Physics-informed machine learning for lead-free dual-absorber perovskites: unveiling the synergy of Cs2NaIrCl6/Cs2AgBi0.75Sb0.25Br6

Abstract

This study explores the enhancement of device performance by designing a new led-free architecture that employs a dual-absorber configuration. The device architecture is constructed around a dual-absorber configuration involving Cs2NaIrCl6 and Cs2AgBi0.75Sb0.25Br6, with a PEIE layer incorporated to operate as the ETL. The designed dual-absorber solar cell configuration was examined through numerical simulation using SCAPS under standard AM1.5G one-sun illumination conditions. This work investigates the effect of physics-driven band alignment and interface optimization on device performance in order to realize efficient and sustainable photovoltaic energy conversion. The optimized device exhibits a power conversion efficiency of 24.86%, together with a short-circuit current density (Jsc) of 18.27 mA cm−2, an open-circuit voltage (Voc) of 1.52 V, and a fill factor (FF) of 89.36%. In combination with numerical simulations, machine learning (ML) techniques were employed for performance prediction and parameter evaluation. By training on part of the simulation dataset, the Random Forest and XGBoost models accurately forecasted PCE and identified the key parameters controlling Jsc, Voc, and FF. Using machine learning with SCAPS simulations gave consistent results, which is a good way to speed up optimization and help design high-performance, lead-free dual-absorber PSCs.