Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 12: Scanning probe microscopy: light matter interaction at atomic scales
O 12.2: Talk
Monday, March 9, 2026, 15:15–15:30, HSZ/0403
Machine Learning for Atomic-Scale Molecular Discovery from Tip-Enhanced Raman Spectroscopy — Harshit Sethi1, Markus Juntilla1, •Orlando J Silveira1, and Adam S Foster1,2 — 1Aalto University, Espoo, Finland — 2Kanazawa University, Kanazawa, Japan
Tip-enhanced Raman spectroscopy (TERS) combines the nanoscale spatial resolution of scanning probe microscopy with the chemical specificity of Raman spectroscopy, offering a powerful tool for resolving both molecular structure and composition. However, extracting precise atomic-scale information from complex TERS hyperspectral data remains a significant challenge. In this work, we address this gap by developing an automated machine learning framework to decode simulated TERS images for molecular structure discovery. Our approach uses convolutional neural networks (CNNs) to analyze hyperspectral datasets generated by modeling the plasmonic enhancement at a probe tip and the vibrational properties of planar molecules via density functional theory. To handle varying numbers of vibrational modes, we introduce a spectral binning procedure that standardizes the input for network training. This model provides a foundational proof of concept for the autonomous identification of molecular structures from TERS data, advancing the development of automated platforms for atomic-scale materials characterization and discovery.
Keywords: Tip-enhanced Raman; Machine learning; Spectroscopy