Study of vacancy ordering and the boson peak in metastable cubic Ge-Sb-Te using machine learning potentials

Abstract

The mechanism of the vacancy ordering in metastable cubic Ge-Sb-Te (c-GST) that underlies the ultrafast phase-change dynamics and prominent thermoelectric properties remains elusive. Achieving a comprehensive understanding of the vacancy-ordering process at an atomic level is challenging because of enormous computational demands required to simulate disordered structures on large temporal and spatial scales. In this study, we investigate the vacancy ordering in c-GST by performing large-scale molecular dynamics simulations using machine learning potentials. The initial c-GST structure with randomly distributed vacancies rearranges to develop a semi-ordered cubic structure with layer-like ordered vacancies after annealing at 700~K for 100~ns. The vacancy ordering significantly affects the lattice dynamical properties of c-GST. In the initial structure with fully disordered vacancies, we observe a boson peak, usually associated with amorphous solids, that consists of localized modes at $\sim$0.575~THz. The boson peak modes are highly localized around specific atomic arrangements of straight vacancy-Te-vacancy trios. As vacancies become ordered, the boson peak disappears and the Debye-Waller thermal \textit{B} factor of Te decreases substantially. This finding indicates that the c-GST undergoes a transition from amorphous-like to crystalline-like solid state by thermal annealing in low-frequency dynamics.