Predictive insights and optimization of $${\textrm{Rb}_{2}\textrm{AgInCl}_{6}}$$ double perovskite solar cells using SCAPS 1D and machine learning

Abstract

In this study, a high-performance, lead-free double perovskite solar cell (PSC) with an ITO/SnS/ \({\textrm{Rb}_{2}\textrm{AgInCl}_{6}}\) / \({\textrm{Cu}_{2}\textrm{NiSnS}_{4}}\) /Pt architecture was comprehensively investigated to address the need for accelerated photovoltaic design. A combined framework of SCAPS-1D simulations and machine learning (ML) was implemented to unravel non-linear relationships and evaluate eight predictive models. Among these, the top-performing Categorical Boosting (CatBoost) model demonstrated superior predictive accuracy, achieving a remarkable testing \(R^2\) of 0.9994 and a 5-fold cross-validation score of 0.9991. Furthermore, correlation matrices and SHAP analysis revealed that acceptor density had the greatest influence on all photovoltaic outputs ( \(J_{sc}\) , \(V_{oc}\) , FF, and PCE). Through this data-driven optimization, the PSC achieved a remarkable efficiency leap, boosting the power conversion efficiency (PCE) from \(18.37\%\) to an optimized \(19.76\%\) . These findings underscore the efficacy of integrating device simulations with machine learning for the accelerated design of stable, highly efficient, and environmentally sustainable lead-free perovskite photovoltaics.