Wafer surface smoothness detection method and system for visual inspection machine

Abstract

The invention relates to the technical field of image data processing, in particular to a wafer surface smoothness detection method and system for a visual inspection machine. The method comprises the following steps: acquiring a grayscale image of a wafer surface image; equally dividing the grayscale image into a plurality of blocks; determining a first qualification coefficient of the blocks; edge pixel points in the blocks are obtained; determining a second qualification coefficient of the blocks; and clustering the grayscale image by using a Naive hierarchical clustering algorithm so as to identify a smooth area in the grayscale image and realize wafer surface smoothness detection. According to the method, the first qualified coefficient and the second qualified coefficient are introduced, multiple factors in the blocks are comprehensively considered, and the characteristics of the blocks can be described more comprehensively and meticulously, so that the unsmooth region and the etching texture region are effectively distinguished, the interference of the etching texture region on the judgment of the unsmooth region is reduced, the clustering accuracy is improved, and the clustering efficiency is improved. Therefore, the unsmooth area can be identified more accurately, and the accuracy of wafer surface smoothness detection is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and in particular to a wafer surface smoothness detection method and system for a visual inspection machine. Background Art

[0002] With the continuous advancement of emerging technologies such as artificial intelligence, 5G, the Internet of Things, and quantum computing, the semiconductor industry, as the foundation supporting these technologies, is experiencing unprecedented rapid development. As the core of semiconductor products, the smoothness inspection of wafers is crucial to ensure efficient manufacturing, while traditional manual visual inspection is inefficient and inaccurate, and needs to be replaced by automated, high-precision inspection technology.

[0003] An existing method for detecting the surface smoothness of a wafer is the Naive Hierarchical Clustering Algorithm. The Naive Hierarchical Clustering Algorithm does not require the number of clusters to be defined in advance and can effectively process complex and irregular data, which makes the Naive Hierarchical Clustering Algorithm an effective tool for detecting the surface smoothness of a wafer.

[0004] However, when using the Naive hierarchical clustering algorithm for image processing, in order to reduce the amount of calculation and improve the efficiency of the algorithm, the image will be divided into several equal-sized image blocks before clustering begins, and each image block will be used as an initial cluster for subsequent clustering. During the clustering process, the two clusters will be continuously merged according to the distance between them until only one cluster remains in the cluster set, and the clustering result is obtained; the distance between the two clusters is often quantified into a characteristic factor by analyzing the gray value performance in each image block. However, when detecting the surface smoothness of the chip, the image blocks of the potential rough area will have similar gray value performance with the image blocks of some etched texture areas on the chip surface. The image blocks of the etched texture area will greatly affect the judgment of the potential rough area, thereby reducing the accuracy of the clustering results of the Naive hierarchical clustering algorithm, and further affecting the accuracy of the detection of the surface smoothness of the chip. Summary of the invention

[0005] In order to solve the problem that the rough area on the wafer surface has a similar gray value performance as the etched texture area on the wafer surface, and the etched texture area will greatly affect the judgment of the potential rough area, resulting in reduced accuracy when clustering using the Naive hierarchical clustering algorithm, thereby affecting the accuracy of wafer surface smoothness detection, the present invention provides a wafer surface smoothness detection method and system for a visual inspection machine.

[0006] In a first aspect, the present invention provides a wafer surface smoothness detection method for a visual inspection machine, which adopts the following technical solution: A method for detecting the smoothness of a wafer surface for a visual inspection machine, comprising: obtaining a grayscale image of the wafer surface image; equally dividing the grayscale image into a plurality of blocks, and determining a first qualification coefficient of the block according to the median of the grayscale values of all pixel points in the grayscale image, the mean value of the grayscale values of all pixel points in the block, and the number of mode values among the grayscale values of all pixel points in the block; recording the pixel points with gradient amplitude changes in any neighborhood direction within the block as edge pixel points; in response to the block containing edge pixel points, determining a second qualification coefficient of the block according to the first qualification coefficient, the number of edge pixel points in the block, the number of pixel points within the minimum circumscribed rectangle of all edge pixel points in the block, and the information entropy value of the gradient amplitudes corresponding to all edge pixel points in the block, otherwise, assigning the first qualification coefficient of the block to the second qualification coefficient of the block; clustering the grayscale image using the Naive hierarchical clustering algorithm according to the second qualification coefficient to obtain a dendrogram of the Naive hierarchical clustering algorithm, so as to identify the smooth regions in the grayscale image and realize the smoothness detection of the wafer surface.

