Inter-image brightness compensation method and wafer defect detection system

Abstract

The invention provides an inter-image brightness compensation method and a wafer defect detection system, and belongs to the field of image processing and semiconductor detection, and the method comprises the steps: dividing an input image into grid blocks with fixed sizes according to grids, obtaining a local grid block image and a corresponding mask binary image, acquiring a brightness compensation kernel corresponding to the local grid block image; acquiring a local grid block image subjected to brightness compensation; and splicing to acquire a complete adjacent Die reference image subjected to brightness compensation. According to the method, the brightness compensation between the images is carried out in a grid square mode, the local brightness difference between the reference image and the to-be-detected image can be effectively reduced, the robustness of a compensation result of the local brightness of the high-resolution image is improved by introducing a blocking strategy of the grid square mode and a self-adaptive brightness compensation kernel solving mechanism, and the robustness of the local brightness of the high-resolution image is improved. Accuracy and stability of subsequent image differential detection are remarkably improved, and popularization and application in the field of semiconductor optical detection and image processing are facilitated.

Description

Technical Field

[0001] This invention belongs to the field of semiconductor inspection and relates to a method for image brightness compensation using a grid block approach, particularly a method for calculating image brightness compensation during defect detection in semiconductor manufacturing, which improves the accuracy and stability of differential detection results. Background Technology

[0002] In semiconductor manufacturing, optical inspection technology is widely used for the detection and classification of surface defects on wafers. Image difference analysis is a mainstream method that involves calculating the difference between a test image and a reference image. By comparing and analyzing the differences between the two images pixel by pixel, the location of potential defects can be determined. The reference image is typically taken from a defect-free, normal area under the same process conditions, such as adjacent chips (die-to-die), to ensure that the two images have the same structure. Under ideal conditions, the reference image and the test image have highly consistent imaging brightness, ensuring that the difference results accurately reflect structural changes between the images, rather than the effects of brightness deviations.

[0003] During the inspection process, images are usually acquired by a high-precision optical microscope or a line scan camera. However, due to factors such as the stability of the light source, the reflection angle, and the exposure time, there is often a significant brightness difference between the reference image and the image under test. This brightness difference will seriously interfere with the quality of the difference map, resulting in the detection of a large number of nuisances, which will affect the identification of the true defect.

[0004] Traditional methods to address these issues typically employ histogram matching based on sliding windows or linear brightness matching based on the global image to preprocess the reference image. However, these methods often have their limitations. For instance, linear brightness matching adjusts the reference image's brightness by calculating the global average brightness ratio, thus failing to address local brightness unevenness within the image. Histogram matching, on the other hand, adjusts brightness by matching grayscale distributions, which can disrupt local contrast. Therefore, these methods are not suitable for high-precision wafer inspection scenarios. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide an image inter-brightness compensation method and a wafer defect detection system using a grid block method, so as to solve the problems existing in the above-mentioned prior art.

[0006] An image-to-image brightness compensation method using a grid-block approach includes: S1, acquiring a target detection die image of the same resolution, adjacent die reference images, and corresponding binary mask images of the regions of interest as input images; S2, using a gridding method, dividing the input image into fixed-size grid blocks according to a grid, and acquiring local grid block images and corresponding binary mask images; S3, acquiring a brightness compensation kernel corresponding to the local grid block images based on the local grid block images and the binary mask images; S4, applying the corresponding brightness compensation kernel to the local grid block images of each adjacent die reference image based on the brightness compensation kernel, and acquiring brightness-compensated local grid block images; S5, stitching the brightness-compensated grid block images together according to the order of adjacent die reference images to obtain complete brightness-compensated adjacent die reference images.

[0007] Furthermore, the process of obtaining the local grid block image and the corresponding binary mask image includes: S21, obtaining the minimum overlapping pixel size and grid block size of the input; S22, calculating the starting position of all grid blocks in the input image based on the width and height information of the input image, the minimum overlapping pixel size, and the grid block size; S23, dividing the input image and obtaining the local grid block image based on the starting position, grid block size, image to be tested, adjacent die reference images, and binary mask image of the region of interest.

