Exploration of highly stable and highly efficient new lead-free halide perovskite solar cells by machine learning

<h2>Summary</h2> Density functional theory (DFT) calculations have always been an important tool for predicting and discovering new materials. Using machine learning (ML) technology and 488 datasets computed with the Perdew-Burke-Ernzerhof (PBE) functional, we constructed and trained ML models, subsequently expanding the perovskite database to 177,264 datasets. Finally, high-throughput screening technology was adopted. Through a three-stage screening process evaluating stability, band gap, and photovoltaic performance, the 177,264 datasets were reduced to 434. After high-throughput screening, the spectroscopic limited maximum efficiency (SLME) values of the remaining perovskite solar cells with practical considerations all well exceed 20%, and the selected four perovskite cells exceed 23%. The four groups of perovskite solar cells with the highest SLME values were all lead free. This study provides valuable insights for advancing the development of green lead-free perovskite solar cells with enhanced efficiency and stability.