Machine learning prediction of dual absorber lead-free perovskite solar cells for boosting PCE

Curtailing the toxicity level of perovskites is a considerable obstacle resisting the wide-scale commercialization of perovskite solar cells (PSCs). This study investigates the impact of implementing several charge transport layers (CTLs) on the performance of the proposed lead-free Cs₂TiCl₆/ Cs₂AgBiI₆ PSC employing SCAPS-1D simulations. Additionally, the effect of variations in thickness, doping, and defect concentrations of each layer has been considered to optimize the performance of the proposed device. Furthermore, various machine learning models have been trained to estimate the performance of the proposed device through a generated dataset consisting of [Formula: see text] unique data points. Results reveal that employing high quality Cs₂TiCl₆ layer of [Formula: see text] thickness and [Formula: see text] donor doping density, above a [Formula: see text] Cs₂AgBiI₆ absorber with [Formula: see text] acceptor doping density can theoretically achieve a power conversion efficiency (PCE) of [Formula: see text] and a short circuit current density (J_SC) of [Formula: see text]. Moreover, the extreme gradient boosting (XGB) model has been demonstrated to be the most effective model to predict the performance of the proposed PSC, yielding the lowest root mean square error ([Formula: see text]), and the highest coefficient of determination ([Formula: see text]) among the other examined models. The findings highlight the capability of optimally engineered dual-absorber PSCs to be considered as eco-friendly, competitive alternatives to the conventional lead-based PSCs.