ADI-to-AEI image generation result judgment method and device

Abstract

The invention discloses a method for judging an ADI-to-AEI image generation result. The method comprises the following steps: identifying the similarity between a generated image and an AEI image by adopting a discriminator and forming a similarity value; the generated image is an image generated by inputting the ADI image into the image generation network. Contour line judgment is conducted, specifically, multiple layers of first contour lines corresponding to different depths of the generated image and multiple layers of second contour lines corresponding to different depths of the AEI image are obtained, and the depth of the first contour line and the depth of the second contour line on the same layer are the same; and obtaining a contour line difference value between the first contour line and the second contour line of each layer. And judging an ADI-AEI image generation result by combining the similarity value and the difference value of each layer of contour line. The invention further discloses a device for judging the image generation result from the ADI to the AEI. According to the invention, the accuracy of the ADI-to-AEI image generated by the image generation network can be improved.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor integrated circuit manufacturing, and in particular to a method for judging the result of ADI to AEI image generation; the present invention also relates to a device for judging the result of ADI to AEI image generation. Background Art

[0002] In recent years, artificial intelligence led by deep neural networks has developed rapidly, especially the generative artificial intelligence boom at the end of 2022, which has continued to this day. As the cornerstone and pioneer of the high-tech industry, the chip manufacturing industry, especially the wafer foundry, will naturally not be absent.

[0003] Graph-to-graph based generative models have various applications.

[0004] like Figure 1 As shown, it is an input and output diagram of a building generated based on a label in an existing deep neural network. When the label image 101 is input into the model corresponding to the deep neural network, a building image 102 is output. The model corresponding to the deep neural network needs to be machine-learned so that the finally generated building image 102 is consistent with the actual building image.

[0005] like Figure 2 As shown, it is an input and output diagram of generating photos based on a map in an existing deep neural network; when the map 103 is input into the model corresponding to the deep neural network, the photo 104 will be output.

[0006] In semiconductor manufacturing, the graphics on the wafer need to be defined by the layout, and the graphics will gradually shift at different stages of semiconductor manufacturing. The image-to-image generative model can be applied to the semiconductor manufacturing process. At this time, it can be used to study the potential correlation between images at different stages in the process, and then obtain the final layout appearance through certain input information.

[0007] like Figure 3 As shown, there are images of different stages in the existing semiconductor manufacturing process; the layout includes a design layout 201, and the design layout 201 needs to be transferred to a mask to form a mask layout 203. Before forming the mask layout 203, an OPC correction 202 needs to be performed, and the OPC correction 203 includes an OPC correction based on a rule or a model.

[0008] After the mask is made, it will be used in wafer manufacturing. The pattern transfer includes:

[0009] Exposure is performed, which results in a light intensity map 205 , which can be simulated by an optical model 204 .

[0010] An after-development inspection (ADI) image 207 is performed. The ADI image 207 can be simulated by a photoresist model 206 .

[0011] After that, etching is performed to form an after-etch inspection (AEI) image 209 . The AEI image 209 is related to a chemical reaction and there is no good model to simulate it.

[0012] Since there is no good model to simulate the ADI image 207 to the AEI image 209, a deep neural network can be used to implement a generative model for generating AEI images based on ADI images, which can realize the generation of ADI to AEI images. In the deep neural network of the generative model for generating AEI images based on ADI images, a machine learning method needs to be used for learning. During the learning process, a discriminator needs to be used to judge the image generation results. If the generation results are accurately judged, the accuracy of the final generated image will also be improved.

[0013] like Figure 4 FIG. 1 is a flowchart of a method for judging the result of the existing ADI to AEI image generation, comprising the following steps:

[0014] The ADI image 301 can be converted by the graphics generation network 302 to output the generated image 303 .

[0015] The generated image 303 and the corresponding AEI image 304 in the sample set are input into the discriminator 305 for similarity comparison and forming a similarity value.

[0016] As shown by arrow line 306 , the similarity value will be fed back to the image generation network 302 , and the image generation network 302 will further iterate according to the fed-back similarity value to adjust the model parameters, and finally make the generated image 303 and the AEI image 304 closer, thereby improving the accuracy of the generated image 303 .

[0017] As can be seen from the above, for the image-to-image model of ADI generating AEI, the existing technology uses image similarity to determine whether the image generation is successful or not. However, image similarity can easily pass (PASS) if the background is similar. However, in some specific graphics, there are actually obvious deviations in some details.