[0008] Furthermore, when the edges of the grid blocks are insufficient to cover the width and height of the entire grid block, a boundary adjustment strategy is adopted to shift the last row or column of blocks to the left or upward to ensure that the size of all blocks remains consistent.

[0009] Furthermore, the solution process for the brightness compensation kernel includes: S31, obtaining the defined brightness compensation kernel size; S32, defining and initializing the autocorrelation matrix and cross-correlation matrix based on the brightness compensation kernel size; S33, obtaining the autocorrelation matrix and cross-correlation matrix based on the local grid block image and the corresponding mask binary image; S34, applying a diagonal enhancement strategy to the autocorrelation matrix based on the autocorrelation matrix to obtain the enhanced autocorrelation matrix; S35, obtaining the brightness compensation kernel corresponding to the local grid block image based on the enhanced autocorrelation matrix and cross-correlation matrix.

[0010] Further, the process of obtaining the enhanced autocorrelation matrix and cross-correlation matrix includes: S351, obtaining the input local grid block image, mask binary image, and defined brightness compensation kernel size; S352, obtaining the effective region in the local grid block image based on the mask binary image; S353, obtaining a sliding window of the local grid block image based on the reference image based on the effective region of the reference image and the brightness compensation kernel size; S354, calculating the autocorrelation matrix based on the sliding window and the effective region of the reference image; adding a regularization term to the diagonal of the autocorrelation matrix to obtain the enhanced autocorrelation matrix; S355, calculating the cross-correlation matrix based on the sliding window, the image to be tested, and the effective region of the reference image; S356, solving the linear equation system based on the enhanced autocorrelation matrix and cross-correlation matrix to obtain the solution vector, and reconstructing it into the brightness compensation kernel of the corresponding local grid block image.

[0011] Furthermore, when obtaining the enhanced autocorrelation matrix in step S354, the autocorrelation matrix... A diagonal enhancement strategy is adopted in the autocorrelation matrix. Add a small regularization term to the main diagonal elements To ensure the autocorrelation matrix It is an invertible matrix: ;in It is the identity matrix. , .

[0012] Furthermore, solving for the brightness compensation kernel set in step S356 includes: based on the inverse matrix of the autocorrelation matrix. and cross-correlation matrix Construct a system of linear equations to solve for the brightness compensation kernel vector. The system of linear equations is as follows: The brightness compensation kernel vector is solved using Cholesky decomposition or least squares method. , ; the set of local grid squares and the corresponding brightness compensation kernel vector Reconstruction Brightness compensation kernel matrix The brightness compensation kernel set is obtained. .

[0013] Furthermore, the method for obtaining the local grid image with brightness compensation in S4 includes: S41, using a brightness compensation kernel... For local reference mesh blocks The convolution operation on the set yields a local reference mesh image after brightness compensation. S42. After performing local compensation on all local regions, a set of grid block image compensation results is obtained.

[0014] Furthermore, the set of compensation results for grid block images... All the brightness-compensated reference images are arranged in the original order of all dies, and a strategy of subsequent blocks overwriting previous blocks is adopted to avoid repeated accumulation, finally resulting in a complete brightness-compensated reference image. .

[0015] The present invention also provides a wafer defect detection system, including a wafer stage, an image acquisition module and a processor. The processor performs brightness compensation on the scanned image based on the aforementioned inter-image brightness compensation method, thereby improving the accuracy and stability of defect detection.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The present application performs brightness compensation between images by means of grid blocks, which can effectively reduce the local brightness difference between the reference image and the image to be tested. It is especially suitable for high-resolution image brightness compensation scenarios. By introducing a grid block-based segmentation strategy and an adaptive brightness compensation kernel solution mechanism, the robustness of the local brightness compensation results of high-resolution images is improved, and the accuracy and stability of subsequent image scoring and detection are significantly improved, which facilitates its application in the fields of semiconductor optical detection and image processing. Attached Figure Description