[0007] The beneficial effects are as follows: By using multiple features, such as the median, mean value, and the number of mode values, to determine the first qualification coefficient of the block, it can effectively distinguish the non-smooth regions and etched texture regions on the surface, avoiding misjudgment caused by the similarity of the gray values between the two; By judging whether the block contains edge pixel points and calculating features such as the number of edge pixel points and the minimum circumscribed rectangle, it can accurately identify the regions with obvious unevenness or defects on the surface, further improving the accuracy of the smoothness detection; Marking the pixel points with gradient amplitude changes within the block and calculating the information entropy value can strengthen the processing of the image texture complexity, reduce the influence of noise and non-target regions on the clustering algorithm, and thus improve the detection accuracy.

[0008] Further, the grayscale image is obtained by performing grayscale processing on the acquired wafer surface image using the weighted average method.

[0009] Further, the equally dividing the grayscale image into a plurality of blocks includes: taking any corner of the grayscale image as a reference point, equally dividing the grayscale image into a plurality of blocks. If there is a remaining area that cannot be evenly divided into blocks, then according to the order of first row and then column, and according to the size of the remaining area, it is divided, including: pre-dividing the remaining area according to the side length of the adjacent blocks to obtain a plurality of pre-divided areas. In response to the row length and column length of the pre-divided area being both less than , merging the pre-divided area into the adjacent blocks, otherwise, taking the pre-divided area as a block; wherein, is the preset side length of the block, is the minimum side length of the preset block.

[0010] Further, the first qualification coefficient satisfies: ; where is the first qualification coefficient of the th block, is the median of the gray values of all pixel points in the grayscale image, is the th block, and is the mean of the gray values of all pixel points within the is the th block, and is the number of the mode values among the gray values of all pixel points within the is a hyperparameter, is the absolute value symbol.

[0011] The beneficial effect is that by calculating the difference between the median of all pixel points in the grayscale image and the mean of the gray values of pixel points in each block, combined with the number of mode values, the anomalies and differences in the gray distribution within the block can be captured more accurately, making it more sensitive when identifying non-smooth regions and reducing misjudgments caused by subtle surface texture changes.

[0012] Further, the gradient magnitude is calculated using the Sobel operator.

[0013] Further, the gradient magnitude is calculated using the Canny operator.

[0014] Further, the second qualification coefficient satisfies: ; where is the second qualification coefficient of the th block, is the number of edge pixel points within the th block, is the th block, and is the number of pixel points within the minimum bounding rectangle of all edge pixel points within the is the first qualification coefficient of the th block, is the th block, and is the information entropy value of the gradient magnitudes corresponding to all edge pixel points within the is the standard normalization function.

[0015] The beneficial effects are: by calculating the second qualified coefficient of each block, the distribution of edge pixels within the image block can be described more accurately, while taking into account the number of pixels in the minimum circumscribed rectangle and the entropy value of the gradient amplitude, the details of the edge information in the image can be better captured; the second qualified coefficient can effectively enhance the robustness of the algorithm to image noise and uneven lighting by combining the first qualified coefficient and the entropy value of the gradient amplitude; by introducing a standard normalization function and normalizing the second qualified coefficient of the block, the deviation caused by differences in image size or differences in blocking methods can be eliminated, thereby making the quality assessment between multiple blocks more consistent.

[0016] Furthermore, the grayscale image is clustered using the Naive hierarchical clustering algorithm according to the second qualified coefficient to obtain a dendrogram of the Naive hierarchical clustering algorithm, including: using the difference in the second qualified coefficients between the blocks as the distance between the blocks when the Naive hierarchical clustering algorithm processes the image; and clustering the grayscale image using the Naive hierarchical clustering algorithm in a single-link manner to generate a clustering result, i.e., a dendrogram of the Naive hierarchical clustering algorithm.

