Grain Boundary Movement in Single-Layer Hexagonal Boron Nitride: Insights from Molecular Dynamics Simulation using Machine-Learned Potentials.

Grain boundaries are commonly found in materials regardless of the growth method and their presence is not always desirable. While there could be some unique properties that come with grain boundaries, many material applications rely on pristine, single crystalline materials. Being able to stimulate and predict the motion of grain boundaries would be beneficial in developing methodology for improving the quality of materials for which single crystalline structures are desirable. In this work, we present first the details of a machine-learned potential that we have developed using the Allegro architecture and attest to its robustness. We next present results of molecular dynamics simulations of hexagonal boron nitride (h-BN) that investigate the movement of 4|8 grain boundaries. These simulations are carried out with ~10,000 atoms to allow for mimicking realistic size of the system. Our calculations of the activation energy barriers show that the initial movement requires a large amount of energy (~2.2eV). However, subsequent movements of the grain boundary unit need a much lower the barrier (<0.5eV). Our results suggest that if the first movement could be stimulated, then the rest of the grain boundary has a high chance of following that motion to directionally diffuse to the edge of the h-BN sheet. These results provide some guidelines for removing grain boundaries in the h-BN which, when defect-laden, is a promising material for both catalysis and single photon emission.