ÄûÃʵ¼º½

Title of the thesis: Image interpretation methods for high-resolution scanning probe microscopy

Thesis defender: Lauri Kurki
Opponent: Associate Professor Oliver Hofmann, TU Graz, Austria
Custos: Professor Adam Foster, Aalto University School of Science

Interfaces play an important role in both natural and technological systems, and atomic-scale interactions and structures strongly influence their properties. Conventional optical microscopes cannot resolve structures this small, so atomic force microscopy (AFM) and scanning tunnelling microscopy (STM) are widely used to image matter at the atomic level. Although these techniques provide highly detailed images, interpreting them is not straightforward. Even slight three-dimensional features in the sample can distort image contrast, and in chemically complex samples the image alone does not reveal which species are actually present.

The aim of this work has been to improve the methods used for identifying samples from high-resolution images. Traditional identification relies on simulations and requires substantial manual effort and computational resources, as the researcher must simulate a large number of possible structures and compare them with the measured image. If this process can be automated, researchers can focus on studying new phenomena rather than spending time on manual identification. The main goal of this dissertation was therefore to develop a machine-learning-based tool for automatic interpretation of STM images. In addition, the work advances identification methods for AFM.

The results fall into two categories. First, traditional simulation-based identification was applied to study water molecules and organosilicon compounds. Second, machine-learning methods were developed in two subsequent studies, where organic molecules were identified automatically without external guidance. With these machine-learning approaches, identification takes only seconds, whereas traditional methods require much more time.

This research forms part of a broader effort to automate the entire imaging workflow, from operating the microscope to interpreting the resulting images. Although this dissertation focuses on only part of that challenge, it takes important steps toward fully automated atomic-scale imaging.

Keywords: scanning tunnelling microscopy, atomic force microscopy, machine learning, neural networks

Contact information: lauri.v.kurki@gmail.com 

Thesis available for public display 7 days prior to the defence at .