Multidimensional high-throughput screening for mixed perovskite materials with machine learning

Abstract

Mixed halide inorganic perovskites exhibit exceptional stability and photovoltaic performance and are considered to be promising photovoltaic materials. However, the chemical diversity of these materials presents a vast screening space, making it challenging to efficiently identify high-performance materials solely through theoretical calculations or experiments. To address this challenge, in this work, we introduce a multidimensional high-throughput screening strategy that combines machine learning with first-principles calculations, specifically designed to identify MHIPs with optimal bandgap and light absorption properties. The bandgap and light absorption models have achieved determination coefficients (r2) of 0.9896 and 0.9833, with root mean square errors of 0.1890 eV and 0.2190 105 eV · cm−1, respectively, demonstrating the high precision and reliability of the models. In the present work, the generation of 306 521 candidate materials through mixed B-site elements is reported, leading to the successful identification of 295 materials with ideal characteristics for MHIPs via screening. Subsequently, an in-depth density functional theory validation is conducted on 20 of these materials. The research results demonstrate that Cs2AgBi0.5Sb0.25Ir0.25I6 and CsSn0.75Ge0.25I3 exhibit outstanding performance, making them the most promising candidate materials for practical applications. These results fully confirm the scientific validity and effectiveness of our screening strategy, laying a solid foundation for the exploration and optimization of high-performance perovskite solar cell materials.