A “2-omega” technique for measuring

anisotropy of thermal conductivity

Cite as: Rev. Sci. Instrum. 83, 124903 (2012); https://doi.org/10.1063/1.4770131
Submitted: 24 September 2012 . Accepted: 22 November 2012 . Published Online: 13 December 2012

Ashok T. Ramu, and John E. Bowers

ARTICLES YOU MAY BE INTERESTED IN

Thermal conductivity measurement from 30 to 750 K: the 3ω method

Review of Scientific Instruments 61, 802 (1990); https://doi.org/10.1063/1.1141498

, and methods for measurements of thermal properties

Review of Scientific Instruments 76, 124902 (2005); https://doi.org/10.1063/1.2130718

A 3 omega method to measure an arbitrary anisotropic thermal conductivity tensor

Review of Scientific Instruments 86, 054902 (2015); https://doi.org/10.1063/1.4918800

Rev. Sci. Instrum. 83, 124903 (2012); https://doi.org/10.1063/1.4770131 83, 124903

© 2012 American Institute of Physics.

 REVIEW OF SCIENTIFIC INSTRUMENTS 83, 124903 (2012)

A “2-omega” technique for measuring anisotropy of thermal conductivity

Ashok T. Ramua) and John E. Bowers
Department of Electrical and Computer Engineering, University of California, Santa Barbara, Santa Barbara,
California 93117, USA
(Received 24 September 2012; accepted 22 November 2012; published online 13 December 2012)
A popular method of measuring the thermal conductivity of thin films and substrates, the “3-omega”
method, is modified to yield a new technique for measuring the anisotropy in thermal transport in bulk
materials. The validity of the proposed technique is established by measuring the thermal conductiv-
ity of strontium titanate, which is expected to be isotropic because of its cubic unit cell. The technique
is then applied to rutile TiO2 . The analysis of experimental results on (100) and (001) TiO2 reveals
that the anisotropy is a function of the crystalline quality, as quantified by the effective thermal con-
ductivity obtained through conventional “3-omega” measurements. The advantages of the proposed
technique are similar to those of the standard “3-omega” method, namely the simplicity of sample
preparation and measurement, and negligible errors due to radiation because of the small volume of
material being heated. For anisotropy determination, the proposed technique has the additional ad-
vantage that a single sample is sufficient to determine both components of the thermal conductivity,
namely the values in and perpendicular to the plane of cleavage. This is significant for materials in
which there is a large variation in the crystalline quality from sample to sample. For such materials, it
is unreliable to use two different samples, one for measuring the thermal conductivity in each direc-
tion. Experimental data are analyzed using a 3D Fourier-series based method developed in this work.
The proposed method determines each component of the thermal conductivity with an estimated ac-
curacy of about 10%. © 2012 American Institute of Physics. [http://dx.doi.org/10.1063/1.4770131]

I. INTRODUCTION conductivity of bulk materials about 500 μm thick, which

information is inaccessible through conventional 3ω mea-
Anisotropy in thermal transport remained a rarely ex-
surements. Instead of a single metal line serving both as the
plored topic1 until the advent of thermoreflectance tech-
heater and as the thermometer, two separate parallel lines are
niques. Since then, thermoreflectance has been frequently
used (Fig. 1); the measured temperature oscillation (hereafter
used in thin film and substrate anisotropy determination.2–5
abbreviated as TO) can then be expected on physical grounds
Another technique for thermal conductivity measurement, the
to be more sensitive to the thermal conductivity in the
3ω method,6 has gained popularity of late because of the
plane of the lines than to the value perpendicular to it. This
simplicity of the measurement. The two methods have been
expectation is borne out by 2D Fourier-domain finite element
shown7 to give results consistent within measurement error
method (FEM) simulations and by analytical expressions for
when the required experimental conditions for both methods
the measured TO.9 We show that the analytical expression
are met. In the conventional 3ω method, a metal line with
as well as the 2D FEM solution is consistent with a 3D
four contact pads (two outer pads for sourcing current, and
Fourier-series solution presented in this work, thus ruling out
two inner ones for measuring voltage) is used both as the heat
end effects due to the finite heater/thermometer length. We
source, through Joule heating by a sinusoidal current source,
will refer to the proposed method as the “anisotropic 2ω”
and as the thermometer, through its temperature coefficient
technique for reasons that will become clear later in the paper.
of resistance. The desired thermal properties of the under-
The technique is first applied to a control sample, stron-
lying material appear in the third harmonic of the voltage
tium titanate, which is expected to be isotropic due to its cubic
across the inner (voltage) pads, hence the name of the method.
unit cell. We then investigate the thermal conductivity of ru-
Anisotropy in thin films of low thermal conductivity on sub-
tile TiO2 , which has a tetragonal unit cell. The ratio of the
strates of higher thermal conductivity can be determined by
conductivity in the [001] direction to that in the [100] direc-
the standard 3ω method by varying the heater width.8 How-
tion is found to lie between 1.73 and 2.36 for the three samples
ever, varying the heater width in the conventional 3ω method
whose anisotropies were studied using the new technique. We
is not a feasible means of extracting anisotropy in the thermal
find a correlation between the anisotropy ratio and the crys-
conductivity of thick films and substrates. This is because the
talline quality, with the latter quantified by the average of the
measured 3ω voltage has only a logarithmic dependence on
thermal conductivities in the two directions.
this parameter, and also because the standard 3ω method is
sensitive to interfacial thermal impedances.
In this paper, we modify the 3ω method to yield a
novel technique for extracting the anisotropy in the thermal II. THE THEORY
Figure 1 shows the geometry of the proposed technique.
a) E-mail: ashok.ramu@gmail.com. For the samples considered in this work, the axes x, y, and z