[0017] The beneficial effects are: by using the second qualified coefficient to measure the difference between image blocks, the similarity between different image regions can be described more accurately, so that the Naive hierarchical clustering algorithm can perform more refined clustering according to the real image features; compared with the traditional clustering method, the Naive hierarchical clustering algorithm clusters by a single link method, so that the image can obtain a hierarchical structure, i.e., a dendrogram, more quickly during the processing, thereby improving the computational efficiency of the clustering process and reducing a large amount of computation.

[0018] Furthermore, the method of detecting the smoothness of the chip surface by identifying the smooth area in the grayscale image includes: cropping the dendrogram according to a preset cluster distance to obtain a plurality of cropped clusters; in response to the number of blocks in the cropped cluster being less than a preset roughness threshold, determining that the grayscale image area corresponding to the cropped cluster is a non-smooth area; in response to the ratio of the area of all non-smooth areas to the area of the grayscale image being greater than a preset abnormal threshold, determining that the chip surface quality is unqualified, and completing the detection of the chip surface smoothness.

[0019] The beneficial effects are as follows: by cropping the grayscale image according to the dendrogram and judging the smoothness of the area according to the size of the cluster, the smooth area and the rough area on the chip surface can be accurately distinguished. By comparing the number of blocks in the rough area with the preset roughness threshold, the irregular or uneven surface can be effectively identified, thereby improving the accuracy of detection; by calculating the ratio of the area of the rough area to the area of the entire image, the overall smoothness of the chip surface can be quickly judged.

[0020] In a second aspect, the present invention provides a wafer surface smoothness detection system for an inspection machine, adopting the following technical solution: A wafer surface smoothness detection system for an inspection machine includes: a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned wafer surface smoothness detection method for an inspection machine is implemented.

[0021] By adopting the above technical solution, the above-mentioned wafer surface smoothness detection method for an inspection machine is generated into a computer program and stored in the memory to be loaded and executed by the processor, so as to manufacture a terminal device according to the memory and the processor, which is convenient to use.

[0022] The present invention has the following technical effects: By introducing the first qualification coefficient and the second qualification coefficient, multiple factors such as the median, mean, mode quantity, number of edge pixel points, number of pixel points within the minimum circumscribed rectangle of edge pixel points, and information entropy value of the gradient amplitude of edge pixel points in the block are comprehensively considered, which can more comprehensively and meticulously characterize the features of the block, thereby effectively distinguishing the non-smooth area and the etched texture area, reducing the interference of the etched texture area on the judgment of the non-smooth area, and improving the accuracy of clustering; when using the Naive hierarchical clustering algorithm for clustering, the second qualification coefficient is used as the distance metric between blocks. Since the second qualification coefficient synthesizes various information, compared with clustering only relying on gray values, it can more accurately reflect the similarity and difference between blocks, making the clustering result more reasonable, and being able to more clearly distinguish the smooth area and the non-smooth area, avoiding mis-clustering caused by similar gray values, thereby improving the accuracy of clustering; the accuracy of clustering directly affects the accuracy of the wafer surface smoothness detection. By accurately identifying the non-smooth area in the gray image, the smoothness of the wafer surface can be more accurately judged. For the inspection machine in industrial production, it means that wafers with non-conforming surfaces can be more effectively screened out, improving the reliability of product quality detection, reducing the outflow of defective products or misjudgment situations caused by inaccurate detection, thereby reducing production costs and improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is the flowchart of the method in an embodiment of the wafer surface smoothness detection method for an inspection machine of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] An embodiment of the present invention discloses a method for detecting the surface smoothness of a wafer for a visual inspection machine, referring to Figure 1 , including steps S1 - S6: S1: Obtain a grayscale image of the wafer surface image.

[0026] Specifically, the grayscale image is obtained by performing grayscale processing on the wafer surface image collected by a high-definition camera using the weighted average method.

[0027] S2: Divide the grayscale image into several equal-sized blocks.

[0028] Specifically, the step of dividing the grayscale image into several equal-sized blocks includes: Taking any corner of the grayscale image as a reference point, dividing the grayscale image into several blocks of equal size. If there is a remaining area that cannot be evenly divided into blocks, then divide it according to the order of rows first and columns second, and according to the size of the remaining area, including: Pre-divide the remaining area according to the side length of the adjacent blocks to obtain several pre-divided areas. In response to the row length and column length of the pre-divided areas both being less than , merge the pre-divided areas into the adjacent blocks, otherwise, take the pre-divided areas as a block; Among them, is the preset side length of the block, is the preset minimum side length of the block.

