Macro defect detection method in semiconductor manufacturing and high-resolution camera

Abstract

The invention discloses a Macro defect detection method in semiconductor manufacturing and a high-resolution camera, and relates to the technical field of defect detection. The method comprises the following steps: acquiring a global image of a wafer, dividing the global image into a plurality of partition images, correlating and querying historical Macro defect data according to corresponding position areas on the wafer, obtaining a defect probability list, dynamically constructing a preferential matching defect subset for each position area, extracting the features of each partition image, and determining the defect probability list according to the preferential matching defect subset. The method comprises the following steps: firstly, carrying out first matching on a defect subset, carrying out rapid matching with a preferential matching defect subset, outputting a Macro defect type, calculating the matching reliability of the Macro defect type of a position region corresponding to each partition image based on a wafer manufacturing process and material optical characteristics, and if the matching reliability of a certain region is smaller than a reliability threshold value, calling a global Macro defect set for matching. Stable and reliable technical support can be provided for quality control in the semiconductor manufacturing process, and the yield and production efficiency of semiconductor products are improved.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and in particular to a method for detecting macro defects in semiconductor manufacturing and a high-resolution camera. Background Technology

[0002] In semiconductor manufacturing processes, the detection of wafer surface defects is a crucial step in ensuring product yield and reliability. Defects are generally classified into micro defects and macro defects. Macro defects mainly refer to relatively large-scale defects (typically on the order of micrometers to millimeters) that can be directly observed by optical means, such as scratches, contamination, particles, missing patterns, and cracks. These defects not only directly affect device performance but also often reflect systemic problems in process equipment or environmental control; therefore, comprehensive and rapid detection and monitoring of these defects is essential.

[0003] Currently, the detection of wafer surface defects in semiconductor production lines mainly relies on automated optical inspection systems. To balance detection accuracy and efficiency, existing technologies typically employ two main approaches: First, using a high-magnification, high-pixel-resolution local scanning imaging system, which uses a precision motion platform to drive the wafer or camera for field-by-field or region-by-region scanning. This method can obtain extremely high local imaging detail, meeting the accuracy requirements for detecting minute defects. However, its detection speed is limited by the small single field of view area, long mechanical movement, and long image stitching time, resulting in excessively long inspection times for the entire wafer, making it difficult to meet the stringent throughput requirements of modern high-volume production lines. Second, using a relatively low-resolution camera with a larger field of view for rapid surface or line scanning. While this method improves detection speed, the large physical size of the wafer corresponding to a single pixel significantly reduces the ability to identify and locate minute or low-contrast macro defects, easily leading to missed or false detections, and failing to meet the defect detection accuracy requirements of advanced processes.

[0004] Therefore, existing technologies present a prominent contradiction: high-precision inspection methods are often slow and have low throughput; while high-speed inspection methods struggle to guarantee the accuracy and reliability of detecting various macro defects across the entire wafer. This contradiction becomes increasingly acute as semiconductor device feature sizes continue to shrink and wafer sizes increase. Production lines urgently need an inspection solution that can achieve rapid and comprehensive inspection of macro defects across the entire wafer without sacrificing, or even improving, inspection accuracy.

[0005] In summary, designing a novel detection method and system that can significantly improve detection speed while meeting the requirements of semiconductor production lines for defect detection sensitivity, positioning accuracy, and classification accuracy, and achieve efficient and rapid inspection of macro defects across the entire wafer, has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This invention provides a method for detecting macro defects in semiconductor manufacturing, comprising: Step S1: Acquire a global image of the wafer and divide it into multiple partition images; Step S2: Based on the location area on the wafer corresponding to each partition image, query historical macro defect data to obtain a list of macro defect probabilities for each location area; Step S3: Based on the Macro defect probability list for each location region, dynamically construct a priority matching defect subset for each location region; Step S4: Extract the features of each partition image, perform fast matching with the priority matching defect subset of its corresponding location region, and output the Macro defect type of the location region corresponding to each partition image. Step S5: Based on the wafer manufacturing process and material optical properties, calculate the matching reliability of the Macro defect type in the corresponding region of each partition image. If the matching reliability of a certain region is less than the reliability threshold, call the preset global Macro defect set for matching and re-output the Macro defect type of the corresponding region.

[0007] The method for detecting macro defects in semiconductor manufacturing, as described above, includes the following sub-steps: acquiring a global image of the wafer and dividing it into multiple partition images. Step S11: Acquire a high-resolution global image of the wafer using optical inspection equipment; Step S12: Establish coordinate mapping based on wafer notches, divide the high-resolution global image into multiple partition images, and associate them with the coordinate positions of the wafer.

