Purdue University

Purdue e-Pubs
Birck and NCN Publications Birck Nanotechnology Center

5-1-2012

Thermal conductivity of bulk and thin-film silicon:

A Landauer approach
Changwook Jeong
Birck Nanotechnology Center, Purdue University

Supriyo Datta
Birck Nanotechnology Center, Purdue University, [email protected]

Mark S. Lundstrom
Birck Nanotechnology Center, Purdue University, [email protected]

Follow this and additional works at: http://docs.lib.purdue.edu/nanopub

Part of the Nanoscience and Nanotechnology Commons

Jeong, Changwook; Datta, Supriyo; and Lundstrom, Mark S., "Thermal conductivity of bulk and thin-film silicon: A Landauer
approach" (2012). Birck and NCN Publications. Paper 1201.
http://dx.doi.org/10.1063/1.4710993

This document has been made available through Purdue e-Pubs, a service of the Purdue University Libraries. Please contact [email protected] for
additional information.
 7KHUPDOFRQGXFWLYLW\RIEXONDQGWKLQILOPVLOLFRQ$/DQGDXHUDSSURDFK
&KDQJZRRN-HRQJ6XSUL\R'DWWDDQG0DUN/XQGVWURP

&LWDWLRQ-$SSO3K\VGRL

9LHZRQOLQHKWWSG[GRLRUJ
9LHZ7DEOHRI&RQWHQWVKWWSMDSDLSRUJUHVRXUFH-$3,$8YL
3XEOLVKHGE\WKH$,33XEOLVKLQJ//&

$GGLWLRQDOLQIRUPDWLRQRQ-$SSO3K\V
-RXUQDO+RPHSDJHKWWSMDSDLSRUJ
-RXUQDO,QIRUPDWLRQKWWSMDSDLSRUJDERXWDERXWBWKHBMRXUQDO
7RSGRZQORDGVKWWSMDSDLSRUJIHDWXUHVPRVWBGRZQORDGHG
,QIRUPDWLRQIRU$XWKRUVKWWSMDSDLSRUJDXWKRUV

Downloaded 10 Sep 2013 to 128.46.221.64. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
 JOURNAL OF APPLIED PHYSICS 111, 093708 (2012)

Thermal conductivity of bulk and thin-film silicon: A Landauer approach

Changwook Jeong, Supriyo Datta, and Mark Lundstrom
Network for Computational Nanotechnology, Birck Nanotechnology Center, Purdue University,
West Lafayette, Indiana 47907, USA

(Received 1 February 2012; accepted 31 March 2012; published online 3 May 2012)
The question of what fraction of the total heat flow is transported by phonons with different
mean-free-paths is addressed using a Landauer approach with a full dispersion description of
phonons to evaluate the thermal conductivities of bulk and thin film silicon. For bulk Si, the results
reproduce those of a recent molecular dynamic treatment showing that about 50% of the heat
conduction is carried by phonons with a mean-free-path greater than about 1 lm. For the in-plane
thermal conductivity of thin Si films, we find that about 50% of the heat is carried by phonons with
mean-free-paths shorter than in the bulk. When the film thickness is smaller than 0.2 lm, 50% of
the heat is carried by phonons with mean-free-paths longer than the film thickness. The cross-plane
thermal conductivity of thin-films, where quasi-ballistic phonon transport becomes important, is
also examined. For ballistic transport, the results reduce to the well-known Casimir limit [H. B. G.
Casimir, Physica 5, 495–500 (1938)]. These results shed light on phonon transport in bulk and
thin-film silicon and demonstrate that the Landauer approach provides a relatively simple but
accurate technique to treat phonon transport from the ballistic to diffusive regimes. V
C 2012 American

Institute of Physics. [http://dx.doi.org/10.1063/1.4710993]

