A method for intelligently identifying surface defects of nanoimprint wafers
Through filtering and grayscale processing technology, three types of grayscale value combinations are constructed to statistics the combination of deviation and grayscale deviation, which solves the problem of low recognition accuracy of wafer surface defects in the existing technology, and achieves higher recognition accuracy and reliability.
Patent Information
- Application Number
- CN202411342501.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-25
AI Technical Summary
The existing wafer surface defect identification method has the problem of low accuracy, and the traditional method is inefficient and susceptible to subjective factors, making it difficult to effectively identify complex defects.
Using filtering and grayscale processing technology, three types of grayscale value combinations are constructed, and the deviation degree and grayscale deviation degree are combined are statistically combined, and the recognition accuracy is improved through a dual judgment mechanism.
显著提高了缺陷识别的精度和可靠性,增强了对复杂缺陷模式的识别能力,减少噪点影响,提高了检测的准确性。
Smart Images

Figure CN118864466B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image processing, and in particular to a method for intelligently identifying surface defects of nanoimprint wafers. Background Art
[0002] In modern semiconductor manufacturing, nanoimprint technology has been widely used due to its good molding ability and high resolution. However, defect detection on the wafer surface is still a key link in the manufacturing process. Since tiny defects on the wafer surface may cause serious degradation of circuit performance, it is particularly important to develop effective defect identification methods. Traditional defect inspection usually relies on manual visual inspection or simple image processing technology, which is inefficient and easily affected by subjective factors. Therefore, it is crucial to adopt intelligent and automated defect identification technology.
[0003] The existing wafer surface defect recognition method uses a high-resolution camera to obtain images of the wafer surface, uses edge detection, texture analysis and other technologies to extract image features, and then classifies the features through machine learning or deep learning algorithms to identify potential defects. There are many types of wafer surface defects with complex shapes, and the defects are different on each wafer, so there is a problem of low wafer defect recognition accuracy. Summary of the invention
[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides: a method for intelligently identifying surface defects of nanoimprint wafers, which solves the problem of low wafer defect identification accuracy in the prior art.
[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for intelligently identifying surface defects of nanoimprint wafers, comprising the following steps:
[0006] S1, filtering the wafer surface image to obtain a filtered image;
[0007] S2, grayscale processing is performed on the filtered image to obtain a grayscale image;
[0008] S3, counting the number of occurrences of each first gray value combination, second gray value combination and third gray value combination on the grayscale image, wherein the first gray value combination, the second gray value combination and the third gray value combination are A combination of two grayscale values within a range;
[0009] S4, obtaining a combined deviation degree of the local area according to the number of occurrences of each first gray value combination, second gray value combination and third gray value combination;
[0010] S5, obtaining the grayscale deviation of the local area according to the difference in grayscale value between the grayscale image and the local area;
[0011] S6. Add the combined deviation and grayscale deviation of the same local area to obtain a defect degree value. When the defect degree value is greater than the defect degree threshold, there is a defect in the area.
[0012] Furthermore, the S1 comprises the following sub-steps:
[0013] S11. On the wafer surface image, within the neighborhood of each pixel point, calculate the difference between the pixel value of each pixel point and the pixel value of each other pixel point within the neighborhood range, and take the absolute value of the difference;
[0014] S12, when the absolute value of each difference in the neighborhood is greater than the difference threshold, the pixel is a noise point;
[0015] S13, weighting the pixel value of each other pixel point within the neighborhood of the noise point to obtain a new pixel value of the noise point;
[0016] S14, obtaining a filtered image according to the new pixel value of each noise point and the pixel value of the unprocessed pixel point.
[0017] Furthermore, S3 includes the following sub-steps:
[0018] S31, select pixel point (x, y), pixel point (x+1, y), pixel point (x, y+1) and pixel point (x+1, y+1) on the grayscale image, where x is the horizontal coordinate of the pixel point and y is the vertical coordinate of the pixel point;
[0019] S32, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x+1, y) to obtain a first grayscale value combination;
[0020] S33, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x, y+1) to obtain a second grayscale value combination;
[0021] S34, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x+1, y+1) to obtain a third grayscale value combination;
[0022] S35, Settings A sliding window of size slides on the grayscale image, and each sliding length is 1 pixel;
[0023] S36, counting the number of occurrences of each first gray value combination, second gray value combination, and third gray value combination after the grayscale image slides.
