Defect detection method for reticle mask

Abstract

The invention discloses a defect detection method for a reticle mask, which relates to the technical field of mask defect detection and comprises the following steps of: automatically calling and solidifying detection parameters according to an instruction multiplying power and a mask name; performing quadratic polynomial fitting normalization correction on the reference image according to the gray level of the light calibration point; dividing a detection area, extracting ORB features, and obtaining sub-pixel homonymy point pairs through Hamming distance and RANSAC; and performing differential defect detection after solving an affine matrix to precisely align the Die image. According to the method, manual re-judgment can be replaced, the detection time is shortened to the second level, the defect recognition accuracy is larger than or equal to 99%, and the efficient and accurate requirements of 28 nm nodes are met.

Description

Technical Field

[0001] The present invention relates to the technical field of mask defect detection, and in particular to a defect detection method for a reticle mask. Background Art

[0002] In semiconductor lithography, the integrity of the reticle pattern directly determines wafer yield. The industry generally uses automated mask inspection systems (such as the KLA-Tencor series) that integrate high-resolution optical imaging, image processing, and defect classification systems to perform 100% mask scanning. This type of equipment typically offers two algorithm modes: Die-to-Database (D2DB): compares the die to be tested with the layout database pixel by pixel; Die-to-Die (D2D): Performs layout-free comparison between adjacent dies, suitable for repetitive graphics scenarios.

[0003] Although existing technologies have achieved high-throughput image acquisition, the "secondary review" and "fine classification" of defects still rely heavily on manual labor: The inspection system only outputs the coordinates of suspected defects and a grayscale difference map. The operator needs to confirm the authenticity of each defect one by one under a high-power microscope and manually mark the type (opaque spot, pindot, pin hole, black convex / white convex, break / short circuit, thinline, particle, contact hole abnormality, etc.).

[0004] Manual judgment is easily affected by subjective experience, visual fatigue, and differences in ambient lighting, resulting in an error rate of >15% and a re-judgment time of >30 minutes for a single mask. This cannot meet the dual requirements of "high precision + high productivity" for advanced nodes of 28 nm and below.

[0005] In recent years, deep learning-based defect classification algorithms have been introduced, but existing solutions remain in the single-point application of "offline training-online inference": Training data requires a lot of manual labeling, which is costly; The reasoning model is separated from the D2D / D2DB comparison algorithm, making it impossible to utilize the difference features generated during the comparison phase, resulting in information redundancy. It is sensitive to sub-pixel positioning errors and lighting differences, and is prone to false defects. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to address the deficiencies of the prior art and specifically provide a defect detection method for a reticle mask, which is as follows: 1) In a first aspect, the present invention provides a defect detection method for a reticle mask, the specific technical solution of which is as follows: Adjusting detection parameters of a device used to perform defect detection on the mask to be tested based on the detection magnification information in the detection instruction and the name of the mask to be tested, and storing the detection parameters in association with the name of the mask to be tested; Based on the grayscale values ​​of the light calibration points in the reference image corresponding to the mask to be tested, the reference image is corrected for light source brightness, and a corrected reference image is generated; The corrected reference image is divided into detection areas, and the ORB feature points corresponding to the test die image and the reference ORB feature points corresponding to the reference die image are extracted from the division results. The Hamming distance matching and RANSAC are used to eliminate false matches and establish sub-pixel homonymous point pairs. Combined with sub-pixel same-name point pairs, the conversion matrix for converting the die image to be tested into the reference die image is determined. Based on the conversion matrix, the die image to be tested and the reference die image are aligned, and defect detection is performed based on the alignment results.

[0007] The beneficial effects of the defect detection method for reticle masks provided by the present invention are as follows: By associating and storing detection parameters with mask template names, one-click quick switching of detection configurations for different process nodes is achieved, improving equipment versatility and line change efficiency; the grayscale value of the light calibration point is used to correct the brightness of the light source, effectively eliminating the grayscale deviation caused by uneven lighting and improving the accuracy of subsequent comparisons; ORB+RANSAC is used to construct sub-pixel same-name point pairs and generate a transformation matrix to achieve high-precision geometric alignment of the die to be tested and the reference die, significantly reducing the false defect rate caused by position errors; the entire process is automated to complete parameter adjustment, brightness correction, area division, feature matching and defect detection, completely replacing manual re-judgment, shortening the detection time from minutes to seconds, and increasing the defect recognition accuracy to more than 99%, meeting the accuracy and production capacity requirements of mask defect detection at advanced nodes of 28 nm and below.

