Solving label errors related to electron beam defect images
The method enhances defect classification accuracy in electron beam defect images by cleaning the dataset and applying supervised training with label rejection rules, addressing the inaccuracies caused by manual labeling.
WO2026142784A1PCT designated stage Publication Date: 2026-07-02APPL MATERIALS ISRAEL LTD +1
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- APPL MATERIALS ISRAEL LTD
- Filing Date
- 2025-10-20
- Publication Date
- 2026-07-02
Smart Images
Figure US2025051689_02072026_PF_FP_ABST
Abstract
An electron beam defect classification system, that includes (a) an imager that is configured to generate an electron beam image of a region of a sample by illuminating the region with an electron beam; and (b) a processing circuit configured to process the electron beam image to provide a defect classification result. The processing circuit includes an artificial intelligence processor that is configured to apply artificial intelligence processing on the electron beam image to provide the defect classification result. The artificial intelligence processing represented an artificial intelligence process that exhibits an accuracy that is contributed to a training of the artificial intelligence process using a dataset that has undergone a label error reduction process.
Need to check novelty before this filing date? Find Prior Art