Josh Kimmel March 2, 1999

Physics 152, Special Topics Paper

Thermoelectric Materials

ABSTRACT

When a thermal gradient is applied to a solid, it will be accompanied by an electric field

in the opposite direction; this is known as the thermoelectric effect. The physics principles that
are used to explain this phenomenon and how these principles relate to the research of materials
that exhibit the largest thermoelectric effect will be reviewed. Thermoelectric material
applications include refrigeration or electrical power generation. Current state of the art
materials and research being done to find better materials will be discussed. Theoretically, there
is no limit to the thermoelectric effect, and a substantial research effort is currently underway.

INTRODUCTION

Thomas Seebeck discovered the thermoelectric effect in materials in 1823 when he

noticed that a voltage drop occurs across a material with a temperature gradient. However, like
many other discoveries, it took many years to find useful applications and materials for this
discovery. In the 1950s, Abraham Ioffe showed that doped semiconductors showed a much
larger thermoelectric effect than other materials. This led to research of binary semiconductors
as thermoelectric materials in the 1950s, with Bi2Te3 having the greatest thermoelectric effect at
room temperature. The research led a good understanding of the properties displayed by a good
thermoelectric material. It was not until the 1990s that interest increased again due to new
materials discoveries with the potential for excellent thermoelectric properties.1

PHYSICS PRINCIPLES FOR THE THERMOELECTRIC EFFECT, AND THE FIGURE

OF MERIT, Z

When a thermal gradient is applied to a solid, it will be accompanied by an electric field

in the opposite direction; this is known as the thermoelectric effect, and was discovered by
Thomas Seebeck in 1826. It is easiest to show this effect using a one-dimensional model. A
thermal gradient is applied along a line taking point x to be halfway between the low and high
temperature sides. The electrical field, E, generated from a thermal gradient can be conveniently
written as
where v is the electron velocity, and τ is the relaxation time. x-vτ represents the position of the
last electron collision on the high temperature side of x, and x+vτ represents the position on the
low temperature side. To convert to the three dimensional case, consider v=vx, and assume that
the electron velocities are equivalent in three dimensions, i.e. < vx2>=< vy2>=< vz2>=1/3 v2, so
that

The mean velocity, vE, due to the electric field, E is

vE=-eEτ/m
where e is the electron charge and m the electron mass. The mean thermal and electronic
velocities are equal and are travelling in opposite direction, so vQ=-vE. Solving for Q we get

where cv is the heat capacity for free electrons

substituting cv into the equation for the Seebeck Coefficient, Q we get

A material’s Seebeck Coefficient, Q, electrical conductivity, σ,and thermal conductivity,

κ, all relate to the efficiency of a thermoelectric material by the figure of merit, Z
Z=Q2 σ/κ
where Z is in unit (1/°K). A more useful method of describing the thermoelectric efficiency of a
material system is by using the dimensionless figure of merit, ZT, where T is the temperature of
interest. The dimensionless figure of merit, ZT will be used throughout the paper to compare the
efficiencies of different material systems. 1,2

APPLICATIONS

One of the two main applications of thermoelectrics is for refrigeration purposes, and is
called Peltier refrigerators. An electrical current applied across a material will cause a
temperature differential, which can be used for cooling. Bi2Te3 is the thermoelectric material of
choice for this application, ZT=0.9 at room temperature. Bi2Te3 materials cannot meet the
efficiencies to compete with compressor based refrigeration systems, which is equivalent to
ZT=3. Peltier thermoelectrics are more reliable than compressor based refrigerators, and are
used in situations where reliability is critical. NASA uses thermoelectric materials to cool deep
power. The principle would be to use exhaust heat generated by another application such as an
automobile engine, and create electrical power.1

