Machine Learning-Driven Design and Optimization of Lead-Free Perovskite Solar Cells Using SCAPS Simulations

The research examines the photovoltaic efficiency of conventional perovskite solar cells, particularly focusing on tin-based absorber layers (CH3NH3SnX3). Perovskite solar cells (PSCs) are highly promising for solar energy, offering strong absorption and efficiency, yet commercialization is hindered by issues with stability and lead toxicity. To address these issues, the research utilizes the SCAPS simulator to analyze device efficiency and employs machine learning (ML) models, generating a dataset of 2785 points by varying absorber layers, thickness, doping levels, temperature, and defect density. Five ML algorithms-support vector regression, linear regression, random forest, neural network, and XGBoost-are implemented, with random forest yielding the best results (R²=92 %). Additionally, the study examines nine design parameters within an Al/FTO/ETL/CH3NH3SnX3/HTL/Au architecture, revealing maximum power conversion efficiencies of 19.67% for CH3NH3SnBr3 and 28.21 % for CH3NH3SnI3. The research also investigates various Electron Transport Layers (ETLs) and systematically evaluates their effects on cell performances. The findings indicate that absorber layer and other parameters composition significantly affect the efficiency, while defect levels have the highest impact on the proposed cell. An optimized Al/FTO/WS2/CH3NH3SnI3/PTAAI Au configuration, achieving a 36.55% efficiency, is found to be the most promising approach for enhancing tin-based PSCs, without requiring complex simulations.