Attention-Enhanced YOLO Model for High-Precision Detection of Internal Chip Defects in Ultrasonic Imagery

Scanning Acoustic Microscopy (SAM) is an effective non-destructive testing technology widely used for detecting internal defects in integrated circuits. However, the current majority of ultrasonic defect detection primarily relies on traditional visual inspection methods, which suffer from low efficiency and high false detection rates. To tackle the above challenges, we present an ultrasonic defect detection approach built on the YOLOv8n model, enhanced through the integration of attention mechanisms and refined loss function optimization. Firstly, the Efficient Channel Attention (ECA) mechanism is integrated into the backbone of YOLOv8n to trade off detection accuracy and computational complexity. Furthermore, the Weighted Intersection over Union (WIoU) loss function is employed to refine feature capture accuracy while minimizing false positives in irrelevant regions. Compared to other state-of-the-art (SOTA) models, the proposed approach demonstrates improved performance in terms of both detection accuracy and speed on the dataset collected in this study. Through meticulous optimization, the proposed method in this paper is expected to further enhance the efficiency and measurement accuracy of chip internal defect detection in semiconductor manufacturing.