SurfaceScience21 – wissenschaftliches Programm

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

O: Fachverband Oberflächenphysik

O 109: Poster Session VIII: Poster to Mini-Symposium: Machine learning applications in surface science III

O 109.5: Poster

Donnerstag, 4. März 2021, 13:30–15:30, P

Prediction of energetics in nucleation and non-equilibrium growth using machine learning — •Thomas Martynec¹, Sabine H. L. Klapp¹, and Stefan Kowarik² — ¹Technische Universität Berlin — ²Karl-Franzens-Universität Graz

Machine learning is playing an increasing role in the discovery of novel materials and may also facilitate the search for optimum growth conditions of crystals and thin films of these materials. We demonstrate that a convolutional neural network that is trained on snapshots of surface configurations can predict the underlying lateral binding energy and diffusion barrier. Specifically, a single KMC image of the morphology is sufficient to determine the energy barriers with high accuracy for energies in the range of 100 - 550 meV. The CNN can also make correct predictions for images with noise and lower than atomic-scale resolution. We expect our machine learning method to be of use for both, fundamental studies of growth kinetics and for faster optimization of low defect materials growth.