Crystal Structure Prediction of Mixed-Halide Perovskites at Room Temperature Using Machine-Learning Potentials

Cesium lead halide perovskites (CsPbI₃) have attracted much attention as a potential candidate for an all-inorganic, high-efficiency solar cell. However, they suffer from short lifetimes because the perovskite phase is unstable at room temperature. To overcome this, recent experiments introduced another halogen element such as Br to stabilize the perovskite structure, which has lower power conversion efficiency than the pure phase. Several computational studies investigated the thermodynamic stability of mixed halides with density functional theory (DFT) calculations and confirmed that the mixing can enhance the stability. However, those computational studies are based on known prototype structures of CsPbI₃, which risks missing unknown phases in mixed-halide perovskites. The heuristic crystal structure prediction (CSP) based on DFT calculations would be necessary in this respect, but it is computationally too expensive to evaluate the energies of numerous candidates.<br/>In this study, we conduct CSP in mixed-halide perovskites (CsPbI<i>_x</i>Br_3-<i>x</i>, CsPbI<i>_x</i>Cl_3-<i>x</i>) with SPINNER, a recently developed in-house code for CSP using a neural network potential.[1] Using SPINNER, we can explore structures beyond prototypes, with computational costs far less than with the DFT-based approaches. As a result, we identify the most stable phases at 0 K for each stoichiometry. We also estimate the transition temperatures between phases by calculating finite-temperature free energy. The computed transition temperatures of CsPbI<i>_x</i>Br_{3-<i>x </i>}are consistent with experimental observations. In case of Cl, the results indicate that γ-phase with a smaller composition than Br is stable at room temperature. We also find that the predicted γ-phases with mixed-halide are expected to have good optical properties as in CsPbI₃perovskites. We anticipate that the present findings can contribute to extending the lifespan of solar cells based on all-inorganic lead halide perovskites.<br/>[1] S. Kang et al., npj Comput. Mater., 8, 108 (2022)