[0008] Based on the above solution, the present invention can also be improved as follows.

[0009] Furthermore, the process of determining the light source brightness correction for the reference image is as follows: Multiple light calibration points are randomly selected at the junction of the transparent area and the non-transparent area in the reference image, and the grayscale values ​​corresponding to the multiple light calibration points are obtained. Then, a light distribution model is constructed through quadratic polynomial fitting. The reference image is normalized using the light distribution model to complete the light source brightness correction.

[0010] The beneficial effects of the above further scheme are: By randomly sampling multiple light calibration points at the junction of transparent and non-transparent areas and using quadratic polynomial fitting to construct an illumination distribution model, pixel-level normalization of the entire reference image is achieved, effectively eliminating the grayscale gradient error caused by uneven light sources in different areas of the mask, thereby reducing the illumination difference noise in the subsequent image comparison stage by more than two orders of magnitude, significantly reducing the number of false defects, and improving the grayscale contrast of real defects, thereby simultaneously improving the reliability and detection rate of sub-pixel defect detection.

[0011] Furthermore, the process of establishing sub-pixel homonymous point pairs is as follows: The feature point pairs consisting of the ORB feature points to be tested and the reference ORB feature points are roughly matched by Hamming distance matching, and the feature point pairs that meet the preset requirements are finely matched by eliminating false matches through RANSAC to obtain sub-pixel homonymous point pairs.

[0012] The beneficial effects of the above further scheme are: After completing the rough matching through Hamming distance, RANSAC is then used to iteratively eliminate false matches, which can obtain high-precision, evenly distributed sub-pixel homonymous point pairs in one go. This reduces the error in the subsequent affine matrix solution to within 0.05 pixel, thereby reducing the false defect rate caused by feature point drift by more than 90%, while ensuring that real defects can still be stably detected under the most stringent 28nm node.

[0013] Furthermore, the specific process of determining defect detection based on the alignment processing result is as follows: Perform differential processing on the aligned die image to be tested and the reference die image, and extract the suspected defect area through binary segmentation; After pre-processing the suspected defect area, feature analysis is performed using a preset analysis algorithm to determine the final defect result.

[0014] The beneficial effects of the above further scheme are: Through differential and binary segmentation, suspected defect areas are quickly identified, and morphological preprocessing is then used to remove background noise, so that subsequent feature analysis only needs to be performed within a range of 2%-5% of pixels, reducing the overall detection time by 70%. At the same time, a preset algorithm is used to control multi-dimensional features such as area, width, height, and grayscale mean, reducing the false defect rate to below 0.3%, ensuring 100% detection and automatic classification of real defects ≥3 pixels on the 28nm node mask.

[0015] Furthermore, it also includes: Based on the feature identifier of the mask to be tested, the offset between the mask to be tested and the defect detection equipment is calculated, and the position of the mask to be tested is corrected based on the offset.

[0016] The beneficial effects of the above further scheme are: By using mask feature representations to calculate and compensate for translational / rotational offsets between the device and the mask in real time, the initial error of subsequent ORB matching and affine alignment is reduced by an order of magnitude, thereby improving the overall positioning accuracy to the sub-pixel level, significantly reducing the number of false defects caused by platform drift, and directly shortening the entire process inspection time by approximately 20%.

[0017] 2) In a second aspect, the present invention further provides a defect detection system for a reticle mask, the specific technical solution of which is as follows: The initialization module is used to adjust the detection parameters of the equipment used for defect detection of the mask to be tested based on the detection magnification information and the name of the mask to be tested in the detection instruction, and store the detection parameters in association with the name of the mask to be tested; The light source brightness correction module is used to: perform light source brightness correction on the reference image based on the grayscale value of the light calibration point in the reference image corresponding to the mask to be tested, and generate a corrected reference image; The detection area determination module is used to divide the detection area of ​​the rectified reference image, extract the ORB feature points of the test die image and the reference ORB feature points of the reference die image, eliminate false matches by using Hamming distance matching and RANSAC, and establish sub-pixel homonymous point pairs. The detection module is used to: determine the conversion matrix for converting the die image to be tested into the reference die image by combining sub-pixel-level same-name point pairs, align the die image to be tested with the reference die image based on the conversion matrix, and perform defect detection based on the alignment results.

[0018] Based on the above solution, the present invention can also be improved as follows.

