Revealing the crystallization dynamics of Sb-Te phase change materials by large-scale simulations

Understanding the crystallization dynamics of chalcogenide phase-change materials (PCMs) is crucial for optimizing their performance in data storage and neuromorphic computing devices. However, experimental investigation is a great challenge due to the fast speed of transformations, and it is also difficult by ab initio molecular dynamics (AIMD) because of the huge cost of computing resources and the small system which AIMD can deal with. To solve the above problems, a machine learning potential for Sb-Te phases is developed using the neuroevolution potential (NEP) framework. The performance of the NEP potential is compared with density functional theory calculations, with training errors of 15.6 meV per atom, 155.8 meV angstrom-1 and 42.4 meV per atom for energy, force, and virial, respectively. Using the NEP potential, the physical properties of Sb, Sb2Te, SbTe, Sb2Te3, and Te components are carefully evaluated, including lattice parameters, melting points and radial distribution functions, and the results show that the NEP predictions are reliable. Moreover, the investigation on crystallization dynamics explored by large-scale molecular dynamics simulations with a model size of 15 nm3 and over 2 ns, reveals different crystallization features, with growth-dominated in Sb2Te and nucleation-dominated in SbTe and Sb2Te3. Finally, this work provides a more efficient approach for performing large-scale simulations of PCMs, facilitating atomic-level studies of PCM devices. Using an efficient and accurate machine learning potential, large-scale crystallization dynamics of Sb-Te phase change materials are achieved.