PHYSICAL REVIEW B 72, 125418 共2005兲

Thermal conductivity of simple and tubular nanowire composites in the longitudinal direction

Ronggui Yang* and Gang Chen

Mechanical Engineering Department, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA

Mildred S. Dresselhaus
Department of Physics and Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology,
Cambridge, Massachusetts 02139, USA
共Received 29 March 2005; published 13 September 2005兲

This work establishes a generic model to study phonon transport and the thermal conductivity of periodic
two-dimensional nanocomposites in the longitudinal direction 共along the wire axis direction兲. More specifi-
cally, the generic model is applied to study the thermal conductivity of silicon-germanium composites with
simple silicon nanowire and tubular silicon nanowire inclusions in a germanium matrix, and cylindrical nano-
porous silicon materials. The results show that the effective thermal conductivity changes not only with the
volumetric fraction of the constituents but also with the radius of the nanowires and cylindrical pores due to the
nature of the ballistic phonon transport. The smaller the wire/pore diameter, the smaller is the thermal con-
ductivity of the periodic two-dimensional nanocomposites for a given volumetric fraction. Composites with
tubular nanowire inclusions have a lower effective thermal conductivity than simple nanowire composites due
to the introduction of additional surface scattering through the pores associated with tubular nanowires. Results
of this study can be used to direct the development of both high-efficiency thermoelectric materials and thermal
interface material containing high-thermal-conductivity particle or wire inclusions.

DOI: 10.1103/PhysRevB.72.125418 PACS number共s兲: 68.65.⫺k, 44.10.⫹i, 63.22.⫹m

I. INTRODUCTION dispersion in periodic structures16 Due to the short wave-

Nanocomposite and nanoporous materials are used in a length of the dominant phonon heat carriers, the phonon
variety of applications and will play more important roles in scattering at interfaces is often diffuse.17 The diffuse inter-
the future with rapid progress being made in making control- face scattering inside the nanostructure materials cannot only
lable nanostructures, such as nanowires, nanotubes, and reduce the phonon mean free path but can also destroy the
nanoparticles.1 Thermal properties of nanocomposites are coherence of phonons. Due to the loss of coherence, classical
important for applications of particle-laden polymers in mi- size effect models such as the phonon Boltzmann equation
croelectronics as thermal interface materials2,3 and nano- are expected to be applicable to a wide range of nanostruc-
porous materials for sensor development.4,5 Our interests in tures. In our previous article, the phonon Boltzmann equa-
the thermal conductivity of nanocomposites arise from their tion is applied to study the classical size effect on the thermal
potential applications in high efficiency thermoelectric conductivity of periodic nanowire composites 共two-
materials.6,7 The efficiency and energy density of thermo- dimensional composites兲 normal to the wire axis direction.7
electric devices are determined by the dimensionless thermo- In this article, we study the thermal conductivity of periodic
electric figure of merit of the thermoelectric materials ZT two-dimensional nanocomposites with simple nanowire and
= S2␴T / k, where S is the Seebeck coefficient, ␴ is the elec- tubular nanowire inclusions, and cylindrical nanoporous ma-
trical conductivity, k is the thermal conductivity, and T is the terials in the longitudinal direction, i.e., along the wire axis
absolute temperature.8 Significant advances for increasing direction.
ZT have been made, based on new ideas on how to engineer
electron and phonon transport.9 Nanocomposites may II. THEORETICAL MODEL AND NUMERICAL
achieve high ZT through a thermal conductivity reduction SIMULATION
while maintaining essentially the same electron transport
performance.7 In Fig. 1共a兲, a temperature gradient is applied along the
There are not many theoretical studies on the thermal con- axial direction 共z direction兲 on a periodic two-dimensional
ductivity of nanocomposites despite its importance for prac- nanocomposite with tubular nanowire inclusions. Since the
tical applications. The prevailing approach is to include the transport is periodic in both x and y directions, the transport
interface thermal resistance, or Kapitza resistance,10 with the inside the nanocomposites can be represented by that in the
Fourier heat conduction theory.11–15 However, Fourier heat small unit cell shown in Fig. 1共b兲 applying the totally specu-
conduction theory is based on the diffusion picture and is not lar phonon reflection boundary conditions along the x and y
applicable when the phonon mean free path is longer than boundaries. However simulation of phonon transport in a
the characteristic length of the nanocomposites such as the 3-D unit cell shown in Fig. 1共b兲 is still a big challenge that
particle diameter and/or interparticle separation distance. An- involves both Cartesian and cylindrical coordinates. As an
other approach in the investigation of the nanocomposite approximation, we further convert the outer surface of the
thermal conductivity is through the calculation of the phonon square unit cell into a circle, i.e., to approximate a square

