Thermoelectric Materials: Principles, Structure, Properties, and Applications

During the 1990s there was a heightened interest in the eld of thermoelectrics driven by the need for more efcient materials for electronic refrigeration and power generation (Nolas et al. 2002, Tritt et al. 2000a, Tritt et al. 1998, Tritt et al. 1997). Proposed industrial and military applications of thermoelectric (TE) materials are generating increased activity in this eld by demanding higher performance, nearroom-temperature TE materials than those presently in use. Thermoelectric refrigeration is an environmentally green method of small-scale localized cooling in computers, infrared detectors, electronics, and opto-electronics as well as many other applications. However, most of the electronics and optoelectronics technologies typically require only small scale or localized spot cooling of small components which do not impose a large heat load. If signicant economical cooling can be achieved the resulting cold computing could produce speed gains of 30 200% in some CMOS computer processors. Cooling is perceived by many as the fundamental limit to electronic system performance. Cooling of laser diodes and infrared detectors to temperatures (T ), between 100 K and 200 K, would greatly improve performance and sensitivity and thus is extremely important to many technologies. Thus, the potential payoff for the development of low-temperature thermoelectric refrigeration devices is great, and the requirement for compounds with properties optimized over wide temperature ranges has led to a much-expanded interest in new thermoelectric materials. Utilization of Peltier coolers in relation to refrigeration of biological specimens/samples is an emerging application of thermoelectrics. Power generation applications are being investigated by the automotive industry as a means to develop electrical power from waste engine heat for use in the next generation vehicle. These uses range from power generation utilizing waste engine heat from the exhaust and radiator cooling system to seat coolers for comfort or electronic component cooling. Of course, the deep space applications of NASAs Voyager and Cassini missions using radioactive thermoelectric generators (RTGs) are well established. Given the present energy needs experienced in the United States there is even a more pressing need to investigate alternative energy conversion technologies in this country, e.g., the thermal to electrical energy conversion from natural heat gradients that thermoelectric technologies provide. This has already been identied as important in many European and Asian countries. An overview of the state-of-the-art materials, a theoretical and experimental discussion of the basic principles, as well as an overview of some of the recent develop-

ments and materials is given in the texts of Tritt and Nolas (Nolas et al. 2001a, Tritt 2000a).

1. Thermoelectric Phenomena
1.1 Seebeck and Peltier Effects A discussion of thermoelectric effects and devices should start with the most fundamental phenomena, the Seebeck effect or thermopower (Morelli 1997), (see Thermoelectric Power: Theoretical Concepts). In the early 1800s Seebeck observed that if two dissimilar materials were joined together and the junctions were held at different temperatures (T and T DT) a voltage difference (DV) was developed that was proportional to the temperature difference (DT). The ratio of the voltage developed to the temperature gradient (DV/DT) is related to an intrinsic property of the materials called the Seebeck coefcient (a) or the thermopower. The Seebeck coefcient is very low for metals (only a few mV K1) and much larger for semiconductors (typically a few 100 mV K1). A related effect (the Peltier effect) was discovered a few years later by Peltier, who observed that if an electrical current is passed through the junction of two dissimilar materials, heat is either absorbed or rejected at the junction depending on the direction of the current. This effect is due to the difference in Fermi energies (discussed later: Sect. 4.1) of the two materials. The Seebeck effect and the Peltier effect are illustrated in Figs. 1 and 2, respectively. These two effects are related to each other, as shown in the denition of the Peltier coefcient, P;
P aT 1

The rate that the Peltier heat is liberated or rejected at the junction (QP) is given by:
QP aIT 2

where, I is the current through the junction and T is the temperature in kelvin. There are also a number of thermomagnetic effects such as the Hall, Ettingshausen, and Nernst effects that are beyond the scope of this article. The reader is referred to the text by Nolas et al. (2001a) for a discussion of these effects. 1.2 Thermoelectric ModulesDevices

The Peltier effect is the basis for many modern day TE refrigeration devices and the Seebeck effect is the basis for TE power generation devices (see Thermoelectric Devices: Refrigeration and Power Generations With No Moving Parts). The versatility of TE materials is illustrated in Fig. 3 which shows a diagram of a thermoelectric couple composed of an n-type (negative thermopower and electron carriers) and a p-type (positive thermopower and hole 1

Thermoelectric Materials: Principles, Structure, Properties, and Applications

Figure 1 A schematic illustrating the Seebeck effect of a couple of two dissimilar materials, in which a voltage difference (DV) is generated that is proportional to the temperature gradient (DT) between the ends of the couple. The Seebeck effect is the basis for power generation discussed in the text.

Figure 3 A schematic illustrating a thermoelectric couple or module made of an n-type (electrons, closed circle arrows) material and a p-type (holes, open circle arrows) material. This diagram illustrates both power generation (Seebeck effect) and electronic refrigeration (Peltier effect) capabilities of the thermoelectric module.

carriers) semiconductor material connected through metallic electrical contact pads. Both refrigeration and power generation may be accomplished using the same module as shown in Fig. 3. A TE module or device is built up of an array of these couples, which are arranged electrically in series and thermally in parallel. Thermoelectric energy conversion utilizes the Peltier heat generated when an electric current is passed through a thermoelectric material to provide a temperature gradient with heat being absorbed on the cold side, transferred through (or pumped by) the thermoelectric materials and rejected at the sink, thus providing a refrigeration capability. The advantages of TE solid state energy conversion are compactness, quietness (no moving parts), and localized heating or cooling. Many of the potential applications and importance of these devices have already been discussed in the background and applications section.

