A Review of the New Thermoelectric Materials zyxw H. J. Goldsmid' and G. S. Nola? 'School of Physics, University of New South Wales, Sydney 2052, Australia. 'Research and Development Division, Marlow Industries, Dallas, Texas 75238, USA. zyx hjgoldsmid@phys.unsw.edu.au Abstract Most of the materials that are used in today's thermoelectric generators and refrigerators were first developed many years ago. However, during the last decade promising results have been reported for several new systems of compounds and alloys. These include the skutterudites, the clathrates and the half-Heusler alloys. There are also a number of layer-structured compounds, some of which may be regarded as developments of the bismuth telluride system. Certain of these materials seem to possess the highest known figures of merit at least at elevated temperatures. Moreover, it is doubtful that any of the new systems have yet been fully optimized. The problem is that there are large numbers of compounds that remain to be investigated. Here we shall try to formulate some principles that will allow us to select those systems that are likely to yield the best long-term results. We discuss, for example, whether it is better for a material to have a high power factor zyxwvuts or a low lattice thermal conductivity. We also consider the relative worth of a large or small effective mass for the charge carriers. Then, of course, one must take account of problems associated with the structure and preparation of the materials. The potential of the new materials for use zyxwvuts as low-dimensional thermoelectrics will be included in our discussion. Introduction There has been a resurgence of interest in thermoelectric energy conversion during the past few years[l]. This has come about in part fiom dissatisfaction with the conversion techniques that are employed at present and in part eom the formulation of new ideas for the improvement of the performance of thermocouples. These ideas are of two kinds: (i) physical changes that affect the properties of existing materials and (ii) new systems of material that may have superior thermoelectric properties. A newcomer to the field must feel bewildered by the number of options that are open to researchers. In particular, there is the question as to which of the systems is most likely to lead to practical thermoelements with an improved figure of merit. In this paper the different options will be reviewed and it is hoped that useful guidelines will be developed. Existing thermoelectric materials Although it is not the only factor that determines the choice of material, the figure of merit zyxwvutsr z, or its dimensionless equivalent zT, is the most important. The figure of merit is defined from the relation[2] zyxwvuts aLcr a zyxwvuts z=- zyxwvutsrqpon 0-7803-5908-9/00/$10 .OO 0200 1 I EEE where a is the Seebeck coefficient, cr is the electrical conductivity and is the thermal conductivity. It is a common practice nowadays[3] to use the term powerfactor to denote do or, sometimes, &UT. This is based on the idea that it involves only electronic properties whereas il always contains a large lattice contribution. The most widespread thermoelectric application is that of refiigeration at ordinary temperatures and, for this purpose, the best present-day materials are the alloys of BizTe3 with SbTe3 and Bi2Se3. These alloys, when optimized, have a zyxwv zT value of about unity[4]. For thermoelectric generation other materials must be used unless the heat source is at a rather low temperature. Until quite recently the best moderate temperature generator materials were based on PbTe[5] and similar compounds[6]. At higher temperatures, Si-Ge alloys have been employed[7]. Bi-Sb alloys have been used as negative-branch thermoelements for low-temperature refiigeration[8]. Occasionally values of zT in excess of unity have been encountered but materials with zT of the order of 2 or more are still awaited. Physical form BizTQ was first used for refiigeration in the early 1950's. Very soon it was shown that it could be improved if it was combined as a solid solution with one or more isomorphous compounds. This, in effect, was an improvement resulting from a physical effect i.e. the preferential scattering of phonons rather than electrons or holes by the disordered components[9]. Another effect that probably remains to be exploited is the preferential scattering of phonons by the grain boundaries in a polycrystalline material. It seems that boundary scattering can reduce the lattice conductivity in polycrystalline materials even when the grain size is substantially larger than the phonon mean fiee path[lO]. On the other hand, the carrier mobility is unlikely to be affected until the grain size and the fiee path length of the charge carriers are comparable. This effect is expected to be particularly noticeable in solid solutions in which the lattice conductivity is already small. Perhaps the most exciting development of recent years has been the theoretical and experimental work on low- dimensional thermoelectrics[ 111. The changes in the band structure that are brought about by making at least one of the dimensions comparable with the lattice constant can, in principle, improve the power factor. Hopefully, the density of states is increased while the carrier mobility remains unchanged. At the same time, the lattice conductivity may become less through the scattering of phonons at the interfaces. Many people think, in fact, that the reduction of the lattice conductivity may lead to a greater improvement 1 20th International Conference on Thermoelectrics (2001)