ARTICLE IN PRESS

Physica B 349 (2004) 270280

Molecular dynamics study of the lattice thermal conductivity

of Kr/Ar superlattice nanowires
Yunfei Chena,*, Deyu Lib, Juekuan Yangc, Yonghua Wuc,
Jennifer R. Lukesd, Arun Majumdare
a

Department of Mechanical Engineering and China Education Council Key Laboratory of MEMS,
Southeast University, Nanjing, 210096, PR China
b
Department of Mechanical Engineering, University of California, Berkeley, CA 94720, USA
c
Department of Mechanical Engineering and China Education Council Key Laboratory of MEMS,
Southeast University, Nanjing, 210096, PR China
d
Department of Mechanical Engineering and Applied Mechanics, University of Pennsylvania, Philadelphia, PA 19104-6315, USA
e
Department of Mechanical Engineering, University of California, Berkeley, CA 94720, USA
Received 4 December 2003; received in revised form 16 March 2004; accepted 17 March 2004

Abstract
The nonequilibrium molecular dynamics (NEMD) method has been used to calculate the lattice thermal
conductivities of Ar and Kr/Ar nanostructures in order to study the effects of interface scattering, boundary
scattering, and elastic strain on lattice thermal conductivity. Results show that interface scattering poses signicant
resistance to phonon transport in superlattices and superlattice nanowires. The thermal conductivity of the Kr/Ar
superlattice nanowire is only about 13 of that for pure Ar nanowires with the same cross-sectional area and total length
due to the additional interfacial thermal resistance. It is found that nanowire boundary scattering provides signicant
resistance to phonon transport. As the cross-sectional area increases, the nanowire boundary scattering decreases,
which leads to increased nanowire thermal conductivity. The ratio of the interfacial thermal resistance to the total
effective thermal resistance increases from 30% for the superlattice nanowire to 42% for the superlattice lm. Period
length is another important factor affecting the effective thermal conductivity of the nanostructures. Increasing the
period length will lead to increased acoustic mismatch between the adjacent layers, and hence increased interfacial
thermal resistance. However, if the total length of the superlattice nanowire is xed, reducing the period length will lead
to decreased effective thermal conductivity due to the increased number of interfaces. Finally, it is found that the
interfacial thermal resistance decreases as the reference temperature increases, which might be due to the inelastic
interface scattering.
r 2004 Elsevier B.V. All rights reserved.
PACS: 63.20.
e; 63.22.+m, 68.65.g
Keywords: Superlattice nanowire; Molecular dynamics simulation; Thermal conductivity; Interface scattering effect

*Corresponding author. Tel.: 86253792178; fax: 86253791414.

Email-address: [email protected] (Y. Chen).
0921-4526/$ - see front matter r 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.physb.2004.03.247

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280

1. Introduction
Semiconductor superlattices (SL) are of great
interest due to their potential applications in
thermoelectric and optoelectronic devices [1,2].
Superlattice structures provide the possibility of
decreasing materials thermal conductivity while
retaining their electrical conductivity, thus achieving a high thermoelectric gure-of-merit and
improving the performance of thermoelectric
devices. A large amount of experimental and
theoretical work [38] has been carried out to
study the effects of lattice period and interface on
the thermal conductivity of various kinds of
superlattice lms. Results showed that the superlattice may have a much lower thermal conductivity than the value for each of the two materials
composing the superlattice structure along both
the in-plane and the cross-plane directions. Different mechanisms have been proposed to explain
the reduction of the thermal conductivity, including interface phonon scattering due to acoustic
impedance mismatch, phonon scattering by crystal
imperfections at the interface, phonon spectrum
mismatch, and mini-band formation. However,
quantitative analysis of the relative importance of
these different mechanisms has not been completed yet and it is not clear which mechanism
contributes most to the reduction of the thermal
conductivity under different conditions. Experimental results [6,7] indicate that as the period of
the superlattice decreases, the thermal conductivity
along the cross-plane direction also decreases, and
for very short period superlattice, its thermal
conductivity could even fall below the alloy limit,
i.e. if the two materials mixed homogeneously into
an alloy. The in-plane thermal conductivity
follows a similar pattern but it jumps to higher
values for some very short period superlattice [8].
This phenomenon cannot be explained solely by
diffuse mismatch theory about phonon scattering
at the interface. Other mechanisms such as
acoustic impedance mismatch, mini-band formation or phonon spectrum mismatch have to be
taken into account.
Theoretically, lattice dynamics and particle
transport models are usually used to study the
thermal conductivity of superlattice structures.

