Semiconductor chip defect detection method and system based on high-precision UV TDI

Abstract

The invention provides a semiconductor chip defect detection method and system based on high-precision UV TDI, and the method comprises the steps: collecting semiconductor chip image data under ultraviolet light through a UV TDI line-scan digital camera, carrying out the preprocessing, and extracting defect features through an image processing algorithm. And comparing the features with a preset defect database, identifying defect types and severity, and classifying the defect types and severity. Based on the information, a defect detection report is generated. The UV TDI technology of the invention combines high sensitivity of ultraviolet light and high resolution and high speed of a line-scan digital camera. An adopted image processing algorithm comprises edge contour recognition, feature extraction and the like. In the feature extraction stage, comprehensive feature vectors are constructed, and the efficiency is improved through dimension reduction processing. The established defect database can identify and classify defects through comparison. When an unknown defect is encountered, a new feature vector can be added into a database so as to improve the adaptability and accuracy of detection.

Description

Technical Field

[0001] The present invention relates to the technical field of chip defect detection, and particularly to a semiconductor chip defect detection method and system based on high-precision UV TDI. Background Art

[0002] In the semiconductor manufacturing industry, the quality control of chips is a key factor in ensuring product performance and reliability. With the rapid development of semiconductor technology, the structure of chips has become increasingly complex and the size has been continuously reduced, which poses higher requirements for defect detection in the chip manufacturing process. Traditional defect detection methods, such as manual visual inspection and automatic detection systems based on visible light, often have problems such as low detection accuracy, slow speed, and being easily affected by lighting conditions, and are difficult to meet the high-precision and high-efficiency requirements of modern semiconductor manufacturing.

[0003] In particular, under ultraviolet light (UV) irradiation, some defects on the surface of semiconductor chips will exhibit more obvious characteristics, which may be difficult to detect under visible light. Therefore, using ultraviolet light for defect detection has potential advantages. However, how to efficiently use ultraviolet light image data for defect detection remains a technical problem.

[0004] Traditional image processing algorithms often face problems such as large image noise, blurred edges, and difficult feature extraction when processing semiconductor chip image data. These problems limit the accuracy and reliability of defect detection. Summary of the Invention

[0005] The present invention aims to at least solve the technical problem of low accuracy in defect detection existing in the prior art, and particularly innovatively proposes a semiconductor chip defect detection method and system based on high-precision UV TDI.

[0006] To achieve the above object of the present invention, the present invention provides a semiconductor chip defect detection method based on high-precision UV TDI, and the method includes: S1. Obtain semiconductor chip image data, which is collected by a UV TDI linear array camera under ultraviolet light irradiation; S2. Preprocess the collected semiconductor chip image data; S3. Use an image processing algorithm to extract defect features from the preprocessed semiconductor chip image data; S4. Based on the comparison and analysis of the defect features with a preset defect database, identify the defect type, judge the severity, and classify the defect type; S5. Generate a defect detection report based on the defect type and severity.

[0007] As an alternative embodiment of the present invention, optionally, in step S3, extracting defect features using an image processing algorithm includes: S301. Perform image enhancement processing on the semiconductor chip image data and remove noise; S302. Identify the edge contours in the semiconductor chip image data; S303. Perform morphological processing on the objects in the semiconductor chip image data based on the edge contours; S304. Extract features from the semiconductor chip image data after morphological processing to obtain feature vectors; S305. Reduce the dimension of the defect features, screen the defect features after dimension reduction, and obtain defect features.

[0008] As an alternative embodiment of the present invention, optionally, in step S302, identifying the edge contours includes: S3021. Perform Gaussian filtering on the semiconductor chip image data, and the expression is: , where represents the semiconductor chip image data after filtering, represents the semiconductor chip image data, represents the convolution operation, represents the Gaussian function, represents the standard deviation of the Gaussian function; S3022. Perform gradient calculation based on the semiconductor chip image data after Gaussian filtering, and the expression is: , , where represents the gradient in the direction, represents the Sobel convolution kernel in the direction, represents the gradient in the direction, represents the Sobel convolution kernel in the direction; S3023. Calculate the gradient magnitude and gradient direction based on the gradient calculation, and the expression is: , , where S3024. Perform non-maximum suppression based on the gradient magnitude and gradient direction. For each pixel point , check the adjacent pixel values in the gradient direction of the pixel point . If the gradient magnitude of the current pixel is not the pixel point If it is a local maximum in the direction, then set this pixel point to zero; S3025. Set two thresholds and threshold > . Based on the threshold and threshold , traverse the gradient magnitude image. If the pixel value is greater than the threshold , then the pixel value is a strong edge point; if the pixel value is between the threshold and threshold , then the pixel value is a weak edge point; if the pixel value is less than the threshold , then the pixel point is a non-edge point; S3026. Obtain a complete edge contour by connecting the weak edge points to the strong edge points.

