Crystal grain image interception and alignment method based on multi-feature pattern cooperative positioning

Abstract

The invention relates to a crystal grain image interception and alignment method based on multi-feature pattern cooperative positioning. The method comprises the following steps: aligning a detection camera with a standard crystal grain on a wafer, shooting, determining a standard crystal grain image in the images shot by the detection camera, selecting a feature pattern on the standard crystal grain image, and storing the feature pattern in an image identification template; determining a position difference parameter between the center coordinate of each characteristic pattern and the starting point coordinate of the standard crystal grain image; global scanning is identified through an image identification template; determining a starting point coordinate of the to-be-intercepted crystal grain image; intercepting a crystal grain image from each image shot by the detection camera according to the coordinates of the starting point and the image parameters of the standard crystal grain image; and aligning the crystal grain image with a standard crystal grain image through affine transformation. According to the invention, global errors can be suppressed, high-precision positioning is realized, and robustness and fault-tolerant capability are enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and particularly relates to a method for intercepting and aligning crystal grain images based on collaborative positioning of multiple feature patterns. Background Art

[0002] Wafer inspection is a key link in semiconductor manufacturing. It is necessary to perform complete image interception and alignment on each crystal grain through high-precision image interception technology to complete subsequent image stitching, defect detection or feature analysis. Traditional methods usually rely on single feature matching or line-by-line scanning for positioning, and there are the following problems:

[0003] Low efficiency: When processing ultra-large wafer images (such as 300mm wafers), traditional pixel-by-pixel traversal and matching algorithms (such as SIFT, ORB) have high computational complexity and are difficult to meet real-time requirements.

[0004] Cumulative error: Crystal grain positioning depends on local feature matching, and is prone to false matching due to image noise and periodic texture interference, resulting in subsequent crystal grain interception offset or rotation angle deviation.

[0005] Insufficient robustness: If the recognition of a single feature pattern fails (such as uneven illumination, defect occlusion), traditional methods cannot adaptively switch the matching strategy, resulting in process interruption. Summary of the Invention

[0006] The present invention provides a method for intercepting and aligning crystal grain images based on collaborative positioning of multiple feature patterns, which is used to solve at least one of the technical problems mentioned in the background art.

[0007] The technical solution of the present invention is as follows: A method for intercepting and aligning crystal grain images based on collaborative positioning of multiple feature patterns, comprising:

[0008] S10: Align the detection camera with the standard crystal grains on the wafer and take a picture. Determine the standard crystal grain image in the image taken by the detection camera, select N feature patterns on the standard crystal grain image, where N≥4, and save the feature patterns to the image recognition template;

[0009] S20: Determine the position difference parameter between the center coordinates of each feature pattern and the starting point coordinates of the standard crystal grain image;

[0010] S30: Perform a line-by-line global scan on the wafer through the detection camera, and identify N feature patterns in each image taken by the detection camera through the image recognition template;

[0011] S40: Obtain the center coordinates of the recognized feature patterns, and determine the starting point coordinates of the crystal grain image to be intercepted according to the center coordinates and the corresponding position difference parameters;

[0012] S50: Crop the grain images from each image captured by the detection camera according to the starting point coordinates and the image parameters of the standard grain image.

[0013] S60: Align the grain images with the standard grain image through affine transformation according to at least three characteristic patterns in the grain images and the characteristic patterns in the standard grain image.

[0014] Furthermore, the selection rules for the characteristic patterns include: the characteristic patterns are unique in the standard grain image; the width and height dimensions of the characteristic patterns are greater than 64×64 pixels and the edge intensity > 50, the spacing between adjacent characteristic patterns in the width direction is greater than 20% of the width of the camera image, and the spacing between adjacent characteristic patterns in the height direction is greater than 20% of the height of the camera image.

[0015] Furthermore, the step S30 includes:

[0016] S310: Scan the first i rows of the wafer row by row through the detection camera, and identify the characteristic patterns in the images of the first i rows captured by the detection camera through the image recognition template.

