Method for repairing mask substrate coating defects

Abstract

This invention discloses a method for repairing adhesive coating defects on a mask substrate, belonging to the field of mask substrate adhesive coating repair technology. It aims to solve the problem of poor repair results for defects in the adhesive coating area of ​​a mask substrate. This invention utilizes differential enhancement to strengthen defect contrast. A machine learning model accurately classifies defect types and attributes through geometric, textural, and optical features, significantly improving the comprehensiveness and accuracy of defect identification and providing a reliable basis for repair. It employs multi-band light source imaging and differential enhancement fusion technology, combined with a machine learning classification model, to achieve comprehensive defect identification and accurate classification. A full-process data traceability and visualization storage mechanism is established, integrating original defect data, repair process parameters, and finished product information, standardizing data formats, and presenting them visually.

Description

Technical Field

[0001] This invention relates to the field of mask substrate coating repair technology, specifically a method for repairing coating defects on mask substrates. Background Technology

[0002] In existing mask substrate coating processes, defect repair faces numerous technical bottlenecks. Traditional defect identification relies heavily on single-light source imaging, making it difficult to simultaneously address various defect types such as surface scratches, adhesive bubbles, and hidden microcracks, leading to frequent missed detections. Defect classification depends on manual judgment, which is highly susceptible to experience-based variations, resulting in low accuracy in type and attribute classification and directly impacting the targeted nature of repair. During the repair process, existing methods lack a systematic assessment of repair conditions, often leading to resource waste due to incorrect priority judgments; insufficient path planning accuracy can easily cause secondary damage due to positioning deviations; and the strong generality of repair methods fails to tailor processes to specific defects such as bubbles and missed coatings, resulting in poor stability of repair effects. Furthermore, cleaning processes often employ a uniform workflow without adapting cleaning agents to the defect type and repair method, easily leaving impurities; inconsistent quality inspection standards and the lack of a full-process data recording and traceability mechanism make it difficult to support process optimization, limiting the improvement of mask substrate yield. Therefore, a repair solution integrating accurate identification, intelligent classification, targeted repair, and end-to-end control is urgently needed. Summary of the Invention

[0003] The purpose of this invention is to provide a method for repairing adhesive coating defects on a mask substrate. It utilizes differential enhancement to strengthen defect contrast, and a machine learning model accurately classifies defect types and attributes through geometric, textural, and optical features, significantly improving the comprehensiveness and accuracy of defect identification and providing a reliable basis for repair. Employing multi-band light source imaging and differential enhancement fusion technology, combined with a machine learning classification model, it achieves comprehensive defect identification and accurate classification. A full-process data traceability and visualization storage mechanism is established, integrating original defect data, repair process parameters, and finished product information, standardizing data formats, and presenting them visually. This method can solve the problems in existing technologies.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] Methods for repairing adhesive coating defects on mask substrates include:

[0006] First, preliminary defect image identification is performed on the mask substrate; then, the defect types and attributes are classified based on the identified defect images; repair conditions are evaluated based on the defect types and attributes; repair path planning is performed on the mask substrate to be repaired based on the evaluation results; after the repair path planning is completed, a repair plan is generated; the defect areas of the mask substrate are repaired according to the repair plan; the repaired mask substrate is cleaned; after the cleaning is completed, the repair quality of the mask substrate is inspected, and the key parameters of the mask substrate that passes the repair quality inspection are stored.

[0007] Preferably, preliminary defect image recognition of the mask substrate includes:

[0008] The substrate material of the mask substrate is a quartz glass substrate. The surface of the mask substrate is irradiated by a multi-band light source, which includes ultraviolet light, visible light and near-infrared light.

[0009] When a multi-band light source illuminates the surface of a mask substrate, an imaging system matched with the multi-band light source is used for synchronous image acquisition.

[0010] The synchronously acquired images undergo image preprocessing, which includes noise reduction and illumination correction.

[0011] The preprocessed images are used to identify defects and background contrast using differential enhancement algorithms, which include ultraviolet image enhancement, visible light image enhancement, and near-infrared image enhancement.

[0012] The differentiated enhanced images are then fused, and the defect information of the different bands in the fused images is then sorted out.

[0013] After processing, the final defect image is obtained.

[0014] Preferably, the defect type and attributes are classified based on the identified defect images, including:

[0015] By using threshold segmentation and morphological operations, the defect region in the defect image is identified, and the correct background region is excluded.

[0016] Connectivity analysis is performed on the identified defect regions to distinguish overlapping or adjacent defects. After the distinction is completed, the boundaries of each connected region are marked, and the independent defect targets in the defect image are obtained after marking.

[0017] The obtained defect targets are subjected to geometric parameter calculation, texture feature extraction, and optical property analysis.

[0018] The calculation of geometric parameters includes the calculation of basic morphological parameters and positional parameters. After the calculation of geometric parameters, the geometric features of the spatial morphology of the defect target are obtained.

[0019] Texture feature extraction includes grayscale distribution features and spatial structure features. After texture feature extraction, the grayscale distribution pattern of the defect surface in the defect target is obtained.

[0020] Optical property analysis involves extracting the optical response characteristics of defective targets under different optical bands;

[0021] The data from geometric parameter calculation, texture feature extraction, and optical property analysis are used to build a defect classification model based on machine learning.

[0022] Based on the constructed model, the defect target is automatically classified into defect types, and the attributes are confirmed according to the defect type. The attributes include size attributes, location attributes, and severity.

[0023] After the attributes are confirmed, the type and attribute data of the defect area in the defect image are obtained.

[0024] Preferably, the assessment of repair conditions is based on the defect type and attributes, including:

[0025] First, design the assessment index system for repair conditions, which includes indicators for the scope of impact, repairability, and potential risks.

[0026] After the design of the repair condition assessment index system is completed, the historical defect data in the database is combined to assign weights to each index. After the weights are assigned, they are integrated into the assessment model in matrix form.

[0027] The type and attribute data of the defect region in the defect image are mapped into the evaluation model, and a single index score is calculated according to the evaluation index system of repair conditions.

[0028] The single-index score calculation result is multiplied by the corresponding weight and then summed to obtain the comprehensive hazard index of each defect in the defect image.

[0029] Priority levels are ranked according to the numerical range of the comprehensive hazard index, with priority levels including high priority, medium priority, and low priority.

