Semantic image segmentation for semiconductor-based applications
Methods and systems for determining information for a specimen are provided. One system includes a computer subsystem and one or more components executed by the computer subsystem. The one or more components include a semantic segmentation model configured for assigning labels to each of multiple pixels in an image responsive to what is represented in each of the multiple pixels. The image is an image of a specimen generated by an imaging subsystem. The computer subsystem is configured for determining information for the specimen from the assigned labels and without a reference image for the specimen.
1. A system configured to determine information for a specimen, comprising:
a computer subsystem; and
one or more components executed by the computer subsystem;
wherein the one or more components comprise a semantic segmentation model configured for:
assigning labels to each of multiple pixels in an image responsive to what is represented in each of the multiple pixels, wherein the image is an image of a specimen generated by an imaging subsystem;
assigning a first of the labels to one or more of the multiple pixels in which a defect is represented and a second of the labels to one or more additional of the multiple pixels in which one or more structures on the specimen are represented; and
assigning a third of the labels to one or more further of the multiple pixels in which one or more additional structures on the specimen formed under the one or more structures are represented; and
wherein the computer subsystem is configured for determining information for the specimen from the assigned labels and without a reference image for the specimen.
2. The system of claim 1 , wherein the information determined for the specimen comprises predicted defect locations on the specimen.
3. The system of claim 1 , wherein the information determined for the specimen comprises predicted repeater defect locations on the specimen.
4. The system of claim 1 , wherein the semantic segmentation model is further configured for assigning a fourth of the labels to one or more other of the multiple pixels in which the defect is not represented, and wherein said assigning is performed without a reference image for the specimen.
5. The system of claim 1 , wherein the information determined for the specimen comprises one or more defect attributes of one or more predicted defect locations on the specimen.
6. The system of claim 1 , wherein assigning the first of the labels comprises assigning the first of the labels to the one or more of the multiple pixels in which one or more instances of a first type of a defect of interest are represented, and wherein the semantic segmentation model is further configured for assigning one or more other of the labels to one or more other of the multiple pixels in which the one or more instances of the first type of the defect of interest are not represented.
7. The system of claim 1 , wherein assigning the first of the labels comprises assigning the first of the labels to the one or more of the multiple pixels in which a first defect of interest is represented and a fourth of the labels to one or more different of the multiple pixels in which a second defect of interest is represented, wherein the semantic segmentation model is further configured for assigning one or more other of the labels to one or more other of the multiple pixels in which the first and second defects of interest are not represented, and wherein the first and second defects of interest are different types of defects of interest.
8. The system of claim 1 , wherein the labels assigned to each of the multiple pixels are selected from a predefined set of labels learned by the semantic segmentation model during training of the semantic segmentation model.
9. The system of claim 1 , wherein the semantic segmentation model does not comprise any fully connected layers.
10. The system of claim 1 , wherein the computer subsystem is further configured for training the semantic segmentation model for said assigning the labels with a training set comprising one or more patch images of one or more defects of interest generated by a deep learning model and labels assigned to each of multiple pixels in the one or more patch images.
11. The system of claim 10 , wherein the deep learning model comprises a generative adversarial network.
12. The system of claim 10 , wherein the training set further comprises one or more additional patch images of the one or more defects of interest or one or more additional defects of interest generated by the imaging subsystem and additional labels assigned to each of multiple pixels in the one or more additional patch images.
13. The system of claim 1 , wherein the computer subsystem is further configured for training the semantic segmentation model for said assigning the labels with a training set comprising one or more patch images of one or more defects of interest generated by the imaging subsystem and labels assigned to each of multiple pixels in the one or more patch images.
14. The system of claim 1 , wherein the imaging subsystem is a light-based imaging subsystem.
15. The system of claim 1 , wherein the imaging subsystem is an electron-based imaging subsystem.
16. The system of claim 1 , wherein the specimen is a wafer.
17. A non-transitory computer-readable medium, storing program instructions executable on a computer system for performing a computer-implemented method for determining information for a specimen, wherein the computer-implemented method comprises:
assigning labels to each of multiple pixels in an image responsive to what is represented in each of the multiple pixels by inputting the image into a semantic segmentation model included in one or more components executed by the computer system, wherein the image is an image of a specimen generated by an imaging subsystem, and wherein assigning the labels comprises;
assigning a first of the labels to one or more of the multiple pixels in which a defect is represented and a second of the labels to one or more additional of the multiple pixels in which one or more structures on the specimen are represented; and
assigning a third of the labels to one or more further of the multiple pixels in which one or more additional structures on the specimen formed under the one or more structures are represented; and
determining information for the specimen from the assigned labels and without a reference image for the specimen.
18. A computer-implemented method for determining information for a specimen, comprising:
assigning labels to each of multiple pixels in an image responsive to what is represented in each of the multiple pixels by inputting the image into a semantic segmentation model included in one or more components executed by a computer subsystem, wherein the image is an image of a specimen generated by an imaging subsystem, and wherein assigning the labels comprises;
assigning a first of the labels to one or more of the multiple pixels in which a defect is represented and a second of the labels to one or more additional of the multiple pixels in which one or more structures on the specimen are represented; and
assigning a third of the labels to one or more further of the multiple pixels in which one or more additional structures on the specimen formed under the one or more structures are represented; and
determining information for the specimen from the assigned labels and without a reference image for the specimen, wherein said inputting and determining are performed by the computer subsystem.