Low thermal conductivity of the layered oxide (Na,Ca)Co2 O4 : Another example of a

phonon glass and an electron crystal

K. Takahata, Y. Iguchi, D. Tanaka, and T. Itoh

Department of Applied Physics, Waseda University, Tokyo 169-8555, Japan

I. Terasaki∗
arXiv:cond-mat/0003069v1 [cond-mat.str-el] 6 Mar 2000

Department of Applied Physics, Waseda University, Tokyo 169-8555, Japan

and also Precursory Research for Embryonic Science and Technology, Japan Science Technology, Tokyo 102-0074, Japan
(August 14, 2018)
The thermal conductivity of polycrystalline samples of (Na,Ca)Co2 O4 is found to be unusually
low, 20 mW/cmK at 280 K. On the assumption of the Wiedemann-Franz law, the lattice thermal
conductivity is estimated to be 18 mW/cmK at 280 K, and it does not change appreciably with
the substitution of Ca for Na. A quantitative analysis has revealed that the phonon mean free path
is comparable with the lattice parameters, where the point-defect scattering plays an important
role. Electronically the same samples show a metallic conduction down to 4.2 K, which strongly
suggests that NaCo2 O4 exhibits a glass-like poor thermal conduction together with a metal-like
good electrical conduction. The present study further suggests that a strongly correlated system
with layered structure can act as a material of a phonon glass and an electron crystal.

PACS numbers: 72.15.Eb, 72.15.Jf, 72.80.Ga, 72.15.Lh

Thermoelectric materials have recently attracted a ters. Namely this compound has a poor thermal con-
renewed interest as an application to a clean energy- duction like a glass and a good electric conduction like
conversion system.1 The conversion efficiency of a ther- a crystal, which Slack10 named a material of “a phonon
moelectric material is characterized by the figure of merit glass and an electron crystal”. It should be mentioned
Z = S 2 /ρκ, where S, ρ and κ are the thermopower, the that rattling is not the only reason for the low ther-
resistivity and the thermal conductivity, respectively. At mal conductivity, where point defects and/or solid so-
a temperature T , a dimensionless value of ZT is required lutions significantly reduce the thermal conductivity of
to be more than unity for a good thermoelectric ma- Lax (Fe,Co)4 Sb12 (x < 1)11 and Co1−x Mx Sb3 (M =Fe,
terial, which is, however, difficult to realize. We have Ni, and Pt).12,13 Nevertheless a search for materials hav-
found a large thermopower (100 µV/K at 300 K) and a ing rattling sites is a recent trend for thermoelectric-
low resistivity (200 µΩcm at 300 K) for NaCo2 O4 single material hunting. Through this search, BaGa16 Ge30 14
crystals.2 These parameters suggest that NaCo2 O4 is a and Tl2 M Te5 (M =Sn and Ge)15,16 have been discovered
potential thermoelectric material. An important finding as potential thermoelectric materials with low thermal
is that the transport properties are difficult to understand conductivity.
in the framework of a conventional one-electron picture A preliminary study of the thermal conductivity of
based on band theories. We have proposed that strong polycrystalline NaCo2 O4 , which has no rattling sites, re-
electron-electron correlation plays an important role in vealed a low value of 15-20 mW/cmK at 300 K.17 This is
the enhancement of the thermopower.2–4 Very recently indeed unexpectedly low, because a material consisting
Ando et al.5 have found that the electron specific-heat co- of light atoms such as oxygens will have a high thermal
efficient of NaCo2 O4 is as large as 48 mJ/molK2 , which is conductivity. In fact, polycrystalline samples of a high-
substantially enhanced from the free-electron value, pos- temperature superconducting copper oxide show a higher
sibly owing to the strong correlation. value of 40-50 mW/cmK at 300 K.18,19 This is qualita-
In addition to a large thermopower and a low resistiv- tively understood from its crystal structure as schemati-
ity, a thermoelectric material is required to show a low cally shown in the inset of Fig. 1. NaCo2 O4 is a layered
thermal conductivity. In view of this, a filled skutterudite oxide, which consists of the alternate stack of the CoO2
Ce(Fe,Co)4 Sb12 shows quite interesting properties.6–9 layer and the Na layer. The CoO2 layer is responsible for
The most remarkable feature of this compound is that the electric conduction, whereas the Na layer works only
“filled” Ce ions make the lattice thermal conductivity as a charge reservoir to stabilize the crystal structure.
several times lower than that for an unfilled skutterudite The most important feature is that the Na ions randomly
CoSb3 .6 The Ce ions are weakly bound in an oversized occupy 50% of the regular sites in the Na layer. The Na
atomic cage so that they will vibrate independently from layer is highly disordered like an amorphous solid, and
the other atoms to cause large local vibrations.9 This vi- it looks like a glass for the in-plane phonons. Thus sig-
bration and the atomic cage are named “rattling” and nificant reduction of the thermal conductivity is likely to
a “rattling site”, respectively. As a result, the phonon occur in the sandwicth structure made of the crystalline
mean free path can be as short as the lattice parame- metallic layers and the amorphous insulating layers.20