1098-0121/2005/72共12兲/125418共7兲/$23.00 125418-1 ©2005 The American Physical Society

YANG, CHEN, AND DRESSELHAUS PHYSICAL REVIEW B 72, 125418 共2005兲

FIG. 1. 共Color online兲 共a兲. Periodic two-

dimensional nanocomposite 共composite with tu-
bular nanowire inclusions兲; 共b兲 cross-sectional
view of a unit cell: a square unit cell cross section
is approximated as a circular cross section; 共c兲 by
the approximation in 共b兲, the transport in nano-
composites becomes phonon transport in core-
shell cylindrical structures; 共d兲 periodic silicon
nanowire composites; and 共e兲 cylindrical nano-
porous silicon material.

unit cell cross section as a circular cross section, as is often Carlo simulation shows that the thermal conductivity of a
used to study fluid flow problems.18,19 The problem then be- random nanocomposite is very close to a periodic
comes phonon transport in core-shell cylindrical structures, composite.20 In addition, the phonon Boltzmann transport
as shown in Fig. 1共c兲. This approximation results in a 10% model is based on the following assumptions: 共1兲 the phonon
error. Figure 1共c兲 shows the generic phonon transport model wave effect can be excluded; 共2兲 the frequency-dependent
we developed for core-shell nanostructures, which consists scattering rate in the bulk medium can be approximated by
of a tubular core layer and a shell layer. Here we use the using an average phonon mean free path 共MFP兲 ⌳. Assump-
following notations: r0 is the inner radius of the core layer, r1 tion 共1兲 can be well justified since the wavelength for domi-
is the outer radius of the core layer, and r2 is the outer radius nant phonons responsible for thermal transport is around
of the shell. We assume partially diffuse and partially specu- 1 nm. The justification for these assumptions can be found in
lar surface scattering at the inner surface of the core layer Refs. 7 and 21. In our simulation, we used the phonon MFP
and totally specular reflection at the outer surface of the shell and the group velocity v listed in Ref. 21, which are obtained
layer, to be consistent with what occurs at the outer surface by approximating the dispersion of the transverse and the
of the unit cell shown in Fig. 1共b兲, i.e., to represent the pho- longitudinal-acoustic phonons with simple sine functions and
non transport in nanocomposites. The specularity parameter neglecting the optical phonon contribution to the thermal
at the inner surface of the core layer and the outer surface of conductivity k. This estimation leads to longer mean free
the shell layer is represented by p1 and p2, respectively, path than using simple kinetic theory expression k = 31 Cv⌳,
where p 共=p1 or p2兲 = 0 corresponds to diffuse scattering and which is consistent with experiments from Goodson’s
p = 1 corresponds to specular scattering at the surface. This group.22
generic model can be used to simulate a variety of nanocom- In terms of the total phonon intensity I,23,24 the 2-D pho-
posites by changing some of the input parameters. For ex- non Boltzmann equation under the single mode relaxation
ample, when the inner radius of the tubular core layer r0 = 0 time approximation in cylindrical coordinates can be written
and the interface specularity p1 = 1 at r0 = 0, the structure rep- as
resents periodic two-dimensional composites with simple ␮⳵ 1 ⳵ ⳵Ii Ii − Ioi
wire inclusions as shown in Fig. 1共d兲. When the same mate- 共rIi兲 − 共 ␩ I i兲 + ␨ = − , 共1兲
r ⳵r r ⳵␾ ⳵z ⌳i
rial is used for the core and the shell layers, the interface
between the core and shell layers disappears and the inter- where subscript i 共=1 , 2兲 denotes properties of the core and
face scattering at r1 dies out; the model represents the pho- shell material ⌳i is the average phonon MFP. ␮, ␩, and ␨ are
non transport inside a nanoporous medium with cylindrical the directional cosines defined as
pores along the pore direction as shown in Fig. 1共e兲.
␮ = sin ␪ cos ␾, ␩ = sin ␪ sin ␾, ␨ = cos ␪ , 共2兲
Clearly the model developed here is strictly valid for pe-
riodic composites only. In reality most of composites made where ␪ and ␾ are the polar and azimuthal angles, respec-
might have random microstructures. Our study using Monte tively, as shown in Fig. 2. In the phonon Boltzmann transport