271

Some recent lattice dynamics work was carried

out by Hyldgaad and Mahan [9] and Tamura and
Tanaka [10]. In their work, a simple cubic lattice
model [9] and a face-centered cubic model [10]
were used to calculate the group velocity of
acoustic phonons in the cross-plane direction.
Their results suggested that reduction of the
phonon group velocity in superlattices could lead
to a reduced thermal conductivity. Considering
their lattice models were too simple to provide
details of phonon dispersion spectra, Kiselev et al.
[11] used a diatomic unit cell model to simulate the
dispersion spectra of the Si/Ge superlattice. Due to
the large mass of Ge atoms in comparison to Si,
the most probable acoustic phonons in Si layers at
room temperature have no counterpart in the
phonon spectra of the Ge layers. In other words, a
phonon at a given frequency in the Si layer may
not be able to proceed into the Ge layer without
scattering with, or into, one or more phonons of
different frequency. This leads to highly efcient
trapping of high-energy phonons in the Si layer
and a drastic reduction of the superlattice thermal
conductivity. Although qualitative agreement can
be obtained through lattice dynamics, it is difcult
to compare those results quantitatively with
experimental results for different materials under
different temperatures due to the fact that lattice
dynamics can only model very simple systems. The
particle transport model treats phonons as individual particles and solves the Boltzmann transport
equation (BTE) to study the phonon transport in
microstructures. Simkin and Mahan [12] showed
that for layers thinner than the mean free path
(mfp) of phonons, the wave aspect of phonons
must be taken into account and wave theory must
be applied, while for layers thicker than the mfp of
phonons, the particle treatment of phonons was
acceptable and BTE could be used to study the
phonon transport. The rst study on the thermal
conductivity of superlattices was carried out by
Ren and Dow [13]. They modeled the thermal
conductivity of ideal superlattice by combining the
BTE with a quantum mechanical treatment of the
additional scattering process caused by the minibands. However, their predicted results could not
match the experimental data. Based on the BTE
and the assumption of partially specular and

ARTICLE IN PRESS
272

Y. Chen et al. / Physica B 349 (2004) 270280

partially diffuse boundary scattering, some interesting results were obtained by Chen et al. [1417]
for superlattice thermal conductivity along both
in-plane and cross-plane directions. In his model
the reduction of the superlattice thermal conductivity was mainly attributed to diffuse scattering at
the interface.
Because some assumptions must be introduced
to get closed form solutions, the theoretical results
usually deviate signicantly from the experimental
data and sometimes the thermal conductivity
reduction mechanisms cannot be readily explained
from those theoretical studies. Classical Molecular
Dynamics (MD) simulation provides an alternative approach to investigate heat transport in
nanostructures. Given the interaction potential
between atoms, the force acting on each atom can
be calculated. Based on Newtons second law, the
motion of a large number of atoms can be
described. Without any further assumptions,
statistical physical properties can be derived from
the ensemble of atoms. In particular, if the size of
the nanostructure is smaller than the phonon mfp,
it is questionable to use the BTE to describe
phonon transport, while MD can be conveniently
used to analyze the effects of size connement on
lattice thermal conductivities. Volz et al. [18]
demonstrated by MD simulation that Si nanowire
thermal conductivity could be two orders of
magnitude smaller than the corresponding bulk
value and they further argued that by adjusting the
specularity parameter, results from the solution of
the BTE could t their MD simulation. Liang et al.
[19] applied MD simulation to investigate the
effects of atomic mass ratio in the alternating
layers of a superlattice on the lattice thermal
conductivity. Their results indicated that the
thermal conductivity has a minimum value for
some specic atomic mass ratio. Abramson et al.
[20] studied the effects of interface number and
elastic strain on the lattice thermal conductivity of
Kr/Ar superlattices with MD simulation. It was
argued that increase of the interface number per
unit length does not necessarily result in decreased
lattice thermal conductivity from their simulation
results. Daly et al. [21,22] reported MD simulation
of a classical face centered cubic (FCC) lattice
model to study the effects of interface roughness

and isotope scattering on thermal conductivities.