[0030] Finally, based on priority ranking, it is determined whether the defective areas in the defective images meet the repair criteria. Preferably, based on the evaluation results, a repair path is planned for the mask substrate to be repaired, including:

[0031] The defect areas in the defect images that meet the repair standards are identified, and the substrate edges and preset positioning marks or pattern features in the identified defect areas are used as registration reference areas.

[0032] Retrieve the original design drawing of the mask substrate from the database, extract the reference area corresponding to the registration reference area of ​​the original design drawing, and perform pixelation image conversion on the corresponding reference area and the registration reference area in the same proportion.

[0033] After the corresponding reference region and the registration reference region are converted to pixelated images, coordinate system I and geometric distortion correction are performed.

[0034] After coordinate system one and geometric distortion correction are completed, coarse registration is performed on the corresponding reference area and registration reference area. The coarse registration is performed by using a template matching algorithm to match the corresponding reference area and registration reference area and calculate the translation amount of the two in the X and Y axis directions.

[0035] After coarse registration is completed, fine registration is performed. Fine registration involves calculating the coordinate deviation of the feature points between the corresponding reference region and the registration reference region after coarse registration using the least squares method, then fitting the transformation matrix, and adjusting the registration reference region. After adjustment, the alignment error of the feature points between the corresponding reference region and the registration reference region is within 1 pixel.

[0036] Image coordinates are extracted and physical coordinates are transformed in the registration reference area after fine registration, and the size and shape parameters of defects in the registration reference area are recorded;

[0037] The trajectory is designed based on the defect type and shape, and then a global movement path is generated by combining the repair sequence and the defect trajectory.

[0038] Finally, the generated global movement path is converted into path instruction parameters, which serve as the repair path for defective areas in the mask substrate.

[0039] Preferably, after the repair path planning is completed, a repair plan is generated, including:

[0040] A repair method library is constructed based on the defect types of the mask substrate, including bubble defects, scratch defects, missing coating defects, and uneven adhesive layer defects.

[0041] The severity of a defect is set according to its type, including minor, moderate, and severe.

[0042] Repair methods were determined based on the type of defect. The repair methods were as follows: for bubble defects, local laser heating and negative pressure adsorption were used; for scratch defects, mechanical micro-polishing and local glue filling were used; for missed coating defects, precise glue application and template filling were used; and for uneven glue layer defects, local glue scraping and secondary thin coating were used.

[0043] Based on the confirmed repair method, the repair parameters of the repair method are set according to the severity of the defect;

[0044] Feasibility verification of repair methods and parameters is conducted, including historical data comparison, simulated repair testing, and risk prediction.

[0045] Based on the feasibility verification results, a final remediation plan is generated, which includes remediation methods, remediation parameters, remediation sequence, and auxiliary requirements.

[0046] Preferably, the defective areas of the mask substrate are repaired according to the repair plan, including:

[0047] According to the repair plan, the repair equipment should first be calibrated and its parameters adapted.

[0048] After calibration and parameter adaptation are completed, the parameters in the repair plan are imported into the corresponding repair equipment, including positioning parameters, tool parameters and path parameters;

[0049] The repair equipment identifies the imported parameters and then locates the defect on the mask substrate according to the repair path.

[0050] After the location is determined, repair operations are performed, which include defect removal and defect filling operations.

[0051] Among them, defect removal operations include laser ablation and bubble degassing; defect filling operations include micro-dot filling and scratch repair.

[0052] During the repair operation, the processing area is captured in real time by a multi-band imaging system. If a deviation is detected, the multi-band imaging system automatically adjusts its parameters; if the deviation exceeds the threshold, it will trigger an immediate pause.

[0053] The final process completes the repair of defective areas in the adhesive coating on the mask substrate.

[0054] Preferably, the repaired mask substrate undergoes a cleaning process, including:

[0055] The type of cleaning agent is determined according to the repair method and the type of defect. For bubble defects, a weakly alkaline organic solvent is selected; for scratch defects, ultrapure water containing surfactants is selected; and for defects such as missed coating and uneven adhesive layer, a photoresist-specific stripper is selected.

[0056] After confirming the type of cleaning agent, set the temperature and concentration parameters for the cleaning agent;

[0057] After the temperature and concentration parameters are set, a graded cleaning operation is performed, which includes local cleaning, full-area rinsing, and drying.

[0058] Among them, local cleaning is for the defect repair area of ​​the adhesive coating area on the mask substrate. A micro nozzle with a diameter of 0.5 mm is used to spray cleaning agent on the repair area, and the spray flow rate is controlled at 0.1 mL / min. Then, a nanofiber brush with a diameter of 5 μm is used to sweep across the surface of the mask substrate to remove the attached residue through mechanical friction. The brush head pressure is <0.01 N.

[0059] The full-area rinsing involves placing the mask substrate into a clean rinsing tank, injecting ultrapure water, ultrasonically cleaning for 1 minute, replacing the ultrapure water after ultrasonic cleaning, and then spray rinsing for 30 seconds.

[0060] The drying process involves blowing the surface of the mask substrate with 99.999% pure nitrogen gas at a 30° angle at a flow rate of 5 L / min, followed by vacuum drying of the repair area for 5 minutes. The vacuum drying pressure is -0.08 MPa and the temperature is 40°C.

[0061] The mask substrate after the graded cleaning operation is subjected to heat treatment and light treatment. The heat treatment is used for the adhesive layer repair area. The process is to put the mask substrate after the graded cleaning operation into a constant temperature oven for heating treatment. The heating treatment process is: room temperature - 60℃ - 90℃ - 120℃. After the heating treatment, the temperature is cooled to 60℃ and then taken out.

[0062] Photolithography is used to repair areas of photoresist. The process is as follows: UV light irradiation involves using a UV light source to irradiate the repair area at a power density of 100 mW / cm². 2 The irradiation time is adjusted according to the thickness of the adhesive layer; the visible light irradiation treatment involves irradiating the entire substrate with a white light source for 20 minutes, and monitoring the light transmittance of the repaired area during the irradiation process. If the light transmittance deviates from that of the normal area by less than 1%, the irradiation is stopped.

[0063] After the heat treatment and photo-treatment processes are completed, the mask substrate is cleaned.

[0064] Preferably, after cleaning, the mask substrate undergoes repair quality inspection, and key parameters of the mask substrates that pass the repair quality inspection are stored, including:

[0065] Before performing repair quality inspection on the cleaned mask substrate, the repair quality inspection parameters are set. Specifically, based on the parameters of the preliminary defect image recognition, the wavelength, illumination angle, and camera focal length of the multi-band light source in the repair quality inspection are unified.

