München 2019 – wissenschaftliches Programm

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

P: Fachverband Plasmaphysik

P 5: Helmholtz Graduate School I

P 5.1: Hauptvortrag

Montag, 18. März 2019, 14:00–14:30, HS 21

On plasma-surface model coupling realized through machine learning — •Jan Trieschmann, Florian Krüger, Tobias Gergs, and Thomas Mussenbrock — Brandenburg University of Technology Cottbus-Senftenberg, Electrodynamics and Physical Electronics, Siemens-Halske-Ring 14, 03046 Cottbus, Germany

The time and length scales of the dynamics at the solid surface and in the gas-phase during sputter deposition span orders of magnitude. While methods to kinetically describe the surface processes and the gas-phase transport take advantage of solving the respective sub-problem independently, their consistent coupling is a prerequisite. A viable plasma-surface model bridges these sub-models, allowing for complex surface and gas compositions encountered in reactive sputtering. For this objective, a machine learning plasma-surface interface is proposed based on a multilayer perceptron network. The latter has been trained and verified with sputtered particle energy and angular distributions obtained from TRIDYN simulations [1] for Ar sputtering an Al-Ti composite target. As verified with reference data, the trained network accurately predicts sputtered particle distributions for arbitrary incident ion energy distributions, which have not been previously trained. To conclude, additional examples where machine learning model interfaces may establish a reliable sub-model coupling are discussed.

This work is supported by the German Research Foundation (DFG) in the frame of transregional collaborative research centre SFB-TR 87.

[1] W. Möller and W. Eckstein, Nucl. Instr. and Meth. B2, 814 (1984)