125418-2
THERMAL CONDUCTIVITY OF SIMPLE AND TUBULAR… PHYSICAL REVIEW B 72, 125418 共2005兲

into 24 points for 0 to ␲ 共not 0 to 2␲ due to symmetry兲. Then

Eq. 共3兲 can be written in discrete form as
2
I0i共r,z兲 = 兺 兺 Ii共r,z, ␨n, ␾m兲wnw⬘m .
4␲ m n
共4兲

The weights wn and wm ⬘ satisfy 兺m兺nwnwm⬘ = 2␲.

Following the conventional artifice of Carlson and
Lathrop25 and Lewis and Miller,29 which maintain phonon
radiative energy conservation and permits minimal direc-
tional coupling, the angular derivative term can be written as
follows:
FIG. 2. Phonon transport in cylindrical coordinate.
⳵ nm nm ␣n,m+1/2In,m+1/2 − ␣n,m−1/2In,m−1/2
共␩ Ii 兲 = i i
, 共5兲
equation 关Eq. 共1兲兴, there are two coordinate systems 关shown ⳵␾ ⬘
w nw m
in Fig. 2兴: spatial coordinates 共r, ␾c, and z兲 and directional where wnw⬘m is a weight and ␣n,m±1/2 is the coefficient for
coordinates 共␪ and ␾兲 for the transport, which correspond to the angular derivative term determined from the nondiver-
the movement of carriers in spatial and momentum space. As gent flow condition as following the recursive equation:
a phonon travels through a curved geometry such as in the
cylindrical coordinates, the propagating direction is con- ␣n,m+1/2 − ␣n,m−1/2 = 共wnwm⬘ 兲␮n,m . 共6兲
stantly varying, even though the phonon does not physically
change its direction. This is why an additional term, the sec- For this work, we are only interested in the size effect
ond term of the left-hand side in Eq. 共1兲, exists. The diffi- occurring in the radial direction, not in the wire axis direc-
culty of numerical simulation is also due to this term. There tion. However, because heat flows along the axial direction,
exist many procedures to address this term in the neutron the simulation requires choosing a proper length and the cor-
transport literature.25 Ioi in Eq. 共1兲 is determined by the Bose- responding boundary conditions at the two ends along the
Einstein distribution of phonons and depends on the local transport direction. When heat is enforced to flow in the axial
equilibrium temperature. In nanostructures, local equilibrium direction, the temperature is different in the two ends. How-
cannot be established and thus the temperature obtained ever, simply assuming the temperature difference in the two
should not be treated in the same way as in the case of local ends might induce artificial size effect in the axial direction.
thermal equilibrium. An energy balance shows that Ioi can be To get rid of the artificial size effect in the axial direction, the
calculated from26 periodic boundary condition on the phonon distribution de-
viation proposed in our earlier work7 is used. This periodic

冕
boundary condition requires that the deviation of phonon in-
1 tensity in each direction at each point in one boundary is the
Ioi共r,z兲 = Ii共r,⍀兲d⍀
4␲ 4␲
same as the deviation in the corresponding point and direc-