[0066] Then, the process of image acquisition and image preprocessing in the preliminary defect image recognition is adopted to acquire images of the mask substrate in the repair quality inspection, and to preprocess the images of the mask substrate after image acquisition.

[0067] Using the substrate edge or fixed mark as a reference, the repaired image is aligned with the initial defect image at the pixel level, and the difference image is generated using the pixel grayscale difference analysis method.

[0068] The generated difference images are evaluated using performance metrics, including flatness evaluation, uniformity evaluation, and optical property consistency evaluation.

[0069] Based on the performance evaluation results, determine whether the difference image is within the acceptable range. If it is within the acceptable range, the repair of the adhesive coating defect area on the mask substrate is acceptable; if it is not within the acceptable range, the repair of the adhesive coating defect area on the mask substrate is unacceptable.

[0070] The failed mask substrates are marked and then repaired a second time until they pass the repair test.

[0071] Preferably, after the cleaning process is completed, the mask substrate is subjected to repair quality inspection, and the key parameters of the mask substrate that passes the repair quality inspection are stored. The process also includes:

[0072] Confirm the key parameter data of the repaired mask substrate, including the original defect data, repair process data, quality inspection data and finished product information. Finished product information includes inspection level, applicable scenario, manual review comments and identification code.

[0073] Next, the data format of the key parameter data is standardized, including image data, text data, and numerical data;

[0074] Finally, visualization tools are used to visualize and transform the key parameter data and their corresponding data formats. After the transformation is completed, the data is stored and backed up.

[0075] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0076] 1. The method for repairing adhesive coating defects on a mask substrate provided by this invention employs multi-band light source imaging and differential enhancement fusion technology, combined with a machine learning classification model, to achieve comprehensive identification and accurate classification of defects. Multi-band light covers the ultraviolet, visible, and near-infrared bands, capturing both surface and latent defects; differential enhancement strengthens defect contrast; and the machine learning model accurately classifies defect types and attributes through geometric, textural, and optical features, significantly improving the comprehensiveness and accuracy of defect identification and providing a reliable basis for repair.

[0077] 2. The method for repairing adhesive coating defects on a mask substrate provided by this invention ensures reasonable repair priority based on repair condition evaluation using historical data; multi-level registration technology ensures path planning accuracy within 1 pixel; and customized repair methods are designed according to defect type with dynamic parameter adjustment, real-time monitoring of deviations and loss prevention, significantly reducing the risk of secondary damage and improving repair success rate and quality stability.

[0078] 3. The method for repairing adhesive coating defects on a mask substrate provided by this invention establishes a full-process data traceability and visualization storage mechanism, integrating original defect data, repair process parameters, and finished product information, standardizing data formats, and presenting them visually. This not only enables full-chain quality traceability but also provides data support for process optimization. By analyzing high-frequency defects and repair effects, the process can be continuously iterated to improve the overall manufacturing yield and production efficiency of the mask substrate. Attached Figure Description

[0079] Figure 1 This is a schematic diagram of the steps in the method for repairing adhesive defects on a mask substrate according to the present invention;

[0080] Figure 2 This is a schematic diagram of the process for repairing defects in the adhesive coating of the mask substrate according to the present invention. Detailed Implementation

[0081] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0082] To address the shortcomings of existing technologies, such as incomplete defect identification, significant image noise interference, ambiguous defect classification, and limited repair evaluation metrics with unreasonable priority determination, please refer to [link to relevant documentation]. Figure 1 and Figure 2 This embodiment provides the following technical solution:

[0083] Methods for repairing adhesive coating defects on mask substrates include:

[0084] First, preliminary defect image identification is performed on the mask substrate; then, the defect types and attributes are classified based on the identified defect images; repair conditions are evaluated based on the defect types and attributes; repair path planning is performed on the mask substrate to be repaired based on the evaluation results; after the repair path planning is completed, a repair plan is generated; the defect areas of the mask substrate are repaired according to the repair plan; the repaired mask substrate is cleaned; after the cleaning is completed, the repair quality of the mask substrate is inspected, and the key parameters of the mask substrate that passes the repair quality inspection are stored.

[0085] Preliminary defect image identification of the mask substrate includes:

[0086] The substrate material of the mask substrate is a quartz glass substrate. The surface of the mask substrate is irradiated by a multi-band light source, which includes ultraviolet light, visible light and near-infrared light.

[0087] When a multi-band light source illuminates the surface of a mask substrate, an imaging system matched with the multi-band light source is used for synchronous image acquisition.

[0088] The synchronously acquired images undergo image preprocessing, which includes noise reduction and illumination correction.

[0089] The preprocessed images are used to identify defects and background contrast using differential enhancement algorithms, which include ultraviolet image enhancement, visible light image enhancement, and near-infrared image enhancement.

[0090] The differentiated enhanced images are then fused, and the defect information of the different bands in the fused images is then sorted out.

[0091] After processing, the final defect image is obtained.

[0092] Specifically, considering the transparent / semi-transparent properties of glass substrates, ultraviolet (UV), visible, and near-infrared (NIIR) light can respond to different types of defects: UV light is more sensitive to microbubbles and chemical residues in the adhesive layer; visible light excels at identifying surface scratches and uneven adhesive layers; and NIIR light can penetrate the surface to capture near-surface latent defects. Multi-band combination avoids the "blind spots" of a single light source for specific defects, significantly improving defect detection coverage. Synchronous illumination with the imaging system ensures strict spatial and scale matching of images from different bands, providing a precise foundation for subsequent image fusion. This avoids defect position shifts due to time differences, reducing information misalignment errors during fusion. Noise reduction eliminates ambient light interference and sensor noise, while illumination correction suppresses uneven brightness caused by reflections from the glass substrate, making the image background more uniform and eliminating interference in distinguishing defects from the background, thus improving the stability of subsequent identification. Targeted enhancements are applied to the defect response characteristics of different bands: the UV image enhances the grayscale difference between the adhesive layer and the substrate, the visible light image improves the sharpness of scratch edges, and the NIIR image highlights the contours of latent defects. This "on-demand enhancement" mode maximizes the contrast between defects and the background, solving the problem that defects on the glass substrate surface are similar in grayscale to the background and are difficult to identify. By fusing multi-band enhanced images, the defect information captured by each band is summarized to form a complete defect map. This not only retains the advantages of a single band but also makes up for its limitations. The final output defect image is more comprehensive and accurate, providing a reliable basis for subsequent defect classification and repair.

