High-Throughput Computations of Cross-Plane Thermal Conductivity in Multilayer Stanene

Abstract

Computational materials science based on data-driven approach has gained increasing interest in recent years. The capability of trained machine learning (ML) models, such as an artificial neural network (ANN), to predict the material properties without repetitive calculations is an appealing idea to save computational time. Thermal conductivity in single or multilayer structure is a quintessential property that plays a pivotal role in electronic applications. In this work, we exemplified a data-driven approach based on ML and high-throughput computation (HTC) to investigate the cross-plane thermal transport in multilayer stanene. Stanene has attracted considerable attention due to its novel electronic properties such as topological insulating features with a wide bandgap, making it an appealing candidate to ferry current in electronic devices. Classical molecular dynamics simulations are performed to extract the lattice thermal conductivities (κL). The calculated cross-plane κL is orders of magnitude lower than its lateral counterparts. Impact factors such as layer number, system temperature, interlayer coupling strength, and compressive/tensile strains are explored. It is found that κL of multilayer stanene in the cross-plane direction can be diminished by 86.7% with weakened coupling strength, or 66.6% with tensile strains. A total of 2700 κL data are generated using HTC, which are fed into 9 different ANN models for training and testing. The best prediction performance is given by the 2-layer ANN with 30 neurons in each layer.