1
 In this paper, we report on measurements and quanti- temperature θD as
tative analyses on the thermal conductivity of polycrys-
talline samples of (Na,Ca)Co2 O4 from 15 to 280 K. The h̄v 1
θD = (6π 2 N ) 3 .
observed thermal conductivity is like that for a disor- kB
dered crystal, and is insensitive to the substitution of Ca
for Na. These results imply that the phonon mean free We employ θD =350 K from the recent specific-heat
path is as short as the lattice parameters, and a semi- data,5 and get ℓph = 6.7 Å for 20 mW/cmK, which is
quantitative analysis reveals that the point-defect scat- comparable with the in-plane lattice parameter (3 Å).
tering due to the solid solution of Na ions and vacancies This picture is intuitively understood from the fact that
effectively reduces the lattice thermal conductivity down the Na layer is highly disordered. Note that the observed
to 15-20 mW/cmK. On the other hand, the electrical re- data of 20 mW/cmK includes the electron thermal con-
sistivity remains metallic down to 4.2 K, which means ductivity, and thus the obtained value of 6.7 Å gives the
that the electron mean free path is much longer than the upper limit of the phonon mean free path.
lattice parameters. Thus NaCo2 O4 can be a material of Figure 2 summarizes the thermoelectric parameters of
a phonon glass and an electron crystal, whose conduction NaCo2 O4 . In Fig. 2(a) are shown the thermal conduc-
mechanisms are qualitatively different from those of the tivity (the same data as x=0 in Fig. 1) and the fig-
“rattler” model of the filled skutterudite.7–9 ure of merit calculated using the resistivity and the ther-
Polycrystalline samples of Na1.2−x Cax Co2 O4 (x=0, mopower of the same sample. We also plot the electron
0.05, 0.10 and 0.15) were prepared through a solid- thermal conductivity (κel ) estimated from the resistiv-
state reaction. Starting powders of NaCO3 , CaCO3 ity on the assumption of the Wiedemann-Franz law as
and Co3 O4 were mixed and calcined at 860◦ C for 12 κel = L0 T /ρ (L0 = π 2 kB 2
/3e2 is the Lorentz number).
h. The product was finely ground, pressed into a pel- κel is 10% of κ, and the heat conduction is mainly deter-
let, and sintered at 920◦C for 12 h. Since Na tends to mined by the phonons. The figure of merit is 10−4 K−1
evaporate during calcination, we added 20 % excess Na. above 100 K, which is largest among oxides,1 but does
Namely we expected samples of the nominal composition not yet reach the criteria of ZT = 1. Much progress is
of Na1.2−x Cax Co2 O4 to be Na1−x Cax Co2 O4 , which we thus needed to realize oxide thermoeletcrics.
will denote as (Na,Ca)Co2 O4 . In Fig. 2(b), the resistivity and the thermopower are
The thermal conductivity was measured using a plotted as a function of temperature, which reproduce
steady-state technique in a closed refrigerator pumped the pioneering work on the Na-Co-O system by Molenda
down to 10−6 Torr. The sample was pasted on a copper et al.23 The temperature dependence of the resistivity is
block with silver paint (Dupont 4922) to make a good essentially the same as that for the in-plane resistivity of
thermal contact with a heat bath, and on the other side the single crystals, though the magnitude is much higher
of the sample a chip resistance heater (120 Ω) was pasted owing to the grain-bondary scattering. It should be noted
to supply heat current. Temperature gradient was mon- that the resistivity exhibits metallic conduction down to
itored by a differential thermocouple made of Chromel- 4.2 K without any indication of the localization. This
Constantan, while temperature was monitored with a re- implies that the electron mean free path is much longer
sistance thermometer (Lakeshore CERNOX 1050). than the lattice parameters. Previously we showed that
Figure 1 shows the thermal con- the electron mean free path of the single crystal is as
ductivity of (Na,Ca)Co2 O4 . The substitution of Ca for long as 230 Å at 4.2 K along the in-plane direction.2
Na only slightly decreases the thermal conductivities of We can therefore say that the phonon mean free path is
(Na,Ca)Co2 O4 . This makes a remarkable contrast to the much shorter than the electron mean free path. This is
change of the resistivity with the Ca substitution.3,4 The nothing but a material of a phonon glass and an electron
magnitude (20 mW/cmK at 280 K) is as low as that of crystal.10
a conventional thermoelectric material such as Bi2 Te3 ,21 Here we will compare the measured thermal conductiv-
which is consistent with the previous study.17 ity with the phonon-scattering theory by Callaway.24,25
Let us make a rough estimate of the phonon mean free The total scattering rate τ −1 is given as the sum of three
path (ℓph ) for NaCo2 O4 at 280 K. In the lowest order scattering rates as
approximation, the lattice thermal conductivity κph is
expressed by22 τ −1 = τpd −1 + τph−ph −1 + τ0 −1
= Aω 4 + Bω 2 + v/L
1
κph = cvℓph ,
3 where τpd −1 , τph−ph −1 and τ0 −1 are the scattering rates
for the point-defect scattering, the phonon-phonon scat-
where c and v are the lattice specific heat and the sound tering, and the boundary scattering, respectively. For a
velocity. Since we consider a moderately high temper- phonon frequency ω, the three scattering rates are writ-
ature region where phonons are sufficiently excited, we ten as Aω 4 , Bω 2 and v/L, where A, B and L are charac-
assume c = 3N kB (N is the number of atoms per unit teristic parameters. According to Ref. 26, A is expressed
volume). The sound velocity is associated with the Debye P
as A = Ω0 fi (1 − Mi /M )2 /4πv 3 , where Ω0 is the unit

