Integrated first-principles calculation and machine learning-assisted optimization of high-efficiency lead-free Sb ₂ Se ₃ /Cs Qiaochu Sun, Yuhan Zhang, Jiang Zhao

Perovskite solar cells (PSCs) are widely investigated due to their low fabrication cost, compatibility with lowtemperature solution processing and the continual improvement in power conversion efficiency (PCE). The efficiency of singlejunction architectures is restricted by the Shockley–Queisser limit, which constrains further advances. This study develops a high-performance, lead-free PSC architecture utilizing Sb ₂ Se ₃ and Cs ₂ TiI ₆ as complementary absorbers. First-principles calculations were performed to evaluate the intrinsic structural, electronic and optical properties of bulk Sb ₂ Se ₃ and Cs ₂ TiI ₆ . The interfacial energy alignment and charge transport in the double-absorber device were subsequently evaluated using SCAPS-1D, through which the device parameters were systematically optimized and the PCE increased from 15.73% to 29.57%. By incorporating realistic losses, including series and shunt resistances, temperature, illumination effects and radiative recombination, a corrected efficiency of 25.79% was achieved. To predict photovoltaic parameters and identify the governing performance factors, five machine-learning models were trained on 1500 simulated datasets, with the light gradient boosting machine (LGBM) exhibiting the highest predictive accuracy. SHapley Additive exPlanation (SHAP) analysis identified the bulk defect density of Sb ₂ Se ₃ as the dominant factor influencing PCE. These findings underscore the potential of double-absorber inorganic PSCs and demonstrate that integrating first-principles calculations with machine learning can significantly accelerate the development of scalable, high-efficiency photovoltaics.