[0009] As an optional embodiment of the present invention, optionally, in step S303, the morphological processing of the object in the semiconductor chip image data based on the edge contour includes: S30301. Perform dilation processing on the edge contour; S30302. Perform erosion processing on the dilated edge contour; S30303. Perform opening operation and closing operation to optimize the edge contour again; S30304. Based on the optimized edge contour, segment the object in the semiconductor chip image data.

[0010] As an optional embodiment of the present invention, optionally, the expression for obtaining the feature vector in step S304 is: ; ; ; ; ; ; ; ; wherein, represents the feature vector, represents the weight of the texture feature, represents the texture feature, represents the weight of the shape feature, represents the shape feature, represents the weight of the statistical feature, Represents statistical features, Represents the weight of the image energy, Represents the image energy, Represents the weight of the image contrast, Represents the image contrast, Represents the weight of the image correlation, Represents the image correlation, Represents the weight of the image homogeneity, Represents the image homogeneity, Represents the weight of the area, Represents the total area of the object, Represents the area of the smallest rectangle enclosing the object, Represents the weight of the perimeter, Represents the total length of the object boundary, Represents the weight of the convex hull ratio of the object, Represents the ratio of the convex hull area of the object to the object area, Represents the weight of the average value of the image gray level, Represents the average value of the image gray level, Represents the weight of the dispersion degree of the image gray level distribution, Represents the dispersion degree of the image gray level distribution, Represents the weight of the asymmetry of the image gray level distribution, Represents the asymmetry of the image gray level distribution, Represents the weight of the sharpness degree of the image gray level distribution, Represents the sharpness degree of the image gray level distribution.

[0011] As an alternative embodiment of the present invention, optionally, in step S4, a comparative analysis is performed based on the defect features and a preset defect database to identify the defect type, judge the severity, and classify the defect type, including: S401. Establish a defect database, where the defect database includes known defect types, feature vectors corresponding to the known defect types, severity descriptions of the defects, and classification labels; S402. Compare the selected defect features with the feature vectors corresponding to the defect types in the defect database one by one using a distance metric method to obtain a comparison result; S403. Based on the comparison result, determine the defect type that best matches the defect features as the recognition result; S404. Query the severity description of the defect corresponding to the identified defect type in the defect database to obtain the severity level; S405. Assign a classification label to the semiconductor chip image data to be detected from the defect database based on the identified defect type and severity level.

[0012] As an optional embodiment of the present invention, optionally, establishing the defect database in step S401 includes: S4010. When the identified defect type is a defect type unknown in the defect database, add the feature vector of the unknown defect type to the defect database and update the defect database.

[0013] On the other hand, the present invention also provides a semiconductor chip defect detection system based on high-precision UV TDI. The system includes the semiconductor chip defect detection method based on high-precision UV TDI; The system further includes: An image acquisition module for acquiring semiconductor chip image data; An image preprocessing module connected to the image acquisition module for preprocessing the semiconductor chip image data; A defect feature extraction module connected to the image preprocessing module for extracting defect features from the preprocessed image data using an image processing algorithm; A defect identification and classification module connected to the defect feature extraction module for performing comparative analysis based on the extracted defect features and a preset defect database, identifying the defect type, judging the severity, and classifying the defect type; A report generation module connected to the defect identification and classification module for generating a defect detection report based on the defect type and severity.

[0014] Advantages of the present invention: The present invention uses a high-precision UV TDI linear array camera to collect image data of semiconductor chips under ultraviolet light irradiation. The UV TDI technology combines the high sensitivity of ultraviolet light and the high resolution and high speed characteristics of the linear array camera, and can capture the subtle features on the chip surface under ultraviolet light, providing high-quality raw data for subsequent image processing. After collecting the image data, this solution performs a series of preprocessing operations, including denoising, enhancing contrast, etc., to improve the image quality and reduce the influence of noise on subsequent processing. This solution adopts advanced image processing algorithms, including edge contour recognition, morphological processing, feature extraction, etc., to accurately extract the defect features in the semiconductor chip image. In particular, in edge contour recognition, through steps such as Gaussian filtering, gradient calculation, non-maximum suppression, and threshold processing, the edge contour in the chip image can be accurately recognized, providing a reliable basis for subsequent feature extraction. In the feature extraction stage, this solution comprehensively considers various features such as texture features, shape features, statistical features, image energy, image contrast, image correlation, image homogeneity, area, perimeter, convex hull ratio, average value of image gray level, dispersion degree, asymmetry, and sharpness, and constructs a comprehensive feature vector. At the same time, through dimensionality reduction processing, redundant information is removed, and the effectiveness and calculation efficiency of the feature vector are improved. This solution establishes a defect database containing known defect types, feature vectors, severity descriptions, and classification labels. By comparing the extracted defect features with the feature vectors in the database using the distance metric method, the defect type can be accurately identified and its severity can be judged. At the same time, according to the recognition result, the corresponding classification label is assigned to the image data of the semiconductor chip to be detected. When encountering an unknown defect type, this solution allows the feature vector of the new defect type to be added to the defect database and the database to be updated. This ensures that the database can be continuously improved with the accumulation of detection practices, and improves the adaptability and accuracy of defect detection.