2
cell volume, Mi is the mass of an atom, P fi is the frac- correlated layer, S 2 /ρ can be increased through the mass
tion of an atom with mass Mi , and M = fi Mi is the enhancement due to the spin fluctuation.1,29 Recently we
average mass. We calculated A for (Na,Ca)Co2 O4 by proposed that the effective mass of NaCo2 O4 is enhanced
following the method in Ref. 11, where Na (23g/mol), as much as that for CePd3 .20 Meanwhile, in the disor-
Ca (40g/mol) and ✷ (vacancy) make a solid solution dered insulating layer, the lattice thermal conductivity
in the ratio of Na:Ca:✷=(1-x):x:1. B is a temperature- can be minimized by the disorder that causes little effect
dependent parameter, which is proportional to T at high on the electric conduction. In this context, it will work
temperatures (B ∼ CT ). It should be √ noted that the as a material of a phonon glass and an electron crystal.
phonon-phonon scattering gives κ ∝ 1/ ACT at high This scenario might be compared with the thermo-
temperatures in the presence of a large A.25 As clearly electric superlattices extensively studied by Dresselhaus
shown in Fig. 1, κ for (Na,Ca)Co2 O4 increases with T , et al.30,31 At present, no band calculation of NaCo2 O4
implying that the phonon-phonon scattering is neglegibly is available, but the band calculation of isostructural
small. Thus L corresponding to an inelastic scattering LiCoO2 shows that the valence bands do not show
length is the only fitting parameter. any sub-band structure expected from the 2D quantum
In Fig. 3, the measured κph (= κ − κel ) of NaCo2 O4 confinement.32 This means that the electronic states of
is compared with two theoretical curves. Sample #1 is NaCo2 O4 are not very anisotropic in a one-electron pic-
the same sample as shown in Fig. 1, and Sample #2 is ture. This situation is essentially identical to the band
another sample prepared in a different run. Curve A is picture of high-temperature superconductors. We think
the calculation using the phonon-scattering theory,24,25 that the enhancement of the thermopower of NaCo2 O4
where L=0.2 µm is used. As expected, the point-defect should not be attributed to the quantum confinement of
scattering quite effectively reduces the thermal conduc- the semiconductor superlattices, but to the strong corre-
tivity by two or three orders of magnitude. The Ca sub- lation.
stitution effect is also consistently explained as shown in In summary, we prepared polycrystalline samples of
the inset, where data points (as indicated by open circles) (Na,Ca)Co2 O4 and measured the thermal conductivity
in different runs are added to show the reproducibility. from 15 to 280 K. We have found that the phonon mean
Although the solid solution of Na and ✷ dominates κph , free path is 6.7 Å at 280 K, which is much shorter
the theory predicts a small correction due to the substi- than the electron mean free path. This means that
tution of Ca (as indicated by the solid line), which is in (Na,Ca)Co2 O4 acts as a materials of a phonon glass and
good agreement with the observation. This directly indi- an electron crystal, though it has no rattling sites. We
cates that the point-defect scattering plays an important have compared the experimental data with the phonon-
role in reducing κph . A problem is the physical mean- scattering theory and the minimum thermal conductivity,
ing of L=0.2 µm: It is much longer than the electron and have found that the point-defect scattering plays an
mean free path (10-102 Å), but much shorter than the important role.
grain size (10 µm). Possible candidates are the average The authors would like to appreciate J. Takeya and Y.
distance of stacking faults and/or interlayer disorder. Ando for technical help for thermal-conductivity mea-
The absence of the phonon-phonon scattering means surement. They also thank T. Kawata, T. Kitajima, T.
that the phonon lifetime is extremely short,27 and is Takayanagi, T. Takemura, and T. Nonaka for collabora-
rather characteristic of the thermal conductivity of a tion.
glass. Curve B is the calculation of the minimum thermal
conductivity κmin by Cahill et al.,28 which has been com-
pared with κph for a glass. Although the calculated κmin
is one order of magnitude smaller than the measured κph ,
such a deviation is also seen in other disordered crystals.
Note that κph ∝ T 3 is not seen for NaCo2 O4 at low ∗
Corresponding author. Electronic address:
temperatures, which is a hallmark of disordered crystals.
[email protected].
Since κph ∝ T 3 is usually seen below 10 K, this is pos- 1
For a recent review, see G. D. Mahan, Solid State Physics
sibly because the measurement temperature was high. 51, 81 (1998).
NaCo2 O4 consists of the sandwich structure of the amor- 2
I. Terasaki, Y. Sasago and K. Uchinokura, Phys. Rev. B56,
phous and crystalline layers, and the heat conduction 12685 (1997).
process is perhaps in between that for a mixed crystal 3
T. Itoh, T. Kawata, T. Kitajima, and I. Terasaki, Pro-
and an amorphous solid. Thus it should be further ex- ceedings of the 17th International Conference on Thermo-
plored which curve is more likely to capture the essential electrics (ICT ’98) May 24-28, 1998, Nagoya, Japan, (The
feature of the heat conduction in NaCo2 O4 . Institute of Electrical and Electronics Engineers, Piscat-
We propose that a layered material consisting of a away, 1998) p. 595.
strongly correlated conducting layer and disordered insu- 4
T. Kawata, Y. Iguchi, T. Itoh, K. Takahata, and I.
lating layer can be a promising thermoelectric material. Terasaki, Phys. Rev. B60, 10584 (1998).
5
If the heavy-fermion system is realized in the strongly Y. Ando, N. Miyamoto, K. Segawa, T. Kawata, and I.