[0018] like Figure 5 It is a comparison diagram of the ADI image, AEI image and generated image corresponding to the existing method for judging the generation result of ADI to AEI image; Figure 5In the figure, the ADI image 301, AEI image 304 and the generated image 303 corresponding to the box 307 are detected by the discriminator 305 and are passed, but the analysis of the ends of each graphic shows that Figure 5 The graphics of the areas indicated by the dotted boxes corresponding to the ends are further enlarged, and the enlarged graphics are graphics 308a, 308b and 308c respectively. It can be seen that there is an obvious deviation between graphics 308b and 308c, which will affect the accuracy of the generated image. Summary of the invention

[0019] The technical problem to be solved by the present invention is to provide a method for judging the generation result of an ADI to AEI image, which can improve the accuracy of the ADI to AEI image generated by an image generation network. To this end, the present invention also provides a device for judging the generation result of an ADI to AEI image.

[0020] In order to solve the above technical problems, the method for judging the ADI to AEI image generation result provided by the present invention includes:

[0021] A discriminator is used to discriminate the similarity between the generated image and the AEI image and form a similarity value; the generated image is an image generated by inputting the ADI image into the image generation network.

[0022] Perform contour line judgment, including:

[0023] A plurality of first contour lines corresponding to different depths of the generated image and a plurality of second contour lines corresponding to different depths of the AEI image are obtained, and the depth of the first contour line and the depth of the second contour line of the same layer are the same.

[0024] A contour line difference between the first contour line and the second contour line of each layer is obtained.

[0025] The similarity value and the contour line difference of each layer are combined to determine the ADI to AEI image generation result.

[0026] A further improvement is that the contour line determination further includes:

[0027] A plurality of maximum contour line differences are obtained according to the maximum values ​​of the contour line differences of each layer.

[0028] The maximum contour line difference is used as a supervision criterion and added into the loop iteration of the image generation network.

[0029] A further improvement is that the image generation network includes a deep neural network.

[0030] A further improvement is that the loop iteration of the image generation network is the loop iteration of the image generation network in machine learning.

[0031] In order to solve the above technical problems, the device for judging the ADI to AEI image generation result provided by the present invention includes: a discriminator and a contour line judgment device.

[0032] The discriminator is used to discriminate the similarity between the generated image and the AEI image and form a similarity value; the generated image is an image generated by inputting the ADI image into the image generation network.

[0033] The contour line judgment device is used to achieve:

[0034] A plurality of first contour lines corresponding to different depths of the generated image and a plurality of second contour lines corresponding to different depths of the AEI image are obtained, and the depth of the first contour line and the depth of the second contour line of the same layer are the same.

[0035] A contour line difference between the first contour line and the second contour line of each layer is obtained.

[0036] The device for judging the ADI to AEI image generation result combines the similarity value and the contour line difference of each layer to judge the ADI to AEI image generation result.

[0037] A further improvement is that the contour line criterion device is also used to implement:

[0038] A plurality of maximum contour line differences are obtained according to the maximum values ​​of the contour line differences of each layer.

[0039] The maximum contour line difference is used as a supervision criterion and added into the loop iteration of the image generation network.

[0040] A further improvement is that the image generation network includes a deep neural network.

[0041] A further improvement is that the loop iteration of the image generation network is the loop iteration of the image generation network in machine learning.

[0042] The present invention adds contour line judgment on the basis of using a discriminator to identify the similarity of images, and multiple layers of contour line differences can be obtained in the contour line judgment. The similarity value obtained by the discriminator and the multiple layers of contour line differences are combined to judge the ADI to AEI image generation result. The contour line difference further reflects the detail deviation between images, which can avoid the defect that the similarity value of the discriminator cannot reflect the detail deviation between images when the backgrounds are similar. Therefore, the present invention can improve the accuracy of the ADI to AEI image generated by the image generation network.

