Presentation Information
[1H08【榊奨励賞受賞講演】]Development of surface and interface analysis techniques integrating multi-dimensional spectromicroscopy and machine learning
*Naoka Nagamura1,2,3 (1. National Institute for Materials Science, 2. Tokyo University of Science, 3. Tohoku University)
Charge transfer processes are essential to devices from semiconductors to batteries. As devices miniaturize, heterointerfaces and defects increasingly affect performance, making spatially-resolved charge imaging critical. We developed an operando scanning photoelectron microscopy (SPEM) system with ~70 nm spatial resolution under bias, enabling XPS mapping of devices such as transistors, Li-ion batteries, and photocatalysts. The resulting multi-dimensional spectra present big data challenges, unsuited to conventional peak fitting. We propose a machine-learning-based method, using Gaussian Mixture models and EM algorithms for automated, high-throughput peak detection, implemented in our Python package “EMPeaks.” We also explore ML-based structure recognition from RHEED patterns, sparse modeling for high-resolution spectral imaging, and property extraction from Fermi surface imaging datasets. This talk will highlight our systems, analysis methods, and device application examples.