Simulation results predicted the similar trends for
the lattice thermal conductivities of GaAs/AlAs
superlattice along both in-plane and cross-plane
directions compared with the experimental data
[23]. In their model, it was also demonstrated
that the lattice thermal conductivity of GaAs/AlAs
superlattice had a minimum value with different
layer thickness. This conclusion supported the
hypothesis that zone folding was the dominant
effect on lattice thermal conductivity in the
short period superlattice. However, it should be
noted that no lattice mismatch between the
alternative materials of the superlattice was
considered in their model. Volz [24] introduced
the conjugate gradient method to minimize the
potential energy of Si/Ge superlattices in order to
relax the elastic strain on the alternating layers.
Simulation results predicted an increasing trend of
the superlattice thermal conductivity with the layer
thickness.
Recently, experimental investigation on the
thermal conductivity of Si and Si/SiGe superlattice
nanowires has been carried out [25,26]. Their
results showed that for Si nanowires, the thermal
conductivity could be greatly reduced compared
with that of bulk Si because of the strong nanowire
boundary scattering. For Si/SiGe nanowires, the
thermal conductivity was below that of the twodimensional Si/SiGe superlattice lms, which was
ascribed to the additional scattering mechanism
provided by the nanowire boundary. In this paper,
we apply molecular dynamics to study the lattice
thermal conductivity of Kr/Ar superlattice nanowires in order to investigate the effects of interface
scattering, nanowire boundary scattering, and
period length.

2. Theoretical model and analysis

Nonequilibrium molecular dynamics (NEMD)
was used to calculate the lattice thermal conductivities of solid Ar and Kr/Ar superlattice nanowires in the present study. The Lennard-Jones
(L-J) potential was used to represent the interac-

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280
Table 1
Simulation parameters

KrKr

Variables

Parameters

e1 (J)
s1 (m)
Lattice constant A1 (m)
Atom mass M1 (Kg)

2:25  10
21
3:65  10
10
5:69  10
10
1:39  10
25

e2 (J)
s2 (m)
Lattice constant A2 (m)
Atom mass M2 (Kg)

21

1:67  10
3:4  10
10
5:3  10
10
6:63  10
26

e12 (J)
s12 (m)

p
e1 e2
s1 s2 =2

this way, heat ux was introduced into the

simulation domain and a temperature gradient
was set up along the heat ux direction. From the
heat ux J and the temperature gradient, the
thermal conductivity of the nanowires can be
calculated from the Fourier law
K

ArAr

KrAr

TH

Kr

Ar

Kr

Ar

z
y
o

Lp
x
L

Fig. 1. Physical model of Kr/Ar superlattice nanowire.

tion between two particles




sij 12
sij 6
Vij r 4eij


;
r
r

where subscripts i and j stand for either argon or

krypton particles, eij and sij represent the energy
and length scale of the potential, and r denotes the
distance between the two particles. Table 1 gives
the selected parameters and the simulation domain
for the Kr/Ar superlattice nanowire is shown in
Fig. 1.
The simulation domain consisted of a hot bath
with a high-constant temperature TH ; a cold bath
with a low-constant temperature TL and alternating Kr/Ar layers. Constant heat ux was added to
the hot bath and the same amount of heat ux was
subtracted from the cold bath at the same time. In

273

J
;
ArT

where A is the cross-sectional area and rT is the

temperature gradient. The temperature gradient
can be obtained according to the equation
rT

TH
TL
;
Lp  N

where Lp and N are the length and the number of

periods for the superlattice nanowire shown in Fig.
1. TH and TL temperatures of the bath atoms in
contact with either side of the nanowire. Phonon
scattering at the interfaces between the hot and
cold baths and the superlattice structure leads to
temperature jumps between the thermal bath
layers nearest the nanowire and the rst adjacent
layer of the nanowire [27,28], which results in a
smaller temperature gradient within the wire than
that imposed externally. In our simulation, Kr
atoms were selected to form the hot bath and Ar
atoms were selected as the cold bath material in
order to minimize the interfacial thermal resistance
on either side. Fixed boundary conditions on the
outer surface of the nanowire were introduced to
avoid evaporation from the nanowire domain
during the simulation process. In the xed
boundary regions, the atoms were xed in their
equilibrium positions. They interacted with the
atoms in the nanowire through the L-J potential.
The parameters are listed in Table 1. In Fig. 1 the
xed boundary regions in the x direction are
depicted by the dashed lines. The thickness of the
xed boundary region was set as 1 unit cell (UC).
It has been shown [18] that for Si nanowires, the
difference between xed boundary condition and
free boundary condition was less than 10%.
According to Lukes et al. [27], two important
criteria for the simulation are that the effective
thermal ux in each atomic layer is nearly the same
as that imposed externally, and that the temperature proles are reasonably close to linear. The

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280

274

proles should not be expected to be completely

linear, since the thermal conductivity is temperature dependent. In order to satisfy the two criteria,
computation time should be selected as long as
possible for the system to reach steady-state. In
this paper, to avoid undue computational burden,
values roughly 20 times of the characteristic
diffusion time were found sufcient to satisfy the
criteria. For the thermal conductivity simulations
reported here, the values of the effective thermal
ux were all within 4% of the imposed ux and the
temperature proles were reasonably linear.
Local temperature calculation is another issue
to be concerned in MD simulation, which relates
to the size of the region over which temperature is
dened. To dene a local temperature, local
thermodynamic equilibrium must be set up, which
means that the temperature must be dened over a
region with characteristic length larger than the
phonon mfp. However, it has been shown that for
classical MD simulation, local temperature can be
dened for each layer of atoms as long as there are
enough atoms in each layer [30]. In this paper, Tp
in the prole was evaluated from the equation
Tp ma