[0029] Implementers can set the side length and the minimum side length according to the specific implementation situation. For example, the side length is 15 and the minimum side length is 10.

[0030] S3: Determine the first pass coefficient of the block.

[0031] It should be noted that by analyzing the gray value performance within each image block, the first qualification coefficient of each image block is obtained. Since most regions on the wafer surface will have very close gray values, that is, the background region, and there will be gray value variations in the potential non-smooth regions due to differences in light reflection, the gray value of the background region on the wafer surface will be closer to the median gray value of the entire wafer surface image compared to the potential non-smooth regions. However, there will be some etching textures in the background region on the wafer surface, and these etching texture regions and the background region both belong to the normal regions on the wafer surface, and there will be a high degree of gray value consistency within both the etching texture regions and the background region on the wafer surface. Therefore, first, analyze the difference between the gray value within an image block and the median gray value in the gray image of the wafer surface image. The smaller the difference, the larger the first qualification coefficient. Then, continue to analyze the consistency of the gray values within an image block. The larger the consistency, the larger the first qualification coefficient.

[0032] Determine the first qualification coefficient of the block based on the median of the gray values of all pixel points in the gray image, the mean of the gray values of all pixel points within the block, and the number of mode values among the gray values of all pixel points within the block.

[0033] Specifically, the first qualification coefficient satisfies: ; In the formula, is the first qualification coefficient of the th block, is the median of the gray values of all pixel points in the gray image, is the mean of the gray values of all pixel points within the th block, is the number of mode values among the gray values of all pixel points within the th block, is a hyperparameter, is the absolute value symbol.

[0034] Implementers can set the hyperparameter according to the specific implementation situation. For example, 0.01. The existence of the hyperparameter is to prevent from making the formula result meaningless.

[0035] Among them, the smaller it is, the closer the gray value of the th block is to the median of the gray values of the entire wafer surface, and the larger the first qualification coefficient of the th block; the larger it is, the greater the gray value consistency within the th block, and it can also indicate that the The closer the gray value of a block is to the median gray value of the entire wafer surface, the greater the credibility, and then the greater the first qualification coefficient of the th block.

[0036] S4: Obtain the edge pixel points within the block.

[0037] Pixels with gradient magnitude changes in any neighborhood direction within the block are recorded as edge pixel points, where the neighborhood direction in the present invention is the 4-neighborhood direction.

[0038] S5: Determine the second qualification coefficient of the block.

[0039] It should be noted that the first qualification coefficient in the combined area of the etched texture area and the background area on the wafer surface will be close to that of the potential non-smooth area. Therefore, it is also necessary to analyze the texture feature performance within each image block, and combine the first qualification coefficient to obtain the second qualification coefficient of each image. Through the second qualification coefficient, it is possible to better distinguish between the potential non-smooth area and the normal area (background area, etched texture area). The difference in the texture feature performance between the potential non-smooth area and the etched texture area is that the area composed of edge pixel points within the etched texture area is relatively regular, while the shape of the area composed of edge pixel points in the potential non-smooth area is relatively irregular (this is due to the different formation causes of the two areas. The etched texture is produced according to a consistent specification, and the generation of the potential non-smooth area is more random). However, there may also be some potential non-smooth areas with relatively regular edge shapes. Then, continue to analyze the consistency of the gradient change intensity of the pixel points with gradient magnitude changes within an image block. If the consistency is poor, it is a potential non-smooth area; otherwise, it is an etched texture area. After analyzing the above features, if the texture feature performance within an image block more conforms to the texture feature performance of the potential non-smooth area, the second qualification coefficient corresponding to the image block will be smaller; if there are no edge pixel points within an image block, then the image block does not have the above features, and the value of the first qualification coefficient of the image block can be assigned to the second qualification coefficient for subsequent steps.

[0040] In response to the block containing edge pixel points, determine the second qualification coefficient of the block according to the first qualification coefficient, the number of edge pixel points within the block, the number of pixel points within the minimum circumscribed rectangle of all edge pixel points within the block, and the information entropy value of the gradient magnitudes corresponding to all edge pixel points within the block; otherwise, assign the first qualification coefficient of the block to the second qualification coefficient of the block.

