An Expandable Yield Prediction Framework Using Explainable Artificial Intelligence for Semiconductor Manufacturing

Round 1

Reviewer 1 Report (New Reviewer)

In general, the article has a reasonable framework and clear logic, but the following questions still need to be answered.

1. In the introduction, the contribution of this paper should be emphasized. At present, the author does not highlight the innovation of the article

2.The literature review part can be appropriately added, and the research literature in the last three years is relatively fewer

3. The author proposed“The Pearson correlation coefficients are calculated, and the dimensions of input data are reduced in case of coefficient values exceeding 0.95 [23].”

The author points out that which indicators should be eliminated through correlation coefficient judgment to reduce the input dimension? Then, if PCA is used for dimension reduction, will the effect be better.

Through PCA combined with neural network method, the literature (10.1109/ACCESS.2019.2920091) found that the prediction accuracy could be significantly improved. If the author refer and cite this article, the idea could be put into the future research.

4. The author selects many models. At present, the popular model is the deep learning model. Generally speaking, the prediction effect of the deep learning model will be significantly improved compared with most models, such as MLP, RF, SVR, KNN. Whether the author has considered adding a deep learning model to better improve the prediction effect.

5.The author chooses RMSE and MAE as the performance indicators of the calculation model. However, it is necessary to list the formula and have the literature basis. It is suggested to quote the literature: https://doi.org/10.3390/app11136199

6.Figure 4 is somewhat vague. the author can make it clearer

7.Lines 285-312 can be placed in the discussion section. At the same time, in this chapter, the author can also discuss the significance and value of this article in this field.

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf

Reviewer 2 Report (New Reviewer)

The paper is investigating the yield prediction for semiconductor manufacturing using the novel SHAP technique. The experiments are well-conducted. Readers may find the paper interesting. The authors successfully claimed their novelty, main output, and limitations. I liked the paper. However, the text is not ready for publication now. It should be more informative. The authors preferred a writing style for scholars who are familiar with these techniques. This style needs to be edited. I recommend a moderate revision.

Issue 1: The abstract is concise, but it should involve the names of the machine learning algorithms you used—at least the best performers in your analysis (to boost the visibility in search engines)—and some sort of summary of your quantitative results (accuracy and/or the input variables that you found as the most contributing ones).

Issue 2: Instead of using LGBM, ADA, and XGB, I recommend using LightGBM, AdaBoost, and XGBoost throughout the text. These abbreviations are more commonly used. Do not forget that scholars may search with a keyword combination like "LightGBM+SHAP."

Issue 3: I guess it would be better to move the details of the data set to materials and methods. In the results, you can give a dictionary of tuned hyperparameters with optimal values.

Issue 4 (Critical): I think that the authors need to prepare a table involving all input variables and their descriptive statistics (mean, max, min, etc.), including the data type (numerical, categorical). Furthermore, they need to mention the number of categories for each categorical data set (how many labels exist after the one-hot encoding process?)

Issue 5 (Optional): The authors may need to give formulae for RMSE and MAE in the methods.

Issue 6 (Critical): The paper lacks a good explanation of the SHAP method. I highly recommend adding a sub-section in Materials and Methods involving the mathematical explanations (please refer to these papers for simple math: https://doi.org/10.1007/s11069-022-05793-y and https://doi.org/10.1016/j.ress.2022.109045) and the treeSHAP library (please check Lundberg’s paper in Nature Machine Intelligence). What is a Shapley value? What are the key benefits of SHAP? I guess you get my point.

Issue 7 (Critical): In the results section, the authors need to give details about the SHAP summary plot, waterfall plot, etc. I mean, you are definitely familiar with this output, but please be more informative for those who are not familiar with the typical plots of the SHAP library (especially for the readers in your field).

 

Issue 8 (a question): I see that your best regressor is Random Forest. You started with a good story involving a variety of regressors for your experiments and using SHAP to explain your variables locally. However, what if your best predictor was SVM, KNN, or Lasso? How would you use the SHAP library? How can your result change once you use KernelSHAP or LinearSHAP?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (New Reviewer)

The author has carefully revised the article and it is suitable for publication in this journal.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

The reported work is related to estimation of semiconductor manufacturing yield. However, no clear evidence regarding semiconductor processes is given in the manuscript. The novelty part of the manuscript is poor because the work is only related to estimation. 

 

Author Response

We appreciate your dedicated time and effort in providing feedback on our manuscript and are grateful for the insightful comments. We have incorporated most of the suggestions made by the reviewers and revised the manuscript accordingly.

Please take a look at the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors had presented an interesting work, I really enjoy it. Before publication, I have the following concerns, I suggested the authors revised them.

1.     The abstract should contain a brief overview of the findings from the research, which is missing in the abstract.

2.     The authors discussed the monitoring processes in production lines in the introduction, thus I suggested the authors take a detailed literature study about this area, For instance, you can refer to doi.org/10.1016/j.procir.2018.03.207 and doi.org/10.1007/s13369-022-06946-8 .

3.     The manuscript lacks a statement about the limits of research and reality more applied in practice.

4.     Correct all the grammatical mistakes and typo errors

Author Response

We appreciate your dedicated time and effort in providing feedback on our manuscript and are grateful for the insightful comments. We have incorporated most of the suggestions made by the reviewers and revised the manuscript accordingly.

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Fig 4 and Fig 5 not clear.

Explain the figures in the text in numerical terms 

Author Response

We appreciate your dedicated time and effort in providing feedback on our manuscript and are grateful for the insightful comments. We have incorporated most of the suggestions made by the reviewers and revised the manuscript accordingly.

Please see the attachment.

Author Response File: Author Response.pdf