冕 冕
tion in the other boundary of the simulation domain in the
2␲ ␲
1 heat transport direction. The equation can be expressed as
= Ii共r,z, ␪, ␾兲sin ␪d␪d␾ 共3兲
4␲ 0 0
I共r,LZ, ␪, ␾兲 − I0共r,LZ兲 = I共r,0, ␪, ␾兲 − I0共r,0兲, 共7兲
where LZ is the simulation domain length. It can be proved
and the corresponding temperature obtained is a measure of
that the temperature difference between the two ends should
the local energy density.
be independent of r and can be related to
Equation 共1兲 is similar to the photon radiative transport
equation 共RTE兲.26 The key is to solve for the intensity distri- Civi关T共r,LZ兲 − T共r,0兲兴
bution I共r , ⍀兲. A variety of solution methods are available in Io共r,LZ兲 − Io共r,0兲 = = const. 共8兲
4␲
the thermal radiative transfer literature.25,27 For phonon
transport in nanostructures, the challenge is to reduce the In our simulation, we superimposed T共r , LZ兲 − T共r , 0兲 = 1 K in
“ray effect,” which often happens similarly in thermal radia- the above equation. If we do not superimpose such a
tion in optical thin limit. In our previous work,28 double temperature difference in the program, Eq. 共7兲 will automati-
Gauss-Legender quadratures have been used to replace the cally converge to a constant temperature difference
conventional SN quadratures for the discrete ordinate method T共r , LZ兲 − T共r , 0兲 value. The converged value varies with
and this approach was shown to successfully resolve the ray simulated structures. But the final results of the thermal con-
effect problem in our phonon transport simulation in nano- ductivity value do not depend on whether the temperature
structures. In the present work, we extend the previous work difference is superimposed. However, the calculation is
in Cartesian coordinates to cylindrical coordinates. The much faster when the temperature difference value is super-
method separately discretizes the integrating points in ␨ imposed. Other than the temperature difference shown in Eq.
= cos ␪ 共the angle ␪兲 and in the angle ␾ using the Gauss- 共8兲, Eq. 共7兲 is apparently a more demanding boundary con-
Legender quadrature. To obtain high accuracy, ␮ is dis- dition, which ensures directional heat flux conservation in
cretized into 120 points from −1 to 1 and ␾ is discretized the two boundaries along the transport direction. For trans-

125418-3
YANG, CHEN, AND DRESSELHAUS PHYSICAL REVIEW B 72, 125418 共2005兲

port along the axial direction, the periodicity is arbitrary. So

we can choose an arbitrary length LZ and use the periodic
boundary condition to obtain thermal conductivity values
that are independent of LZ. If the boundary conditions at the
two ends are assumed as emitted temperature condition,28
one often needs to simulate a domain more than three times
longer than the phonon mean free path, which can be many
times longer than the size in the radial direction. With the
proposed boundary condition, the simulation domain length
can be adjusted according to the size of the structure in the
radial direction. We also note that the nature of the Boltz-
mann equation requires iterations to obtain convergent re-
sults with the boundary conditions we have defined.
As in previous work, a nonuniform grid system and the
step scheme is used for spatial discretization to accurately FIG. 3. Thermal conductivity of silicon germanium nanocom-
capture the physics of the transport phenomena and to mini- posite which is comprised of a germanium matrix with silicon wire
mize the calculation time. The step scheme is used for spatial inclusions as a function of the silicon wire radius and the volumet-
discretization. The equation is solved by iteration over the ric ratio.
value of the equivalent equilibrium intensity I0i共r , z兲
⬘ . Assuming constant spe-
= 共2 / 4␲兲兺m兺nIi共r , z , ␨n , ␾m兲wnwm
U2共T兲v2
cific heat over a wide temperature range, we can write effec- Td12共T兲 = , 共13兲
tive temperature, which is a measure of the local energy U1共T兲v1 + U2共T兲v2
density inside the medium, as
where U is the volumetric internal energy. In a bulk Si- Ge
4␲I0共r,z兲 composite, the effective thermal conductivity increases lin-
T共r,z兲 = . 共9兲
C i兩 v i兩 early as the cross-sectional area or volumetric ratio of Si
共⌽Si兲 increases since Si has much higher thermal conductiv-
The heat flux along the axial direction at every point can be ity than Ge. For Si- Ge nanocomposites, the effective thermal
accordingly written as conductivity decreases as the radius of wire inclusion de-
creases due to the relative increasing interface scattering area
qz共r,z兲 = 兺 兺 I共x,y, ␨n, ␾m兲␨nwnwm
⬘, 共10兲 per unit volume. When the radius of the wire inclusion is
m n
larger than 500 nm, the effective thermal conductivity ap-
The surface heat flux in the axial direction can be calculated proaches asymptotically that of macroscale composites. This
as means that the interface scattering is negligible compared to
the internal thermal resistance in Si wires and the Ge matrix