[0017] S320: Obtain the center coordinates of all identified characteristic patterns, and transform the center coordinates from the image coordinate system to the global coordinate system.

[0018] S330: Establish a minimized residual optimization function according to the transformed center coordinates of all characteristic patterns in the first i rows and the theoretical grid coordinates.

[0019] S340: Solve the minimized residual optimization function through nonlinear least squares method to obtain the optimized grid parameters.

[0020] S350: Generate the predicted center coordinates of all grains in the global coordinate system according to the optimized grid parameters and the center coordinates of the reference grain in the global coordinate system.

[0021] S360: Determine the search area according to the predicted center coordinates, and the detection camera scans the remaining area of the wafer row by row, and identifies the characteristic patterns in the search area in the images captured by the detection camera through the image recognition template.

[0022] Furthermore, the step S320 includes:

[0023] Perform a logical operation on the center coordinates in the image coordinate system and the pre-calibrated external parameter matrix to obtain the center coordinates in the global coordinate system.

[0024] Furthermore, calculate the similarity between each image captured by the detection camera and the characteristic patterns in the image recognition template through the NCC template matching algorithm.

[0025] If the similarity meets the preset threshold, it is determined that the feature pattern is recognized.

[0026] Further, the step S340 includes:

[0027] Iteratively solve the minimized residual optimization function through the Levenberg - Marquardt algorithm. When the stop iteration condition is reached, output the optimized grid parameters. The stop iteration condition includes: the residual change rate < 1e - 6 or reaching the maximum number of iterations.

[0028] Further, the step S360 includes:

[0029] According to the predicted center coordinates of the grains in the global coordinate system, the width and height of the image, and the image resolution, determine the predicted starting point coordinates of the grains in the global coordinate system;

[0030] According to the predicted starting point coordinates of the grains in the global coordinate system and the position gap parameter, determine the center coordinates of all feature patterns in the global coordinate system;

[0031] Expand a preset inner margin outward from the center coordinates of all feature patterns in the global coordinate system to determine the search area.

[0032] Further, the step S10 includes:

[0033] After performing Gaussian filtering and gray - value normalization on all feature patterns, save them to the image recognition template.

[0034] Further, the step S60 includes: Pair at least 3 feature patterns in each grain image with the feature patterns in the standard grain image, and through the affine transformation matrix, stack the center coordinates in the image coordinate system of the paired feature patterns to form an over - determined system of equations;

[0035] Through the matrix decomposition algorithm, calculate the minimized sum of squared residuals of the over - determined system of equations.

[0036] Advantages of the present invention:

[0037] 1. Global error suppression and high - precision positioning

[0038] Eliminate cumulative error: Through the feature pattern data of the first four rows of grains, use the least - squares method to uniformly optimize the wafer rotation angle and grain spacing, control the positioning residual within the sub - pixel level (such as ±0.1 pixel), and avoid the error accumulation problem of traditional row - by - row matching.

[0039] High - precision affine transformation: Based on the dynamic affine alignment of multiple feature pattern point pairs, compensate for local deformations (such as stage vibration, thermal expansion), and the alignment accuracy reaches ±0.5μm (at a resolution of 0.5μm / pixel).

[0040] 2. Significantly improved computational efficiency

[0041] Predictive search: By optimizing the grid parameters, the search area for the target grain is predicted, and the search range is reduced to 1% - 5% of the original image. The single-frame image processing time is reduced from 10 ms to 0.1 ms, supporting high-speed scanning (>500 mm / s).

[0042] 3. Enhanced robustness and fault tolerance

[0043] Multi-feature pattern collaborative fault tolerance: For each grain, N independent feature patterns are selected, N≥4, supporting dynamic elimination of abnormal points (such as dirt and occlusion). Only 3 valid points are required to complete the affine transformation, ensuring the continuity of the process.