[0024] Furthermore, the S4 comprises the following sub-steps:
[0025] S41, cut the grayscale image to obtain image area;
[0026] S42, calculating a first deviation according to the number of times the first gray value combination appears in the image area and the number of times the first gray value combination appears in the grayscale image;
[0027] S43, calculating a second deviation according to the number of times the second gray value combination appears in the image area and the number of times the second gray value combination appears in the grayscale image;
[0028] S44, calculating a third deviation according to the number of times the third gray value combination appears in the image area and the number of times the third gray value combination appears in the grayscale image;
[0029] S45. Add the first deviation, the second deviation and the third deviation to obtain a combined deviation.
[0030] Furthermore, S42, S43 and S44 each include the following sub-steps:
[0031] A1, taking the ratio of the number of times the gray value combination appears in the image area to the number of times the gray value combination appears in the grayscale image as a proportional coefficient;
[0032] A2, normalizing the number of times the gray value combination appears in the gray image to obtain a compensation coefficient;
[0033] A3. Multiply the compensation coefficient by the proportional coefficient to obtain the degree of deviation.
[0034] Furthermore, the formula for calculating the degree of deviation in S42, S43 and S44 is: , where ε is the deviation, F H,i is the number of times the i-th first gray value combination, the second gray value combination or the third gray value combination appears in the image area, F W,i is the number of times the i-th first gray value combination, second gray value combination or third gray value combination in the image area appears in the grayscale image, i is a positive integer, and K is the number of types of the first gray value combination, second gray value combination or third gray value combination in the image area.
[0035] Furthermore, the S5 comprises the following sub-steps:
[0036] S51, cut the grayscale image to obtain image area;
[0037] S52, counting the number of times each gray value appears in the image region and the number of times each gray value appears in the grayscale map in each image region, and calculating the grayscale deviation coefficient;
[0038] S53, calculating the grayscale difference according to the difference between the grayscale value of the image area and the grayscale image;
[0039] S54, multiplying the grayscale deviation coefficient and the grayscale difference to obtain the grayscale deviation of the local area.
[0040] Furthermore, the formula for calculating the grayscale deviation coefficient in S52 is: , where γ is the grayscale deviation coefficient, F H,n is the number of times the nth gray value appears in the image area, F W,n is the number of times the nth grayscale value appears in the grayscale image in the image area, n is a positive integer, and L is the number of grayscale values in the image area.
[0041] Furthermore, the formula for calculating the grayscale difference in S53 is: , where d is the grayscale difference, G H,k is the gray value of the kth pixel in the image area, is the grayscale mean on the grayscale image, k is a positive integer, M is the number of pixels in the image area, and tanh is the hyperbolic tangent function.
[0042] The beneficial effects of the present invention are as follows: the present invention reduces the influence of noise through filtering processing. The grayscale values are combined in pairs within the range to construct three grayscale value combinations. The combined deviation of the local area is determined according to the number of occurrences of the three grayscale value combinations. The grayscale deviation of the local area is obtained by taking into account the difference in grayscale value between the grayscale image and the local area. This achieves the simultaneous consideration of the distribution of a single grayscale value and the distribution of the three grayscale value combinations, greatly enhancing the ability to recognize complex defect patterns and improving the accuracy of defect recognition. A dual judgment mechanism is established by simultaneously calculating the combined deviation and the grayscale deviation, which significantly improves the accuracy and reliability of defect recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 The present invention is a flow chart of an intelligent method for identifying surface defects of nanoimprint wafers. DETAILED DESCRIPTION
[0044] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0045] like Figure 1 As shown, a method for intelligently identifying surface defects of a nanoimprint wafer comprises the following steps:
[0046] S1, filtering the wafer surface image to obtain a filtered image;
[0047] S2, grayscale processing is performed on the filtered image to obtain a grayscale image;
[0048] S3, counting the number of occurrences of each first gray value combination, second gray value combination and third gray value combination on the grayscale image, wherein the first gray value combination, the second gray value combination and the third gray value combination are A combination of two grayscale values within a range;
[0049] S4, obtaining a combined deviation degree of the local area according to the number of occurrences of each first gray value combination, second gray value combination and third gray value combination;
[0050] S5, obtaining the grayscale deviation of the local area according to the difference in grayscale value between the grayscale image and the local area;
[0051] S6. Add the combined deviation and grayscale deviation of the same local area to obtain a defect degree value. When the defect degree value is greater than a defect degree threshold, there is a defect in the area.
[0052] In this embodiment, the defect level threshold is a threshold set for the defect level value.
