ETD

• density functional theory
• displacive phase transitions
• machine learning potentials
• melting
• molecular dynamics
• non-volatile memory
• phase change materials

Germanium Telluride (GeTe) is one of the simplest phase-change materials (PCMs), a class of compounds able to reversibly switch between crystalline and disordered phases with drastic changes in electrical and optical properties. Despite its simple stoichiometry, GeTe exhibits remarkable behavior, making it the prototypical system to study the fundamental mechanisms of PCMs, which are central to nonvolatile memory technologies. Understanding these transformations requires largescale atomistic simulations, but the cost of density functional theory (DFT) limits their scope, while classical molecular dynamics (MD) relies on accurate interatomic potentials.
In this work, we develop a machine learning interatomic potential (MLP) for ground-state GeTe. After detailed convergence studies, a DFT dataset of crystalline and liquid configurations is generated to train the chosen model. The MLP is then validated against DFT and tested in MD simulations, showing reliable performance for both phases. This framework enables accurate and efficient simulations of GeTe, paving the way for large scale studies of PCM behavior relevant to memory applications.