A wafer appearance defect detection method based on machine vision

Abstract

The present invention discloses a method for detecting wafer appearance defects based on machine vision, comprising the following steps: S1. image acquisition, converting the wafer image into a grayscale image; S2. preprocessing, performing mean filtering on the image, then downsampling, and performing top-hat and black-hat operations respectively; S3. defect extraction and determination: thresholding the image using a threshold to obtain a binary image; performing connected region calculation on the binary image to obtain several connected regions, calculating the average grayscale difference between the connected regions and the surrounding background; fitting the connected regions to obtain straight lines, calculating the standard deviation of the distances from the edge points of the connected regions to the straight lines; determining the slopes between the obtained connected regions, and merging two connected regions into one connected region when the slopes are consistent and relatively close; calculating the length of the connected region, and determining that the connected region is scratched if it is greater than a threshold. The present invention can improve detection efficiency and achieve high accuracy through comprehensive determination.

Description

Technical Field

[0001] The present invention relates to the field of wafer detection technology, and in particular to a method for detecting wafer appearance defects based on machine vision. Background Art

[0002] Wafer substrates are typically made of sapphire, silicon carbide, and silicon, with sapphire being the most widely used. However, various factors can cause defects such as scratches, pits, and bubbles during the manufacturing process, making it essential to inspect finished products during the final inspection process on the manufacturing line.

[0003] Machine vision captures wafer images, uses algorithms to analyze them, and measures defects. This fatigue-free method enables non-contact, rapid, and long-term online defect detection. Replacing traditional manual vision methods with machine vision offers a new approach to wafer defect detection research. Therefore, research into automated defect detection methods based on machine vision has become a key development direction in wafer appearance inspection.

[0004] The invention application with publication number CN108648168A discloses a method for detecting surface defects of IC wafers. The method spatially aligns a standard image and an image to be tested to obtain a difference image. The differences between the image to be tested and the standard image are then found. The maximum inter-class variance method is used to binarize the difference image, and then a morphological method is used to process the image to obtain a defect image. For the defect image, the defect is determined by the number of segments of the closed boundary contour, but determination based on the number of segments is prone to deviation.

[0005] The invention application with publication number CN114910480A discloses a wafer surface defect detection method based on machine vision, which uses an unsupervised learning classification model to perform wafer surface defect detection, and the detection process is complicated. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for detecting wafer appearance defects based on machine vision to improve detection efficiency and accuracy. To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] The present invention discloses a method for detecting wafer appearance defects based on machine vision, comprising the following steps:

[0008] S1. Image Acquisition

[0009] The wafer image is collected, converted into a grayscale image, and the black outer area at the edge of the grayscale image is removed to obtain the image P of the area to be inspected.

[0010] S2. Preprocessing

[0011] The image P is mean filtered and then downsampled to obtain the image S. The image S is subjected to top-hat and black-hat operations to obtain images T1 and T2.

[0012] S3. Defect extraction and determination

[0013] Extract and determine defects from the images T1 and T2, respectively, wherein the defects include at least scratch defects;

[0014] a11. Threshold the image using a threshold μ1 to obtain a binary image B1; wherein the threshold μ1 is the threshold of binarization after a top-hat operation or a black-hat operation.

[0015] a12. Perform connected area calculation on the binary image B1 to obtain several connected areas A51, and calculate the average grayscale difference between the connected area A51 and the surrounding background; if the average grayscale difference is greater than the set value, continue with the screening in step S33; if the average grayscale difference is less than the set value, it is determined that there is no defect and the process ends.

[0016] a13. Fit the connected area A51 to obtain a straight line, calculate the standard deviation of the distance from the edge points of the connected area A51 to the straight line, and if the standard deviation is less than the set threshold L1, mark it as a connected area A52.

[0017] a14. Determine the slopes of the connected areas A52 obtained. If the slopes of two connected areas A52 are consistent and relatively close, merge them into one connected area A53.

[0018] a15. Calculate the length of the connected area A53. If it is greater than the threshold L2, determine that the connected area A53 is a scratch and calculate the number of scratches.

[0019] S4. Merge the defects determined by images T1 and T2 to obtain the final wafer appearance defects.