Np
X

v2i =3Np kB ;

i1

where ma ; Np are the mass and the number of

atoms in a local plane, vi the velocity of each atom,
and kB the Boltzmann constant. The criteria for
Eq. (4) to be valid are that local thermodynamic
equilibrium is established in each layer, and the
simulation temperature is above the Debye temperature of the material so that the heat capacity is
temperature independent. According to Hafskjold
[30], at least 30 atoms are needed in each layer in
order to reach local equilibrium and to dene an
instantaneous temperature. So for a superlattice
nanowire with a square cross-section, the width of
the cross-section should be up to at least 4UC. On
the other hand, Lukes et al. [27] demonstrated that
prolonging the simulation time was more efcient
for a system to reach equilibrium compared with
increasing the number of atoms in the simulation
system. So in our simulation, the lower limit of the
cross-sectional area of the superlattice nanowires
was selected as 4UC  4UC: As the simulation

temperature in this paper was below the Debye

temperatures of Kr and Ar (72, 92 K; respectively), quantum modication should be introduced to correct both the MD temperature and the
thermal conductivity. However, it has been suggested [31] that the classical MD model could give
acceptable prediction for lattice thermal conductivity as long as the temperature is higher than
one-fourth of the Debye temperature. Since the
reference temperature in our simulation system is
above one-fourth of the Debye temperature for
both Kr and Ar, for simplicity, we still used Eq. (4)
to estimate the local temperature.
Temperature discontinuity appeared in the
temperature prole at interfaces between Kr and
Ar layers along the superlattice nanowires, which
was caused by the interfacial thermal resistance.
The interfacial thermal resistance can be calculated
from the following equation [32]:
Rint

DT
;
J=A

where A is the cross-sectional area, DT the

temperature discontinuity at the interface and J
the heat ux. The interfacial thermal resistance
depends on the number of the phonons incident on
the interface, the energy carried by each phonon
and the probability that each phonon is transmitted across the interface. Two limiting models
have been proposed to describe phonon transport
across an interface: the acoustic mismatch model,
which assumes specular reection of phonons at
the interface, and the diffuse mismatch model,
which assumes that all phonons incident on the
interface will be diffusively scattered. However, it
is difcult to argue if the interfacial thermal
resistance can be calculated from these two models
because of the strong assumptions made in these
two theories. On the other hand, in MD simulation, the interfacial thermal resistance can be
readily calculated from Eq. (5) according to the
classical denition of interfacial thermal resistance. The total thermal resistance along the
nanowire can be described as
Rtot nRint NRA RB

where n is the number of the interface, N the

number of periods and n 2N
1: RA ; RB are

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280

Keff

Lp
:
RA RB 2Rint
Rint =N

As we know, under atmospheric pressure, for an

interface between two bulk materials, if there is no
interface imperfection, the interfacial thermal
resistance only depends on temperature and the
materials that constitute the interface. Once the
materials and the temperature are determined, the
interfacial thermal resistance should be a constant.
Thus, from Eq. (7), it can be concluded that the
effective thermal conductivity of superlattice nanowire will depend on the length and number of
periods. It predicts that the thermal conductivity
of the superlattice nanowire increases with the
period length. However, for superlattice nanowires, due to the fact that the period length may be
less than the phonon mean free path, the thermal
resistance in the alternating layers and the interfacial thermal resistance are all layer thickness
dependent, so the relation between the effective
thermal conductivity and the period length might
be nonlinear.

3. Simulation results and discussion

3.1. Effect of the number of interfaces
NEMD was used to predict lattice thermal
conductivities of pure Ar and Kr/Ar superlattice
nanowires. Fig. 2 shows the thermal conductivities
of Ar and Kr/Ar superlattice nanowires versus
nanowire length for a xed period length of 8UC.
For the Kr/Ar superlattice nanowire, one period is
composed of two layers and the thickness of each
layer is 4UC. The total length of the Ar nanowire
was chosen to be the same as that of the Kr/Ar

0.45
Thermal conductivity (W/K.m)

the thermal resistance of each Ar and Kr layer,

respectively. Eq. (6) is only an approximate
expression for the total thermal resistance. In fact,
due to the difference of the reference temperature
in different layers along the superlattice nanowire,
the thermal resistances RA ; RB should not be
exactly the same for different sections. Assuming
that RA ; RB are the same along the whole
nanowire, the effective thermal conductivity of
the superlattice nanowire can be approximated as

