First-Principles and Machine-Learning Assisted High-Throughput Screening for Novel Inorganic Perovskite Solar Cell

In recent years, perovskite-based solar cells have shown a dramatic growth of their efficiency whereas the conventional silicon-based solar cells reached the theoretical limit through decades. However, the hybrid organic-inorganic perovskites (HOIP) which enable the rapid improvement of photovoltaic efficiency suffers from lack of long-term stability. To overcome the instability which originate from their organic components, the development of fully inorganic perovskites with comparable performance could be the solution. In this work, we first investigate the optical and electrical properties of all known ABX₃ inorganic perovskites from the inorganic crystal structure database (ICSD) using density functional theory (DFT) calculations. Though we found several new candidate materials for photovoltaic absorbers, their properties are not expected as promising as reported best materials. To discover new all-inorganic perovskite materials that have good performance and stability simultaneously, we suggest so-called ‘double-anion perovskite’ as new promising perovskites for solar-cell. While ABB’X-type double perovskite can only replace B-site with element having the same oxidation state, double-anion perovskites can expand possible combinations of A-B metals if chalcogen and halogen elements are mixed in the anion sites. To search the most stable structure of a given combination, we adopt a machine-learning potential based crystal structure prediction method, which is called ‘SPINNER’. By combining DFT calculations and machine-learning potential methods, we suggest new all-inorganic double-anion perovskites for photovoltaic absorbers.