3
 32
Terasaki, Phys. Rev. B60, 10580 (1998). M. K. Aydinol, A. P. Kohan, G. Ceder, K. Cho, and J.
6
D. T. Morelli and G. P. Meisner, J. Appli. Phys. 77, 3777 Joannopoulos, Phys. Rev. B56, 1354 (1997).
(1995).
7
B. C. Sales, D. Mandrus, and R. K. Williams, Science 272,
1325 (1996).
8
B. C. Sales, D. Mandrus, B. C. Chakoumakos, V. Keppens (Na,Ca)Co2O4

Thermal Conductivity (mW/cmK)

and J. R. Thompson, Phys. Rev. B56, 15081 (1997). x=0
9
V. Keppens, D. Mandrus, B. C. Sales, B. C. Chakoumakos, 20 x=0.05
P. Dai, R. Coldrea, M. B. Maple, D. A. Gajewski, E. J. x=0.10
x=0.15
Freeman, and S. Bennington, Nature 395, 876 (1998).
10
G. A. Slack, in CRC Handbook of Thermoelectronics, edited
by D. M. Rowe, p. 407 (CRC Press, Boca Raton, 1995).
11
G. P. Meisner, D. T. Morelli, S. Hu, J. Yang, and C. Uher, Co
Phys. Rev. Lett. 80, 3551 (1998). 10
12
S. Katsuyama, Y. Shichijo, M. Ito, K. Majima, and H. Na
Nagai J. Appl. Phys. 84, 6708 (1998). (50% vacant)
13
H. Anno, K. Matsubara, Y. Notohara, T. Sakakibara, and
H, Tashiro J. Appl. Phys. 84, 3780 (1999).
14
G. S. Nolas, J. L. Cohn, G. A. Slack, and S. B. Schujman,
Appl. Phys. Lett. 73, 178 (1998). 0
15 0 100 200 300
J. W. Sharp, B. C. Sales, D. G. Mandrus, and B. C. Chak-
oumakos, Appl. Phys. Lett. 74, 3794 (1999). Temperature (K)
16
B. C. Sales, B. C. Chakoumakos, D. Mandrus, and J. W. FIG. 1. The thermal conductivity of (Na,Ca)Co2 O4 plot-
Sharp, J. Solid State Chem. 146, 528 (1999). ted as a function of temperature. The inset shows the
17
H. Yakabe, K. Kikuchi, I. Terasaki, Y. Sasago, and K. Uchi- schematic picture of the crystal structure of NaCo2 O4 .
nokura, in Proceedings of the 16th International Conference
on Thermoelectricity (ICT ’97), Aug. 26-19, 1997, Dres-
den, Germany, (The Institute of Electrical and Electronics 2
Engineers, Piscataway, 1998) p. 523. (a) Z
18
J. L. Cohn, C. P. Popoviciu, Q. M. Lin, and C. W. Chu, 20
κ (mW/cmK)

