Machine learning–enhanced SCAPS optimization of 28.38%-efficient bifacial triple-cation perovskite solar cells for building integration

A very appealing path toward next-generation building-integrated photovoltaics (BIPVs) is represented by bifacial perovskite solar cells (BPSCs), which combine customizable optoelectronic properties with dual-side light harvesting. In this work, a comprehensive and unified optimization framework that goes beyond conventional One-dimensional Solar Cell Capacitance Simulator (SCAPS-1D)-based studies is developed by integrating advanced device architecture, interface engineering, and data-driven modeling. Leveraging SCAPS-1D simulations, this study evaluates n-i-p device architectures incorporating Csx(FA0.4MA0.6)1-xPbI2.8Br0.2 triple-cation perovskite absorber, while systematically screening a wide range of electron transport layers (ETLs) such as BaSnO3, ZnOS, Nb2O5, Ag2S, and hole transport layers (HTLs) including CuAlO2, Me-4PACz, NiCo2O4, CuPc, CuCrO2, alongside BA2PbI4 2D-perovskite (2D-PVK) for passivation purposes. Unlike prior studies, a simultaneous multi-parameter optimization of absorber thickness, defect density, and band alignment was performed under both front and rear illumination, leading to high efficiencies up to 28.38% with bifaciality factors (Bf) approaching unity. A Random Forest (RF) machine learning (ML) model, combined with Shapley Additive Explanations (SHAP), was used to quantitatively identify the dominant physical factors and uncover nonlinear structure–performance relationships beyond the capabilities of traditional simulation methods.This integrated physics-based and data-driven methodology enables enhanced and well-balanced bifacial performance while providing deep mechanistic insight into defect-limited transport and interfacial recombination.