[0019] Furthermore, the process of determining the light source brightness correction for the reference image is as follows: Multiple light calibration points are randomly selected at the junction of the transparent area and the non-transparent area in the reference image, and the grayscale values ​​corresponding to the multiple light calibration points are obtained. Then, a light distribution model is constructed through quadratic polynomial fitting. The reference image is normalized using the light distribution model to complete the light source brightness correction.

[0020] Furthermore, the process of establishing sub-pixel homonymous point pairs is as follows: The feature point pairs consisting of the ORB feature points to be tested and the reference ORB feature points are roughly matched by Hamming distance matching, and the feature point pairs that meet the preset requirements are finely matched by eliminating false matches through RANSAC to obtain sub-pixel homonymous point pairs.

[0021] Furthermore, the specific process of determining defect detection based on the alignment processing result is as follows: Perform differential processing on the aligned die image to be tested and the reference die image, and extract the suspected defect area through binary segmentation; After pre-processing the suspected defect area, feature analysis is performed using a preset analysis algorithm to determine the final defect result.

[0022] Furthermore, it also includes: The correction module is used to calculate the offset between the mask to be tested and the defect detection equipment based on the feature identifier of the mask to be tested, and perform position correction on the mask to be tested based on the offset.

[0023] 3) In a third aspect, the present invention further provides an electronic device, comprising a processor coupled to a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor so that the electronic device implements any of the above methods.

[0024] 4) In a fourth aspect, the present invention further provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor to enable a computer to implement any of the above methods.

[0025] It should be noted that the beneficial effects achieved by the technical solutions of the second to fourth aspects of the present invention and the corresponding possible implementation methods can be found in the above-mentioned technical effects of the first aspect and its corresponding possible implementation methods, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 This is a flow chart of a defect detection method for a reticle mask according to an embodiment of the present invention; Figure 2 This is a second flow chart of a defect detection method for a reticle mask according to an embodiment of the present invention; Figure 3 1 is a flow chart of light source brightness correction in a defect detection method for a reticle mask according to an embodiment of the present invention; Figure 4 Schematic diagram of a detection area of ​​a defect detection method for a reticle mask according to an embodiment of the present invention; Figure 5 Schematic diagram of defect detection results of a defect detection method for a reticle mask according to an embodiment of the present invention; Figure 6 This is a structural framework diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0028] like Figure 1 As shown, a defect detection method for a reticle mask according to an embodiment of the present invention includes the following steps: S1, adjusting detection parameters of a device used to perform defect detection on the mask to be tested based on the detection magnification information in the detection instruction and the name of the mask to be tested, and storing the detection parameters in association with the name of the mask to be tested; S2, based on the grayscale values ​​of the light calibration points in the reference image corresponding to the mask to be tested, performing light source brightness correction on the reference image and generating a corrected reference image; S3, divide the detection area of ​​the corrected reference image, extract the ORB feature points of the test die image and the reference ORB feature points of the reference die image in the division result, use Hamming distance matching and RANSAC to eliminate false matches, and establish sub-pixel homonymous point pairs; S4, combining the sub-pixel level same-name point pairs, determines the conversion matrix for converting the die image to be tested into the reference die image, aligns the die image to be tested with the reference die image based on the conversion matrix, and performs defect detection based on the alignment result.

[0029] The beneficial effects of the defect detection method for reticle masks provided by the present invention are as follows: By associating and storing detection parameters with mask template names, one-click quick switching of detection configurations for different process nodes is achieved, improving equipment versatility and line change efficiency; the grayscale value of the light calibration point is used to correct the brightness of the light source, effectively eliminating the grayscale deviation caused by uneven lighting and improving the accuracy of subsequent comparisons; ORB+RANSAC is used to construct sub-pixel same-name point pairs and generate a transformation matrix to achieve high-precision geometric alignment of the die to be tested and the reference die, significantly reducing the false defect rate caused by position errors; the entire process is automated to complete parameter adjustment, brightness correction, area division, feature matching and defect detection, completely replacing manual re-judgment, shortening the detection time from minutes to seconds, and increasing the defect recognition accuracy to more than 99%, meeting the accuracy and production capacity requirements of mask defect detection at advanced nodes of 28 nm and below.

[0030] It's important to note that the "reticle under test" refers to the high-precision quartz or glass template used in the photolithography process to transfer chip patterns onto silicon wafers. In this scenario, it specifically refers to the mask currently undergoing defect inspection. It's covered with repetitively arranged die cells, and the goal of inspection is to identify any pattern defects or contamination within these cells.