[0017] Figure 1 This is a flowchart of the image brightness compensation method using a grid block approach according to the present invention. Figure 2 This is a schematic diagram of the image grid block division; Figure 3 This is a diagram illustrating the areas of user interest. Figure 4 This is a schematic diagram of image edge augmentation operation; Figure 5 This is a schematic diagram of the image sliding window operation; Figure 6 This is a schematic diagram of the splicing and reconstruction of local brightness compensation results. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that the terms "system," "device," "unit," and / or "module" as used in this specification are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0020] Flowcharts are used in this specification to illustrate the operations performed by the system according to embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0021] A method for inter-image brightness compensation using a grid-block approach, see [link to relevant documentation]. Figure 1 The methods for brightness compensation between images include the following.

[0022] Preparation: Define the high-resolution image input to the current die as the image to be tested. The high-resolution image of the adjacent die input is used as the reference image. The two images have the same size and alignment, are both single-channel grayscale images, and have a resolution defined as follows: Generally, it can achieve Or higher.

[0023] See Figure 3 Define a user-defined Region of Interest (ROI) mask, denoted as the ROI binary mask image. This mask is used to identify The effective region for brightness compensation needs to be calculated. The region of interest is represented by a binary mask image. Resolution and reference image The same, and represented in the form of a binary image, with the effective region's pixel value being... or The pixel value of the invalid region is .

[0024] The specific steps of the method are as follows.

[0025] Step 1, see Figure 2 The input image to be tested Reference image Binary mask image of the region of interest The image is divided into several local grid blocks of fixed size.

[0026] Define grid blocks The width and height are respectively and The minimum overlapping pixel size is and ,in and These represent the overlap dimensions of adjacent grid squares in the horizontal and vertical directions, respectively.

[0027] To ensure the effectiveness of grid division, the grid blocks The width and height must meet the requirements ,and The minimum overlapping pixel size must meet the following requirements. ,and .

[0028] Calculate grid blocks The number of grids that can be divided in the horizontal (x) and vertical (y) directions. and .

[0029] , .

[0030] Step 2: Based on the number of grid cells, grid blocks The initial coordinate sequence in the horizontal and vertical directions can be defined as follows: and And it is expressed as follows.

[0031] ; .

[0032] Among them, when or At that time, the corresponding grid block The edges are insufficient to cover the entire For the region, the solution employs a boundary adjustment strategy, meaning that the last row or column of the patch will be shifted horizontally or upwards. The starting coordinates of the corrected edge are then determined. and And it is expressed as follows.

[0033] ; .

[0034] This strategy ensures that all patch sizes remain consistent, thus guaranteeing consistent matrix dimensions during subsequent brightness compensation kernel calculations. The corresponding local mesh image patches for the three input images are shown below.

[0035] ; ; .

[0036] Step 3: Construct the translation set and define brightness compensation. The size is ,in If the number is odd, the padding size is... Based on fill size Define the translation set S (shiftlist): .

[0037] Step 4: For the three types of local mesh image patches obtained in Step 1 Define the current local mesh image block, and represent it as follows.

[0038] .

[0039] Step 5: For the local mesh image patch R defined in Step 4, establish a set of translation reference matrices for the current local mesh image patch. .

[0040] ;in, Translation amount For each translation amount See Figure 5 By using reference images The sliding window operation is as follows.

[0041] ;in It's a cyclic translation operation; for edge pixels, the coordinates are taken from the original image. Fill the pixel values ​​into the new image If it goes beyond the boundary, it wraps around to the other side. See also Figure 4 The convolution operation can be equivalent to performing boundary expansion processing on the original image, thereby obtaining an expanded image containing neighborhood information before the convolution calculation. Based on the obtained translated reference image... .

[0042] Step 6: For the translation reference matrix set Ref obtained in Step 5 and the image to be tested T, input Ref and T into the local correlation modeling module, and perform weighted correlation operation on the reference image and the image to be tested under the constraint of the weight matrix W to construct the autocorrelation matrix A and the correlation matrix B.

