Thermal Transport of GeTe/Sb2Te3 Superlattice by Large-Scale Molecular Dynamics with Machine-Learned Potential

The thermal conductivity of chalcogenide Ge–Sb–Te alloy superlattices (SLs), such as GeTe/Sb₂Te₃, is pivotal for their application in phase-change memory and potential thermoelectric uses. However, the complexity of adjustable SL configurations and inefficient fabrication techniques poses significant challenges for experimental investigations. Additionally, the large size of typical SLs complicates ab initio molecular dynamics simulations, while classical molecular dynamics lacks effective interatomic potentials for these alloys. To overcome these obstacles, we developed a machine-learned potential for GeTe/Sb₂Te₃ SLs using the neuroevolution potential (NEP) framework. The NEP’s performance was evaluated against density functional theory calculations, yielding training root-mean-square errors of 1.54 meV per atom for energy, 66.29 meV/Å for force, and 24.13 meV per atom for virial, and was confirmed by accurately predicting lattice parameters and phonon dispersion relations. Utilizing this model, nonequilibrium molecular dynamics simulations were conducted to investigate the thermal conductivities of ∼60 nm GeTe/Sb₂Te₃ SLs at 300 K, further validated by homogeneous nonequilibrium molecular dynamics calculations. The results indicate nondiffusive thermal transport with conductivities ranging from 0.290 to 0.388 W/mK, with a minimum conductivity observed at the 1:2 SL configuration. The coherent-to-incoherent phonon transport transition was observed in the 1:4 SLs as the lattice period varies. This study provides a robust framework for exploring the thermal transport properties of Ge–Sb–Te superlattices, offering significant insights for future research.