[ThA1] III-V Nanostructures: From Spintronics to Advanced Lasers

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)

[ThA1-04] Deep Learning–Based Quantitative Analysis of InAs/GaAs Quantum Dot Ensembles via AFM Image Segmentation

〇Jinkwan Kwoen¹, Masahiro Kakuda¹, Yasuhiko Arakawa¹ (1. The Univ. of Tokyo (Japan))

Accurate segmentation of Stranski–Krastanov (S–K) III–V quantum dots (QDs) in atomic force microscopy (AFM) images is essential for reliable quantitative analysis and for establishing meaningful correlations between structural and optical properties. In this work, a deep learning–based segmentation approach was applied to InAs/GaAs S–K QDs and compared with a conventional rule-based method. While both approaches detected a comparable number of QDs, the rule-based method suffered from missed detections, insufficient separation of adjacent dots, and inaccurate boundary delineation. In contrast, the deep learning approach, implemented using Cellpose, provided more robust segmentation, resulting in clearer bimodal height distributions and physically meaningful correlations between AFM-derived height statistics and photoluminescence peak width.