[0043] In addition, compared with other methods of identifying generated images, the identification method of the present invention is more practical because contour lines of different heights directly correspond to critical dimension (CD) measurement, and the correction of the optical proximity correction (OPC) model of the layout uses the CD value. Therefore, the supervised method of the present invention is more suitable for machine learning of image generation networks in semiconductor manufacturing and is more meaningful. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0045] Figure 1 It is an input and output graph of a building generated based on labels in an existing deep neural network;

[0046] Figure 2 It is the input and output graph of an existing deep neural network for generating photos based on maps;

[0047] Figure 3 These are images of different stages in the existing semiconductor manufacturing process;

[0048] Figure 4 It is a flow chart of the existing method for judging the result of ADI to AEI image generation;

[0049] Figure 5 It is a comparison diagram of the ADI image, AEI image and generated image corresponding to the existing method for judging the generation result of ADI to AEI image;

[0050] Figure 6 is a flow chart of a method for judging the ADI to AEI image generation result according to an embodiment of the present invention;

[0051] Figure 7 It is a flow chart of a method for judging the result of ADI to AEI image generation according to a preferred embodiment of the present invention;

[0052] Figure 8 It is a schematic diagram of obtaining multiple layers of first contour lines corresponding to different depths of a generated image in the method for judging the ADI to AEI image generation result in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] like Figure 6 As shown, it is a flow chart of a method for judging the result of generating the ADI to AEI image 404 according to an embodiment of the present invention; Figure 7 FIG. 1 is a flow chart of a method for determining the result of generating the ADI to AEI image 404 in a preferred embodiment of the present invention. Figure 7 The method for judging the result of generating the ADI to AEI image 404 in the embodiment of the present invention includes:

[0054] Step S101 , using the discriminator 405 to discriminate the similarity between the generated image 403 and the AEI image 404 and form a similarity value; the generated image 403 is an image generated by inputting the ADI image 401 into the image generation network 402 .

[0055] In the method of the embodiment of the present invention, the image generation network 402 includes a deep neural network.

[0056] Step S102, performing contour line judgment, including:

[0057] Multiple layers of first contour lines corresponding to different depths of the generated image 403 and multiple layers of second contour lines corresponding to different depths of the AEI image 404 are obtained, and the depth of the first contour line and the depth of the second contour line of the same layer are the same.

[0058] like Figure 8 , which is a schematic diagram of obtaining multiple layers of first contour lines corresponding to different depths of a generated image in a method for judging an ADI to AEI image generation result according to an embodiment of the present invention, Figure 8 4 shows an enlarged view of the contour map 406 , in which a graphic 408 in the contour map 406 is enlarged. The graphic 408 includes an inner contour 408 a and an outer contour 408 b. Figure 8 The cross-sectional structure diagram along line 409 is also shown. From the cross-sectional structure, it can be seen that a groove is formed after patterning of the photoresist (PR) 410. The groove is the structure of the pattern 408. At different depths of the groove, the width of the groove, that is, the CD value, is different. Figure 8 In the embodiment, as the depth of the groove increases, the width of the groove gradually decreases. Therefore, the inner contour line 408a corresponds to the edge contour of the groove at the darker dotted line 411a, and the inner contour line 408b corresponds to the edge contour of the groove at the shallower dotted line 411b. Figure 8 In the figure, only two layers of contours are shown. Figure 8 Taking the first contour line as an example, the multi-layer second contour lines can be obtained by the same method.

[0059] By comparing the first contour line and the second contour line, contour line differences between the first contour line and the second contour line of each layer can be obtained.

[0060] Step S103 , combining the similarity value and the contour line difference of each layer to determine the ADI to AEI image 404 generation result.

[0061] In the method of the embodiment of the present invention, the contour line determination further includes:

[0062] A plurality of maximum contour line differences are obtained according to the maximum values ​​of the contour line differences of each layer.

[0063] The maximum contour line difference is used as a supervision criterion and added to the loop iteration of the image generation network 402.

[0064] like Figure 7 As shown, the arrow line 407 represents the combination of the image judgment result and the contour analysis result formed by the discriminator 405 to realize the supervision of the generated image 403, which is fed back to the image generation network 402 and the model parameters of the image generation network 402 are adjusted through loop iterations, so that the final generated image 403 meets the requirements.

[0065] In the method of the embodiment of the present invention, the loop iteration of the image generation network 402 is the loop iteration of the image generation network 402 in machine learning.

[0066] The method of the embodiment of the present invention adds contour line judgment on the basis of using the discriminator 405 to identify the similarity of images, and multiple layers of contour line differences will be obtained in the contour line judgment. The similarity value obtained by the discriminator 405 and the multiple layers of contour line differences are combined to judge the generation result of the ADI to AEI image 404. The contour line difference further reflects the detail deviation between the images, which can avoid the defect that the similarity value of the discriminator 405 cannot reflect the detail deviation between the images when the backgrounds are similar. Therefore, the method of the embodiment of the present invention can improve the accuracy of the ADI to AEI image 404 generated by the image generation network 402.