[0008] The method for detecting macro defects in semiconductor manufacturing, as described above, includes the following sub-steps: Based on the location region on the wafer corresponding to each partition image, historical macro defect data is queried to obtain a list of macro defect probabilities for each location region. Step S21: Based on the entire historical wafer manufacturing process, collect historical macro defect data and construct a historical macro defect dataset; Step S22: Based on the location region on the wafer corresponding to each partition image, perform retrieval and statistics in the historical Macro defect dataset to obtain a list of Macro defect probabilities for each location region.

[0009] The method for detecting macro defects in semiconductor manufacturing, as described above, includes the following sub-steps: Dynamically constructing a priority matching defect subset for each location region based on a list of macro defect probabilities for each region. Step S31: Filter the list of Macro defects for each location region to construct an initial candidate defect subset for each location region; Step S32: Adjust and sort the Macro defect types in the initial candidate defect subsets of each location region to generate a priority matching defect subset for each location region.

[0010] The method for detecting macro defects in semiconductor manufacturing, as described above, includes the following sub-steps: extracting features from each partition image, quickly matching them with a priority subset of defects in their corresponding location regions, and outputting the macro defect type for each partition image's corresponding location region. Step S41: Extract image features of each partition image in parallel using a multi-scale adaptive feature extraction network; Step S42: Match the image features of each partition image with the priority matching defect subset of its corresponding location region to obtain the Macro defect type of the location region corresponding to each partition image.

[0011] The method for detecting macro defects in semiconductor manufacturing, as described above, involves calculating the matching reliability of macro defect types for corresponding regions in each image partition based on the wafer manufacturing process and material optical properties. If the matching reliability of a certain region is less than a reliability threshold, a preset global macro defect set is called for matching, and the corresponding macro defect type is re-output. This includes the following sub-steps: Step S51: Based on the wafer manufacturing process and material optical properties, calculate the matching reliability of the Macro defect type in the corresponding location area of ​​each partition image using a multi-source reliability fusion algorithm; Step S52: If the matching reliability of the Macro defect type in the corresponding region of the partitioned image is less than the preset reliability threshold, trigger the re-matching of the low reliability region, call the preset global Macro defect set for matching, and re-output the Macro defect type of the corresponding region.

[0012] The present invention also provides a high-resolution camera for detecting macro defects in semiconductor manufacturing, comprising: The image acquisition and segmentation module acquires a global image of the wafer and divides it into multiple partition images; The defect probability list generation module retrieves the macro defect probability list for each location region by querying historical macro defect data based on the location area on the wafer corresponding to each partition image. The priority matching defect subset construction module dynamically constructs a priority matching defect subset for each location region based on the Macro defect probability list of each location region. The defect type matching module extracts the features of each partition image and performs fast matching with the priority matching defect subset of its corresponding location region, outputting the Macro defect type of the location region corresponding to each partition image; The defect matching result verification module calculates the matching reliability of the Macro defect type in the corresponding region of each partition image based on the wafer manufacturing process and material optical properties. If the matching reliability of a certain region is less than the reliability threshold, the preset global Macro defect set is called for matching, and the Macro defect type of the corresponding region is re-output.

[0013] As described above, a high-resolution camera for macro defect detection in semiconductor manufacturing includes an image acquisition module, specifically comprising: The image acquisition submodule acquires high-resolution global images of the wafer using optical inspection equipment; The image segmentation submodule establishes a coordinate mapping based on wafer notches, divides the high-resolution global image into multiple partition images, and associates them with the wafer's coordinate positions.

[0014] As described above, a high-resolution camera for macro defect detection in semiconductor manufacturing includes a defect probability list generation module, which specifically comprises: The historical macro defect dataset construction submodule collects historical macro defect data based on the entire historical wafer manufacturing process and constructs a historical macro defect dataset. The defect probability list acquisition submodule retrieves the probability list of macro defects for each location region by searching and statistically analyzing the historical macro defect dataset based on the location region on the wafer corresponding to each partition image.

[0015] As described above, a high-resolution camera for macro defect detection in semiconductor manufacturing includes a priority matching defect subset construction module, which specifically includes: The initial candidate defect subset construction submodule filters the Macro defect probability list for each location region and constructs an initial candidate defect subset for each location region. The priority matching defect subset generation submodule adjusts and sorts the Macro defect types in the initial candidate defect subsets of each location region to generate priority matching defect subsets for each location region.