[0044] 4. System compatibility and scalability

[0045] Flexibly adapt to different process nodes, compatible with multiple types of feature patterns (edge markers, circuit structures, etc.). Low hardware dependence: Only a standard industrial camera and a stage grating ruler are required, without additional high-precision sensors, reducing the deployment cost. Description of the drawings

[0046] Figure 1 is the flowchart of the present invention.

[0047] Figure 2 is the schematic diagram of the standard grain in the present invention. Detailed implementation manners

[0048] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0049] In the technical solution of the present invention, Figure 1 is the flowchart provided according to the specific process of a method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning of the present invention. As Figure 1 shown, the present invention includes:

[0050] S10: Align the detection camera with the standard grains on the wafer and take pictures. Determine the standard grain image in the image taken by the detection camera, select N feature patterns on the standard grain image, N≥4, and save the feature patterns to the image recognition template.

[0051] Move the detection camera to the center of the standard die on the wafer, and select a rectangular frame to determine the position of the standard die in the image. As Figure 2 shown, Figure 2 a standard die image is formed within the cross-shaped rectangular frame in start , determine the coordinates, pixel width, and height (X start , Y die , W die , H Figure 2 ) of the standard die image in the original image, and frame the feature patterns inside the standard die image.

[0052] The selection rules for the feature patterns include: the width and height dimensions of the feature patterns are greater than 64×64 pixels and the edge intensity > 50, the spacing between adjacent feature patterns in the width direction is greater than 20% of the camera image width, and the spacing between adjacent feature patterns in the height direction is greater than 20% of the camera image height.

[0053] Select 4 unique regions for the standard die image, and preferably satisfy:

[0054] The framed feature patterns are unique in the standard die image (no duplicates)

[0055] The area size of the framed feature patterns is greater than 64×64 pixels (corresponding to a physical size of 32μm×32μm);

[0056] The gray-scale gradient of the feature patterns is significant, specifically, the Sobel edge intensity > 50;

[0057] The center distance between adjacent feature patterns needs to be greater than 20% of the width / height of the camera image

[0058] All feature patterns need to be Gaussian filtered and gray-scale normalized, and then saved to the image recognition template to generate the image recognition template, that is, the NCC template.

[0059] Perform Gaussian filtering (standard deviation σ = 1.0) on each feature pattern to reduce noise, normalize the gray scale to [0, 255], and store it as the NCC template.

[0060] S20: Determine the position difference parameter between the center coordinates of each feature pattern and the starting point coordinates of the standard die image.

[0061] Record the position difference parameter offset between the center of each feature pattern and the starting point of the standard die image, where the starting point of the standard die image is the starting point of the rectangular frame.

[0062]

[0063] Among them, are the coordinates of the center point of the feature pattern, and the label (X start , Y start ) are the starting point coordinates of the standard grain image, and are the position gap parameters.

[0064] S30: The inspection camera performs a global scan of the wafer row by row, and through the image recognition template, N feature patterns are recognized in each image captured by the inspection camera.

[0065] In the full-wafer scanning stage, parallel feature pattern recognition is performed on the images captured by the inspection camera. Specifically, the NCC matching algorithm is used.

[0066] The similarity between each image captured by the inspection camera and the feature pattern in the image recognition template is calculated through the NCC template matching algorithm; if the similarity meets the preset threshold, it is determined that the feature pattern has been recognized.

[0067] Calculate the similarity between the image recognition template and the image area. If it meets the preset threshold, it is considered that the feature pattern has been recognized, and record the center coordinates (x pattern_center , y pattern_center ) of the feature pattern in the global coordinate system.

[0068] The specific steps of S30 include:

[0069] S310: The inspection camera scans the first i rows of the wafer row by row, and through the image recognition template, the feature pattern is recognized in the first i rows of images captured by the inspection camera.

[0070] The grains of the wafer are arranged in a grid pattern. During the wafer scanning process, each grain (column number j = 0, 1,..., N) in the first four rows (row number i = 0, 1, 2, 3) performs the following operations:

[0071] The inspection camera captures an image and recognizes the feature patterns of the four corner points through NCC template matching.

