SurfaceScience21 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 71: Mini-Symposium: Machine learning applications in surface science I
O 71.2: Vortrag
Mittwoch, 3. März 2021, 14:15–14:30, R1
Automated Tip Functionalization and Image interpretation with Machine Learning in Atomic Force Microscopy — Benjamin Alldritt1, Chen Xu1, Prokop Hapala3, Niko Oinonen2, •Fedor Urtev1, Lauri Kurki2, Ondrej Krejci1, Filippo Federici Canova1, Juho Kannala1, Peter Liljeroth1, and Adam Foster1 — 1Aalto University, Espoo, Finland — 2University of Helsinki, Finland — 3Czech Academy of Sciences, Prague, Czechia
Atomic force microscopy (AFM) is ubiquitous nanoscale characterisation technique to measure a 3D map of surface roughness at atomic resolutions [1]. AFM data interpretation and quantitative analysis for complex mixtures of molecules and bulky 3D molecules can be difficult [2], due to the complex nature of contrast in AFM images, and need significant acceleration and automation to make AFM technique available to a wide range of laboratories and clinics. Here, we introduce a machine learning (ML) approach both for the preparation of AFM experiments and for data interpretation in AFM. For the first objective our method involves a convolutional neural network (CNN) that has been trained to analyse the quality of a CO-terminated tip. For the interpretation of AFM images, we introduce ML image descriptors characterising the molecular configuration, allowing us to predict the molecular structure directly. [1] L. Gross et al., Science, vol. 325, no. 5944, (2009). [2] O. M. Gordon and P. J. Moriarty, Mach. Learn. Sci. Technol., vol. 1, no. 2, (2020).