[0016] As described above, a high-resolution camera for macro defect detection in semiconductor manufacturing includes a defect type matching module, specifically comprising: The image feature extraction submodule extracts image features from each partition image in parallel through a multi-scale adaptive feature extraction network. The Macro defect type acquisition submodule matches the image features of each partition image with the priority matching defect subset of its corresponding location region to obtain the Macro defect type of the corresponding location region of each partition image.

[0017] As described above, a high-resolution camera for macro defect detection in semiconductor manufacturing includes a defect matching result verification module, which specifically comprises: The matching reliability calculation submodule, based on the wafer manufacturing process and material optical properties, calculates the matching reliability of the Macro defect type in the corresponding location area of ​​each partition image through a multi-source reliability fusion algorithm. The verification matching submodule, if the matching reliability of the Macro defect type in the corresponding location area of ​​the partitioned image is less than the preset reliability threshold, triggers the verification matching of the low reliability area, calls the preset global Macro defect set for matching, and re-outputs the Macro defect type of the corresponding area.

[0018] The beneficial effects achieved by this invention are as follows: This invention can significantly reduce the matching redundancy of irrelevant macro defect types, improve detection speed, solve the problems of large cross-regional defects being easily fragmented and local high-incidence defects being easily missed, ensure the integrity and specificity of macro defect detection, reduce the risk of false detection caused by environmental interference and process differences, achieve a synergistic improvement in the efficiency, accuracy and integrity of macro defect detection, provide stable and reliable technical support for quality control in the semiconductor manufacturing process, and improve the yield and production efficiency of semiconductor products. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of a macro defect detection method in semiconductor manufacturing provided in Embodiment 1 of this application; Figure 2 This is a schematic diagram of a high-resolution camera for detecting macro defects in semiconductor manufacturing, provided in Embodiment 2 of this application. Detailed Implementation

[0021] 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, not all, of the embodiments of the present invention. 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.

[0022] Example 1

[0023] like Figure 1 As shown in Embodiment 1 of this application, a method for detecting macro defects in semiconductor manufacturing is provided. The method includes the following steps: Step S1: Acquire a global image of the wafer and divide it into multiple partition images; Furthermore, acquiring a global image of the wafer and dividing it into multiple partition images includes the following sub-steps: Step S11: Acquire a high-resolution global image of the wafer using optical inspection equipment; Specifically, the optical inspection equipment uses a high-resolution optical imaging system specifically designed for wafer inspection. Its core components include, but are not limited to, a high-frame-rate CMOS image sensor, a high numerical aperture telecentric microscope lens, a multi-mode programmable illumination module, and a real-time noise reduction and image enhancement module. The high-frame-rate CMOS image sensor has a pixel resolution of no less than 12 million pixels and a pixel size ≤3.5μm, used to capture microscopic details on the wafer surface. The high numerical aperture telecentric microscope lens has a numerical aperture ≥0.85, used to reduce optical distortion and ensure consistent imaging magnification between the wafer's central and edge regions. The multi-mode programmable illumination module is used for bright-field illumination, dark-field illumination, preset-angle oblique illumination, and multi-band visible light switching within an adjustable wavelength range. It adaptively matches the optimal illumination mode based on the wafer's material optical properties, highlighting the wafer surface's texture features, defect contours, and macroscopic morphology information. The real-time noise reduction and image enhancement module uses adaptive threshold filtering and edge sharpening algorithms to suppress ambient light interference and image noise, improving image grayscale levels and detail clarity. The aforementioned optical inspection equipment is used to acquire high-resolution global images of the wafer with no obvious noise and with key information that can be accurately identified. The key information includes, but is not limited to, the chip outline, suspected defect prototypes, and edge morphology on the wafer surface.

[0024] Step S12: Establish coordinate mapping based on wafer notches, divide the high-resolution global image into multiple partition images, and associate them with the coordinate positions of the wafer; Specifically, using the wafer edge notch as a physical reference, a unified wafer physical coordinate system is established. By identifying specific alignment marks associated with the wafer notch in the high-resolution global image, a mapping relationship between the wafer physical coordinate system and the image pixel coordinate system is constructed, achieving coordinate alignment between the high-resolution global image and the wafer. Based on the wafer's die layout design data (including die standard size, horizontal and vertical spacing, number of rows and columns, and other related parameters), a virtual grid corresponding one-to-one with the wafer die distribution is generated on the high-resolution global image that has completed coordinate alignment. The image area covered by each virtual grid is defined as a partition image. For areas where the wafer edge cannot form a complete die, its corresponding image area is also divided into independent partitions to ensure that every spatial area of ​​the wafer has a corresponding partition image. At the same time, based on the row and column numbers of each partition image in the virtual grid, a unique partition identifier is assigned to each partition image. Through the coordinate mapping relationship, the partition identifier is associated with the physical coordinates on the wafer corresponding to the partition image, achieving a unique association between each partition image and a specific area of ​​the wafer.

