CN-121982016-A - Method, apparatus and storage medium for defect detection

Abstract

Methods, apparatuses, and storage media for defect detection are provided according to example embodiments of the present disclosure. The method comprises the steps of obtaining a test image of a part to be detected of a wafer and a reference image of a reference part corresponding to the part to be detected of the wafer, respectively generating corresponding multi-scale feature representations of the test image and the reference image in a defect detection model based on the test image and the reference image, generating difference features based on the corresponding multi-scale feature representations by using the defect detection model, wherein the difference features represent differences between the part to be detected and the reference part, and generating a defect detection result for the part to be detected based on the difference features. In this way, the accuracy and applicability of defect detection may be improved.

Inventors

• Request for anonymity

Assignees

• 全芯智造技术股份有限公司

Dates

Publication Date

20260505

Application Date

20260330

Claims (15)

1. 1. A method for defect detection, comprising: Acquiring a test image of a part to be detected of a wafer and a reference image of a reference part corresponding to the part to be detected in the wafer; generating respective multi-scale feature representations of the test image and the reference image based on the test image and the reference image, respectively, using a defect detection model; Weighting the respective multi-scale feature representation using a feature processing module of the defect detection model, generating the difference feature based on the processed respective multi-scale feature representation, and And generating a defect detection result for the part to be detected based on the difference characteristic.
2. 2. The method for defect detection of claim 1, wherein generating respective multi-scale feature representations of the test image and the reference image comprises: Inputting the test image and the reference image into the defect detection model respectively, and And extracting corresponding multi-scale characteristic representations of the test image and the reference image on a plurality of scales respectively by utilizing the defect detection model.
3. 3. The method for defect detection of claim 1, wherein weighting the respective multi-scale feature representations comprises: And weighting corresponding multi-scale feature representations of the test image and the reference image by utilizing the feature processing module so that a target area has higher influence on the processed corresponding multi-scale feature representations than other areas except the target area, wherein the target area is a difference area between the test image and the reference image.
4. 4. The method for defect detection of claim 1, wherein the difference features comprise a difference feature map, and generating difference features based on the processed respective multi-scale feature representations comprises performing at least one of the following for pixels in the test image and the reference image: computing differences between the multi-scale feature representations processed at the pixel to generate the difference feature map; At this pixel, stitching the processed corresponding multi-scale feature representations to generate the difference feature map, or At the pixel, the processed respective multi-scale feature representations are weighted fused to generate the difference feature map.
5. 5. The method for defect detection of claim 1, wherein the defect detection result comprises a defect probability map, each pixel value in the defect probability map being used to indicate a confidence that the corresponding location of the portion to be detected has a defect.
6. 6. The method for defect detection of claim 1, wherein prior to generating the respective multi-scale feature representations of the test image and the reference image, the method further comprises: registering the test image with the reference image so that the corresponding structures of the part to be detected and the reference part are matched under the same pixel coordinate system.
7. 7. The method for defect detection of claim 6, wherein registering the test image with the reference image comprises at least one of: Integral registration based on displacement and rotation relationship between the test image and the reference image, or Local registration is performed based on local geometric deformations between the test image and the reference image.
8. 8. A method for training a defect detection model, comprising: Constructing a plurality of training samples, the training samples in the plurality of training samples comprising a predetermined test image, a predetermined reference image, and a labeling image for one or more defective areas in the predetermined test image, and Inputting a defect detection model with the predetermined test image and the predetermined reference image as paired inputs to generate a defect detection result for the predetermined test image; Based on the defect detection result and the annotation image, updating the defect detection model with a composite loss function, the composite loss function including a first loss term for measuring classification accuracy and a second loss term for improving discrimination of the one or more defect regions in the event of sample class distribution imbalance.
9. 9. The method for training a defect detection model of claim 8, wherein generating defect detection results for the predetermined test image comprises: Generating respective multi-scale feature representations of the predetermined test image and the predetermined reference image, respectively, using a feature extraction module in the defect detection model; Generating, with a feature processing module in the defect detection model, a difference feature based on the respective multi-scale feature representation, the difference feature representing a difference between a portion of the predetermined test image to be detected and a reference portion of the predetermined reference image, and And generating a defect detection result for the predetermined test image based on the difference feature by using a decoding module in the defect detection model.
10. 10. The method for training a defect detection model of claim 9, wherein the feature extraction module comprises two feature extraction branches with shared weights, and generating respective multi-scale feature representations of the predetermined test image and the predetermined reference image comprises: inputting the predetermined test image and the predetermined reference image into the two feature extraction branches, respectively, and In the two feature extraction branches, respective multi-scale feature representations of the predetermined test image and the predetermined reference image are extracted on a plurality of scales, respectively, based on the shared weights.
11. 11. The method for training a defect detection model of claim 10, wherein the two feature extraction branches are symmetrical dual-branch coding structures.
12. 12. The method for training a defect detection model of claim 9, wherein the feature processing module comprises an attention module, and generating the difference feature comprises: processing respective multi-scale feature representations of the predetermined test image and the predetermined reference image with the attention module such that a target region, which is a region of difference between the predetermined test image and the predetermined reference image, affects the processed respective multi-scale feature representations to a higher extent than other regions other than the target region, and The difference features are generated based on the processed respective multi-scale feature representations.
13. 13. The method for training a defect detection model of claim 12, wherein the attention module is disposed on a jump connection path between the feature extraction module and the decoding module.
14. 14. An electronic device, comprising: At least one processing unit, and At least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, which when executed by the at least one processing unit, cause the electronic device to perform the method of any one of claims 1 to 7 or 8 to 13.
15. 15. A computer readable storage medium, having stored thereon a computer program, characterized in that the computer program is executable by a processor to implement the method according to any of claims 1 to 7 or 8 to 13.