I. INTRODUCTION in-plane thermal conductivities with constant and frequency-

dependent p suggested that the frequency-dependent model
Recent molecular dynamics (MD) simulations2 have
provides a more accurate description for the in-plane thermal
shown that in bulk silicon (Si) about 50% of the heat is car-
conductivity.11 We show in this paper that with the addition
ried by phonons with mean-free-paths (MFPs) greater than
of a model for surface roughness scattering, the Landauer
about 1 lm—a fact that is surprising and that the authors of model used for bulk Si also accurately describes in-plane
Ref. 2 could not explain with a simple, Callaway model with thermal transport. We then use the model to address the
Debye approximation.3 These results raise similar questions question raised by the authors of Ref. 2: “How do phonons
about heat transport in thin Si films. In this paper, we show with various mean-free-paths contribute to thermal conduc-
that a simple Landauer model, essentially a Callaway model tion in thin Si films?”
with full phonon dispersion, accurately reproduces the The measured cross-plane thermal conductivity of thin
results of Ref. 2. We also show that the same model Si films has been found to be even lower than the in-plane
describes the in-plane thermal conductivity of silicon thin thermal conductivity.12 In some models,13–15 the reduction
films and find that about 50% of the heat is carried by pho- in cross-plane thermal conductivity is modeled with the
nons with mean-free-paths shorter than in the bulk. When Boltzmann transport equation (BTE) by including a phonon-
the film thickness is smaller than 0.2 lm, 50% of the heat boundary scattering time (sb ), which is assumed to be equal
is carried by phonons with mean-free-paths longer than the to the average time between “boundary scattering events:”
film thickness. Finally, we apply the technique to cross-plane sb ¼ L=ð2tÞ, where L is the length of the conductor and t the
thermal transport in Si. The results shed light on thermal group velocity. Ballistic phonon transport (i.e., no phonon
transport in thin Si films and demonstrate that the Landauer scattering within thin film), which becomes important in
approach provides a simple and accurate treatment of lattice cross-plane transport is typically described by an equation of
thermal conductivity that is useful for analyzing experiments phonon radiative transport (EPRT),16 which has been devel-
and for designing materials and structures. oped based on the Boltzmann equation and the analogy
The findings of Minnich et al.2 for bulk Si raise ques- between phonons and photons. In the ballistic limit and with
tions about heat conduction in thin Si films and how the heat the Debye approximation, the EPRT yields a familiar black-
is carried by phonons with different MFPs. In silicon on in- body radiation law for phonons,17 q ¼ rðTH4  TC4 Þ, where r
sulator (SOI) films, it is well-known that the in-plane thermal is the Stefan-Boltzmann constant for phonons. This result
conductivity decreases as the film thickness decreases due to has also been derived by Casimir,1 who treated a perfectly
the increasing importance of surface roughness scattering.4–7 diffusive surface (p ¼ 0) as if it absorbed all phonons inci-
The influence of surface roughness is usually modeled dent upon it and reemitted them at a rate depending on the
by using either a constant5–9 or frequency-dependent4,10 absolute temperature of the surface according to the theory
specularity parameter, p, representing the probability of of blackbody radiation. We will show in this paper that the
specular phonon boundary scattering. As an example, for same Landauer approach used to describe diffusive phonon
perfectly diffusive scattering p ¼ 0 and for perfectly specular transport can be simply extended to accurately describe bal-
scattering p ¼ 1. The quantitative comparison between the listic and quasi-ballistic transport.

0021-8979/2012/111(9)/093708/6/$30.00 111, 093708-1 V

C 2012 American Institute of Physics

Downloaded 10 Sep 2013 to 128.46.221.64. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
093708-2 Jeong, Datta, and Lundstrom J. Appl. Phys. 111, 093708 (2012)

