CN-122023359-A - Intelligent wafer pattern defect detection method and system based on multi-mode data fusion

CN122023359ACN 122023359 ACN122023359 ACN 122023359ACN-122023359-A

Abstract

The application discloses a multi-mode data fusion wafer pattern defect intelligent detection method and system, and relates to the field of industrial vision, wherein the method comprises the steps of configuring an industrial vision imaging system, synchronously collecting a first light path image and a second light path image of an optical substrate wafer pattern region, carrying out space registration and image enhancement processing to obtain a multi-mode registration image, extracting a first characteristic image and a second characteristic image, generating a pattern multi-characteristic image after fusion, carrying out comparison analysis with a preset gold template to locate a pattern abnormal candidate region, inputting the pattern multi-characteristic image and the pattern abnormal candidate region into a pre-trained AI vision recognition model, carrying out depth detection on the pattern multi-characteristic image under the guidance of the pattern abnormal candidate region, and outputting a defect recognition result. The application solves the technical problems of low recognition precision and high false missing judgment rate of the existing wafer pattern defect detection, and achieves the technical effects of improving the defect recognition precision and reducing the false missing judgment rate.

Inventors

FENG XIANGXU
WU MING
XU YUNMING
TANG CHANGYONG
SHEN JIE

Assignees

浙江蓝特光学股份有限公司
浙江蓝创光电科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The intelligent wafer pattern defect detection method based on multi-mode data fusion is characterized by comprising the following steps of: Configuring an industrial vision imaging system comprising at least a first imaging optical path and a second imaging optical path; Synchronously acquiring a first light path image and a second light path image of an optical substrate wafer pattern area through the industrial vision imaging system, wherein a preset optical function pattern is formed on the surface of the optical substrate wafer; Performing spatial registration and image enhancement processing on the first optical path image and the second optical path image to obtain a multi-mode registration image; based on the multi-mode registration image, extracting a first feature image and a second feature image, and generating a pattern multi-feature image after fusion; Comparing and analyzing the pattern multi-feature map with a preset golden template, and positioning a pattern abnormal candidate region; Inputting the pattern multi-feature map and the pattern abnormality candidate region into a pre-trained AI visual recognition model, and carrying out depth detection on the pattern multi-feature map by the AI visual recognition model under the guidance of the pattern abnormality candidate region, and outputting a defect recognition result.
2. The intelligent detection method for defects of a multi-modal data fusion wafer pattern according to claim 1, wherein comparing the pattern multi-feature map with a preset golden template for analysis, and locating a pattern anomaly candidate region comprises: Calculating the difference between the pattern multi-feature map and the golden template pixel by pixel to generate a pixel difference map; performing multi-scale Gaussian pyramid decomposition on the pixel difference graph to obtain a plurality of difference subgraphs; And respectively carrying out self-adaptive threshold segmentation on the plurality of difference subgraphs, mapping the segmented abnormal regions back to the original image scale, merging and morphologically optimizing, and removing isolated regions with the area smaller than a preset value to obtain the pattern abnormal candidate regions.
3. The intelligent detection method for defects of a multi-mode data fusion wafer pattern according to claim 2, wherein the golden template is a golden template feature map, and is generated by inputting standard reference data into the same feature extraction and fusion process as that of the pattern multi-feature map; wherein, the standard reference data is: Acquiring image data from a similar optical substrate wafer sample confirmed to be defect-free by using the industrial visual imaging system and generating a multi-mode registration image; or according to the design layout data of the optical functional pattern, calculating and rendering the generated multi-mode registration simulation image through the optical imaging simulation model.
4. The method for intelligently detecting defects of a wafer pattern according to claim 1, wherein the pattern multi-feature map and the pattern anomaly candidate region are input into a pre-trained AI visual recognition model, the AI visual recognition model performs depth detection on the pattern multi-feature map under the guidance of the pattern anomaly candidate region, and outputs a defect recognition result, and the method comprises the following steps: the AI visual recognition model is a twin network based on an encoder-decoder architecture and is integrated with a spatial attention guiding module; The spatial attention guiding module takes the pattern abnormal candidate region as a spatial attention mask, performs feature modulation on the input pattern multi-feature map, enhances the feature response in the candidate region, suppresses the feature response of the non-candidate region and generates an attention modulation feature map; The first branch of the twin network takes a local feature block corresponding to a candidate region in the attention modulation feature map as input to extract local detail features of defects, and the second branch of the twin network takes the whole attention modulation feature map as input to extract global context features containing the relationship between the candidate region and the peripheral pattern; and the feature fusion layer of the twin network fuses the local detail features and the global context features, performs defect semantic segmentation and classification based on the fused features, and outputs the defect recognition result.