2. The Figure of Merit

2.1 Denition and Description of the Figure of Merit The potential of a material for thermoelectric applications is determined in large part to a measure of the materials dimensionless gure of merit,
Figure 2 A schematic illustrating the Peltier effect between two dissimilar materials (A and B). As a current (I) ows through the junction of the two materials, heat is either absorbed or rejected depending on the direction of the current through the two materials. The Peltier effect is the basis for electronic refrigeration discussed in the text. ZT a2 sT a2 T k rk 3

where, a is the Seebeck coefcient, s the electrical conductivity, r the electrical resistivity, and k the total thermal conductivity (k kL kE ; the lattice

Thermoelectric Materials: Principles, Structure, Properties, and Applications

and electronic contributions, respectively). The power factor, a2sT, (or a2T/r) is typically optimized as a function of carrier concentration (typically around 1019 carriers cm3), through doping, to give the largest ZT. High mobility carriers are most desirable in order to have the highest electrical conductivity. Semiconductors have been primarily the materials of choice for thermoelectric applications. The ZT for a single material is somewhat incomplete since an array of TE couples are utilized in a device or module. There are two materials in the TE couple, which is shown in Fig. 3; an n-type and a p-type. The resulting gure of merit for the couple (or a device) is typically given by:
ZT ap an 2 T rn kn 1=2 rp kp 1=2 4

The ZT is not a xed quantity for the couple and can depend on the relative dimensions of the branches. ZT is maximized when the product RK is minimized, where R is the total couple resistance and K is the couple thermal conductance. This is accomplished when:
Ln Ap Lp An   rp kn 1=2 rn kp 5

and the gure of merit for the couple is given by Eqn. (4). 2.2 Coefcient of Performance and Efciency

The best thermoelectric materials that are currently in devices have a value of ZTE1. This value, ZTE1, has been a practical upper limit since the 1970s, yet no theoretical or thermodynamic reason exists for a ZTE1 as an upper barrier. The value of ZT (Eqn. (3)) can be increased by decreasing kL or by increasing either a or s. However, s is tied to the electronic thermal conductivity, kE , through the WiedemannFranz relationship (see Boltzmann Equation and Scattering Mechanisms) and the ratio is essentially constant at a given temperature. Some of the goals of research efforts in the early part of the twenty-rst century are to nd new materials which either raise the current efciency of thermoelectric devices (increase ZT ) or have the capability of operating in new and broader temperature regimes, especially at lower temperatures, To250 K. Since the 1970s, alloys based on the Bi2Te3 system [(Bi1xSbx)2(Te1xSex)3] and the Si1xGex system have been extensively studied and optimized for their use as thermoelectric materials to perform a variety of solid state TE refrigeration and power generation applications (Goldsmid 1986, Rowe 1995). Due to the extensive investigations of these traditional thermoelectric materials there appears little room for future improvement, and thus, entirely new classes of compounds will have to be investigated (Tritt 1996). Figure 4 shows the ZT as a function of temperature for the Bi2Te3 and Si1xGex materials as

The coefcient of performance (f) (refrigeration mode) of the thermoelectric couple is determined by the cooling power or net heat pumping rate (QC ) over the power input (W ) and is given below:
j
2 QC ap an ITC K DT 1 2I R I ap an DT IR W

where TC TH is the cold side (hot side) temperature and DT THTC. The terms in the numerator are, in order, the Peltier heat, the thermal conductance heat ow (Fourier heat) due to the DT, and the Joule heating term due to the total resistance of the couple. The efciency (Z) (power generation mode) of the thermoelectric couple is given by the power input to the load (W) over the net heat ow rate (QH ), where QH is positive for heat ow from the source to the sink and is given below:
Z W I ap an DT IR 2 QH K DT ap an ITH 1 2I R 7

Thus, one can see that the efciency (Z) is proportional to 1 ZT 1=2 .

Figure 4 A plot of the gure of merit (ZT) as a function of temperature for several bulk thermoelectric materials as indicated in the gure above. The Bi2Te3 material is the current state-of-the-art refrigeration material and the Si1-xGex alloy is the current state-of-the-art power generation material. Several bulk materials are currently under investigation and show some promise, as shown in the gure above.

Thermoelectric Materials: Principles, Structure, Properties, and Applications

well as many of the more recently developed bulk materials, which are discussed in Sect. 4. 2.3 Thermopower or Seebeck Coefcient gap is large enough to effectively minimize minority carrier contributions. In a measurement of the Seebeck effect, what is actually measured is the difference in the Seebeck coefcients of the two dissimilar materials, aAB , which is called the relative Seebeck of the couple. The relative Seebeck coefcient is given by aAB DV/DT, where aAB aA aB , which includes both the sample (aA ) and the lead (aB ) contributions. The absolute thermopower of an unknown sample is then given by the measured aAB , and a known lead value, (aB ), such as Pb or some other known standard. 2.4 Electrical Conductivity

The thermoelectric power or Seebeck coefcient (a) can be thought of as the heat per carrier over temperature or more simply the entropy per carrier, aEC/q where C is the specic heat and q is the charge of the carrier (Chaiken 1990). For the case of a classical gas each particle has an energy of 3/2 kB T , where kB is the Boltzmann constant. The thermopower is thus approximately kB =e, where e is the charge of the electron. Let us estimate an expected magnitude for the thermopower of metals and semiconductors. For metals, the heat per carrier is essentially a product of the electronic specic heat and the temperature divided by the number of carriers (N ), i.e., (aECelT/N) and then a is approximately:
  Cel kB kB T aE E Q e EF 8

The electrical conductivity is related to the density of charge carriers (n) and their mobility (m), typically given by
s nem 11

The mobility is given by:

m et=me 12

where EF is the Fermi energy (related to the chemical potential of the material) which is basically the energy, such that at TE0, all states above this energy are vacant, and all the states below are occupied. The quantity, kB/eE87mVK1, is a constant, which represents the thermopower of a classical electron gas. Metals have thermopower values much less than 87 mV K1 (on the order of 110 mV K1) and decrease with decreasing temperature, i.e., EF ckB T ). In a semiconductor, a charged particle must rst be excited across an energy gap EG. In this case the thermopower is approximated by:
  Cel kB E G aE E Q e kB T 9

where m is the mobility of the carriers, me is the effective mass, and t is the mean scattering time between collisions for the carriers. In metals there are many carriers and states available for conduction, typically nE1022 carriers cm3. The electrical conductivity is then very large for metals; on the order of 106 (O cm)1. Again for semiconductors, the carriers must be thermally excited across a gap for conduction to occur.
  EG sEs0 exp kB T 13

