Visual detection method and system for surface defects of semiconductor wafer

Abstract

The invention discloses a semiconductor wafer surface defect visual detection method and system, and the method comprises the following steps: obtaining a wafer image, calculating a deflection angle through Hough linear transformation, completing the rotation correction of a wafer, searching all contours in the wafer after the rotation correction, reading the area of the contours, and selecting different filters to remove noise according to the noise type, the method has the advantages that the gray level distribution condition in the neighborhood is analyzed according to the standard deviation of the pixel gray values in the neighborhood, the image resolution is improved, the interference is reduced, the crystal grain contour information is protected, the image is enhanced, the interested area is intensified on purpose, the background area is inhibited, the image contrast is improved, and the image which is more beneficial to computer analysis is obtained. The method has the advantages that the gray level distribution condition in the neighborhood is analyzed through the standard deviation of the pixel gray values in the neighborhood; according to the requirements of inhibiting the crystal grain edge and enhancing the defect edge, image enhancement is performed on pixels at different positions, and according to the EnFcm algorithm, the gray level of clustering data is changed into the gray level from the gray level, so that the segmentation efficiency is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of visual inspection of wafer surface defects, and specifically relates to a method and system for visual inspection of semiconductor wafer surface defects. Background Art

[0002] With the booming development of the semiconductor industry, the quality requirements for semiconductor wafer materials are getting higher and higher. Wafers are commonly used in the manufacture of semiconductor intelligent chips, and their excellent quality and performance play a crucial role in the field of integrated circuits. In order to meet the social needs for the production quantity and quality of wafers, wafer manufacturing enterprises continuously improve their processing technologies. The raw material of wafers is single-crystalline silicon, and it takes dozens of cumbersome processes to manufacture wafers from single-crystalline silicon. With the improvement of process complexity, the sources of grain surface defects show a diversified trend. Wafer surface defects seriously affect the product yield rate and even become an important factor hindering the development of enterprises.

[0003] At present, domestic small and medium-sized wafer manufacturing enterprises still use the traditional manual defect detection method for screening, that is, using the human eye to observe whether there are defects on the surface of each grain with the assistance of an electron microscope. Traditional detection mainly relies on people's subjective consciousness and industry experience as the judgment basis, which varies from person to person and cannot follow a unified judgment standard. There are many drawbacks in the manual detection method and it cannot meet the needs of large-scale production. The deficiencies of this method are mainly reflected in the following aspects: 1) Long-term high-intensity work will cause visual fatigue, resulting in an increase in the misdetection rate; 2) Restricted by the physiological structure of people, long-term work will cause physical and mental fatigue and reduce production efficiency; 3) The judgment of defects mainly relies on people's subjective consciousness and cannot be summarized into a standard, resulting in uneven product quality; 4) To meet the large-scale production of enterprises, a large number of inspection personnel need to be recruited. Currently, the labor cost is constantly increasing. Therefore, a method and system for visual inspection of semiconductor wafer surface defects are provided to solve the above problems. Summary of the Invention

[0004] To solve the problems in the above background art, the technical solution adopted by the present invention to solve the technical problems is: A method for visual inspection of semiconductor wafer surface defects, which includes the following steps:

[0005] S1: Obtain a wafer image, calculate the deflection angle through the Hough line transform and complete the wafer rotation correction. After the rotation correction, find all the contours in the wafer and read the contour areas;

[0006] S2: During the process of grain contour extraction, the image needs to be preprocessed;

[0007] S3: Then perform clustering segmentation on the extracted image to distinguish the defects and the background with similar colors and sizes in the wafer surface image;

[0008] S4: Next, perform contour feature extraction on the image that has been clustered and segmented in S3. After clustering and segmentation, the foreground image and background of the grains have been segmented into two categories, clearly showing the contour features of the grains and defect points. By analyzing the defect type features, it is concluded that when there are defects on the grain surface, the outer contour of the grain deforms or there are nested contours.

[0009] S5: Through the contour matching image contour matching algorithm of Hu moments, calculate the similarity of the Hu invariant moments of the contour. The larger the similarity value, the higher the similarity, and the higher the detection requirement. Furthermore, the surface defects of the wafer can be classified.

