Molecular Dynamics Study of Silicon Carbide Using an Ab Initio-Based Neural Network Potential: Effect of Composition and Temperature on Crystallization Behavior

Structure and diffusion dynamics of silicon carbide (Si1–xCx) are investigated via molecular dynamics computer simulations with ab initio-based neural network potentials, exploring the effect of composition and temperature on crystallization behaviors. A neural network potential is developed to describe high-dimensional potential energy surfaces of silicon carbide (SiC) systems, reproducing first-principles results on their potential energies and forces. The phase behavior of amorphous Si1–xCx below its experimental melting point is systematically demonstrated by analyzing the structural and dynamic properties as a function of temperature and carbon concentration x in the composition range 0 ≤ x ≤ 0.5 and the temperature range T = 2000–2600 K, compared to available experiments. The phase of Si1–xCx is characterized by analyzing the pair correlation function, coordination number, tetrahedral order parameter, SiC tetrahedron fraction, Si disordered fraction, and excess entropy. Our results indicate that the system undergoes the crystallization by organizing the short- and medium-range order as the carbon content increases, where the critical carbon fraction for crystallization increases with temperature. The addition of carbon to silicon results in the phase separation into liquid Si and crystal SiC as well as the partial crystallization of Si1–xCx. The self-diffusivity of Si1–xCx is also evaluated to understand how the structural change caused by the crystallization works on diffusion dynamics. The diffusion dynamics of Si1–xCx becomes slower with increasing carbon content and decreasing temperature, which significantly slows down with onset of the crystallization.