The viability of thermoelectric energy conversion as an alternative to more traditional technologies depends on the availability of materials with a high thermoelectric figure of merit (ZT). Such materials would be characterized by low thermal conductivities and high electrical conductivities and Seebeck coefficients. The intricacy of this materials optimization problem caused a stagnation of the field for several decades until the advent of modern, powerful nanostructuring techniques. Among them, the synthesis of nanograined materials has several advantages, such as relying only on intrinsic properties and being able to produce bulk samples directly. Still, experimental resources are scarce and so far researchers have focused their efforts on improving known compounds with high bulk ZT, or alternatively on cheap and readily available materials like silicon.
This presentation is devoted to the results of a fully ab-initio high-throughput screening of a large library of nanograined thermoelectrics. All the phenomenological coefficients involved in ZT were directly calculated for each material without experimental input. The first class explored comprised the 78,768 ternary compounds with half-Heusler prototype available in the aflowlib.org repository. Comparison shows that many half-Heusler candidates can have higher performance than those from elementary group-IV and binary III-V semiconductors. More specifically, values of ZT near 3 are plausible at high temperatures. Good candidates can be found for use either as type-p or as type-n thermoelectrics.
The factors underlying the advantages of half Heuslers in this context are analyzed, and the distribution of ZT over this class of materials studied in terms of their constituent elements, leading to practical recipes that can help guide experimental studies. For instance, good candidates are more likely to appear if two of their elements come from the first columns of the periodic table. Comparison with experimental data for several well-known half Heuslers shows that thermoelectric performance in the bulk and in nanograined form do not necessarily go hand in hand, underlining the importance of looking for the right material for each regime.
In order to go beyond nanograined structures, a software tool chain is presented that implements lattice thermal conductivity calculations for semiconductors in bulk and nanowire forms. At its core is a solver for the Boltzmann transport equation based on ab-initio input. Using the same library of half Heusler as test systems, the difficulties of a brute-force high-throughput approach to this problem are discussed, and an alternative based on machine-learning techniques is introduced.
All are welcome to attend this colloquium.