Thermal conductivity modeling using machine learning potentials: application to crystalline and amorphous silicon

First principles-based modeling on phonon dynamics and transport using density functional theory and the Boltzmann transport equation has proven powerful in predicting thermal conductivity of crystalline materials, but it remains unfeasible for modeling complex crystals and disordered solids due to the prohibitive computational cost to capture the disordered structure, especially when the quasiparticle 'phonon' model breaks down. Recently, machine learning regression algorithms show great promises for building high-accuracy potential fields for atomistic modeling with length scales and timescales far beyond those achievable by first principles calculations. In this work, using both crystalline and amorphous silicon as examples, we develop machine learning-based potential fields for predicting thermal conductivity. The machine learning-based interatomic potential is derived from density functional theory calculations by stochastically sampling the potential energy surface in the configurational space. The thermal conductivities of both amorphous and crystalline silicon are then calculated using equilibrium molecular dynamics, which agree well with experimental measurements. This work documents the procedure for training the machine learning-based potentials for modeling thermal conductivity and demonstrates that machine learning-based potential can be a promising tool for modeling thermal conductivity of both crystalline and amorphous materials with strong disorder. (C) 2019 Elsevier Ltd. All rights reserved.