[0031] In another embodiment of the present solution, the process of adjusting the detection parameters of the device includes: Step 1: Issue the test instructions The software layer sends the detection magnification (such as ×150, ×200) and the name of the mask to be tested (such as "N28-LV-001") as keywords to the machine control unit.

[0032] Step 2: Recipe library matching The machine searches the local recipe database for an entry with the same name based on the mask template name; if it does not exist, it selects a default template based on the inspection magnification.

[0033] The recipe entry contains all the parameters required for the corresponding magnification: Optical magnification and numerical aperture (NA); light source power and exposure time; image acquisition resolution (pixels / μm); defect size conversion factor; grayscale threshold, card control index, ORB feature extraction threshold and other algorithm parameters Step 3: Closed-loop calibration of machine hardware Magnification switching: Drive the motor to adjust the objective lens group so that the error between the actual magnification and the recipe setting value is less than 0.1%.

[0034] Light source power adjustment: Using real-time feedback from an optical power meter, the LED current is adjusted in a closed loop to ensure that the grayscale value at the calibration point falls within the 45%–55% dynamic range specified in the recipe.

[0035] Pixel size calibration: Use the standard grid plate to calibrate the pixel size at the current magnification. The defect size conversion factor is updated by real-time calibration based on the actual size.

[0036] Reset the platform movement resolution: Recalculate the stepper motor pulse equivalent according to the magnification to ensure that the scanning step size matches the recipe.

[0037] Step 4: Parameter Solidification and Associative Storage All of the above real-time calibration results are written back to the recipe entry and stored in the machine log along with the mask template name and detection timestamp, enabling one-click reproduction of the "name-parameter" relationship.

[0038] In another embodiment of this solution, before correcting the brightness of the light source, the mask to be tested is further corrected, specifically: After successful initialization, the algorithm immediately loads the inspection magnification, product name, and ROI parameters issued by the software into memory. It then reads the reference template image corresponding to the product name from the template library and extracts ORB feature points from its four corners and center. Simultaneously, the newly acquired original image to be tested is transferred to the memory buffer. The algorithm performs ORB descriptor matching on the two images, first using Hamming distance brute force matching to obtain coarse matching point pairs. RANSAC is then used to iteratively eliminate false matches, retaining ≥8 pairs of the most confident same-name points. Based on these same-name points, the translation vector (dx, dy) and rotation angle θ of the image to be tested relative to the reference template are calculated to generate a 2×3 affine matrix. This matrix is ​​used by the control system to drive a high-precision XYθ stage for real-time compensation. Once the stage is in place, the image is acquired again and the matching-calculation-compensation closed loop is repeated until (dx, dy) is less than 0.05 pixels and θ is less than 0.002 rad, confirming that the product has achieved sub-pixel positioning.

[0039] In another embodiment of the present invention, the reference image corresponding to the mask to be measured refers to an image of the mask to be measured recognized by the device. Based on the above content, the process of light source brightness correction is specifically as follows: Read the entire mask image that has been turned and aligned with coordinates. First, traverse all the horizontal transition edges between the transparent area (white high grayscale area) and the non-transparent area (black low grayscale area) of the image, randomly pick up no less than 16 edge points as candidate light calibration points, and record their (x, y) coordinates and corresponding grayscale g; perform quadratic polynomial fitting on these (g, x, y) triplets to obtain a two-dimensional illumination distribution model I(x, y) = ax² + by² + cxy + dx + ey + f, where the coefficients are solved by the least squares method; use this model to calculate the theoretical grayscale value of each pixel in the entire image, and compare it with the measured grayscale to obtain a normalized coefficient matrix; finally, multiply the original image pixel by pixel by the coefficient matrix to complete brightness normalization, output the corrected reference image, and attach a light distribution map of the same size for subsequent threshold segmentation call, at this point the light source brightness correction is complete.

[0040] In another embodiment of this solution, the process of dividing the corrected reference image into regions is specifically as follows: The normalized reference image is read into memory. The geometric coordinates of the pattern layer, cutting path, alignment mark and blank area are first parsed using the preset mask layout rule file. The image is binarized into pattern area and non-pattern area through threshold segmentation. Morphological opening and closing operations are then used to remove isolated noise points and close small gaps. The entire image is divided into three types of areas according to the rules: the pattern defect detection area is formed by expanding the boundaries of all valid chip patterns by 2μm; the contamination defect detection area covers the cutting path and blank area and expands by 1μm; the shielding area is directly taken from the alignment mark and known non-functional area. Then, a corresponding binary mask is generated for each area, and the mask and the corrected image are synchronously sent to the next step of ORB feature extraction.

