Comparative optimization of multiple Cs₂SnI₆-based perovskite solar cell architectures using SCAPS-1D simulations and machine learning

Perovskite solar cells (PSCs) represent a leading technology for clean energy generation. Among perovskite materials, Cs₂SnI₆ provides an environmentally friendly alternative with excellent structural and electronic stability, serving as the primary absorber in multiple device architectures. In this study, four PSC architectures — n-i-p, HTL-free, bilayer, and CNT-based — are analyzed using SCAPS-1D simulations. Six key parameters are optimized, yielding maximum power conversion efficiencies (PCEs) of 27.36% (HTL-free), 26.89% (n-i-p), 23.32% (bilayer), and 20.34% (CNT-based), confirming the HTL-free configuration as the most efficient. The optimized HTL-free (FTO/TiO₂/Cs₂SnI₆/Ni) device exhibits PCE = 27.36%, V oc = 1.224 V, J sc = 26.86 mA/cm², and FF = 83.23%. Validation of the SCAPS-1D model against experimental data shows deviations below 6%. Machine learning models (ANN, RF, MLR, ANFIS) are employed, with the ANN (6−16−4) achieving the highest predictive accuracy. The machine learning method predicts the HTL-free structure as the best-performing device among all architectures. For the HTL-free structure, the ANN predicts PCE values closely matching simulation results (R² = 0.991–0.993, MSE = 0.0048–0.0055). Shapley Additive ExPlanations (SHAP) analysis identifies absorber thickness as the most important parameter, followed by back contact work function, as key performance determinants. Correlation analysis shows that both thickness and back contact are strongly correlated with performance parameters. These findings provide quantitative guidance for material selection, interface engineering, and the design of high-efficiency, lead-free PSCs for sustainable energy applications.