 
The Landauer approach is widely used to treat ballistic kB2 TL p2
electron18 and phonon19,20 transport in nanostructures. As Kph ¼ hMph ihhTph ii; (2c)
3h
recently shown for electrons21 and phonons,22 the method can
also be applied to diffusive transport in bulk materials. The Ðwhere the averagehXifor any quantity X is defined as hXi 
approach reduces to the BTE for diffusive transport but is X Wph dðhxÞ while the average hhXii is defined as hM Xi=
more physically transparent and also provides some computa- hMi. From Eq. (2c), the expressions for the lattice thermal
tional advantages. Section II presents a brief summary of the conductivity, jph ¼ Kph ðL=AÞ, can be expressed as
formalism for phonon transport as discussed in detail in Ref.  2 
k TL p2
22. In Sec. III, results for the thermal conductivity of bulk Si jph ¼ B hMph =Aihhkph iieff ; (3a)
3h
are presented and compared to recent molecular dynamics
simulations. In Sec. IV, the in-plane and cross-plane thermal where A is the cross-sectional area of the conductor and
conductivities are discussed. Sections III and IV also illustrate hhkph iieff is the effective MFP given as
a general approach for extracting a well-defined mean-free-
1 1
path for phonons from measured thermal conductivity data. In hhkph iieff ¼ hhTph iiL ¼ hhðk1
ph þ L Þ ii: (3b)
Sec. V, we discuss the ballistic limit of phonon transport and
relate the Landauer expressions to the well-known Casimir From Eq. (2c), the ballistic thermal conductance per area
formula.1 Finally, our conclusions are summarized in Sec. VI. Kph BAL =A can be defined as
 2 
kB TL p2
II. APPROACH Kph BAL =A ¼ hMph =Ai; (3c)
3h
This paper is an application of the approach presented
in Ref. 22. The Landauer formula for heat current (IQ ) is so the thermal conductivity is expressed as
expressed as jph ¼ ðKph BAL =AÞhhkph iieff : (3d)
ð
1 1 Since hMph i and Kph BAL =A can be readily obtained from the
IQ ¼ hxÞðTph Mph Þhxðn1  n2 Þ;
dð (1a)
h 0 bandstructure, the hhkph iieff can be estimated by taking the
ratio of measured jph to the Kph BAL =A. Note that Eqs. (3a)
where Tph is the transmission at a given energy hx, Mph is and (3b) holds for all transport regimes. In the ballistic limit,
the number of conducting channels at a given energy, and n1 L  kph , hhkph iieff ¼ L and in the diffusive limit, L  kph ,
and n2 are Bose-Einstein distributions for the two contacts hhkph iieff ¼ hhkph ii with hhkph iibeing the average MFP in
across which heat flows.18 The transmission Tph is given as18 the diffusive limit. In the quasi-ballistic limit, the appropriate
Tph ¼ kph ðxÞ=ðL þ kph ðxÞÞ; (1b) effective MFP is given by Eq. (3b).
It was shown in Ref. 22 that given an accurate phonon
where kph ðxÞ is the mean-free-path for backscattering and L dispersion, Mph ðxÞ can be readily computed by a simple
the length of the conductor. Equation (1a) applies to the bal- numerical technique—the “band counting” method. To eval-
listic limit (L  kph ðxÞ) for which Tph ¼ 1; the quasi- uate Mph ðxÞ in this work, a full band description of phonon
ballistic regime (L  kph ðxÞ) for which Tph ¼ kph ðxÞ=ðL þ dispersion was obtained from the Tersoff24 interatomic pair
kph ðxÞÞ as well as to the diffusive limit (L  kph ðxÞ) for potential model within the General Utility Lattice Program
which Tph ¼ kph ðxÞ=L. In the diffusive limit, it was shown (GULP).25 Then it is straightforward to compute Kph BAL =A.
that the Landauer expression for lattice thermal conductivity
is essentially equivalent to the conventional expression from III. BULK THERMAL CONDUCTIVITY
BTE.22 Note that the product Tph Mph is proportional to the In this section, the phonon thermal conductivity of bulk
well-known “transport distribution” for electrons.23 Si will be evaluated and compared to a recent MD calcula-
For a small temperature gradient (DT), thermal conduct- tion.2 Figure 1(a) displays the energy-resolved Mph ; kph ; and
ance (Kph ¼ IQ =DT) is Wph at 300 K for bulk Si. Note that the entire phonon disper-
 2 ð sion participates in conduction since Wph is almost constant.
k TL p2 þ1
Kph ¼ B dðhxÞðTph Mph ÞWph ; (2a) This is in sharp contrast to the case of electrons, in which the
3h 0
important energies are near the bottom of the band. The bal-
where kB2 TL p2 =3h is the quantum of thermal conductance listic thermal conductance, Kph BAL , is readily evaluated from
with TL being the lattice temperature and Wph is a “window Eq. (3c). By comparing Kph BAL to the measured conductivity,
function” given by22 jph ,26 the average MFP, hhkph ii, is readily extracted from
   Eq. (3d). The results in Fig. 1(b) show that hhkph ii ’ 135 nm
3  hx 2 @n0 at T ¼ 300K. (As will be discussed in Sec. V, when compar-
Wph ðhxÞ ¼ 2  : (2b)
p kB T L @ðhxÞ ing this result to the conventional mean-free-path, lph , it is
important to remember that the Landauer mean-free-path (or
The integral of the window function, Wph , from 0 to 1 is 1, mean-free-path for backscattering) is 4/3 times longer.)22
just like the derivative of the Fermi function ð@f0 =@EÞ To examine how heat is conducted by phonons with dif-
appearing in the expression for electrical conductivity. The ferent mean-free-paths, we need expressions for the spectral
thermal conductance, Eq. (2a), can be also expressed as phonon mean-free-path for backscattering,22

Downloaded 10 Sep 2013 to 128.46.221.64. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
093708-3 Jeong, Datta, and Lundstrom J. Appl. Phys. 111, 093708 (2012)

FIG. 2. The cumulative thermal conductivity, jph , as a function of energy is

plotted for diffusive (scattering) and ballistic (no scattering) cases. For the
ballistic case, all energy channels equally contribute to jph . With scattering,
low-energy channels mainly contribute to jph because high-energy phonons
have very short mean-free-paths.

