Effect of the step-flow growth on defect nucleation in SiC epitaxy by first-principles simulations and machine learning interatomic potentials

SiC is an emerging material for power electronics due to its superior properties, like a wide band gap, compared to conventional Si-based technologies¹. Defect-free epitaxial growth remains one of the main challenges in maximizing wafer yield. Stacking faults and triangular defects occur due to the mixing of two competing phases², called the 3C and 4H polytype with different stacking sequences, and cause leakage currents and device breakdown³.

Chemical vapor deposition of 4H-SiC with H₂ carrier and Si- and C-containing precursor gasses on 4° off-axis cut substrates ensues step-flow growth. We use ab-initio calculations to study the driving forces of the underlying growth mechanisms. For bulk and surface systems, we observe marginal differences in the stability between 3C and 4H⁴. The adatom adsorption on step sites is strongly favored over terrace sites, ensuring that 4H stacking is retained. We suspect that 3C defects are caused by terrace nucleation when adsorption at the step is blocked. Studying the growth kinetics by large-scale ab-initio simulation is challenging, and the available classical potentials are not suited to describe the Si-C-H accurately. We resort to training an accurate machine learning potential on ab-initio data^5-7.