[0025] Step S2: Based on the location area on the wafer corresponding to each partition image, query historical macro defect data to obtain a list of macro defect probabilities for each location area; Furthermore, based on the location region on the wafer corresponding to each partition image, historical macro defect data is queried to obtain a list of macro defect probabilities for each location region, including the following sub-steps: Step S21: Based on the entire historical wafer manufacturing process, collect historical macro defect data and construct a historical macro defect dataset; Specifically, based on the entire historical wafer manufacturing process, comprehensive data on various historical macro defects generated from core process steps to final product inspection are collected to construct a historical macro defect dataset. Each macro defect record includes a coordinate location identifier field, a defect classification code field, a process context field, and a defect feature field. The coordinate location identifier field records the coordinates of the wafer and corresponding die where the macro defect occurs. The defect classification code field records the type of macro defect, defining it through standardized classification coding rules (e.g., SCRATCH_01 represents a specific type of scratch). The process context field records the specific process step, corresponding equipment number, and batch number at the time the macro defect occurred. The defect feature field records the key image features of the macro defect. The historical macro defect dataset is updated in real-time with new macro defect data throughout the entire production line manufacturing process.

[0026] Step S22: Based on the location region on the wafer corresponding to each partition image, perform retrieval and statistics in the historical Macro defect dataset to obtain a list of Macro defect probabilities for each location region; Specifically, for each partition image of the wafer, the associated location region on the wafer is identified through its partition identifier. Simultaneously, the wafer's process context data is collected. This process context data includes the process step the wafer is in and the machine number of the process equipment used. Using the coordinates of the wafer location region corresponding to the partition image as the retrieval reference, and combining the process context data, a spatiotemporal context defect retrieval algorithm is used to perform correlation retrieval and statistics in the historical macro defect dataset, calculating the probability of occurrence of various macro defects in each location region. The spatiotemporal context defect retrieval algorithm is as follows: Based on the coordinates of the target location region associated with the target partition image, a search is performed in the historical macro defect dataset to obtain multiple historical macro defect data. The search scope includes historical macro defect data of the target location region, as well as all historical macro defect data within a circular or square spatial neighboring region with a radius equal to a preset number of grains centered at the corresponding coordinates. Based on the grain spacing between the historical macro defect location region and the target location region, a spatial weight that decreases with increasing distance is assigned to each historical macro defect data. Based on the process context fields of each historical macro defect data and the process context data of the current wafer, process flow similarity is calculated, and a process context weight is assigned to each historical macro defect data based on this similarity. The historical macro defect data were grouped and statistically analyzed according to macro defect type, and the probability calculation formula was used. Calculate the probability of occurrence of each macro defect type in the corresponding location region of the target partition image, where, For the first Macro-type defect type The probability of occurrence in the target location area The range of values ​​is , The number of types of macro defects appearing in the target location region. The first one retrieved from the historical Macro defect dataset Macro-like defects The number of historical Macro defect data entries The range of values ​​is , For the first Macro-like defects The Spatial weights of historical data For the first Macro-like defects The The process context weight of each historical data point This represents the total number of historical Macro defect records retrieved from the historical Macro defect dataset. The range of values ​​is , For the first Spatial weights of historical data For the first The process context weights of the historical data are calculated. The above retrieval and calculation are performed in parallel on the location regions corresponding to each partition image, and the Macro defect type-occurrence probability association data corresponding to each location region is output to form a Macro defect probability list for each location region.