0034-6748/2012/83(12)/124903/8/$30.00 83, 124903-1 © 2012 American Institute of Physics

124903-2 A. T. Ramu and J. E. Bowers
FIG. 1. The geometry of the proposed technique (not to scale): (a) profile
view: the heater and thermometer lines lie in the x-y plane and are parallel to
the y-axis, which goes into the plane of the paper. (b) Plan view: the lines are
equally long and parallel to each other, with D L.
may be aligned with the principal crystallographic directions
of the lattice, and the heater/thermometer lines are always
aligned parallel to the y-axis. For each sample considered,
the assignment of crystallographic axes to x, y, and z will
be explicitly specified. It is to be noted that expressions
(1)–(4) below apply only to the case of diagonal, possibly
anisotropic thermal conductivity matrices. That this is the
case for tetragonal crystals cannot be proven rigorously by
symmetry arguments; however, Ref. 12 notes that this is most
likely the case for cubic and tetragonal crystals. We therefore
denote the components of the thermal conductivity matrix in
the directions x, y, and z as kx , ky , and kz , respectively, and the
off-diagonal elements of the matrix are assumed zero.
In the geometry of Fig. 1, it can be shown that the ther-
mometer temperature oscillation is selectively sensitive to kx ,
and relatively insensitive to ky (see below) and to kz (Ref. 9).
Additionally, in contrast to the standard 3ω method, it has
been shown9 by FEM simulations that the two-line scheme
of Fig. 1 is not sensitive to interfacial thermal impedances.
Physically, this is because the temperature gradient is approx-
imately perpendicular to the interface beneath the heater and
approximately parallel to the interface beneath the thermome-
ter. These observations are essential to this work.
The 3ω method, if performed conventionally on an
anisotropic substrate (with a single line used as both the heater
and thermometer), yields the following expression9, 10 for the
TO profile on the heater/thermometer:
   
phase P 1 kx
Theater = ln + a . (1)
π Lk̃ 2 2ωCv (wheater /2)2
Rev. Sci. Instrum. 83, 124903 (2012)
Here, ω = 2π f is the angular frequency of the heating current,
phase
Theater is the component of the TO at a frequency 2ω that is in-
phase with the applied power of magnitude P, L is the length
of the heater/thermometer, kx is the thermal conductivity in
the plane of the wire and perpendicular to it, wheater is the
heater width, Cv is the volumetric heat capacity, a is a con-
stant equal to 0.93 in theory,11 and k̃ is an “effective” thermal
conductivity in the two directions perpendicular to the heater
line, as discussed below. In theory, both kx and k̃ (and hence
kz , from Eq. (4)) may be determined from Eq. (1) for the stan-
dard 3ω method, by varying the frequency ω and the heater
width. In practice, this does not work because the standard 3ω
method is sensitive to interfacial thermal impedances, which
cause the value of a to deviate from the theoretical value.
phase
Thus, k̃ alone may be determined from the slope of Theater
as a function of the logarithm of ω.
For the geometry of Fig. 1, in which separate heater and
thermometer lines of equal length parallel to each other are
used, it can be shown9 that
P
Ttherm =
π Lk̃ wtherm wheater
 wheater /2  wtherm /2
· K0 (q (D + τ − x)) dτ dx.
x=−wheater /2 τ =−wtherm /2
(2)
In addition to the definitions above, wtherm is the width
of the dedicated thermometer line, K0 is the zeroth-order
modified Bessel function of the second kind, D is the
heater-thermometer separation, and Ttherm is the frequency-
dependent, complex temperature oscillation amplitude of the
thermometer, whose real and imaginary parts give, respec-
tively, the components at frequency 2ω in phase and in
quadrature with the applied power. Only the in-phase com-
phase
ponent, Ttherm is of interest to us, since only this component
is selectively sensitive to the thermal conductivity component
kx . The 1/|q| is the characteristic length scale over which the
surface temperature oscillation profile decays to 0, where q is
given by