Thus, the thermopower is larger than the characteristic value 87 mV K1 and increases with decreasing temperature. Semiconductors can exhibit either electron conduction (negative thermopower) or hole conduction (positive thermopower). The thermopower for different carrier types is given by a weighted average to their electrical conductivity values (se and sp )
aE ae se ap sp se sp 10

The conductivity can occur through the contributions of both holes and electrons
s neme pemh 14

It is necessary to dope the semiconductors with either donor or acceptor states to allow extrinsic conduction of the appropriate carrier type, electrons, or holes, respectively. It is apparent the total thermopower will be lower than that of either of the individual contributions unless the direct band 4

where n and me are the electron concentration and mobility, respectively, and p and mh are the hole concentration and mobility, respectively. There are two primary ways to achieve a high conductivity in a semiconductor, either by having a very small gap to excite across (EGokBT) or by having very high mobility carriers. Typical values of the electrical conductivity lie between 104 and 104 (O cm)1, however these are somewhat arbitrary boundaries. 2.5 Thermal Conductivity

The thermal conductivity, k, is related to the transfer of heat through a material, either by the electrons or

Thermoelectric Materials: Principles, Structure, Properties, and Applications

via quantized vibrations of the lattice called phonons such that
k kL kE 15

(where kL and kE are the lattice and electronic contributions, respectively). The electrical conductivity and the thermal conductivity are interrelated, in that s is tied to kE through the WiedemannFranz relationship (see Boltzmann Equation and Scattering Mechanisms):
kE L0 sT 16

The ratio is essentially constant at a given temperature. The Lorentz number is given by
L0   p2 kB 2 3 e 17

where, L0 2.45 108 W O K2 (or L0 2.45 108 (V2 K2)). This assumes an elastic scattering mechanism and that the mean scattering time for the electrons and the phonons is the same. In many areas of research related to new TE materials, attempts are being made to reduce the lattice part of the thermal conductivity essentially to its minimum value, i.e., where a minimum lattice thermal conductivity is achieved (that is, when all the phonons have a mean free path equal to the interatomic spacing of the constituent atoms). This is being attempted by scattering phonons in different frequency ranges utilizing a variety of methods such as mass uctuation scattering (a mixed crystal, in ternary and quaternary compounds), grain boundary scattering due to the size of the grains and interface scattering in thin lms or multilayer systems. In addition, there is the relatively recent idea of rattling scattering or inserting new atoms in open structure or cage-like systems to essentially rattle around inside the cages and thus resonantly scatter the heat carrying or acoustic phonons. The aspect of thermal conductivity reduction utilizing the rattling scattering in the cage structure will become much clearer when this aspect is discussed in more detail later in this article in relation to classes of materials called skutterudites, lled skutterudites, and clathrates.

coefcient and electrical conductivity (both in the numerator of ZTsee Eqns. (3) and (4)) are strong functions of the doping level and chemical composition. These quantities must therefore be optimized for good thermoelectric performance. The thermal conductivity of complex materials can often be modied by chemical substitutions. Understanding these various effects, and selecting optimization strategies can be an exceedingly difcult problem, because in complex materials there are often many possible degrees of freedom. Slack suggested that the best thermoelectric material would behave as a phononglass, electron-crystal (PGEC) that is, it would have the electrical properties of a crystalline material and the thermal properties of an amorphous or glass-like material. In general, the successful candidate material requires several specic characteristics. The candidate material is typically a narrow bandgap semiconductor (EGE10 kBT or E0.25 eV at 300 K). Also, the mobility of the carriers must remain high, (mE2000 cm2 (V s)1) while the thermal conductivity must be minimized. This high ratio of mobility to thermal conductivity imparts certain demands on the structure of the material. To maintain a high mobility, the electronegativity difference, Dw, between the elements of the material must be as small as possible, preferably less than approximately Dwo0.5 per bond on average, using the Allred scale. All these demands insure that the structural elements will be drawn from the heavy metal or metalloid set (p-band electrons) of the periodic table, such as: Bi, Sb, Se, Te, Ge, Sn, etc. It is only these elements which have such small electronegativity differences. In addition, many of the compounds formed by these elements are narrow bandgap semiconductors. Currently there are no theoretical or thermodynamic limits to the possible values of ZT . Slack estimated that an optimized PGEC material would have values of ZTE4 in the 77300 K temperature range (Slack 1995). This gives encouragement that such materials may be possible. Thus, a systematic search and subsequent thorough investigation of promising materials might eventually yield these much-needed materials for the next-generation TE devices. Whether one is investigating thin lm or bulk thermoelectrics materials, the PGEC approach may well be the best strategy to employ.