[0072] S320: Obtain the center coordinates of all recognized feature patterns and convert the center coordinates from the image coordinate system to the global coordinate system.

[0073] After recognizing the feature patterns of the four corner points, record the center coordinates of the feature pattern in the image coordinate system.

[0074] According to the mapping relationship between the image coordinate system and the global coordinate system, convert the center coordinates of the feature pattern from the image coordinate system to the global coordinate system, and record the center coordinates (X i,j , Y i,j ) in the global coordinate system.

[0075] The specific conversion method is as follows:

[0076] Perform a logical operation on the center coordinates in the image coordinate system and the pre-calibrated external parameter matrix to obtain the center coordinates in the global coordinate system.

[0077] Application of the external parameter matrix: Through the pre-calibrated external parameter matrix T, convert the center coordinates (u center , v center ) = (W / 2, H / 2) (where W and H are the width and height of the image) in the image coordinate system to the center coordinates in the global coordinate system:

[0078]

[0079] Among them, (u center , v center ) are the center coordinates in the image coordinate system, (X global_center , Y global_center ) are the center coordinates in the global coordinate system, and T is the calibrated external parameter matrix.

[0080] The definition of the global coordinate system is as follows: Taking the center of the wafer as the origin, the X / Y axes are parallel to the movement direction of the stage, and the pixel resolution is the same as that of the detection camera (such as 0.5μm / pixel).

[0081] S330: Establish a minimization residual optimization function based on the center coordinates and theoretical grid coordinates of all feature patterns in the first i rows.

[0082] Among them, the theoretical grid coordinates are based on the position of the first standard grain as the reference point, and the subsequent theoretical positions of all grains are deduced, that is:

[0083]

[0084] Among them, d x is the horizontal spacing between adjacent grains, d y is the vertical spacing between adjacent grains, and θ is the global rotation angle of the wafer. It is assumed that the arrangement of all grains is a standard grid, that is, the spacing is consistent and the rotation angle is consistent.

[0085] The minimization residual optimization function is used to minimize the residual between the center coordinates and theoretical grid coordinates of all grains in the first four rows.

[0086]

[0087] Parameter definition: d x is the horizontal spacing between adjacent grains, d y is the vertical spacing between adjacent grains, and θ is the global rotation angle of the wafer.

[0088] S340: Solve the minimized residual optimization function by non - linear least squares method to obtain the optimized grid parameters.

[0089] Iteratively solve the minimized residual optimization function by Levenberg - Marquardt algorithm. When the iteration stop condition is reached, output the optimized grid parameters. The iteration stop conditions include: the residual change rate < 1e - 6 or the maximum number of iterations is reached. Among them, the optimized grid parameters are θ, (dx, dy).

[0090] Use Levenberg - Marquardt algorithm to iteratively solve θ, (dx, dy), and set the initial values as follows:

[0091] θ0 = 0: Assume no initial rotation.

[0092] Estimate according to the mean value of the adjacent grain spacings in the first four rows.

[0093] Iteration termination conditions: the residual change rate < 1e - 6 or the maximum number of iterations (e.g., 100 times).

[0094] S350: Generate the predicted center coordinates of all grains in the global coordinate system according to the optimized grid parameters and the center coordinates of the reference grain in the global coordinate system.

[0095] Among them, the reference grain can be determined by those skilled in the art and can be the grain in the first row and the first column.

[0096] The optimized global rotation angle θ opt , the horizontal spacing between adjacent grains and the vertical spacing Then:

[0097]

[0098] is the measured center coordinate of the reference grain, the predicted center coordinate of the grain, θ opt , are the optimized grid parameters.

[0099] S360: Determine the search area according to the predicted center coordinates. The inspection camera scans the remaining area of the wafer row by row, and identifies the characteristic pattern in the search area of the image captured by the inspection camera through the image recognition template.