[0020] Furthermore, the defects also include poor polishing defects; the number of connected areas A53 is calculated, and if the number is greater than a threshold value N1, it is determined that the wafer has poor polishing defects.

[0021] The threshold μ1 ranges from 3 to 20, the threshold L2 is 300 pixels, the threshold L1 is 1 pixel, and the threshold N1 is 20 pixels.

[0022] Preferably, in step S33, the connected area A51 is fitted with the following straight line equation to obtain a straight line:

[0023]

[0024] Among them, x i ,y iare the X-axis and Y-axis coordinates of the i-th edge point, and n is the number of edge points.

[0025] The calculation formula for the standard deviation σ of the distance from the edge points to the straight line of the connected area A51 is as follows:

[0026]

[0027] Among them, L K is the distance from the kth edge point to the fitting line, and n is the number of edge points.

[0028] Preferably, the connected region calculation adopts a two-pass scanning method:

[0029] (1) Start from the upper left corner of the image and traverse the pixels, find the first point with a pixel value of 255, and set label = 1.

[0030] (2) When the left neighbor pixel and the upper neighbor pixel of the pixel have invalid values, a new label value, label++, is set for the pixel and the set is recorded.

[0031] (3) When one of the left neighboring pixels or the upper neighboring pixels of the pixel has a valid value, the label of the valid value pixel is assigned to the label value of the pixel.

[0032] (4) When both the left neighbor pixel and the upper neighbor pixel of the pixel have valid values, the smaller label value is selected and assigned to the label value of the pixel.

[0033] (5) Update the label of each point to the smallest label in its set.

[0034] After the scan is completed, the pixels with the same label value in the image form the same connected area.

[0035] Preferably, the defects also include small pit defects; the extraction and determination steps are as follows:

[0036] a31. Threshold the image T1 or T2 using a threshold μ2 to obtain a binary image B2; wherein the threshold μ2 is the threshold of the binarization after the top hat operation or the black hat operation.

[0037] a32. Calculate the connected regions of the binary graph B2 to obtain several connected regions A61.

[0038] a33. Calculate the average grayscale of the middle and outer circles of the connected area A61, as well as the roundness of the middle black connected area. If the roundness of the middle black connected area is greater than 0.7, the average grayscale of the middle is less than 30, and the average grayscale of the outer circle is greater than 220, mark it as a connected area A62.

[0039] a34. Calculate the area of ​​the connected region A62. If it is larger than a preset threshold value S1, the connected region is determined to be a small pit and the wafer has a small pit defect.

[0040] Furthermore, the calculation formulas for the average grayscale grey(A61) in the middle of the connected area A61, the average grayscale grey(ΔA61) in the outer circle, and the roundness YD of the middle black connected area are as follows:

[0041]

[0042]

[0043]

[0044] Where f(x,y) is the grayscale value of the coordinate (x,y) in the figure, A represents the area of ​​the region, P represents the perimeter of the region, and the connected region that satisfies the above formula is marked as connected region A62.

[0045] The threshold μ2 has a value range of 3-20, and the threshold S1 is 50 square pixels.

[0046] Furthermore, the defects also include bubble defects; the extraction and determination steps are as follows:

[0047] a41. Threshold the image using a threshold μ3 to obtain a binary image B3; wherein the threshold μ3 is the threshold of binarization after top-hat operation or black-hat operation.

[0048] a42. Calculate the connected regions of the binary graph B3 to obtain several connected regions A71.

[0049] a43. Calculate the average grayscale of the connected area A71. If the average grayscale is greater than 240, mark it as the connected area A72.

[0050] a44. Calculate the area of ​​the connected region A72. If it is larger than a preset threshold value S2, the connected region is determined to be a bubble and the wafer has a bubble defect.

[0051] Preferably, the calculation formula of the average grayscale grey(A71) of the connected area A71 is as follows:

[0052]

[0053] Where f(x,y) is the grayscale value of the coordinate (x,y) in the figure.

[0054] The threshold μ3 has a value range of 3-20, and the threshold S2 is 500 square pixels.

