the effective thermal
conductivity varies with the thermal conductivity ratio of filler to
matrix (kf =km ). It should be noted that for a given structure the
ratio of the effective thermal conductivity to the matrix thermal
conductivity (keff =km ) is only a function of kf =km [37]. From Fig. 3,
we can see that keff =km first significantly increases with the
increase of kf =km and then saturates at around 1000 (the relative
increment is within 2% with even larger kf =km ). It shows that it is
unnecessary to seek for fillers with extremely high thermal con-
ductivity, since the enhancement of the effective thermal conduc-
tivity is limited. On the other hand, if the thermal conductivity of
the low conductivity matrix can be enhanced, the effective thermal
conductivity will be enhanced proportionally. The different depen-
dence on the matrix and filler is due to the fact that the matrix is
thermally connected while the fillers are not [38]. If further com-
paring the thermal conductivity of three different structures with
different volume fractions, we find that the larger volume fraction
always gives higher effective thermal conductivity. In addition, we
also calculate the effective thermal conductivity of three DD
358 Z. Tong et al. / International Journal of Heat and Mass Transfer 100 (2016) 355–361

Fig. 2. (a) The ratio of effective thermal conductivity to matrix thermal conductivity (keff/km) varies with the ratio of the gap distance to the diameter of the sphere (d/D) for a
FCC structure with sphere to matrix thermal conductivity ratio of infinity using our method (square dot) and the analytical results (red curve [16,36]). The top view of FCC
structure is shown in the inset, where D is the diameter of the sphere, L is the lateral length of the cube, and d is the gap distance. (b) The ratio of effective thermal
conductivity to matrix thermal conductivity (keff/km) varies
pffiffi
with the volume fraction f for a FCC structure using our method (circular dot) and the analytical results (red curve
[16,36]). The vertical dashed line corresponds to f ¼ 62 p. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this
article.)

Fig. 3. In the left-bottom axis, these curves describe keff/km as a function of kf/km. In
the right-top axis, these curves describe the effective thermal conductivity as a
function of the thermal conductivity of filler at a given matrix (km = 0.2 W/m K).
models with different volume fractions by changing the size of the
small sphere. From Fig. 3, we find that the increment of the effec-
tive thermal conductivity of DD model with low volume fraction is
not obvious when compared to that of FCC model, but it is more
significant with higher volume fraction corresponding that the
smaller sphere is nearly in contact with the large sphere.
We further consider the widely used alumina–silicone compos-
ite, in which the thermal conductivity of alumina and the silicone
can be assumed to be 30 W/m K and 0.2 W/m K, respectively. With
the thermal conductivity ratio kf =km to be 150, the effective ther-
mal conductivity values of the alumina–silicone composites can
be determined from Fig. 3, which are 1.71 W/m K for SC composite
model, 3.65 W/m K for FCC composite model, and 4.36 W/m K for
DD composite model. Compared to FCC model, the relative incre-
ment of the volume fraction of the DD model is just 7%, but the
increment of the effective thermal conductivity is nearly 20%. It
is shown that the effective thermal conductivity of composites
with high filler volume fraction is very sensitive to the variation
of the filler loading. Even for a small amount of extra fillers, the
enhancement of thermal conductivity is significant.
In addition, we compare our results with the theoretical models
in the case of alumina–silicone composites, and the comparisons
are shown in Table 1. We can see that the result of Maxwell model
is nearly 50% less than that of present work and the Bruggeman
symmetric model gives nearly 5 times larger value than the simu-
lation results. Since these models are just derived at relatively low
volume fractions, it is easy to understand that they cannot work for
the high loading particle-filled composites.
We also compare our results with the percolation model devel-
oped by Devpura et al. [34]. Following their method we developed
a MATLAB code to implement the transfer matrix method. As this
method works only for cubic fillers, following the suggestion by
Devpura et al., the volume fraction was multiplied by the correc-
tion factor of 0.637 to produce the results for spherical particles.
The thermal conductivity values of the two components are also
30 W/m K and 0.2 W/m K, to model the alumina–silicone compos-
ites. The corresponding width and height of the system were con-
sidered to be 25, the length was to be 800. To reduce the statistical
uncertainty, for each volume fraction, the final value is achieved by
averaging the calculated value of 20 independent systems. Surpris-
ingly, the percolation model gives relatively smaller effective ther-
mal conductivity than our simulation results, but the difference is
within 10%. This actually validates their percolation model to pre-
dict the thermal conductivity of high filler loading composite
materials.
3.3. Effect of the contact area
In the previous discussions, the particles in these high filler
loading composite models are assumed to be perfectly spherical
and nearly tangential to each other. A valid question would be
what happens if the two neighboring spheres are exactly tangen-
tial to each other. However, such a problem is ill defined in the
framework of FEM because it has infinitely small feature size.
Moreover, in reality such a model may not even be necessary to
describe the contact of two particles because the surface of the par-
ticles more or less has certain roughness. Based on these consider-
ations, we replace the contact point for the tangential spheres with
a cylindrical ‘‘cushion”, as shown in Fig. 4(a). Note that while build-
ing these models the two spheres are kept perfectly tangential to
each other, and we just replace the region close to the tangential
point with a cylindrical cushion. Therefore, the half of the thickness
of the cushion t and its area DS is related by
DS ¼ ptðD  tÞ; ð2Þ
 Z. Tong et al. / International Journal of Heat and Mass Transfer 100 (2016) 355–361 359

