Descriptors to identify asymmetric conduction bipolar materials with ultra-high thermoelectric power factors*

Low bandgap thermoelectric materials suffer from bipolar effects at high temperatures, with increased electronic thermal conductivity and reduced Seebeck coefficient, leading to reduced power factor and low ZT figure of merit. However, the presence of strong conduction asymmetries between the conduction and valence bands can allow high-phonon limited electronic conductivity at finite Seebeck coefficient values, leading to largely enhanced power factors. The power factors that can be achieved are significantly larger compared to their maximum unipolar counterparts, allowing for doubling of the ZT figure of merit. Using both advanced electronic Boltzmann transport calculations for realistic bandstructures as well as simplified parabolic electronic bands, including all relevant energy dependencies of scattering rates, we develop a series of descriptors which can guide machine learning studies in identifying such classes of materials with extraordinary power factors at nearly undoped conditions. For this we test more than 1500 analytical bandstructures and their features, and more than 60 possible descriptors, to identify the most promising ones that contain: i) only bandstructure features for easy identification from material databases, and ii) bandstructure and transport parameters that provide much higher correlations, but are somewhat more difficult to find parameters for. We validate the descriptors in cases from the half-Heusler material family and identify possible candidates.