adjusted to match measured jph vs. T (Fig. 1(b)) rather than

to obtain the same MFP distribution as the MD calculations.
Our results obtained from full phonon dispersion (blue solid
line) are in good agreement with the recent MD simulations2
—50% of the heat conduction is attributed to phonons with
MFP >1 lm. The reason is that the phonons with MFP
> 1 lm are low-energy acoustic phonons near the Brillouin
FIG. 1. (a) For bulk Si, energy-resolved number of conducting channels zone center which do not suffer a lot from umklapp scatter-
(Mph ), window function (Wph ), and the mean-free-path for backscattering
ing, which compensates for the fact that there is a small per-
(kph ) are plotted at 300 K. Note that the entire spectrum of Mph participates
in conduction since Wph is almost constant. The mean-free-path shows that centage of low energy channels. As shown in Fig. 1(b), the
low-energy acoustic phonons have long kph since they do not suffer a lot average MFP for bulk Si (hhkph ii ¼ jph =ðKph BAL =AÞ) is
from umklapp scattering. The spectral kph are calculated based on the relaxa- about 0.135 lm at room temperature. Therefore, the com-
tion time approximation for umklapp scattering,27 point defect scattering,28
and crystalline boundary scattering.26 (b) The thermal conductivity jph (left
monly used average MFP does not give a clear picture of
axis) and the average MFP hhkph ii (right axis) are plotted as a function of which phonons carry the heat since the hhkph ii includes a sig-
temperature. Experimental results are obtained from Ref. 26. Good agree- nificant number of high energy modes with very small MFPs.
ment between calculation and experiment is observed.

kph ðxÞ ¼ ð4=3Þtph ðxÞsph ðxÞ ¼ ð4=3Þlph ðxÞ; (4)

where tph ðxÞis the spectral phonon group velocity at fre-

quency x, sph ðxÞ the phonon momentum relaxation time,
and lph ðxÞ ¼ tph ðxÞsph ðxÞ. For sph ðxÞ, the relaxation time
approximation (RTA) is used for umklapp scattering,27 point
defect scattering,28 and crystalline boundary scattering
C=T
rates:26 s1
u ¼ Bx Te
2
; s1 1
d ¼ Dx ; and sb ¼ htðxÞi=
4

ðF  lÞ, respectively. Typical parameters to fit the jph vs. T

for bulk Si are used:22 B ¼ 2.8  1019 s/K, C ¼ 140 K,
F ¼ 0.4, and l ¼ 7.16  103 m. Parameter D ¼ 1.32  1045
s3 is analytically determined from the isotope concentration, FIG. 3. We plot the computed cumulative thermal conductivity (jph ) as a
so the value given in Ref. 26 is used for bulk Si. Figure 1(b) function of the MFP for backscattering (kph ) using three different phonon
dispersion models: full phonon dispersion, a sine-type dispersion, and a
shows that the resulting fit is excellent. Figure 2 shows the Debye model. The computed results are compared to the MD simulations
cumulative distribution function of thermal conductivity as a obtained from Ref. 2. The MD simulation is plotted while taking into
function of energy with and without scattering. It can be account the difference between a conventional MFP for scattering (lph ) and
seen that all energy channels contribute to the ballistic ther- the MFP for backscattering, i.e., kph ¼ ð4=3Þlph . Our results with full phonon
dispersion (solid line) is in good agreement with the recent MD simulations,2
mal conductance. When scattering is included, however, which showed 50% of the heat conduction is attributed to phonons with
high energy channels contribute very little to jph because MFP >1 lm. The MFP distribution is not correctly predicted by a simple
high energy phonons have very short MFPs. sine-type dispersion model (dotted line) or a Debye model (dashed line).
Inset: Computed thermal conductivity jph vs. temperature is plotted for the
To find how the heat is carried by phonons with different three phonon dispersion models and is compared to experiment.26 Note that
MFPs, the spectral analysis (cumulative jph vs. kph ) is pre- regardless of the phonon dispersion model used, we can fit well the experi-
sented in Fig. 3. Note that the scattering parameters are mental data by adjusting scattering parameters.

