Advanced data mining in field ion microscopy

TL;DR

This paper reviews recent advances in data mining and machine learning techniques for processing field ion microscopy images, aiming to improve atomic resolution imaging and materials characterization.

Contribution

It introduces new machine learning and image processing methods for extracting data from FIM images, enhancing analysis of atomic structures and defects.

Findings

ML algorithms enable automated atom detection

Advanced image processing improves data extraction

Energy minimization enhances spatial resolution

Abstract

Field ion microscopy (FIM) allows to image individual surface atoms by exploiting the effect of an intense electric field. Widespread use of atomic resolution imaging by FIM has been hampered by a lack of efficient image processing/data extraction tools. Recent advances in imaging and data mining techniques have renewed the interest in using FIM in conjunction with automated detection of atoms and lattice defects for materials characterization. After a brief overview of existing routines, we review the use of machine learning (ML) approaches for data extraction with the aim to catalyze new data-driven insights into high electrical field physics. Apart from exploring various supervised and unsupervised ML algorithms in this context, we also employ advanced image processing routines for data extraction from large sets of FIM images. The outcomes and limitations of such routines are discussed, and we conclude with the possible application of energy minimization schemes to the extracted point clouds as a way of improving the spatial resolution of FIM.