6. Outlook
Field ion microscopy has provided a unique vantage point for atomic scale characterization of materials. The inherent resolution of FIM and the ability to also gather three dimensional information from materials through 3DFIM, puts the technique far ahead of any other contemporary microscopy technique. But the lack of advanced data processing and extraction routines are one of the reasons that likely hindered FIM from becoming a mainstream characterization technique. This article has detailed a burgeoning framework that will allow full data extraction, which we hope will support the renewed interest in FIM and its extension to 3DFIM. The pure image processing techniques have helped us to semi-automate the data extraction from FIM images. The application of machine learning algorithms to our data not only showed the behavior of field evaporation but also helped us to improve the accuracy of the reconstructed data. The layer by layer field evaporation behavior was evident when using PCA or Isomap algorithms on the 3DFIM data. Also using Isomap clustered the images better with respect to number of atoms an image had on its first terrace. The semi automated data extraction routine based on image processing has been helpful not only to improve the data extraction but also helps us to get much more accurately labeled data which can be used as a training dataset for supervised machine learning. With the use of such advanced algorithms for data extraction, we hope not only to completely automate the data extraction from 3DFIM but also identify and characterize various material defects. Apart from data extraction, machine learning has been gaining popularity in identifying fundamental physics phenomena [51]. The physics of image formation in FIM is still not completely understood, and it is our firm belief that machine learning can be a powerful tool in this direction.
The discussed methods have been implemented in a set of Python routines. These routines are available upon request to the authors.