275

0.40
Ar Nanowire
Ar/Kr superlattice nanowire

0.35
0.30
0.25
0.20
0.15
2

4
6
Period number

Fig. 2. Thermal conductivities of Ar and Kr/Ar superlattice

nanowires.

superlattice nanowire. Also, the two kinds of

nanowires are of the same cross-sectional area as
4UC  4UC: The temperatures in the hot bath and
the cold bath are 50 and 10 K; respectively, and
the reference temperature is 30 K: Fig. 2 shows
that the thermal conductivity of the Kr/Ar
nanowire is only about one-third of that of the
Ar nanowire, which we believe is due to the
additional interface scattering in the Kr/Ar superlattice nanowire. For Ar nanowire, its thermal
conductivity increases as the length of the nanowire increases up to 40UC. We believe that this is
due to the scattering from the heat source and heat
sink. If the length of the simulation domain is
shorter than the phonon mfp, then the nanowire
thermal conductivity should be length dependent,
i.e. the thermal conductivity increases with the
length of the simulation domain. The phonon mfp
for Ar [31,33] is estimated to be approximately
5 nm (1016UC for Ar and Kr) at 30 K: However,
low frequency phonons have longer mfps while
high frequency phonons have shorter mfps, which
means that even for a system of a length longer
than the phonon mfp, the effects of phonon
scattering at the heat source and heat sink may
still have some effect on the total thermal
conductivity. In our calculation, the thermal
conductivity of Ar nanowire increases with the
nanowire length until the length exceeds 40UC,
which we think comes from the scattering of long
wavelength phonons at the heat source and heat

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280

0.33

superlattice
SL nanowire

0.30
K (W/K.m)

0.27
0.24
0.21
0.18
0.15
6

12
15
18
Period length (UC)

21

Fig. 3. Thermal conductivities of Kr/Ar superlattice and

superlattice nanowires.

0.33

0.22

Rint

0.20

Rint/R

0.30
0.27

0.18

0.24

0.16

0.21

0.14

0.18

0.12

0.15

0.10

3.2. Effect of period length on the thermal

conductivity and the interfacial thermal resistance
In order to study the effects of period length,
Kr/Ar superlattice and superlattice nanowires of
different period length were investigated with a
xed total length of 48UC and a cross-sectional
area of 4UC  4UC: Fig. 3 gives thermal conductivity versus period length, which reveals the
effects of the interface and nanowire boundary on
the lattice thermal conductivity. For nanowires of
xed total length, the lattice thermal conductivity
increases with the period length because fewer
interfaces are present in the longer period superlattice nanowire. In addition, the interfacial
thermal resistance is not a constant as the period
length changes. It increases as the period length
increases, as shown in Fig. 4. This phenomenon is
attributed to the interface strain effects on phonon
transport [20]. The atomic spacing on either side of
the interface tends to either slightly contract (for

24

Rint/R

sink. Typically for simulations on bulk materials

one should not see a size dependence and for those
cases size effects would be an unphysical artifact.
However for materials with characteristic dimensions on or below the order of a phonon mfp, there
will be a size dependence and it is a true, physical
effect [27,29]. Thermal conductivity in that case is
not intrinsic but is dominated by boundary effects.
Nonetheless, this size dependent phenomenon
cannot be observed for Kr/Ar superlattice nanowire, as shown in Fig. 2. With the increase of
nanowire length, the thermal conductivities of Kr/
Ar superlattice nanowires remain almost constant
and a very slight decreasing trend appears when
the period number exceeds 5. The reason might be
that the effects of phonon scattering at the
boundaries of the hot and cold baths are masked
by the interface scattering. The effects of phonon
scattering on the heat source and heat sink
boundaries play the same role as the interface
scattering so for superlattice nanowires, the data
of the thermal conductivities for different length of
the nanowire superlattice seem to be consistent
with formula (7), i.e. for a xed period length
superlattice nanowire, its thermal conductivity
keeps constant.

Rint (K.m2/W)

276

12
15
18
Period length (UC)

21

24

Fig. 4. Interfacial thermal resistance of Kr/Ar superlattice

nanowires.

