Integrated SCAPS-1D modeling and ML approaches for performance enhancement of lead-free Ca₃AsI₃ PSCs

Perovskite solar cells (PSCs) have emerged as promising next-generation photovoltaic technologies due to their remarkable power conversion efficiencies and low fabrication costs. However, the toxicity and environmental concerns associated with lead-based perovskites have accelerated the search for sustainable lead-free alternatives. In this work, a numerical and machine learning (ML) investigation of a Pb-free Ca₃AsI₃-based perovskite solar cell (PSC) is carried out using SCAPS-1D simulation. The device architecture, configured as FTO/C₆₀/Ca3AsI3/CuO/Pt, was optimized by adjusting key design variables including absorber layer thickness, acceptor and defect density, bandgap, and carrier concentrations. Under optimized conditions, the device achieved a power conversion efficiency (PCE) of 27.11%, with an open-circuit voltage (V OC ) of 1.18 V, a short-circuit current density (J SC ) of 26.03 mA cm⁻², and a fill factor (FF) of 88.27%. Among the ten machine learning models evaluated, the Gradient Boosting Regressor (GBR) exhibited the highest predictive accuracy, achieving validation R² scores of 99.89% with a low RMSE of 0.1122. Five-fold cross-validation confirmed the robustness and generalization capability of the model. Furthermore, SHAP feature importance and Pearson correlation analyses revealed that absorber bandgap, defect density, and carrier concentrations are the most influential parameters governing device performance. The findings of this study can help for better understanding of the influence of material and device parameters on the photovoltaic performance of lead-free Ca₃AsI₃-based perovskite solar cells, thereby facilitating their design and optimization for sustainable energy applications.