Machine-learning for atom probe tomography (T02#)

• Subject Area: Technical Chemistry
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 405553726

Atom probe tomography (APT) is a burgeoning characterization technique that provides compositional mapping of materials in three-dimensions at the near-atomic scale. Within the Collaborative Research Center HoMMage, APT has found application across projects on Permanent Magnets (Project A: Maximum Hysteresis) and Magnetocalorics (Project B: Minimum Hysteresis), and is performed within the service project Z01. The data from APT takes the form of a mass spectrum that provides the composition, and a point cloud that reflects the distribution of all the elements within the region-ofinterest of the material being investigated. Material-relevant data, including the local composition of a phase along with its morphology, or even partial structural information, must be extracted from this point cloud through the use of data mining techniques – including cluster-finding or nearest-neighbour algorithms today classified as machine-learning but used in the APT community for many decades. Data reconstruction and processing is most often done in the commercially-available software, which, being a Swiss-army knife, is not perfectly adapted to the analytical needs of the user, which has led to many adhoc data processing techniques developed by individual groups targeted to their needs. There is hence a need to facilitate access of these state-of-the-art tools through the commercial software, in particular when it comes to cutting-edge application of machine-learning algorithms to mine atom probe data, and build novel methods upon these. Project T02 is a partnership with Cameca Instruments Inc., the market leader with over 90% of the commercial atom probes in operation worldwide. Cameca has recently offered to give priority to PI Gault’s group to access a new version of their software with a python console that facilitate interacting with external packages or suite of functions. The aim of project T02 is threefold: first interface with the world-leading manufacturer of atom probes to ensure that the data reconstruction, processing and extraction software integrates all the state-of-the-art methods that have been developed, first those developed during Phase I of CRC HoMMage, but also more broadly; second, ensure interoperability of the software and data processing approaches across platforms, in liaison with Z-INF to guarantee compliance with FAIR data principles and facilitate bridging to atomistic simulations in strong collaboration with the CRC’s simulation groups (A05 Hongbin Zhang, A06 Baixiang Xu); third, establish world-leadership through novel developments of machine-learning-based approaches. T02 builds on the competence and expertise of PI Gault in APT, including on data reconstruction, processing and interfacing with other techniques, particularly electron microscopy and spectroscopy, along with the development of machine learning based data processing techniques. Strong interactions between T02 and Z01/Z02 will be established.

• DFG Programme: CRC/Transregios (Transfer Project)
• Subproject of: TRR 270: Hysteresis Design of Magnetic Materials for Efficient Energy Conversion: HoMMage
• International Connection: USA
• Applicant Institution: Technische Universität Darmstadt
• Business and Industry: CAMECA ; Science & Metrology Solutions.
• Project Head: Dr. Baptiste Jean Germain Gault, Ph.D.