[0093] Based on the identified defect images, defects are classified according to their type and attributes, including:

[0094] By using threshold segmentation and morphological operations, the defect region in the defect image is identified, and the correct background region is excluded.

[0095] Connectivity analysis is performed on the identified defect regions to distinguish overlapping or adjacent defects. After the distinction is completed, the boundaries of each connected region are marked, and the independent defect targets in the defect image are obtained after marking.

[0096] The obtained defect targets are subjected to geometric parameter calculation, texture feature extraction, and optical property analysis.

[0097] The calculation of geometric parameters includes the calculation of basic morphological parameters and positional parameters. After the calculation of geometric parameters, the geometric features of the spatial morphology of the defect target are obtained.

[0098] Texture feature extraction includes grayscale distribution features and spatial structure features. After texture feature extraction, the grayscale distribution pattern of the defect surface in the defect target is obtained.

[0099] Optical property analysis involves extracting the optical response characteristics of defective targets under different optical bands;

[0100] The data from geometric parameter calculation, texture feature extraction, and optical property analysis are used to build a defect classification model based on machine learning.

[0101] Based on the constructed model, the defect target is automatically classified into defect types, and the attributes are confirmed according to the defect type. The attributes include size attributes, location attributes, and severity.

[0102] After the attributes are confirmed, the type and attribute data of the defect area in the defect image are obtained.

[0103] Specifically, the three-dimensional design of impact range indicators and potential risk indicators constructs a complete assessment framework from three dimensions: the severity of the defect, the feasibility of repair, and subsequent risks. This avoids the one-sidedness of assessments caused by single indicators and ensures that the assessment results are more in line with actual repair needs. Weighting is performed using historical defect repair effect data from the database, ensuring that the weights of each indicator are based on empirical evidence rather than subjective experience, thus improving the objectivity and adaptability of the assessment model. This is especially valuable for assessing new types of defects. The calculation method of single-indicator scoring and weighted summation transforms qualitative descriptions into comparable comprehensive hazard indices, solving the problem of "difficulty in directly comparing different types of defects." This makes the severity of each defect quantifiable and rankable. High, medium, and low priorities are divided through the comprehensive hazard index, allowing for priority treatment of defects that have a significant impact on the mask substrate performance and are easy to repair, avoiding wasting resources on low-hazard defects and improving the overall efficiency of the repair process. The matrix integration and streamlined operation of the assessment model reduce individual differences in human judgment, ensuring consistent assessment results across different batches and operators. At the same time, clear repair standards avoid vague decisions of "repairable or not repairable," providing a clear basis for subsequent repair path planning.

[0104] Based on defect type and attributes, assess repair conditions, including:

[0105] First, design the assessment index system for repair conditions, which includes indicators for the scope of impact, repairability, and potential risks.

[0106] After the design of the repair condition assessment index system is completed, the historical defect data in the database is combined to assign weights to each index. After the weights are assigned, they are integrated into the assessment model in matrix form.

[0107] The type and attribute data of the defect region in the defect image are mapped into the evaluation model, and a single index score is calculated according to the evaluation index system of repair conditions.

[0108] The single-index score calculation result is multiplied by the corresponding weight and then summed to obtain the comprehensive hazard index of each defect in the defect image.

[0109] Priority levels are ranked according to the numerical range of the comprehensive hazard index, with priority levels including high priority, medium priority, and low priority.

[0110] Finally, based on the priority ranking, it is determined whether the defective areas in the defective images meet the repair standards.

[0111] Specifically, the three-dimensional design of impact range indicators and potential risk indicators constructs a complete assessment framework from three dimensions: the severity of the defect, the feasibility of repair, and subsequent risks. This avoids the one-sidedness of assessments caused by single indicators and ensures that the assessment results are more in line with actual repair needs. Weighting is performed using historical defect repair effect data from the database, ensuring that the weights of each indicator are based on empirical evidence rather than subjective experience, thus improving the objectivity and adaptability of the assessment model. This is especially valuable for assessing new types of defects. The calculation method of single-indicator scoring and weighted summation transforms qualitative descriptions into comparable comprehensive hazard indices, solving the problem of "difficulty in directly comparing different types of defects." This makes the severity of each defect quantifiable and rankable. High, medium, and low priorities are divided through the comprehensive hazard index, allowing for priority treatment of defects that have a significant impact on the mask substrate performance and are easy to repair, avoiding wasting resources on low-hazard defects and improving the overall efficiency of the repair process. The matrix integration and streamlined operation of the assessment model reduce individual differences in human judgment, ensuring consistent assessment results across different batches and operators. At the same time, clear repair standards avoid vague decisions of "repairable or not repairable," providing a clear basis for subsequent repair path planning.

[0112] To address the shortcomings of existing technologies, such as low accuracy in repair path planning, lack of specificity in repair methods, unreasonable parameter settings, and insufficient feasibility verification, please refer to [link to relevant documentation]. Figure 1 and Figure 2 This embodiment provides the following technical solution:

[0113] Based on the evaluation results, a repair path is planned for the mask substrate to be repaired, including:

[0114] The defect areas in the defect images that meet the repair standards are identified, and the substrate edges and preset positioning marks or pattern features in the identified defect areas are used as registration reference areas.

[0115] Retrieve the original design drawing of the mask substrate from the database, extract the reference area corresponding to the registration reference area of ​​the original design drawing, and perform pixelation image conversion on the corresponding reference area and the registration reference area in the same proportion.

[0116] After the corresponding reference region and the registration reference region are converted to pixelated images, coordinate system I and geometric distortion correction are performed.

[0117] After coordinate system one and geometric distortion correction are completed, coarse registration is performed on the corresponding reference area and registration reference area. The coarse registration is performed by using a template matching algorithm to match the corresponding reference area and registration reference area and calculate the translation amount of the two in the X and Y axis directions.

[0118] After coarse registration is completed, fine registration is performed. Fine registration involves calculating the coordinate deviation of the feature points between the corresponding reference region and the registration reference region after coarse registration using the least squares method, then fitting the transformation matrix, and adjusting the registration reference region. After adjustment, the alignment error of the feature points between the corresponding reference region and the registration reference region is within 1 pixel.

[0119] Image coordinates are extracted and physical coordinates are transformed in the registration reference area after fine registration, and the size and shape parameters of defects in the registration reference area are recorded;

[0120] The trajectory is designed based on the defect type and shape, and then a global movement path is generated by combining the repair sequence and the defect trajectory.

