Ion Migration at Metal Halide Perovskite Grain Boundaries Elucidated with a Machine Learning Force Field.

Metal halide perovskites are promising optoelectronic materials with excellent defect tolerance in carrier recombination, believed to arise largely from their unique soft lattices. However, weak lattice interactions also promote ion migration, leading to serious stability issues. Grain boundaries (GBs) have been experimentally identified as the primary migration channels, but the relevant mechanism remains elusive. Using molecular dynamics with a machine learning force field, we directly model ion migration at a common CsPbBr GB. We demonstrate that the as-built GB model, containing 6400 atoms, experiences structural reconstruction over several nanoseconds, and only Br atoms diffuse after that. A fraction of Br atoms near the GB either migrate toward the GB center or along the GB through different migration channels. Increasing the temperature not only accelerates the ion migration via the Arrhenius activation but also allows more Br atoms to migrate. The activation energies are much lower at the GB than in the bulk due to large-scale structural distortions and favorable non-stoichiometric local environments available at GBs. Making the local GB composition more stoichiometric by doping or annealing can suppress the ion migration. The reported results provide valuable atomistic insights into the GB properties and ion migration in metal halide perovskites.