Graphic wafer defect detection method based on adaptive local frequency domain filtering
Through adaptive local frequency domain filtering and improved peak-center of mass method, the problem of defect extraction in the defect detection of graphics wafers in the prior art is solved, efficient detection and positioning of complex graphics wafers is achieved, and the detection is improved. Robustness and applicability.
Patent Information
- Application Number
- CN202410894267.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-04
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-04
AI Technical Summary
In the detection of graphics wafer defects, post-processing algorithms are mostly simple spatial domain processing or common frequency domain processing, making it difficult to effectively extract defects of complex graphics wafers. In addition, artificial intelligence algorithms have high demand for computing resources, and The detection of complex graphics defects is not applicable.
The graphical wafer defect detection method based on adaptive local frequency domain filtering is adopted, and frequency domain filtering is performed by composite second-order differential and normalized frequency filtering factors, focusing on the low-frequency part of the defect, suppressing noise, and combining with improved peak- Center of mass method for defect positioning.
The defect contrast and peak signal-to-noise ratio of defect images are improved, and the precise detection and positioning of surface defects of different types of graphic wafers is achieved, which reduces the computing resource requirements and improves the robustness and applicability of detection.
Smart Images

Figure CN118918059A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of defect detection, and in particular relates to a graphic wafer defect detection method based on adaptive local frequency domain filtering. Background Art
[0002] Semiconductor testing equipment runs through the entire process of hundreds of processes in semiconductor chip production and is an effective means to ensure IC manufacturing quality and yield. Wafer defects include random defects mainly caused by surface contaminants and systematic defects caused by the manufacturing process. The introduction of systematic defects will greatly reduce the yield and increase manufacturing costs.
[0003] Current semiconductor defect detection mainly includes scanning electron microscopy (SEM), transmission electron microscopy (TEM), atomic force microscopy (AFM) and optical detection. For non-destructive and high-precision global wafer defect detection, optical detection is the most important detection method. Optical detection is based on optical principles and often requires a series of image post-processing to accurately locate defects. Defect detection for graphic wafers is more complex and challenging, and post-processing algorithms play a vital role in improving the signal-to-noise ratio and contrast of defects.
[0004] In 2021, Chen Shiwei from Zhejiang University performed homomorphic filtering on images to improve the impact of uneven lighting, obtained standard control images through a threshold iteration method based on statistics, and obtained defects through differentiation (Chen Shiwei. Research on multi-scanning graphical wafer inspection technology based on bright and dark field imaging [D]. Master's thesis of Zhejiang University, 2021).
[0005] Chinese patent CN201210585636.8 proposes a Fourier filter with multiple switchable blocking units, which can select the direction of blocking filtering by switching different filters to complete wafer defect detection on graphics in different directions.
[0006] Chinese patent CN202210988249.2 proposes a period extraction method for a periodic structure and a wafer defect detection method, which performs peak search on the sampling spectrum twice to obtain the precise value of the period of the periodic structure, and performs defect detection and defect classification on the image to be inspected by obtaining the differential image of the median image of the target image and the adjacent image.
[0007] Chinese patent CN202310527789.5 inputs the image to be inspected into a preset defect detection model trained based on the unsupervised algorithm PaDim to obtain a detection result characterizing whether the wafer has defects, and segments an image representing the defect area from the camera image through a preset image segmentation strategy.
[0008] Most of the existing post-processing algorithms are simple spatial domain processing (such as Laplacian operator image enhancement, mean, Gaussian, bilateral filtering, etc.) and common frequency domain processing (low-pass, high-pass filters, etc.), which are not applicable to complex graphic wafer defect detection and it is difficult to complete defect extraction through a single method. Artificial intelligence algorithms rely on a large amount of labeled data and parameter debugging, and at the same time place high demands on computing resources. Summary of the invention
[0009] In order to overcome the shortcomings of the above-mentioned prior art, the present invention proposes a graphic wafer defect detection method based on adaptive local frequency domain filtering, which effectively improves the defect contrast and peak signal-to-noise ratio of the defect image. Combined with the improved peak-centroid method proposed in the present invention to extract defects, different types of graphic wafer surface defects can be accurately detected and located.