[0055] Furthermore, in step a14, the method for determining whether two connected regions are close is as follows: taking the midpoints P1 and P2 of the two connected regions respectively, measuring the distance d between P1 and P2, and if the distance d satisfies the following formula, then the two connected regions are determined to be close;

[0056]

[0057] where x P1 、x P2 are the X-axis coordinates of point P1 and point P2, P1 、y P2 They are the Y-axis coordinates of points P1 and P2 respectively.

[0058] Due to the adoption of the above structure, the present invention has the following beneficial effects:

[0059] 1. This invention uses connected region calculation to extract and identify defects using grayscale images, simplifying the calculation and improving detection efficiency. The method also calculates the average grayscale difference of the connected region, the standard deviation of the distance from the edge point to the fitted line, and the slope, and then comprehensively determines whether the connected region is a scratch. This comprehensive determination is highly accurate.

[0060] 2. The present invention obtains two images through top hat operation and black hat operation, extracts and judges them respectively, and finally merges the judged defects to make the judgment more accurate.

[0061] 3. The present invention can comprehensively extract and determine defects such as poor scratches, poor polishing, poor small pits, and poor bubbles, thereby realizing defect classification, improving detection accuracy, and facilitating rework in later production. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is the wafer image acquired in step S1.

[0063] Figure 2 is the image after mean filtering.

[0064] Figure 3 It is an image of several connected regions A51 obtained by performing connected region calculation on the binary image B1.

[0065] Figure 4 It is an image of the connected area A53 obtained by merging the connected areas.

[0066] Figure 5 It is the test result display. DETAILED DESCRIPTION

[0067] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the present invention is described in further detail below with reference to the accompanying drawings and specific embodiments.

[0068] The threshold value in the present invention refers to the difference value in the machine image processing software, which is the difference between two pixels.

[0069] Example 1

[0070] This embodiment discloses a method for detecting wafer appearance defects based on machine vision, which can detect scratch defects on wafers.

[0071] This embodiment includes the following steps:

[0072] S1. Image Acquisition

[0073] like Figure 1 As shown, a wafer image is captured and converted into a grayscale image. The black outer area at the edge of the grayscale image is removed to obtain an image P of the area to be inspected.

[0074] S2. Preprocessing

[0075] Perform mean filtering on the image P and then downsample it to obtain image S. Perform top hat operation and black hat operation on image S to obtain images T1 and T2.

[0076] S3. Defect extraction and determination

[0077] Extract and determine defects from images T1 and T2, respectively, where the defects include at least scratch defects.

[0078] a11. Use the threshold μ1 to threshold the image and obtain the binary image B1, as shown in Figure 2 As shown in Figure 2, the threshold μ1 is the binarization threshold after top-hat or black-hat operation. The value range of the threshold μ1 is 3-20, and the threshold μ1 is based on comprehensive considerations of factors such as the camera and light source.

[0079] S32. Calculate the connected regions of the binary graph B1 to obtain several connected regions A51, such as Figure 3 Calculate the average grayscale difference between the connected area A51 and the surrounding background.

[0080] The connected region calculation adopts a two-pass scanning method, which includes the following steps:

[0081] (1) Start from the upper left corner of the image and traverse the pixels, find the first point with a pixel value of 255, and set label = 1.

[0082] (2) When the left neighbor pixel and the upper neighbor pixel of the pixel have invalid values, a new label value, label++, is set for the pixel and the set is recorded.

[0083] (3) When one of the left neighboring pixels or the upper neighboring pixels of the pixel has a valid value, the label of the valid value pixel is assigned to the label value of the pixel.

[0084] (4) When both the left neighbor pixel and the upper neighbor pixel of the pixel have valid values, the smaller label value is selected and assigned to the label value of the pixel.

[0085] (5) Update the label of each point to the smallest label in its set.

[0086] After the scan is completed, the pixels with the same label value in the image form the same connected area.

[0087] a13. Fit connected area A51 to a straight line. Calculate the standard deviation of the distances from the edge points of connected area A51 to the straight line. If the standard deviation is less than a set threshold L1, mark it as connected area A52. L1 is set to 1, meaning that when the standard deviation is less than 1 pixel, connected area A51 is considered a straight line and marked as connected area A52. Otherwise, when the standard deviation is greater than 1 pixel, connected area A51 is considered not a straight line.