j 2ωCv
q= . (3)
kx
These expressions are derived in Ref. 9, where it is also
proved that modified Bessel functions of order higher than
0 must vanish, despite the lack of cylindrical symmetry in the
temperature oscillation profile in an anisotropic substrate. For
small anisotropies, it is further shown that k̃ approximately
equals (kx + kz )/2 where kz is the component of the thermal
conductivity in the z-direction, i.e., perpendicular to the plane
in which the lines lie.
Reference 10 states an expression for k̃ valid for arbi-
trary anisotropy. We verified this expression using FEM sim-
ulations for the anisotropies considered in this work, and we
will use this expression henceforth:

k̃ = kx · kz . (4)
We further define, for later convenience, the following
quantity η, called the anisotropy ratio or simply anisotropy.
124903-3 A. T. Ramu and J. E. Bowers
2 deposition of 20 nm Ti followed by 200 nm Pt. Consider
TO in phase with applied power (K) Fourier−domain FEM simulation
Analytical solution, Eq. 2
3D Fourier−series, L=2 mm
1.5
3D Fourier−series, L=0.5 mm
1 the applied heating power) is of interest to us. In order to
0.5
0 resistance oscillation at frequency 2ω gives rise to a small
−0.5 1
2 3
10 10 10 10
Freq. (Hz)
FIG. 2. Comparison of the theoretical approaches used in this work for a typ-
ical parameter set. The 2D analytical expression Eq. (2) agrees very well with
the 2D FEM solution. Agreement is also excellent with the 3D Fourier series
solution of the Appendix, above a heater/thermometer length L = 0.5 mm.
Below this length, end effects cause the calculated TO to deviate from the 2D
approaches because the infinitely long heat-source approximation implicit in
the latter becomes less valid.
Our aim in this work is to determine this quantity.
kz
η= . (5)
kx
Note that ky is the component parallel to the heater and
thermometer lines. 3D Fourier-series solutions showed that
with kx = 10 W/m-K, kz = 10 W/m-K, D = 100 μm, and
L = 2 mm (typical values for the samples considered in this
work), varying ky from 1 to 30 W/m-K had little effect (<5%)
on the TO of the thermometer in the configuration of Fig. 1.
Thus, ky can only be determined by having a second set of
lines on the same sample, perpendicular to the first. We do
not do so in this work. In the samples considered, either the
x, y directions are equivalent to each other, or one of them is
equivalent to the z direction.
Equation (2) agrees very well with 2D Fourier-domain
FEM simulations, as exemplified by Fig. 2. The agreement
with the 3D Fourier-series solution (see the Appendix) is also
shown; the agreement is good for L > 0.5 mm. End effects due
to the small heater length relative to the heater-thermometer
separation distort the measured TO for L < 0.5 mm. Since
we have used L = 2 mm for all experiments reported in this
work, the three theoretical approaches are essentially equiv-
alent. The simulation parameters for Fig. 2 are as follows:
kx = ky = 9 W/m-K, kz = 5.2 W/m-K, Cv = 2.902
× 106 J/m3 -K, D = 100 μm, and wtherm = wheater = 20 μm.
L is varied for the 3D Fourier-series solution, with P/L nor-
malized to 1 W/cm. The theoretical curves reported in the rest
of this work were generated by the 3D Fourier-series method,
and the fits to experiment were verified by the other two ap-
proaches.
III. EXPERIMENTAL METHODOLOGY
The 2 mm long heater and thermometer lines separated
by 100 μm are fabricated directly on the substrates by
photolithography followed by electron beam evaporation-
Rev. Sci. Instrum. 83, 124903 (2012)
a heater line excited at a frequency ω = 2π f. This creates
temperature oscillations in the substrate at a frequency 2ω.
Let the TO underneath the thermometer be given by Tmeas .
phase
Only the real part Tmeas (the component in-phase with
measure this, a dc “carrier” current IDC is passed through the
thermometer. Due to the temperature coefficient of resistance
α of the thermometer wire, the resistance has a component
that oscillates at a frequency 2ω, and the dc current times the
voltage at a frequency 2ω in addition to the dominant dc
component, IDC Rtherm where Rtherm is the resistance between
4
the inner terminals of the thermometer at the average temper-
ature. Rtherm is measured using the four-point probe method.
This dc component can easily be filtered out by means of a
capacitor, and the voltage component at 2ω measured using
a lock-in amplifier, hence the name “anisotropic 2ω” that we
use for the proposed technique in the rest of this work. In
the standard 3ω technique, the carrier current is necessarily
ac, and analog circuitry is required to filter out the dominant
voltage component.6
In a practical implementation, the thermometer is fed by
a 1 K resistor connected to a dc voltage source of 10–15 V.
The dc “carrier” voltage across the inner pads of the ther-
mometer line is recorded; IDC can then be calculated knowing
Rtherm . The heater of resistance Rheater between its inner termi-
nals, as measured by the four-point probe method, is fed by a
100  resistor, which is connected to the 5 V-rms ac voltage
output of a lock-in amplifier. The resistors help reduce noise
pick-up at the connection between the voltage sources and the
respective lines.
The lines are then probed and connected as shown in
Fig. 3. The inner pads of the thermometer are connected
directly to the two differential inputs of the lock-in amplifier
(model SRS830), with the latter operating in the “ac-coupled”
mode. The lock-in amplifier then eliminates the dc carrier
voltage by dropping it across a capacitor, and reads only the