[0015] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and / or additional aspects and advantages of the present invention will become apparent and easy to understand from the description of the embodiments in conjunction with the following drawings, wherein: Figure 1 is a flowchart of the semiconductor chip defect detection method based on high-precision UV TDI in Embodiment 1 of the present invention; Figure 2 is a structural diagram of the semiconductor chip defect detection system based on high-precision UV TDI in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0018] Embodiment 1 As Figure 1 shown, a semiconductor chip defect detection method based on high-precision UV TDI, the method includes: S1. Obtain semiconductor chip image data, which is collected by a UV TDI linear array camera under ultraviolet light irradiation; It should be noted that in step S1, the semiconductor chip image data is collected by a UV TDI linear array camera under ultraviolet light irradiation. Specifically, it can make full use of the high resolution and high speed advantages of the UV TDI technology, as well as the high sensitivity of ultraviolet light to the fine features on the chip surface, so as to ensure that the collected image data has high quality and rich information. In the subsequent image processing steps, by performing preprocessing operations such as denoising and enhancing the contrast of the image data, the image quality can be further improved, laying a solid foundation for subsequent edge contour recognition and feature extraction.

[0019] S2. Preprocess the collected semiconductor chip image data; It should be noted that in step S2, preprocessing the collected semiconductor chip image data aims to eliminate image noise and enhance image contrast, providing high-quality image data for subsequent edge detection and feature extraction. By adopting appropriate preprocessing algorithms, such as denoising algorithms and contrast enhancement algorithms, the signal-to-noise ratio and clarity of the image can be significantly improved, thus ensuring the accuracy and reliability of subsequent processing.

[0020] S3. Adopt an image processing algorithm to extract defect features from the preprocessed semiconductor chip image data; It should be noted that the defect feature extraction in step S3 specifically includes multiple links such as edge contour recognition, morphological processing, and feature vector construction. Edge contour recognition can accurately depict the object edges in the chip image, and morphological processing optimizes the edge contours through steps such as dilation, erosion, opening operation, and closing operation to improve the accuracy of feature extraction. Feature vector construction comprehensively considers various image features, including texture features, shape features, statistical features, etc., and constructs a comprehensive and effective feature vector, providing a reliable basis for subsequent defect recognition and classification. By adopting advanced image processing algorithms and comprehensive feature vector construction, this embodiment can accurately extract the defect features in the semiconductor chip image.

[0021] S4. Based on the comparison and analysis of the defect features with a preset defect database, identify the defect type, judge the severity, and classify the defect type; It should be noted that in step S4, by comparing and analyzing the defect features through a preset defect database, the rapid and accurate identification of the defect type in the semiconductor chip image data is achieved. This database contains known defect types, their corresponding feature vectors, descriptions of the severity of the defects, and classification labels, enabling the system to compare the extracted defect features with the information in the database, thereby determining the defect type, severity, and classification label. This step improves the efficiency and accuracy of defect detection. At the same time, the system has the ability of self-learning and updating. When encountering an unknown defect type, it can add its feature vector to the defect database and update the database, thus continuously enhancing the detection ability and adaptability of the system.

[0022] S5. Generate a defect detection report based on the defect type and severity.

[0023] It should be noted that the defect detection report generated in step S5 details information such as the defect type, severity, and classification label found during the detection process, providing strong data support for the production quality control of semiconductor chips. This report can not only help production personnel promptly discover and repair defects in the chips, improving product quality, but also provide valuable reference for subsequent process improvement and product design.