Table 1
Comparisons between our simulation results and the theoretical models. Here we assume kf = 30 W/m K and km = 0.2 W/m K.

Model Expressions SC (W/m K) FCC (W/m K) DD (W/m K)

Vol (%) – 52.20 73.83 79.05
Maxwell keff kf þ2km þ2f ðkf km Þ 0.83 1.77 2.27
km
¼ kf þ2km f ðkf km Þ
Bruggeman symmetric k k kf keff 8.93 18.42 20.75
ð1  f Þ kmmþ2keff þ f kf þ2keff
¼0
eff

Percolation – 1.61 ± 0.15 3.32 ± 0.14 4.28 ± 0.13

FEM – 1.71 3.65 4.36

km is the thermal conductivity of matrix, kf is the thermal conductivity of filler, f is the volume fraction of fillers.

Fig. 4. (a) The cylindrical ‘cushion’ model. Here kc is the thermal conductivity of the cushion, t is half of the thickness of the cushion, and DS is the area of the cushion (contact
area). (b) The effective thermal conductivity varies with the ratio of the contact area to the surface area of the particle (S is the surface area of the particle).

where D is the diameter of the particle. Due to the complexity of
building such models, we only consider the SC arrangement of par-
ticles to understand the importance of the contact. It should be
noted that the thermal conductivity of the thin layer cannot be
determined directly because it is related to too many factors,
including the surface roughness, pressure, and the complicated
micro/nanoscale thermal transport effects [39]. Nevertheless, we
can estimate the upper and lower limit of the thermal conductivity
of the thin layer, which are the thermal conductivity of filler and
matrix, respectively. The simulation is performed assuming the fil-
ler and matrix are alumina and silicone, as mentioned above. The
effective thermal conductivity is then calculated with different con-
tact areas and the thermal conductivity values of the cushion are
chosen to be 0.2, 1.0, 3.0, and 30.0 W/m K. When the thermal con-
ductivity of the cushion is 0.2 W/m K, the particles are separated
by a layer of matrix material, corresponding to a poor contact. If
it is 30.0 W/m K, this is an ideal contact (particles are perfectly
connected).
The simulation results are shown in Fig. 4(b), in which the
lower horizontal axis is the ratio of the cushion area to the particle
surface area, and the upper horizontal axis is the ratio of the thick-
ness of the corresponding thin layer to the diameter of the particle.
From Fig. 4(b), we can see that the effect of the contact area
behaves completely differently for cushions of different thermal
conductivities. When the thermal conductivity of the cushion is
larger than 3.0 W/m K, the effective thermal conductivity of the
composite significantly increases with contact area. On the other
hand, if the cushion thermal conductivity is below 1.0 W/m K,
the effective thermal conductivity decreases with contact area.
The trend is depending on whether a conductive pathway can be
formed in the structure. It can also be noticed that when the cush-
ion thermal conductivity is larger than 3.0 W/m K, even a very
small contact area can significantly enhance the overall thermal
conductivity. Although it is previously known that forming con-
ductive pathway is important to make high conductivity composite
material, this simulation results show that the detail at the contact
between particles is also a key factor to obtain high effective ther-
mal conductivity of composites.
3.4. Effect of the interfacial thermal resistance
In the simulation results presented above, the effect of interface
thermal resistance has not been considered. The results should be
useful if the particles are relatively large so that such effect can be
neglected. Sometimes it is practically needed to add smaller parti-
cles into the TIMs. Therefore, it is also necessary to quantify the
effect of interface thermal resistance for high volume fraction
particle-filled composite. We employed a core–shell model as
shown in the inset of Fig. 5. The interfacial thermal resistance
was modeled as a very thin layer between fillers and matrix (i.e.,
d=D ¼ 0:01 [26] in the present work), and the value of the interfa-
cial thermal resistance was incorporated in the dimensionless
parameter Biot number
Rb km
Bi ¼ ; ð3Þ
D
where Rb is the interfacial thermal resistance, km is the thermal con-
ductivity of the matrix, D is the diameter of fillers. The critical diam-
eter d0 (=2Rbkm) is defined by setting the Biot number equal to 1
[40]. The critical diameter has clear physical meaning. If the particle
diameter is less than the critical diameter, the thermal conductivity
of the composite will be less than that of the matrix and it will be
greater than that of matrix otherwise. Due to the complexity of such
structures, we did not consider the DD model here. To ensure the
accuracy, the gap between the outer shells of neighboring spheres
is kept 0.001D. Due to the thin shell layer, the volume fractions of
360 Z. Tong et al. / International Journal of Heat and Mass Transfer 100 (2016) 355–361

