Dresden 2026 – wissenschaftliches Programm

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

O: Fachverband Oberflächenphysik

O 26: Plasmonics and nanooptics: Fabrication, characterization and applications – Poster

O 26.1: Poster

Montag, 9. März 2026, 18:00–20:00, P2

Extending multiphysics crystallization models for phase-change materials with parameters derived from machine learning — •Luis Schüler^{1, 2}, José Moreira², Andrei Lupuleasa², Wenjing Lei², Keerthika Kalavalapudi², Matthias Wuttig², Thomas Taubner², and Dmitry Chigrin^{1, 2} — ¹AMO GmbH, Aachen — ²I. Institute of Physics (IA), RWTH Aachen

Optical metasurfaces based on dielectric or metallic scatterers such as nanoantennas provide advanced control over light-matter interactions. Dynamic tuning can be realized with phase-change materials (PCMs), which can be reversibly switched between amorphous and crystalline states with a strong difference in refractive index. Especially metallic nanostructures can significantly influence the crystallization dynamics of PCMs, resulting in deviations from the expected spectral behavior of the metasurface. Therefore, Multiphysics simulations that couple electrodynamics, heat transport, and crystallization kinetics are essential for understanding these effects and optimizing device geometry. However, it is experimentally challenging to obtain key crystallization parameters, such as viscosity or free energy density, over the relevant temperature range. Here, we use machine-learned interatomic potentials trained molecular dynamics data to compute the required material parameters. Furthermore, we use machine learning to predict the viscosity of chalcogenide-based PCMs using data from related glasses. This approach broadens the range of phase-change materials that can be reliably simulated and provides a pathway toward the systematic design of dynamically tunable metasurfaces.

Keywords: multiphysics simulations; phase-change materials; machine learning; molecular dynamics simulations