Phys. Rev. B59, 3823 (1999).

Z (10 K )
-4 -1
19
M. Ikebe, H. Fujishiro, T. Naito, and K. Noto, J. Phys. κ
1
Soc. Jpn 63, 3107 (1994).
20
I. Terasaki, in Proceedings of the 18th International Con- 10
ference on Thermoelectricity (ICT ’99), Aug. 29-Sept. 2,
1999, Baltimore, Maryland, (The Institute of Electrical and κel=L0T/ρ
Electronics Engineers, Piscataway, 2000) (in press). 0 0
21
T. Caillat, M. Carle, P. Pierrat, H. Scherrer, and S. Scher- 3 (b) 80
rer, J. Phys. Chem. Solids 53, 1121 (1992). S
ρ (mΩcm)

22
N. W. Ashcroft and N. D. Mermin, Solid State Physics 60

S (µV/K)
(Hot-Saunders, Philadelphia, 1976) p. 22. 2
23 40
J. Molenda, C. Delmas, and P. Hargenmuller, Solid State
Ionics 9&10, 431 (1983). ρ
24
J. Callaway Phys. Rev. 113, 1046 (1959). 1 20
25
J. Callaway and H. C. von Baeyer, Phys. Rev. 120, 1149 NaCo2O4
(1960). 0
26
0
P. G. Klemens, Phys. Rev. B119, 507 (1960) 0 100 200 300
27
Another possibility is that the phonon in NaCo2 O4 is quite Temperature (K)
harmonic. In this case, there are little disspation channels, FIG. 2. The thermoelectric parameters of polycrystalline
and κph would be much higher. NaCo2 O4 . (a) The thermal conductivity (κ) and the figure
28
D. G. Cahill, S. K. Watson, and R. O. Pohl, Phys. Rev. of merit (Z); (b) The resistivity (ρ) and the thermopower
B46, 6131 (1992). (S). Note that the electron thermal conductivity (κel ) evalu-
29
B. C. Webb, A. J. Sievers, and T. Mihalisin, Phy. Rev. ated through the Wiedemann-Franz law is shown by the solid
Lett. 57, 1951 (1986). curve, where L0 is the Lorentz number (= π 2 kB
2
/3e2 ).
30
L. D. Hicks and M. S. Dresselhaus, Phys. Rev. B47, 12727
(1993).
31
M. S. Dresselhaus, Proceedings of the 17th International
Conference on Thermoelectrics (ICT ’98) May 24-28, 1998,
Nagoya, Japan, (The Institute of Electrical and Electronics
Engineers, Piscataway, 1998) p.29 and references therein.

4
 Lattice Thermal Conductivity (mW/cmK)
NaCo2O4 Curve A
L = 0.2 µm
Sample #2
10

Curve B
5 κmin
Sample #1

κph (mW/cmK)
20

10
1 (Na, Ca)Co2O4

0
0.5 3 0 0.05 0.1 0.15
~T Ca-content x

10 50 100
Temperature (K)
FIG. 3. The lattice thermal conductivity (κph ) of
NaCo2 O4 . The open and closed circles represent κph mea-
sured for Sample #1 and #2, where the electron thermal con-
ductivity was estimated through the Wiedemann-Franz law.
Curve A is the calculation proposed by Callaway.24,25 Curve
B is the minimum thermal conductivity proposed by Cahill
et al.28 The inset shows κph of (Na,Ca)Co2 O4 at 200 K. The
solid line is the same calculation as Curve A.