sely proportional to the particle size, so keff/km decreases with the

Fig. 5. The ratio of the effective thermal conductivity to the matrix thermal
conductivity (keff/km) varies with Biot number, calculated by FEM (dots), Cheng et al.
(solid curve [21]), and Maxwell–Garnet model (dashed curve [20]). The inset is the
FCC core–shell model used to investigate the effect of interfacial thermal resistance.
A thin shell is used to model the interfacial thermal resistance between fillers and
matrix.
fillers in these models are slightly smaller than those in Section 3.2.
The values of the ratio of effective thermal conductivity to matrix
thermal conductivity (keff/km) are plotted in Fig. 5 as a function of
Biot number. In the same figure, for comparison purpose, we also
show the values calculated by using the formula derived by Cheng
et al. [21] (based on Rayleigh’s multipole expansion method) for
cubic arrays of spheres with interface thermal resistance. Based
on this model, the effective thermal conductivity is given by
decrease of the particle size. This is because when the particle size
is smaller, the interfaces become denser and the interface thermal
resistance becomes dominant. As such, for a fixed interfacial ther-
mal resistance, there is a critical diameter, denoted by d0 in Fig. 5.
The particles can increase the thermal conductivity of the compos-
ite when the diameter of the particle is larger than d0, and it will
decrease the thermal conductivity otherwise. We notice that the
results calculated by the formula provided by Cheng et al. agree
well with the FEM results. However, there exists a discrepancy
between Maxwell–Garnet and the FEM results (and Cheng’s for-
mula), especially for small Bi. We believe this is still mainly due
to the dipole approximation of the Maxwell–Garnet model and it
is typically valid for low volume fraction (less than 40% [4]). For
larger Bi, since the interface resistance becomes dominated, the
results from different models agree better with each other.
As for alumina–silicone composite, the value of the interfacial
thermal resistance has not been directly measured. We then take
the value to be (3.75 ± 1.25 ) 108 m2 K/W, based on the mea-
surement of Putnam et al. [42] for the interface thermal resistance
between alumina and PMMA. It is assumed here that the interfacial
thermal resistance between alumina and polymers (i.e., silicone
and PMMA) does not differ significantly. Then we can determine
that the critical diameter of alumina for alumina–silicone compos-
ite is 15 nm, and the diameter corresponding to the effective ther-
mal conductivity is nearly independent (when Bi is less than 103)
of the Biot number is 7.5 lm. Therefore, we can see that the effec-
tive thermal conductivity strongly depends on the particle diame-
ter when 15 nm < D < 7:5 lm, and it becomes lower than the
thermal conductivity of the matrix when D < 15 nm. Our simula-
tion result can be a good guidance to the choice of the size of the
alumina fillers to fabricate high thermal conductivity TIMs.