[0043] Wherein, the autocorrelation matrix A is used to characterize the statistical correlation characteristics of the reference image R within a local region, and the cross-correlation matrix B is used to characterize the correlation characteristics between the reference image R and the image to be tested T; the weight matrix W is used to limit the image positions participating in the calculation, so that image positions that do not meet the preset conditions are not included in the correlation calculation. T, R, and M are the current local grid image, i.e. .

[0044] T represents the image to be tested. The corresponding local mesh image, where R represents the reference image. The corresponding local mesh image, where M represents the binary mask image of the region of interest. The corresponding local mesh image, This represents the set of reference translation matrices constructed based on the local mesh image R.

[0045] Step 7: Calculate the autocorrelation matrix obtained in Step 6. A diagonal enhancement strategy is adopted in the autocorrelation matrix. Add a small regularization term to the main diagonal elements The enhanced autocorrelation matrix is ​​obtained. To ensure the autocorrelation matrix It is an invertible matrix: ;in It is the identity matrix. , .

[0046] Step 8: For the enhanced autocorrelation matrices obtained in Steps 6 and 7... and cross-correlation matrix Construct a system of linear equations to solve for the brightness compensation kernel vector. , .

[0047] Solve using Cholesky decomposition or least squares method. , .

[0048] The local mesh set obtained in step one and the corresponding brightness compensation kernel vector obtained in step eight are used as examples. Reconstruction Brightness compensation kernel matrix The brightness compensation kernel set is obtained as follows.

[0049] .

[0050] Step 9: For local reference mesh blocks The brightness compensation kernel obtained in step eight is used. For local reference mesh blocks The convolution operation is performed on the set as follows.

[0051] .

[0052] Received This is the local reference grid image after brightness compensation. After local compensation for each local area, the set of grid block image compensation results is as follows.

[0053] .

[0054] Step 10: Compensation result set for the grid block image obtained in Step 9. See Figure 6 This invention employs a strategy of subsequent blocks overwriting preceding blocks to avoid redundant accumulation, ultimately obtaining a complete reference image after brightness compensation: ;in This represents a union operation within a spatial domain that covers overlapping regions.

[0055] A wafer defect detection system (not shown) includes a wafer stage, an image acquisition module, and a processor. The processor performs brightness compensation on the scanned image based on the aforementioned inter-image brightness compensation method, thereby improving the accuracy and stability of defect detection.

[0056] The method of image brightness compensation using a grid block approach of the present invention significantly reduces the local brightness difference between the reference image and the image under test, while solving the problem of local detail distortion of the reference image caused by global compensation. The local grid block approach enables the local kernel to adapt to the illumination changes of each region, providing higher quality images for subsequent image difference calculation and detection algorithms, and improving the accuracy and stability of wafer defect detection.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for inter-image brightness compensation using a grid block approach, characterized in that: Inter-image brightness compensation methods include: S1. Obtain the target image of the target detection die with the same resolution, the reference images of adjacent dies, and the corresponding binary mask image of the region of interest as the input image; S2. Using a gridding method, based on the input image, the input image is divided into grid squares of fixed size according to the grid, and the local grid square image and the corresponding mask binary image are obtained. S3. Based on the local grid block image and the mask binary image, obtain the brightness compensation kernel corresponding to the local grid block image; S4. Based on the brightness compensation kernel, apply the corresponding brightness compensation kernel to the local grid block image of each adjacent Die reference image to obtain the brightness-compensated local grid block image. S5. Based on the brightness-compensated grid block image, stitch them together in the order of adjacent die reference images to obtain a complete brightness-compensated adjacent die reference image.

2. The image brightness compensation method according to claim 1, characterized in that, The process of obtaining the local grid block image and the corresponding mask binary image includes: S21. Obtain the minimum overlapping pixel size and grid square size of the input; S22. Based on the width and height information of the input image, the minimum overlapping pixel size, and the grid square size, calculate the starting position of all grid squares in the input image; S23. Based on the starting position, grid size, image to be tested, adjacent die reference images, and binary mask image of the region of interest, divide the input image and obtain local grid image.

