Performance boost in novel CBAI/SNMO double perovskite solar cells via GO-induced back surface field optimization

Abstract

This investigation presents a comprehensive SCAPS-1D and machine-learning-assisted investigation of four environmentally benign double-perovskite solar cell architectures: Device I (Cs₂BiAgI₆/WS₂/FTO), Device II (Sm₂NiMnO₆/WS₂/FTO), Device III (Sm₂NiMnO₆/Cs₂BiAgI₆/WS₂/FTO), and Device IV (GO/Sm₂NiMnO₆/Cs₂BiAgI₆/WS₂/FTO). Systematic simulation demonstrates that integrating both Sm₂NiMnO₆ and Cs₂BiAgI₆ into a dual-absorber configuration (Device III) significantly improves light harvesting, band alignment, and charge separation compared to the single-absorber reference devices. The inclusion of a graphene oxide (GO) back-surface-field layer in Device IV further suppresses interfacial recombination, boosts hole reflection, and strengthens the built-in electric field, enabling the highest performance with V_OC = 1.16 V, J_SC = 35.72 mA cm⁻², FF = 79.62%, and PCE = 32.99%. The study reveals that absorber defect density, acceptor concentration, and interface trap states strongly dictate recombination losses across all structures. To accelerate optimization, ensemble machine-learning models (Random Forest, Gradient Boosting, and XGBoost) trained on extensive SCAPS-generated datasets achieve excellent prediction accuracy (R² > 0.98), while SHAP interpretability identifies absorber defect density and electron affinity as the dominant parameters influencing device efficiency. The combined physics-based and data-driven workflow provides clear, quantitative design rules for high-performance, eco-friendly double-perovskite solar cells and establishes a versatile path toward scaling next-generation, eco-friendly photovoltaic technologies.