High-Resolution, High-Sensitivity Crystal Structure Observation Using STEM:
We used STEM to observe the crystal sites occupied by single-atom dopants in Eu-doped SiAlON phosphors synthesized by Hirosaki et al. at NIMS. Although these atoms cannot be detected in conventional TEM images, Eu atoms are visible in high signal-to-noise ratio (SNR) images obtained through multiple measurements using annular dark-field (ADF) imaging. Through simulation-based analysis, we also identified the TEM sample thickness and imaging conditions, thereby determining the positions of the single-atom Eu dopants.
Quantitative measurement of ADF images provides information regarding atomic number Z. We found that as Z increases and the acceleration voltage decreases, the relationship deviates from the so-called power law for Z. By performing quantitative comparisons with simulations, it is also possible to identify atoms in ADF images.

Microstructural Analysis Using 4D-STEM and Machine Learning:
The technique of acquiring a large number of diffraction patterns while scanning the STEM probe is called 4D-STEM. We first observed atomic resolution in 2011. In 2021, we were the first to combine non-negative matrix factorization (NMF), a machine learning technique, with 4D-STEM. Furthermore, we independently developed a clustering method based on the similarity of diffraction patterns and used it to analyze the nanosized crystalline structures precipitated in metallic glasses through heat treatment (2024). Using these methods, we have also succeeded in polarity mapping of single-layer MoS₂ (2025). Some of our scripts for DigitalMicrograph (Gatan Inc.) are available on our group’s website and on GitHub.