2025-11-11

半導体多層構造パターンのSEM画像に対する輪郭抽出アプローチの有効性評価

• Event: 第45回ナノテスティングシンポジウム(NANOTS2025)
• 村上慎治, 大家政洋, 岡本陽介, 中澤伸一, 丸山浩太郎, 山崎裕一郎, 御堂義博, 三浦典之
  
  第45回ナノテスティングシンポジウム(NANOTS2025), pp.40–43, Nov. 2025.

ABSTRACT

We evaluated two contour-extraction approaches for metrology on high-voltage SEM images: CVNet, which integrates an Active Contour Model with deep learning, and a segmentation-based method using U-Net+FCN. Both methods were assessed using synthetic SEM images containing controlled noise, low contrast, multilayer overlap, and other SEM-specific artifacts. Performance was compared across multiple pattern geometries in terms of contour extraction, region segmentation, and overlay displacement sensitivity relevant to overlay and EPE measurements. U-Net+FCN achieved slightly higher overall accuracy, while CVNet showed stronger consistency between contour- and region-based metrics. Both methods exhibited comparable sensitivity for overlay displacement measurement.

Key words: High voltage, EBI, SEM, D2DB, CD, EPE, Overlay, CVNet