differential ac voltage at a frequency equal to twice that of
the ac heating voltage. Only the component of the differential
phase
voltage in phase with the heating power, V2ω is of interest
to us. The frequency is swept from 20 Hz to 2000 Hz. The
phase
voltage V2ω , and the ac voltage across the inner pads of the
heater V1ω are both recorded at each frequency. The voltage
signals are converted into temperature oscillation amplitudes
using the following equation:
√ phase
β 2Lcm Rheater V2ω
phase
Tmeas = · · . (6)
αIDC Rtherm 2
V1ω
The TO has been scaled to an applied heating power of 1
W/cm input power/unit length, so that it can directly be
compared
√ to simulations and analytical expressions. The
factor 2 converts rms values to amplitudes (all ac voltages
mentioned here are rms values). Lcm is the length of the wires
in units of cm. In all our experiments, we used 2 mm lines, so
Lcm = 0.2. β is a correction factor to account for the fact that
the heater voltage, and hence the power, is measured across
the inner pads of the heater line, and thus does not span the
124903-4 A. T. Ramu and J. E. Bowers
FIG. 3. Schematic of the measurement circuit for thermal conductivity anisotropy determination.
true length of the line. In our experiment, the inner voltage
pads are 75 μm from each end of the 2 mm heater line, and
β = 2.00/1.85 = 1.081. The rest of the quantities in Eq. (6)
have been defined earlier in this section. We note that in our
implementation, Rheater and Rtherm are of the order of 100 .
In order to extract kx and kz separately, we first perform
a standard 3ω sweep and determine k̃ (the effective thermal
conductivity) from the slope of the in-phase component of the
TO as a function of the logarithm of the frequency, follow-
ing the usual procedure16 (see Eq. (1)). We then conduct the
experiment described above, and determine from simulations
the values of kx and kz that best fit the experimental data, under
the constraint that kx and kz be related through Eq. (4). In all
measurements reported in this work, the heater-thermometer
separation is 100 μm, which represents a good compromise
between signal level and the sensitivity to the thermal con-
ductivity component kx in the plane of the lines.
IV. EXPERIMENTAL RESULTS AND DISCUSSION
We first investigate strontium titanate (SrTiO3 or STO), a
crystal that is expected to be isotropic because of its cubic unit
cell. The x, y, and z axes are chosen to lie along the equivalent
100 directions. The value of k̃ is first determined from stan-
dard 3ω sweeps to be 10.2 W/m-K. Next, the anisotropic 2ω
technique is used to determine the two components separately.
We obtain an anisotropy ratio η (Eq. (5)) ranging from 1.0
to 1.17 depending on the location of the heater-thermometer
pair. This proves that the method is feasible, and that it is not
overly biased in favor of the direction in the plane of the mea-
surement over that perpendicular to it, or vice versa. Fig. 4
shows the fit for the heater-thermometer pair with η = 1. The
overall fit is excellent. The simulated in-phase component of
the TO is also shown for anisotropies of 0.67 and 1.5, from
which it can be seen that the experiment is sensitive enough
to rule out significant anisotropy where it is not present. For
the heat capacity (Cv ) of SrTiO3 , we use 2.75×106 J/m3 -K.14
Rutile TiO2 has a tetragonal unit cell, with lattice con-
stants a = b = 0.45933 nm and c = 0.29592 nm, where c
is along the [001] direction and a and b are along the equiv-
alent [010] and [100] directions. Rutile is known to have an
Rev. Sci. Instrum. 83, 124903 (2012)
anisotropic thermal conductivity. Reports on the ratio of con-
ductivity in the [001] direction to that in the [100] direction
vary from 1.4 to 1.7 at room-temperature.1 A value of 2 has
been reported at 68 ◦ C in Ref. 13, which, however, quotes a
value of 1.4 close to room-temperature. In this work, we mea-
sured both (100) and (001) oriented single crystal rutile TiO2 ,
supplied by MTI Corporation, Inc. Our results are at variance
with hitherto reported values for the conductivity ratio. For
the heat capacity (Cv ) of TiO2 , we use 2.902×106 J/m3 -K.15
Figure 5 shows experimental results on (001) TiO2 , sam-
ple number M3-001. The x and y axes are along the equiva-
lent [100] and [010] directions, while z is oriented along the
[001] direction. Standard 3ω sweeps yield an effective ther-
mal conductivity k̃ of 7.5 W/m-K. Three pairs of values of
kx and kz consistent with this k̃ are shown; of these, the best
fit is obtained with kx = 5.3 W/m-K and kz = 10.6 W/m-K,
and the conductivity ratio [001]/[100] is 2. This result (using
9 μm wide heater and thermometer lines) was consistent with
measurements using another pair of lines, each 18 μm wide
and on the same sample.
2
Expt.
TO in phase with applied power (K)
kx=kz=10.2 W/m−K
1.5
kx=8.3, kz=12.5 W/m−K
kx=12.5, kz=8.3 W/m−K
1
STO is isotropic
from this data set
0.5
0
−0.5 1 2 3 4
10 10 10 10
Freq. (Hz)
FIG. 4. Experimental data on SrTiO3 (this work) compared to theoretical
curves obtained through the 3D Fourier-series method (see the Appendix).
The widths of the heater and thermometer lines are 22 μm as measured by
optical microscopy. The data are consistent with η = 1, with the simulated
curves for η = 1.5 and η = 0.67 also shown for comparison. This rules out
significant anisotropy in the thermal conductivity of SrTiO3 as determined by
the anisotropic 2ω method.
124903-5 A. T. Ramu and J. E. Bowers Rev. Sci. Instrum. 83, 124903 (2012)