3. Investigating New Thermoelectric Materials

3.1 Phonon-glass: Electron-crystal Approach Slack has described the chemical characteristics of materials that might be candidates for a good thermoelectric material (Slack 1995). The candidates need to be narrow bandgap semiconductors with high-mobility carriers, while the thermal conductivity has to be low. In semiconductors, the Seebeck

3.2

Minimum Thermal Conductivity (kmin)

The thermal conductivity of complex materials can often be modied by chemical substitutions or alloy scattering of the phonons. While grain boundary scattering can be an important consideration for reducing the lattice thermal conductivity, as discussed, it can also be detrimental to carrier mobility. Understanding the various effects and selecting optimization strategies can be an exceedingly difcult 5

Thermoelectric Materials: Principles, Structure, Properties, and Applications

problem due to the fact that in complex materials there are often many possible degrees of freedom. Slack introduced the concept of a minimum lattice thermal conductivity (kmin). The lattice thermal conductivity is given by:
kL EvS CLph 18

thermal conductivity is zero (kL 0)! We will also assume the scattering in this system is elastic and that the WiedemannFranz relationship, slightly rearranged (kE/s L0T, where L0 Lorentz number (L0E2.45 108 V2 K2)) is well behaved in this material. Then we can rewrite Eqn. (3):
ZT a2 sT a2 kE L0 19

where, vS is the velocity of sound, C is the heat capacity, and Lph is the mean free path of the phonons. At high temperatures (T4E300 K) the sound velocity and the heat capacity are essentially temperature independent in typical materials. Therefore, the magnitude and the temperature dependence of kL is basically determined by the mean free path of the phonons. Slack dened the minimum thermal conductivity, kmin , as the thermal conductivity when the mean free path is essentially limited by the interatomic distance between the atoms within the crystal (Slack 1979). Typical analysis of kmin results in values of kminE0.250.5 W m1K1 (Slack 1995). These values will essentially result in a practical limit for new thermoelectric materials, and effectively yields an upper limit of ZTE2 at room temperature, ZTE1.5 at 100 K, and at high temperatures, a ZTE4 may be achievable. This analysis is consistent with estimations of practical limits of ZT by both Slack (1995) and Goldsmid (1986). Based on the predictions by Slack, the ideal thermoelectric materials will probably be those containing at least three elements, two or more of which form a highly crystalline framework, while the third element occupies a site in which it can create localized disorder (e.g., rattlers). More recently this concept has been successfully employed to reduce kL in compounds that possess these rattling atoms, while maintaining relatively high power factors. One example is the skutterudites, discussed later. The framework should be well ordered for the efcient mobility of charge, while the less ordered element rattles around to inhibit phonon conduction. It was such a recipe, which originally led to the correct prediction that lled skutterudites should have good thermoelectric properties (Slack and Tsoukala 1994). 3.3 Minimum Thermopower

Therefore for a material to be a viable TE material it must possess a minimum thermopower which is directly related to the value of ZT and L0. Given this description, in order to achieve a certain value of ZT the material would require that a (L0)0.5 157 mV K1 for a ZT 1 and a (2L0)0.5 225 mV K1 for a ZT 2.

3.4

Bulk Materials Approach

The discussion thus far indicates that new materials must eventually be able to achieve certain minimum values of important parameters to be considered for a potential thermoelectric material. It does not matter if a material has a kLEkmin. If it cannot be tuned or doped to attain a minimum thermopower of aE150 mV K1 then it would not be able to achieve a ZTE1. Unless, of course, a material is in serious violation of the WiedemannFranz relationship. Some of the materials that are under investigation are highlighted in Sect. 4 and a summary is given as to why these materials are interesting for potential TE applications in Sect. 5. Each of the materials under investigation for thermoelectrics has its own set of challenges that must be overcome.

4. Recent Results and Promising Materials

4.1 Complex Chalcogenides One of the important issues relative to the development of low-temperature thermoelectric materials is the identication of mechanisms which might give high thermopower (a) at low temperatures. Possibilities include phonon drag, heavy fermion materials, Kondo systems, (see Kondo Systems and Heavy Fermions: Transport Phenomena) and material which exhibit phase transitions, as well as quasi-onedimensional materials. Low-dimensional systems are known to be susceptible to van Hove singularities (or cusps) in their density of states, D(E), electronic phase transitions, and exotic transport phenomena which can add structure in D(E) near their Fermi energy, EF. In most materials, at temperatures far from a phase transition, the electrical conductivity and thermopower are related to the electron density of states near the Fermi energy D(EF). The

There are certain practical limits for each of the parameters used to calculate ZT. These practical limits must be possible in order to achieve a material viable for thermoelectric applications. For example, in Bi2Te3, to achieve a ZTE1 at T 320 K, sE1 m O-cm, sE225 mV K1, and kE1.5 W m1K1. We have already discussed the ZT barrier which, in effect is given by minimizing the thermal conductivity. It is practical to investigate materials where the electronic and lattice terms are comparable, on the order of 0.751 W m1K1. Let us look at the hypothetical situation of a material where the lattice 6

Thermoelectric Materials: Principles, Structure, Properties, and Applications

conductivity is proportional to D(EF) while a is proportional to
  1 dDE aE D dE E EF 20

conductivity. These materials exhibit a ZTE0.6 at TE300 K. The practical use of these materials is limited by the extreme toxicity of the oxides of the Tl, however other compounds or derivatives not containing Tl may be possible.

evaluated at E EF . Hence, as D is increased, s typically increases while a decreases. This is illustrated more precisely in the Mott relation given below:
 2  2    p kB T 1 d s a e s dE E EF 3 21

4.2

Thermoelectric Oxide Materials (NaxCo2O4)

