Machine learning-enhanced novel design and performance optimization of M3SbI3 (M = Ba and Ca) based dual absorber perovskite solar cells

Abstract

The growing demand for renewable energy necessitates the development of sustainable, high-performance, eco-conscious solar cells. This research proposes a novel lead-free dual-absorber perovskite solar cell (DAPSC) utilizing Ca₃SbI₃ as the top absorber and Ba₃SbI₃ as the bottom absorber layer. The device structure, Al/FTO/SnS₂/Ca₃SbI₃/Ba₃SbI₃/CBTS/Au, was analyzed by employing SCAPS-1D to evaluate photovoltaic performance under standard AM1.5G illumination. The dual-absorber configuration exhibited a considerably improved power conversion efficiency (PCE) of 36.03%, short-circuit current density (J_sc) of 32.18 mA cm⁻², open-circuit voltage (V_oc) of 1.305 V, and fill factor (FF) of 85.84%, outperforming single-absorber perovskite solar cells. We further employed a Random Forest Regression (RFR) model to forecast the proposed device performance. We found that the machine learning (ML) model achieved excellent predictive performance with an average coefficient of determination (R²) of 0.9635, mean absolute error (MAE) of 0.4506, and root mean square error (RMSE) of 0.6253. Moreover, feature importance analysis validated by SHAP (Shapley Additive exPlanations) summary plots and correlation matrices revealed that operating temperature and absorber layer parameters (doping, thickness, and defect level) were the most critical factors influencing photovoltaic performance, while electron and hole transport layer properties also played significant roles. These outcomes reveal that the proposed lead-free DAPSC model presents enhanced performance, stability, and environmental compatibility for photovoltaic applications.