[0053] The S1 comprises the following sub-steps:
[0054] S11. On the wafer surface image, within the neighborhood of each pixel point, calculate the difference between the pixel value of each pixel point and the pixel value of each other pixel point within the neighborhood range, and take the absolute value of the difference;
[0055] S12, when the absolute value of each difference in the neighborhood is greater than the difference threshold, the pixel is a noise point;
[0056] S13, weighting the pixel value of each other pixel point within the neighborhood of the noise point to obtain a new pixel value of the noise point;
[0057] S14, obtaining a filtered image according to the new pixel value of each noise point and the pixel value of the unprocessed pixel point.
[0058] The present invention effectively identifies abnormal pixels by calculating the difference between the pixel value at the center and other pixel values within the neighborhood of each pixel. In order to find defects as much as possible, the present invention calculates the difference between the pixel value at the center and other pixel values in the neighborhood. When the absolute values of the eight differences are all greater than the difference threshold, the pixel is a noise point, avoiding classifying the pixel at the defect as a noise point.
[0059] In this embodiment, the difference threshold is a threshold set for the absolute value of the difference.
[0060] In this embodiment, the weighting formula of step S13 is: , where P is the new pixel value of the noise point, P m is the pixel value of the mth pixel in the neighborhood, where m is a positive integer.
[0061] The S3 comprises the following sub-steps:
[0062] S31, select pixel point (x, y), pixel point (x+1, y), pixel point (x, y+1) and pixel point (x+1, y+1) on the grayscale image, where x is the horizontal coordinate of the pixel point and y is the vertical coordinate of the pixel point;
[0063] S32, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x+1, y) to obtain a first grayscale value combination;
[0064] S33, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x, y+1) to obtain a second grayscale value combination;
[0065] S34, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x+1, y+1) to obtain a third grayscale value combination;
[0066] S35, Settings A sliding window of size slides on the grayscale image, and each sliding length is 1 pixel;
[0067] S36, counting the number of occurrences of each first gray value combination, second gray value combination, and third gray value combination after the grayscale image slides.
[0068] In the present invention, it is equivalent to setting A sliding window of size is used to traverse the grayscale image horizontally to find the number of times each first grayscale value combination occurs horizontally, to traverse the grayscale image vertically to find the number of times each second grayscale value combination occurs vertically, and to traverse the grayscale image diagonally to find the number of times each third grayscale value combination occurs on the diagonal.
[0069] The present invention constructs three different grayscale value combinations by selecting pixels on the grayscale image and their neighboring pixels, effectively capturing the characteristics of the local area. A 2*2 sliding window is set and slid on the grayscale image to achieve a comprehensive scan of the entire image, which not only enhances the sensitivity to tiny defects, but also improves the accuracy of defect recognition by counting the number of occurrences of different grayscale value combinations. In particular, by traversing in the horizontal, vertical and diagonal directions, the distribution characteristics of the first, second and third grayscale value combinations are comprehensively analyzed.
[0070] The S4 comprises the following sub-steps:
[0071] S41, cut the grayscale image to obtain image area;
[0072] S42, calculating a first deviation according to the number of times the first gray value combination appears in the image area and the number of times the first gray value combination appears in the grayscale image;
[0073] S43, calculating a second deviation according to the number of times the second gray value combination appears in the image area and the number of times the second gray value combination appears in the grayscale image;
[0074] S44, calculating a third deviation according to the number of times the third gray value combination appears in the image area and the number of times the third gray value combination appears in the grayscale image;
[0075] S45. Add the first deviation, the second deviation and the third deviation to obtain a combined deviation.
[0076] The present invention obtains the grayscale image by cutting Next, according to different gray value combinations, the number of occurrences in the image area is counted, and the deviations of the first, second, and third gray value combinations are calculated to reflect the gap between the image area and the grayscale image. By combining various deviations, a more comprehensive defect recognition basis is formed to improve the accuracy of detection.
[0077] In this embodiment, the grayscale image is sheared N-1 times horizontally and N-1 times vertically to obtain image areas. In this embodiment, N is greater than 10.
[0078] S42, S43 and S44 all include the following sub-steps:
[0079] A1, taking the ratio of the number of times the gray value combination appears in the image area to the number of times the gray value combination appears in the grayscale image as a proportional coefficient;
[0080] A2, normalizing the number of times the gray value combination appears in the gray image to obtain a compensation coefficient;
[0081] A3. Multiply the compensation coefficient by the proportional coefficient to obtain the degree of deviation.
[0082] In this embodiment, the formula for calculating the deviation in S42, S43 and S44 is: , where ε is the deviation, F H,i is the number of times the i-th first gray value combination, the second gray value combination or the third gray value combination appears in the image area, F W,iis the number of times the i-th first gray value combination, second gray value combination or third gray value combination in the image area appears in the grayscale image, i is a positive integer, and K is the number of types of the first gray value combination, second gray value combination or third gray value combination in the image area.
