A large number of atom probe tomography (APT) datasets from past experiments were collected into a database to conduct statistical analyses. An effective way of handling the data is shown, and a study on hydrogen is conducted to illustrate the usefulness of this approach. We propose to handle a large collection of APT spectra as a point cloud and use a city block distance-based metric to measure dissimilarity between spectra. This enables quick and automated searching for spectra by similarity. Since spectra from APT experiments on similar materials are similar, the point cloud of spectra contains clusters. Analysis of these clusters of spectra in this point cloud allows us to infer the sample materials. The behavior of contaminant hydrogen is analyzed and correlated with voltage, electric field, and sample base material. Across several materials, the H2+ /H+ ratio is found to decrease with increasing field, likely an indication of postionization of H2+ ions. The absolute amounts of H2+ and H+ are found to frequently increase throughout APT experiments.
Keywords: atom probe tomography; cluster; data analysis; data mining; database; hydrogen; machine learning; mass-to-charge spectrum; t-SNE; zirconium.
© The Author(s) 2023. Published by Oxford University Press on behalf of the Microscopy Society of America.