keff 1 þ 2f b1  2ð1  f Þf2 b21 4. Conclusions

¼ ;
km 1  f b1  2ð1  f Þf2 b21
where keff is the effective thermal conductivity, km is the thermal
conductivity of matrix, f is the volume fraction of fillers, f2 is the
geometric parameter which has been tabulated as a function of
the volume fraction by McPhedran et al. [41]. By definition the
dimensionless parameter b1 ¼ kþ2þ2R k1R
where k ¼ kf =km and
R ¼ 2Rb kf =D ¼ 2kBi. In addition, we also provide the values calcu-
lated by using the Maxwell–Garnet model [20] with the effective
thermal conductivity given by,
keff ½kf ð1 þ 2aÞ þ 2km  þ 2f ½kf ð1  aÞ  km 
¼ ;
km ½kf ð1 þ 2aÞ þ 2km   f ½kf ð1  aÞ  km 
where km is the thermal conductivity of matrix, kf is the thermal
conductivity of filler, and f is the volume fraction of fillers. By defi-
nition the dimensionless parameter a ¼ 2Rb km =D ¼ 2Bi.
In our simulations, the thermal resistance of the shell
Rs ¼ Rb þ Rm ¼ d=ks , where Rm is the matrix intrinsic thermal resis-
tance corresponding to the thickness of d, and ks is the thermal
conductivity of the shell. The effect of the interface thermal resis-
tance on the overall thermal conductivity is shown in Fig. 5, from
which we can see that the interfacial thermal resistance has a sig-
nificant effect for high filler loading composites. Furthermore, one
can see that the variation of keff/km is within 2.5% when Bi is less
than 103 but significantly decreases in the range of 103 to 0.5,
and then decreases to less than 1 when Bi is less than 0.5. This phe-
nomenon can be understood from the following two aspects. For
fixed particle size, the Biot number is proportional to the interfacial
thermal resistance. Then keff/km decreases with the increase of
interfacial thermal resistance, as it induces additional resistance.
For a fixed interfacial thermal resistance, the Biot number is inver-
ð4Þ
To summarize, FEM simulations on the heat conduction of
close-packed particulate composite materials are performed. The
effects of thermal conductivity of fillers, volume fraction, the prox-
imity effect of adjacent particles, and interfacial thermal resistance,
on the overall thermal conductivity of the composite, were inves-
tigated. A few conclusions can be drawn. First, the effective ther-
mal conductivity of the composite first increases with the
thermal conductivity of fillers and then saturates when the filler
thermal conductivity is 1000 times larger than the matrix. Second,
the effective thermal conductivity of composites with high filler
volume fraction is sensitive to the variation of the filler loading
ð5Þ
especially at the nearly contact state. Even for small enhancement
of filler loading, the enhancement of thermal conductivity is signif-
icant. Third, details at the contact region between adjacent parti-
cles are crucial to the effective thermal conductivity of
composites. More effective thermal contacts between particles
can help to build the conductive pathway and significantly
enhance the overall thermal conductivity. Fourth, the interfacial
thermal resistance and particle size is also an important factor
for Biot number larger than 0.001. In the case of alumina–silicone
composites, interface thermal resistance for particles larger than a
few microns is generally negligible. The particles with diameter
smaller than 15 nm will have negative contributions to the thermal
conductivity.
Acknowledgement
The authors gratefully acknowledge the support from HUAWEI
TECHNOLOGIES CO., LTD. Fruitful discussions with Yan Xu and
Wenbin Liu are gratefully appreciated.
 Z. Tong et al. / International Journal of Heat and Mass Transfer 100 (2016) 355–361 361