Description

Method, apparatus and storage medium for defect detection Technical Field Embodiments of the present disclosure relate generally to the field of integrated circuit technology and, more particularly, relate to a method, apparatus, and storage medium for defect detection. Background With the continued development of integrated circuit fabrication processes and packaging technology, chip systems are evolving toward higher levels of integration, more complex structures, and finer manufacturing scales. In this process, the structural consistency and manufacturing reliability inside or between chips has a significant impact on the final product performance and yield. Especially in the wafer level manufacturing and advanced packaging scenarios, a plurality of chip units with identical or similar structures are often manufactured and combined at the same time, and the manufacturing process is inevitably affected by a plurality of factors such as process fluctuation, material difference, environmental factors and the like. Therefore, how to effectively analyze the structural differences between the chip units in a complex manufacturing environment, and to ensure the detection accuracy while simultaneously considering the detection efficiency and the adaptability, is a problem that needs to be focused in the fields of integrated circuit manufacturing and quality control. Disclosure of Invention In a first aspect of the present disclosure, a method for defect detection is provided. The method comprises the steps of obtaining a test image of a part to be detected of a wafer and a reference image of a reference part corresponding to the part to be detected in the wafer, respectively generating corresponding multi-scale feature representations of the test image and the reference image by using a defect detection model based on the test image and the reference image, generating difference features by using the defect detection model based on the corresponding multi-scale feature representations, wherein the difference features represent differences between the part to be detected and the reference part, and generating defect detection results for the part to be detected based on the difference features. In a second aspect of the present disclosure, a method for training a defect detection model is provided. The method includes constructing a plurality of training samples, wherein the training samples in the plurality of training samples comprise a predetermined test image, a predetermined reference image and a labeling image for one or more defect areas in the predetermined test image, and inputting a defect detection model with the predetermined test image and the predetermined reference image as paired inputs to generate a defect detection result for the predetermined test image, and updating the defect detection model based on the defect detection result and the labeling image by using a composite loss function, wherein the composite loss function comprises a first loss term for measuring classification accuracy and a second loss term for improving distinguishing ability for the one or more defect areas in case of sample class distribution imbalance. In a third aspect of the present disclosure, an electronic device is provided. The electronic device includes a processor, and a memory coupled to the processor. The memory has instructions stored therein that, when executed by the processor, cause the electronic device to perform a method according to the first or second aspect of the present disclosure. In a fourth aspect of the present disclosure, a computer-readable storage medium is provided. The computer readable storage medium has a computer program stored thereon. The computer program, when executed by a processor, implements the method according to the first or second aspect of the present disclosure. As will be appreciated from the following description, according to an embodiment of the present disclosure, a test image of a portion of a wafer to be inspected and a reference image of a reference portion of the wafer corresponding to the portion to be inspected are first acquired. Further, based on the test image and the reference image, respective multi-scale feature representations of the test image and the reference image are generated, respectively, using the defect detection model. Still further, using the defect detection model, a difference feature is generated based on the respective multi-scale feature representation, the difference feature representing a difference between the portion to be detected and the reference portion. Based on the difference characteristics, a defect detection result for the portion to be detected is generated. It should be understood that what is described in this summary is not intended to limit the critical or essential features of the embodiments of the disclosure nor to limit the scope of the disclosure. Other features of the present disclosure will become apparent from the