Simulation of the Crystal-Amorphous Phase Transition of Ge₂Sb₂Te₅ via a Machine-Learned Interatomic Potential

Phase change materials such as Ge₂Sb₂Te₅ (GST) are ideal candidates for next-generation, non-volatile, solid-state memory due to the ability to retain binary data in the amorphous and crystal phases, and rapidly transition between these phases to write/erase information. Thus, there is wide interest in using molecular modeling to study GST. Recently, a Gaussian Approximation Potential (GAP) was trained for GST to reproduce Density Functional Theory (DFT) energies and forces at a fraction of the computational cost [Zhou et al. Nature Electronics 6, 746–754 (2023)]; however, simulations of large length and time scales are still challenging using this GAP model. We present a machine-learned (ML) potential, implemented using the Atomic Cluster Expansion (ACE) framework, that shows comparable accuracy to the GAP of Zhou et al. and performs three orders of magnitude faster. We train ACE-ML potentials both directly from DFT, as well as using a recently introduced indirect learning approach where the potential is trained instead from an ML potential, in this case, GAP. Indirect learning allows us to consider a significantly larger training set than could be generated using DFT. The substantial speedup of the ACE model compared to existing potentials -- especially when using GPU acceleration -- allows us to perform nanosecond simulations of device-scale samples with only modest computational resources, which we use to demonstrate repeated cycling between the crystal and amorphous phases. We look to use this potential to investigate the crystal nucleation and dynamic optical properties of GST.