[0083] In the present invention, the same gray value combination is classified into one category.
[0084] When the number of gray value combinations in the image area is the same as the number of gray value combinations in the grayscale image, F W,i =F H,i , then the scale factor is 1, which is the largest scale factor, and the image area is part of the grayscale image. Therefore, F W,i Greater than or equal to F H,i , the smaller the scale factor, the more times the gray value combination appears on the grayscale image, so the possibility that the gray value combination is the gray value combination of the defect is lower. W,i =F H,i When , this gray value combination only appears in the image area. Therefore, the gray value combination is most likely to be in the defect area. At the same time, the compensation coefficient is set to compensate the proportional coefficient through the compensation coefficient to further improve the judgment accuracy of the gray value combination at the defect. The compensation coefficient is related to the number of times the gray value combination appears in the grayscale image. The more times the gray value combination appears in the grayscale image, the lower the possibility of the gray value combination in the defect area. The fewer times the gray value combination appears in the grayscale image, the higher the possibility of the gray value combination in the defect area.
[0085] The S5 comprises the following sub-steps:
[0086] S51, cut the grayscale image to obtain image area;
[0087] S52, counting the number of times each gray value appears in the image region and the number of times each gray value appears in the grayscale map in each image region, and calculating the grayscale deviation coefficient;
[0088] S53, calculating the grayscale difference according to the difference between the grayscale value of the image area and the grayscale image;
[0089] S54, multiplying the grayscale deviation coefficient and the grayscale difference to obtain the grayscale deviation of the local area.
[0090] In the present invention, the shearing methods of S51 and S41 are the same.
[0091] The formula for calculating the grayscale deviation coefficient in S52 is: , where γ is the grayscale deviation coefficient, F H,n is the number of times the nth gray value appears in the image area, FW,n is the number of times the nth grayscale value appears in the grayscale image in the image area, n is a positive integer, and L is the number of grayscale values in the image area.
[0092] In the present invention, the same grayscale value is classified into one category.
[0093] The formula for calculating the grayscale difference in S53 is: , where d is the grayscale difference, G H,k is the gray value of the kth pixel in the image area, is the grayscale mean on the grayscale image, k is a positive integer, M is the number of pixels in the image area, and tanh is the hyperbolic tangent function.
[0094] When the number of times a gray value appears in the image area is the same as the number of times the gray value appears in the grayscale image, F W,n =F H,n , then the grayscale deviation coefficient is 1, which is the largest grayscale deviation coefficient. The image area is part of the grayscale image. Therefore, F W,n Greater than or equal to F W,n , the smaller the scale factor, the higher the number of times the gray value appears on the grayscale image, so the possibility that the gray value is the gray value of the defect is lower. W,n =F H,n When , the gray value only appears in the image area, so the gray value is most likely to be in the defect area. Then, based on the difference between the gray value in the image area and the gray value on the grayscale map, the gray difference is calculated. When the gray difference is larger, the gray value in the image area deviates more seriously from the overall gray mean, and the image area is more likely to be a defect area.
[0095] In this embodiment, the defect level threshold and the difference threshold are specifically set based on experiments or experience.
[0096] The present invention reduces the influence of noise by filtering. Grayscale values are combined in pairs within a certain range to construct three grayscale value combinations. The combined deviation of the local area is determined according to the number of occurrences of the three grayscale value combinations. The grayscale deviation of the local area is obtained according to the difference in grayscale value between the grayscale image and the local area. The distribution of the single grayscale value and the distribution of the three grayscale value combinations are considered at the same time, which greatly enhances the recognition ability of complex defect patterns and improves the defect recognition accuracy. The combined deviation and grayscale deviation of the same local area are added to obtain the defect degree value. When the defect degree value is greater than the defect degree threshold, there is a defect in the area. The two deviation characteristics of the local area are considered at the same time, which significantly improves the accuracy and reliability of defect recognition.