[0010] As a preferred technical solution of the present invention, in S1, the wafer rotation correction is to perform binarization processing on the wafer image, select multiple local regions to extract straight lines of the wafer grooves using the Hough line transform, calculate the average tilt angle based on the extracted straight lines, and determine whether the tilt angle is greater than the threshold. If it is greater than the threshold, send the rotation angle to the transplant manipulator for rotation, and capture the image of the wafer after rotation. If it is less than the threshold, it means that this angle will not affect the grain extraction and the grain extraction can be directly performed without rotation; if the read contour area is less than the set lower limit or greater than the set upper limit, it is determined that this contour is not a grain contour, and continue to read the next contour information. Otherwise, it is determined that this contour is a grain contour, record the coordinate information of this contour, and extract the grain according to the grain contour coordinate information.

[0011] As a preferred technical solution of the present invention, when preprocessing the image in S2, different filters are selected according to the noise type to remove noise, reduce interference, protect the grain contour information, and enhance the image. The region of interest is strengthened purposefully while the background region is suppressed, improving the image contrast to obtain an image that is more conducive to computer analysis; the image preprocessing is performed using morphological closing reconstruction. Morphological closing reconstruction is to use a defined structural element to extract the shape feature information in the image, and then perform feature analysis and target recognition on the image.

[0012] As a preferred technical solution of the present invention, enhancing the image can improve the contrast of the grayscale image, highlight the feature information of the grain edges and defects, and reduce the complexity of the subsequent image processing and analysis process.

[0013] As a preferred technical solution of the present invention, in S3, the clustering segmentation uses the idea of EnFcm to improve the segmentation efficiency of the K-means algorithm. Utilizing the characteristic that the number of gray levels in the grayscale image is much smaller than the number of pixels, the grayscale histogram is used as the clustering sample for segmentation, and the distance measure between the same gray value and the clustering center is equal.

[0014] As a preferred technical solution of the present invention, in S4, defect detection uses the characteristics of the contour tree and contour moments to detect whether there are defects on the grain surface.

[0015] As a preferred technical solution of the present invention, in the S4, the contour extraction algorithm takes a point in the connected region as the starting point, gradually scans the pixel points, tracks its contour, and marks the pixels on the contour.

[0016] As a preferred technical solution of the present invention, in the S5, the moment is often used in image processing to represent the object after image segmentation.

[0017] The semiconductor wafer surface defect vision detection system includes a grain extraction module, an image preprocessing module, an image segmentation module, and a defect recognition module. The grain extraction module is used for wafer rotation and contour extraction. The image preprocessing module is used for image filtering and image enhancement. The image segmentation module is used for gray-level statistics and clustering segmentation. The defect recognition module is used for contour analysis and contour matching.

[0018] As a preferred technical solution of the present invention, the image segmentation module refers to dividing the image into multiple independent regions with specific properties, and the algorithm extracts all the contours in the image, including inner contours and outer contours.

[0019] The present invention has the following advantages: After automatically acquiring the wafer image through the semiconductor wafer surface defect vision detection system of the present invention, the deflection angle is calculated by using the Hough line transformation, and the wafer rotation correction and grain extraction are completed. Based on the research on defect types such as redundant substances, scratches, and crystal corner defects on the grain surface, the noise reduction effects of different filtering algorithms are compared. The gray-level distribution in the neighborhood is analyzed through the standard deviation of the pixel gray values in the neighborhood. According to the requirements of suppressing the grain edge and enhancing the defect edge, image enhancement is performed on the pixels at different positions. By analyzing the characteristics of defective grains, a defect detection method based on contour features is proposed to calculate and judge the wafer defects, so as to achieve high accuracy, high detection efficiency, and effectively avoid the situation of misdetection in traditional manual detection. Description of the Drawings

[0020] Figure 1 is the flowchart of the detection method of the preferred embodiment of the present invention;

[0021] Figure 2 is the schematic diagram of the operating principle of the semiconductor wafer surface defect vision detection of the preferred embodiment of the present invention;

[0022] Figure 3 is the schematic diagram of the wafer rotation process of the preferred embodiment of the present invention;

[0023] Figure 4 is the schematic diagram of the wafer image before rotation of the preferred embodiment of the present invention;

[0024] Figure 5It is a schematic diagram of the rotated wafer image of the preferred embodiment of the present invention. Detailed implementation manners

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0026] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific implementation manners.