Downloaded 10 Sep 2013 to 128.46.221.64. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
093708-4 Jeong, Datta, and Lundstrom J. Appl. Phys. 111, 093708 (2012)

To illustrate the effect of phonon dispersion model, two thermal conductivity data of SOI film, yielding that the D for
simple approximations of phonon dispersion are assumed. SOI film is 2 larger than D for bulk Si. (Note that for bulk
The first is a Debye model, x ¼ ts k; where ts is sound Si we used the parameter D ¼ 1.32  1045 s3 analytically
velocity and k is a wave vector, and the second is a sine-type determined from the isotope concentration.)28 Assuming that
dispersion model, x ¼ x0 sinðpk=2k0 Þ;where x0 is the maxi- the length of conductor, L, is much greater than kph ðxÞ,
mum phonon frequency and k0 is the Debye cutoff wave vec- transport is diffusive and Tph ¼ kph ðxÞ=L in Eq. (2a). For
tor.29 As shown in the inset of Fig. 3, jph vs. temperature these calculations, we retain the bulk phonon dispersions.
computed from the two simple models almost overlap with The results are shown in Fig. 4. Previous experimental data
the results of full phonon dispersion and match well the are in good agreement with the calculations, and considering
measured conductivity. But this requires an increase in the B point defects produces a better fit to thicker Si layers in SOI.
parameter for umklapp scattering by a factor of 4.5 for the Next, we examine the cross-plane phonon thermal con-
Debye model and a factor of 3 for the sine-type model. It can ductivity of thin Si film layers. For cross-plane thermal trans-
be clearly seen that while the simple phonon models can fit port in thin films of thickness less than 1 lm, quasi-
the measured thermal conductivity by adjusting fitting param- ballistic transport becomes important. Therefore, the
eters, the MFP distribution of the simple models does not assumption of diffusive transport no longer holds. The cross-
agree well with that of MD simulation. Thus, the question of plane thermal conductivity is conventionally computed with
how phonons with different MFPs carry the heat, which is the BTE by including a phonon-boundary scattering time.
important to know when designing thermoelectric devices is For example, sb ¼ L=ð2tÞ (Refs. 13–15) is commonly used,
not correctly addressed by the simple phonon models. but the physical significance is unclear. In contrast to the pre-
vious work, we have included only scattering processes
IV. IN-PLANE AND CROSS-PLANE THERMAL within the thin film—no interface resistances are considered.
CONDUCTIVITIES FOR THIN FILMS In a Landauer picture, we are assuming ideal reflection-less
contacts at the top and bottom of the film, so our calculations
Having verified that a simple Landauer model with full
will provide an upper limit that does not consider the inter-
phonon dispersion accurately reproduces the results of MD
face resistances that may occur in practice. For the transmis-
simulations,2 we turn next to heat transport in thin Si films.
sion, the expression, Tph ¼ kph ðxÞ=ðL þ kph ðxÞÞ, is used to
For thin Si films, phonon boundary scattering significantly
describe quasi-ballistic transport.
influences the thermal conductivity. In this section, phonon
Figure 4 shows our calculation of the cross-plane ther-
thermal conductivity of thin Si film layers along the in-plane
mal conductivity which is defined as jph ¼ Kph ðL=AÞ com-
and the cross-pane direction will be evaluated as a function
pared to a recent experimental observation.12 In agreement
of Si layer thickness.
with the one available measurement,12 our calculations show
For in-plane thermal conduction in thin films, we
a much reduced thermal conductivity in the cross-plane. This
consider the surface roughness of boundaries with a
result occurs even though we have not included possible
frequency-dependent specularity parameter. The scattering
interface resistances and can be understood from Eq. (3a). In
time reduction due to boundary scattering in the thin film has
the ballistic limit, the effective mean-free-path approaches
been examined by a solution of the BTE (Refs. 4 and 30)
the thickness of the film. Note that a better fit for the cross-
and the corresponding MFP of thin film (kph;thin ) compared
plane measurement could be obtained by increasing the
to that of bulk Si (kph;bulk ) was given as
parameter, D, in the point defect scattering rate formula (we
 ð  
3ð1  pÞ 1 1 1
kph;thin ðxÞ ¼ kph;bulk ðxÞ 1  
2d 1 t3 t5

1  expðdtÞ
 dt ; (5a)
1  p expðdtÞ

where d ¼ ð4=3ÞdSi =kph;bulk with dSi being the thickness of Si

thin film, and the specularity parameter p is given by10
!
16p3 g2SOI
p ¼ exp  ; (5b)
K2ph

where Kph is the phonon wavelength and gSOI is the surface

roughness which was estimated to be between 0.2 and 1 nm
for SOI wafers.31 To consider the impact of additional FIG. 4. Thermal conductivity (jph ) vs. silicon layer thickness at room tem-
perature is plotted. In-plane experimental data4–7 and calculations are shown
imperfections associated with the SOI wafers such as point by open squares and blue lines, respectively. For our calculations, a surface
defects, stacking faults, and dislocations, we use an approxi- roughness of 0.5 nm is used, which is a typical value for SOI wafers. We
mate formula since it is not clear which type of defect is assume that the point defect scattering rate for SOI wafer is 2 larger than
that of bulk Si. The results with the point defects (blue solid line) give a bet-
dominant. The same point defect scattering rate formula ter fit for thicker Si layers. The cross-plane experimental data12 and calcula-
used for bulk Si (s1
d ¼ Dx ) was used except that the pre-
4
tions are shown by filled circle and red lines, respectively. It is assumed that
factor, D, for SOI film is adjusted to fit the experimental the Mph ðxÞ for the thin film is the same as that for bulk Si.