[0024] In summary, this embodiment uses a high-precision UV TDI linear array camera to collect image data of semiconductor chips under ultraviolet light irradiation. The UV TDI technology combines the high sensitivity of ultraviolet light and the high resolution and high speed characteristics of the linear array camera, enabling it to capture the subtle features of the chip surface under ultraviolet light, providing high-quality raw data for subsequent image processing. After collecting the image data, this solution performs a series of preprocessing operations, including denoising, enhancing contrast, etc., to improve the image quality and reduce the impact of noise on subsequent processing. This solution uses advanced image processing algorithms, including edge contour recognition, morphological processing, feature extraction, etc., to accurately extract the defect features in the semiconductor chip image. In particular, in edge contour recognition, through steps such as Gaussian filtering, gradient calculation, non-maximum suppression, and threshold processing, the edge contours in the chip image can be accurately recognized, providing a reliable basis for subsequent feature extraction. In the feature extraction stage, this solution comprehensively considers various features such as texture features, shape features, statistical features, image energy, image contrast, image correlation, image homogeneity, area, perimeter, convex hull ratio, average value of image grayscale, dispersion, asymmetry, and sharpness, and constructs a comprehensive feature vector. At the same time, through dimensionality reduction processing, redundant information is removed, improving the effectiveness and computational efficiency of the feature vector. This solution establishes a defect database containing known defect types, feature vectors, severity descriptions, and classification labels. By comparing the extracted defect features with the feature vectors in the database using the distance metric method, the defect type can be accurately identified and its severity can be judged. At the same time, according to the recognition result, the corresponding classification label is assigned to the image data of the semiconductor chip to be detected. When encountering an unknown defect type, this solution allows the feature vector of the new defect type to be added to the defect database and the database to be updated. This ensures that the database can be continuously improved with the accumulation of detection practices, improving the adaptability and accuracy of defect detection.

[0025] As an alternative embodiment of the present invention, optionally, the extraction of defect features using an image processing algorithm in step S3 includes: S301. Perform image enhancement processing on the semiconductor chip image data and remove noise; S302. Identify the edge contours in the semiconductor chip image data; S303. Perform morphological processing on the objects in the semiconductor chip image data based on the edge contours; S304. Perform feature extraction on the semiconductor chip image data after morphological processing to obtain a feature vector; S305. Perform dimensionality reduction on the defect features, screen the defect features after dimensionality reduction, and obtain defect features.

[0026] It should be noted that in step S301, image enhancement processing and denoising are to improve the signal-to-noise ratio and contrast of image data, making the defect features in the image more obvious and facilitating subsequent edge contour recognition and feature extraction. In step S302, edge contour recognition is one of the key steps in defect feature extraction. By accurately depicting the object edges in the chip image, it can provide an accurate basis for subsequent morphological processing and feature extraction. In step S303, morphological processing can further improve the accuracy of feature extraction by optimizing the edge contour. In step S304, feature extraction is the key to constructing a comprehensive and effective feature vector. By comprehensively considering various image features, it provides a reliable basis for subsequent defect recognition and classification. In step S305, dimensionality reduction processing and feature screening are to remove redundant information, improve the effectiveness and computational efficiency of the feature vector, and thus ensure the accuracy and real-time performance of defect detection.

[0027] As an alternative embodiment of the present invention, optionally, the edge contour recognition in step S302 includes: S3021. Perform Gaussian filtering on the semiconductor chip image data, and the expression is: , where represents the semiconductor chip image data after filtering, represents the semiconductor chip image data, represents the convolution operation, represents the Gaussian function, represents the standard deviation of the Gaussian function; S3022. Perform gradient calculation based on the semiconductor chip image data after Gaussian filtering, and the expression is: , , where represents the gradient in the direction, represents the Sobel convolution kernel in the direction, represents the gradient in the direction, represents the Sobel convolution kernel in the S3023. Calculate the gradient magnitude and gradient direction based on the gradient calculation, and the expression is: , , where represents the magnitude of the gradient, represents the direction of the gradient; S3024. Perform non-maximum suppression based on the gradient magnitude and gradient direction. Based on each pixel point , check the pixel point The adjacent pixel values in the gradient direction. If the gradient magnitude of the current pixel is not the local maximum in the direction of this pixel point then set this pixel point to zero; S3025. Set two thresholds and threshold , > . Based on the threshold and threshold traverse the gradient magnitude image. If the pixel value is greater than the threshold , then the pixel value is a strong edge point; if the pixel value is between the threshold and the threshold , then the pixel value is a weak edge point; if the pixel value is less than the threshold , then the pixel point is a non-edge point; S3026. Obtain the complete edge contour by connecting the weak edge points to the strong edge points.