Qz共z兲 = 冕 r2

0
qz共r,z兲2␲rdr. 共11兲
and the effective value can be predicted by the Fourier law.
For nanowire composites with the wire inclusion radius less
than 150 nm, there exists a minimum thermal conductivity as
When the simulation is converged, Qz共z兲 = const. After local the volumetric ratio of Si changes. The trend is similar as
effective temperature distribution and heat flux are obtained, those predicted for core-shell Si- Ge nanowires. When the
the thermal conductivity calculation is straightforward. The volumetric ratio of Si 共⌽Si兲 in the composites is large, the
effective thermal conductivity k of the core-shell structure effective thermal conductivity has the same trend as that of
can be obtained as the macroscale composite. When the volumetric ratio of Si
共⌽Si兲 in the composites is small, the effective thermal con-
Q zL Z ductivity increases as the fraction of the low thermal conduc-
k= . 共12兲 tivity component increases, which is contrary to the behavior
␲r2关T共r,LZ兲 −
2
T共r,0兲兴
of bulk composites. This is because the effective thermal
conductivity of the Si wire keSi is decreased to well below the
III. RESULTS AND DISCUSSIONS bulk Ge thermal conductivity due to interface scattering. The
effective thermal conductivity of the Ge matrix keGe in-
Figure 3 shows results for the thermal conductivity of a creases as 共r2 − r1兲 / r1 increases, since the scattering surface
silicon-germanium nanocomposite which consists of a ger- per unit volume of Ge decreases. The effective thermal con-
manium matrix with silicon wire inclusions 关as shown in Fig. ductivity of the core-shell structure can be written as
1共d兲兴 as a function of the silicon wire radius and the volu- 关keSi
* *
r12 + keGe 共r22 − r12兲兴 / r22, thus accounting for the existence
metric ratio of Si, ⌽Si = r12 / r22, which can be related to the of a minimum value.
atomic ratio ␥Si through ␥Si = ⌽Si / 关⌽Si + 共1 − ⌽Si兲asi / aGe兴, As stated in the Introduction, most past studies on the
where a is the lattice constant. The phonon scattering at the thermal conductivity of nanocomposites were based on the
silicon-germanium interface is assumed to be diffuse and the Fourier diffusion theory together with consideration of the
phonon transmissivity at the interface can be calculated as30 thermal boundary resistance. To examine the validity of such