[0027] Please refer to Figures 1-5 , the visual inspection method for semiconductor wafer surface defects of the present invention includes the following steps:

[0028] S1: Obtain the wafer image, calculate the deflection angle through the Hough line transform and complete the wafer rotation correction. After the rotation correction, find all the contours in the wafer, read the contour area. If the area is less than the set lower limit or greater than the set upper limit, it is determined that the contour is not a grain contour, and continue to read the next contour information. Otherwise, it is determined that the contour is a grain contour, record the contour coordinate information, and extract the grain according to the grain contour coordinate information;

[0029] S2: During the grain contour extraction process, the image needs to be preprocessed. Different filters are selected according to the noise type to remove noise, reduce interference, protect the grain contour information, and enhance the image. The region of interest is strengthened purposefully while the background region is suppressed to improve the image contrast and obtain an image more conducive to computer analysis;

[0030] S3: Then perform clustering segmentation on the extracted image to distinguish the defects and the background with similar colors and sizes on the wafer surface image; at the same time, it has the characteristics of little influence by light and strong adaptability;

[0031] S4: Then perform contour feature extraction on the image after clustering segmentation in S3. After clustering segmentation, the grain foreground image and the background have been segmented into two categories, clearly showing the contour features of the grains and defect points. By analyzing the defect type features, it can be concluded that when there are defects on the grain surface, the outer contour of the grain deforms or there are nested contours;

[0032] S5: Through the contour matching image contour matching algorithm of Hu moments, calculate the similarity of the Hu invariant moments of the contours. The larger the similarity value, the higher the similarity, and the higher the detection requirements. Furthermore, the surface defects of the wafer can be classified. Moments are often used in image processing to represent objects after image segmentation. Moments can be specific weighted image pixel concentrations or a series of functions similar to moments.

[0033] In S1, for wafer rotation correction, the wafer image is binarized, and multiple local regions are selected to extract straight lines of the wafer grooves using the Hough line transform. Calculate the average tilt angle based on the extracted straight lines, and determine whether the tilt angle is greater than the threshold. If it is greater than the threshold, send the rotation angle to the transfer manipulator for rotation, and capture the image of the rotated wafer. If it is less than the threshold, it means that this angle will not affect the grain extraction, and the grain extraction can be directly performed without rotation.

[0034] In S2, image preprocessing adopts morphological closing reconstruction. Morphological closing reconstruction uses a defined structural element to extract the shape feature information in the image, thereby realizing feature analysis and target recognition of the image. Image enhancement can improve the contrast of grayscale images, highlight the feature information of grain edges and defects, and reduce the complexity of subsequent image processing and analysis. Image enhancement can be divided into linear grayscale enhancement and non-linear grayscale enhancement. The principle of linear grayscale enhancement is to perform grayscale value operations on the value of each pixel point using a linear function or several piecewise linear functions. Non-linear grayscale enhancement uses non-linear functions for grayscale value operations. Non-linear functions include exponential functions, logarithmic functions, power-exponential functions, etc.

[0035] In S3, clustering segmentation uses the idea of EnFcm to improve the segmentation efficiency of the K-means algorithm. Utilize the characteristic that the number of gray levels in a grayscale image is much smaller than the number of pixels, and use the grayscale histogram as the clustering sample for segmentation. The distance measure between the same gray value and the clustering center is equal. In S4, defect detection uses the characteristics of the contour tree and contour moments to detect whether there are defects on the surface of the grains. Extract the contour of the grain image after segmentation. If there is a nested contour in the contour tree, that is, a sub-level contour exists, the grain is marked as defective. Otherwise, based on the template grain contour, use the contour similarity to detect the contour. If the matching similarity is less than the threshold, it is marked as defective, otherwise it is marked as a good product. The contour extraction algorithm starts from a point in the connected region, gradually scans the pixel points, and tracks its contour, marking the pixels on the contour. When the contour is completely closed, a scan is completed, and the pixel point scanning continues until a new connected domain is found.

[0036] Vision inspection method and system for semiconductor wafer surface defects, including a grain extraction module, an image preprocessing module, an image segmentation module, and a defect recognition module. The grain extraction module is used for wafer rotation and contour extraction. The image preprocessing module is used for image filtering and image enhancement. The image segmentation module is used for gray-level statistics and clustering segmentation. The defect recognition module is used for contour analysis and contour matching. The image segmentation module refers to dividing an image into multiple independent regions with specific properties. The algorithm extracts all the contours in the image, including inner contours and outer contours.

[0037] Among them, the wafer inspection results are shown in Table 1 below:

[0038] Table 1 Wafer Inspection Results Table

[0039] From the test results in Table 1, it can be seen that the recognition accuracy rate of the vision inspection algorithm for redundant defects is 100%. There is 1 misdetection for both the scratch and crystal corner defect types, and 2 normal grains are misdetected. A small number of misidentifications of normal grains will not affect the enterprise production. At the same time, the overall detection accuracy rate is 99.87%, within the allowable error range. Through comprehensive analysis, this detection algorithm has accurate recognition and can meet the actual production requirements.