[0088] The connected area A51 is fitted with a straight line using the following fitting line equation:

[0089]

[0090] Among them, x i ,y i are the X-axis and Y-axis coordinates of the i-th edge point, and n is the number of edge points.

[0091] The calculation formula for the standard deviation σ of the distance from the edge points to the straight line of the connected area A51 is as follows:

[0092]

[0093] Among them, L K is the distance from the kth edge point to the fitting line, and n is the number of edge points.

[0094] a14. Determine the slopes of the connected areas A52 obtained. If the slopes of two connected areas A52 are consistent and close to each other, merge them into one connected area A53. Figure 4 shown.

[0095] The method to determine whether two connected regions are close is as follows: establish an XY axis scale on the image, take the midpoints P1 and P2 of the two connected regions respectively, and measure the distance d between P1 and P2. If the distance d satisfies the following formula, the two connected regions are considered close;

[0096]

[0097] where x P1 、x P2 are the X-axis coordinates of point P1 and point P2, P1 、y P2 They are the Y-axis coordinates of points P1 and P2 respectively.

[0098] a15. Calculate the length of the connected area A53. If it is greater than a threshold value L2, the connected area A53 is determined to be a scratch. In this embodiment, the threshold value L2 is 300 pixels. That is, if the length of the connected area A53 is greater than 300 pixels, the wafer has a scratch defect.

[0099] Repeat steps a11 to a15 for image T2 to obtain the defect of image T2.

[0100] S4. Merge the defects determined by images T1 and T2 to obtain the final wafer appearance defects.

[0101] like Figure 5 As shown, in this embodiment, the defect detected by image T1 is 1 scratch, and the result obtained by detecting image T2 is 0 scratches. The detection results of the two images are merged to obtain the final detection result. There is a scratch defect in the illustrated box area of ​​the wafer, and the number of scratches is 1.

[0102] Example 2

[0103] This embodiment discloses a method for detecting wafer appearance defects based on machine vision, which can detect defects such as scratches, polishing defects, small pits, and bubbles on wafers.

[0104] This embodiment continues to perform the following steps of detection based on the scratch detection of the wafer in the first embodiment.

[0105] a2. Polishing defect detection

[0106] The number of connected areas A53 is calculated. If the number is greater than a threshold value N1, it is determined that the wafer has a poor polishing defect.

[0107] In this embodiment, the threshold value N1 is 20. That is, when the number of connected areas A53 is greater than 20, it is determined that the wafer is poorly polished.

[0108] a3. Detection of small pit defects.

[0109] In a machine vision imaging system, small pits are imaged as black in the middle and white on the outside, and are approximately circular. The extraction and determination steps are as follows:

[0110] a31. Threshold image T1 or T2 using threshold μ2 to obtain binary image B2. Here, threshold μ2 is the binarization threshold after top-hat or black-hat operation. The value range of threshold μ2 is 3-20, and it is based on a comprehensive consideration of factors such as the camera and light source.

[0111] a32. Calculate the connected regions of the binary graph B2 to obtain several connected regions A61.

[0112] a33. Calculate the average grayscale of the middle and outer circles of the connected area A61, as well as the roundness of the middle black connected area. If the roundness of the middle black connected area is greater than 0.7, the average grayscale of the middle is less than 30, and the average grayscale of the outer circle is greater than 220, mark it as a connected area A62.

[0113] The calculation formulas for the average grayscale grey(A61) in the middle of the connected area A61, the average grayscale grey(ΔA61) in the outer circle, and the roundness YD of the middle black connected area are as follows:

[0114]

[0115]

[0116]

[0117] Where f(x,y) is the grayscale value of the coordinate (x,y) in the figure, A represents the area of ​​the region, P represents the perimeter of the region, and the connected region that satisfies the above formula is marked as connected region A62.

[0118] a34. Calculate the area of ​​connected region A62. If it is greater than a pre-set threshold S1, the connected region is determined to be a pit, and the wafer is considered to have a pit defect. In this embodiment, threshold S1 is 50 square pixels. That is, if the area of ​​connected region A62 is greater than 50 square pixels, the connected region is determined to be a pit. Pixels, also known as resolution, refer to the array of horizontal and vertical pixels that can be displayed. A square pixel is the number of pixels per square.

