Simulation and machine learning driven optimization of Rb2SnBr6-based lead-free perovskite solar cells using diverse ETLs for enhanced photovoltaic performance

Abstract

In this work, n–i–p planar heterojunction perovskite solar cells (PSCs) were simulated using SCAPS-1D, employing Rb₂SnBr₆ as a lead-free, stable, and cost-effective absorber. Fluorine-doped tin oxide (FTO) served as the transparent substrate, with gold (Au) as the rear contact, and three electron transport layers (ETLs) – ZnSe, In₂S₃, and CdZnSe – were evaluated. Device I (ZnSe) achieved the best performance with a PCE of 28.73%, V_OC of 0.868 V, J_SC of 38.09 mA cm⁻², and FF of 86.86%. Device II (In₂S₃) and device III (CdZnSe) exhibited lower efficiencies of 26.62% and 24.04%, respectively. To further analyze performance, a random forest (RF) machine learning model was applied, with SHAP values identifying the most influential parameters. The RF model demonstrated high accuracy (R² = 0.8825) and strong agreement between predicted and actual efficiencies. These results highlight the potential of Rb₂SnBr₆, particularly with a ZnSe ETL, for developing high-efficiency, eco-friendly PSCs.