[0027] Step S3: Based on the Macro defect probability list for each location region, dynamically construct a priority matching defect subset for each location region; Furthermore, based on the Macro defect probability list for each location region, dynamically constructing a priority matching defect subset for each location region includes the following sub-steps: Step S31: Filter the list of Macro defects for each location region to construct an initial candidate defect subset for each location region; Specifically, a base probability threshold is set based on the sensitivity of the process to macro defects in the target location area. Macro defect types with a probability greater than the base probability threshold are filtered out from the macro defect probability list of the target location area to construct an initial candidate defect subset for the target location area. The arithmetic mean of the occurrence probabilities of all macro defects is then calculated based on the macro defect probability list of the target location area. With the maximum value ,when hour, The amplification factor indicates the presence of a single, high-incidence macro defect risk in the target location area, and is dynamically adjusted using a threshold formula. The base probability threshold is dynamically adjusted, whereby... The adjusted probability threshold. Based on the basic probability threshold, As a regulating factor, To be the maximum probability of occurrence, The arithmetic mean of the occurrence probabilities is used to re-select the Macro defect types of the target location region based on the adjusted probability threshold, and an initial candidate defect subset of the target location region is constructed. The above operation is performed in parallel for each location region to form an initial candidate defect subset of each location region.

[0028] Step S32: Adjust and sort the Macro defect types in the initial candidate defect subsets of each location region to generate a priority matching defect subset for each location region; Specifically, for each Macro defect type in the initial candidate defect subset of the target location region. , This is an index identifier for the Macro defect type within the initial candidate defect subset of the target location region. The range of values ​​is , To determine the number of Macro defect types within the initial candidate defect subset for the target location region, retrieve the Macro defect types from the historical Macro defect dataset. Historical Macro defect data appearing in the spatially adjacent area of ​​the target location region are analyzed using the spatial clustering intensity formula. Calculate Macro Defect Types The spatial cluster strength, where, Macro defect type The spatial cluster intensity, the higher the value, the more macro defect type it represents. The more pronounced the spatial clustering characteristics in the target location area, the better. The number of historical Macro defect data retrieved. The range of values ​​is , Macro defect type Spatial influence radius parameter, For target location area The center coordinates, For the first The center coordinates of the region where the historical Macro defect data is located.

[0029] Simultaneously, by combining the process context data of the target location area, the types of macro defects are statistically analyzed in the historical macro defect dataset. The number of occurrences within the same or highly similar process context is calculated using the process correlation formula. The degree of enhancement of process-related correlation is calculated, among which, Macro defect type Enhanced process relevance This is the process-related gain coefficient. Macro defect type The number of times they appear in the same or highly similar process context Macro defect type Total number of occurrences across all process contexts.

[0030] Combining the occurrence probability of various macro defect types in the target location area, the spatial cluster strength, and the process-related enhancement degree, a dynamic priority probability formula is used. Calculate the priority match rate for each Macro defect type in the initial candidate defect subset, where, Macro defect type Priority matching rate Macro defect type The probability of occurrence in the target location area Macro defect type Spatial cluster intensity in the target location region, Macro defect type Enhanced process relevance in the target location region.

[0031] Based on the priority matching rate of each macro defect type, the macro defect types in the initial candidate defect subset of the target location region are sorted in descending order. Defect feature fields for each macro defect type are extracted from the historical macro defect dataset and used as standard defect image features. These features are then associated and bound one-to-one with the macro defect types in the sorted initial candidate defect subset, forming a priority matching defect subset for the target location region. This process is repeated in parallel for each location region to form a priority matching defect subset for each region.

[0032] Step S4: Extract the features of each partition image, perform fast matching with the priority matching defect subset of its corresponding location region, and output the Macro defect type of the location region corresponding to each partition image. Furthermore, features of each partition image are extracted, and a priority matching defect subset of its corresponding location region is used for fast matching. The output of the Macro defect type of the location region corresponding to each partition image includes the following sub-steps: Step S41: Extract image features of each partition image in parallel using a multi-scale adaptive feature extraction network; Specifically, a multi-scale adaptive feature extraction network is used to extract image features of each partition image in parallel. The multi-scale adaptive feature extraction network is a lightweight encoder-decoder structure. The encoder includes three parallel convolutional branches, which use convolutional kernels of different sizes to convolve the input partition image to capture multi-scale visual features from pixel-level details to region-level structures. The decoder fuses feature maps from different branches through skip connections and outputs image features of a fixed-dimensional, high-information-density partition image, which comprehensively represents the texture, edge, shape and gray-level distribution of each partition image.

[0033] Step S42: Match the image features of each partition image with the priority matching defect subset of its corresponding location region to obtain the Macro defect type of the location region corresponding to each partition image; Specifically, based on the ranking of Macro defect types in the priority matching defect subset corresponding to the location region of the target partition image, the image features of the target partition image are sequentially compared with the standard defect image features bound to each Macro defect type in the priority matching defect subset. Simultaneously, based on the defect feature fields of each Macro defect type in the historical Macro defect dataset, the defect matching strictness coefficient corresponding to the current wafer's process context, and the preset image quality compensation coefficient based on the location characteristics on the wafer, an image feature similarity threshold is dynamically set. If the similarity of the currently matched Macro defect type is greater than or equal to the image feature similarity threshold, this Macro defect type is determined to be the Macro defect type of the target partition image, and the matching of the remaining defect types in the priority matching defect subset is terminated. If the similarity of the currently matched Macro defect type is less than the image feature similarity threshold, the matching of the next Macro defect type in the priority matching defect subset continues. If no Macro defect type satisfying the image feature similarity threshold is found after matching the entire priority matching defect subset, the target partition image is determined to have no matching defects, and the process directly jumps to step S5 for verification matching of low-reliability regions. The above matching operation is performed in parallel on all partitioned images to obtain the Macro defect type of the corresponding location region of each partitioned image.

[0034] When a target partition image matches a macro defect type in a corresponding location region, it automatically checks several adjacent location regions. If the same macro defect type exists in consecutive adjacent location regions, it is determined that the target region has a clustered distribution of the same macro defect. The priority matching defect subsets of adjacent location regions with the same macro defect type are temporarily merged and deduplicated to form a joint priority matching subset for the defect cluster region. The ranking of this same macro defect type is promoted to the first position. Using the joint priority matching subset, the macro defect type matching is performed again for all location regions in the cluster region, and the macro defect type of this defect cluster region is output.

[0035] Based on the Macro defect type matching results of the corresponding regions or defect cluster regions of the above-mentioned partitioned images, the matched Macro defect type is output.

[0036] Step S5: Based on the wafer manufacturing process and material optical properties, calculate the matching reliability of the Macro defect type of the corresponding location region in each partition image. If the matching reliability of a certain region is less than the reliability threshold, call the preset global Macro defect set for matching and re-output the Macro defect type of the corresponding region. Furthermore, based on the wafer manufacturing process and material optical properties, the matching reliability of the Macro defect type in the corresponding region of each partition image is calculated. If the matching reliability of a certain region is less than the reliability threshold, a preset global Macro defect set is called for matching, and the Macro defect type of the corresponding region is re-output, including the following sub-steps: Step S51: Based on the wafer manufacturing process and material optical properties, calculate the matching reliability of the Macro defect type in the corresponding location area of ​​each partition image using a multi-source reliability fusion algorithm; Specifically, based on the wafer manufacturing process and material optical properties, a multi-source reliability fusion algorithm is used. Calculate the matching reliability of Macro defect types for corresponding regions in each image partition, where, For the corresponding location area of ​​the partitioned image The reliability of matching Macro defect types For matching enhancement coefficients, Matching similarity for Macro defect types, Configure weights for matching similarity. This represents the historical probability of the Macro defect type occurring in the corresponding region of the partitioned image. This is the smoothing coefficient for probability stability. Assign weights to the probability of historical occurrences. For spatial consistency, It is a very small positive number. It is a space-consistent lower bound protection function. Configure weights for spatial consistency. To adjust the slope, For process relevance, As a benchmark threshold for process relevance, Assign weights to process relevance. To configure the number of weights, The range of values ​​is .

[0037] Step S52: If the matching reliability of the Macro defect type in the corresponding region of the partitioned image is less than the preset reliability threshold, trigger the re-matching of the low reliability region, call the preset global Macro defect set for matching, and re-output the Macro defect type of the corresponding region. Specifically, a reliability threshold is dynamically set based on statistical analysis of historical review results. For any image region, if the matching reliability is higher than the reliability threshold, the matching result of the Macro defect type is considered highly reliable, and this result is directly used as the final output Macro defect type. If the matching reliability is lower than the reliability threshold, the matching result of the Macro defect type is considered insufficiently reliable, and a review matching is initiated for the corresponding region of the image partition. Image features are re-extracted from the image partition, and similarity calculation is performed with each Macro defect feature in the preset global Macro defect set. The Macro defect type with the highest similarity is selected as the final output Macro defect type for the region corresponding to the image partition. At the same time, the historical Macro defect dataset is automatically updated based on the relevant data from this review.

[0038] Example 2

[0039] like Figure 2 As shown, Embodiment 2 of this application provides a high-resolution camera for detecting macro defects in semiconductor manufacturing, comprising: Image acquisition and segmentation module 21 acquires a global image of the wafer and divides it into multiple partition images; Furthermore, the image acquisition module 21 includes the following sub-modules: The image acquisition submodule acquires high-resolution global images of the wafer using optical inspection equipment; The image segmentation submodule establishes a coordinate mapping based on wafer notches, divides the high-resolution global image into multiple partition images, and associates them with the wafer's coordinate positions; The defect probability list generation module 22, based on the location area on the wafer corresponding to each partition image, queries historical macro defect data to obtain the macro defect probability list for each location area. Furthermore, the defect probability list generation module 22 includes the following sub-modules: The historical macro defect dataset construction submodule collects historical macro defect data based on the entire historical wafer manufacturing process and constructs a historical macro defect dataset. The defect probability list acquisition submodule retrieves the probability list of macro defects for each location region by searching and statistically analyzing the historical macro defect dataset based on the location region on the wafer corresponding to each partition image. The priority matching defect subset construction module 23 dynamically constructs a priority matching defect subset for each location region based on the Macro defect probability list of each location region. Furthermore, the priority matching defect subset construction module 23 includes the following sub-modules: The initial candidate defect subset construction submodule filters the Macro defect probability list for each location region and constructs an initial candidate defect subset for each location region. The priority matching defect subset generation submodule adjusts and sorts the Macro defect types in the initial candidate defect subsets of each location region to generate the priority matching defect subsets for each location region. The defect type matching module 24 extracts the features of each partition image and performs fast matching with the priority matching defect subset of its corresponding location region, outputting the Macro defect type of the location region corresponding to each partition image; Furthermore, the defect type matching module 24 includes the following sub-modules: The image feature extraction submodule extracts image features from each partition image in parallel through a multi-scale adaptive feature extraction network. The Macro defect type acquisition submodule matches the image features of each partition image with the priority matching defect subset of its corresponding location region to obtain the Macro defect type of the corresponding location region of each partition image; The defect matching result verification module 25 calculates the matching reliability of the Macro defect type in the corresponding region of each partition image based on the wafer manufacturing process and material optical properties. If the matching reliability of a certain region is less than the reliability threshold, the preset global Macro defect set is called for matching, and the Macro defect type of the corresponding region is re-output. Furthermore, the defect matching result verification module 25 includes the following sub-modules: The matching reliability calculation submodule, based on the wafer manufacturing process and material optical properties, calculates the matching reliability of the Macro defect type in the corresponding location area of ​​each partition image through a multi-source reliability fusion algorithm. The verification matching submodule, if the matching reliability of the Macro defect type in the corresponding location area of ​​the partitioned image is less than the preset reliability threshold, triggers the verification matching of the low reliability area, calls the preset global Macro defect set for matching, and re-outputs the Macro defect type of the corresponding area; Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor; The memory is used to store one or more program instructions; A processor is used to run one or more program instructions to execute a macro defect detection method in semiconductor manufacturing.

[0040] Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are used by a processor to provide a method for detecting macro defects in semiconductor manufacturing.

[0041] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions that, when executed on a computer, cause the computer to perform the aforementioned method for detecting macro defects in semiconductor manufacturing.

[0042] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0043] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0044] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0045] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0046] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0047] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0048] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0049] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting macro defects in semiconductor manufacturing, characterized in that, include: Step S1: Acquire a global image of the wafer and divide it into multiple partition images; Step S2: Based on the location area on the wafer corresponding to each partition image, query historical macro defect data to obtain a list of macro defect probabilities for each location area; Step S3: Based on the Macro defect probability list for each location region, dynamically construct a priority matching defect subset for each location region; Step S4: Extract the features of each partition image, perform fast matching with the priority matching defect subset of its corresponding location region, and output the Macro defect type of the location region corresponding to each partition image. Step S5: Based on the wafer manufacturing process and material optical properties, calculate the matching reliability of the Macro defect type in the corresponding region of each partition image. If the matching reliability of a certain region is less than the reliability threshold, call the preset global Macro defect set for matching and re-output the Macro defect type of the corresponding region.

2. The method for detecting macro defects in semiconductor manufacturing as described in claim 1, characterized in that, Based on the location region on the wafer corresponding to each partition image, the historical macro defect data is queried to obtain a list of macro defect probabilities for each location region, including the following sub-steps: Step S21: Based on the entire historical wafer manufacturing process, collect historical macro defect data and construct a historical macro defect dataset; Step S22: Based on the location region on the wafer corresponding to each partition image, perform retrieval and statistics in the historical Macro defect dataset to obtain a list of Macro defect probabilities for each location region.