5. The method for intelligently detecting defects of a wafer pattern according to claim 1, wherein the first imaging optical path is a bright field or dark field microscopic imaging optical path for acquiring a pattern of high-contrast two-dimensional geometry, and the second imaging optical path is a spectral imaging optical path for acquiring a material component of a pattern area or a polarization imaging optical path for acquiring film characteristics of the pattern area or a confocal imaging optical path for acquiring subsurface information of the pattern area.
6. The intelligent wafer pattern defect detection method based on multi-modal data fusion as set forth in claim 5, wherein performing spatial registration and image enhancement processing on the first and second optical path images to obtain a multi-modal registered image includes: extracting a first characteristic point set from the first light path image, extracting a second characteristic point set from the second light path image, and matching the first characteristic point set with the second characteristic point set by using a characteristic descriptor to obtain a plurality of pairs of matched characteristic point pairs; screening correct matching pairs from the plurality of pairs of matching feature points, and calculating a space geometric transformation matrix from the second light path image to the first light path image by using the correct matching pairs; Registering the second light path image by applying the space geometric transformation matrix to generate a second light path registration image; Denoising and self-adapting contrast enhancement processing are respectively carried out on the first light path image and the second light path registration image, so as to obtain a first enhanced light path image and a second enhanced light path image; And splicing the first enhanced light path image and the second enhanced light path image along the channel dimension, and performing inter-mode standardization processing on the spliced multi-channel image to generate the multi-mode registration image.
7. The method for intelligently detecting defects in a multi-modal data fusion wafer pattern as recited in claim 6, wherein the inter-modal normalization process is a Z-score normalization process.
8. The method for intelligently detecting defects of a wafer pattern by multi-modal data fusion according to claim 1, wherein the steps of extracting a first feature map and a second feature map based on the multi-modal registration image, and generating a pattern multi-feature map after fusion include: Respectively processing first-mode image data and second-mode image data in the multi-mode registration image through a feature extraction network to obtain a first feature map and a second feature map; The first feature map and the second feature map are subjected to global average pooling after channel dimension splicing to obtain a plurality of global description vectors, wherein each channel corresponds to one global description vector; inputting the global description vectors into a multi-layer perceptron, and learning and outputting a plurality of weight coefficients; Normalizing the plurality of weight coefficients by using a Softmax function to obtain a plurality of channel attention weights; and weighting corresponding channels of the first characteristic map and the second characteristic map by using a plurality of channel attention weights, and then carrying out channel combination to generate the pattern multi-characteristic map.
9. The intelligent detection method for defects of a multi-modal data fusion wafer pattern as set forth in claim 1, further comprising: carrying out statistical analysis on the defect identification results of the same batch to generate a defect distribution map; Positioning a deviation source of a patterning process according to the defect distribution map; And generating parameter control suggestions according to the deviation sources.
10. The intelligent detection system for wafer pattern defects by multi-mode data fusion is characterized in that the system is used for implementing the intelligent detection method for wafer pattern defects by multi-mode data fusion according to any one of claims 1-9, and the system comprises: an industrial vision imaging system configuration module for configuring an industrial vision imaging system comprising at least a first imaging optical path and a second imaging optical path; The optical path image acquisition module is used for synchronously acquiring a first optical path image and a second optical path image of an optical substrate wafer pattern area through the industrial visual imaging system, wherein a preset optical function pattern is formed on the surface of the optical substrate wafer; the image registration module is used for carrying out space registration and image enhancement processing on the first optical path image and the second optical path image to obtain a multi-mode registration image; The feature extraction fusion module is used for extracting a first feature map and a second feature map based on the multi-mode registration image, and generating a pattern multi-feature map after fusion; the pattern abnormal candidate region positioning module is used for comparing and analyzing the pattern multi-feature map with a preset gold template to position a pattern abnormal candidate region; The defect recognition module is used for inputting the pattern multi-feature map and the pattern abnormal candidate region into a pre-trained AI visual recognition model, and the AI visual recognition model carries out depth detection on the pattern multi-feature map under the guidance of the pattern abnormal candidate region and outputs a defect recognition result.

