research-article

Image Hash Layer Triggered CNN Framework for Wafer Map Failure Pattern Retrieval and Classification

Authors:

Minghao Piao,

Yi Sheng,

Jinda Yan,

Cheng Hao JinAuthors Info & Claims

ACM Transactions on Knowledge Discovery from Data, Volume 18, Issue 4

Article No.: 75, Pages 1 - 26

https://doi.org/10.1145/3638053

Published: 13 February 2024 Publication History

Get Access

Abstract

Recently, deep learning methods are often used in wafer map failure pattern classification. CNN requires less feature engineering but still needs preprocessing, e.g., denoising and resizing. Denoising is used to improve the quality of the input data, and resizing is used to transform the input into an identical size when the input data sizes are various. However, denoising and resizing may distort the original data information. Nevertheless, CNN-based applications are focusing on studying different feature map architectures and the input data manipulation is less attractive. In this study, we proposed an image hash layer triggered CNN framework for wafer map failure pattern retrieval and classification. The motivation and novelty are to design a CNN layer that can play as a resizing, information retrieval-preservation method in one step. The experiments proved that the proposed hash layer can retrieve the failure pattern information while maintaining the classification performance even though the input data size is decreased significantly. In the meantime, it can prevent overfitting, false negatives, and false positives, and save computing costs to a certain extent.

Appendix

A Evaluation Metric

A confusion matrix is a table that is often used to evaluate the performance of a classification. The data used for the evaluation should have labeled targets. The Table 5 shows the confusion matrix, and in the Table 5:

True Positive (TP): TP is the correctly predicted positive value which means that the value of the actual class positive is predicted as positive.

True Negative (TN): TN is the correctly predicted negative value which means that the value of the actual class negative is predicted as negative.

False Positive (FP): FP is the incorrectly predicted positive value which means that the value of the actual class negative is predicted as positive.

False Negative (FN): FN is the incorrectly predicted negative value which means that the value of the actual class positive is predicted as negative.

We can calculate Accuracy, Precision, and Recall based on these four parameters.

\begin{equation} Accuracy=\frac{TP+TN}{TP+TN+FP+FN}=\frac{Correctly \; Predicted}{All \; Positives \; and \; Negatives} \end{equation}

(6)

\begin{equation} Precision=\frac{TP}{TP+FP}=\frac{True \; Positive}{Total \; Predicted \; Positives} \end{equation}

(7)

\begin{equation} Recall=\frac{TP}{TP+FN}=\frac{True \; Positive}{Total \; Actual \; Positives} \end{equation}

(8)

B More Details of Experimental Results

B.1 More Details of Machine Learning based Hash Evaluation

More details of the Table 2 and Table 3 are shown in Appendix Table 6, Table 7, and Table 8. The evaluation is performed on five data sets, respectively.

B.2 More Details of Hash-CNN vs. 2D-CNN on Different Resized Images

The Figure 12 shows the average accuracy comparison of 2D-CNN and Hash-CNN. 2D-CNN is trained on resized images of 16 × 16, 32 × 32, and 64 × 64, respectively. Figure 13 shows the precision and recall comparison between Hash-CNN and 2D-CNN on training data.

B.3 More Details of Hash-CNN vs. 2D-CNN on Padded Image

The Figure 14 shows the max and min training accuracy of 2D-CNN and Hash-CNN during 100 epochs, and the Figure 15 shows the max and min test accuracy of 2D-CNN and Hash-CNN during 100 epochs. We can see that 2D-CNN outperforms Hash-CNN on training data. However, on the test data, as shown in the Figure 15, the performance of Hash-CNN becomes closer to 2D-CNN.

Fig. 14.

Fig. 15.

The Table 9 shows the comparison between 2D-CNN and Hash-CNN on the training and test data. We can see that the average precision of both models is higher than the accuracy. That means that false positives are a smaller part of our study. The average recall of 2D-CNN on training is lower than precision, while the average recall of Hash-CNN is higher than the other two metrics on training and test data. We are concerned with reducing false negatives and false positives, so we can say that Hash-CNN is better than 2D-CNN.

Source Code Availability Statement

All models, or codes that support the findings of this study are proprietary and may only be provided upon reasonable request to [email protected].

References

[1]

M. Abd. Al Rahman, Sebelan Danishvar, and Alireza Mousavi. 2021. An improved capsule network (WaferCaps) for wafer bin map classification based on DCGAN data upsampling. IEEE Transactions on Semiconductor Manufacturing 35, 1 (2021), 50–59.

Abstract

A Evaluation Metric

B More Details of Experimental Results

B.1 More Details of Machine Learning based Hash Evaluation

B.2 More Details of Hash-CNN vs. 2D-CNN on Different Resized Images

B.3 More Details of Hash-CNN vs. 2D-CNN on Padded Image

Source Code Availability Statement

References

Cited By

Index Terms

Recommendations

CNN and ensemble learning based wafer map failure pattern recognition based on local property based features

Ore Image Classification Based on Improved CNN

Channel modulus normalization for CNN image classification

Comments

Information

Published In

Publisher

Publication History

Permissions

Check for updates

Author Tags

Qualifiers

Funding Sources

Contributors

Other Metrics

Bibliometrics

Article Metrics

Other Metrics

Citations

Cited By

Login options

Full Access

View options

PDF

eReader

Full Text

Share

Share this Publication link

Share on social media

Affiliations