High throughput screening of thermal interface materials by machine learning

Till now, it remains a challenge for effective prediction and screening of novel materials with high thermal conductivity, as well as further optimization of the interface thermal resistance. Normally, people have to spend long time on tedious calculations when predicting and screening these materials. In this paper, I combined machine learning with molecular dynamics simulations to investigate the thermal conductive properties of materials with the aim of significantly reducing computational consumption. I first applied molecular dynamics simulations to obtain the relevant properties of materials, then generated models for predicting physical properties by machine learning, and finally made predictions of thermophysical properties of materials. The use of machine learning significantly reduces the prediction time compared to direct molecular dynamics simulations. Especially when the XGBoost and the neural network models are employed, the prediction efficiency is significantly improved. This work guides a new way for the future screening of high-performance thermal interface materials.