Description

Intelligent wafer pattern defect detection method and system based on multi-mode data fusion Technical Field The application relates to the field of industrial vision, in particular to an intelligent detection method and system for defects of a wafer pattern based on multi-mode data fusion. Background Wafer pattern defect detection is a core link for guaranteeing the production yield and reliability of optical devices, and directly affects the performance stability and the service life of downstream optical modules. The current mainstream detection technology takes single-dimension imaging as a core, captures the pattern information of the wafer surface through an optical imaging device, and completes defect identification and judgment by matching with a traditional image comparison or basic algorithm. The technology is limited by single-dimensional information acquisition capability, is difficult to cover the characteristic difference of different cause defects, has weak anti-interference capability, causes insufficient resolving power for low-contrast defects and hidden defects, is easy to generate the phenomena of missing detection and misjudgment, and cannot meet the quality control requirement of high-precision production. In the related art at the present stage, the defect detection of the wafer pattern has the technical problems of low recognition precision and high false alarm rate of missed detection. Disclosure of Invention According to the intelligent detection method and system for the defects of the wafer pattern, an industrial visual imaging system comprising a geometric imaging light path and a material/film/subsurface information imaging light path is built, multi-mode images of the wafer pattern of an optical substrate are synchronously acquired, spatial registration and enhancement processing are carried out on the images, multi-mode registration images are generated, multi-mode features are extracted and fused from the multi-mode registration images, a pattern multi-feature image is obtained, the pattern multi-feature image is compared with a defect-free gold template, an abnormal candidate region of the pattern is positioned, the multi-feature image and the candidate region are input into a pre-training AI visual identification model, defect depth detection is completed under the guidance of the candidate region, and identification results are output. The application provides an intelligent detection method for defects of a wafer pattern through multi-mode data fusion, which comprises the steps of configuring an industrial visual imaging system, synchronously collecting a first optical path image and a second optical path image of an optical substrate wafer pattern area through the industrial visual imaging system, carrying out space registration and image enhancement processing on the first optical path image and the second optical path image to obtain a multi-mode registration image, extracting a first characteristic image and a second characteristic image based on the multi-mode registration image, generating a pattern multi-characteristic image after fusion, comparing the pattern multi-characteristic image with a preset gold template to locate a pattern abnormal candidate area, inputting the pattern multi-characteristic image and the pattern abnormal candidate area into a pre-trained AI visual recognition model, and carrying out depth detection on the pattern multi-characteristic image by the AI visual recognition model under the guidance of the pattern abnormal candidate area to output a defect recognition result. In a possible implementation manner, comparing and analyzing the pattern multi-feature map with a preset golden template to locate a pattern abnormal candidate region, performing the following processing of calculating differences between the pattern multi-feature map and the golden template pixel by pixel to generate a pixel difference map, performing multi-scale Gaussian pyramid decomposition on the pixel difference map to obtain a plurality of difference subgraphs, performing self-adaptive threshold segmentation on the plurality of difference subgraphs respectively, mapping the segmented abnormal region back to the original map scale for merging and morphological optimization, and removing isolated regions with areas smaller than a preset value to obtain the pattern abnormal candidate region. In a possible implementation mode, the golden template is a golden template feature map and is generated by inputting standard reference data into a feature extraction and fusion process which is the same as that of extracting the pattern multi-feature map, wherein the standard reference data is a multi-mode registration simulation image generated by acquiring image data from a similar optical substrate wafer sample confirmed to be defect-free by using the industrial vision imaging system or calculating and rendering the multi-mode registration simulation image through an optical i