Based on Deep Learning CD-SEM Image Defect Detection System

Yan, Shijia; Ding, Shenglan; Wang, Sen; Luo, Cong; Li, Lei; Ai, Juan; Shen, Qiang; Xia, Qing; Li, Zhi; Cheng, Qilin; Li, Shilin; Dai, Hongwei; Hu, Xiangang

doi:10.1109/cstic55103.2022.9856857

Cited by 3 publications

(3 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the majority of these past works use SEM image only as input, whereas the current work uses both SEM image and design layout image for more accurate detection. For example, works such as [3,4,5,6] similarly use YOLO but the focus is on finding certain patterns in SEM images without comparing with intended design, which means that if a pattern such as contact is missing, it would not get detected as defect as there is no reference for comparison. Combining SEM image and layout image into a single input as 2 channels has been done in [7] but the work uses Generative Adversarial Network (GAN) model for defect detection and is constrained to only two defect types.…”

Section: Related Workmentioning

confidence: 99%

Fast and accurate automatic wafer defect detection and classification using machine learning based SEM image analysis

Choi,

Xie,

Greeneltch

et al. 2024

Metrology, Inspection, and Process Control XXXVIII

View full text Add to dashboard Cite

Performing accurate and timely Scanning Electron Microscope (SEM) image analysis to identify wafer defects is crucial as it directly impacts manufacturing yield. In this paper, a machine learning (ML) based approach for analyzing SEM images (from wafer inspection machines) to locate and classify wafer defects is proposed. A state-of-the-art one-stage objection detection model called YOLOv8 (You Only Look Once version 8) is used as it offers a good balance between accuracy and inference speed. Experimental results confirm that an ensemble model composed of multiple YOLOv8 models can predict 6 types of defects with a mean Average Precision (mAP) of 0.789 (at IoU=0.5) for unseen test data consisting of real-world SEM images from 5 wafer fabs that have varying image qualities.

show abstract

Section: Related Workmentioning

confidence: 99%

Fast and accurate automatic wafer defect detection and classification using machine learning based SEM image analysis

Choi,

Xie,

Greeneltch

et al. 2024

Metrology, Inspection, and Process Control XXXVIII

View full text Add to dashboard Cite

show abstract

“…They use a classification model to identify the type of defect and generate the heatmap showing class activation maps to localize defect positions. Yan et al 5 proposed a defect inspection system using object detection model. They use a pretrained object detection model and apply transfer learning to reduce the cost of collecting many SEM images and labels.…”

Section: Deep Learning-based Defect Inspectionmentioning

confidence: 99%

“…They use a classification model to identify the type of defect and generate the heatmap showing class activation maps to localize defect positions. Yan et al 5 . proposed a defect inspection system using object detection model.…”

Section: Related Workmentioning

confidence: 99%

Trainable die-to-database for large field of view e-beam inspection

Fukuda

Ouchi

Ishikawa

et al. 2022

J. Micro/Nanopattern. Mats. Metro.

View full text Add to dashboard Cite

Extreme ultraviolet lithography has advanced microfabrication of semiconductor devices toward the sub-10-nm generation. In this situation, stochastic defects increase and hence process evaluation requires an entire wafer inspection at high speed. To satisfy this requirement, a large field of view (FoV) inspection with low-resolution enables us to inspect an entire wafer within an acceptable time because the throughput of e-beam inspection depends on imaging resolution. However, low-resolution images are difficult to inspect at high precision using conventional methods because of a smaller photographed defect size and worse signal-to-noise ratio. Moreover, deformation caused by the manufacturing process and larger distortion caused by large FoV result in false detections when we apply die-to-database (D2DB) inspection. To solve these issues, we propose trainable D2DB inspection, which predicts a pixel-value distribution of normal images from a corresponding design layout. The proposed method is robust to lowresolution images because it considers noise and acceptable deformation as variance of the learned distribution. In addition, by introducing a model to predict a misalignment between a design layout and inspection image, trainable D2DB becomes robust to image distortion. Experiments show that trainable D2DB can perform high-precision inspection on images with large noise and image distortion.

show abstract