論文ID: 2025EAP1012
Defect pattern detection in wafer bin maps (WBMs) is crucial for enhancing wafer quality, as it prevents the escalation of defects and the squandering of resources. To address this, we introduce a quantum-based Support Vector Machine (SVM) leveraging the quantum approximate optimization algorithm, termed QAOASVM. We employ Inception V3 to extract efficient and compact features from WBMs and apply QAOASVM for training and testing the data. When recognizing WBMs with two mixed defect types, our method outperforms the original mixup approach by over 7.4%, and achieves a 0.7% accuracy improvement compared to the Inception V3+SVM method. For mixed defect samples containing more than two types of defects (three-mixed and four-mixed), we observe gains of at least 4.2% (relative to SVM) and 0.9% (relative to Inception V3+SVM), respectively. Additionally, our method surpasses other state-of-the-art Convolutional Neural Network (CNN) methods in both single-type and mixed-type defect pattern recognition. Furthermore, QAOASVM requires only O(Nd) time, which is significantly more efficient than the O(N3) time complexity of traditional SVM. In summary, QAOASVM achieves higher accuracy with significantly reduced computational time.
