Machine Learning-Driven Optimization of Transport Layers in MAPbI₃ Perovskite Solar Cells for Enhanced Performance

This study aims to analyse the performance of MAPbI3-based perovskite solar cells (PSCs) by integrating machine learning (ML) models with the SCAPS-1D simulator. An extensive dataset of 28,182 PSCs, combinations of six-electron transport layers, ten-hole transport layers, and MAPbI3 absorber layer by varying thickness of each layer, has been generated in the SCAPS-1D simulator. In this research work, among those eight ML models, the XGBoost algorithm shows high accuracy for predicting the power conversion efficiency (PCE) of the cell, achieving root mean square error (RMSE) of 0.052 and a coefficient of determination (R2) of 0.999. Using Pearson correlation and Shapley Additive Explanations (SHAP), the most effective configuration for high-performance PSCs was identified by evaluating parameter significance. SCAPS-1D simulations revealed an optimal configuration comprising 200nm WS2, 900nm MAPbI3, and 500nm CBTS thin layer, achieving a PCE of 24.34%. Further adjustments in doping densities increased the PCE to 34.65%. This research highlights the critical importance of precise material and structural optimization to improve PSC performance. The integration of ML with traditional simulation techniques provides a robust foundation for PSC research, supporting further experimental validation and potential large-scale applications, ultimately advancing more efficient and durable photovoltaic technologies.