High-Throughput Theoretical Analysis of All-Inorganic Mixed Perovskites Using Deep Learning

All-inorganic perovskite solar cells (PSCs) have gained attention for next-generation photovoltaics due to their superior thermodynamic and optoelectronic stability over hybrid perovskites. Traditionally, the discovery of potential perovskite materials has relied on laborious and costly experiments, but advances in computational methods and artificial intelligence now enable high-throughput exploration. In this study, we present a deep learning workflow that integrates density functional theory (DFT) calculations with a state-of-the-art graph neural network (GNN) model to predict the properties of all-inorganic mixed perovskites suitable for stable single-junction PSCs. Our findings suggest that the stability of perovskites can be improved by incorporating a high chlorine (Cl) ratio, while the associated bandgap widening can be controlled by adjusting the elemental ratio at the B-site. Moreover, mixing Cl with bromine (Br) at the X-site and tin (Sn) with calcium (Ca) or Strontium (Sr) at the B-site yields the lowest mixing energies among nonlead perovskites, a key factor in mitigating phase segregation in mixed compositions. Overall, this workflow provides an effective approach for the discovery of highly functional perovskite materials within a significantly reduced time frame.