[0010] In order to achieve the above object, the technical solution adopted by the present invention is:
[0011] A method for detecting patterned wafer defects based on adaptive local frequency domain filtering comprises the following steps:
[0012] Step 1: preprocessing the collected wafer image to obtain image I;
[0013] Step 2: Perform the second-order difference in the column direction on image I to obtain image I dx ;
[0014] Step 3: Image I dx Perform the second-order difference in the row direction to obtain image I dxy , and cut off the area where the complete row and column second-order difference is not performed, and get image I C ;
[0015] Step 4: Image I C Perform local frequency domain analysis to obtain the normalized frequency filter factor S i,j ;
[0016] Step 5: Apply the normalized frequency filter factor to perform frequency domain filtering to obtain image I FC ;
[0017] Step 6: Image I based on statistics FC Selecting an abnormal area, wherein the abnormal area is determined to have a defect;
[0018] Step 7: Locate the defect.
[0019] Compared with traditional spatial domain filtering methods and frequency domain filtering methods, the present invention proposes an adaptive local frequency domain filtering defect detection method for graphic wafers. Composite second-order differences are performed according to the periodicity of the graphic wafer, focusing on the low-frequency part where the defect is located, and frequency domain filtering is performed according to the corresponding frequency filtering factor. Noise is effectively suppressed without losing defect characteristics, thereby improving defect contrast.
[0020] At the same time, the present invention also proposes an improved peak-centroid method based on the selected abnormal area to achieve accurate positioning of defects.
[0021] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 The present invention is a flow chart of a method for detecting defects in a graphic wafer based on adaptive local frequency domain filtering.
[0023] Figure 2 The original image is collected based on a dark field microscope according to one embodiment of the present invention.
[0024] Figure 3 This is an image after the original image is rotated according to an embodiment of the present invention.
[0025] Figure 4 This is a region of interest image provided according to an embodiment of the present invention.
[0026] Figure 5 This is an image after second-order difference of columns provided by an embodiment of the present invention.
[0027] Figure 6 A composite second-order difference image is provided according to an embodiment of the present invention.
[0028] Figure 7 This is an image obtained by removing the boundary of the composite second-order difference image provided by one embodiment of the present invention.
[0029] Figure 8 A normalized frequency filter factor graph is provided according to an embodiment of the present invention.
[0030] Fig. 9 A frequency domain filtered image based on a normalized frequency filter factor is provided according to an embodiment of the present invention.
[0031] Fig.10 This is a Gaussian filtered image provided according to an embodiment of the present invention.
[0032] Fig.11 This is a frequency domain low-pass filtered image provided by the present invention according to an embodiment.
[0033] Fig.12 This is an abnormal area determination diagram provided according to an embodiment of the present invention.
[0034] Fig.13 This is a diagram of abnormal regions in a filtered image provided according to an embodiment of the present invention.
[0035] Fig.14 A filter image defect localization map provided according to an embodiment of the present invention.
[0036] Fig.15 This is an original image defect location map provided according to an embodiment of the present invention. DETAILED DESCRIPTION
[0037] The embodiments of the present invention are described in detail below with reference to the accompanying drawings and defect detection examples.
[0038] refer to Figure 1 As shown, the method for detecting defects in a patterned wafer based on adaptive local frequency domain filtering of the present invention mainly comprises the following steps:
[0039] Step 1: pre-process the collected wafer image to obtain image I.
[0040] The present invention is applicable to wafer images collected by bright field and dark field optical microscopes.
[0041] In the embodiment of the present invention, a dark field microscope is used to collect images. Figure 2 As shown, the test sample is a 55nm line width periodic pattern, and there are three defects in the periodic sample: a 55×300nm horizontal bridge, a 55nm diameter cylinder, and a 55×320nm vertical bridge, simulating a graphic wafer, in which one of the full-period graphics simulates a Cell.
[0042] The preprocessing methods of this step mainly include wafer posture adjustment and region selection. Among them, posture adjustment is to rotate the image according to the rotation angle α, where α = atan [(y2-y1) / (x2-x1)], (x1, y1) and (x2, y2) are any two feature points in the image. Region selection is to select the wafer area in the image and exclude the area outside the wafer. The size of the obtained image I is M×N, a unit area on the wafer is called a Cell, and the size of each Cell is M c ×N c , then image I contains A complete unit area.