[0028] It should be noted that in step S3021, Gaussian filtering, as a linear smoothing filter, can effectively suppress image noise, reduce the interference of high-frequency components, and make the edge features in the image smoother and more obvious. By selecting an appropriate standard deviation of the Gaussian function, the smoothing degree of the filter can be controlled to achieve the best filtering effect. In step S3022, the gradient calculation is based on the Sobel convolution kernel to perform convolution operations on the image, which can calculate the gradients of the image in the x-axis and y-axis directions, thereby reflecting the change of the image gray value. In step S3023, by calculating the gradient magnitude and direction of each pixel point, it can be determined whether the pixel point is an edge point and the direction of the edge. In step S3024, non-maximum suppression is an edge thinning technique. By suppressing non-edge pixel points and retaining the local maximum in the gradient direction, a refined edge contour can be obtained. This step can reduce the redundant information of the edge and improve the continuity and accuracy of the edge. In step S3025, by setting two thresholds, the pixel points in the gradient magnitude image can be divided into strong edge points, weak edge points, and non-edge points. Strong edge points have a higher gradient magnitude and usually correspond to obvious edges in the image; weak edge points have a lower gradient magnitude but still play an important role when connecting strong edge points; non-edge points have the lowest gradient magnitude and usually correspond to flat areas in the image. In step S3026, by connecting weak edge points to strong edge points, a complete edge contour can be obtained. This step can fill in the discontinuous points in the edge contour and make the edge more continuous and smooth. By adopting the above edge contour recognition method, this embodiment can accurately recognize the edge contour in the semiconductor chip image.

[0029] As an alternative embodiment of the present invention, optionally, in step S303, the morphological processing of the object in the semiconductor chip image data based on the edge contour includes: S30301. Performing dilation processing on the edge contour; S30302. Performing erosion processing on the edge contour after dilation processing; S30303. Further optimizing the edge contour by using opening operation and closing operation; S30304. Segmenting the object in the semiconductor chip image data based on the optimized edge contour.

[0030] It should be noted that in step S30301, the dilation processing can expand the range of the edge contour, fill the small holes and concave parts in the edge contour, and make the edge contour more complete and continuous. This step helps to improve the connectivity of the object in the image. In step S30302, the erosion processing can narrow the range of the edge contour, remove the small protruding parts on the edge contour, and make the edge contour smoother and more regular. By alternately performing dilation and erosion, fine adjustment of the edge contour can be achieved, improving its accuracy and reliability. In step S30303, the opening operation can remove the small objects on the edge contour while keeping the main structure of the edge contour unchanged, helping to eliminate the noise and interference factors in the image. The closing operation can fill the discontinuous points and small holes on the edge contour, making the edge contour more complete and continuous. By combining the opening operation and the closing operation, the morphology of the edge contour can be further optimized, improving its quality and stability. In step S30304, based on the optimized edge contour, the object in the semiconductor chip image data can be segmented. This step can distinguish different objects in the image, providing an accurate basis for subsequent feature extraction and defect recognition. By adopting the above morphological processing method, this embodiment can optimize the edge contour in the semiconductor chip image, improving the accuracy and reliability of feature extraction.

[0031] As an alternative embodiment of the present invention, optionally, the expression for obtaining the feature vector in step S304 is: ; ; ; ; ; ; ; ; Wherein, Represents a feature vector, Represents the weight of the texture feature, Represents the texture feature, Represents the weight of the shape feature, Represents the shape feature, Represents the weight of the statistical feature, Represents the statistical feature, Represents the weight of the image energy, Represents the image energy, Represents the weight of the image contrast, Represents the image contrast, Represents the weight of the image correlation, Represents the image correlation, Represents the weight of the image homogeneity, Represents the image homogeneity, Represents the weight of the area, Represents the total area of the object, Represents the area of the smallest rectangle enclosing the object, Represents the weight of the perimeter, Represents the total length of the object boundary, Represents the weight of the object convex hull ratio, Represents the ratio of the object convex hull area to the object area, Represents the weight of the average value of the image gray level, Represents the average value of the image gray level, Represents the weight of the dispersion degree of the image gray level distribution, Represents the dispersion degree of the image gray level distribution, Represents the weight of the asymmetry of the image gray level distribution, Represents the asymmetry of the image gray level distribution, Represents the weight of the sharpness of the image gray level distribution, Represents the sharpness of the image gray level distribution.

[0032] As an alternative embodiment of the present invention, optionally, in step S4, based on the comparison and analysis of the defect features with a preset defect database, the defect type is identified, the severity is judged, and the classification of the defect type includes: S401. Establish a defect database, where the defect database includes known defect types, feature vectors corresponding to the known defect types, severity descriptions of the defects, and classification labels; S402. Compare the selected defect features with the feature vectors corresponding to the defect types in the defect database one by one using a distance metric method to obtain a comparison result; S403. Based on the comparison result, determine the defect type that best matches the defect feature as the recognition result; S404. Based on the recognized defect type, query the defect database for the severity description of the defect corresponding to this defect type to obtain the severity level; S405. Based on the recognized defect type and severity level, assign a classification label to the semiconductor chip image data to be detected from the defect database.

