Wafer image anomaly detection method, device, equipment and medium
Flat field correction is performed using a dark field reference frame and a preset flat field model. Frequency domain feature vectors are extracted and self-supervised learning is conducted. A target prototype library is constructed for anomaly detection, which solves the problem of wafer image detection methods being sensitive to illumination and material fluctuations, and improves detection accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHONGJIA MICROVISION (SHENZHEN) SEMICONDUCTOR TECHNOLOGY CO LTD
- Filing Date
- 2026-03-17
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, wafer image inspection methods are sensitive to illumination and material fluctuations, resulting in low inspection accuracy and a high risk of missed detections.
Flat field correction is performed using a dark field reference frame and a preset flat field model to remove slowly changing low-frequency background. Frequency domain feature vectors of bandpass energy, directionality, and periodic signatures are extracted. A target prototype library is constructed using self-supervised/contrastive learning. Anomaly detection is performed by combining density, distance, and neighborhood inconsistency terms.
It effectively reduces the impact of illumination and system drift on detection, enhances the stable characterization of periodic and stripe-like patterns, reduces the dependence on large-scale annotation, and improves the accuracy of anomaly detection.
Smart Images
Figure CN122415440A_ABST