Enhanced Design\nof Kesterite Solar Cells through High-Throughput\nScreening and Machine Learning Approaches

Kesterite Cu<sub>2</sub>ZnSnS<sub>4</sub> (CZTS) is regarded\nas\none of the most promising materials for thin-film solar cells due\nto its high light absorption capability, composition of earth-abundant\nand nontoxic elements, and ease of low-cost mass production. Although\nthe certified power conversion efficiency (PCE) of kesterite solar\ncells has exceeded 14%, this efficiency remains significantly below\nthe Shockley–Queisser limit. In this study, we generated a\nPerdew–Burke–Ernzerhof (PBE) band gap data set encompassing\n263 64-atom species for high-throughput screening by substituting\nelements at different sites in A<sub>2</sub>BCX<sub>4</sub> quaternary\nkesterite materials. Additionally, we utilized a symbolic regression\nmethod based on genetic programming to explore the functional relationship\namong the oxidation state, ionic radius, and electronegativity of\nkesterites with PBE band gaps. Simultaneously, we employed decision\ntree models (XGBoost, LightGBM, CatBoost, and random forest) and convolutional\nneural network (CNN) models (CustomCNN, VGG16, DenseNet121, Xception,\nand EfficientNetV2B0) to predict band gaps, achieving a coefficient\nof determination (<i>R</i><sup>2</sup>) of up to 0.93. Furthermore,\nwe selected 54 kesterite materials with PBE band gaps ranging from\n0.4 to 1.5 eV for detailed electronic structure calculations with\nHeyd–Scuseria–Ernzerhof (HSE06) functional and investigated\nthe effects of B-site atomic substitutions on the performance of solar\ncell materials. Compared to Ag<sub>2</sub>CaSnSe<sub>4</sub>, Ag<sub>2</sub>SrSnSe<sub>4</sub> exhibits fewer deep defects and richer\nshallow defects, which contribute to an increased carrier concentration\nand reduced charge and energy losses, making it a superior candidate\nfor solar cell applications.