125418-4
THERMAL CONDUCTIVITY OF SIMPLE AND TUBULAR… PHYSICAL REVIEW B 72, 125418 共2005兲

an approach, we compare the effective thermal conductivity

obtained from the Boltzmann equation with that of the effec-
tive medium approach 共EMA兲 developed by Nan et al.,
which gives the anisotropic effective thermal conductivity
values of nanowire composites as13
k P共1 + ␣兲 + km + ⌽ P关k P共1 − ␣兲 − km兴
k11 = k22 = km ,
k P共1 + ␣兲 + km − ⌽ P关k P共1 − ␣兲 − km兴
共14兲

k33 = 共1 − ⌽ P兲km + ⌽ Pk P , 共15兲

where k11 and k22 are the effective composite thermal con-
ductivity across the wire axis direction, k33 is the effective
thermal conductivity in the longitudinal direction, km is the
thermal conductivity of the host material, k P is the bulk ther-
mal conductivity of the nanowire inclusion materials, ⌽ P is FIG. 4. Comparison of the thermal conductivity of nanowire
the volume fraction of nanowire inclusion, and ␣ is a dimen- composites in the direction perpendicular to the wire axial direction
sionless parameter defined as a = ak / a p for nanowire compos- obtained from phonon Boltzmann equation simulation and from the
ites. a p is the radius of nanowire inclusions and ak = Rkm, effective medium approximation 共EMA兲 based on the Fourier law
where R is the interface thermal resistance which can be and thermal boundary resistance, demonstrating that the EMA un-
calculated as21 derpredicts size effects.

4 4共U1v1 + U2v2兲 pore radius decreases since the scattering area per unit vol-
R= ⬇ . 共16兲
Td12U1v1 U 1v 1U 2v 2 ume increases. The implication of this study is that nanop-
ores can possibly be used to further reduce the thermal con-
Apparently, Eq. 共15兲 shows that the EMA model does not ductivity of nanowire composites. In a recent report, Zhao et
consider the size effect on the thermal conductivity for nano- al. showed that the effective thermal conductivity is further
wire composites in longitudinal direction, i.e., when the tem- reduced and thus the thermoelectric figure of merit ZT is
perature gradient is applied along the wire axis direction. increased in a Bi2Te3 composite with tubular Bi2Te3 nano-
Such a result is clearly contrary to the solution of the Bolt- wire inclusions.31
zmann equation presented above that shows a strong size The model has also been used to study the thermal con-
dependence. We further compare k11, the thermal conductiv- ductivity of tubular nanowire composites as shown in Fig.
ity perpendicular to the wire axis, with our previous 1共a兲. Again the results shown here assume a totally diffuse
calculations.7 As our previous calculation was done for surface scattering at the pore surface. The effective thermal
square wire inclusions, an effective diameter D = 2a p conductivity of tubular nanowire composites is a function
= 4AC / P is used, where AC is the cross-sectional area and P of the pore radius inside the tubular silicon wire r0, the
is the perimeter of the cross section, to convert the square outer radius of the silicon shell r1 共silicon shell thickness is
geometry into a circular geometry. Figure 4 compares the defined as r1 − r0兲, and the volumetric ratio of silicon
thermal conductivity in nanowire composites perpendicular ⌽Si = 共r12 − r02兲 / 共r22 − r02兲 in the composites. Figure 6 shows
to wire axis obtained from the phonon Boltzmann equation the effect of the silicon core layer thickness of the tubular
simulation and the effective medium approximation 共EMA兲.
As we can see, the effective medium approach based on
incorporating the thermal boundary resistance into the solu-
tions of the Fourier heat conduction law leads to erroneous
results that underpredict the size effects.
The same code can be used to simulate the thermal con-
ductivity of a nanoporous medium. Figure 5 shows results
for the thermal conductivity of porous silicon along the cy-
lindrical pore direction 关as shown in Fig. 1共e兲兴. Here r0 is the
pore radius and the shell thickness 共r2 − r0兲 is determined by
the porosity as ⌽o = r02 / r22. For simplicity, only the results
assuming diffuse surface scattering at the pore surface are
reported. For macroscale porous materials, the effective ther-
mal conductivity decreases linearly as the porosity increases.
When the pore radius is less than two to three times the
phonon mean free path in the silicon matrix, the effective
thermal conductivity is not only a function of porosity, but FIG. 5. 共Color online兲 Thermal conductivity of porous silicon
also a function of pore radius. For composites with the same along the cylindrical pore direction as a function of the pore radius
porosity, the effective thermal conductivity decreases as the and porosity.

