Machine learning–guided SCAPS-1D design of lead-free CH₃NH₃SnI₃ perovskite solar cells with 40.17% simulated PCE

Abstract We integrate large-scale SCAPS-1D simulation with machine learning (ML) to co-optimize hole-transport layers (HTLs) and absorber engineering in lead-free perovskite solar cells. Screening HTL material/thickness/doping/defects in tandem with absorber selection and thickness across 35,585 device configurations, we identify CH₃NH₃SnI₃ as the preferred absorber and an optimal architecture FTO/WS₂/CH₃NH₃SnI₃/V₂O₅/Pt. Device-level factors (metal work function, series/shunt resistance, interface trap density, and spectrum) were varied to quantify contributions to power-conversion PCE (PCE). Five ML regressors were trained to predict PCE from device descriptors; a tuned Random Forest achieved R 2 = 0.96 (MSE = 0.210) on a held-out test set, enabling rapid exploration of the design space. The best configuration reaches a theoretical/simulated PCE of 40.17% under idealized SCAPS-1D conditions, attributed to a low-doping absorber regime and reduced interfacial recombination; capacitance–voltage analysis corroborates improved charge extraction/accumulation. Sensitivity analyses (series/shunt and interface traps) clarify the assumptions under which this performance is attainable. Beyond a single optimum, the model yields actionable design rules for lead-free PSCs, providing a reproducible, data-driven pathway toward high PCE while highlighting the need for interface passivation and experimental validation.