[0119] a4. Bubble defects

[0120] In the machine vision imaging system, the judgment principle is: if the image is white, the average grayscale is greater than 240, and the area is greater than 500, it is a bubble.

[0121] The extraction and determination steps are as follows:

[0122] a41. Threshold image T1 or T2 using threshold μ3 to obtain binary image B3. Here, threshold μ3 is the binarization threshold after top-hat or black-hat operation. The value range of threshold μ3 is 3-20. Threshold μ2 is determined by comprehensive considerations of factors such as the camera and light source.

[0123] a42. Calculate the connected regions of the binary graph B3 to obtain several connected regions A71.

[0124] a43. Calculate the average grayscale of the connected area A71. If the average grayscale is greater than 240, mark it as the connected area A72. The calculation formula for the average grayscale grey(A71) of the connected area A71 is as follows:

[0125]

[0126] Where f(x,y) is the grayscale value of the coordinate (x,y) in the figure;

[0127] a44. Calculate the area of ​​connected region A72. If it is greater than a preset threshold S2, the connected region is determined to be a bubble, and the wafer has a bubble defect. In this embodiment, threshold S2 is 500 square pixels, meaning that if the area of ​​connected region A72 is greater than 500, the connected region is determined to be a bubble.

[0128] In the above steps, the order of performing defect inspections on the wafer for scratches, polishing, pits, and bubbles can be changed arbitrarily.

[0129] S4. Merge the defects determined by images T1 and T2 to obtain the final wafer appearance defects.

[0130] You can first inspect all defect types on image T1, then inspect all defect types on image T2, and finally merge the defects. You can also inspect both images T1 and T2 for a certain defect type, then switch to another defect and inspect both images separately until all defect types are inspected, and finally merge the defects.

[0131] For example, the defects detected by image T1 are 1 scratch, 2 small pits, and 1 bubble, and the defects detected by image T2 are 2 scratches and 1 small pit. The detection results of the two images are merged to obtain the final detection result: there are scratch defects, small pit defects, and bubble defects, the number of scratches is 3, the number of small pits is 3, and the number of bubbles is 1.

[0132] The detection method of the present invention is particularly suitable for detecting appearance defects of wafers with sapphire substrates, and has high detection efficiency and good practicability.

[0133] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for detecting wafer appearance defects based on machine vision, characterized in that: The following steps are involved: S1. Image Acquisition Acquire a wafer image, convert the wafer image into a grayscale image, remove the black outer area at the edge of the grayscale image to obtain an image P of the area to be inspected; S2. Preprocessing Perform mean filtering on image P and then downsample it to obtain image T. Perform top hat operation and black hat operation on image S to obtain images T1 and T2 respectively. S3. Defect extraction and determination Extract and determine defects from the images T1 and T2, respectively, wherein the defects include at least scratch defects; a11. The image is thresholded using a threshold μ1 to obtain a binary image B1; wherein the threshold μ1 is the threshold value of the binarization after the top hat operation or the black hat operation; a12 binary image B1 connected area calculation to obtain several connected areas A51, calculate the average grayscale difference between the connected area A51 and the surrounding background; if the average grayscale difference is greater than the set value, then proceed to step S33 screening, if the average grayscale difference is less than the set value, it is determined that there is no defect, the process ends; a13. Fit the connected area A51 to a straight line, calculate the standard deviation of the distance from the edge points of the connected area A51 to the straight line, and if the standard deviation is less than the set threshold L1, mark it as a connected area A52; a14. Determine the slope between the connected areas A52. When the slopes of the two connected areas A52 are consistent and close to each other, merge them into one connected area A53. a15. Calculate the length of the connected area A53. If it is greater than the threshold value L2, the connected area A53 is determined to be scratched, and the number of scratches is calculated; The defects also include small pit defects; the extraction and determination steps are as follows: a31. The image is thresholded using a threshold μ2 to obtain a binary image B2; wherein the threshold μ2 is the threshold value after the top-hat operation or the black-hat operation; a32. Calculate the connected regions of the binary graph B2 to obtain several connected regions A61; a33. Calculate the average grayscale of the middle and outer circles of connected area A61, as well as the roundness of the middle black connected area. If the roundness of the middle black connected area is greater than 0.7, the average grayscale of the middle is less than 30, and the average grayscale of the outer circle is greater than 220, mark it as connected area A62. In step a33, the average grayscale of the middle of the connected area A61 Average grayscale of the outer ring , and the calculation formula of the circularity YD of the middle black connected area is as follows: Where f(x,y) is the grayscale value of the coordinate (x,y) in the figure, A represents the area of ​​the region, P represents the perimeter of the region, and the connected region that satisfies the above formula is marked as connected region A62; a34. Calculate the area of ​​the connected region A62. If it is greater than the preset threshold S1, the connected region is determined to be a small pit, and the wafer has a small pit defect. S4. Merge the defects determined in images T1 and T2 to obtain the final wafer appearance defects.