Downloaded 10 Sep 2013 to 128.46.221.64. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
093708-5 Jeong, Datta, and Lundstrom J. Appl. Phys. 111, 093708 (2012)

used the same D as for the in-plane calculation, 2 larger in-plane direction as well as to quasi-ballistic transport in
than D for bulk Si), but the comparison to experiment is thin films along the cross-plane direction. Although the BTE
clouded by uncertainties in possible interface resistances. with an additional boundary scattering (sb ) succeeds in mod-
Nevertheless, the calculation shows the reduction in thermal eling, the bulk and thin film experiments, including sb , can-
conductivity that should be expected for the thin film itself. not accurately describe the ballistic limit (i.e., the Casimir
Next, we turn to the question of which phonons play the limit, q ¼ rðTH4  TC4 Þ) for which Majumdar used the
dominant role in the heat conduction in thin films. Figure 5 is EPRT.16 In this section, we show that the Landauer approach
a plot of k50% and hhkph iieff vs. thickness of thin Si films at reduces to the Casimir limit under the appropriate condi-
room temperature along the in-plane and the cross-plane trans- tions. We also relate the MFP for backscattering in the Lan-
port directions. Here, k50% is the MFP at which the cumulative dauer model to the commonly used MFP for scattering and
jph is equal to 50%, and the effective MFP, hhkph iieff , is discuss some limitations of the Landauer approach.
extracted according to Eq. (3d). The cross-plane direction dis- In the ballistic limit (Tph ¼ 1) and at a temperature
plays about 2 smaller k50% and hhkph iieff than the in-plane much lower than the Debye temperature where Mph ðxÞ is
direction. It can be also seen that hhkph iieff is always less than given as Mph ðxÞ ¼ Að3x2 =4pt2s Þ with ts being the velocity
k50% for the same reason as for the bulk—hhkph iieff places too of sound. The heat flux (q) can be expressed from Eq. (1a)
for a small temperature gradient (DT) as
much emphasis on the high energy modes with very small
MFPs. However, the difference between hhkph iieff and k50% ð  2  
IQ 1 1 3x dn @x
decreases as the thickness of the thin film decreases. This q¼ ¼ dðhxÞ hx DT ; (6a)
A h 0 4pt2s dx dT
occurs for both in-plane and cross-plane conduction but for
different reasons. For cross-plane conduction, as the thickness Ð1
where n ¼ 1=ðex  1Þ and x  hx=kB T: Using 0 x4
of thin films (i.e., L in Eq. (1b)) decreases, the transmission ðdn=dxÞdx ¼ 4p4 =15; the heat flux of Eq. (6a) is given as
Tph in Eq. (1b) for phonons with short MFPs increases more
rapidly than that for phonons with large MFPs. This results in q ¼ rDðT 4 Þ; (6b)
a decrease of k50% . For in-plane conduction, low-energy
acoustic phonons, which have large MFPs (kph >1 lm) where r ¼ p2 kB4 =40h3 t2s is the Stefan-Boltzmann constant
leading to the large k50% values in the bulk, suffer a lot from for phonons.17 This result shows that the Landauer approach
boundary scattering in thin films with thickness <1 lm, and correctly reduces to the Casimir result in the ballistic limit
therefore, the k50% decreases rapidly with film thickness. with the Debye approximation. Majumdar16 showed that the
Finally, we note that for the in-plane direction, about 50% of Casimir limit can be also obtained from the EPRT and that
the heat conduction is carried by phonons with a mean-free- the use of the Fourier law causes significant errors for con-
path greater than the thickness of the thin film when the thick- duction across the film. The EPRT, however, does not accu-
ness is smaller than 0.2 lm. rately predict the thermal conductivity reduction for both the
in-plane and the cross-plane conduction of Si thin films13
V. DISCUSSION due to the gray approximation. Finally, note that more gener-
So far, we have applied the Landauer approach to diffu- ally, for temperatures above the Debye temperature, the bal-
sive transport in bulk materials and thin films along the listic heat flux is q ¼ ðKph BAL =AÞDT, where the ballistic
thermal conductance is given by Eq. (3c).
It was also shown from the EPRT (Ref. 16) with the
gray approximation that the Fourier law can be used for all
transport regimes if the effective mean-free-path for scatter-
ing, hhlph iieff , is used instead of the commonly used average
MFP for scattering hhlph ii. According to Eq. (44) in Ref. 16

