Over 60% of the energy in the United States is wasted, most of it as heat. This amounts to staggering losses of natural and economic resources, and although some of this heat is Carnot heat that is unavoidable, even recovering a small fraction of the remaining waste heat would lead to economic and environmental benefits. Thermoelectric materials are a class of materials that can generate power from heat, but their widespread deployment has been limited because thermoelectric materials are currently inefficient, made from rare elements, and decompose at high temperatures when operated in air. Researchers have sought to develop oxide thermoelectric materials to overcome these shortfalls, but development of oxide materials is still relatively new, and has lacked guiding principles that have led to significant advances in traditional thermoelectric materials.

The work presented here outlines the development of design principles for oxide thermoelectric materials, which involved the creation of a thermoelectric materials database, identification of the property space of interest, and the experimental preparation and characterization of materials in this property space. This work also highlights the development of machine learning models to create a materials recommendation engine. This recommendation engine has discovered a new class of thermoelectric materials with unexpected chemistry, and is being used to suggest other new material compositions likely to exhibit promising thermoelectric performance.