[0033] It should be noted that in step S401, the defect database is established based on a large number of known defect types and corresponding feature vectors, and these feature vectors are obtained through previous defect detection and analysis processes. The severity description and classification label of the defect are based on the evaluation and classification of professionals to ensure the accuracy and reliability of the database. In step S402, the distance metric method can evaluate the similarity between the defect feature to be detected and the feature vectors of known defect types by calculating the distance between them. The smaller the distance, the higher the similarity, thereby determining the best-matching defect type. In step S403, based on the comparison result, select the defect type with the smallest distance as the recognition result, which helps to accurately determine the defect type existing in the semiconductor chip image to be detected. In step S404, by querying the defect database, the severity level and classification label of the recognized defect type can be obtained. Finally, based on step S405, by adopting the above defect recognition method, this embodiment can achieve accurate recognition and classification of defects in semiconductor chip images.

[0034] As an optional embodiment of the present invention, optionally, the expression of the distance metric method in step S402 is: ; where, represents the selected defect feature vector and the feature vector of the th known defect type in the defect database the distance between them, represents the dimension of the feature vector, represents the th weight of the feature component, represents the th value of the feature component in the selected defect feature vector, represents the th known defect type in the defect database th value of the feature component in the feature vector, represents the th standard deviation of the feature component.

[0035] As an alternative embodiment of the present invention, optionally, establishing the defect database in step S401 includes: S4010. When the identified defect type is a defect type unknown in the defect database, adding the feature vector of the unknown defect type to the defect database and updating the defect database.

[0036] It should be noted that in step S4010, when no matching item can be found for the identified defect type in the existing defect database, it is regarded as an unknown defect type. At this time, the feature vector of this unknown defect type can be added to the defect database and the database record can be updated. This step helps to expand the coverage of the defect database and improve its adaptability and practicality. By continuously updating the defect database, it can be ensured that the defect recognition system can accurately identify and classify newly emerging defect types. At the same time, this also reflects the flexibility and scalability of the method of the present invention.

[0037] Embodiment 2 As Figure 2 shown, a semiconductor chip defect detection system based on high-precision UV TDI, the system includes a semiconductor chip defect detection method based on high-precision UV TDI; The system further includes: An image acquisition module for acquiring semiconductor chip image data; in this embodiment, the image acquisition module is an image sensor based on high-precision UV TDI, which has high resolution and high sensitivity and can achieve precise acquisition of semiconductor chip images.

[0038] An image preprocessing module connected to the image acquisition module for preprocessing the semiconductor chip image data; in this embodiment, the preprocessing includes operations such as denoising, enhancing contrast, and correcting image distortion to improve the image quality and the accuracy of subsequent processing. Through the processing of the image preprocessing module, the noise and interference factors in the image can be eliminated, the detailed features of the image can be enhanced, and high-quality image data can be provided for subsequent edge contour recognition and morphological processing.

[0039] A defect feature extraction module connected to the image preprocessing module for extracting defect features from the preprocessed image data by using image processing algorithms; specifically, the defect feature extraction module can identify and analyze the key features in the image, such as texture, shape, statistical characteristics, and image energy, contrast, correlation, homogeneity, etc. These features are crucial for defect recognition. Through image processing technology, the defect feature extraction module can accurately extract the feature vectors related to defects and provide strong support for subsequent defect recognition.

[0040] A defect recognition and classification module, connected to the defect feature extraction module, for performing comparative analysis based on the extracted defect features and a preset defect database, identifying the defect type, judging the severity, and classifying the defect type; A report generation module, connected to the defect recognition and classification module, for generating a defect detection report based on the defect type and severity.

[0041] The defect feature extraction module includes: An image enhancement and denoising sub-module for enhancing the semiconductor chip image data and removing noise; An edge contour recognition sub-module for recognizing the edge contours in the image; specifically, the edge contour recognition sub-module uses image processing algorithms such as Canny edge detection, etc., which can accurately recognize the edge contours in the image. These edge contours are the basis for subsequent morphological processing and defect feature extraction. Through the processing of the edge contour recognition sub-module, it can ensure that the key features in the image are accurately captured. The edge contour recognition sub-module can also adjust parameters according to actual needs to adapt to the characteristics and requirements of different semiconductor chip images, improving the flexibility and adaptability of the system.

[0042] A morphological processing sub-module for performing morphological processing such as dilation, erosion, opening operation, and closing operation on the edge contours; A feature vector extraction sub-module for extracting feature vectors using the edge contours after morphological processing; The defect recognition and classification module includes: A defect database establishment sub-module for establishing a defect database, which includes known defect types, feature vectors corresponding to the known defect types, severity descriptions of the defects, and classification labels; A feature comparison sub-module for using the distance metric method to compare the selected defect features with the defect types in the defect database one by one to obtain comparison results; specifically, the feature comparison sub-module can efficiently process a large amount of defect feature data, and by comparing with the feature vectors in the defect database, quickly find the most matching defect type.