[0040] The above shows and describes the basic principles, main features, and advantages of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to include all changes falling within the meaning and scope of the equivalent elements of the claims in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0041] Other parts not detailed in the present invention belong to the prior art, so they will not be elaborated here.

[0042] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for visually inspecting surface defects of semiconductor wafers, characterized in that: The following steps are involved: S1: Obtain wafer images, calculate the deflection angle through Hough linear transformation and complete wafer rotation correction. After rotation correction, find all contours in the wafer and read the contour area; S2: In the process of grain contour extraction, image preprocessing is required; S3: Then cluster and segment the extracted image to distinguish the defects and background in the wafer surface image with similar colors and sizes; S4: Then, contour feature extraction is performed on the image that has been clustered and segmented in S3. After clustering and segmentation, the foreground image and background of the grain have been segmented into two categories, and the contour features of the grain and the defect point are clearly displayed. By analyzing the defect type features, it is concluded that when there are defects on the surface of the grain, the outer contour of the grain is deformed or there is a nested contour; S5: The HU moment contour matching image contour matching algorithm is used to calculate the similarity of the contour HU invariant moment. The larger the similarity value, the higher the similarity, and the higher the detection requirement, and the wafer surface defects can be classified.

2. The semiconductor wafer surface defect visual inspection method according to claim 1, characterized in that: The wafer rotation correction in S1 is to perform binarization processing on the wafer image, select multiple local areas to use Hough linear transformation to extract straight lines from the wafer grooves, calculate the average tilt angle based on the extracted straight lines, and judge whether the tilt angle is greater than a threshold value. If it is greater than the threshold value, the rotation angle is sent to the transplanting robot for rotation, and the image of the rotated product circle is taken. If it is less than the threshold value, it means that the angle will not affect the grain extraction and no rotation is required to directly extract the grain; if the read contour area is less than the set lower limit or greater than the set upper limit, it is judged that the contour is not a grain contour, and the next contour information is read. Otherwise, the contour is judged to be a grain contour, the contour coordinate information is recorded, and the grain is extracted according to the grain contour coordinate information.

3. The semiconductor wafer surface defect visual inspection method according to claim 1, characterized in that: In the S2, when preprocessing the image, different filters are selected according to the noise type to remove noise, reduce interference, protect grain contour information, and enhance the image, purposefully strengthen the area of ​​interest while suppressing the background area, improve image contrast, and obtain an image that is more conducive to computer analysis; the image preprocessing adopts morphological closed reconstruction, and the morphological closed reconstruction uses limited structural elements to extract shape feature information in the image, and then perform feature analysis and target recognition on the image.

4. The semiconductor wafer surface defect visual inspection method according to claim 3, characterized in that: The enhanced image can improve the contrast of the grayscale image, highlight the characteristic information of the grain edge and defects, and reduce the complexity of subsequent image processing and analysis processes.

5. The semiconductor wafer surface defect visual inspection method according to claim 1, characterized in that: The cluster segmentation in S3 adopts the idea of ​​EnFcm to improve the segmentation efficiency of the K-means algorithm. The grayscale histogram is used as a cluster sample for segmentation by utilizing the characteristic that the grayscale level in the grayscale image is much smaller than the number of pixels. The distance measurement between the same grayscale value and the cluster center is equal.

6. The semiconductor wafer surface defect visual inspection method according to claim 1, characterized in that: The defect detection in S4 utilizes the characteristics of the contour tree and contour moment to detect whether there are defects on the surface of the grain.

7. The semiconductor wafer surface defect visual inspection method according to claim 1, characterized in that: The contour extraction algorithm in S4 takes a point in the connected area as a starting point, scans pixel points step by step, tracks its contour, and marks pixels on the contour.

8. The semiconductor wafer surface defect visual inspection method according to claim 1, characterized in that: The S5 moment is often used in image processing to represent objects after image segmentation.

9. A semiconductor wafer surface defect visual inspection system, characterized in that: Applied to the detection method described in any one of claims 1 to 8 above, the detection system includes a grain extraction module, an image preprocessing module, an image segmentation module, and a defect recognition module, the grain extraction module is used for wafer rotation and contour extraction, the image preprocessing module is used for image filtering and image enhancement, the image segmentation module is used for grayscale statistics and cluster segmentation, and the defect recognition module is used for contour analysis and contour matching.

10. The semiconductor wafer surface defect visual inspection system according to claim 9, characterized in that: The image segmentation module divides the image into multiple independent regions with specific properties, and the algorithm extracts all contours in the image, including inner contours and outer contours.

Images