Mayur Singh1,Satish Kumar1
Georgia Institute of Technology1
Mayur Singh1,Satish Kumar1
Georgia Institute of Technology1
The excellent thermomechanical properties of SiC have made it an attractive material for applications at extreme environment such as space entry vehicles subjected to high temperatures. There has been extensive study and modelling of phases of SiC phases, most popular of which includes the 3C-SiC and 6H-SiC phases. These studies have led to the development of empirical potentials for the SiC phases. MD studies have been performed for studying SiC’s properties using these empirical potentials but failed to capture the phase transformations of SiC at various temperatures. Machine Learning potentials are very promising to address this challenge. Neural network potentials (NNPs) have shown great promise at capturing complex physics of interatomic forces with ab-initio accuracy. In this work, we use Convolutional Neural Network potentials trained on DFT computed structures for the various phases of SiC and investigate the variation in thermal expansion coefficients of SiC at various temperatures. Our CNN potentials constructs 3D images of the atomic neighborhood for 3D convolution to predict the forces and energy. We implement this potential into the JAX-MD software package to perform MD simulations to take advantage of the Just-in-Time compilation and native GPU support of JAX-MD.