[0043] A defect recognition and classification sub-module for determining the most matching defect type based on the comparison results and querying the defect database to obtain the severity level and classification label; specifically, the defect recognition and classification sub-module can accurately judge the defect types existing in the semiconductor chip image to be detected and give the corresponding severity level and classification label. The realization of this function depends on the accuracy and integrity of the defect database, as well as the efficient processing ability of the feature comparison sub-module.

[0044] A database update sub-module, which is used to add the feature vector of the unknown defect type to the defect database and update the defect database when the identified defect type is an unknown defect type in the defect database.

[0045] The principle of the semiconductor chip defect detection system is as follows: The system first obtains the image data of the semiconductor chip through the image acquisition module. These data are then transferred to the image preprocessing module for necessary preprocessing operations such as grayscale conversion, filtering, etc., to improve the image quality and reduce noise interference. The preprocessed image data is sent to the defect feature extraction module, which uses image processing algorithms such as image enhancement, denoising, edge contour recognition, and morphological processing to extract the defect features in the semiconductor chip image. These feature vectors accurately reflect the defect conditions on the chip surface. The extracted feature vectors are then transferred to the defect identification and classification module. The defect database built in this module stores a large number of known defect types and their corresponding feature vectors, severity descriptions, and classification labels. By adopting efficient comparison algorithms such as the distance metric method, the module can compare the extracted defect features with the known defect types in the database one by one to quickly determine the most matching defect type. At the same time, it can also query the database to obtain the severity level and classification label of this defect type. If the identified defect type has no matching item in the database, it is regarded as an unknown defect type. At this time, the database update sub-module will come into play. It will automatically add the feature vector of this unknown defect type to the defect database and update the database record, so as to ensure the continuous accuracy and adaptability of the defect identification system. Finally, the report generation module automatically generates a detailed defect detection report according to the defect type and severity. This report not only provides an important basis for the production quality control of semiconductor chips, but also provides valuable reference for subsequent product improvement and optimization. The working process of the entire detection system realizes a high degree of automation and intelligence, greatly improving the efficiency and accuracy of semiconductor chip defect detection.

[0046] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and purposes of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. A semiconductor chip defect detection method based on high-precision UV TDI, characterized in that The method includes: S1. Obtain semiconductor chip image data, which is collected by a UV TDI linear array camera under ultraviolet light irradiation; S2. Use a contrast enhancement algorithm to enhance the contrast of the semiconductor chip image data; S3. Adopt an image processing algorithm to extract defect features from the semiconductor chip image data. When extracting defect features, it also includes identifying edge contours and morphological processing; S4. Based on the comparison and analysis of the defect features with a preset defect database, identify the defect type, judge the severity level, and classify the defect type, including: S401. Establish a defect database and update the database. The defect database includes known defect types, feature vectors corresponding to the known defect types, descriptions of the severity levels of the defects, and classification labels; S402. Compare the selected defect features with the feature vectors corresponding to the defect types in the defect database one by one using a distance metric method to obtain comparison results; S403. Based on the comparison results, determine the defect type that best matches the defect features as the recognition result; S404. Query the description of the severity level of the defect corresponding to the defect type in the defect database based on the identified defect type to obtain the severity level; S405. Based on the identified defect type and severity level, assign a classification label to the semiconductor chip image data to be detected from the defect database; S5. Generate a defect detection report based on the defect type and severity level.

2. The semiconductor chip defect detection method based on high-precision UV TDI according to claim 1, characterized in that, In step S3, using an image processing algorithm to extract defect features includes: S301. Perform image enhancement processing on the semiconductor chip image data and remove noise; S302. Identify the edge contours in the semiconductor chip image data; S303. Based on the edge contours, perform morphological processing on the objects in the semiconductor chip image data; S304. Perform feature extraction on the semiconductor chip image data after morphological processing to obtain feature vectors; S305. Reduce the dimension of the defect features, screen the defect features after dimension reduction, and obtain defect features.

3. The semiconductor chip defect detection method based on high-precision UV TDI according to claim 2, wherein, In step S302, identifying edge contours includes: S3021. Perform Gaussian filtering on the semiconductor chip image data, with the expression: , where represents the semiconductor chip image data after filtering, represents the semiconductor chip image data, represents the convolution operation, represents the Gaussian function, represents the standard deviation of the Gaussian function; S3022. Perform gradient calculation based on the semiconductor chip image data after Gaussian filtering. The expression is: , , where represents the gradient in the axis direction, represents the Sobel convolution kernel in the axis direction, represents the Sobel convolution kernel in the axis direction; S3023. Calculate the gradient magnitude and gradient direction based on the said gradient, and the expressions are: , , where represents the magnitude of the gradient, represents the direction of the gradient; S3024. Perform non-maximum suppression based on the gradient magnitude and gradient direction. Based on each pixel point , check the adjacent pixel values in the gradient direction of the pixel point . If the gradient magnitude of the current pixel is not the local maximum in the direction of this pixel point , then set this pixel point to zero. S3025. Set two thresholds and threshold , > . Based on the thresholds and threshold , traverse the gradient magnitude image. If the pixel value is greater than the threshold , the pixel value is a strong edge point; if the pixel value is between the threshold and threshold , the pixel value is a weak edge point; if the pixel value is less than the threshold , the pixel point is a non-edge point; S3026. Obtain a complete edge contour by connecting the weak edge points to the strong edge points.

