Abstract:
Identifying defective semiconductor wafers is a crucial and complicated aspect
of the manufacturing process. Developed systems confront difficulties in
capturing intricate defect patterns and the long-range relationship of defects. In
addressing these issues, this research proposes a hybrid system that comprises a
modified Mobilenet structure and the error-correcting output code (ECOC)
technique-based support vector machine (SVM) classifier. In the proposed
method, for feature extraction purposes, modified MobileNet architecture is
designed 1) by using the Swish function instead of the ReLU function in the
depth-wise separable convolution block of MobileNet structure for detecting
complex defective patterns and 2) by incorporating the multi-head attention
mechanism in MobileNet architecture for capturing long-range dependencies
on defective wafers. The ECOC-SVM approach is used to classify wafer defects.
Also, the histogram equalization technique is employed to improve the
visibility of small defects. The real-world semiconductor wafer dataset WM
811K is used in this study. The proposed system with histogram equalization
achieved a superior testing accuracy of 98.55% and better average values of
AUC (99.74%), recall (93.34%), precision (95.64%) and F1-score (94.42%). On the
other hand, the original version of the MobileNet model achieved testing
accuracy of 95.31%. The proposed model correctly identified complex patterns
in scratch, edge-ring, and donut types of defect wafers compared with other
models by using the swish function in the model. The comparative analyses of
the proposed system with other developed systems and state-of-the-art systems
are given in this research.