[0121] Finally, the generated global movement path is converted into path instruction parameters, which serve as the repair path for defective areas in the mask substrate.

[0122] Specifically, the substrate edge, preset positioning marks, and pattern features are used as registration references. These areas have strong geometric stability and high recognizability, which can avoid reference offset caused by deformation or interference in defect areas, providing a reliable reference for subsequent registration and ensuring the spatial reference consistency of path planning. Coarse registration quickly calculates the translation amount through template matching, reducing global deviation. Fine registration uses the least squares method to fit the transformation matrix, controlling the feature point alignment error within 1 pixel. Combined with geometric distortion correction, it effectively eliminates coordinate deviation caused by glass substrate processing errors and image acquisition distortion, ensuring that the actual defect position matches the planned path with millimeter-level or even micrometer-level accuracy, solving the repair deviation caused by "misalignment between drawings and physical objects". To address the defect problem, image coordinate extraction and physical coordinate transformation convert pixel-scale defect parameters into actual physical units. This allows path planning to directly interface with the motion control parameters of the repair equipment, avoiding coordinate scale transformation errors and improving the executability of path instructions. Customized trajectories based on defect type and shape, combined with priority sorting, reduce idle movement of the repair equipment, optimize the work sequence, and significantly improve repair efficiency per unit time. The global path is converted into a unified format of path instruction parameters, allowing different models of repair equipment to directly call them, avoiding individual differences caused by manual programming. This ensures that the repair path for the same defect remains consistent across different batches and different equipment, improving repair stability in large-scale production.

[0123] After the repair path planning is completed, a repair plan is generated, including:

[0124] A repair method library is constructed based on the defect types of the mask substrate, including bubble defects, scratch defects, missing coating defects, and uneven adhesive layer defects.

[0125] The severity of a defect is set according to its type, including minor, moderate, and severe.

[0126] Repair methods were determined based on the type of defect. The repair methods were as follows: for bubble defects, local laser heating and negative pressure adsorption were used; for scratch defects, mechanical micro-polishing and local glue filling were used; for missed coating defects, precise glue application and template filling were used; and for uneven glue layer defects, local glue scraping and secondary thin coating were used.

[0127] Based on the confirmed repair method, the repair parameters of the repair method are set according to the severity of the defect;

[0128] Feasibility verification of repair methods and parameters is conducted, including historical data comparison, simulated repair testing, and risk prediction.

[0129] Based on the feasibility verification results, a final remediation plan is generated, which includes remediation methods, remediation parameters, remediation sequence, and auxiliary requirements.

[0130] Specifically, for four typical defects—bubbles, scratches, missed coatings, and uneven adhesive layers—a dedicated repair method library has been built: laser heating combined with negative pressure adsorption directly addresses the need for gas removal from bubbles; mechanical micro-polishing and adhesive replenishment adapt to the physical morphology repair of scratches; precise adhesive application and template replenishment solve the problem of filling missed coating areas; and localized scraping and secondary thin coating optimize the thickness consistency of uneven adhesive layers. This type-method-oriented approach avoids the inefficiency or misoperation of generic repair methods, ensuring that each defect has a suitable solution. Parameters are set according to minor, moderate, and severe severity levels to avoid "over-repair" or "under-repair," ensuring that the repair intensity is precisely matched with the severity of the defect, thus improving the stability of repair quality. Historical data comparison draws on repair experience of similar defects, and simulated repair tests verify the rationality of parameters through digital twins. Risk prediction helps avoid potential problems in advance. This triple verification mechanism significantly reduces the probability of failure in actual repairs, especially for the repair of new or complex defects. Combining the compatibility of defect location, type, and repair methods, the work sequence is optimized to reduce waiting time, while avoiding mutual interference between different processes, improving the continuity of the repair process. The environmental parameters, tool specifications, and personnel qualifications required for repair are clearly defined, ensuring seamless connection from design to execution, avoiding repair interruptions or quality fluctuations due to the lack of auxiliary conditions, and enhancing the practical operability of the solution.

[0131] To address the problems in existing technologies, such as repair positioning errors, susceptibility to secondary damage, insufficient targeted cleaning resulting in significant residue, inconsistent testing standards leading to inaccurate assessments, and disorganized and difficult-to-trace data storage, please refer to [the relevant documentation / reference]. Figure 1 and Figure 2 This embodiment provides the following technical solution:

[0132] The defective areas of the mask substrate are repaired according to the repair plan, including:

[0133] According to the repair plan, the repair equipment should first be calibrated and its parameters adapted.

[0134] After calibration and parameter adaptation are completed, the parameters in the repair plan are imported into the corresponding repair equipment, including positioning parameters, tool parameters and path parameters;

[0135] The repair equipment identifies the imported parameters and then locates the defect on the mask substrate according to the repair path.

[0136] After the location is determined, repair operations are performed, which include defect removal and defect filling operations.

[0137] Among them, defect removal operations include laser ablation and bubble degassing; defect filling operations include micro-dot filling and scratch repair.

[0138] During the repair operation, the processing area is captured in real time by a multi-band imaging system. If a deviation is detected, the multi-band imaging system automatically adjusts its parameters; if the deviation exceeds the threshold, it will trigger an immediate pause.

[0139] The final process completes the repair of defective areas in the adhesive coating on the mask substrate.

[0140] Specifically, the equipment is calibrated before repair and adapted according to the solution parameters to ensure that the equipment output is highly consistent with the solution requirements. This precise "equipment-solution" connection avoids repair deviations caused by equipment errors from the source, providing stable hardware support for subsequent operations. Positioning parameters, tool parameters, and path parameters are synchronously imported into the equipment, and automatic positioning of the defect location is achieved in combination with the repair path, with positioning accuracy down to the micrometer level. Compared with manual positioning, this significantly reduces operational errors, and is especially suitable for repairing tiny defects on mask substrates. In defect removal operations, laser ablation precisely removes excess adhesive layers, and bubble degassing is used to directionally expel gas. In defect filling operations, micro-dot adhesive filling controls the amount of adhesive to achieve smooth coverage of missed areas, and scratch repair restores surface smoothness through a combination of adhesive application and polishing. This "removal-filling" categorized processing allows each operation to directly target the core of the defect, avoiding the limitations of a single repair method. A multi-band imaging system simultaneously captures images during repair, and automatically fine-tunes equipment parameters to correct minor deviations by comparing the actual trajectory with the planned path in real time. When the deviation exceeds a threshold, the process is immediately paused to prevent the defect from expanding or causing new damage. This "real-time monitoring-dynamic adjustment-emergency loss prevention" mechanism minimizes the risks in the repair process. From parameter import and location to repair operation and deviation adjustment, the entire process requires no manual intervention, reducing individual differences in human operation and significantly increasing the repair volume per unit time. At the same time, the standardized automated process ensures consistent repair quality across different batches, facilitating quality control in large-scale production.