MATERIAL PROPERTIES

To maximize the dimensionless figure of merit, ZT, the Seebeck coefficient and electrical
conductivity are maximized, while simultaneously minimizing the thermal conductivity. The
best thermoelectric materials will have a small electronic contribution to the thermal
conductivity, κel, thus maximizing the electronic conductivity while minimizing the thermal
conductivity. The Weidmann-Franz law states that
κel/σT=L0=2.44 x 108 V2/°K2
Therefore the electronic contribution of the thermal conductivity is a constant (Lorenz
number=L0) at a given temperature. The total thermal conductivity, is the sum of the electronic
and lattice contributions κ= κel+ κlat. Finding materials that minimize the lattice contribution to
the thermal conductivity will maximize the dimensionless figure of merit.
Metals are poor thermoelectric materials because they have a low Seebeck coefficient,
and large electronic contribution to thermal conductivity, so σ and κwill cancel each other out.
Insulators have a high Seebeck coefficient, and a small electronic contribution to thermal
conductivity, however their charge density and therefore electrical conductivity are low leading
to a low thermoelectric effect. The best thermoelectric materials are between metals and
insulators; i.e. semiconductors with an electronic density of 1019/cm3, see Figure 1.
The next step is finding semiconductors with a low lattice contribution to the thermal
conductivity, thus maximizing the dimensionless figure of merit. The materials with the lowest
κlat have large a large number of atoms in the unit call, a large average atomic mass, and a large
average coordination number per atom thus lowering the mean free path. Another method of
reducing κlat is finding a unit cell with cages, or voids, with loosely bound atoms that can “rattle”
and thus scatter phonons.
There are four groups of materials that are being researched. Normal broadband
semiconductors, semiconductors with rattling atoms or molecules, “correlated” metals or
superconductors, and superlattices, also called multiple quantum well materials.1
NORMAL BROADBAND SEMICONDUCTORS

There already has been a substantial effort in finding p-type or n-type binary semiconductors
with high dimensionless figures of merit. The current state of the art materials include Bi2Te3;
ZT=0.9 at room temperature, PbTe; ZT=1.2 at 700°K, and SiGe; ZT=0.6 at 1000°K.
Large values of electronic conductivity are necessary for good thermoelectric materials;
thus large values of electron or hole mobility are needed. It can be shown that maximizing ZT is
equivalent to maximizing U/κlat, where U is the weighted mobility for electrons and holes
U=Nµm*3/2
N is the number of equivalent parabolic bands, µ is the electron or hole mobility, and m* is the
effective mass. Covalently bonded semiconductors will have the highest effective mobility. To
maximize ZT, we need superconductors with U>0.02 m2V-1s-1, and κlat<3 Wm-1°K -1, see Figure
2. These quantities are a good starting point to determine the thermoelectric efficiency of
semiconductor materials. Research is focused towards finding ternary materials that maximize
ZT.1,3

Figure 2 – Lines of Constant ZT in the Weighted Mobility versus κLattice

“RATTLING” SEMICONDUCTORS

The main approach to reducing the lattice contribution of the thermal conductivity has
been to lower the mean free path by using heavy atoms within large unit cells. However, the
lowest thermal conductivity is seen in amorphous structures. The idea is to find a material that
group of the periodic table. To form the loosely bound void, the R atom radius has to be smaller
than the void radius of the unit cell. Nolas and Slack proved this principle by adding La – 1.79A,
Nd – 1.70A, and Sm 1.69A to the IrSb3 binary (not filled) skutterudite unit cell with a void
radius of 2.04A. An order of magnitude decrease in the κlat was exhibited by filling the cages in
the binary skutterudite crystals.4 This effect is also shown in Figure 4 by adding La or Ce to the
CoSb3 binary skutterudite crystals to form RFe3CoSb12 where R represents La or Ce. The best
dimensionless figure of merit at room temperature is only ZT=0.25, but these structures show
great promise at higher temperatures.1,4

Figure 3 – Filled Skutterudite Crystal Figure 4 – κLattice versus Temperature for

