Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 9: Topical Session: Physics-driven Artificial Intelligence for Materials II

MM 9.7: Vortrag

Montag, 9. März 2026, 17:45–18:00, SCH/A251

Development of a GRACE Machine-Learning Potential for Modeling SiC Epitaxial Growth — •Anders Vesti, Thomas Hammerschmidt, and Ralf Drautz — Ruhr Universtät Bochum, Bochum, Germany

Silicon carbide (SiC) is a highly attractive wide band-gap semiconductor for power electronics due to its high breakdown field and low on-state resistance. However, the widespread adoption of SiC-based devices is hindered by challenges in epitaxial growth, including uncontrolled polytype switching and defect formation, which ultimately increase production costs.

In this work, we present the development and benchmarking of a physics-driven machine-learning interatomic potential for SiC based on the GRACE formalism. Starting from a general-purpose foundational model, we refit the potential using comprehensive C, Si, and SiC datasets to construct a specialized model tailored for simulating SiC epitaxy.

We validate the resulting GRACE potential against density functional theory (DFT) calculations and available experimental data for Si, C, and SiC. The developed model provides a basis for testing proposed growth mechanisms in SiC epitaxy.

Keywords: Machine-learned interatomic potential; Silicon carbide; GRACE formalism; Epitaxial growth