[0141] The repaired mask substrate undergoes cleaning, including:

[0142] The type of cleaning agent is determined according to the repair method and the type of defect. For bubble defects, a weakly alkaline organic solvent is selected; for scratch defects, ultrapure water containing surfactants is selected; and for defects such as missed coating and uneven adhesive layer, a photoresist-specific stripper is selected.

[0143] After confirming the type of cleaning agent, set the temperature and concentration parameters for the cleaning agent;

[0144] After the temperature and concentration parameters are set, a graded cleaning operation is performed, which includes local cleaning, full-area rinsing, and drying.

[0145] Among them, local cleaning is for the defect repair area of ​​the adhesive coating area on the mask substrate. A micro nozzle with a diameter of 0.5 mm is used to spray cleaning agent on the repair area, and the spray flow rate is controlled at 0.1 mL / min. Then, a nanofiber brush with a diameter of 5 μm is used to sweep across the surface of the mask substrate to remove the attached residue through mechanical friction. The brush head pressure is <0.01 N.

[0146] The full-area rinsing involves placing the mask substrate into a clean rinsing tank, injecting ultrapure water, ultrasonically cleaning for 1 minute, replacing the ultrapure water after ultrasonic cleaning, and then spray rinsing for 30 seconds.

[0147] The drying process involves blowing the surface of the mask substrate with 99.999% pure nitrogen gas at a 30° angle at a flow rate of 5 L / min, followed by vacuum drying of the repair area for 5 minutes. The vacuum drying pressure is -0.08 MPa and the temperature is 40°C.

[0148] The mask substrate after the graded cleaning operation is subjected to heat treatment and light treatment. The heat treatment is used for the adhesive layer repair area. The process is to put the mask substrate after the graded cleaning operation into a constant temperature oven for heating treatment. The heating treatment process is: room temperature - 60℃ - 90℃ - 120℃. After the heating treatment, the temperature is cooled to 60℃ and then taken out.

[0149] Photolithography is used to repair areas of photoresist. The process is as follows: UV light irradiation involves using a UV light source to irradiate the repair area at a power density of 100 mW / cm². 2 The irradiation time is adjusted according to the thickness of the adhesive layer; the visible light irradiation treatment involves irradiating the entire substrate with a white light source for 20 minutes, and monitoring the light transmittance of the repaired area during the irradiation process. If the light transmittance deviates from that of the normal area by less than 1%, the irradiation is stopped.

[0150] After the heat treatment and photo-treatment processes are completed, the mask substrate is cleaned.

[0151] Specifically, targeted agents are selected based on the characteristics of different defect repair residues: Adhesive fragments remaining after bubble defect repair are easily dissolved by weakly alkaline organic solvents; tiny particles after scratch repair can be removed by ultrapure water emulsification with surfactants; and photoresist residues after repair of missed coatings and uneven adhesive layers are efficiently decomposed by specialized stripping agents. This targeted "defect-reagent" combination ensures cleaning effectiveness while avoiding corrosion of the substrate or adhesive layer by the cleaning agent. The temperature and concentration parameters of the cleaning agent are optimized according to the agent characteristics and residue type, ensuring cleaning efficiency while avoiding adhesive swelling or incomplete cleaning due to improper parameters, providing a stable foundation for subsequent operations. Local cleaning uses precise spraying from micro-nozzles and gentle pressure friction from nanofiber brushes to remove stubborn residues in the repaired area. The 0.5mm nozzle and low-pressure brush head avoid interference with surrounding areas. Full-area rinsing combines ultrasonic vibration and spraying to remove trace impurities from the entire substrate. Drying uses a combination of high-purity nitrogen purging and vacuum drying to prevent water stains. This three-stage operation forms a comprehensive cleaning process. A complete cleaning chain of "local-global-drying" is employed. Heat treatment promotes cross-linking and curing of the adhesive layer in the repair area through stepped heating, avoiding cracking caused by direct high temperatures. In the phototreatment, ultraviolet light is used to directionally cure the photoresist, while visible light irradiation combined with transmittance monitoring ensures optical consistency between the repaired and normal areas. These two treatments synergistically enhance the physical and optical stability of the repaired area. From 0.5mm micro-nozzles and 5μm fiber brushes to 99.999% high-purity nitrogen, parameter precision is controlled at the micron and milliliter levels. Especially for the high-precision requirements of the mask substrate, this ensures no residue, no damage, and no optical distortion after cleaning, providing reliable samples for subsequent quality inspection. The cleaning process is shown in the table below.

[0152]

[0153]

[0154] After cleaning, the mask substrate undergoes repair quality inspection, and key parameters of the mask substrates that pass the repair quality inspection are stored, including:

[0155] Before performing repair quality inspection on the cleaned mask substrate, the repair quality inspection parameters are set. Specifically, based on the parameters of the preliminary defect image recognition, the wavelength, illumination angle, and camera focal length of the multi-band light source in the repair quality inspection are unified.

[0156] Then, the process of image acquisition and image preprocessing in the preliminary defect image recognition is adopted to acquire images of the mask substrate in the repair quality inspection, and to preprocess the images of the mask substrate after image acquisition.

[0157] Using the substrate edge or fixed mark as a reference, the repaired image is aligned with the initial defect image at the pixel level, and the difference image is generated using the pixel grayscale difference analysis method.

[0158] The generated difference images are evaluated using performance metrics, including flatness evaluation, uniformity evaluation, and optical property consistency evaluation.

[0159] Based on the performance evaluation results, determine whether the difference image is within the acceptable range. If it is within the acceptable range, the repair of the adhesive coating defect area on the mask substrate is acceptable; if it is not within the acceptable range, the repair of the adhesive coating defect area on the mask substrate is unacceptable.

[0160] The failed mask substrates are marked and then repaired a second time until they pass the repair test.

[0161] Confirm the key parameter data of the repaired mask substrate, including the original defect data, repair process data, quality inspection data and finished product information. Finished product information includes inspection level, applicable scenario, manual review comments and identification code.

[0162] Next, the data format of the key parameter data is standardized, including image data, text data, and numerical data;

[0163] Finally, visualization tools are used to visualize and transform the key parameter data and their corresponding data formats. After the transformation is completed, the data is stored and backed up.

[0164] Specifically, the detection parameters are unified with the initial identification, and the image acquisition and preprocessing processes are reused to ensure that the data before and after repair are compared under the same standards. Pixel-level alignment and grayscale difference analysis quantify defect changes, and multi-dimensional performance evaluation covers physical and optical indicators. Combined with clear acceptance criteria, the accuracy of quality inspection is guaranteed from a technical perspective, providing a reliable data source for subsequent parameter storage. Non-conforming products are marked and repaired a second time until they are qualified, avoiding missed defects. The secondary repair is based on the difference image to locate the problem, and the process is optimized in a targeted manner to form a closed loop of "inspection-rework-re-inspection", effectively reducing the risk of non-conforming products leaving the product and ensuring that the stored parameters correspond to high-quality finished products. The stored data covers the original defect information, repair process parameters, quality inspection results, and complete information of the finished product, realizing full-process data traceability from defect generation to finished product delivery. This "full life cycle record" provides a complete basis for subsequent quality analysis and process optimization, unifies the format standards of image, text, and numerical data to avoid data loss or misinterpretation due to format confusion, and uses visualization tools to convert multiple types of data into intuitive charts, combined with a backup mechanism to ensure data security. Visual presentation facilitates the rapid extraction of key information, providing data support for process iteration; standardized storage and backup meet industry compliance requirements and lay the foundation for tracing the source.

[0165] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0166] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for repairing a mask substrate coating defect, characterized in that, The method comprises the following steps: First, the mask substrate is subjected to preliminary defect image recognition; According to the recognized defect image, the defect type and attribute are classified; According to the defect type and attribute, the repair condition is evaluated; According to the evaluation result, the repair path of the mask substrate to be repaired is planned; After the repair path planning is completed, the repair scheme is generated; According to the repair scheme, the defect area of the mask substrate is repaired; After the repair is completed, the mask substrate is subjected to cleaning treatment; After the cleaning treatment is completed, the repair quality of the mask substrate is detected, and the mask substrate with qualified repair quality is subjected to key parameter storage; The preliminary defect image recognition of the mask substrate comprises the following steps: The substrate material of the mask substrate is quartz glass substrate, and a multi-band light source is used to irradiate the surface of the mask substrate, wherein the multi-band light source comprises ultraviolet light, visible light and near-infrared light; When the multi-band light source irradiates the surface of the mask substrate, an imaging system matched with the multi-band light source is used for synchronous image acquisition; The synchronous acquired images are subjected to image preprocessing, and the image preprocessing comprises noise reduction processing and illumination correction; The acquired images after the image preprocessing are subjected to differential enhancement algorithm, and the defect and background contrast of the images are confirmed, wherein the differential enhancement algorithm comprises ultraviolet image enhancement, visible light image enhancement and near-infrared image enhancement; The images after the differential enhancement are fused, and the defect information of different bands in the fused images is arranged; After the arrangement is completed, the final defect image is obtained; According to the recognized defect image, the defect type and attribute are classified, which comprises the following steps: Threshold segmentation and morphological operation are used to confirm the defect area in the defect image, and the correct background area is excluded; Connected domain analysis is performed on the confirmed defect area, and overlapping or adjacent defects are distinguished, and after the distinction is completed, the boundary of each connected domain is marked, and the independent defect target in the defect image is obtained after the marking; The obtained defect target is subjected to geometric parameter calculation, texture feature extraction and optical characteristic analysis; The geometric parameter calculation comprises calculation of basic morphological parameters and position parameters, and the geometric feature of the spatial form of the defect target is obtained after the geometric parameter calculation; The texture feature extraction comprises gray distribution feature and spatial structure feature, and the gray distribution rule of the defect surface in the defect target is obtained after the texture feature extraction; The optical characteristic analysis is to extract the optical response characteristics of the defect target under different light bands; The data of the geometric parameter calculation, the texture feature extraction and the optical characteristic analysis are subjected to model construction based on a machine learning defect classification model; According to the constructed model, the defect type of the defect target is automatically classified, and the attribute is confirmed according to the defect type, wherein the attribute comprises size attribute, position attribute and severity; After the attribute confirmation is completed, the type and attribute data of the defect area in the defect image are obtained. According to the defect type and attribute, the repair condition is evaluated, which comprises the following steps:

2. The method of claim 1, wherein the method further comprises: First, a repair condition evaluation index system is designed, and the repair condition evaluation index system comprises influence range index, repairability index and potential risk index; ​ After the design of the repair condition evaluation index system is completed, the historical defect data in the database is combined to assign weights to each index. After the weight assignment is completed, the evaluation model is integrated in matrix form; The type and attribute data of the defect area in the defect image are mapped into the evaluation model, and single-index scoring calculation is performed according to the repair condition evaluation index system; The single-index scoring calculation result is multiplied by the corresponding weight and then summed to calculate the comprehensive hazard index of each defect in the defect image; According to the numerical range of the comprehensive hazard index, the priority level is sorted, including high priority, medium priority and low priority; Finally, whether the defect area in the defect image meets the repair standard is determined according to the priority level sorting.

3. The method of claim 2, wherein the method further comprises: According to the evaluation result, the repair path planning of the mask substrate to be repaired is carried out, including: The defect area in the defect image that meets the repair standard is identified, and the substrate edge in the identified defect area and the pre-set positioning mark or pattern feature are used as the registration reference area; The original design drawing of the mask substrate is called from the database, and the reference area corresponding to the registration reference area in the original design drawing is extracted, and the corresponding reference area and the registration reference area are converted into the same scale pixel image; After the corresponding reference area and the registration reference area pixel image conversion is completed, the coordinate system is unified and the geometric distortion is corrected; After the coordinate system unification and the geometric distortion correction are completed, the corresponding reference area and the registration reference area are coarsely registered, wherein the coarse registration is to match the corresponding reference area and the registration reference area by using a template matching algorithm, and the translation amount of the two in the X and Y axis directions is calculated; After the coarse registration is completed, fine registration is performed, which is to calculate the coordinate deviation of the feature points by using the least square method, then to perform transformation matrix fitting, and to adjust the registration reference area, so that the alignment error of the feature points in the corresponding reference area and the registration reference area after adjustment is within 1 pixel; The image coordinates of the registration reference area after fine registration are extracted and the physical coordinates are converted, and the size and shape parameters of the defects in the registration reference area are recorded; According to the defect type and shape, the trajectory design is carried out, and then the global movement path is generated in combination with the repair sequence and the defect trajectory; Finally, the generated global movement path is converted into path instruction parameters, which are used as the repair path of the defect area in the mask substrate.

