Wafer color image consistency evaluation and defect detection method

Abstract

The invention discloses a wafer color image consistency evaluation and defect detection method. Firstly, the consistency of a wafer color image is evaluated from three dimensions of image quality, physical attributes and pixel difference, machine hardware is adjusted and monitored accordingly, the consistency and high quality of the wafer color image are guaranteed, difference is carried out on a wafer image to be detected and a standard template based on a crystal grain minimum repetitive unit during detection, and the detection accuracy is improved. And finally determining a defect area through weighted integration of multiple channels. The method can improve the accuracy of the defect detection effect and the consistency of results of different machines, is simple and feasible in implementation and calculation process, and can further meet the requirements of the semiconductor industry for automatic optical detection.

Description

Technical Field

[0001] Applied in the field of semiconductor automatic optical inspection, it provides a method for consistency evaluation and defect detection of wafer color images. Background Art

[0002] In the semiconductor industry, wafer manufacturing is extremely complex and sophisticated, typically involving hundreds or even thousands of process steps. This production process is inevitably subject to issues such as raw material contamination, environmental particulate matter, and process risks, leading to surface defects such as scratches, cracks, particles, and bubbles, which in turn impact production efficiency, product performance, and ultimately yield. The industry uses automated optical inspection technology to monitor macroscopic defects during wafer manufacturing. This involves capturing images of the wafer surface and identifying defects through image processing.

[0003] Due to considerations such as image resolution, scanning speed, and cost-effectiveness, the industrial vision field mostly uses grayscale cameras. However, to further observe defect morphology, identify defect types, and improve detection accuracy, wafer automated optical inspection is transitioning to three-dimensional color images. This also leads to factors such as the machine's lighting conditions, lens distortion, and shooting angle having a greater impact on image consistency. To eliminate hardware differences between machines and facilitate subsequent inspection, consistency evaluation and hardware parameter adjustment must be performed during the image acquisition phase to obtain consistent and high-quality wafer color images on the machine side. Image consistency evaluation generally begins by calculating information such as the image's pixel distribution or key features, and then quantifies the differences based on statistical metrics. Typical methods include histograms, mean square error, and feature matching. The histogram method directly compares the pixel histograms of two images, without considering the specific pixel distribution of the images and unable to capture local information of the images. The mean square error method directly calculates the square mean of the corresponding pixel differences. It focuses primarily on pixel-level differences, ignoring structural and semantic information, and is very sensitive to geometric transformations and noise in the images. The feature matching method evaluates image consistency by calculating the number and similarity of feature points between images. It can overcome the shortcomings of the first two methods to a certain extent, but the computational complexity is high, and the results are often different from visual intuition. Commonly used image processing methods can be divided into two categories: traditional methods and deep learning. The latter has developed rapidly in recent years, with processing effects improving by leaps and bounds, but it requires a lot of manual annotation and computing resources, and has difficulty in covering small samples and unknown defects. Its practicality in this field is still insufficient. Based on versatility considerations, defects are currently mainly identified by comparing the image to be tested with a template or between different units. Summary of the Invention

[0004] In response to the problem that traditional grayscale cameras cannot accurately identify wafer defects and hardware differences lead to different results between machines, the present invention proposes a method for wafer color image consistency evaluation and defect detection. First, the consistency of the wafer color image is evaluated from three dimensions, including image quality, physical properties, and pixel differences. Based on this, the camera parameters and machine hardware conditions such as illumination are monitored and adjusted to ensure the high quality of the wafer color image. Then, with the grain as the basic unit, the standard template is extracted and the three channels are differentiated respectively to preliminarily obtain the difference information. Then, through weighted integration of each channel, the defect area is finally determined. This method can improve the accuracy of defect detection and the consistency of results from different machines, and the implementation and calculation process are simple and easy, which can further meet the semiconductor industry's needs for automatic optical inspection.

[0005] The technical solution adopted by the present invention is: a method for wafer color image consistency evaluation and defect detection, comprising the following steps:

[0006] Step 1: Combine various working conditions and measure and monitor the tool hardware through wafer color image consistency evaluation to obtain the hardware parameters related to tool consistency that need to be adjusted;

[0007] Step 2: For the batch of wafers to be tested, first create a standard template, then collect a color image of the wafer to be tested, and perform differential integration with the standard template to obtain the defect area.

[0008] The various working conditions include:

[0009] After the first device is installed and the hardware parameters are calibrated, a wafer sample image is collected as a reference image. The newly installed machine initially collects images of the same wafer sample and performs a consistency evaluation with the reference image, thereby adjusting the hardware parameters until the score meets the threshold requirements;

[0010] b. For the wafer samples to be tested or the images to be tested of the same batch of wafers collected regularly during the operation of the machine, their consistency is evaluated with other machines or historical images to monitor the machine hardware parameters. If the threshold is exceeded, the hardware parameters are readjusted.

[0011] Step 1 of the color image consistency evaluation includes:

[0012] 1) Calculate the physical properties of the wafer image, image quality clarity D, and pixel difference S for the image to be tested and the reference image. e ;

[0013] The physical properties include: center C(X,Y), radius R, roundness E and deflection angle θ;

[0014] 2) First calculate the image quality consistency score of the two images:

[0015] S q=α1·(D1-D2) 2 +α2·(P1-P2) 2 +α3·(T1-T2) 2 , where the weight α1+α2+α3=1, D, P, and T are the image quality clarity, signal-to-noise ratio, and structural similarity respectively;

[0016] Then calculate the physical property consistency score of the two images:

[0017] S p =β1·[(X1-X2) 2 +(Y1-Y2) 2 ]+β2·(R1-R2) 2 +β3·[(E1-E2) 2 +(E1+E2) / 2·(θ1-θ2) 2 ], where weight β1+β2+β3=1;

[0018] Finally, the mean square error method is used to calculate the difference S between the two images e , set the weight γ1+γ2+γ3=1, and calculate the final consistency score: S=γ1S q +γ2S p +γ3S e ;

[0019] 3) If S q If the value is greater than the set threshold, the camera focus needs to be adjusted to improve image clarity and the camera flat field correction needs to be performed to improve the signal-to-noise ratio.

[0020] If S p If the value is greater than the set threshold, the imaging distance and camera pitch angle need to be adjusted to improve the image size and roundness;

[0021] If S q and S p If the conditions are met and the final S score is greater than the set threshold, the camera optical axis direction needs to be adjusted to improve the consistency between units.

[0022] The method for calculating the physical properties of the image includes:

[0023] Image preprocessing: converting the acquired color image into a grayscale image and extracting the wafer area using the Blob algorithm;

[0024] Create N measuring calipers evenly around the wafer circumference, perform edge detection on the caliper area to obtain N edge points, and fit a circle using the least squares method and RANSAC algorithm to obtain the center C(X,Y) and radius R.

[0025] The wafer contour is obtained by fitting N edge points, and the integral of the area between the wafer contour and the fitting circle is calculated as the roundness E, and the integral of the product of the area between the wafer contour and the fitting circle and the angle between them is calculated as the deflection angle θ;

[0026] The image quality calculation method includes:

[0027] Image preprocessing: converting the acquired color image into a grayscale image, extracting the wafer area using the Blob algorithm, and calculating the width W and height H of the wafer area;

[0028] The Sobel operator is used to calculate the horizontal and vertical image gradient G of each point in the wafer area. x (x,y) and G y (x,y), calculate the comprehensive image gradient Calculating image clarity

[0029] Calculate the signal-to-noise ratio of the entire image Where k is the scaling factor, MAX and MIN are the maximum and minimum values ​​of the local variance of all pixels in the image, respectively.

[0030] The pixel difference calculation method includes:

[0031] Image preprocessing: convert the acquired color image into a grayscale image, extract the wafer area using the Blob algorithm, translate the center of mass to align it, and rotate the wafer notch to align it;

[0032] The three channels of the wafer area are differentially processed and combined: Where V1 and V2 are the pixel values ​​of a certain point in the two images respectively, and C represents the number of channels.

[0033] The defect detection includes:

[0034] 1) Fix the hardware parameters related to machine consistency, adjust the illumination, collect color images of defect-free wafer samples, pre-process and segment the smallest units, and fuse them to generate a standard template;

[0035] 2) Fix the hardware parameters related to machine consistency, adjust the illumination, collect color images of the wafer to be tested, pre-process and segment the smallest unit, and perform differential integration with the above template to obtain the defect area.

[0036] The adjusting illumination comprises:

[0037] Collect the wafer color image at the preset illumination, convert it into a grayscale image, and count the number of pixels N that are greater than the threshold grayscale. pixel , which is less than the empirical threshold T min Then increase the illumination k1(T min -N pixel), greater than the empirical threshold T max Then reduce the illumination k2(N pixel -T max ), where k1 and k2 are empirical coefficients;

[0038] After adjusting the illumination, collect the wafer image again and repeat the above steps until T min <N pixel <T max If the above standards are not met before the given maximum number of adjustments, the illumination adjustment fails and the machine alarms.

[0039] The defect detection includes:

[0040] a) Image preprocessing: First, the acquired color image is converted into a grayscale image. The wafer area is extracted from the grayscale image as the ROI (region of interest) using the Blob algorithm.

[0041] b) converting the grayscale image of the above area into the frequency domain, analyzing the maximum frequency component, combining the wafer layout information to determine the size of the minimum repeating unit, using the sliding window method to identify the boundary of the repeating unit, and accurately segmenting the minimum unit of the wafer;

[0042] c) selecting a number of repeating units obtained by segmenting the color image of the defect-free wafer sample, each having different radii and positions covering the wafer;

[0043] Take the average width and height of the above units as the size of the template to be tested to avoid lens distortion;

[0044] d) unifying the smallest unit obtained by segmenting the color image of the wafer to be tested to the size of the template to be tested by bilinear interpolation and shape correction, thereby obtaining several units to be tested of different sizes;

[0045] e) Differentiate the three channels of the unit under test from the standard template, mark the areas where the difference and area are greater than the threshold, calculate the coordinates of the above areas in the wafer image, and output the image and coordinates of the area as the defect detection result.

[0046] The standard template is a series of standard units of different sizes obtained by pre-collecting defect-free wafers through steps a) to d).

[0047] The present invention has the following beneficial effects and advantages:

[0048] In response to the problem that traditional grayscale cameras cannot accurately identify wafer defects and hardware differences lead to different results between machines, the present invention proposes a method for wafer color image consistency evaluation and defect detection. First, the consistency of the wafer color image is evaluated from three dimensions, including image quality, physical properties, and pixel differences. Based on this, the camera parameters and machine hardware conditions such as illumination are monitored and adjusted to ensure the high quality of the wafer color image. Then, with the grain as the basic unit, the standard template is extracted and the three channels are differentiated respectively to preliminarily obtain the difference information. Then, through weighted integration of each channel, the defect area is finally determined. This method can improve the accuracy of defect detection and the consistency of results from different machines, and the implementation and calculation process are simple and easy, which can further meet the semiconductor industry's needs for automatic optical inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the hardware platform of the present invention.

[0050] Figure 2 This is a flow chart of image consistency evaluation of the present invention.

[0051] Figure 3 This is a defect detection flow chart of the present invention.

[0052] Figure 4 This is a comparison diagram of the template wafer and the test wafer of the present invention.

[0053] Figure 5 Schematic diagram of detecting local defects of wafers according to the present invention.

[0054] In the figure, 1 is the wafer station, 2 is the two-dimensional slide module, 3 is the reflector, 4 is the line scan camera, 5 is the light source, 6 is the wafer to be measured, A is the camera pitch angle, B is the optical axis direction, and d is the imaging distance. DETAILED DESCRIPTION

[0055] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, the specific implementation methods of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth many specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the invention. Therefore, the present invention is not limited to the specific implementation methods disclosed below.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art of the art to which the present invention pertains. The terms used in the specification of the invention herein are for the purpose of describing specific embodiments only and are not intended to limit the present invention.

[0057] like Figure 1As shown in FIG, it is a schematic diagram of the hardware platform of the present invention, which includes a wafer station 1, a two-dimensional slide module 2, a reflector 3, a line scan camera 4, and a light source 5.

[0058] The wafer station 1 is used to place the wafer 6 to be tested, and is installed on a two-dimensional slide module. The wafer station moves in a plane on the two-dimensional slide module 2, changing the local viewing area of ​​the wafer entering the camera, so that the line scan camera can scan the entire area of ​​the wafer to be tested; finally, it is spliced ​​into a complete wafer image.

[0059] Reflector 3, placed above the wafer to be measured, is used to shorten the optical path and project the wafer image onto the line scan camera;

[0060] The line scan camera 4 is used to capture a partial projection image of the wafer to be tested and output a complete wafer image.

[0061] Light source 5 is placed above the wafer to be tested to provide suitable illumination; the white strip light source is used to ensure uniform imaging of the wafer surface.

[0062] A wafer color image consistency evaluation and defect detection method of the present invention includes:

[0063] Step 1: Measure and monitor the tool hardware through wafer color image consistency evaluation, including the matching between tools (ensuring high consistency in parameters such as size, center of mass, roundness, and clarity of wafer images captured by different tools) and the stability of the tools themselves (ensuring uniformity of light source imaging and camera acquisition frequency). Determine the hardware parameters related to tool consistency that need adjustment, including camera focal length f, camera flat field correction C, imaging distance d, camera pitch angle A, and optical axis direction B.

[0064] Step 2: For the batch of wafers to be tested, first create a standard template, then collect a color image of the wafer to be tested, perform differential integration with the standard template, and obtain the defect area.

[0065] Among them, in step one, for the first equipment to be installed and the hardware parameters to be calibrated, the wafer sample image is collected as the reference image, the newly installed machine initially collects the image of the same wafer sample and evaluates the consistency with the reference image, and adjusts the hardware parameters until the score meets the threshold requirement; for the operation of the machine, the wafer samples to be tested or the images to be tested of the same batch of wafers are regularly collected, and the consistency is evaluated with other machines or historical images, and the stability of the machine hardware is monitored. If the threshold is exceeded, the hardware parameters are readjusted.

[0066] Among them, such as Figure 2 As shown, the color image consistency evaluation in step 1 includes:

[0067] 1) For the collected images, calculate the image physical properties (center C(X,Y), radius R, roundness E and deflection angle θ), image quality clarity D, pixel difference S e ;

[0068] 1a) Methods for calculating image physical properties include:

[0069] Image preprocessing: converting the acquired color image into a grayscale image and extracting the wafer area using the Blob algorithm;

[0070] Create N measuring calipers evenly around the wafer circumference, perform edge detection on the caliper area to obtain N edge points, and fit a circle using the least squares method and RANSAC algorithm to obtain the center C(X,Y) and radius R.

[0071] The wafer contour is obtained by fitting N edge points, and the integral of the area between the wafer contour and the fitting circle is calculated as the roundness E, and the integral of the product of the area between the wafer contour and the fitting circle and the angle between them is calculated as the deflection angle θ;

[0072] 1b) The image quality calculation method includes:

[0073] Image preprocessing: converting the acquired color image into a grayscale image, extracting the wafer area using the Blob algorithm, and calculating the width W and height H of the wafer area;

[0074] The Sobel operator is used to calculate the horizontal and vertical image gradient G of each point in the wafer area. x (x,y) and G y (x,y), calculate the comprehensive image gradient Calculating image clarity

[0075] Calculate the signal-to-noise ratio of the entire image Where k is the scaling factor, MAX and MIN are the maximum and minimum values ​​of the local variance of all pixels in the image respectively;

[0076] 1c) Pixel difference calculation method includes:

[0077] Image preprocessing: convert the acquired color image into a grayscale image, extract the wafer area using the Blob algorithm, translate the center of mass to align it, and rotate the wafer notch to align it;

[0078] The three channels of the wafer area are differentially processed and combined: Where V1 and V2 are the pixel values ​​of a certain point in the two images respectively, and C represents the number of channels;

[0079] 2) First calculate the image quality consistency score of the two images: S q =α1·(D1-D2)2 +α2·(P1-P2) 2 +α3·(T1-T2) 2 , where the weight α1+α2+α3=1, D, P, and T are the image quality clarity, signal-to-noise ratio, and structural similarity, respectively.

[0080] Then calculate the physical property consistency score of the two images: S p =β1·[(X1-X2) 2 +(Y1-Y2) 2 ]+β2·(R1-R2) 2 +β3·[(E1-E2) 2 +(E1+E2) / 2·(θ1-θ2) 2 ], where weight β1+β2+β3=1.

[0081] Finally, the mean square error method is used to calculate the difference S between the two images e , set the weight γ1+γ2+γ3=1, and calculate the final consistency score: S=γ1S q +γ2S p +γ3S e ;

[0082] 3) If S q If the value is greater than the set threshold, the camera focal length f needs to be adjusted to improve image clarity and the camera flat field correction C needs to be adjusted to improve the signal-to-noise ratio. The camera focal length f and the camera flat field correction C are internal parameters of the camera.

[0083] If S p If the image size is greater than the set threshold, the imaging distance d and the camera pitch angle A need to be adjusted to improve the image size and roundness.

[0084] If S q and S p If the conditions are met and the final S score is greater than the set threshold, the camera optical axis direction B needs to be adjusted to improve the consistency between units.

[0085] 4) After adjusting the above parameters, you also need to adjust the illumination parameters:

[0086] Fix other hardware parameters of the machine, adjust illumination, collect color images of defect-free wafer samples, pre-process and segment the smallest units, and fuse them to generate a standard template;

[0087] Fix other hardware parameters of the machine, adjust the illumination, collect color images of the wafer to be tested, pre-process and segment the smallest unit, and perform differential integration with the above template to obtain the defect area.

[0088] Collect the wafer color image at the preset illumination, convert it into a grayscale image, and count the number of pixels N that are greater than the threshold grayscale. pixel, which is less than the empirical threshold T min Then increase the illumination k1(T min -N pixel ), greater than the empirical threshold T max Then reduce the illumination k2(N pixel -T max ), where k1 and k2 are empirical coefficients;

[0089] After adjusting the illumination, collect the wafer image again and repeat the above steps of counting grayscale pixels until T min <N pixel <T max If the above standards are not met before the given maximum number of adjustments, the illumination adjustment fails and the machine alarms.

[0090] Among them, such as Figure 3 As shown, the defect detection in step 2 includes:

[0091] 1) Standard template production process:

[0092] 1a) Image preprocessing: First, the acquired color image is converted into a grayscale image. The wafer area is extracted from the grayscale image as the ROI (region of interest) using the Blob algorithm.

[0093] 1b) Convert the grayscale image of the above area to the frequency domain, analyze the maximum frequency component, and combine the wafer layout information (for example, rectangular patterns of different sizes on the wafer are repeating units) to determine the size of the minimum repeating unit. Use the sliding window method to identify the boundaries of the repeating unit and accurately segment the wafer into the smallest unit.

[0094] 1c) Selecting a number of repeating units obtained by segmenting the color image of the defect-free wafer sample, so as to cover as many different radii and positions of the wafer as possible;

[0095] Due to factors such as lens distortion, the above-mentioned unit sizes may be inconsistent and not rectangular. The average width and height of the above-mentioned units are taken as the size of the standard template;

[0096] 1d) The above units are unified into the size of a standard template by bilinear interpolation and shape correction to obtain several standard units of different sizes, and the above standard units are pixel-fused to obtain a standard template.

[0097] 2) Actual testing process:

[0098] 2a) Image preprocessing: First, the acquired color image is converted into a grayscale image. The wafer area is extracted from the grayscale image as the ROI (region of interest) using the Blob algorithm.

[0099] 2b) Convert the grayscale image of the above area to the frequency domain, analyze the maximum frequency component, and combine the wafer layout information (for example, rectangular patterns of different sizes on the wafer are repeating units) to determine the size of the minimum repeating unit. Use the sliding window method to identify the boundaries of the repeating unit and accurately segment the wafer into the smallest unit.

[0100] 2c) selecting a number of repeating units obtained by segmenting the color image of the defect-free wafer sample, so as to cover as many different radii and positions of the wafer as possible;

[0101] Due to factors such as lens distortion, the above-mentioned unit sizes may be inconsistent and not rectangular. The average width and height of the above-mentioned units are taken as the size of the standard template;

[0102] 2d) unifying the smallest unit obtained by segmenting the color image of the wafer to be measured to the size of a standard template through bilinear interpolation and shape correction, thereby obtaining several standard units of different sizes;

[0103] 2e) Differentiate the three channels of the standard cell with the standard template, mark the areas where the difference and area are greater than the threshold, calculate the coordinates of the above areas in the wafer image, and output the image and coordinates of the area as the defect detection result.

[0104] like Figure 4 As shown in the figure, (a) is the standard template wafer image, (b) is the wafer image to be tested, and the wafer to be tested is printed with a size of 100*100um 2 、200*200um 2 、300*300um 2 、400x400um 2 、500x500um 2 、650x650um 2 、800x800um 2 、1000x1000um 2 、1250x1250um 2 、1500x1500um 2 and 2000x2000um 2 defects.

[0105] like Figure 5 As shown in the figure, (a) is a partial magnified view of a standard template wafer and (b) is a magnified effect view of a local defect detection of a wafer to be detected using the method of the present invention. As can be seen from the figure, the defect is larger than 200*200um. 2 All defects are detected.

[0106] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should be regarded as within the scope of protection of the present invention.

Claims

1. A wafer color image consistency evaluation and defect detection method, characterized in that: The following steps are involved: Step 1: Combine various working conditions and measure and monitor the tool hardware through wafer color image consistency evaluation to obtain the hardware parameters related to tool consistency that need to be adjusted; Step 2: For the batch of wafers to be tested, first create a standard template, then collect a color image of the wafer to be tested, and perform differential integration with the standard template to obtain the defect area.

2. The method for wafer color image consistency evaluation and defect detection according to claim 1, characterized in that: Various working conditions include: a. After the first equipment is installed and the hardware parameters are calibrated, wafer sample images are collected as reference images. The newly installed machine initially collects images of the same wafer sample and evaluates the consistency with the reference image, thereby adjusting the hardware parameters until the score meets the threshold requirements; b. For the wafer samples to be tested or the images to be tested of the same batch of wafers collected regularly during the operation of the machine, they are evaluated for consistency with other machines or historical images, thereby monitoring the machine hardware parameters. If the threshold is exceeded, the hardware parameters are readjusted.

3. The method for wafer color image consistency evaluation and defect detection according to claim 1, wherein: Step 1 of the color image consistency evaluation includes: 1) Calculate the physical properties of the wafer image, image quality clarity D, and pixel difference S for the image to be tested and the reference image. e The physical properties include: center C(X,Y), radius R, roundness E and deflection angle θ; 2) First calculate the image quality consistency score of the two images: S q =α1·(D1-D2) 2 +α2·(P1-P2) 2 +α3·(T1-T2) 2 , where the weight α1+α2+α3=1, D, P, and T are the image quality clarity, signal-to-noise ratio, and structural similarity respectively; Then calculate the physical property consistency score of the two images: S p =β1·[(X1-X2) 2 +(Y1-Y2) 2 ]+β2·(R1-R2) 2 +β3·[(E1-E2) 2 +(E1+E2) / 2·(θ1-θ2) 2 ], where weight β1+β2+γ3=1; Finally, the mean square error method is used to calculate the difference S between the two images e , set the weight γ1+γ2+γ3=1, and calculate the final consistency score: S=γ1S q +γ2S p +γ3S e ; 3) If S q If the value is greater than the set threshold, the camera focus needs to be adjusted to improve image clarity and the camera flat field correction needs to be performed to improve the signal-to-noise ratio. If S p If the value is greater than the set threshold, the imaging distance and camera pitch angle need to be adjusted to improve the image size and roundness; If S q and S p If the conditions are met and the final S score is greater than the set threshold, the camera optical axis direction needs to be adjusted to improve the consistency between units.

4. The method for wafer color image consistency evaluation and defect detection according to claim 3, wherein: The method for calculating the physical properties of the image includes: Image preprocessing: converting the acquired color image into a grayscale image and extracting the wafer area using the Blob algorithm; Create N measuring calipers evenly around the wafer circumference, perform edge detection on the caliper area to obtain N edge points, and fit a circle using the least squares method and RANSAC algorithm to obtain the center C(X,Y) and radius R. The wafer contour is obtained by fitting N edge points, and the area integral between the wafer contour and the fitting circle is calculated as the roundness E. The integral of the area between the wafer contour and the fitting circle and the angle between them is calculated as the deflection angle θ.

5. The method for wafer color image consistency evaluation and defect detection according to claim 3, wherein: The image quality calculation method includes: Image preprocessing: converting the acquired color image into a grayscale image, extracting the wafer area using the Blob algorithm, and calculating the width W and height H of the wafer area; The Sobel operator is used to calculate the horizontal and vertical image gradient G of each point in the wafer area. x (x,y) and G y (x,y), calculate the comprehensive image gradient Calculate image clarity Calculate the signal-to-noise ratio of the entire image Where k is the scaling factor, MAX and MIN are the maximum and minimum values ​​of the local variance of all pixels in the image, respectively.

6. The method for wafer color image consistency evaluation and defect detection according to claim 3, wherein: The pixel difference calculation method includes: Image preprocessing: convert the acquired color image into a grayscale image, extract the wafer area using the Blob algorithm, translate the center of mass to align it, and rotate the wafer notch to align it; The three channels of the wafer area are differentially processed and combined: Where V1 and V2 are the pixel values ​​of a certain point in the two images respectively, and C represents the number of channels.

7. The method for wafer color image consistency evaluation and defect detection according to claim 1, wherein: The defect detection includes: 1) Fix the hardware parameters related to machine consistency, adjust the illumination, collect color images of defect-free wafer samples, pre-process and segment the smallest units, and fuse them to generate a standard template; 2) Fix the hardware parameters related to machine consistency, adjust the illumination, collect color images of the wafer to be tested, pre-process and segment the smallest unit, and perform differential integration with the above template to obtain the defect area.

8. The method for wafer color image consistency evaluation and defect detection according to claim 7, wherein: The adjusting illumination comprises: Collect the wafer color image at the preset illumination, convert it into a grayscale image, and count the number of pixels N that are greater than the threshold grayscale. pixel , which is less than the empirical threshold T min Then increase the illumination k1(T min -N pixel ), greater than the empirical threshold T max Then reduce the illumination k2(N pixel -T max ), where k1 and k2 are empirical coefficients; After adjusting the illumination, collect the wafer image again and repeat the above steps until T min <N pixel <T max If the above standards are not met before the given maximum number of adjustments, the illumination adjustment fails and the machine alarms.

9. The method for wafer color image consistency evaluation and defect detection according to claim 7, wherein: The defect detection includes: a) Image preprocessing: First, the acquired color image is converted into a grayscale image. The wafer area is extracted from the grayscale image as the ROI (region of interest) using the Blob algorithm. b) converting the grayscale image of the above area into the frequency domain, analyzing the maximum frequency component, combining the wafer layout information to determine the size of the minimum repeating unit, using the sliding window method to identify the boundary of the repeating unit, and accurately segmenting the minimum unit of the wafer; c) selecting a number of repeating units obtained by segmenting the color image of the defect-free wafer sample, each having different radii and positions covering the wafer; Take the average width and height of the above units as the size of the template to be tested to avoid lens distortion; d) unifying the smallest unit obtained by segmenting the color image of the wafer to be tested to the size of the template to be tested by bilinear interpolation and shape correction, thereby obtaining several units to be tested of different sizes; e) Differentiate the three channels of the unit under test from the standard template, mark the areas where the difference and area are greater than the threshold, calculate the coordinates of the above areas in the wafer image, and output the image and coordinates of the area as the defect detection result.

10. The method for wafer color image consistency evaluation and defect detection according to claim 7, wherein: The standard template is a series of standard units of different sizes obtained by pre-collecting defect-free wafers through steps a) to d).

Images