4. The semiconductor chip defect detection method based on high-precision UV TDI according to claim 2, wherein, In step S303, based on the edge contours, performing morphological processing on the objects in the semiconductor chip image data includes: S30301. Perform dilation processing on the edge contours; S30302. Perform erosion processing on the edge contours after dilation processing; S30303. Use opening operation and closing operation to optimize the edge contours again; S30304. Based on the optimized edge contours, segment the objects in the semiconductor chip image data.

5. The semiconductor chip defect detection method based on high-precision UV TDI according to claim 2, wherein The expression for obtaining feature vectors in step S304 is: ； ； ； ； ； ； ； ； Among them, represents the feature vector, represents the weight of the texture feature, represents the texture feature, represents the weight of the shape feature, represents the shape feature, represents the weight of the statistical feature, represents the statistical feature, represents the weight of the image energy, represents the image energy, represents the weight of the image contrast, represents the image contrast, represents the weight of the image correlation, represents the image correlation, represents the weight of the image homogeneity, represents the image homogeneity, represents the weight of the area, represents the total area of the object, represents the area of the smallest rectangle enclosing the object, represents the weight of the perimeter, represents the total length of the object boundary, represents the weight of the object convex hull ratio, represents the ratio of the object convex hull area to the object area, represents the weight of the average value of the image gray level, represents the average value of the image gray level, represents the weight of the dispersion degree of the image gray level distribution, represents the dispersion degree of the image gray level distribution, represents the weight of the asymmetry of the image gray level distribution, represents the asymmetry of the image gray level distribution, represents the weight of the sharpness of the image gray level distribution, represents the sharpness of the image gray level distribution.

6. The semiconductor chip defect detection method based on high-precision UV TDI according to claim 1, characterized in that, The expression for the distance metric method in step S402 is: ； Among them, represents the screened defect feature vector and the feature vector of the nth known defect type in the defect database, represents the dimension of the feature vector, represents the weight of the nth feature component, represents the value of the nth feature component in the screened defect feature vector, represents the value of the nth feature component in the feature vector of the nth known defect type in the defect database, represents the standard deviation of the nth feature component.

7. The semiconductor chip defect detection method based on high-precision UV TDI according to claim 1, characterized in that Establishing a defect database in step S401 includes: S4011. When the identified defect type is a defect type unknown in the defect database, add the feature vector of the unknown defect type to the defect database and update the defect database.

8. A semiconductor chip defect detection system based on high-precision UV TDI, characterized in that, The system includes the semiconductor chip defect detection method based on high-precision UV TDI according to any one of claims 1 to 7. The system further includes: An image acquisition module for acquiring semiconductor chip image data. An image preprocessing module connected to the image acquisition module for preprocessing the semiconductor chip image data. A defect feature extraction module connected to the image preprocessing module for extracting defect features from the preprocessed image data using an image processing algorithm. A defect identification and classification module connected to the defect feature extraction module for performing comparative analysis based on the extracted defect features and a preset defect database, identifying the defect type, judging the severity, and classifying the defect type. A report generation module connected to the defect identification and classification module for generating a defect detection report based on the defect type and severity.

9. The semiconductor chip defect detection system based on high-precision UV TDI according to claim 8, characterized in that, The defect feature extraction module includes: An image enhancement and denoising sub-module for enhancing and denoising the semiconductor chip image data. An edge contour recognition sub-module for recognizing the edge contour in the image. A morphological processing sub-module for performing morphological processing such as dilation, erosion, opening operation, and closing operation on the edge contour. A feature vector extraction sub-module for extracting a feature vector using the edge contour after morphological processing. The defect identification and classification module includes: A defect database establishment sub-module for establishing a defect database, which includes known defect types, feature vectors corresponding to the known defect types, severity descriptions of defects, and classification labels. A feature comparison sub-module for using a distance metric method to compare the selected defect features with the defect types in the defect database one by one to obtain a comparison result. A defect identification and classification sub-module for determining the most matching defect type based on the comparison result and querying the defect database to obtain the severity level and classification label. A database update sub-module for adding the feature vector of the unknown defect type to the defect database and updating the defect database when the identified defect type is a defect type unknown in the defect database.

Images