[0100] Determine the predicted starting point coordinates of the grains in the global coordinate system according to the predicted center coordinates of the grains in the global coordinate system, the width and height of the image, and the image resolution.

[0101] Among them, the predicted center coordinates of the grains in the global coordinate system are (Xglobal_center , Y global_center ).

[0102] Predicted starting point coordinates of grains in the global coordinate system:

[0103]

[0104] Where res is the pixel resolution (such as 0.5 μm / pixel), and W and H are the width and height of a single image respectively (such as 4096×4096 pixels).

[0105] Determine the center coordinates of all feature patterns in the global coordinate system based on the predicted starting point coordinates of grains in the global coordinate system and the position difference parameter.

[0106] Determine the search area by expanding a preset inner margin outward from the center coordinates of all feature patterns in the global coordinate system.

[0107] Based on the optimized position difference parameter, the theoretical areas of all grains in the global coordinate system in the images captured by the detection camera can be predicted. For the subsequently scanned grains (the area where the row number i≥4 is the remaining area of the circular wafer), the search area of the feature pattern can be deduced according to the previously recorded position difference parameter between the center of each feature pattern and the starting point of the grain. The search area is a rectangle centered on the theoretical coordinates, and expands outward by a preset inner margin (such as 50 pixels).

[0108] S40: Obtain the center coordinates of the recognized feature patterns, and determine the starting point coordinates of the grain image to be intercepted based on the center coordinates and the corresponding position difference parameter.

[0109] Search for feature patterns only within the search area according to the predicted pixel coordinate search area, and inversely obtain the starting point coordinates of the newly scanned grains through the position difference parameter.

[0110] S50: Intercept the grain image in each image captured by the detection camera according to the starting point coordinates and the image parameters of the standard grain image.

[0111] S60: Align the grain image with the standard grain image through affine transformation according to at least 3 feature patterns in the grain image and the feature patterns in the standard grain image.

[0112] Pair at least 3 feature patterns in each grain image with the feature patterns in the standard grain image, and form an overdetermined system of equations by stacking the center coordinates in the image coordinate system of the paired feature patterns through the affine transformation matrix;

[0113] Calculate the minimum residual sum of squares of the overdetermined system of equations through the matrix decomposition algorithm.

[0114] Calculation of Affine Transformation Matrix:

[0115] Based on the recognition result of the characteristic pattern, multiple pairs of characteristic pattern pixel points of the standard grain image and the grain image (currently captured) can be obtained:

[0116] The central coordinates of the four characteristic patterns of the standard grain image in the image coordinate system are:

[0117] The central coordinates of the four characteristic patterns of the grain image (currently captured) in the image coordinate system are:

[0118] The affine transformation matrix is established in the form of:

[0119]

[0120] If the number of matching point pairs is N≥3, the total matrix equation (overdetermined system of equations) can be obtained through point pair stacking:

[0121] A·p=b

[0122] where A is a 2N×6 coefficient matrix

[0123] p=[a 11 ,a 12 ,t x ,a 21 ,a 22 ,t y T is the parameter vector to be solved

[0124] b is a 2Nx1 observation vector

[0125] The solution of the overdetermined system of equations (number of equations > number of unknowns) is obtained by minimizing the sum of squared residuals:

[0126]

[0127] Its analytical solution is:

[0128] p=(A T A) -1 A T b

[0129] Numerically stable matrix decomposition algorithms such as singular value decomposition (SVD) or QR decomposition can be used.

[0130] After aligning the current grain image with the standard grain image through affine transformation and saving the image, it can be used for subsequent production of the gold template or feature extraction.