125418-5
YANG, CHEN, AND DRESSELHAUS PHYSICAL REVIEW B 72, 125418 共2005兲

FIG. 7. The solid thermal conductivity of the composites de-

creases as the pore radius increases due to the increasing surface
scattering per unit volume.

As = ␲共r22 − r02兲. For comparison, the Si wire radius 共or outer

shell radius of the tubular nanowire兲 is fixed. Figure 7 also
shows that the solid thermal conductivity decreases as the
pore radius increases due to the increasing surface scattering
per unit volume.

IV. CONCLUSION

We study in this paper the thermal conductivity of peri-

odic two-dimensional nanocomposites along the wire axis
direction, including simple nanowire composites, nano-
porous medium, and composites with tubular nanowire in-
clusions. The results show that the effective thermal conduc-
FIG. 6. The effect of the silicon core layer thickness and the tivity changes not only with the volumetric fraction of the
pore size of tubular silicon wire inclusions on the effective thermal constituents but also the radius of the nanowire, pore, and
conductivity of the nanocomposites. tubular nanowire inclusions due to the nature of the ballistic
phonon transport. These results are in contradiction with the
silicon wire inclusions on the effective thermal conductivity existing theory on the thermal conductivity of composites,
of the nanocomposites. From Fig. 6, the effective thermal which lead to effective transport properties depending only
conductivity is smaller when the Si core layer thickness is on the volume fraction but independent of size. The smaller
thinner for the same volumetric ratio of Si and same inner the wire/pore diameter, the smaller is the thermal conductiv-
pore radius. A smaller Si shell thickness means smaller Ge ity of periodic two-dimensional nanocomposites for a given
shell thickness. This gives a smaller effective thermal con- volumetric fraction. Composites with tubular nanowire inclu-
ductivity for both the Si and Ge layers and thus a smaller sions have both lower effective and solid thermal conductiv-
effective thermal conductivity of the nanocomposites. Com- ity than simple nanowire composites due to the introduction
parison of Figs. 6共a兲 and 6共b兲 shows that the effective ther- of surface scattering through the pores. We also show that the
mal conductivity of the composites decreases as the radius of effective medium approach based on incorporating the ther-
the inner pores increases for fixed Si shell thickness and mal boundary resistance into the solution based on the Fou-
volumetric ratio of Si due to the increase of surface scatter- rier heat conduction law leads to erroneous results that un-
ing area per unit volume. derpredict the size effects. Results of this study can be used
Figure 7 shows that the tubular Si nanowire composite to direct the development of both high efficiency thermoelec-
has lower solid thermal conductivity than that of simple tric materials and thermal interface materials with high ther-
nanowire composite due to the additional surface scattering mal conductivity particle or wire inclusions.
introduced through the inner pore surface. Here the solid
thermal conductivity kS is defined as kS = QL / AS⌬T, where Q ACKNOWLEDGMENTS
is the heat flux, L is the length of the simulation domain
along the direction where the temperature difference ⌬T is The work is financially supported by NASA Contract No.
applied, and AS is the solid part of the cross-sectional area NAS3-03108.