hhlph ii
hhlph iieff ¼ ; (7a)
1 þ ð4=3Þðhhlph ii=LÞ

where the hhlph ii is the average MFP for scattering that can
be obtained from a classical kinetic theory (i.e., jph ¼
ð1=3ÞCV ts hhlph ii with CV being the specific heat). A physical
interpretation of Eq. (7a), however, was not given: For exam-
FIG. 5. Along the in-plane (in) and the cross-plane (cross) transport direc-
tions, k50% and hhkph iieff are plotted as a function of the thickness of the thin
ple, where does the value of 4/3 in the denominator come
Si films at room temperature. Here, k50% is the MFP at which the cumulative from? We can easily show that Eq. (7a) follows directly
thermal conductivity (jph ) is equal to 50%, and the effective MFP, from Eq. (3b) of the Landauer approach. Using Eq. (4) in
hhkph iieff , is obtained from Eq. (3d). The cross-plane direction displays Eq. (7a) to convert from MFP to MFP for backscattering,
about 2 smaller k50% and hhkph iieff than the in-plane direction. hhkph iieff is
always less than k50% since hhkph iieff places too much emphasis on the high we find
energy modes with very small MFPs. Note that for the in-plane direction,  1
about 50% of the heat conduction is carried by phonons with a mean-free- 1 1
path greater than the thickness of the thin film when the thickness is smaller hhkph iieff ¼ þ ; (7b)
than 0.2 lm (blue symbol).
hhkph ii L

Downloaded 10 Sep 2013 to 128.46.221.64. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions
093708-6 Jeong, Datta, and Lundstrom J. Appl. Phys. 111, 093708 (2012)

