Discovery and optimization of thermoelectrics materials using high-throughput calculations and machine learning

The development of thermoelectric devices requires new high-performance materials. Traditionally, this has been done in two ways: by optimising known thermoelectric materials or by discovering new materials with promising properties. However, although the experimental investigations are the most efficient way to improve and/or to discover thermoelectric properties, these methods are time-consuming, which limits the possible advances in thermoelectric materials. At the same time, the rise of high-speed calculations has made possible to explore and/or to optimize, in a systematic and accelerated manner, families of known thermoelectric materials as well as new families.In this presentation, I will review the advances in high-throughput calculations related to the identification of new families of stable compounds for thermoelectric applications, as well as the screening of compounds in the same family to find the most promising one. We will also look at the recent contribution of machine learning methods to these investigations.