Dislocation-limited Thermal Transport in Silicon Carbide Dislocations from Machine-learning Molecular Dynamics

Abstract Silicon carbide (SiC) emerges as a promising wide-bandgap semiconductor featuring high breakdown field and outstanding thermal conductivity, making it promising for high-power application. Although extensive efforts have been devoted to understanding and controlling dislocations in SiC, the fundamental relationship between dislocations and thermal transport remains insufficiently understood. The major challenge lies in the giant supercell size required for reliable dislocation modelling. Here, we employ machine-learning molecular dynamic with neuroevolution potential (NEP) designed for modeling extended defects in SiC to overcome the computational challenge for phonon-dislocation interaction. Using this approach, we calculated the thermal conductivity over a range of temperatures and dislocation densities for typical dislocation types in SiC. The phonon-edge-dislocation scattering was found to be nonlinearly dependent on the dislocation density, which was attributed to the long-ranged dislocation-induced normal strain field. Spectral heat-current analysis further reveals that dislocations primarily scatter long-wavelength acoustic phonons below 20 THz. This work provides microscopic and quantitative perspective into the phonon-dislocation scattering mechanisms.