which is precisely the Landauer result if we make the gray insights into phonon transport from the ballistic to diffusive
approximation, hhTph ii ¼ hhkph ii=ðL þ hhkph iiÞ. regimes in both nanostructures and bulk materials.
The value of 4/3 in Eq. (7a) comes from the difference
between the MFP for scattering and the MFP for backscatter-
ACKNOWLEDGMENTS
ing. The usual definition of mean-free-path is the average
distance that a carrier travels before scattering. In the Lan- This work was supported by MARCO Materials Struc-
dauer approach, kph ðxÞ is the mean-free-path for backscat- tures and Devices (MSD) Focus Center and computational
tering and has a specific meaning; it is the length at which services were provided by the Network for Computational
the transmission Tph ¼ kph ðxÞ=ðL þ kph ðxÞÞ drops to one- Nanotechnology (NCN). The authors also acknowledge illu-
half, and the inverse of the kph ðxÞ is interpreted as the proba- minating discussions with N. Mingo, E. Pop, R. Venkatasu-
bility per unit length that a positive flux is converted into a bramanian, K. Goodson, A. M. Marconnet, and J. Maassen.
negative flux. Following the proper definition of kph ðxÞ,21 it
can be shown that in 1D, kph ðxÞ ¼ 2lph ðxÞ, in 2D,
1
H. B. G. Casimir, Physica 5, 495–500 (1938).
2
A. J. Minnich, M. S. Dresselhaus, Z. F. Ren, and G. Chen, Energy Environ.
kph ðxÞ ¼ ðp=2Þlph ðxÞ, and in 3D, kph ðxÞ ¼ ð4=3Þlph ðxÞ: Sci. 2, 466–479 (2009).
We have shown that the Landauer approach provides a 3
J. Callaway, Phys. Rev. 113, 1046 (1959).
4
simple but physically insightful description of diffusive M. Asheghi, M. N. Touzelbaev, K. E. Goodson, Y. K. Leung, and S. S.
transport, quasi-ballistic transport, and ballistic transport, but Wong, J. Heat Transfer 120, 30 (1998).
5
M. Asheghi, Y. K. Leung, S. S. Wong, and K. E. Goodson, Appl. Phys.
it does have limitations. For example, for problems like Lett. 71, 1798–1800 (1997).
cross-plane thermal transport, we made the assumption of 6
Y. S. Ju and K. E. Goodson, Appl. Phys. Lett. 74, 3005–3007 (1999).
ideal contacts (i.e., that are reflection-less and that maintain 7
W. Liu and M. Asheghi, J. Appl. Phys. 98, 123523 (2005).
a near-equilibrium thermal population of phonons). The role
8
P. Chantrenne, J. L. Barrat, X. Blase, and J. D. Gale, J. Appl. Phys. 97,
104318 (2005).
of contacts is problem-specific and should be considered on 9
Y. F Zhu, J. S Lian, and Q. Jiang, Appl. Phys. Lett. 92, 113101 (2008).
a case-by-case basis. Problems involving space and time de- 10
J. M Ziman, Electrons and Phonons: The Theory of Transport Phenomena
pendent transport and multi-dimensional transport tend to be in Solids (Oxford University Press, New York, 1960).
11
M. Maldovan, J. Appl. Phys. 110, 034308 (2011).
easier to handle with the Boltzmann equation, but for 1D, 12
P. E. Hopkins, C. M. Reinke, M. F. Su, R. H. Olsson, E. A. Shaner, Z. C.
steady-state transport, the Landauer approach provides sig- Leseman, J. R. Serrano, L. M. Phinney, and I. El-Kady, Nano Lett. 11,
nificantly more physical insight as well as computational 107–112 (2010).
13
advantages in computing the transport distribution (or num- D. P. Sellan, E. S. Landry, J. E. Turney, A. J. H. McGaughey, and C. H.
Amon, Phys. Rev. B 81, 214305 (2010).
ber of channels, MðxÞ). 14
D. P. Sellan, J. E. Turney, A. J. H. McGaughey, and C. H. Amon, J. Appl.
Phys. 108, 113524 (2010).
15
A. J. H. McGaughey, E. S. Landry, D. P. Sellan, and C. H. Amon, Appl.
VI. SUMMARY AND CONCLUSION Phys. Lett. 99, 131904 (2011).
16
A. Majumdar, ASME Trans. J. Heat Transfer 115, 7–16 (1993).
In this paper, we showed that a simple Landauer model 17
E. T. Swartz and R. O. Pohl, Rev. Mod. Phys. 61, 605 (1989).
18
in the diffusive limit with a full phonon dispersion reprodu- S. Datta, Electronic Transport in Mesoscopic Systems (Cambridge Univer-
ces the results of more sophisticated molecular dynamics sity Press, 1997).
19
D. E. Angelescu, M. C. Cross, and M. L. Roukes, Superlattices Micro-
simulations of phonon transport in bulk Si. For thin Si films, struct. 23, 673–689 (1998).
the same approach also accurately describes the measured 20
N. Mingo, Phys. Rev. B 68, 113308 (2003).
21
in-plane (diffusive transport) and cross-plane (quasi-ballistic C. Jeong, R. Kim, M. Luisier, S. Datta, and M. Lundstrom, J. Appl. Phys.
transport) thermal conductivities, jph vs. thickness of the Si 107, 023707 (2010).
22
C. Jeong, S. Datta, and M. Lundstrom, J. Appl. Phys. 109, 073718 (2011).
layer. The spectral analysis of cumulative thermal conductiv- 23
G. D. Mahan and J. O. Sofo, Proc. Natl. Acad. Sci. U.S.A. 93, 7436–7439
ity as a function of a MFP demonstrates that the commonly (1996).
used average MFP should be used with caution because it
24
J. Tersoff, Phys. Rev. B 39, 5566 (1989).
does not convey which phonons mainly contribute to the
25
J. D. Gale and A. L. Rohl, Mol. Simul. 29, 291–341 (2003).
26
M. G. Holland, Phys. Rev. 132, 2461 (1963).
heat conduction. In the ballistic limit and with the Debye 27
M. Asen-Palmer, K. Bartkowski, E. Gmelin, M. Cardona, A. P. Zhernov,
approximation, the Landauer model yields the Casimir limit, A. V. Inyushkin, A. Taldenkov, V. I. Ozhogin, K. M. Itoh, and E. E. Hal-
the blackbody radiation law for phonons. The results pre- ler, Phys. Rev. B 56, 9431 (1997).
28
P. G. Klemens, Proc. Phys. Soc. London A68, 1113 (1955).
sented here shed new light on phonon transport in Si struc- 29
C. Dames and G. Chen, J. Appl. Phys. 95, 682–693 (2004).
tures and also show that the Landauer approach provides a 30
E. H. Sondheimer, Adv. Phys. 1, 1–42 (1952).
simple and useful computational approach that gives new 31
W. P. Maszara, J. Electrochem. Soc. 138, 341 (1991).

Downloaded 10 Sep 2013 to 128.46.221.64. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions