CN-122023219-A - Inter-image brightness compensation method and wafer defect detection system

CN122023219ACN 122023219 ACN122023219 ACN 122023219ACN-122023219-A

Abstract

The application provides an inter-image brightness compensation method and a wafer defect detection system, which belong to the field of image processing and semiconductor detection, wherein the method comprises the following steps: dividing an input image into grid squares with fixed size according to grids, acquiring a local grid square image and a corresponding mask binary image, acquiring a brightness compensation kernel corresponding to the local grid square image, acquiring a brightness compensated local grid square image, and splicing to acquire a complete brightness compensated adjacent Die reference image. According to the application, the brightness compensation between images is carried out in a grid square mode, so that the local brightness difference between the reference image and the image to be detected can be effectively reduced, the robustness of the compensation result of the local brightness of the high-resolution image is improved by introducing the blocking strategy in the grid square mode and the self-adaptive brightness compensation kernel solving mechanism, the accuracy and stability of the differential detection of the subsequent image are obviously improved, and the method is convenient to popularize and apply in the fields of semiconductor optical detection and image processing.

Inventors

CAI XIONGFEI
Ke Kexin
CHEN YI

Assignees

苏州矽行半导体技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The inter-image brightness compensation method by the grid square mode is characterized by comprising the following steps: s1, acquiring an image to be detected of a target detection Die, an adjacent Die reference image and a corresponding binary mask image of a region of interest with the same resolution as input images; S2, dividing the input image into grid squares with fixed sizes according to grids based on the input image by adopting a meshing method, and acquiring a local grid square image and a corresponding mask binary image; S3, acquiring a brightness compensation kernel corresponding to the local grid square image based on the local grid square image and the mask binary image; S4, based on the brightness compensation kernels, applying corresponding brightness compensation kernels to local grid square images of each adjacent Die reference image to obtain local grid square images with brightness compensation; and S5, based on the grid square images after brightness compensation, splicing the grid square images according to the sequence of the adjacent Die reference images, and obtaining the complete adjacent Die reference images after brightness compensation.
2. The method of claim 1, wherein the process of obtaining the local grid square image and the corresponding mask binary image comprises: s21, acquiring the input minimum overlapped pixel size and the grid square size; S22, calculating starting point positions of all grid squares in the input image based on the width and height information of the input image, the minimum overlapped pixel size and the grid square size; s23, dividing an input image based on a starting point position, a grid square size, an image to be detected, an adjacent Die reference image and a binary mask image of a region of interest, and acquiring a local grid square image.
3. The method of claim 1 or 2, wherein when the edges of the grid square are insufficient to cover the wide-high area of the complete grid square, a boundary adjustment strategy is used that the last row or column of square Patches is shifted left or up to ensure that the dimensions of all square Patches remain uniform.
4. The method of claim 1, wherein the solving of the luminance compensation kernel comprises: S31, acquiring the defined brightness compensation kernel size; s32, defining and initializing an autocorrelation matrix and a cross-correlation matrix based on the brightness compensation kernel size; S33, acquiring an autocorrelation matrix and a cross correlation matrix based on the local grid square image and the corresponding mask binary image; s34, based on the autocorrelation matrix, implementing diagonal enhancement strategy to the autocorrelation matrix to obtain an enhanced autocorrelation matrix; S35, acquiring a brightness compensation kernel corresponding to the local grid square image based on the enhanced autocorrelation matrix and the enhanced cross correlation matrix.
5. The method of claim 4, wherein the process of obtaining the enhanced auto-correlation matrix and the cross-correlation matrix comprises: S351, acquiring an input local grid square image, a mask binary image and a defined brightness compensation kernel size; s352, acquiring an effective area in the local grid square image based on the mask binary image; s353, based on the effective area and the brightness compensation kernel size of the reference image, acquiring a sliding window of the local grid square image based on the reference image; s354, based on the sliding window and the effective area of the reference image, calculating to obtain an autocorrelation matrix, adding regularization terms into the diagonal of the autocorrelation matrix, and obtaining an enhanced autocorrelation matrix; S355, calculating and obtaining a cross-correlation matrix based on the sliding window, the effective areas of the image to be detected and the reference image; and S356, based on the enhanced autocorrelation matrix and the cross correlation matrix, solving the linear equation set to obtain a solution vector, and reconstructing the solution vector into a brightness compensation core corresponding to the local grid square image.
6. The method for compensating luminance between images according to claim 5, wherein the autocorrelation matrix is obtained by the step S354 of obtaining the autocorrelation matrix In the autocorrelation matrix by adopting a diagonal enhancement strategy Adding a tiny regularization term to the main diagonal element of (2) To ensure an autocorrelation matrix Is a reversible matrix: Wherein Is an identity matrix of the unit cell, , 。
7. The method for luminance compensation between images according to claim 6, wherein solving the luminance compensation kernel set in step S356 comprises: Inverse matrix based on autocorrelation matrix Cross-correlation matrix Constructing a linear equation set for solving the brightness compensation kernel vector The system of linear equations is: ; solving the brightness compensation kernel vector by using Cholesky decomposition or least square method , ; For a set of local grid squares and corresponding luminance compensation kernel vectors Reconstruction of Brightness compensation kernel matrix Obtaining a brightness compensation kernel set 。
8. The method for luminance compensation between images according to claim 7, wherein the step of obtaining the luminance-compensated partial square block image in S4 comprises: S41, adopting brightness compensation core For local reference grid square Performing convolution operation on the set to obtain a local reference grid image subjected to brightness compensation ; And S42, after carrying out local compensation on all the local areas, obtaining a grid square image compensation result set.
9. The method for compensating luminance between images according to claim 8, wherein the result set is a set of compensation results for grid square images All the brightness compensated reference images in the system are arranged according to the original sequence of all the Die, and a strategy that the follow-up blocks cover the preceding blocks is adopted to avoid repeated accumulation, so that the complete brightness compensated reference image is finally obtained 。
10. A wafer defect detection system comprises a wafer motion platform, an image acquisition module and a processor, and is characterized in that the processor performs brightness compensation on an image obtained by scanning based on the brightness compensation method between images according to any one of claims 1-9, so that the accuracy and stability of defect detection are improved.

