AlignMixup-based classification of mixed-type defect patterns in wafer bin maps
- Resource Type
- Conference
- Authors
- Yu, Qingqing; Lin, Degui
- Source
- 2023 4th International Conference on Mechatronics Technology and Intelligent Manufacturing (ICMTIM) Mechatronics Technology and Intelligent Manufacturing (ICMTIM), 2023 4th International Conference on. :480-485 May, 2023
- Subject
- Components, Circuits, Devices and Systems
Computing and Processing
Power, Energy and Industry Applications
Robotics and Control Systems
Training
Semiconductor device modeling
Tensors
Mechatronics
Pattern classification
Production
Data augmentation
Defect pattern classification
mixed-type defects
WBM classification
AlignMixup
- Language
Defect pattern classification of wafer bin maps (WBMs) is important owing to accurate classification is helpful in manufacture procedure improvement and so as to avoid further defects. Though the newest study of single defect WBM has made great progress, research seldom focus on the recognition of mixed-type defect WBMs possibly due to insufficient mixed-type defects data which is necessary during model training. Therefore, we suggest an AlignMixup-based classification approach to train convolutional neural networks with only a single defect WBMs data for mixed-type defects classification. Unlike previous approaches based on synthesizing mixed-type defect samples before model training, our method generates mixed-type defect WBMs adopting AlignMixup a powerful data augmentation method for model training. In the classification of WBM with two defect types, our method makes improvement by 16.9 % compared to previous settlements without consideration of mixed-type defect patterns. We performed on an real-world dataset of WBMs to prove the effectiveness and applicability of this approach. AlignMixup-based classification arrangement performs better than the state-of-the-art mixup approaches on five different benchmarks.