[0043] for Figure 2The image shown is rotated to the design direction, and two points (600, 1430) and (1419, 1410) are taken to calculate the rotation angle, α = atan [(1410-1430) / (1419-600)] = -1.398°. The rotated image can be seen in Figure 3 . Cut the region of interest (477:1764, 1407:2694) to get image I, see Figure 4 .
[0044] Step 2: Perform the second-order difference in the column direction on image I to obtain image I dx .
[0045] In this step, the size of the cell in the column direction, i.e., M c is the spacing, and the second-order difference of image I in the column direction is obtained to obtain image I dx , the implementation method is as follows:
[0046] I dx (x, y) = I(xM c , y)-2I(x, y)+I(x+M c , y)
[0047] Among them, I dx (x, y) represents image I dx The gray value at the pixel coordinate (x, y), M c is the size of a cell area on the wafer in the column direction, I(x, y), I(x+M c ,y)、I(xM c , y) represent the pixel coordinates (x, y) and (x+M c ,y)、(xM c , grayscale value at y).
[0048] In the embodiment of the present invention Figure 4 In the image shown, the size of each Cell is M c ×N c 56×28 pixels, M=1288 pixels, N=1288 pixels, M / M c =23,N / N c =46, then there are 23×46 cells in the image I, and the second-order difference is performed with the size of 56 in the column direction as the spacing. At this time, the cells in the left and right columns of the image cannot be second-order differentiated, and the obtained image I dx See also Figure 5 .
[0049] Step 3: Image I dx Perform the second-order difference in the row direction to obtain image I dxy .
[0050] In this step, the size of the cell in the row direction, that is, N c is the spacing, for image I dx Perform the second-order difference in the row direction to obtain image I dxy , the implementation method is as follows:
[0051] I dxy (x, y) = I dx (x,yN c )-2I dx (x,y)+I dx (x,y+N c )
[0052] Among them, I dxy (x, y) represents image I dxy Gray value at pixel coordinate (x, y), N c is the size of the Cell in the row direction, I dx (x,yN c ), I dx (x,y+N c ) represent images I dx Pixel coordinates (x, yN c )、(x,y+N c ) is the gray value at .
[0053] In the embodiment of the present invention Figure 5 The image shown in image I dx Based on the cell row size 28 as the spacing, the second-order difference is performed. At this time, the upper and lower rows of cells in the image cannot be second-order differentiated, and the obtained image I dxy See also Figure 6 .
[0054] The present invention removes background components by composite second-order difference. This composite second-order difference in the column direction and then in the row direction will amplify the changes in the image and enhance the defect characteristics in two dimensions, making the defects in the image more obvious. dxy The boundary area of the image where the composite second-order difference is not completed, specifically for image I dxy The area where the complete composite second-order difference process cannot be performed due to the unit being at the boundary on the wafer is cut off to obtain image I C , now image I C Divided into (M / Mc-2)×(N / N c -2) area, embodiment 1 of the present invention C There are 21×44 complete cycles in Figure 7 .
[0055] In the present invention, the order of difference in step 2 and step 3 can be interchanged, that is, the second-order difference in the row direction can be performed first, and then the second-order difference in the column direction can be performed on the image after the second-order difference in the row direction, and the final result is the same or similar, and the processing effect is not changed. However, it should be noted that the difference in step 3 must be based on the difference image completed in the previous step, which is a composite row-column second-order difference.
[0056] Step 4: Image I C Perform local frequency domain analysis to obtain the normalized frequency filter factor S i,j .