[0041] In another embodiment of this solution, the process of establishing sub-pixel homonymous point pairs is specifically as follows: ORB feature processing is performed synchronously in the die image to be tested and the corresponding reference die image: first, FAST corner detection is used to extract candidate key points in the mask intersection of the two types of areas, Harris response is calculated for each point and the top N strongest points are retained, the main direction is assigned using the grayscale centroid method, and a 256-bit rBRIEF descriptor is generated; after completion, the descriptor sets of the two images are sent to Hamming distance brute force matching, and the ones with the ratio of the nearest neighbor to the next nearest neighbor distance <0.8 are retained as coarse matches. RANSAC is iterated 1000 times, and 4 pairs of points are randomly selected each time to estimate the homography matrix. Point pairs with reprojection error <1 pixel are judged as inliers, and finally the largest inlier point set is retained as sub-pixel homonymous point pairs.

[0042] In another embodiment of this solution, the process of determining the conversion matrix includes: Extract all sub-pixel homonymous point pairs. First, use least squares to solve the 2×3 affine matrix A=[a11 a12 tx;a21 a22 ty]. Use bilinear interpolation according to A to map the Die image to the reference Die coordinate system pixel by pixel. During interpolation, reversely calculate the floating-point coordinates (x, y) in the image to be tested for each target integer coordinate (x′, y′) and take the weighted average of the grayscale of its 2×2 neighborhood. After the mapping is completed, crop or pad with zeros according to the size of the reference image to ensure that the width and height of the two images are exactly the same. Finally, calculate the mutual information or mean square error of the entire overlapping area as the alignment quality indicator. If the indicator is better than the set threshold, output the aligned image to be tested. Otherwise, return to the feature matching step to re-screen the point pairs and recalculate the matrix until the accuracy requirements are met.

[0043] In another embodiment of this solution, the process of aligning the die image to be tested with the reference die image includes: The aligned Die image to be tested I′test and the reference Die image Iref are subjected to absolute difference pixel by pixel to obtain the grayscale difference image. , according to the pattern area and the pollution area, the corresponding Otsu adaptive threshold T is called to Figure 2 The initial defect mask B is obtained by quantization. An opening operation (erode followed by dilation) is performed on B using a 3×3 structuring element to remove isolated noise points with an area less than 3 pixels, and a closing operation is used to bridge small breaks. The cleaned mask is then labeled with 8-connected domains. For each connected domain, the area A, the width and height (w, h) of the circumscribed rectangle, the average grayscale difference μ_D, the aspect ratio r, and the roundness c are extracted. Finally, these features are input into the multidimensional control thresholds of each defect type in the recipe for judgment. If a connected domain meets any of the real defect conditions, its centroid coordinates, defect type, size, and corresponding thumbnail are recorded to generate a defect list Dlist and directly enter the result merging and output steps.

[0044] In another embodiment of the present solution, the process of performing defect detection according to the alignment processing result includes: The defect list Dlist of the current die is converted to the global coordinates of the entire reticle according to the coordinate system. The Dlist of all detected dies on the same reticle is traversed. The defects in the overlapping area are merged and deduplicated based on IoU>0.5, and sorted by defect type, area, and confidence. The global defect information is packaged into a unified data structure: defect index, global coordinates (x, y), type label, size, grayscale difference, thumbnail path and corresponding die number. The XML / CSV template is called to generate a machine-readable report and upload it to the MES database simultaneously. At the same time, a PDF report containing a defect distribution map and a ZIP package containing a small defect image are output in the customer format. Finally, the "detection completion" signal is returned to the software layer and the memory cache is released, waiting for the next mask template task to be triggered.

[0045] Furthermore, the process of determining the light source brightness correction for the reference image is as follows: Multiple light calibration points are randomly selected at the junction of the transparent area and the non-transparent area in the reference image, and the grayscale values ​​corresponding to the multiple light calibration points are obtained. Then, a light distribution model is constructed through quadratic polynomial fitting. The reference image is normalized using the light distribution model to complete the light source brightness correction.

[0046] Furthermore, the specific process of establishing sub-pixel homonymous point pairs is as follows: The feature point pairs consisting of the ORB feature points to be tested and the reference ORB feature points are roughly matched by Hamming distance matching, and the feature point pairs that meet the preset requirements are finely matched by eliminating false matches through RANSAC to obtain sub-pixel homonymous point pairs.

[0047] Furthermore, the specific process of determining defect detection based on the alignment processing result is as follows: Perform differential processing on the aligned die image to be tested and the reference die image, and extract the suspected defect area through binary segmentation; After pre-processing the suspected defect area, feature analysis is performed using a preset analysis algorithm to determine the final defect result.

[0048] Furthermore, it also includes: Based on the feature identifier of the mask to be tested, the offset between the mask to be tested and the defect detection equipment is calculated, and the position of the mask to be tested is corrected based on the offset.

[0049] Example 1, as Figure 2 As shown in the figure, the software receives the inspection task parameters and completes the initialization configuration of the algorithm module. Based on the inspection recipe name passed to the software, the corresponding template file is loaded. The inspection magnification represents image resolution information and is used for subsequent defect size calculation. The ROI area configuration parameters of the image to be tested are used to capture useful inspection information from the image to be tested. The algorithm outputs a signal indicating whether initialization is successful, loads relevant image data into memory, and configures algorithm operation parameters.

[0050] The captured original image is rotated and translated to align it with the reference image. This is achieved by extracting features from positioning markers, primarily using the ORB feature extraction algorithm to automatically extract positioning markers. ORB (Oriented FAST and Rotated BRIEF) is a computer vision algorithm for feature point detection and description, designed to provide a fast and efficient alternative to SIFT and SURF, two computationally expensive methods. ORB determines the main orientation of each feature point by calculating the grayscale centroid within its neighborhood, a method known as the grayscale centroid method. The grayscale centroid is a point obtained by calculating the weighted average position of the pixels surrounding the feature point, with the weight being its grayscale value. BRIEF (Binary Robust Independent Elementary Features) is a binary string-based descriptor that constructs a descriptor by comparing a pair of pixels (for example, if the intensity of the first pixel is greater than the second, it is recorded as 1, otherwise it is recorded as 0). When generating BRIEF descriptors, ORB does not directly use the original pixel pairs, but rotates and adjusts these pixel pairs according to the calculated main direction, thereby ensuring that feature points can be correctly matched even when the image is rotated.

[0051] like Figure 3As shown in the figure, image brightness differences caused by uneven light sources are corrected to improve the accuracy of subsequent image comparisons. Multiple "grayscale transition points" are selected at the junction of transparent and opaque areas as calibration points. The grayscale values ​​at these locations are sampled as light intensity samples. A quadratic polynomial fit is used to construct a light distribution model. The entire image is then normalized.

[0052] like Figure 4 As shown in the figure, the detection area includes pattern structure area (such as lines, circuit area), pollution defect detection area, and non-functional area (shielded area); the area types that need to be detected are divided, different detection parameters are set according to the less important areas, and the shielded area is excluded to avoid false alarms.

[0053] Feature points are extracted from the die to be tested and the reference die, and a matching is performed to establish a corresponding relationship. The present invention realizes feature matching by extracting ORB features.

[0054] The die image to be tested is converted to the coordinate space of the reference die, ensuring a one-to-one correspondence between pixels. Based on the feature points, the correspondence between the die image to be tested and the reference die image is found. Based on this correspondence, the affine transformation matrix is ​​calculated. Based on bilinear interpolation, the die image to be tested is converted to the coordinate system of the reference die using the affine transformation matrix. To ensure that the image size of the die to be tested and the reference die is consistent, the image size is usually adjusted after the affine transformation to ensure that the image size of the die to be tested and the reference die is consistent.

[0055] Affine transformation is a geometric transformation method widely used in image processing and computer vision. It can realize image translation, rotation, scaling, shearing and other operations. The affine transformation matrix is ​​a 2×3 matrix with the following form: T=[ ], For a two-dimensional point (x, y), after affine transformation, a new point (x', y') is obtained. The transformation formula is: [ ]=[ ]* [ ] Among them: tx and ty represent the translation amount, and a11, a12, a21, and a22 represent rotation, scaling, and shearing.

[0056] In the image alignment task, there are usually at least 3 pairs of non-collinear corresponding points (source_points target_points), and estimates the affine transformation matrix through the least squares algorithm.

[0057] After alignment, the die under test is subtracted from the reference die image to extract abnormal areas and determine whether the suspected defects are real defects. First, based on the formula diff = abs(img1 - img2), the aligned die under test and the reference die are subtracted, and binary segmentation is performed to extract the suspected defect areas. Morphological operations are then performed to remove background noise. Feature analysis (area, width, height, grayscale mean, etc.) is performed, and finally, card control indicators are screened to determine the final true defects.

[0058] For each pair of corresponding pixels (x, y), calculate the grayscale difference between the test die and the reference die: D(x, y) = |Itest(x, y) - Imodel(x, y)|, where D(x, y) represents the pixel value of the difference image at position (x, y); Itest(x, y) and Imodel(x, y) represent the pixel values ​​of the test image and the template image at position (x, y), respectively. Apply a threshold T to the difference image and convert the difference result into a binary image: B(x,y)= This can highlight areas of difference and remove minor differences due to noise or other factors.

[0059] Connected domain analysis is performed on the final binary image to extract and mark each independent defect area. For each connected domain, its area, size and other characteristics can be calculated for further defect analysis.

[0060] Merge and output of inspection results: Summarize the inspection results of all dies and output defect information (coordinates, defect type, size information, defect thumbnail, etc.).

[0061] It should be noted that if Figure 5 As shown, various defects include: opaque spot, pindot, pinhole, black convex, white convex, open circuit, short circuit, thin line, particle, contact hole, etc. Introduction to various defects: Opaque Spot: A white circular spot appears in an originally black area (the light-transmitting area is blocked); Pinhole: Black circular spots appear in previously white areas (light is transmitted through the shaded areas); White Bump is locally darker or lighter (depending on the definition); Black Bump is locally brighter or whiter (depending on how it is used to represent light transmission); Broken Line A line segment is partially missing; Short: There is an additional connecting area between the two lines; Thin Line A line is thinner than other lines; Particle occlusion point.

[0062] In the above embodiments, although the steps are numbered S1, S2, etc., these are only specific embodiments given by the present invention. Those skilled in the art may adjust the execution order of S1, S2, etc. according to actual conditions, which is also within the scope of protection of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0063] The present invention also provides a defect detection system for a reticle mask, and the specific technical solution is as follows: The initialization module is used to adjust the detection parameters of the equipment used for defect detection of the mask to be tested based on the detection magnification information and the name of the mask to be tested in the detection instruction, and store the detection parameters in association with the name of the mask to be tested; The light source brightness correction module is used to: perform light source brightness correction on the reference image based on the grayscale value of the light calibration point in the reference image corresponding to the mask to be tested, and generate a corrected reference image; The detection area determination module is used to divide the detection area of ​​the rectified reference image, extract the ORB feature points of the test die image and the reference ORB feature points of the reference die image, eliminate false matches by using Hamming distance matching and RANSAC, and establish sub-pixel homonymous point pairs. The detection module is used to: determine the conversion matrix for converting the die image to be tested into the reference die image by combining sub-pixel-level same-name point pairs, align the die image to be tested with the reference die image based on the conversion matrix, and perform defect detection based on the alignment results.

[0064] It should be noted that the beneficial effects of the defect detection system for reticle reticles provided in the above-mentioned embodiment are the same as the beneficial effects of the defect detection method for reticle reticles provided in the above-mentioned embodiment, and will not be repeated here. In addition, when implementing the functions of the system provided in the above-mentioned embodiment, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to actual conditions to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above-mentioned embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0065] like Figure 6As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 320, which is coupled to a memory 310. The memory 310 stores at least one computer program 330. The at least one computer program 330 is loaded and executed by the processor 320 to enable the electronic device 300 to implement any of the above methods. Specifically: The electronic device 300 may vary significantly due to different configurations or performance, and may include one or more processors 320 (Central Processing Units, CPUs) and one or more memories 310. The one or more memories 310 store at least one computer program 330, which is loaded and executed by the one or more processors 320 to enable the electronic device 300 to implement the defect detection method for a reticle mask provided in the above embodiment. Of course, the electronic device 300 may also have components such as a wired or wireless network interface, a keyboard, and input / output interfaces for input and output. The electronic device 300 may also include other components for implementing device functions, which are not detailed here.

[0066] A computer-readable storage medium according to an embodiment of the present invention stores at least one computer program, and the at least one computer program is loaded and executed by a processor to enable a computer to implement any of the above methods.

[0067] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, or the like.

[0068] In an exemplary embodiment, a computer program mask or computer program is further provided, the computer program mask or computer program including computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the above methods.

[0069] It should be noted that the terms "first" and "second" in the specification and claims of this application are used to distinguish similar objects and represent a specific order or precedence. The order used for similar objects can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0070] Those skilled in the art will appreciate that the present invention may be implemented as a system, method, or computer program mask. Therefore, the present disclosure may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the present invention may be implemented in the form of a computer program mask embodied in one or more computer-readable media containing computer-readable program code.