[0067] In addition, compared with other identification methods for generating the image 403, the identification method of the embodiment of the present invention is more practical, because the contour lines of different heights directly correspond to the critical dimension (CD) measurement, and the correction of the optical proximity correction (OPC) model of the layout uses the CD value. Therefore, the supervision method of the embodiment of the present invention is more suitable for the machine learning of the image generation network 402 in semiconductor manufacturing, and is more meaningful.

[0068] The device for judging the result of generating the ADI to AEI image 404 in the embodiment of the present invention includes: a discriminator 405 and a contour line judgment device.

[0069] The discriminator 405 is used to discriminate the similarity between the generated image 403 and the AEI image 404 and form a similarity value; the generated image 403 is an image generated by inputting the ADI image 401 into the image generation network 402. The image generation network 402 includes a deep neural network.

[0070] The contour judgment device is used to achieve:

[0071] Multiple layers of first contour lines corresponding to different depths of the generated image 403 and multiple layers of second contour lines corresponding to different depths of the AEI image 404 are obtained, and the depth of the first contour line and the depth of the second contour line of the same layer are the same.

[0072] The contour line difference between the first contour line and the second contour line of each layer is obtained.

[0073] The device for judging the generation result of the ADI to AEI image 404 combines the similarity value and the difference value of the contour lines of each layer to judge the generation result of the ADI to AEI image 404.

[0074] The contour line judgment device is also used to achieve:

[0075] A plurality of maximum contour line differences are obtained according to the maximum values ​​of the contour line differences of each layer.

[0076] The maximum contour line difference is used as a supervision criterion and added to the loop iteration of the image generation network 402.

[0077] The loop iteration of the image generation network 402 is the loop iteration of the image generation network 402 in machine learning.

[0078] The present invention has been described in detail above through specific embodiments, but these do not constitute limitations of the present invention. Without departing from the principle of the present invention, those skilled in the art may also make many variations and improvements, which should also be considered as the protection scope of the present invention.

Claims

1. A method for judging the result of ADI to AEI image generation, characterized in that: include: A discriminator is used to identify the similarity between the generated image and the AEI image and form a similarity value; The generated image is an image generated by inputting the ADI image into the image generation network; Perform contour line judgment, including: Obtaining multiple layers of first contour lines corresponding to different depths of the generated image and multiple layers of second contour lines corresponding to different depths of the AEI image, wherein the depth of the first contour line and the depth of the second contour line of the same layer are the same; Obtaining a contour line difference between the first contour line and the second contour line of each layer; The similarity value and the contour line difference of each layer are combined to determine the ADI to AEI image generation result.

2. The method for judging the ADI to AEI image generation result according to claim 1, characterized in that: The contour line determination further includes: Obtaining multiple maximum contour line differences according to the maximum values ​​of the contour line differences of each layer; The maximum contour line difference is used as a supervision criterion and added into the loop iteration of the image generation network.

3. The method for judging the ADI to AEI image generation result according to claim 2, characterized in that: The image generation network includes a deep neural network.

4. The method for judging the ADI to AEI image generation result according to claim 3, characterized in that: The loop iteration of the image generation network is the loop iteration of the image generation network in machine learning.

5. A device for judging the result of ADI to AEI image generation, characterized in that: include: Discriminator and contour judgment device; The discriminator is used to identify the similarity between the generated image and the AEI image and form a similarity value; The generated image is an image generated by inputting the ADI image into the image generation network; The contour line judgment device is used to achieve: Obtaining multiple layers of first contour lines corresponding to different depths of the generated image and multiple layers of second contour lines corresponding to different depths of the AEI image, wherein the depth of the first contour line and the depth of the second contour line of the same layer are the same; Obtaining a contour line difference between the first contour line and the second contour line of each layer; The device for judging the ADI to AEI image generation result combines the similarity value and the contour line difference of each layer to judge the ADI to AEI image generation result.

6. The device for judging the ADI to AEI image generation result according to claim 5, characterized in that: The contour line criterion device is also used to achieve: Obtaining multiple maximum contour line differences according to the maximum values ​​of the contour line differences of each layer; The maximum contour line difference is used as a supervision criterion and added into the loop iteration of the image generation network.

7. The device for judging the ADI to AEI image generation result according to claim 6, characterized in that: The image generation network includes a deep neural network.

8. The device for judging the ADI to AEI image generation result according to claim 7, characterized in that: The loop iteration of the image generation network is the loop iteration of the image generation network in machine learning.

Images