[0057] In this step, the image I to be analyzed in the local frequency domain C Divided into unit area, In other words, the image after the composite second-order difference of each Cell is described as dCell i,j . Perform frequency domain analysis on each unit area, including:
[0058] a. For the cell area dCell in the i-th column and j-th row i,j Perform a two-dimensional discrete Fourier transform (DFT) to convert the signal in the spatial domain into the frequency domain, that is:
[0059]
[0060] Among them, F i,j (u, v) represents the spectrum at the frequency index (u, v), dCell i,j (m, n) represents dCell i,j The value at the spatial domain index (m, n);
[0061] b. Calculate the amplitude spectrum, amplitude | FFTshift i , (u, v)| is obtained by taking the absolute value of the complex number, as follows:
[0062] Mag i,j (u, v) = |F i,j (u, v)|
[0063] Among them, Mag i,j (u, v) is the amplitude spectrum at the frequency index (u, v), that is, the intensity at this frequency, without considering the phase information, which is used to analyze the energy distribution of the spectrum;
[0064] c. Mag i,j , extract its center The sum of the values in the frequency region SMag i,j ,Right now:
[0065]
[0066] Among them, round means rounding the number to the nearest integer;
[0067] d. SMag i,j Perform normalization processing to obtain the normalized frequency filter factor S i,j ,Right now:
[0068]
[0069] Where SMag is SMag i,j A set of SMags, min(SMag) and max(SMag) are the minimum and maximum values in the set SMag, respectively.
[0070] In the embodiment of the present invention, for Figure 7 The image shown, 1≤i≤21, 1≤j≤44, extracts Mag in step c i,j The total value of all frequency components in the central 14×7 area is Smag i,j :
[0071]
[0072] The final normalization process obtains the frequency filter factor S i,j It is expressed as:
[0073]
[0074] Normalized frequency filter factor see Figure 8 .
[0075] It is worth noting that the present invention can also perform defect detection on non-patterned wafers. In this case, there is no need to perform differentiation, and the process directly proceeds to step 4, where a reasonable period is selected for frequency domain analysis and corresponding frequency domain filtering, and then defect extraction is performed on the filtered image.
[0076] Step 5: Apply the normalized frequency filter factor to perform frequency domain filtering to obtain image I FC .
[0077] In this step, the image I C Each cell area dCell i,j In the frequency domain, the corresponding normalized frequency filter factor is applied to perform frequency domain filtering, including:
[0078] a. dCell i,j , applying the corresponding normalized frequency filter factor S in its frequency domain i,j :
[0079] G i,j (u, v) = F i,j (u,v)·S i,j
[0080] Among them, G i,j (u, v) is the spectrum at the frequency index (u, v) after filtering;
[0081] b. Perform a two-dimensional discrete Fourier inverse transform and transform G i,j (u, v) is transformed back to the spatial domain, and obtained by dCell i,j The result after frequency domain filtering is fdCell i,j :
[0082]
[0083] where fdCell i,j (m, n) is the grayscale value of the pixel index (m, n) in the cell area image of the i-th column and j-th row in the spatial domain after frequency domain filtering, represented by fdCell i,j Forming filtered image I FC .
[0084] In the embodiment of the present invention, fdCell i,j (1≤i≤M / Mc-2, 1≤j≤N / N c -2) Forming filtered image I FC , see Fig. 9 .
[0085] In addition, in order to compare the method proposed by the present invention with the commonly used spatial domain and frequency domain filtering methods, Fig.10 and Fig.11 The images after Gaussian filtering (filter size is 3×3, standard deviation is 2) and frequency domain low-pass filtering (cutoff frequency is 300 pixels) are given respectively. It can be seen that compared with other filtering algorithms, the filtering method based on frequency filtering factor proposed in the present invention can effectively eliminate the image background and improve the defect contrast.
[0086] The present invention adopts a frequency domain filtering method based on normalized frequency filter factors. Since the noise component is located at high frequency, only the low-frequency area reflecting the defect component is focused on, and the image is adaptively localized and filtered according to the frequency domain characteristics of different areas. Compared with traditional spatial domain filtering methods such as mean and median filtering and frequency domain filtering methods such as low-pass and high-pass filtering, this method effectively suppresses background noise without losing defect details and improves defect contrast.
[0087] Step 6: Image I based on statistics FC The abnormal area is selected and judged as having defects.