LaFe3CoSb12, CeFe3CoSb12 and CoSb3

CORRELATED METALS AND SEMICONDUCTORS

Correlated metals and semiconductors, or Kondo insulators are described by rare earth
intermetallic compounds usually containing Ce. Metals are not good thermoelectric materials
due to a low Seebeck coefficient; however, this is not true for rare earth intermetallic compound
where the 4f level is close to the fermi energy. These compounds have a high density of states
near the fermi energy, thus scattering conduction electrons. A simple power law no longer
describes the scattering time, but now exhibits resonant scattering. The Kondo effect, which
relates to the scattering of conduction electrons, shows compounds with large Seebeck
coefficients at low temperatures, reference Figure 5. Compounds such as CePd3 show Q as high
as 115 µV/°K at low temperature. It should be noted that Q or S could represent the Seebeck
coefficient interchangeably. The minimum value of Q, when κlat=0, has to be Q=156 µV/°K to
have the dimensionless figure of merit, ZT=1. The best materials to date have only achieved
dimensionless figure of merit, ZT=0.25 or below.1,5
 Figure 5 – Seebeck Coefficient versus temperature for CePd3, CeIn3, and Bi2Te3

SUPERLATTICES

Superlattices or Multiple Quantum Well (MQW) structures are materials that will only
conduct electrons in two dimensions, thus increasing the density of states with decreasing well
widths. The increase in density of states will increase a materials figure of merit, by increasing
the Q2σ factor, assuming no change in the thermal conductivity. It is also assumed that there is
no change between the electron mobility of the bulk material, and the quantum well, so any
increase in Z would arise through the factor Q2n, where n is the electron density. Values of Q2n
versus well thickness are shown in Figure 6, and give good agreement between experiment and
theory. Lowering the well thickness, and using multiple quantum well structures shows a large
increase in Q2n, and thus Z. Thomas Reinke at Naval Research Laboratories has shown a 20%
improvement in the ZT of multiple quantum well structures versus bulk material. At quantum
wells <10A, phonons will be scattered by the interfaces between quantum wells, thus lowering
the thermal conductivity, and further enhancing the KT value. The theory neglects two physical
effects that will lower the dimensionless figure of merit. Thermal current flows through the
quantum wells and the barrier layer equivalently, thus lowering the electrical conductivity in the
quantum wells. In addition, electron tunneling will occur between quantum wells, which lowers
the systems density of states, thus lowering the ZT value. Theoretically, the largest values for
the dimensionless figure of merit can be reached using this method.1,6,7,8
 Figure 6 - Q2n versus Well Thickness for Multiple Quantum Well PbTe. The black dots
represent experimental data points for PbTe.

CONCLUSIONS

The potential uses of thermoelectric materials as Peltier refrigerators, or for power

generation are limitless if the dimensionless figure of merit, ZT can reach a value of 3 or larger.
There are currently a few applications in practice that require high reliability, most notably
NASA’s use of thermoelectric materials to cool deep space probes to temperatures as low as
160°K. There are four groups of materials that are being researched. Normal broadband
semiconductors, semiconductors with rattling atoms or molecules, “correlated” metals or
superconductors, and superlattices, also called multiple quantum well materials. Semiconductors
with “rattling” atoms, and multiple quantum wells have the greatest potential for improving the
ZT value, and improved values of ZT have already been shown experimentally.
 REFERENCES

1. G. Mahan, B. Sales, and J. Sharp, Physics Today, March 1997, 42-47.

2. N.W. Ashcroft and N.D. Mermin, Solid State Physics, ed. Saunders College, Philadelphia,
1976, 20-25, 47, 52.
3. J.W. Sharp, Mat. Res. Soc. Symp. Proc., 478, 1997, 15-23.
4. G.S. Nolas, G.A. Stark, et.al., J. Appl. Phys., 79, 1996, 4002-4007.
5. R.J. Gambino, W.D. Grobman, and A.M. Toxen, Appl. Phys. Lett., 22, 1973, 506-507.
6. L.D. Hicks and M.S. Dresselhaus, Phys. Rev. B, 47, 1993, 12727-12731.
7. L.D. Hicks, T.C. Harmon, et.al., Phys. Rev. B, 53, 1996, 10493-10496.
8. D.A. Broido and T.L. Reincke, Phys. Rev. B, 51, 1995, 13797-13800