References [22] P. Cheng, C.-T. Hsu, The effective stagnant thermal conductivity of porous
media with periodic structures, J. Porous Media 2 (1) (1999) 2.
[23] J. Ma, C.W. Nan, Effective-medium approach to thermal conductivity of
[1] E. Pop, Energy dissipation and transport in nanoscale devices, Nano Res. 3 (3)
heterogeneous materials, Ann. Rev. Heat Transfer 17 (2014) 303–331.
(2010) 147–169.
[24] K. Ramani, A. Vaidyanathan, Finite element analysis of effective thermal
[2] D.D.L. Chung, Thermal interface materials, J. Mater. Eng. Perform. 10 (1) (2001)
conductivity of filled polymeric composites, J. Compos. Mater. 29 (13) (1995)
56–59.
1725–1740.
[3] R. Prasher, Thermal interface materials: historical perspective, status, and
[25] I.A. Tsekmes, R. Kochetov, P.H.F. Morshuis, J.J. Smit, Modeling the thermal
future directions, Proc. IEEE 94 (8) (2006) 1571–1586.
conductivity of polymeric composites based on experimental observations,
[4] R.J. Linderman, T. Brunschwiler, U. Kloter, H. Toy, B. Michel, Hierarchical nested
IEEE Trans. Dielectr. Electr. Insul. 21 (2) (2014) 412–423.
surface channels for reduced particle stacking and low-resistance thermal
[26] J. Zeng, R. Fu, S. Agathopoulos, S. Zhang, X. Song, H. He, Numerical simulation
interfaces, in: Twenty Third Annual IEEE Semiconductor Thermal
of thermal conductivity of particle filled epoxy composites, J. Electron. Packag.
Measurement and Management Symposium, SEMI-THERM 2007, 2007, pp.
131 (4) (2009) 041006.
87–94.
[27] S.-T. Tu, W.-Z. Cai, Y. Yin, X. Ling, Numerical simulation of saturation behavior
[5] W. Yu, H. Xie, L. Yin, J. Zhao, L. Xia, L. Chen, Exceptionally high thermal
of physical properties in composites with randomly distributed second-phase,
conductivity of thermal grease: synergistic effects of graphene and alumina,
J. Compos. Mater. 39 (7) (2005) 617–631.
Int. J. Therm. Sci. 91 (2015) 76–82.
[28] L.C. Davis, B.E. Artz, Thermal conductivity of metal-matrix composites, J. Appl.
[6] B.Z. Gao, J.Z. Xu, J.J. Peng, F.Y. Kang, H.D. Du, J. Li, S.W. Chiang, C.J. Xu, N. Hu, X.S.
Phys. 77 (10) (1995) 4954–4960.
Ning, Experimental and theoretical studies of effective thermal conductivity of
[29] R.H. Khiabani, Y. Joshi, C.K. Aidun, Thermal properties of particulate TIMs in
composites made of silicone rubber and Al2O3 particles, Thermochim. Acta 614
squeeze flow, Int. J. Heat Mass Transfer 53 (19) (2010) 4039–4046.
(2015) 1–8.
[30] M. Wang, N. Pan, J. Wang, S. Chen, Mesoscopic simulations of phase
[7] Thermal Interface Materials from Fujipoly America|SARCON Thermal
distribution effects on the effective thermal conductivity of microgranular
Management Components, 22-Jan-2016. [Online]. Available: <http://
porous media, J. Colloid Interface Sci. 311 (2) (2007) 562–570.
www.fujipoly.com/usa/products/sarcon-thermal-management-components/>.
[31] Y. Agari, A. Ueda, M. Tanaka, S. Nagai, Thermal conductivity of a polymer filled
[Accessed: 22-Jan-2016].
with particles in the wide range from low to super-high volume content, J.
[8] X. Huang, P. Jiang, T. Tanaka, A review of dielectric polymer composites with
Appl. Polym. Sci. 40 (5–6) (1990) 929–941.
high thermal conductivity, Electr. Insul. Mag. IEEE 27 (4) (2011) 8–16.
[32] D. Ganapathy, K. Singh, P.E. Phelan, R. Prasher, An effective unit cell approach
[9] J.C. Maxwell, A Treatise on Electricity and Magnetism, third ed., Dover, New
to compute the thermal conductivity of composites with cylindrical particles,
York, 1954.
J. Heat Transfer 127 (6) (2005) 553–559.
[10] H.T. Davis, L.R. Valencourt, C.E. Johnson, Transport processes in composite
[33] C. Yue, Y. Zhang, Z. Hu, J. Liu, Z. Cheng, Modeling of the effective thermal
media, J. Am. Ceram. Soc. 58 (9–10) (1975) 446–452.