[0041] Specifically, the gradient magnitude is calculated using the Sobel operator.

[0042] In another embodiment, the gradient magnitude is calculated using the Canny operator.

[0043] Specifically, the second qualification coefficient satisfies: ; In the formula, is the second qualification coefficient of the th block, is the number of inner edge pixels of the th block, is the number of pixels within the minimum bounding rectangle of all the edge pixels in the th block, is the first qualification coefficient of the th block, is the information entropy value of the gradient magnitudes corresponding to all the edge pixels in the th block, is the standard normalization function.

[0044] Among them, The larger it is, it indicates that from the perspective of the gray value performance, the th block is more likely to belong to the normal area of the wafer surface, and then the second qualification coefficient of the th block is larger; represents the ratio of the area of the minimum bounding rectangle of all the pixels (edge pixels) with gradient magnitude changes in the 4-neighborhood direction within the th block. The smaller this value is, it indicates that the shape of the area composed of the edge pixels in the th block is more irregular, and it indicates that its texture feature performance is more in line with the texture feature performance of the potential non-smooth area. Then the second qualification coefficient of the th block will be smaller; The larger it is, it indicates that the consistency of the gradient change intensity of the edge pixels in the th block is worse, which can confirm that the shape of the area composed of the edge pixels in the th block is more irregular with a greater degree of credibility. It can be shown that the texture feature performance within the th block is more in line with the texture feature performance of the potential non-smooth area, and the second qualification coefficient of the th block will be smaller.

[0045] S6: Use the Naive hierarchical clustering algorithm to cluster the grayscale image to identify the smooth areas in the grayscale image and achieve the smoothness detection of the wafer surface.

[0046] According to the second qualification coefficient, use the Naive hierarchical clustering algorithm to cluster the grayscale image to obtain a dendrogram of the Naive hierarchical clustering algorithm, so as to identify the smooth areas in the grayscale image and realize the smoothness detection of the wafer surface.

[0047] Specifically, the step of using the Naive hierarchical clustering algorithm to cluster the grayscale image according to the second qualification coefficient to obtain a dendrogram of the Naive hierarchical clustering algorithm includes: Take the difference in the second qualification coefficient between the blocks as the distance between the blocks when the Naive hierarchical clustering algorithm processes the image. Cluster the grayscale image using the Naive hierarchical clustering algorithm in a single-linkage manner to generate a clustering result, that is, a dendrogram of the Naive hierarchical clustering algorithm.

[0048] Specifically, the step of identifying the smooth areas in the grayscale image and realizing the smoothness detection of the wafer surface includes: Crop the dendrogram according to a preset clustering cluster distance to obtain a number of cropped clusters. In response to the number of blocks within the cropped cluster being less than a preset roughness threshold, determine that the grayscale image area corresponding to the cropped cluster is a non-smooth area. In response to the ratio of the area of all non-smooth areas to the area of the grayscale image being greater than a preset abnormality threshold, determine that the quality of the wafer surface is unqualified, and complete the smoothness detection of the wafer surface.

[0049] Implementers can set the clustering cluster distance, roughness threshold, and abnormality threshold according to specific implementation situations. For example, the clustering cluster distance is 0.3, the roughness threshold is 6, and the abnormality threshold is 0.03.

[0050] An embodiment of the present invention also discloses a wafer surface smoothness detection system for a visual inspection machine, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a method for detecting the smoothness of the wafer surface according to the present invention is implemented.

[0051] The above system also includes a communication bus and a communication interface, as well as other components well known to those skilled in the art. Their settings and functions are known in the art, so they will not be described in detail here.

[0052] The above are all preferred embodiments of the present invention. The protection scope of the present invention is not limited by this. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for detecting the surface smoothness of a wafer used in a visual inspection machine, characterized in that, Comprising: Obtaining a grayscale image of the wafer surface image; Equally dividing the grayscale image into a plurality of blocks; Determining a first qualification coefficient of the block according to the median of the grayscale values of all pixel points in the grayscale image, the mean value of the grayscale values of all pixel points in the block, and the number of the mode values of the grayscale values of all pixel points in the block; Denoting the pixel points with gradient amplitude changes in any neighborhood direction within the block as edge pixel points; In response to the block containing edge pixel points, determining a second qualification coefficient of the block according to the first qualification coefficient, the number of edge pixel points in the block, the number of pixel points within the minimum circumscribed rectangle of all edge pixel points in the block, and the information entropy value of the gradient amplitudes corresponding to all edge pixel points in the block, otherwise, assigning the first qualification coefficient of the block to the second qualification coefficient of the block; Clustering the grayscale image by using the Naive hierarchical clustering algorithm according to the second qualification coefficient to obtain a dendrogram of the Naive hierarchical clustering algorithm, so as to identify smooth regions in the grayscale image and realize the smoothness detection of the wafer surface.