1 15

TO in phase with applied power (K)

TO in phase with applied power (K) Expt. 14
0.8 kx=5.3, kz=10.6 W/m−K
13
kx=5.9, kz=9.5 W/m−K
0.6 12 TiO2 M1−100
kx=4.8, kz=11.6 W/m−K Heater line parallel to [010]
11
0.4 k =10, k =4.7 W/m−K
x z
keff=7.5 W/m−K by 10 k =8, k =5.9 W/m−K
x z
0.2 std. 3−omega sweeps 9 k =9, k =5.2 W/m−K
x z
Expt2
8 Expt1
0 (a)
7 1 2 3
10 10 10
−0.2
Freq. (Hz)

−0.4 1 2 3 4
10 10 10 10 2

TO in phase with applied power (K)

Freq. (Hz)
FIG. 5. Experiment on (001) oriented TiO2 sample M3-001 is consistent
with a conductivity ratio [001]/[100] of 2. The line widths were 9 μm as
revealed by optical microscopy.
Different results were obtained on a different (001) TiO2
sample, labeled M1-001. As before, x and y are along the
equivalent [100] and [010] directions, while z is oriented
along the [001] direction. Standard 3ω sweeps established that
k̃ = 8.5 W/m-K, and the anisotropic 2ω technique was consis-
tent with the values kx = 5.5 W/m-K and kz = 13 W/m-K for
the individual components. The conductivity ratio [001]/[100]
was 2.36 for this sample.
For (100) oriented samples, the x and y axes are not
equivalent. Sample M1-100 had x = [001], y = [010] and
z = [100]. That the metal lines were indeed along [010] was
confirmed by X-ray diffraction experiments. Fig. 6(a) shows
the results of two standard 3ω sweeps on two separate lines,
from which we extracted an effective thermal conductivity
k̃ = 6.85 W/m-K. Three possibilities for the pair of compo-
nents kx and kz are shown. All three possibilities are consis-
tent with the slope of the graph, in accord with Eq. (4), but
none matches the intercept. This is due to interfacial ther-
mal impedance between the heater and the substrate, and
one of the main advantages of the anisotropic 2ω technique
is the elimination of interfacial effects. Fig. 6(b) shows that
kx = 9 W/m-K and kz = 5.2 W/m-K best fits the data from the
anisotropic 2ω experiment. The conductivity ratio [001]/[100]
was 1.73.
Measurements on another (100) TiO2 sample provided
yet another point of verification of the anisotropic 2ω tech-
nique. The metal line (or the y-axis) was aligned along the
[001] direction on the sample M3-100, as confirmed by the
X-ray diffraction (XRD) experiment. The x- and z- axes were,
respectively, along the [010] and [100] directions. These lat-
ter directions being equivalent, we may expect the anisotropy
ratio to be close to 1. This was indeed the case, with k̃ equal
to 5.1 W/m-K and the conductivity ratio [001]/[100] equal
to 1.17.
It may be observed that the TiO2 samples show a large
variation in effective thermal conductivity. The crystalline
quality is assumed to be correlated to the thermal conduc-
tivity. This assumption is partly confirmed by the XRD
“rocking curve” experiment, which shows a very broad, split
Expt.
kx=9 W/m−K, kz=5.2 W/m−K
1.5
kx=8 W/m−K, kz=5.9 W/m−K
k =10 W/m−K, k =4.7 W/m−K
x z
1
Anisotropy ratio=1.73 for M1−100
keff=6.85 W/m−K from std. 3−omega.
0.5
Heater/thermometer along [010]
0
(b)
−0.5 1 2 3 4
10 10 10 10
Freq. (Hz)
FIG. 6. (a) The standard 3ω sweep yields the effective thermal conductivity
k̃. Two different lines have been measured. (b) The anisotropic 2ω technique
yields each component of the thermal conductivity separately. Anisotropy