conductivity of composite materials with FEM based on resistor networks
[11] K. Pietrak, T.S. Wiśniewski, A review of models for effective thermal
approach, Microsyst. Technol. 16 (4) (2010) 633–639.
conductivity of composite materials, J. Power Technol. 95 (1) (2014) 14–24.
[34] A. Devpura, P.E. Phelan, R.S. Prasher, Percolation theory applied to the analysis
[12] L. Rayleigh, LVI. On the influence of obstacles arranged in rectangular order
of thermal interface materials in flip-chip technology, in: The Seventh
upon the properties of a medium, Lond. Edinb. Dublin Philos. Mag. J. Sci. 34
Intersociety Conference on Thermal and Thermomechanical Phenomena in
(211) (1892) 481–502.
Electronic Systems, ITHERM 2000, vol. 1, 2000.
[13] H. Levine, The effective conductivity of a regular composite medium, IMA J.
[35] B. Dan, B.G. Sammakia, G. Subbarayan, S. Kanuparthi, S. Mallampati, The study
Appl. Math. 2 (1) (1966) 12–28.
of the polydispersivity effect on the thermal conductivity of particulate
[14] R.C. McPhedran, D.R. McKenzie, The conductivity of lattices of spheres. I. The
thermal interface materials by finite element method, IEEE Trans. Compon.
simple cubic lattice, Proc. R. Soc. Lond. A: Math. Phys. Eng. Sci. 359 (1978) 45–
Packag. Manuf. Technol. 3 (12) (2013) 2068–2074.
63.
[36] R.T. Bonnecaze, J.F. Brady, A method for determining the effective conductivity
[15] D.R. McKenzie, R.C. McPhedran, G.H. Derrick, The conductivity of lattices of
of dispersions of particles, Proc. R. Soc. Lond. A: Math. Phys. Eng. Sci. 430
spheres. II. The body centred and face centred cubic lattices, Proc. R. Soc. Lond.
(1990) 285–313.
A: Math. Phys. Eng. Sci. 362 (1978) 211–232.
[37] S. Durmaz, A numerical study on the effective thermal conductivity of
[16] A.S. Sangani, A. Acrivos, The effective conductivity of a periodic array of
composite material (Ph.D. thesis), Graduate School of Natural and Applied
spheres, Proc. R. Soc. Lond. A: Math. Phys. Eng. Sci. 386 (1983) 263–275.
Sciences of Dokuz Eylul University, 2004.
[17] A.A. Zick, Heat conduction through periodic arrays of spheres, Int. J. Heat Mass
[38] A. Bjorneklett, L. Haukeland, J. Wigren, H. Kristiansen, Effective medium theory
Transfer 26 (3) (1983) 465–469.
and the thermal conductivity of plasma-sprayed ceramic coatings, J. Mater. Sci.
[18] J.B. Keller, Conductivity of a medium containing a dense array of perfectly
29 (15) (1994) 4043–4050.
conducting spheres or cylinders or nonconducting cylinders, J. Appl. Phys. 34
[39] C. Shao, H. Bao, A molecular dynamics investigation of heat transfer across a
(4) (1963) 991–993.
disordered thin film, Int. J. Heat Mass Transfer 85 (2015) 33–40.
[19] D.P.H. Hasselman, L.F. Johnson, Effective thermal conductivity of composites
[40] A. Devpura, Size effects on the thermal conductivity of polymers laden with
with interfacial thermal barrier resistance, J. Compos. Mater. 21 (6) (1987)
highly conductive filler particles, Microscale Thermophys. Eng. 5 (3) (2001)
508–515.
177–189.
[20] A.G. Every, Y. Tzou, D.P.H. Hasselman, R. Raj, The effect of particle size on the
[41] R.C. McPhedran, G.W. Milton, Bounds and exact theories for the transport
thermal conductivity of ZnS/diamond composites, Acta Metall. Mater. 40 (1)
properties of inhomogeneous media, Appl. Phys. A 26 (4) (1981) 207–220.
(1992) 123–129.
[42] S.A. Putnam, D.G. Cahill, B.J. Ash, L.S. Schadler, High-precision thermal
[21] H. Cheng, S. Torquato, Effective conductivity of periodic arrays of spheres with
conductivity measurements as a probe of polymer/nanoparticle interfaces, J.
interfacial resistance, Proc. R. Soc. Lond. A: Math. Phys. Eng. Sci. 453 (1997)
Appl. Phys. 94 (10) (2003) 6785–6788.
145–161.