Doping can produce very substantial effects in these types of materials and can drastically change their electronic transport. Quantum well systems take advantage of a low-dimensional character through physical connement in thin lm structures to enhance the electronic properties of a given material (Hicks and Dresselhaus 1993). Results on a new system of materials grown by the Kanatzidis group at Michigan State University exhibit very promising low-temperature thermoelectric properties and have yielded some of the highest ZT values achieved below TE250 K (Chung et al. 2000). These materials are based on CsBi4Te6 and exhibit low-dimensional-anisotropic transport behavior. This may be an important aspect of future investigations of new materials. The gure of merit as a function of temperature is shown in Fig. 4 for the CsBi4Te6 material, with a ZTE0.8 at TE220 K. Another family of low-dimensional semiconductors or semimetals, called pentatellurides (HfTe5 and ZrTe5), have also shown promise as potential lowtemperature thermoelectric materials (Littleton et al. 2001). These pentatelluride materials exhibit a thermopower that is relatively large over a broad range at low temperatures, To250 K. These materials do, in fact, exhibit anisotropic transport properties with the high conductivity axis being the growth axis (a-axis) and have a van der Waals gap between the individual layers (perpendicular to the b-axis), cleaving easily along these planes. This is similar to the Bi2Te3 system, which results in interplanar scattering for the phonons. The power factor of the Se-doped pentatellurides exceeds that of the optimally doped Bi2Te3 system [(Bi1xSbx)2(Te1xSex)3] over the temperature range of measurement. However, investigations of the thermal conductivity indicate it is too high (kE48 W m1K1) for use as a TE material. Another group of complex chalcogenides that are of interest are the Tl2SnTe5 and Tl2GeTe5 systems that were identied as potential thermoelectric materials by Sharp et al. (1999). One aspect of these materials is that they exhibit a very low total thermal conductivity, (E0.5 W m1K1) with a lattice thermal conductivity very near that of the estimated minimum thermal

The development of air and water-stable refractory metal oxides as thermoelectric materials has potentially numerous advantages over existing hightemperature TE materials, especially for power generation applications. They would be more rugged and easier to prepare than the conventional materials, and could be expected to have a wide range of operating environments. The class of NaxCo2O4 compounds has displayed surprisingly effective thermoelectric properties (Terasaki et al. 1997). The materials are members of a large class of layered metal bronzes, wherein the transition metals form edge shared octahedral oxide layers, with the alkali metal ions located within the layers. The compounds contain mixed valence Co ions with itinerant electrons within the layers, and nonstoichiometric amounts of Na ions between the layers. There are many possibilities in this system of materials both in single crystal and polycrystalline form.

4.3

Zinc Antimonides (b-Zn4Sb3)

b-Zn4Sb3 has been re-investigated for thermoelectric applications by the group at the Jet Propulsion Laboratory (Caillat et al. 1997). They have measured the thermal conductivity of polycrystalline b-Zn4Sb3 samples, and found it is nearly temperature independent between 300 K and 650 K. This is unusual, since typically a 1/T dependence is observed in kL in this temperature range. b-Zn4Sb3 exhibits values of kLE0.65 W m1K1 at 300 K, nearly two times lower than that of Bi2Te3 alloys. These very low values of kL along with its unusual temperature dependence can be predominantly attributed to the relatively complex crystal structure and to the presence of vacancies in the lattice. The electronic transport properties are typical of a semimetal with low electrical resistivity increasing with increasing temperature (dr/dT 4 0). The Seebeck coefcient also increases with increasing temperature and peaks at 675 K with a value of about 200 mV K1. This relatively high Seebeck value (for a semimetal) is the result of fairly large effective mass. The best ZT obtained to date on polycrystalline samples is about 1.4 at 675 K, as shown in Fig. 4 (Caillat et al. 1997). Above 765 K, b-Zn4Sb3 transforms into g-Zn4Sb3, which has poorer thermoelectric properties. Band structure calculations (Kim et al. 1998) predict a metallic behavior with improved thermoelectric 7

Thermoelectric Materials: Principles, Structure, Properties, and Applications

performance for lower doping levels. However, little success has been obtained experimentally in optimizing the doping level of this compound. The compound b-Zn4Sb3 forms a full range of solid solution with the isostructural compound Cd4Sb3 and further reduction in kL can be obtained for Zn4-xCdxSb3 mixed crystals, but these crystals appear to be even less temperature stable than b-Zn4Sb3 itself. Further optimization of the thermoelectric properties of these materials seems to be limited considering the difculties with doping them and the restricted compositional variations possible. kLE3 W m1K1 in the high ZT sample identied above. They further state that possible grain size effects may be able to lower kL even further in concert with previous theoretical predictions (Sharp et al. 2001). Earlier results presented on reduction of grain boundary scattering on the lattice thermal conductivity appear to have observed these predicted results (Tritt et al. 2001).

4.5

Skutterudites

4.4

Half-Heusler Alloys

Another group of materials, which are under investigation as potential thermoelectric materials are the half-Heusler alloys (Poon 2000). The halfHeusler alloys are intermetallic compounds with the general formula MNiSn where M is a group 4 transition metal (M Zr, Hf, Ti). Half-Heusler alloys have a MgAgAs-type crystal structure, forming three interpenetrating f.c.c. sublattices with one Ni sublattice vacant. Heusler alloys (e.g., MNi2Sn) differ from half-Heusler alloys in having the Ni sublattice fully occupied and are also metallic or half-metallic (see Thermoelectric Devices: Refrigeration and Power Generations With No Moving Parts) and exhibit interesting magnetic properties. Half-Heusler alloys, on the other hand, are small band gap semiconductors with a gap of (EGE0.10.5 eV). The half-Heusler alloys exhibit a high negative thermopower (40 mV K1 to 250 mV K1) and low electrical resistivity values (0.18 mO cm) both of which are necessary for a potential thermoelectric material. Some of the recent investigations have centered around the effect of Sb doping on the Sn site (TiNiSn1-xSbx). The Sb doping leads to a relatively large power factor of 0.21.0 W m1K1 at room temperature for small concentrations of Sb. These values are comparable to that of Bi2Te3 alloys. The power factor is much larger at TE700 K where it is over 4 W m1K1 making these materials very attractive for potential power generation considerations. Unfortunately the thermal conductivity is relatively high for a thermoelectric material (on the order of 1020 W m1K1) and much of the current and future research is related to reducing the thermal conductivity while maintaining the high power factor. Shen et al. (2002) reported the effect of partial substitution of Ni by Pd on the thermoelectric (TE) properties in the ZrNiSn half-Heusler system. This investigation yielded a thermoelectric gure of merit of ZT 0:7 at TE800 K in a sample of Zr0.5Hf0.5 Ni0.8Pd0.2Sn0.99Sb0.01. Shen and co-workers were able to achieve values of kLE2 W m1K1 at TE800 K in a sample of Zr0.5Hf0.5Ni0.5Pd0.5Sn0.99Sb0.01 and 8