Kr atoms) or expand (for Ar atoms) to align with

adjacent atoms because of the lattice mismatch
between these two materials. For short period
superlattice nanowires, the elastic strain causes the
lattice in the adjacent layers to match better with
each other, which poses less resistance to phonon
transport. The difference of elastic strain for
different period length is shown in Fig. 5. The
cross-sectional areas for the two nanowires are
5:19 nm2 and their period lengths are 4.4 and
13:2 nm; respectively. Two and three periods were
simulated for the 13:2 nm and 4:4 nm period
length nanowires, respectively, so the nanowires
are labeled as 2  13:2 nm  5:19 nm2 ; 3 
4:4 nm  5:19 nm2 : The two dashed lines labeled
Kr and Ar stand for the distance between two

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280

nanowire, the nanowire boundary scattering pose

signicant resistance to the phonon transport.

N=3, Lp=4.4nm,A=5.19nm2

Dimensionless distance

N=2,Lp=13.2nm,A=5.19nm2
Kr

0.78

0.76

0.74
Ar
0.72
0

20

40
60
80
100
Atomic plane number along X Axis

Fig. 5. Elastic strain in Kr/Ar superlattice nanowires.

neighboring atomic planes along the x-axis for

pure Kr and pure Ar, respectively. For the longer
period nanowire, the atomic planes far away from
the interface tend to maintain the equilibrium
spacing found in the pure material. For the shorter
period nanowire, however, most of the atoms in
the two layers deviate from their equilibrium
spacing, as shown in Fig. 5. We believe that as
the period length increases, the acoustic impedance
mismatch between the two different material layers
increases, which causes the larger interfacial
thermal resistance as shown in Fig. 4. However,
because the total length of the nanowire is xed,
the ratio of the total interface resistance to the
total effective thermal resistance decreases for
increasing period length. This is due to the fact
that the number of interfaces is reduced for longer
period superlattice nanowires if the total length of
the nanowire is xed.
Fig. 3 also gives the thermal conductivity of a
Kr/Ar superlattice lm in order to compare the
effects of cross-sectional area on the thermal
transport. The period length for the superlattice
is the same as that of Kr/Ar superlattice nanowires. Periodic boundary conditions were imposed
in y and z directions to stand for innite crosssectional area. Simulation results show that the
thermal conductivity of superlattice lms is much
higher than that of superlattice nanowires for the
same period length, which is consistent as the
experimental observations [26]. This means that in
addition to the interface scattering in superlattice

3.3. Effects of cross-sectional area on the thermal

conductivity and the interfacial thermal resistance
At low temperature, thermal conductivity increases with the sample size due to the fact that
phonon-boundary scattering dominates the thermal resistance. For larger cross section, the
probability of phonon-boundary scattering reduces, which leads to higher lattice thermal
conductivity. To explore the effects of different
cross-sectional area on the thermal conductivity of
superlattice nanowires, we simulated thermal
transport in nanowires of different cross-sectional
area as 4UC  4UC; 5UC  5UC; 6UC  6UC
and 7UC  7UC: To save computational time, the
total length of the nanowire was chosen as 24UC,
the period length as 8UC, and the thickness of
each layer as 4UC. The simulation results are
shown in Fig. 6. It can be seen that the thermal
conductivity increases nearly linearly with the size
of the cross section. If we assume constant specic
heat for all four nanowires, kinetic theory predicts
that the thermal conductivity is proportional to
the phonon mfp. The equivalent phonon mfp can
be written as
1
1
1
1

;
l lphonon linterface lboundary

0.30
0.27
K(W/K.m)

0.82

0.80

277

0.24

SL nanowire
SL

0.21
0.18
0.15
4

10

12

14

16

Section area (nm2)

Fig. 6. Thermal conductivity of Kr/Ar superlattice and superlattice nanowires.

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280

where the subscripts phonon, interface and

boundary stand for the effective phonon mfp
determined by the three phonon scattering process,
the interface scattering, and nanowire boundary
scattering, which is related to the cross-section
size. If we do not take into account the modication of the phonon dispersion relation due to the
different diameters of the nanowires, the phonon
mfp due to anharmonic scattering can be approximated as constant. From Eq. (8), it can be deduced
that the effective thermal resistance of the nanowire decreases for increasing cross section, which is
conrmed by the results shown in Fig. 6. Fig. 7
shows that for superlattice nanowires, the ratio of
the interfacial thermal resistance to the total
thermal resistance increases as the cross-sectional
area increases. If the cross-sectional area increased
to innity, i.e. the superlattice nanowire becomes
the superlattice lm, the ratio increased to 42%.
For Kr/Ar superlattice lms shown in Figs. 6
and 7, the period length and period number are the
same as that of the superlattice nanowires. The
difference between the superlattice lms of different simulation domain is only marginal due to the
periodic boundary conditions for superlattice
lms. The value of the thermal conductivity for
superlattice lms is the upper limit of the thermal
conductivity for superlattice nanowires. This is
also true for the ratio of the interface resistance to
the total thermal resistance, as shown in Fig. 7.
Considering the minor difference among the
nanowires diameter for the four types of nanowires, any resulting differences in dispersion
relation will be slight so the change of the phonon

dispersion relations among the nanowires is

neglected [34,35]. The reduced inuence of boundary scattering processes for the thicker nanowires
is the main reason that the value of thermal
conductivity increases with the cross sectional
area.