ratio = 1.73 for this sample. The heater and thermometer lines were each 18
μm wide, as measured under an optical microscope.
(200) peak, indicative of large dislocation density, in the
sample M3-100 that has the lowest k̃ (5.1 W/m-K). Figure 7
compares the “rocking curve” scans of M3-100 and M1-100.
In addition, sample M3-001 (k̃ = 7.5 W/m-K) has a broader
(002) peak, and hence more dislocations, than M1-001
(k̃ = 8.5 W/m-K). Figure 8 shows the conductivity ratio
[001]/[100] as a function of effective thermal conductivity for
the three TiO2 samples whose anisotropies were determined
in this work. Higher anisotropy is seen to be correlated
with higher effective thermal conductivity (and hence better
crystalline quality).
We conclude this section with a brief discussion of the
accuracy and limitations of the anisotropic 2ω technique. Er-
rors due to radiation are negligible due to the small volume
of the material being heated.16 Interfacial impedance has no
effect on the experiment. Its effect on the standard 3ω ex-
periment is non-zero but frequency independent; thus, the
slope of the component of the TO in phase with the applied
power, as a function of the logarithmic frequency, yields k̃
unambiguously.16 Errors in the measured temperature coef-
ficient of resistance α are the single largest source of error
in this experiment. However, each component of the ther-
mal conductivity scales identically with α; therefore, even
though errors in the measurement of α will affect k̃, they
will have no effect on the anisotropy ratio. We have measured
α = 2.4 × 10−3 /K for the metallization used in this work. The
reproduction of the thermal conductivities of SrTiO3 17 and
InSb (not shown) controls to within 10% suggests that α and
124903-6 A. T. Ramu and J. E. Bowers Rev. Sci. Instrum. 83, 124903 (2012)

nish a convenient means of measuring the heat capacities of

bulk materials.

FIG. 7. X-ray diffraction “rocking curve” scan of sample M3-100 (k̃ = 5.1
W/m-K) shows a very broad, split (200) peak indicative of large disloca-
tion density, compared to sample M1-100 (k̃ = 6.85 W/m-K). The crystalline
quality is correlated with the effective thermal conductivity.
V. CONCLUSIONS
We have demonstrated a new technique that builds on
the “3-omega” method to measure the anisotropy in thermal
transport in bulk materials with thermal conductivities less
than about 10 W/m-K. We have validated the technique us-
ing an isotropic control SrTiO3 , and applied it to anisotropic
rutile TiO2 . We estimate the method to be accurate to within
10%. We have found a correlation between the anisotropy and
crystalline quality (quantified by the effective thermal con-
ductivity in the principal crystallographic directions) of TiO2 .
Unlike most previous works, we can use the same sample for
measuring both components of the thermal conductivity of
TiO2 , namely that along the equivalent 010 directions and
that along the [001] direction. The range of applicability of the
proposed technique makes it particularly relevant to the char-
acterization of thermoelectric materials. The findings related
to thermal transport in TiO2 may have applications in resistive
random-access memory devices, where switching dynamics is
mainly controlled by Joule heating.20
ACKNOWLEDGMENTS

