Cs₂AgBiBr₆ lead-free perovskite solar cells: A numerical and ML approach to optimize HTLs for enhanced PV performance

Lead-free perovskite solar cells (PSCs) utilizing cesium silver bismuth bromide (Cs₂AgBiBr₆) are eco-friendly alternatives to traditional lead-based devices. However, their practical application is limited by low power conversion efficiency (PCE), stemming from insufficient charge transport and interfacial recombination losses. This study presents a hybrid numerical–machine learning (ML) framework to optimize Cs₂AgBiBr₆-based PSCs structured as FTO/SnO₂/Cs₂AgBiBr₆/HTL/Au. Device simulations using SCAPS-1D were performed to optimize the device architecture, identifying Cu₂O as the most effective HTL, leading to enhanced photovoltaic performance under optimized conditions. Cu₂O demonstrates effective energy interaction and transport properties, achieving open-circuit voltage (V OC ) of 1.59 V, short-circuit current density (J SC ) of 16.05 mA/cm², fill factor (FF) of 84.05%, and a maximum simulated PCE of 21.39% under optimized and idealized conditions. The simulated quantum efficiency (QE) reaches approximately 92% across a broad spectral range, indicating efficient carrier collection. Performance optimization was further analyzed under varying electrical conditions. To enhance performance prediction, supervised ML models—including Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN)—are trained on a 625-point dataset generated by SCAPS. The Gradient Boosting model exhibits the strongest predictive capability, achieving R² = 0.9992 and RMSE = 0.0344, effectively capturing nonlinear relationships among device parameters. The integrated SCAPS-ML framework provides a systematic and computationally efficient approach for optimizing lead-free PSCs.