[0131] ​Finally, it should be noted that the above specific embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the examples, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for intercepting and aligning grain images based on collaborative positioning of multi-feature patterns, characterized in that Including: S10: Align the detection camera with the standard die on the wafer and take a picture. Determine the standard die image in the image taken by the detection camera. Select N characteristic patterns on the standard die image, where N≥4. Save the characteristic patterns to the image recognition template; S20: Determine the position difference parameters between the center coordinates of each characteristic pattern and the starting point coordinates of the standard die image; S30: Perform a row-by-row global scan of the wafer through the detection camera. Through the image recognition template, identify the N characteristic patterns in each image taken by the detection camera; S40: Obtain the center coordinates of the identified characteristic patterns, and determine the starting point coordinates of the die image to be intercepted according to the center coordinates and the corresponding position difference parameters; S50: Intercept the die image in each image taken by the detection camera according to the starting point coordinates and the image parameters of the standard die image; S60: Align the die image with the standard die image through affine transformation according to at least 3 characteristic patterns in the die image and the characteristic patterns in the standard die image.

2. The method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning according to claim 1, wherein The selection rules of the characteristic patterns include: the characteristic patterns are unique in the standard die image; the width and height dimensions of the characteristic patterns are greater than 64×64 pixels and the edge intensity>50; the spacing between adjacent characteristic patterns in the width direction is greater than 20% of the width of the camera image, and the spacing between adjacent characteristic patterns in the height direction is greater than 20% of the height of the camera image.

3. The method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning according to claim 1, wherein The step S30 includes: S310: Perform a row-by-row scan of the first i rows of the wafer through the detection camera. Through the image recognition template, identify the characteristic patterns in the images of the first i rows taken by the detection camera; S320: Obtain the center coordinates of all the identified characteristic patterns, and convert the center coordinates from the image coordinate system to the global coordinate system; S330: Establish a minimized residual optimization function according to the center coordinates of all the characteristic patterns in the first i rows after conversion and the theoretical grid coordinates; S340: Solve the minimized residual optimization function through the nonlinear least squares method to obtain the optimized grid parameters; S350: Generate the predicted center coordinates of all the dice in the global coordinate system according to the optimized grid parameters and the center coordinates of the reference die in the global coordinate system; S360: Determine the search area according to the predicted center coordinates. The detection camera performs a row-by-row scan of the remaining area of the wafer. Through the image recognition template, identify the characteristic patterns in the search area in the image taken by the detection camera.

4. The method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning according to claim 3, characterized in that The step S320 includes: Perform a logical operation on the center coordinates in the image coordinate system and the pre-calibrated external parameter matrix to obtain the center coordinates in the global coordinate system.

5. The method for intercepting and aligning die images based on multi-characteristic pattern collaborative positioning according to claim 3, characterized in that: Calculate the similarity between each image taken by the detection camera and the characteristic patterns in the image recognition template through the NCC template matching algorithm; If the similarity meets the preset threshold, it is determined that the characteristic pattern is recognized.

6. The method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning according to claim 3, wherein, The step S340 includes: Iteratively solve the minimized residual optimization function through the Levenberg-Marquardt algorithm. When the iteration stop condition is reached, output the optimized grid parameters. The iteration stop condition includes: the residual change rate < 1e-6 or reaching the maximum number of iterations.

7. The method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning according to claim 3, wherein The step S360 includes: Determine the predicted starting point coordinates of the grain in the global coordinate system according to the predicted center coordinates of the grain, the width and height of the image, and the image resolution in the global coordinate system; Determine the center coordinates of all feature patterns in the global coordinate system according to the predicted starting point coordinates of the grain and the position gap parameter in the global coordinate system; Determine the search area by expanding a preset inner margin outward from the center coordinates of all feature patterns in the global coordinate system.

8. The method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning according to claim 1, wherein The step S10 includes: After performing Gaussian filtering and gray value normalization on all feature patterns, save them to the image recognition template.

9. The method for intercepting and aligning grain images based on multi-feature pattern collaborative positioning according to claim 1, wherein The step S60 includes: Pair at least 3 feature patterns in each grain image with the feature patterns in the standard grain image, and through the affine transformation matrix, stack the center coordinates of the paired feature pattern images in the coordinate system to form an overdetermined system of equations; Calculate the minimized sum of residual squares of the overdetermined system of equations through the matrix decomposition algorithm.

Images