An Adjustment Strategy for Tilted Moiré Fringes via Deep Q-Network

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this study, the authors proposed a leveling strategy using a Deep Q-Network (DQN) algorithm to improve lithography overlay accuracy by correcting wafer tilt. A convolutional neural network (CNN) processes tilt images, and the DQN adjusts angles with high precision. Simulations showed angle measurement accuracy of 0.0011 degrees (horizontal) and 0.0043 degrees (vertical), enabling effective tilt correction.

Overall, the manuscript is clearly written. However, the originality and contribution of the paper are low, and the accuracy and depth of the results’ analysis need enhancement. The reviewer has the following comments and concerns:

-      The introduction section needs to be enhanced and explicitly state the main contributions of the paper and what sets it apart from previous research in the field.

-      The authors should validate their findings against previously published work.

-      The conclusions should not just repeat the data. Instead, they should highlight the significance of the findings for the field as well as for general interests.

-      There are repetitive descriptions of similar concepts, such as the interference intensity of the upper and lower parts of the tilted Moiré fringe. This redundancy can be reduced for better clarity.

-      The terms used to describe the tilting scenarios and angular deviations are sometimes inconsistent, which can confuse the reader.

-      While the mathematical models and algorithms are well-explained, there is a lack of contextual explanation about why certain models or approaches were chosen over others. Providing a brief comparison or rationale would enhance understanding.

-      Recent references should be added to the manuscript.

Comments on the Quality of English Language

The language needs improvement, as there are a few grammatical errors and some phrases that should be fixed to make the text clearer and more professional.

Author Response

Thank you very much to the reviewer for your valuable comments.  Based on your comments, the modifications are as follows.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this manuscript, the authors achieved an angular measurement accuracy of 0.0011 degrees for the wafer's horizontal plane tilt environment model and 0.0043 degrees for the vertical plane tilt environment model by the DQN algorithm. These results are impressive. The paper is suitable for publication in Photonics after the following concerns have been addressed.

1. The Deep Q-Network-based algorithm mentioned in the abstract of the paper is used as the environment configuration with four consecutive frames of input values, so why not increase the amount of data input? Generally, the more inputs, the more data can be collected and the smaller the calculation error will be. Here, is it because four frames of data are enough to calculate the environment variables or the algorithm itself can not achieve a higher scale of inputs? Is it the limitation of the algorithm itself or other factors?

2. In this paper, the authors have developed a DQN algorithm Deep Q-Network (DQN) algorithm, including its three components: the CNN-Behaviour network, the CNN-Target network, and the experience replay pool, which parts are the authors' work? What tuning parameters were involved or improved? What improvements were made? It is not clear enough in the paper.

3. In the actual lithography alignment process, the deviation of the manually placed mask from the center of the substrate is usually only a few degrees, while the in-plane and out-of-plane tilt angles in the training model in the article are all based on a 30-degree reference, is this far from the reality?

4. Figure 8 illustrates the measurement error of 0.00025 can be achieved after applying the DQN algorithm leveling. This result is excellent, but how to achieve the accuracy of the motion calibration and detection in practical applications?

Author Response

Thank you very much to the reviewer for your valuable comments.  Based on your comments, the modifications are as follows.

Author Response File: Author Response.pdf