2. A method for detecting the surface smoothness of a wafer used in a visual inspection machine according to claim 1, characterized in that, The grayscale image is a grayscale image of the wafer surface image obtained by performing grayscale processing on the collected wafer surface image by using a weighted average method.

3. A method for detecting the surface smoothness of a wafer for an optical inspection machine according to claim 1, characterized in that, The equally dividing the grayscale image into a plurality of blocks includes: Taking any corner of the grayscale image as the reference point, evenly divide the grayscale image into several sub-blocks. If there is a remaining area that cannot be evenly divided into sub-blocks, then divide it according to the order of rows first and columns second, and according to the size of the remaining area, including: Pre-divide the remaining area according to the side length of adjacent divided blocks to obtain a number of pre-divided areas. In response to the row length and column length of the pre-divided areas both being less than , merge the pre-divided areas into adjacent divided blocks; otherwise, regard the pre-divided areas as one divided block. Among them, is the side length of a preset block, is the minimum side length of a preset block.

4. A method for detecting the surface smoothness of a wafer used in a visual inspection machine according to claim 1, characterized in that, The first qualification coefficient satisfies: ； Wherein, is the first qualification coefficient of the th block, is the median of the gray values of all pixel points in the grayscale image, is the average of the gray values of all pixel points within the th block, is the number of the mode values among the gray values of all pixel points within the th block, is a hyperparameter, is the absolute value symbol.

5. A method for detecting the surface smoothness of a wafer used in a visual inspection machine according to claim 1, characterized in that, The gradient amplitude is calculated by using a Sobel operator.

6. A method for detecting the surface smoothness of a wafer used in a visual inspection machine according to claim 1, characterized in that, The gradient amplitude is calculated by using a Canny operator.

7. A method for detecting the surface smoothness of a wafer used in a visual inspection machine according to claim 1, characterized in that, The second qualification coefficient satisfies: ； In the formula, is the second qualification coefficient of the th block, is the number of inner-edge pixel points of the th block, is the number of pixel points within the minimum circumscribed rectangle of all edge pixel points in the th block, is the first qualification coefficient of the th block, is the information entropy value of the gradient amplitudes corresponding to all edge pixel points in the th block, is the standard normalization function.

8. A method for detecting the surface smoothness of a wafer used in a visual inspection machine according to claim 1, characterized in that, The clustering the grayscale image by using the Naive hierarchical clustering algorithm according to the second qualification coefficient to obtain a dendrogram of the Naive hierarchical clustering algorithm includes: Regarding the difference between the second qualification coefficients of the blocks as the distance between the blocks when the Naive hierarchical clustering algorithm processes the image; Clustering the grayscale image by using the Naive hierarchical clustering algorithm in a single-linkage manner to generate a clustering result, that is, a dendrogram of the Naive hierarchical clustering algorithm.

9. A method for detecting the surface smoothness of a wafer for an eye inspection machine according to claim 1, characterized in that, The identifying smooth regions in the grayscale image and realizing the smoothness detection of the wafer surface includes: Cropping the dendrogram according to a preset clustering cluster distance to obtain a plurality of cropped clusters; In response to the number of blocks within the cropped cluster being less than a preset roughness threshold, determining that the grayscale image region corresponding to the cropped cluster is a non-smooth region; In response to the ratio of the area of all non-smooth regions to the area of the grayscale image being greater than a preset abnormality threshold, determining that the quality of the wafer surface is unqualified and completing the smoothness detection of the wafer surface.

10. A wafer surface smoothness detection system for a visual inspection machine, characterized in that, Comprising: A processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a method for detecting the smoothness of the wafer surface for an eye inspection machine according to any one of claims 1-9 is implemented.

Images