Quenching Thermal Transport in Aperiodic Superlattices: A Molecular Dynamics and Machine Learning Study

Random multilayer (RML) structures, or aperiodic superlattices, can localize coherent phonons and therefore exhibit drastically reduced lattice thermal conductivity compared to their superlattice counterparts. The optimization of RML structures is essential for obtaining ultralow thermal conductivity, which is critical for various applications such as thermoelectrics and thermal barrier coatings. A higher degree of disorder in RMLs will lead to stronger phonon localization and, correspondingly, a lower lattice thermal conductivity. In this work, we identified several essential parameters for quantifying the disorder in layer thicknesses of RMLs. We were able to correlate these disorder parameters with thermal conductivity, as confirmed by classical molecular dynamics simulations of conceptual Lennard-Jones RMLs. Moreover, we have shown that these parameters are effective as features for physics-based machine learning models to predict the lattice thermal conductivity of RMLs with improved accuracy and efficiency.