Regensburg 2025 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 16: Scanning Probe Techniques: Method Development
O 16.8: Talk
Monday, March 17, 2025, 16:45–17:00, H25
Image-to-molecule translation for high-resolution SPM images — •Lauri Kurki1, Jie Huang1, Niko Oinonen1,2, and Adam S. Foster1,3 — 1Aalto University, Finland — 2Nanolayers Research Computing Ltd., UK — 3WPI-NanoLSI, Kanazawa University, Japan
Scanning tunnelling microscopy (STM) and atomic force microscopy (AFM) functionalized with a CO molecule on the probe apex capture sub-molecular level detail of the imaged sample [1]. However, the produced images are often difficult to interpret due to complex tip-sample interactions. To accelerate image analysis, we propose machine learning tools to extract sample properties directly from SPM images.
In recent years, there has been rapid development in image analysis methods in SPM in general and in particular for extracting atomic positions from AFM and STM images [2,3,4]. We build upon these models and achieve improved chemical and physical sensitivity compared to previous results [2]. Additionally, we explore equivariant neural networks [5] and compare their data efficiency and accuracy to traditional deep learning models.
[1] Cai et al., J. Am. Chem. Soc. 2022, 144, 44, 20227-20231 [2] Kurki et al., ACS Nano 2024, 18, 17, 11130*11138 [3] Alldritt et al., Sci. Adv. 2020; 6 : eaay6913 [4] Carracedo-Cosme et al., Nanomaterials 2021, 11, 1658. [5] Cesa et al., arXiv:1911.08251
Keywords: Scanning probe microscopy; Atomic force microscopy; Machine learning; Artificial intelligence; Equivariant deep learning