all k̃ reported in this work are accurate to within 10% as well.
Due to the progressive deterioration of both the signal level
and the sensitivity to the in-plane conductivity, the anisotropic
2ω technique is not indicated (at least in the form described
here) for materials with thermal conductivities much larger
than 10 W/m-K. Also, the heat capacity of the material
must be known with 5% accuracy, which is well within the
range of adiabatic calorimetry.18 Conversely, if the thermal
conductivity is known to be isotropic and is determined by
means of the standard 3ω method experiment, and if α is
calibrated carefully, the technique described here may fur-
2.6
Expt.
2.5 y = 0.38*x − 0.86
Least−squares fit
2.4
Conductivity ratio [001]/[100]
We are indebted to Kevin White (Apple, Inc.), Helena
McGahagan (Mathematics Department, U. C. Santa Barbara),
Youli Li (Materials Research Laboratory, U. C. Santa Bar-
bara), Chong Zhang (ECE Department, U. C. Santa Barbara),
and Hong Lu (Materials Department, U. C. Santa Barbara) for
several helpful discussions, to Jeffrey T. Imamura and Anna
A. Revolinsky for assisting with measurements, and to Tyler
Cain and Susanne Stemmer (Materials Department, UCSB)
for providing us with STO substrates. This work was sup-
ported by the Center for Energy Efficient Materials, an En-
ergy Frontier Research Center funded by the U.S. Depart-
ment of Energy, Office of Basic Energy Sciences (Award No.
DE-SC0001009). The XRD analysis reported in this work
made use of shared experimental facilities of the Materials
Research Laboratory, an NSF MRSEC, supported by NSF
DMR 1121053. The MRL is a member of the NSF-supported
Materials Research Facilities Network (www.mrfn.org).

M1−001
2.3

2.2 APPENDIX: THE FOURIER-SERIES METHOD FOR 3D

SOLUTION OF THE HEAT EQUATION IN AN
2.1 ANISOTROPIC MATERIAL UNDER SINUSOIDAL
2 M3−001 HEAT-FLUX EXCITATION

1.9 Consider a substrate of dimensions Lx and Ly in the plane

of the heater and thermometer, and let h be the thickness of
1.8
M1−100
the substrate. Suppose further that the heater line has a length
1.7 a and a width w, that the thermometer is at a distance D
1.6
from the heater and parallel to it, that the thermometer has a
6.5 7 7.5 8 8.5 9 width wt , and that its TO is averaged over a length a1 (equal
Effective conductivity of [100] and [001] directions (W/m−K)
to the distance between its inner voltage probes). Assuming
FIG. 8. Thermal conductivity ratio vs. effective thermal conductivity on a diagonal anisotropic thermal conductivity matrix, steady
three of the TiO2 samples tested in this work. state conditions, and frequency of oscillation 2ω, the heat
124903-7 A. T. Ramu and J. E. Bowers Rev. Sci. Instrum. 83, 124903 (2012)

equation is given by Although it appears that this series gives zero heat-flux at
2 2 2 z = h, it must be remembered that a Fourier series cannot
∂ T ∂ T ∂ T 2
be differentiated term by term. In order to calculate kz ∂∂zT2 in
kx 2
+ ky 2 + kz 2 = j · 2ωCv · T (A1)
∂x ∂y ∂z p+ 1 πz
Eq. (A1), following Ref. 19, we multiply it by sin ( 2 ) and h
with the following heat-flux boundary condition at the top integrate twice by parts to yield
surface, z = h:  h 
  ∂ 2T p + 12 π z
∂T P   w  w
kz 2 sin dz
−kz = u x+ −u x− z=0 ∂z h
∂z wa 2 2
  
 
a a ∂T 1 2 π2
· u y+ −u y− . (A2) = (−1)p kz − p+ kz
2 2 ∂h 2 h2
Here, u is the unit step function, P is the input power and  h 
p + 12 π z
T(x, y, z) is the (complex) temperature oscillation profile × T sin dz. (A4)
in the substrate. The other boundary conditions are that the z=0 h
TO is 0 + j0 on the sidewalls of the substrate, x = ± L2x Thus, we are led to a surface term to account for the non-zero
L
and y = ± 2y , as well as on the bottom of the substrate, flux at the top boundary. ∂T
∂h
is an abbreviation for ∂T
∂z
evalu-
z = 0. ated at z = h. The series Eq. (A3) can be differentiated term
Thus, we are led to the form by term with respect to both x and y since there are no surface
∞
terms to account for. Substituting Eq. (A3) in Eq. (A1), multi-
 (2m + 1) π x (2n + 1) πy p+ 1 πz
T = amnp cos cos plying both sides by cos (2m+1)πx cos (2n+1)πy sin ( 2 ) , in-
Lx Ly h
m,n,p=0
Lx Ly tegrating over the dimensions of the substrate, and then using
 Eq. (A4) for the term involving the second-derivative in z, we
p + 12 π z
× sin . (A3) get
h