Skutterudites are the next class of materials that have received a lot of attention within the TE community since the 1990s. This is due in large part to the ability to greatly vary kL, which is facilitated by the lling of the voids within the structure with small diameter, large mass interstitials such as trivalent rare-earth ions (Nolas et al. 1999 and references therein, Uher 2000). Specically, the smaller more massive ions that are incorporated within the skutterudite voids result in the lowest kL. This is shown clearly by Nolas and co-workers in an investigation of the thermal conductivity of RECo4Sb9Ge3, with RE La, Nd, Sm, as compared with the unlled IrSb3. The skutterudite structure is cubic (with space group Im3) and the unit cell contains eight AB3 groups where A is the transition metal element Ir, Co, or Rh, and B is the pnictide element such as P, As, and Sb. It contains 32 atoms per unit cell, with the metal atoms on the corners of the eight cubes with six (four atom) Sb rings inside the cubes with two voids in the remaining cubes. In context of the doping scheme for thermoelectrics the general crystal structure of these materials can be best described by the following designation, &2CO8Sb24. The open square, &, indicates the presence of two voids in the structure which are typically lled by the rare earth atoms to produce the most pronounced effects on the thermal conductivity reduction. The other metal and pnictide sites are doped to charge compensate for the additional electrons contributed by the rare earth doping and to enhance the electronic properties. For example, a sample of (CeyFe1-y)xCo4-xSb12 exhibited a ZTE1.4 at TE1000 K as shown by the GM, JPL, and Oak Ridge groups, see Fig. 4. The reduction in the thermal conductivity of lled skutterudites is due the dynamic, or rattling disorder of the void-lling ions that substantially affect the phonon propagation through the lattice. More loosely bound rattlers produce local vibrational modes of lower frequency and are thus more effective in scattering the lowerfrequency, heat-carrying phonons. The heavier and smaller the ion in the voids, the larger the disorder that is produced and therefore the larger the reduction in the lattice thermal conductivity, kL. This concept, rst introduced by Slack, is corroborated by the large atomic displacement parameters (ADPs)

Thermoelectric Materials: Principles, Structure, Properties, and Applications

that have been observed in alkaline earth and lanthanide-lled skutterudites. ADPs have been found to be very useful in thermoelectric materials research (Sales et al. 2000). Investigations of the TE properties of Yb partially lled skutterudites yielded a gure of merit, ZT41 at TE600 K (Nolas et al. 2000a). Similar results have also been observed in a series of Eu-doped Co-based skutterudites, which is shown in Fig. 4 (Lamberton et al. 2002). The thermal conductivity of the Yb- and Eu-lled skutterudites is much lower, with a relatively weak temperature dependence, as compared to that of CoSb3. Only a 19% Yb in CoSb3 dramatically decreases kL as compared to the unlled compound. The rattling motion of the small and massive Yb in the voids of CoSb3 substantially reduces kL in these compounds. Note that although the Yb0.19Co4Sb12 specimen has a low thermal conductivity, it is still much higher than the estimated minimum thermal conductivity of CoSb3, kminE0.3 W m1K1. Although these results are encouraging, even higher ZT values may be forthcoming in skutterudite compounds. atoms that are encapsulated in two different polyhedra and E represents the group 14 elements such as Si, Ge, or Sn. The type II structure is represented by the general formula X8Y16E136. Metallic variants of these compounds are most readily synthesized. Ba containing Si and Ge type I compounds have shown superconducting properties. Ternary type I clathrates, where charge compensation is established by substitutionally doping the tetravalent framework atoms with trivalent atoms such as gallium, result in semiconducting behavior. This was the approach used by Nolas and co-workers in synthesizing polycrystalline semiconducting type I clathrates in order to investigate their transport properties for possible TE applications. One of the more interesting properties of the type I clathrates is the very distinct thermal conductivity. For example, the semiconductor Sr8Ga16Ge30 exhibits a thermal conductivity of similar magnitude and temperature dependence as that of amorphous materials, where it can be lower than that of vitreous silica and very close to that of amorphous germanium. The low-temperature (To1K) data indicate a T2 temperature dependence. Higher temperature data show a minimum, or dip, indicative of resonance scattering. It is clear from these results that in the Sr8Ga16Ge30 compound the traditional alloy phonon scattering, which predominantly scatters the highest frequency phonons, has been replaced by one or more much lower frequency scattering mechanisms. The scattering of the low-frequency acoustic phonons by the rattle modes of the encapsulated Sr atoms results in the low thermal conductivity. It is suggested that low-frequency vibrational modes of the guest atoms inside their oversized polyhedra are responsible for the large reduction in thermal conductivity in these clathrates. Room temperature structural renements from single crystal and powder neutron scattering and x-ray diffraction reveal large atomic displacement parameters (ADPs) for atoms inside the structure of the type I clathrates. This is an indication of localized disorder within these polyhedra beyond typical thermal vibration. Large single crystals have been synthesized by the Oak Ridge group (Sales and co-workers) and extensive investigations of X8Ga16Ge30 (where X Eu, Sr, Ba) have been performed (Sales et al. 2001). The thermal conductivity properties were consistent with those observed earlier, in that they were low, kLo1 W m1K1 (at TE300 K) and the Eu8Ga16Ge30 and Sr8Ga16Ge30 systems exhibited properties of a structural glass while the Ba8Ga16Ge30 system exhibited a temperature dependence consistent with a crystalline material. This study concluded that tunneling states as well as rattling states are necessary to produce a true glass-like thermal conductivity in these materials. These materials have yet to exhibit a ZTE1, but these are relatively new TE materials and much work is ahead. 9