2. The method for detecting wafer appearance defects based on machine vision according to claim 1, wherein: The defects also include poor polishing defects: a2: Calculate the number of connected areas A53. If the number is greater than a threshold value N1, it is determined that the wafer has a poor polishing defect.

3. The method for detecting wafer appearance defects based on machine vision according to claim 2, wherein: The threshold μ1 has a value range of 3-20, the threshold L2 has a value of 300 pixels, the threshold L1 has a value of 1 pixel, and the threshold N1 has a value of 20 pixels.

4. The method for detecting wafer appearance defects based on machine vision according to claim 1, wherein: In step a13, the connected area A51 is fitted with the following straight line equation to obtain a straight line: in, ,y i are the X-axis and Y-axis coordinates of the i-th edge point, and n is the number of edge points; The standard deviation of the distance from the edge point to the straight line of the connected area A51 The calculation formula is as follows: Among them, L K is the distance from the kth edge point to the fitting line, and n is the number of edge points.

5. The method for detecting wafer appearance defects based on machine vision according to claim 1, wherein: The connected area calculation adopts a two-pass scanning method: (1) Start from the upper left corner of the image and traverse the pixels, find the first point with a pixel value of 255, and set label = 1; (2) When the left neighbor pixel and the upper neighbor pixel of the pixel have invalid values, set a new label value for the pixel, label++, and record the set; (3) When one of the left neighboring pixels or the upper neighboring pixels of the pixel has a valid value, the label of the valid value pixel is assigned to the label value of the pixel; (4) When both the left neighbor pixel and the upper neighbor pixel of the pixel have valid values, the smaller label value is selected and assigned to the label value of the pixel; (5) Update the label of each point to the smallest label in its set; After the scan is completed, the pixels with the same label value in the image form the same connected area.

6. The method for detecting wafer appearance defects based on machine vision according to claim 1, wherein: The threshold μ2 has a value range of 3-20, and the threshold S1 is 50 square pixels.

7. The method for detecting wafer appearance defects based on machine vision according to claim 1, wherein: The defects also include bubble defects; the extraction and determination steps are as follows: a41. The image is thresholded using a threshold μ3 to obtain a binary image B3; wherein the threshold μ3 is the threshold value of the binarization after the top hat operation or the black hat operation; a42. Calculate the connected regions of the binary graph B3 to obtain several connected regions A71; a43. Calculate the average grayscale of the connected area A71. If the average grayscale is greater than 240, mark it as the connected area A72. a44. Calculate the area of ​​the connected region A72. If it is larger than a preset threshold value S2, the connected region is determined to be a bubble and the wafer has a bubble defect.

8. The method for detecting wafer appearance defects based on machine vision according to claim 7, wherein: The average grayscale of the connected area A71 The calculation formula is as follows: Where f(x,y) is the grayscale value of the coordinate (x,y) in the figure; The threshold μ3 has a value range of 3-20, and the threshold S2 is 500 square pixels.

9. The method for detecting wafer appearance defects based on machine vision according to any one of claims 1 to 8, wherein: In step a14, the method for determining whether two connected regions are close is as follows: take the midpoints P1 and P2 of the two connected regions respectively, measure the distance d between P1 and P2, and if the distance d satisfies the following formula, then the two connected regions are determined to be close; d≤ in 、 are the X-axis coordinates of points P1 and P2 respectively, 、 They are the Y-axis coordinates of points P1 and P2 respectively.

Images