   
Lx Ly h 2 π
2
2 π
2
1 2 π2
amnp . − (2m + 1) 2 kx − (2n + 1) 2 ky − p + kz − j · 2ωCv
8 Lx Ly 2 h2
 Lx /2  Ly /2
∂T (2m + 1) π x (2n + 1) πy
= dx dy (−1)p kz cos cos (A5)
−Lx /2 −Ly /2 ∂h L x Ly
Equation (A2) gives kz ∂T
∂h
. The integrals in Eq. (A5) can easily be evaluated to yield the complex coefficients amnp :
(−1)p ·P
wahπ 2
· · sin (2m+1)πw
32
(2m+1)(2n+1)2Lx
sin (2n+1)πa
2Ly
amnp = 
. (A6)
2 π2 2 π2 1 2 π2
(2m + 1) L2 kx + (2n + 1) L2 ky + p + 2 h2 kz + j · 2ωCv
x y

The expansion Eq. (A3), together with the coefficients
Eq. (A6), gives the TO profile throughout the substrate. Upon
averaging this temperature oscillation profile over the ther-
mometer’s width wt and the distance a1 between its voltage
probes, we arrive at the (complex) temperature oscillation am-
plitude of the thermometer:
In this work, we have used m = 0 to 400, n = 0 to 160,
and p = 0 to 160. MATLAB R
’s matrix computation algorithms
permit rapid evaluation of the summation of Eq. (A7) despite
a large number of terms. It can be implemented in about 35
lines of MATLAB code, available from the authors upon re-
quest.

1 W. R. Thurber and A. J. H. Mante, Phys. Rev. 139, A1655–A1665

∞ amnp (−1)p Lx Ly (1965).
3D
Ttherm = 2 X. Quelin, B. Perrin, G. Louis, and P. Peretti, Phys. Rev. B 48, 3677–3682
m,n,p=0 wt a1 (2m + 1) (2n + 1) π 2
(1993).
  3 A. J. Schmidt, X. Chen, and G. Chen, Rev. Sci. Instrum. 79, 114902 (2008).
(2m + 1) π D + w2t
· sin
4 M. Bertolotti, A. Ferrari, G. L. Liakhou, R. Li Voti, A. Marras, T. A.

Lx Ezquerra, and F. J. Balta-Calleja, J. Appl. Phys. 78, 5706 (1995).

5 J. Hartmann, P. Voigt, M. Reichling, and E. Matthias, Appl. Phys. B 62,
 
(2m + 1) π D − w2t 493–497 (1996).
− sin 6 D. G. Cahill and R. O. Pohl, Phys. Rev. B 35, 4067–4073 (1987).
Lx 7 Y. K. Koh, S. L. Singer, W. Kim, J. M. O. Zide, H. Lu, D. G. Cahill, A.

Majumdar, and A. C. Gossard, J. Appl. Phys. 105, 054303 (2009).

(2n + 1) π a1 8 B. W. Olson, S. Graham, and K. Chen, Rev. Sci. Instrum. 76, 053901
· 2 sin . (A7) (2005).
2Ly
124903-8 A. T. Ramu and J. E. Bowers Rev. Sci. Instrum. 83, 124903 (2012)

9 A. T. Ramu and J. E. Bowers, J. Appl. Phys. 112, 043516 15 D. de Ligny, P. Richet, E. F. Westrum, Jr., and J. Roux, Phys. Chem. Miner.
(2012). 29, 267–272 (2002).
10 T. Borca-Tasciuc, A. R. Kumar, and G. Chen, Rev. Sci. Instrum. 72, 2139 16 D. G. Cahill, Rev. Sci. Instrum. 61, 802 (1990).

(2001). 17 C. Yu, M. L. Scullin, M. Huijben, R. Ramesh, and A. Majumdar, Appl.

11 S. M. Lee and D. G. Cahill, J. Appl. Phys. 81, 2590 (1997). Phys. Lett. 92, 191911 (2008).
12 H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Solids, 2nd ed. 18 S. Stølen, R. Glöckner, and F. Grønvold, J. Chem. Thermodyn. 28,

(Oxford University Press, New York, 1946). 1263–1281 (1996).

13 K. A. McCarthy and S. S. Ballard, J. Opt. Soc. Am. 41, 1062 19 W. A. Strauss, Partial Differential Equations: An Introduction (Wiley,

(1951). USA/Canada, 1992).

14 D. de Ligny and P. Richet, Phys. Rev. B 53, 3013–3022 (1996). 20 R. Waser, Microelectron. Eng. 86, 1925–1928 (2009).