Die defect detection for integrated circuit using deep learning object detection techniques

Wong, Tack Hwa (2023) Die defect detection for integrated circuit using deep learning object detection techniques. Final Year Project, UTAR.

Preview

PDF Download (6Mb) | Preview

Abstract

Due to advances in semiconductor technology, the complexity of integrated circuit design continues to increase, resulting in ever-smaller defects appearing on these circuits. While some companies still rely on manual inspection for defect detection, these small and hard-to-see defects often lead to high false detection rates due to the human eye's limitations. This study aims to replace manual inspection with an approach that uses object detection to identify subtle defects, which are die rotation and die cracks. The YOLOv5n model is trained to capture ROI and strengthened by incorporating the SAM model to enhance segmentation performance. To address the issue of limited defect images, the StyleGANv2 model is trained to generate extra defect images. The YOLOv7- tiny model has been trained for object detection, with several enhancements made to the network architecture and loss function, pruning is also applied to decrease computational demands. The final model boosts a 3% increase in mAP@0.5 and 2.5% increase in mAP@0.5:0.95, while reducing parameters by 65.34% and GFLOPS by 33.84% compared to the original YOLOv7-tiny model. This study demonstrates that object detection can be an effective method for detecting defects in integrated circuits. The proposed method is able to achieve high accuracy and efficiency.

Actions (login required)