4.6

Clathrates

Another class of materials that is very interesting for thermoelectrics comprises materials called clathrates (e.g., Sr8Ga16Ge30) (Nolas et al. 2001b, Nolas et al. 2000b). These materials, like the skutterudites are also cage-like structures, thus allowing for a rattling mechanism with which to tune the thermal conductivity. Clathrates with frameworks built up from group 14 atoms have been known since the 1970s based on Na8Si46 and NaxSi136 structures. These materials have crystal structures closely related to those of type-I and type-II clathrate hydrates such as (Cl2)8(H2O)46 and (CO2)24(H2O)136. Many different compositions with these two structures are possible. They are of fundamental interest from the perspective of both bonding and their physical properties. Thus far, there has been a substantial amount of work on compounds with the type I crystal structure. A variety of different elements have been encapsulated inside the polyhedra of these structure types, including alkali metal, alkaline earth, and rare earth atoms. The very low glass-like thermal conductivity found in certain type I clathrate compounds has also attracted interest, especially for potential thermoelectric applications. The bonding scheme in the clathrate hydrates is somewhat similar to that of diamond-structured group 14 elements. A similar framework can therefore be built up from elements such as Si, Ge, or Sn. The type I structure can then be represented by the general formula X2Y6E46, where X and Y are alkalimetal, alkaline earth, or rare earth metal guest

Thermoelectric Materials: Principles, Structure, Properties, and Applications

4.7 Thin Films Materials Approach these ternary materials have been attempted. Certainly theoretical guidance, in terms of band structure calculations and modeling, will be essential to the identication of the most promising materials and compositions. In addition, rapid yet accurate characterization of materials is also essential to effectively advance this eld of research. See also: Intermetallic Compounds: Electrical Resistivity; Intermediate Valence Systems; Thermoelectric Devices: Refrigeration and Power Generations With No Moving Parts; Magnetocaloric Effect: From Theory to Practice

Researchers at the Research Triangle Institute (RTI) have demonstrated a signicant enhancement in the thermoelectric gure-of-merit overcoming a major hurdle since the 1960s (Venkatasubramanian et al. 2002). These materials exhibited a ZTE2.4. The enhancement is attributed to creating a nanoengineered material that is efcient in thermal insulation while remaining a good electrical conductor, not easily found in nature. The thermal insulation arises from a complex localization-like behavior for phonons while the electron transmission is facilitated by optimal choice of band-offsets in these semiconductor heterostructures. These nano-engineered TE devices would offer a high coefcient of performance in solid-state refrigeration as well as allow high cooling power densities to be achieved, as high as 700 W cm2. In addition, these thin-lm devices are extremely fast acting, within B10 ms, which is a factor of 23 000 better than state-of-the art bulk TE technology. The combination of performance, power density, and speed can lead to enormous applications in a gamut of different technologies. Also, there have been reports by Harman et al. at Lincoln Labs on PbTe/PbTeSe quantum dot structures that yield a ZTE0.8 at TE300 K and ZTE2 at TE550 K (Harman et al. 2000). These materials have been grown as thick lms, which have been oated off the substrate to yield free-standing lms, which were measured to yield these results. Conservative estimates of the thermal conductivity were used in these calculations.

Bibliography
Caillat T, Fleurial J-P, Borshchevsky A 1997 Preparation and thermoelectric properties of semiconducting Zn4Sb3. J. Phys. Chem. Solids 58, 111925 Chaiken P M 1990 An introduction to thermopower for those who might want to use it to study organic conductors and superconductors. In: Kresin V Z, Little W A (eds.) Organic Superconductors. Plenum Press, New York, pp. 1015 Chung D Y, Hogan T, Brazis P, Rocci-Lane M, Kannewurf C, Bastea M, Uher C, Kanatzidis MG 2000 CsBi4Te6; a highperformance thermoelectric material for low-temperature applications. Science 287, 10248 Goldsmid H J 1986 Electronic Refrigeration. Pion, London Harman T, Taylor P J, Spears D L, Walsh M P 2000 Thermoelectric quantum-dot superlattices with high ZT. J. Electron. Mater. 29, L14 Hicks L D, Dresselhaus M S 1993 Effect of quantum-well structures on the thermoelectric gure of merit. Phys. Rev. B. 47, 1272731 Kim S-G, Mazin II, Singh D J 1998 First-principles study of ZnSb thermoelectrics. Phys. Rev. B 57, 6199203 Lamberton G A Jr, Bhattacharya S, Littleton IV R T, Kaeser M A, Tedstrom R H, Tritt T M, Yang J, Nolas G S 2002 High gure of merit in Eu-lled CoSb3-based skutterudites. Appl. Phys. Lett. 80, 598600 Littleton IV R T, Tritt T M, Korzenski M, Ketchum D, Kolis J W 2001 Effect of Sb substitution on the thermoelectric properties of the Group IV pentatelluride materials M1XYXTe5 (M Hf, Zr and Ti). Phys. Rev. B. Rapid Comm. 64, 1211047 Morelli D T 1997 Thermoelectric devices. In: Trigg G L, Immergut E H (eds.) Encyclopedia of Applied Physics. Wiley-VCH, New York, Vol. 21, pp. 33954 Nolas G S, Johnson D, Mandrus D G 2002 Thermoelectric materials and devices. In: Proc. 2002 Materials Research Society. Warrendale, PA, Vol. 691 Nolas G S, Kaeser M, Littleton IV R T, Tritt T M 2000a High gure of merit in ytterbium-lled skutterudite materials. Appl. Phys. Lett. 77, 18557 Nolas G S, Morelli D T, Tritt T M 1999 Skutterudites: a phonon-glass-electron-crystal approach to advanced thermoelectrics energy conversion applications. Annu. Rev. Mater. Sci. 29, 89116 Nolas G S, Sharp J, Goldsmid H J 2001a Thermoelectrics: Basic Principles and New Materials Developments. Spinger, New York Nolas G S, Slack G A 2001b Thermoelectric clathrates. Am. Sci. 89, 13641