[0088] After obtaining the filtered image I FC After that, first perform the abnormal area kThe selection of the normalized frequency filter factor array S is known, and the normalized frequency filter factor S is calculated i,j The mean
[0089]
[0090] Judge S i,j Is it greater than or equal to Forming the logical array L i,j , thereby defining the abnormal area, namely:
[0091]
[0092] The value of 1 in the logical array corresponds to the value in image I FC In the process, an abnormal region is formed, which reflects the defective cell region.
[0093] In the embodiment of the present invention, Then judge S i,j Is it greater than or equal to That is 0.5092, forming the logical array L i,j , see Fig.12 There are three abnormal areas here, see Fig.13 .
[0094] Step 7: Locate the defect.
[0095] The defect location in this step is in image I FC For each abnormal region in image I, the improved peak-centroid method is used. FC For each abnormal area in the image, perform the following steps to identify and locate the defect:
[0096] a. The kth abnormal area Area k Take the absolute values of the grayscale values of all pixels in the image, sort these absolute values in descending order, select the pixel coordinates corresponding to the first n values in the sorting, and form a coordinate set TP k :
[0097] Sort k =sort(|Area k (:)|,′descend′)
[0098] TP k ={(x, y)||Area k (x,y)|∈Sort k (1:n)}
[0099] Among them, 'descend' means the array is sorted in descending order, kis the sorted array of the absolute values of the grayscale values of pixels in the kth abnormal area, TP k Area is the pixel coordinate set corresponding to the first n values in the array; k (x, y) represents the grayscale value of the pixel coordinate (x, y) in the kth abnormal area;
[0100] b. Calculate the coordinates TP in each abnormal area k The total value of the corresponding numerical value I k :
[0101]
[0102] c. Calculate TP k The weighted average of the elements in , i.e. the defect location in each abnormal area:
[0103]
[0104] in The x and t coordinates of each defect, respectively. k That is, a defect coordinate set Centroid is constructed, and then the defect coordinates are regressed and marked into the rotated image according to the position of the taken region of interest.
[0105] This step adopts the improved peak-centroid method based on the selected abnormal area, which is more applicable and accurate to defect signals with different characteristics and can achieve precise positioning of the defect area and position.
[0106] In the embodiment of the present invention, n=10, that is, the pixel coordinates corresponding to the top 10 values are selected to form a coordinate set TP k The weighted average calculation result is rounded to obtain image I FC The three defect positions (430, 151), (586, 520), (742, 994) are shown in Fig.14 . The coordinates are regressed to the rotated image, and the defect positions are (962, 1585), (1118, 1954), and (1274, 2428), which are marked in the rotated image. Fig.15 .
[0107] The graphic wafer defect detection method of the present invention applies a frequency filter factor to the defect image for frequency domain filtering, effectively improving the defect contrast and peak signal-to-noise ratio of the defect image, and combined with the improved peak-centroid method to extract defects, can accurately detect and locate different types of graphic wafer surface defects.
[0108] In order to evaluate the positioning error of the peak-centroid method defect location, the distance between the defect and the boundary point is calculated, and the known design value is taken as the true value to calculate the relative error.
[0109] The design values of the distances between the three defects and the boundary points are (40μm, 19.8125μm), (50.3μm, 45μm), and (60.1775μm, 80μm), respectively. The distances calculated based on defect positioning are (38.7780μm, 19.6650μm), (49.5420μm, 45.1260μm), and (60.3060μm, 77.8320μm), respectively. The positioning errors are (3.06%, 0.74%), (1.51%, 0.28%), and (0.21%, 2.71%), respectively. The positioning error is less than 5%, which fully demonstrates the effectiveness of the method of the present invention.
[0110] In summary, the present invention provides a method for detecting defects in graphic wafers, which has the following characteristics: a. removing background components by composite second-order difference and enhancing defect characteristics from two dimensions; b. adaptive local frequency domain filtering to suppress background noise and improve defect contrast; c. defect positioning based on the improved peak-centroid method of the selected abnormal area, which is more robust and more applicable and accurate in positioning defects of different types and characteristics.
