Assessing High-Throughput Descriptors for Prediction of Transparent Conductors

The growth of materials databases has yielded significant quantities of data to mine for new energy materials using high-throughput screening methodologies. One application of interest to energy and optoelectronics is the prediction of new high performing p-type transparent conductors (TCs). However, screening methods for such materials have never been validated over the breadth of computed materials properties. In this study, we compile an experimental data set of 74 bulk crystal structures corresponding to known state-of-the-art n-type and p-type TCs and compute a series of corresponding computational descriptor properties. Our goals are to (1) compare computational descriptors to experimentally demonstrated properties of real materials in the data set, (2) determine the ability of ground state, density functional theory (DFT)-based computational screening methodologies to identify these experimentally realized TCs, and (3) use this understanding to estimate reasonable screening cutoffs for four commonly used descriptors. First, stability calculations demonstrate that most materials in the data set have an energy above the convex hull (Ehull) of m*, is found to correlate weakly to conductivity in the p-type materials as compared with n-type materials, which may motivate an increase in the m* cutoff as well. Lastly, we perform an uncertainty analysis and literature comparison for the branch point energy (BPE), a qualitative descriptor for dopability. We find the BPEs of most n-type materials to lie near the conduction band and those of most p-type materials to lie at midgap; this is a significant distinction, suggesting BPE to be a more definitive descriptor for n-type TC materials. By assessing the validity of this simple screening methodology via comparing experimental data to computational descriptors, we aim to motivate and strengthen future materials discovery efforts in the field of transparent conductors and beyond.