4. The method of claim 3, wherein the method further comprises: After the repair path planning is completed, the repair scheme is generated, including: A repair method library is constructed according to the defect type of the mask substrate, including bubble defect, scratch defect, missing coating defect and uneven glue layer defect; The defect severity is set according to the defect type, including slight, moderate and severe; The repair method is confirmed according to the defect type, wherein the repair method is: laser local heating and negative pressure adsorption for bubble defect; mechanical micro-polishing and local glue supplement for scratch defect; precise point gluing and template coating for missing coating defect; local glue scraping and secondary thin coating for uneven glue layer defect; According to the confirmed repair method, the repair parameters of the repair method are set in combination with the defect severity. The feasibility of the repair method and repair parameters is verified, including historical data comparison, simulation repair test and risk prediction; According to the feasibility verification result, the final repair scheme is generated, including repair method, repair parameter, repair sequence and auxiliary requirement.

5. The method of claim 4, wherein the repairing the glue defect of the mask substrate is performed by a laser. According to the repair scheme, the defect area of the mask substrate is repaired, including: According to the repair scheme, the repair equipment is calibrated and the parameters are adapted; After calibration and parameter adaptation, the parameters in the repair scheme are imported into the corresponding repair equipment, including positioning parameters, tool parameters and path parameters; The repair equipment identifies the imported parameters, and after identification, the defect position on the mask substrate is positioned according to the repair path; After positioning, the repair operation is carried out, including defect removal operation and defect filling operation; Among them, the defect removal operation is laser ablation removal and bubble exhaust treatment; The defect filling operation is micro-point glue filling and scratch repair; During the repair operation, the processing area is photographed in real time by the multi-band imaging system, and if deviation is detected, the multi-band imaging system automatically adjusts the parameters; If the deviation exceeds the threshold, it will trigger immediate suspension; Finally, the repair process of the glue defect area on the mask substrate is completed.

6. The method of claim 5, wherein the method further comprises: The mask substrate repaired is cleaned, including: According to the repair method and defect type, the type of cleaning agent is confirmed, wherein the bubble defect selects an organic solvent with weak alkaline composition; The scratch defect selects ultrapure water containing a surfactant; The underfill defect and the uneven glue layer defect select a photoresist special stripper; After the type of cleaning agent is confirmed, the temperature and concentration parameters of the cleaning agent are set; After the temperature and concentration parameters are set, the hierarchical cleaning operation is carried out, including local cleaning, global rinsing and drying treatment; Among them, the local cleaning is for the defect repair area of the glue area on the mask substrate, a micro nozzle with a diameter of 0.5mm is used to spray cleaning agent on the repair area, the spray flow is controlled at 0.1mL / min, and a nanofiber brush with a diameter of 5μm is used to sweep the surface of the mask substrate, and the residual adhered is removed by mechanical friction, the brush head pressure is <0.01N; The global rinsing is to put the mask substrate into a clean rinsing tank and inject ultrapure water, and ultrasonic cleaning is carried out for 1 minute, after ultrasonic cleaning, the ultrapure water is replaced, and then spray rinsing is carried out for 30 seconds; The drying treatment is to use clean nitrogen gas with a purity of 99.999% to blow the surface of the mask substrate at an angle of 30°, the flow rate is 5L / min, and then the repair area is vacuum dried for 5 minutes, wherein the vacuum drying pressure is-0.08MPa and the temperature is 40℃; The mask substrate after hierarchical cleaning operation is subjected to heat treatment and light treatment, the heat treatment is used for the glue layer repair area, and the process is to put the mask substrate after hierarchical cleaning operation into a constant temperature oven for heating treatment, the heating treatment process is: room temperature-60℃-90℃-120℃, after heating treatment, the temperature is lowered to 60℃ and then taken out; The light treatment is used for repairing the photoresist area, and the process is as follows: the ultraviolet light treatment is to use an ultraviolet light source to irradiate the repair area, and the power density is 100 mW / cm², and the irradiation time is adjusted according to the thickness of the glue layer; the visible light treatment is to use a white light source to irradiate the whole substrate for 20 minutes, and the transmittance of the repair area is monitored during the irradiation process, if the transmittance deviation of the repair area is less than 1% compared with the normal area, then stop irradiation; After the process of heat treatment and light treatment is completed, the cleaning treatment of the mask substrate is completed.

7. The method of claim 6, wherein the method further comprises: After the cleaning treatment is completed, the repair quality detection of the mask substrate is carried out, and the key parameters of the mask substrate with qualified repair quality detection are stored, including: Before the repair quality detection of the mask substrate after the cleaning treatment is completed, the repair quality detection parameters are set, wherein, according to the parameters of the preliminary defect image recognition, the wavelength, irradiation angle and camera focal length of the multi-band light source in the repair quality detection are unified; Then, the image acquisition and image preprocessing process in the preliminary defect image recognition is used to acquire the image of the mask substrate in the repair quality detection, and the image preprocessing of the mask substrate after the image acquisition is completed is carried out; Taking the substrate edge or fixed mark as the reference, the repaired image is aligned with the initial defect image at pixel level, and the pixel gray scale difference analysis method is used to generate the difference image; The performance index evaluation of the generated difference image is carried out, including flatness evaluation, uniformity evaluation and optical property consistency evaluation; According to the performance index evaluation result, it is judged whether the difference image is within the qualified judgment range, if yes, the repair of the photoresist defect area on the mask substrate is qualified; if not, the repair of the photoresist defect area on the mask substrate fails; The mask substrate with repair failure is labeled and subjected to secondary repair until the repair is qualified.

8. The method of claim 7, wherein the method further comprises: After the cleaning treatment is completed, the repair quality detection of the mask substrate is carried out, and the key parameters of the mask substrate with qualified repair quality detection are stored, including: The key parameter data of the mask substrate with qualified repair is confirmed, including defect original data, repair process data, quality detection data and product information, the product information includes inspection level, applicable scene, manual review opinion and identification code; Then, the data format of the key parameter data is standardized, including image data, text data and numerical data; Finally, the key parameter data and the corresponding data format are converted into visual data by using a visual tool, and after the conversion is completed, the data is stored and backed up.

Images