Exploring pressure-dependent kinetics of phase transitions in Si and Ge using machine learning interatomic potentials

Silicon (Si) and Germanium (Ge) exhibit several metastable phases promising for the integration of optoelectronics into Si-based devices. Using machine learning (ML) interatomic potentials we explore atomistic mechanisms driving pressure-induced phase transitions in both Si and Ge. In particular, we developed a ML interatomic potential specifically-tailored to study pressure-induced phase transitions for Ge. We exploited the DeePMD-kit, incorporating configurations along minimum energy paths extracted by solid-state Nudged Elastic Band (ss-NEB) and ss-Dimer calculations. We demonstrate how this potential enables detailed exploration of pressure-dependent phase transitions, notably showing a nucleation event. For silicon, we also focus on the pressure-induced phase transitions, leveraging on molecular dynamics simulation (MD) based on an established ML potential. We reveal the competition between the kinetics of local nucleation and full-cell transitions by a synergic exploitation of ss-NEB and pressure-controlled MD simulations.