SKM 2023 – wissenschaftliches Programm

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

O: Fachverband Oberflächenphysik

O 77: Scanning Probe Techniques: Method Development II

O 77.4: Vortrag

Donnerstag, 30. März 2023, 11:15–11:30, REC C 213

Machine learning: radical technique or plus ça change? The case for automated scanning probe microscopy — •Dylan Barker, Adam Sweetman, and Phil Blowey — University of Leeds

Atomic resolution scanning probe microscopy (SPM) provides a critical tool for studying the chemical and electronic structure of surfaces at the single atom scale, however, practically these techniques require a large amount of experimental time to manually prepare the scanning probe tip in-situ, usually via controlled indents into the surface. This apparently simple, but time-consuming process, is potentially an ideal candidate for automation using machine learning and computer vision techniques. Previous attempts to automate the classification of probe tips from topographical images have been made using machine learning methods [1-2], however using prior knowledge of the system in question we find it is also possible to classify the tip state using computationally simple image analysis methods such as Fourier ring correlation and cross-correlation. In this work I will present a comparison between "deterministic" image analysis methods and machine learning approaches for tip state classification. I will also address the known issue of small sample sizes for training ML techniques for SPM image classification, via an automated (scripted) data generation approach.

[1] Rashidi, M & Wolkow, R. A. ACS Nano 12, 5185-5189 (2018).

[2] Gordon, O. et al. Review of Scientific Instruments 90, 103704 (2019).