3.4. The dependence of interfacial thermal

resistance on temperature
The total thermal resistance of a superlattice
nanowire equals the sum of the thermal resistance
in each layer and the interfacial thermal resistance.
From both the acoustic mismatch model and the
diffuse mismatch model, at very low temperature,
the interfacial thermal resistance decreases with
temperature at a rate proportional to T
3 ; due to
the temperature dependence of specic heat. As
the temperature rises, the specic heat approaches
a constant value, so the interfacial thermal
resistance also approaches a constant value. Fig.
8 plots the interfacial thermal resistance and the
thermal conductivity of superlattice nanowires
versus temperature, which shows that as the
temperature increases, both the interfacial thermal
resistance and the nanowire thermal conductivity
decrease in the temperature range of 3555 K: The
reason for the thermal conductivity reduction can
be ascribed to the increased thermal resistance of
each layer due to phononphonon scattering.
However, the reason for the interfacial thermal
resistance decrease is not fully understood. From
0.22

0.23

0.42

0.35
0.34

0.41

SL
SL nanowire

0.40

0.33
0.32

0.39

0.31

0.38

K (W/K. m)

0.22

Rint/R for SL

Rint/R for SL nanowire

K
R
0.20

0.21

0.18

R (K m2/W)

278

0.20
0.16
0.19

10

12

14

16

Fig. 7. Interfacial thermal resistance of Kr/Ar superlattice and

superlattice nanowires.

35

40

45

50

55

Temperature T(K)

Fig. 8. Thermal conductivity and interfacial thermal resistance

of Kr/Ar superlattice nanowire.

ARTICLE IN PRESS
Y. Chen et al. / Physica B 349 (2004) 270280

the acoustic mismatch model, the thermal boundary resistance can be approximated as 14 Cvs a
1 ;
where, C; vs ; and a are the specic heat, the Debye
phonon group velocity, and the averaged phonon
transmission probability. The Debye phonon
group velocity can be regarded as a constant.
The phonon transmission probability is typically
assumed in the literature to be independent of
temperature [32]. The above equation predicts that
the thermal boundary resistance is independent of
temperature. The decrease of the interfacial
thermal resistance cannot be derived from the
diffuse mismatch model either. One possible
reason might be that inelastic scattering occurs at
the interface [32]. At higher temperature, the highfrequency phonons might break into two or more
low-frequency phonons, which will be predominantly transmitted. This inelastic scattering process leads to higher phonon transmission
probabilities, thus the interfacial thermal resistance decreases.

279

due to the larger acoustic impedance mismatch

between the two layers.
If the total length of the superlattice nanowire is
xed, increasing the interface number per unit
length will lead to decreased interfacial thermal
resistance. However, the effective thermal conductivity for the superlattice nanowire decreases
due to the fact that the number of interfaces
increases. Nanowire boundary scattering also
contributes to the total thermal resistance. Increasing the cross-sectional area of superlattice
nanowires leads to higher thermal conductivity.
The ratio of the interfacial thermal resistance
increases with the cross-sectional area, and as an
upper limit, the ratio of the interfacial thermal
resistance to the total thermal resistance reaches
42% for Kr/Ar superlattice lms. It is also found
that the interfacial thermal resistance decreases
with the temperature, which might be explained by
inelastic phonon scattering at the interface.

Acknowledgements
4. Conclusion
In summary, the lattice thermal conductivity of
Ar nanowires and Kr/Ar nanowires were studied
by NEMD. The interfacial thermal resistance in
the Kr/Ar nanowires contributes signicantly to
the thermal resistance and results in a lattice
thermal conductivity that is only one-third of that
of pure Ar nanowires. Interfacial thermal resistance increases with the length of the superlattice
period for the Kr/Ar superlattice nanowires, which
is attributed to increased acoustic impedance
mismatch between adjacent layers for long period
nanowires. For short period nanowires, the
acoustic impedance in the alternating layers beside
the interface matches better due to elastic strain
throughout each layer. The phonons can pass
through the interface more easily, which results in
smaller interfacial thermal resistance. On the other
hand, for longer period superlattice nanowires, the
atoms located far away from the interface maintain their pure crystalline lattice positions so that
the strain is localized near the interface. It is more
difcult for phonons to pass through the interface

Y.C. would like to acknowledge the nancial

support of the Chinese Natural Science Foundation (Project No. 50276011, 50275026) and the 863
High Technology Program (Project No.
2003AA404160).

