Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 47: New methods: Theory
O 47.2: Talk
Tuesday, March 10, 2026, 14:45–15:00, HSZ/0201
Innovative Approaches to Semiconductor Surface Oxidation Studies Using Active Learning and MLIP — •Ondrej Krejci1,2, Shola Adyemi2, Konstantinos Konstantinou2, and Milica Todorović2 — 1Department of Chemistry and Materials Science, Aalto University, Espoo, Finland — 2Department of Mechanical and Materials Engineering, University of Turku, Turku, Finland
Oxygen passivation of InAs surfaces critically affects material performance in electronic devices, but the nature of the oxide surface reconstruction is not well characterized. To address this, we employ a machine learning (ML) driven workflow. Starting from the ζ(4×2) reconstruction of pristine InAs(100) [1], we use Bayesian Optimization [2] to identify oxygen binding sites. This allows us to populate the surface with increasing number of oxygen atoms. The oxide models serve as input for a ML interatomic potential based on the MACE model [3], trained via the active learning method PALIRS [4]. The potential is used for molecular dynamics simulations to identify promising candidates for the oxidized InAs(100) surface reconstruction. Our workflow enables an efficient exploration of configurational space surpassing traditional computational approaches.
[1] Appl. Phys. A 75, 89 (2002)
[2] Npj. Comput. Mat. 5, 35 (2019)
[3] NeurIPS 35, 11423 (2022)
[4] Npj Comput. Mat. 11, 324 (2025)
Keywords: Surface structure; Semiconductor surfaces; Machine learning; Density functional theory; Active learning