125418-6
THERMAL CONDUCTIVITY OF SIMPLE AND TUBULAR… PHYSICAL REVIEW B 72, 125418 共2005兲

*Also at Department of Mechanical Engineering, University of 15 S. Y. Lu and J. L. Song, Chem. Eng. Sci. 51, 4393 共1996兲.
Colorado, Boulder, Colorado 80309, USA. 16
O. L. Lazarenkova and A. A. Balandin, J. Appl. Phys. 89, 5509
1
Y. N. Xia, P. D. Yang, Y. G. Sun, Y. Y. Wu, B. Mayers, B. Gates, 共2001兲; Phys. Rev. B 66, 245319 共2002兲; A. A. Balandin and O.
Y. D. Yin, F. Kim, and Y. Q. Yan, Adv. Mater. 共Weinheim, Ger.兲 L. Lazarenkova, Appl. Phys. Lett. 82, 415 共2003兲.
15, 353 共2003兲; M. Law, J. Goldberger, and P. D. Yang, Annu. 17
B. Yang and G. Chen, Phys. Rev. B 67, 195311 共2003兲.
Rev. Mater. Res. 34, 83 共2004兲. 18
J. A. Fay, Introduction to Fluid Mechanics 共MIT Press, Cam-
2 A. R. Abramson, W. C. Kim, S. T. Huxtable, H. Tan, Y. Wu, A.
bridge, MA, 1994兲.
19
Majumdar, C.-L. Tien, and P. Yang, J. Microelectromech. Syst. J. H. Lienhard, A Heat Transfer Textbook, http://web.mit.edu/
13, 505 共2004兲. lienhard/www/ahtt.html 共2005兲.
3
R. S. Prasher, J. Shipley, S. Prastic, P. Koning, and J. L. Wang, J. 20
M. S. Jeng, R. G. Yang, and G. Chen, Phys. Rev. B 共to be pub-
Heat Transfer 125, 1170 共2003兲. lished兲.
4
V. Lysenko, S. Perichon, B. Remaki, D. Barbier, and B. Cham- 21
G. Chen, Phys. Rev. B 57, 14958 共1998兲.
pagnon, J. Appl. Phys. 86, 6841 共1999兲. 22 K. E. Goodson and Y. S. Ju, Annu. Rev. Mater. Sci. 29, 261
5 A. Drost, P. Steiner, H. Moser, and W. Lang, Sens. Mater. 7, 111
共1999兲.
共1995兲. 23 A. Majumdar, J. Heat Transfer 115, 7 共1993兲.
6
G. Chen, Semicond. Semimetals 71, 2003 共2001兲. 24
G. Chen and C. L. Tien, J. Thermophys. Heat Transfer 7, 311
7
R. G. Yang and G. Chen, Phys. Rev. B 69, 195316 共2004兲. 共1993兲.
8
H. J. Golsmid, Thermoelectric Refrigeration 共Plenum Press, New 25
Computing Methods in Reactor Physics, edited by H. Greenspan,
York, 1964兲. C. N. Kelber, and D. Okrent 共Gorden and Breach, London,
9 T. M. Tritt, ed., Semicond. Semimetals 69–71 共2001兲.
1967兲.
10 P. L. Kapitza, J. Phys. 4, 181 共1941兲; E. T. Swartz and R. O. Pohl, 26 A. A. Joshi and A. Majumdar, J. Appl. Phys. 74, 31 共1993兲.

Rev. Mod. Phys. 61, 605 共1989兲. 27 M. F. Modest, Radiative Heat Transfer, 2nd ed. 共Academic Press,
11
D. P. H. Hasselman and L. F. Johnson, J. Compos. Mater. 21, 508 New York, 2003兲.
共1987兲. 28 R. G. Yang, G. Chen, M. Laroche, and Y. Taur, J. Heat Transfer
12
Y. Benvensite, J. Appl. Phys. 61, 2840 共1987兲; Y. Benvensite and 127, 298 共2005兲.
T. Miloh, ibid. 69, 1337 共1991兲. 29 E. E. Lewis and W. F. Miller, Computational Method of Neutron
13 C.-W. Nan, R. Birringer, D. R. Clarke, and H. Gleiter, J. Appl.
Transport 共Wiley, New York, 1984兲, pp. 156–207.
Phys. 81, 6692 共1997兲; C.-W. Nan, Z. Shi, and Y. Lin, Chem. 30
C. Dames and G. Chen, J. Appl. Phys. 95, 682 共2004兲.
Phys. Lett. 375, 666 共2003兲. 31 X. B. Zhao, X. H. Ji, Y. H. Zhang, T. J. Zhu, J. P. Tu, and X. B.
14
J. D. Felske, Int. J. Heat Mass Transfer 47, 3453 共2004兲. Zhang, Appl. Phys. Lett. 86, 062111 共2005兲.

125418-7