Accelerated Screening of Ternary Chalcogenides for Multifunctional Energy Applications

Chalcogenides have attracted considerable attention in multiple fields of applications, such as optoelectronics, thermoelectrics, transparent contacts, and thin-film transistors. However, the number of synthesized chalcogenides remains relatively low compared to that of oxides. In this study, we performed systematic high-throughput screening combining first-principles calculations and machine learning modeling to identify novel ternary chalcogenides for energy applications. More than 400,000 compounds are considered by exploiting the ion-substitution approach for the 32 most frequent crystal structure prototypes in the database, with their thermodynamic stabilities evaluated by collecting all available binary and ternary chalcogenides from the OQMD database. This gives rise to a comprehensive database, enabling us to quantify the structural maps and model the thermodynamic stabilities via machine learning. In addition, to guide further experimental synthesis, we modeled the synthesizability of such ternary chalcogenides using the CGCNN approach, giving rise to a series of candidates with a high likelihood of being synthesized experimentally. Furthermore, the functional properties of the (meta-)stable compounds were investigated following the magnitude of the band gaps, obtained via self-consistent hybrid functional calculations. Correspondingly, we performed further calculations to identify for potential candidates for photovoltaic and thermoelectric applications, with the properly formulated characteristic figure of merits and implemented computational workflows. Last but not least, machine learning modeling of such physical properties was carried out based on Bayesian optimization, and we demonstrated how to explore such a large chemical space to obtain materials with optimal physical properties, enabling us to perform inverse design of functional chalcogenides in the future.