[0097] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for intelligently identifying surface defects of nanoimprint wafers, characterized in that: The following steps are involved: S1, filtering the wafer surface image to obtain a filtered image; S2, grayscale processing is performed on the filtered image to obtain a grayscale image; S3, counting the number of occurrences of each first gray value combination, second gray value combination and third gray value combination on the grayscale image, wherein the first gray value combination, the second gray value combination and the third gray value combination are A combination of two grayscale values within a range; S4, obtaining a combined deviation degree of the local area according to the number of occurrences of each first gray value combination, second gray value combination and third gray value combination; S5. Obtain the grayscale deviation of the local area according to the number of times the grayscale value appears on the grayscale image; S6. Add the combined deviation and grayscale deviation of the same local area to obtain a defect degree value. When the defect degree value is greater than the defect degree threshold, there is a defect in the area; The S4 comprises the following sub-steps: S41, cut the grayscale image to obtain image area; S42, calculating a first deviation according to the number of times the first gray value combination appears in the image area and the number of times the first gray value combination appears in the grayscale image; S43, calculating a second deviation according to the number of times the second gray value combination appears in the image area and the number of times the second gray value combination appears in the grayscale image; S44, calculating a third deviation according to the number of times the third gray value combination appears in the image area and the number of times the third gray value combination appears in the grayscale image; S45, adding the first deviation degree, the second deviation degree and the third deviation degree to obtain a combined deviation degree; S42, S43 and S44 all include the following sub-steps: A1, taking the ratio of the number of times the gray value combination appears in the image area to the number of times the gray value combination appears in the grayscale image as a proportional coefficient; A2, normalizing the number of times the gray value combination appears in the gray image to obtain a compensation coefficient; A3. Multiply the compensation coefficient by the proportional coefficient to obtain the degree of deviation; The S5 comprises the following sub-steps: S51, cut the grayscale image to obtain image area; S52, counting the number of times each gray value appears in the image region and the number of times each gray value appears in the grayscale map in each image region, and calculating the grayscale deviation coefficient; S53, calculating the grayscale difference according to the difference between the grayscale value of the image area and the grayscale image; S54, multiplying the grayscale deviation coefficient and the grayscale difference to obtain the grayscale deviation of the local area; The formula for calculating the grayscale deviation coefficient in S52 is: , where γ is the grayscale deviation coefficient, F H,n is the number of times the nth gray value appears in the image area, F W,n is the number of times the nth grayscale value appears in the grayscale image in the image area, n is a positive integer, and L is the number of grayscale values in the image area.
2. The method for intelligently identifying surface defects of nanoimprint wafers according to claim 1, characterized in that: The S1 comprises the following sub-steps: S11. On the wafer surface image, within the neighborhood of each pixel point, calculate the difference between the pixel value of each pixel point and the pixel value of each other pixel point within the neighborhood range, and take the absolute value of the difference; S12, when the absolute value of each difference in the neighborhood is greater than the difference threshold, the pixel is a noise point; S13, weighting the pixel value of each other pixel point within the neighborhood of the noise point to obtain a new pixel value of the noise point; S14, obtaining a filtered image according to the new pixel value of each noise point and the pixel value of the unprocessed pixel point.
3. The method for intelligently identifying surface defects of nanoimprint wafers according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31, select pixel point (x, y), pixel point (x+1, y), pixel point (x, y+1) and pixel point (x+1, y+1) on the grayscale image, where x is the horizontal coordinate of the pixel point and y is the vertical coordinate of the pixel point; S32, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x+1, y) to obtain a first grayscale value combination; S33, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x, y+1) to obtain a second grayscale value combination; S34, combining the grayscale value of the pixel point (x, y) with the grayscale value of the pixel point (x+1, y+1) to obtain a third grayscale value combination; S35, Settings A sliding window of size slides on the grayscale image, and each sliding length is 1 pixel; S36, counting the number of occurrences of each first gray value combination, second gray value combination, and third gray value combination after the grayscale image slides.
4. The method for intelligently identifying surface defects of nanoimprint wafers according to claim 1, characterized in that: The formula for calculating the deviation in S42, S43 and S44 is: , where ε is the deviation, F H,i is the number of times the i-th first gray value combination, the second gray value combination or the third gray value combination appears in the image area, F W,i is the number of times the i-th first gray value combination, second gray value combination or third gray value combination in the image area appears in the grayscale image, i is a positive integer, and K is the number of types of the first gray value combination, second gray value combination or third gray value combination in the image area.
5. The method for intelligently identifying surface defects of nanoimprint wafers according to claim 1, characterized in that: The formula for calculating the grayscale difference in S53 is: , where d is the grayscale difference, G H,k is the gray value of the kth pixel in the image area, is the grayscale mean on the grayscale image, k is a positive integer, M is the number of pixels in the image area, and tanh is the hyperbolic tangent function.
Citation Information
Patent Citations
Method and apparatus for detecting defects, and computer readable storage medium
US20190228515A1
Method for selecting a light source for illuminating defects, electronic device, and non-transitory storage medium
US20220198634A1