Description

Inter-image brightness compensation method and wafer defect detection system Technical Field The invention belongs to the field of semiconductor detection, relates to a method for compensating brightness between images in a grid square mode, and particularly relates to a method for calculating brightness compensation between images during defect detection in the semiconductor manufacturing process, which is used for improving accuracy and stability of a differential detection result. Background In the semiconductor manufacturing process, optical inspection techniques are widely used for detecting and classifying defects on the wafer surface. The image difference method is a mainstream detection method, and the method needs to perform difference calculation on an image (TEST IMAGE) to be detected and a reference image (REFERENCE IMAGE), and positions of potential defects are located by comparing and analyzing differences between the two images pixel by pixel. The reference image is typically derived from a normal region that is defect free under the same process conditions, such as a neighboring chip (Die-to-Die), to ensure that the two images have the same structure. Under ideal conditions, the reference image and the image to be detected have imaging brightness which is highly consistent, so that the difference result can accurately reflect structural change among the images, and the influence caused by brightness deviation is avoided. In the detection process, the image is usually obtained by a high-precision optical microscope or a linear array camera, but due to factors such as light source stability, reflection angle, exposure time and the like, a significant brightness difference exists between the reference image and the image to be detected, and the brightness difference can seriously interfere with the quality of a difference image (DIFFERENCE MAP), so that a large number of interference defects (Nuisance) are detected, and the identification of True defects (True defects) is affected. Conventional methods to solve the above problems, a histogram matching method based on a sliding window or a linear brightness matching method based on a global image is generally used to preprocess a reference image. However, these methods often have limitations, such as the linear luminance matching method is to adjust the luminance of the reference image by calculating the global average luminance ratio, so that the problem of uneven local luminance in the image cannot be solved. While the histogram matching method realizes brightness adjustment by matching the gray distribution, the method may destroy local contrast. Therefore, the method is not applicable to a high-precision detection scene of wafer detection. Disclosure of Invention In order to overcome the defects in the prior art, the present invention is directed to an inter-image brightness compensation method and a wafer defect detection system by using a grid square method, so as to solve the problems in the prior art. The inter-image brightness compensation method through the grid square method comprises the steps of S1, obtaining an image to be detected of target detection Die, adjacent Die reference images and corresponding binary mask images of a concerned area with the same resolution ratio as an input image, S2, dividing the input image into grid squares with fixed sizes according to grids based on the input image by adopting a grid method, obtaining a local grid square image and corresponding mask binary images, S3, obtaining brightness compensation kernels corresponding to the local grid square images based on the local grid square images and the mask binary images, S4, applying corresponding brightness compensation kernels to the local grid square images of each adjacent Die reference image based on the brightness compensation kernels, obtaining brightness compensated local grid square images, S5, splicing the adjacent Die reference images according to the sequence of the adjacent Die reference images based on the brightness compensated grid square images, and obtaining complete brightness compensated adjacent Die reference images. Further, the acquisition process of the local grid square image and the corresponding mask binary image comprises the steps of S21, acquiring the input minimum overlapped pixel size and grid square size, S22, calculating starting positions of all grid squares in the input image based on the width and height information of the input image, the minimum overlapped pixel size and the grid square size, S23, dividing the input image based on the starting positions, the grid square size, the image to be detected, the adjacent Die reference image and the attention area binary mask image, and acquiring the local grid square image. Further, when the edges of the grid square are insufficient to cover the wide-high area of the complete grid square, a boundary adjustment strategy is employed in which the square Patch of the last ro