[0111] This detection method is applicable to images collected by optical detection systems, such as bright field and dark field optical microscopes. In addition, this detection method is also applicable to defect detection of non-patterned wafers. The present invention has a significant effect in improving the contrast and peak signal-to-noise ratio of defect images. It is combined with the improved peak-centroid method to extract defects and effectively detect and locate different types of wafer surface defects. It is an adaptive and robust defect detection method that plays an important role in micro-nano wafer defect detection.
[0112] The specific implementation modes of the present invention are described above in conjunction with the accompanying drawings, but these descriptions cannot be understood as limiting the scope of the present invention. The protection scope of the present invention is defined by the attached claims, and any changes based on the claims of the present invention are within the protection scope of the present invention.
Claims
1. A method for detecting pattern wafer defects based on adaptive local frequency domain filtering, characterized in that: The following steps are involved: Step 1: preprocessing the collected wafer image to obtain image I; Step 2: Perform the second-order difference in the column direction on image I to obtain image I dx ; Step 3: Image I dx Perform the second-order difference in the row direction to obtain image I dxy , and cut off the area where the complete row and column second-order difference is not performed, and get image I C ; Step 4: Image I C Perform local frequency domain analysis to obtain the normalized frequency filter factor S i,j ; Step 5: Apply the normalized frequency filter factor to perform frequency domain filtering to obtain image I FC ; Step 6: Image I based on statistics FC Selecting an abnormal area, wherein the abnormal area is determined to have a defect; Step 7: Locate the defect.
2. The method for detecting pattern wafer defects based on adaptive local frequency domain filtering according to claim 1, characterized in that: In the step one, the preprocessing includes wafer posture adjustment and area selection; the posture adjustment is to rotate the image according to the rotation angle α, where α=atan[(y2-y1) / (x2-x1)], (x1, y1) and (x2, y2) are any two feature points in the image; the area selection is to select the wafer area in the image and exclude the area outside the wafer.
3. The method for pattern wafer defect detection based on adaptive local frequency domain filtering according to claim 1, characterized in that: In the step 2, M c is the spacing, and the second-order difference of image I in the column direction is obtained to obtain image I dx , the implementation method is as follows: I dx (x,y)=I(x-M c ,y)-2I(x,y)+I(x+M c ,y) Among them, I dx (x,y) represents image I dx The gray value at the pixel coordinate (x, y), M c is the size of a cell area on the wafer in the column direction, I(x,y), I(x+M c ,y)、I(xM c , y) represent the pixel coordinates (x, y) and (x+M c ,y)、(xM c ,y) is the gray value at that location.
4. The method for detecting pattern wafer defects based on adaptive local frequency domain filtering according to claim 3 is characterized in that: In the step 3, N c is the spacing, for image I dx Perform the second-order difference in the row direction to obtain image I dxy , the implementation method is as follows: AND dxy (x,y)=I dx (x,yN c )-2I dx (x,y)+I dx (x,y+N c ) Among them, I dxy (x,y) represents image I dxy Gray value at pixel coordinate (x, y), N c is the size of the Cell in the row direction, I dx (x,yN c ), I dx (x,y+N c ) represent images I dx The pixel coordinates (x,yN c )、(x,y+N c ) is the gray value at .
5. The method for pattern wafer defect detection based on adaptive local frequency domain filtering according to claim 4 is characterized in that: In the step 4, the image I to be analyzed in the local frequency domain C Divided into unit area, Represents, where M×N is the size of image I; Frequency domain analysis is performed on each unit area, including: a. For the cell area dCell in the i-th column and j-th row i,j Perform a two-dimensional discrete Fourier transform to convert the signal in the spatial domain into the frequency domain, that is: Among them, F i,j (u,v) represents the spectrum at the frequency index (u,v), dCell i,j (m,n) represents dCell i,j The value at the spatial domain index (m,n); b. Calculate the amplitude spectrum as follows: Mag i,j (u,v)=|F i,j (u,v)| Among them, Mag i,j (u,v) is the amplitude spectrum at the frequency index (u,v), that is, the intensity at that frequency, without considering the phase information; c. Mag i,j , extract its center The sum of the values in the frequency region SMag i,j ,Right now: Among them, round means rounding the number to the nearest integer; d. SMag i,j Perform normalization processing to obtain the normalized frequency filter factor S i,j ,Right now: Where SMag is SMag i,j A set of SMags, min(SMag) and max(SMag) are the minimum and maximum values in the set SMag, respectively.