References
[1] X. Fan, G. Zeng, C. LaBounty, J. Bowers, E. Croke,
C. Ahn, S. Huxtable, A. Majumdar, A. Shakouri, Appl.
Phys. Lett. 78 (2001) 1580.
[2] M.R. Kitchin, M.J. Shaw, E. Corbin, J.P. Hagon,
M. Jaros, Phys. Rev. B 61 (2000) 8375.
[3] T. Yao, Appl. Phys. Lett. 51 (1987) 1798.
[4] G. Chen, C.L. Tien, X. Wu, J.S. Smith, J. Heat Trans. 116
(1994) 325.
[5] X.Y. Yu, G. Chen, A. Verma, J.S. Smith, Appl. Phys. Lett.
67 (1995) 3554.
[6] S.-M. Lee, D. Cahill, R. Venkatasubramanian, Appl. Phys.
Lett. 70 (1997) 2957.
[7] W.S. Capinski, H.J. Maris, Physica B 219 (1996) 699.
[8] S.T. Huxtable, A.R. Abramson, C.-L. Tien, Arun Majumdar, Appl. Phys. Lett. 80 (2002) 1737.
[9] P. Hyldgaad, G.D. Mahan, Phys. Rev. B 56 (1997) 10754.
[10] S. Tamura, Y. Tanaka, H.J. Maris, Phys. Rev. B 60 (1999)
2627.

ARTICLE IN PRESS
280

Y. Chen et al. / Physica B 349 (2004) 270280

[11] A.A. Kiselev, K.W. Kim, M.A. Stroscio, Phys. Rev. B 62

(2000) 6896.
[12] M.V. Simkin, G.D. Mahan, Phys. Rev. Lett. 84 (2000)
927.
[13] S.Y. Ren, J.D. Dow, Phys. Rev. B 25 (1982) 3750.
[14] G. Chen, J. Heat Trans. 119 (1997) 220.
[15] G. Chen, Phys. Rev. B 57 (1998) 14958.
[16] G. Chen, M. Neagu, Appl. Phys. Lett. 71 (1997) 2761.
[17] G. Chen, J. Heat Trans. 119 (1996) 220.
[18] S. Volz, G. Chen, Appl. Phys. Lett. 75 (1999) 2056.
[19] X.G. Liang, B. Shi, Materials Sci. Eng. A. 292 (2000) 198.
[20] A.R. Abramson, C.-L. Tien, A. Majumdar, J. Heat Trans.
124 (2002) 963.
[21] B.C. Daly, H.J. Maris, K. Imamura, S. Tamura, Phys.
Rev. B 66 (2002) 024301.
[22] B.C. Daly, H.J. Maris, K. Imamura, S. Tamura, Phys.
Rev. B 67 (2003) 033308.
[23] W.S. Capinski, H.J. Maris, T. Ruf, M. Cardona, K. Ploog,
D.S. Katzer, Phys. Rev. B 59 (1999) 8105.

[24] S. Volz, J.B. Saulnier, G. Chen, P. Beauchamp, Microelectron. J. 31 (2000) 815.

[25] D. Li, Y. Wu, P. Kim, L. Shi, P. Yang, A. Majumdar,
Appl. Phys. Lett. 83 (2003) 2934.
[26] D. Li, Y. Wu, R. Fan, P. Yang, A. Majumdar, Appl. Phys.
Lett. 83 (2003) 3186.
[27] J.R. Lukes, D. Li, X.-G. Liang, C.-L. Tien, J. Heat Trans.
122 (2000) 536.
[28] P.K. Schelling, S.R. Phillpot, P. Keblinski, Phys. Rev. B 65
(2002) 144306.
[29] H.B.G. Casimir, Physica 5 (1938) 595.
[30] B. Hafskjold, S.K. Ratkje, J. Stat. Phys. 78 (1995) 463.
[31] H. Kaburaki, J. Li, S. Yip, Mater. Res. Soc. Symp. Proc.
538 (1998) 503.
[32] E.T. Swartz, R.O. Pohl, Rev. Mod. Phys. 61 (1989) 605.
[33] S. Volz, J.-B. Saulnier, M. Lallemand, B. Perrin,
P. Depondt, M. Mareschal, Phys. Rev. B 54 (1996) 340.
[34] C. Dames, G. Chen, J. Appl. Phys. 95 (2) (2004) 682.
[35] J. Zou, A. Balandin, J. Appl. Phys. 89 (5) (2001) 2932.