Straub, ChristopherChristopherStraubMundinar, SimonSimonMundinarKlonnek, JessicaJessicaKlonnekPixius, ChristopheChristophePixiusRoßkopf, AndreasAndreasRoßkopf2025-11-142025-11-142025-09-24https://publica.fraunhofer.de/handle/publica/49936210.1109/SISPAD66650.2025.11186386We simulate nickel silicidation in one and two space dimensions via physics-informed machine learning. Our machine learning models are solely trained on the governing physical laws in the form of a reaction-diffusion system, without requiring measurement or simulation data. In 1D, our model yields accurate predictions across a parametric temperature range. The 2D process is well approximated away from irregular domain features. Compared to classical state of the art simulations, our models achieve speedups of three orders of magnitude. We further discuss potential extensions to the approach, including the incorporation of measurement data for calibration purposes and enabling broader applicability to process optimization tasks.enconference paper