3. The method for detecting macro defects in semiconductor manufacturing as described in claim 1, characterized in that, Based on the Macro defect probability list for each location region, dynamically constructing a priority matching defect subset for each location region includes the following sub-steps: Step S31: Filter the list of Macro defects for each location region to construct an initial candidate defect subset for each location region; Step S32: Adjust and sort the Macro defect types in the initial candidate defect subsets of each location region to generate a priority matching defect subset for each location region.

4. The method for detecting macro defects in semiconductor manufacturing as described in claim 1, characterized in that, Extracting features from each partition image and performing fast matching with the priority matching defect subset of its corresponding location region, outputting the Macro defect type of the location region corresponding to each partition image includes the following sub-steps: Step S41: Extract image features of each partition image in parallel using a multi-scale adaptive feature extraction network; Step S42: Match the image features of each partition image with the priority matching defect subset of its corresponding location region to obtain the Macro defect type of the location region corresponding to each partition image.

5. The method for detecting macro defects in semiconductor manufacturing as described in claim 1, characterized in that, Based on the wafer manufacturing process and material optical properties, the matching reliability of the Macro defect type in the corresponding region of each partition image is calculated. If the matching reliability of a certain region is less than the reliability threshold, a preset global Macro defect set is called for matching, and the Macro defect type of the corresponding region is re-output. The process includes the following sub-steps: Step S51: Based on the wafer manufacturing process and material optical properties, calculate the matching reliability of the Macro defect type in the corresponding location area of ​​each partition image using a multi-source reliability fusion algorithm; Step S52: If the matching reliability of the Macro defect type in the corresponding region of the partitioned image is less than the preset reliability threshold, trigger the re-matching of the low reliability region, call the preset global Macro defect set for matching, and re-output the Macro defect type of the corresponding region.

6. A high-resolution camera for detecting macro defects in semiconductor manufacturing, characterized in that, include: The image acquisition and segmentation module acquires a global image of the wafer and divides it into multiple partition images; The defect probability list generation module retrieves the macro defect probability list for each location region by querying historical macro defect data based on the location area on the wafer corresponding to each partition image. The priority matching defect subset construction module dynamically constructs a priority matching defect subset for each location region based on the Macro defect probability list of each location region. The defect type matching module extracts the features of each partition image and performs fast matching with the priority matching defect subset of its corresponding location region, outputting the Macro defect type of the location region corresponding to each partition image; The defect matching result verification module calculates the matching reliability of the Macro defect type in the corresponding region of each partition image based on the wafer manufacturing process and material optical properties. If the matching reliability of a certain region is less than the reliability threshold, the preset global Macro defect set is called for matching, and the Macro defect type of the corresponding region is re-output.

7. A high-resolution camera for detecting macro defects in semiconductor manufacturing as described in claim 6, characterized in that, The defect probability list generation module specifically includes: The historical macro defect dataset construction submodule collects historical macro defect data based on the entire historical wafer manufacturing process and constructs a historical macro defect dataset. The defect probability list acquisition submodule retrieves the probability list of macro defects for each location region by searching and statistically analyzing the historical macro defect dataset based on the location region on the wafer corresponding to each partition image.

8. A high-resolution camera for detecting macro defects in semiconductor manufacturing as described in claim 6, characterized in that, Prioritize matching defect subset building modules, specifically including: The initial candidate defect subset construction submodule filters the Macro defect probability list for each location region and constructs an initial candidate defect subset for each location region. The priority matching defect subset generation submodule adjusts and sorts the Macro defect types in the initial candidate defect subsets of each location region to generate priority matching defect subsets for each location region.

9. A high-resolution camera for detecting macro defects in semiconductor manufacturing as described in claim 6, characterized in that, The defect type matching module specifically includes: The image feature extraction submodule extracts image features from each partition image in parallel through a multi-scale adaptive feature extraction network. The Macro defect type acquisition submodule matches the image features of each partition image with the priority matching defect subset of its corresponding location region to obtain the Macro defect type of the corresponding location region of each partition image.

10. A high-resolution camera for detecting macro defects in semiconductor manufacturing as described in claim 6, characterized in that, The defect matching result verification module specifically includes: The matching reliability calculation submodule, based on the wafer manufacturing process and material optical properties, calculates the matching reliability of the Macro defect type in the corresponding location area of ​​each partition image through a multi-source reliability fusion algorithm. The verification matching submodule, if the matching reliability of the Macro defect type in the corresponding location area of ​​the partitioned image is less than the preset reliability threshold, triggers the verification matching of the low reliability area, calls the preset global Macro defect set for matching, and re-outputs the Macro defect type of the corresponding area.

Images