5. Conclusion
From these discussions it is apparent that there are many strategies towards investigating materials in the hope of identifying the next generation TE materials. In 2002 none of these materials have as yet displaced the current state-of-the-art materials (Bi2Te3 and SiGe) which have held that distinction since the 1970s. However, given the many materials that are yet to be investigated, as well as the ongoing research on the most promising new materials, there is certainly much more work ahead. While the thin lm results are very exciting they may be most appropriate for small-scale electronic and opto-electronics applications where small heat loads or low levels of power generation are more appropriate. To address large-scale refrigeration (home refrigerators) or power generation (automotive or industrial applications) requirements, then higher performance bulk materials will have to be developed. The existing bulk TE materials are based on binary compounds and the ternary systems are just beginning to be investigated. A number of potential materials have been identied and discussed in Sect. 4. Only a small fraction of 10

Thermoelectric Materials: Principles, Structure, Properties, and Applications

Nolas G S, Slack G A, Schujman S B 2000b Semiconducting clathrates: a phonon electron crystal material with potential for thermoelectric applications glass. In: Tritt T M (ed.) Semiconductors and Semimetals. Recent Trends in Thermoelectric Materials Research, Academic Press, San Diego, Vol. 69, Chap. 6, pp. 255300 Poon S J 2000 Electronic and thermoelectric properties of halfHeusler alloys. In: Tritt T M (ed.) Semiconductors and Semimetals. Recent Trends in Thermoelectric Materials Research, Academic Press, San Diego, Vol. 70, Chap. 2, pp. 3776 Rowe D M. (ed.) 1995 CRC Handbook of Thermoelectrics. CRC Press, Boca Raton, FL Sales B C, Chakoumakos B C, Jin R, Thompson J R, Mandrus D 2001 Structural, magnetic, thermal, and transport properties of X8Ga16Ge30 (X Eu, Sr, Ba) single crystals. Phys. Rev. B 63, 24511321 Sales B C, Mandrus D G, Chakoumakos B C 2000 Use of atomic displacement parameters in thermoelectric materials research. In: Tritt T M (ed.) Semiconductors and Semimetals. Recent Trends in Thermoelectric Materials Research, Academic Press, San Diego, Vol. 70, Chap. 1, pp. 136 Sharp J W, Sales B C, Mandrus D G, Chakoumakos B 1999 Thermoelectric properties of Tl2SnTe5 and Tl2GeTe5. Appl. Phys. Lett. 74, 37946 Sharp J W, Poon S J, Goldsmid H J 2001 Boundary scattering of phonons and thermoelectric gure of merit. Physica Status Solidi (a) 187, 50716 Shen Q, Chen L, Goto T, Hirai T, Yang J, Meissner G P, Uher C 2002 Effects of partial substitution of Ni by Pd on the thermoelectric properties of ZrNiSn-based half-Heusler compounds. Appl. Phys. Lett. 79, 41657 Slack G A 1979 The thermal conductivity of non-metallic crystals. In: Seitz F, Turnbull D, Ehrenreich H (eds.) Solid State Physics. Academic Press, New York, Vol. 34, pp. 171 Slack G A 1995 New materials and performance limits for thermoelectric cooling. In: Rowe D M (ed.) CRC Handbook of Thermoelectrics, CRC Press, Boca Raton, FL, pp. 40740 Slack G A, Tsoukala V G 1994 Some properties of semiconducting IrSb3. J. Appl. Phys, 76, 166571 Terasaki I, Sasago Y, Uchinokura K 1997 Large thermoelectric power in NaCo2O4 single crystals. Phys. Rev. B 56, R826859 Tritt T M 1996 Thermoelectric materials beat the heat. Science 272, 12767 Tritt, T M (ed.) 2000a Semiconductors and Semimetals, Recent Trends in Thermoelectric Materials Research, Academic Press, San Diego, Vol. 6971 Tritt T M, Nolas G S, Mahan G, Mandras D, Kanatzidis M G (eds.) 2000b New materials for small-scale thermoelectric refrigeration and power generation applications. Proc. 2000 Materials Research Society, Vol. 626, Warrendale PA Tritt T M, Bhattacharya S, Xia Y, Ponnambalam V, Poon S J, Thadhani N 2001 Effect of various grain structures and sizes on the thermal conductivity of Ti-based half-Heusler alloys. Proc. 20th Int. Conf. Thermoelectrics, Beijing, China. ICT 2001, IEEE Press, pp. 712 Tritt T M, Kanatzidis M G, Lyon H B, Mahan G D, (eds.) 1997 Thermoelectric materialsnew directions and approaches. Proc. 1997 Materials Research Society, Vol. 478, Warrendale PA Tritt T M, Kanatzidis M G, Lyon H B, Mahan G D (eds.) 1998 New material for small-scale thermoelectric refrigeration and power generation applications. Proc. 1998 Materials Research Society, Vol. 545, Warrendale, PA Uher C 2000 Skutterudites: prospective novel thermoelectrics. In: Tritt T M (ed.) Semiconductors and Semimetals. Recent Trends in Thermoelectric Materials Research, Vol. 69, Chap. 5, pp. 139254, Academic Press, San Diego Venkatasubramanian R, Siivola E, Colpitts T, OQuinn B C 2002 Thin-lm thermoelectric devices with high room temperature gures of merit. Nature 413, 597602

T. M. Tritt

Copyright r 2002 Elsevier Science Ltd. All rights reserved. No part of this publication may be reproduced, stored in any retrieval system or transmitted in any form or by any means: electronic, electrostatic, magnetic tape, mechanical, photocopying, recording or otherwise, without permission in writing from the publishers. Encyclopedia of Materials: Science and Technology ISBN: 0-08-043152-6 pp. 111 11