Atomistic Mechanism of Perovskite Grain Boundary Healing by Halide Doping: Machine Learning and Ab Initio Analyses

Facile methods of materials fabrication create polycrystalline structures and introduce grain boundary (GB) and other defects that influence performance. Focusing on a typical GB in a popular all-inorganic metal halide perovskite, we combine ab initio and machine learning tools to study the evolution of the GB structural and electronic properties in a halide-rich environment on a nanosecond time scale. We demonstrate that separately, the GB and halide dopant introduce midgap electron and hole trap states. However, the chemical driving force and low barriers allow dopant diffusion to the GB region, resulting in self-passivation of the extended and point defects. Every few tens or hundreds of picoseconds, the halide dopant hops, perturbing geometric and electronic structure and giving rise to trap states within the bandgap. This lasts only transiently, a few picoseconds. Halide doping makes the GB more structurally sound compared to the undoped GB, which exhibits long, 100 ps regions of structural instability and persistent trap states. The transient trap states originate from jammed structures in sub-boundary regions, while unfavorable conformations directly at the GB have sufficient space to relax. Most of the time, the doped GB contains no midgap traps while separating electrons and holes and assisting in exciton dissociation. The demonstrated self-healing of perovskite GBs under halide-rich conditions provides an efficient strategy for improving material properties. The detailed atomistic analysis, obtained with the help of modern machine learning tools, assists in resolving conflicting statements regarding the influence of perovskite GBs on material performance and provides valuable insights into the unusual properties of these important materials.