[0071] Any combination of one or more computer-readable media may be used. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0072] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A defect detection method for a reticle mask, characterized in that: include: Adjusting detection parameters of a device for performing defect detection on the mask to be tested based on the detection magnification information in the detection instruction and the name of the mask to be tested, and storing the detection parameters in association with the name of the mask to be tested; Based on the grayscale values ​​of the light calibration points in the reference image corresponding to the mask to be tested, performing light source brightness correction on the reference image and generating a corrected reference image; The corrected reference image is divided into detection areas, and the ORB feature points of the test die image and the reference ORB feature points of the reference die image are extracted from the division results. The Hamming distance matching and RANSAC are used to eliminate false matches and establish sub-pixel homonymous point pairs. In combination with the sub-pixel level same-name point pairs, a conversion matrix for converting the die image to be tested into the reference die image is determined. Based on the conversion matrix, the die image to be tested and the reference die image are aligned, and defect detection is performed according to the alignment result.

2. The defect detection method for a reticle mask according to claim 1, characterized in that: The specific process of determining the light source brightness correction for the reference image is as follows: A plurality of light calibration points are randomly selected at the junction of the transparent area and the non-transparent area in the reference image, and the grayscale values ​​corresponding to the plurality of light calibration points are obtained. Then, a light distribution model is constructed through quadratic polynomial fitting processing. The reference image is normalized using the light distribution model to complete the light source brightness correction.

3. The defect detection method for a reticle mask according to claim 1, characterized in that: The specific process of establishing sub-pixel homonymous point pairs is as follows: The feature point pairs consisting of the ORB feature points to be tested and the reference ORB feature points are roughly matched by Hamming distance matching, and the feature point pairs that meet the preset requirements are finely matched by eliminating mismatches by RANSAC to obtain the sub-pixel homonymous point pairs.

4. The defect detection method for a reticle mask according to claim 1, characterized in that: The specific process of determining defect detection based on the alignment processing results is as follows: Performing differential processing on the aligned die image to be tested and the reference die image, and extracting suspected defect areas by binary segmentation; After pre-processing the suspected defect area, feature analysis is performed using a preset analysis algorithm to determine the final defect result.

5. The defect detection method for a reticle mask according to claim 1, characterized in that: Also includes: Based on the feature identifier of the mask to be tested, an offset between the mask to be tested and the defect detection equipment is calculated, and a position correction is performed on the mask to be tested based on the offset.

6. A defect detection system for a reticle mask, characterized in that: include: The initialization module is used to adjust the detection parameters of the equipment used for defect detection of the mask to be tested based on the detection magnification information and the name of the mask to be tested in the detection instruction, and store the detection parameters in association with the name of the mask to be tested; The light source brightness correction module is used to: perform light source brightness correction on the reference image based on the grayscale value of the light calibration point in the reference image corresponding to the mask to be tested, and generate a corrected reference image; The detection area determination module is used to: divide the detection area of ​​the corrected reference image, extract the ORB feature points of the test die image and the reference ORB feature points of the reference die image in the division result, eliminate false matches by using Hamming distance matching and RANSAC, and establish sub-pixel homonymous point pairs; The detection module is used to: determine the conversion matrix for converting the die image to be tested into the reference die image in combination with the sub-pixel level same-name point pairs, align the die image to be tested with the reference die image based on the conversion matrix, and perform defect detection according to the alignment result.

7. The defect detection system for reticle masks according to claim 6, characterized in that: The specific process of determining the light source brightness correction for the reference image is as follows: A plurality of light calibration points are randomly selected at the junction of the transparent area and the non-transparent area in the reference image, and the grayscale values ​​corresponding to the plurality of light calibration points are obtained. Then, a light distribution model is constructed through quadratic polynomial fitting processing. The reference image is normalized using the light distribution model to complete the light source brightness correction.

8. The defect detection system for reticle masks according to claim 6, characterized in that: The specific process of establishing sub-pixel homonymous point pairs is as follows: The feature point pairs consisting of the ORB feature points to be tested and the reference ORB feature points are roughly matched by Hamming distance matching, and the feature point pairs that meet the preset requirements are finely matched by eliminating mismatches by RANSAC to obtain the sub-pixel homonymous point pairs.

9. An electronic device, characterized in that: The electronic device includes a processor coupled to a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor so that the electronic device implements the method according to any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable a computer to implement the method according to any one of claims 1 to 5.

Images