3. The image brightness compensation method according to claim 1 or 2, characterized in that: When the edges of a grid block are insufficient to cover the width and height of the entire grid block, a boundary adjustment strategy is adopted to shift the last row or column of blocks to the left or upward to ensure that the size of all blocks remains consistent.

4. The image brightness compensation method according to claim 1, characterized in that, The solution process for the brightness compensation kernel includes: S31. Obtain the defined brightness compensation kernel size; S32. Based on the brightness compensation kernel size, define and initialize the autocorrelation matrix and cross-correlation matrix; S33. Based on the local grid block image and the corresponding mask binary image, obtain the autocorrelation matrix and cross-correlation matrix; S34. Based on the autocorrelation matrix, implement the diagonal enhancement strategy on the autocorrelation matrix to obtain the enhanced autocorrelation matrix; S35. Based on the enhanced autocorrelation matrix and cross-correlation matrix, obtain the brightness compensation kernel corresponding to the local grid block image.

5. The image brightness compensation method according to claim 4, characterized in that, The process of obtaining the enhanced autocorrelation matrix and cross-correlation matrix includes: S351. Obtain the input local grid block image, mask binary image, and defined brightness compensation kernel size; S352. Based on the masked binary image, obtain the effective region in the local grid block image; S353. Based on the effective area of ​​the reference image and the size of the brightness compensation kernel, obtain a sliding window of the local grid block image based on the reference image; S354. Based on the effective region of the sliding window and the reference image, calculate and obtain the autocorrelation matrix; add a regularization term to the diagonal of the autocorrelation matrix to obtain the enhanced autocorrelation matrix; S355. Based on the sliding window, the effective regions of the image to be tested and the reference image, the cross-correlation matrix is ​​calculated and obtained. S356. Based on the enhanced autocorrelation matrix and cross-correlation matrix, solve the linear equation system to obtain the solution vector, and reconstruct it into the brightness compensation kernel of the corresponding local grid block image.

6. The image brightness compensation method according to claim 5, characterized in that: When obtaining the enhanced autocorrelation matrix in step S354, the autocorrelation matrix... A diagonal enhancement strategy is adopted in the autocorrelation matrix. Add a small regularization term to the main diagonal elements To ensure the autocorrelation matrix It is an invertible matrix: ;in It is the identity matrix. , .

7. The image brightness compensation method according to claim 6, characterized in that: Solving for the brightness compensation kernel set in step S356 includes: Based on the inverse matrix of the autocorrelation matrix and cross-correlation matrix Construct a system of linear equations to solve for the brightness compensation kernel vector. The system of linear equations is as follows: ; The brightness compensation kernel vector is solved using Cholesky decomposition or least squares method. , ; The set of local grid squares and their corresponding brightness compensation kernel vectors Reconstruction Brightness compensation kernel matrix The brightness compensation kernel set is obtained. .

8. The image brightness compensation method according to claim 7, characterized in that: The methods for obtaining local grid images with brightness compensation in S4 include: S41, Use brightness compensation kernel For local reference mesh blocks The convolution operation on the set yields a brightness-compensated local reference mesh image. ; S42. After performing local compensation on all local regions, a set of grid block image compensation results is obtained.

9. The image brightness compensation method according to claim 8, characterized in that: Set of compensation results for grid block images All the brightness-compensated reference images are arranged in the original order of all dies, and a strategy of subsequent blocks overwriting previous blocks is adopted to avoid repeated accumulation, finally resulting in a complete brightness-compensated reference image. .

10. A wafer defect detection system, comprising a wafer motion stage, an image acquisition module, and a processor, characterized in that: The processor performs brightness compensation on the scanned image based on the image inter-brightness compensation method according to any one of claims 1-9, thereby improving the accuracy and stability of defect detection.

Images