6. The method for pattern wafer defect detection based on adaptive local frequency domain filtering according to claim 5, characterized in that: In the step 5, the image I C Each cell area dCell i,j In the frequency domain, the corresponding normalized frequency filter factor is applied to perform frequency domain filtering, including: a. dCell i,j , applying the corresponding normalized frequency filter factor S in its frequency domain i,j : G i,j (u,v)=F i,j (u,v)·S i,j Among them, G i,j (u,v) is the spectrum at the frequency index (u,v) after filtering; b. Perform a two-dimensional discrete Fourier inverse transform and transform G i,j (u,v) is transformed back to the spatial domain, and obtained by dCell i,j The result after frequency domain filtering is fdCell i,j : where fdCell i,j (m,n) is the grayscale value of the pixel index (m,n) in the cell area image of the i-th column and j-th row in the spatial domain after frequency domain filtering, represented by fdCell i,j Forming filtered image I FC .
7. The method for pattern wafer defect detection based on adaptive local frequency domain filtering according to claim 1, characterized in that: In step 6, the normalized frequency filter factor S is calculated. i,j The mean Judge S i,j Is it greater than or equal to Forming the logical array L i,j , thereby defining the abnormal area, namely: The value of 1 in the logical array corresponds to the value in image I FC In the process, an abnormal region is formed, which reflects the defective cell region.
8. The method for pattern wafer defect detection based on adaptive local frequency domain filtering according to claim 1, characterized in that: In step 7, the defect location is performed by using the improved peak-centroid method. FC For each abnormal region in the , perform the following steps: a. The kth abnormal area Area k Take the absolute values of the grayscale values of all pixels in the image, sort these absolute values in descending order, select the pixel coordinates corresponding to the first n values in the sorting, and form a coordinate set TP k : Sort k =sort(|Area k (:)|,′descend′) TP k ={(x,y)||Area k (x,y)|∈Sort k (1:n)} Among them, 'descend' means the array is sorted in descending order, k is the sorted array of the absolute values of the grayscale values of pixels in the kth abnormal area, TP k Area is the pixel coordinate set corresponding to the first n values in the array; k (x, y) represents the grayscale value of the pixel coordinate (x, y) in the kth abnormal area; b. Calculate the coordinates TP in each abnormal area k The total value of the corresponding numerical value I k : c. Calculate TP k The weighted average of the elements in , i.e., the defect location in each abnormal area: in The x and y coordinates of each defect, respectively. k That is, a defect coordinate set Centroid is constructed, and then the defect coordinates are regressed and marked into the rotated image according to the position of the taken region of interest.
9. The method for pattern wafer defect detection based on adaptive local frequency domain filtering according to claim 1, characterized in that: The order of differences in step 2 and step 3 is interchanged, that is, firstly the second-order difference in the row direction is performed, and then the second-order difference in the column direction is performed on the image after the second-order difference in the row direction.
10. The method for pattern wafer defect detection based on adaptive local frequency domain filtering according to claim 1, characterized in that: The wafer image is an image collected by a bright field or dark field optical microscope; alternatively, defect detection is performed on a non-patterned wafer, and step 4 is directly entered at this time to select a period for frequency domain analysis and perform corresponding frequency domain filtering, and then perform defect extraction on the filtered image.
Citation Information
Patent Citations
Single-side light-entering-type light guide plate defect extraction method
CN108346141A
Refractory brick surface scratch identification method based on frequency domain filtering enhancement
CN108550138A
Method for fault diagnosis of variable-rotating-speed bearing through variation nonlinear mode decomposition
CN109668733A
Wafer surface defect identification method based on out-of-focus image, storage medium and electronic equipment
CN116485706A
Fabric defect detection method based on deep convolutional neural network and visual saliency
WO2019104767A1
Cited By
Image processing method and device, medium and product
CN120387987A
Wafer image preprocessing method and device, equipment and storage medium
CN120672664A
Wafer surface trace particulate matter identification method and system based on multispectral imaging
CN121640187A
