Inspection and Classification of Semiconductor Wafer Surface Defects Using CNN Deep Learning Networks

Abstract

Due to advances in semiconductor processing technologies, each slice of a semiconductor is becoming denser and more complex, which can increase the number of surface defects. These defects should be caught early and correctly classified in order help identify the causes of these defects in the process and eventually help to improve the yield. In today’s semiconductor industry, visible surface defects are still being inspected manually, which may result in erroneous classification when the inspectors become tired or lose objectivity. This paper presents a vision-based machine-learning-based method to classify visible surface defects on semiconductor wafers. The proposed method uses deep learning convolutional neural networks to identify and classify four types of surface defects: center, local, random, and scrape. Experiments were performed to determine its accuracy. The experimental results showed that this method alone, without additional refinement, could reach a top accuracy in the range of 98% to 99%. Its performance in wafer-defect classification shows superior performance compared to other machine-learning methods investigated in the experiments.

Automated Sewer Defects Detection Using Style-Based Generative Adversarial Networks and Fine-Tuned Well-Known CNN Classifier

“…Through retraining and finetuning the tailored CNNs as feature classifiers, the number of training samples and related computation time can be greatly saved when applied to new/novel tasks. Related research work and applications include: fine-tuned the SqueezeNet to classify steel surface defects [40], fine-tuned the Inception_v3 network to detect cracks by autonomous UAVs [41], trained the ResNet to detect cracks in different materials [42], and classification of semiconductor wafer surface defects [43]. These results show that pretrained CNNs and transfer learning can effectively improve the detection accuracy of structural defects and have great potential for sewer defect detection.…”

Production Flow Analysis in a Semiconductor Fab Using Machine Learning Techniques

“…Machine learning methods have been used before for analysis and decision making in semiconductor fabs. Some examples are work-in-progress prediction [13], lead time prediction [14], dynamic storage dispatching [15], vehicle traffic control [16], and wafer defect detection using image classification [17,18]. The machine learning method is classified as a data-driven approach that is suitable for cases with complicated relationships between many factors [19].…”

A Systematic Review of Deep Learning for Silicon Wafer Defect Recognition

“…Chien et al [56] demonstrated and compared two ways of building a CNN for wafer defect classification: a custommade CNN, carefully designed model for the specific dataset, and a pre-trained CNN fine-tuned by transfer learning. Both methods were equally good and performed better than the machine learning algorithms; SVM, logistic regression, random forest, and soft voting ensemble.…”

“…Their work demonstrated that transfer learning is a faster and more accurate way of wafer map defect classification. Chien et al [56] compared faster-R-CNN pre-trained on COCO and KITTI datasets and found the KITTI pre-trained model better than the other. A balanced data from the four classes of WM-811K was selected for model retraining.…”