Wafer image clustering method and device, equipment and storage medium

Abstract

The application provides a wafer image clustering method and device, electronic equipment and computer readable storage medium. The method comprises: generating a binary image corresponding to a wafer image based on the wafer image containing a defective die; generating topological structure information of the defective die based on pixel information of the defective die in the binary image; generating a contour image corresponding to the binary image based on the topological structure information; performing pixel filling on a contour region in the contour image based on a contour list corresponding to a contour curve in the contour image to generate a filled image; and marking each contour region in the filled image as a clustering result. In this way, the wafer image can be quickly clustered without any prior setting, the time cost required by the clustering algorithm is reduced, and the clustering efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of wafer inspection technology, and in particular to a method, apparatus, electronic device and computer-readable storage medium for clustering wafer images. Background Technology

[0002] The semiconductor chip industry is a vital pillar of the global economy, providing new solutions for various fields and playing an indispensable role in modern society. Wafers are the fundamental raw material for semiconductor devices; as manufacturing technology continues to shrink in size, the manufacturing process becomes increasingly complex and prone to defects. Defects on wafers are particularly susceptible to aggregated defects, which can lead to complete chip failure.

[0003] A wafer image is an image created during the wafer inspection and testing phase by recording test results at spatial locations on the wafer using methods such as probe testing, electrical testing, or defect detection. Transforming the defect clusters observed on the wafer image into meaningful information or knowledge is crucial for improving semiconductor processes and increasing yield.

[0004] However, existing traditional wafer image clustering methods require significant time to cluster defective grains. Therefore, reducing the time cost of traditional clustering algorithms to achieve fast and accurate clustering has become a current research hotspot. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and computer-readable storage medium for clustering wafer images, which can achieve rapid clustering of wafer images without any pre-setting, and reduce the time cost required by the clustering algorithm and improve clustering efficiency.

[0006] According to a first aspect of this application, a clustering method for wafer images is provided. The method includes: generating a binary image corresponding to the wafer image based on a wafer image containing defective granules; generating topological structure information of the defective granules based on pixel information of the defective granules in the binary image; generating a contour image corresponding to the binary image based on the topological structure information; filling the contour regions in the contour image with pixels based on a contour list corresponding to the contour curves in the contour image to generate a filled image; and marking each contour region in the filled image as a clustering result.

[0007] According to a second aspect of this application, a wafer image clustering apparatus is provided. The apparatus includes: a binary image generation module for generating a binary image corresponding to a wafer image based on a wafer image containing defective dies; a structure information generation module for generating topological structure information of defective dies based on pixel information of defective dies in the binary image; a contour image generation module for generating a contour image corresponding to the binary image based on the topological structure information; a filled image generation module for filling contour regions in the contour image with pixels based on a contour list corresponding to contour curves in the contour image to generate a filled image; and a labeling module for labeling each contour region in the filled image as a clustering result.

[0008] According to a third aspect of this application, an electronic device is provided, comprising: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements any of the above-described wafer image clustering methods.

[0009] According to a fourth aspect of this application, a computer-readable storage medium is provided, characterized in that computer program instructions are stored on the computer-readable storage medium, which, when executed by a processor, implement any of the above-described wafer image clustering methods.

[0010] In summary, the wafer image clustering method, apparatus, electronic device, and computer-readable storage medium provided in this application have at least the following beneficial effects: By generating a binary image corresponding to the wafer image, the grains and defects in the wafer image can be clearly distinguished, which helps to more accurately identify and locate defective grains. By generating topological structure information of defective grains based on the pixel information of defective grains in the binary image, defective grains with similar characteristics can be connected together, which facilitates subsequent clustering. Based on the topological structure information, a contour image corresponding to the binary image is generated, which can provide a clearer and more intuitive distinction of wafer surface defects and is beneficial to subsequent clustering operations. Based on the contour list corresponding to the contour curves in the contour image, the contour regions in the contour image are pixel-filled to generate a filled image, and each contour region in the filled image is marked as a clustering result. In this way, the interior of the contour is filled to form the clustering result of the image, thereby achieving the effect of fast clustering of wafer images without any pre-setting, reducing the time cost required by the clustering algorithm, and improving the clustering efficiency. Attached Figure Description

[0011] To more clearly illustrate the specific embodiments of this application or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This diagram illustrates a flowchart of a wafer image clustering method provided in an embodiment of this application.

[0013] Figure 2 This diagram illustrates a topology provided in an embodiment of this application.

[0014] Figure 3 This diagram illustrates a contour image provided in an embodiment of this application.

[0015] Figure 4 This diagram illustrates a filled image provided in an embodiment of this application;

[0016] Figure 5 This diagram illustrates a filled image for marking risk levels according to an embodiment of this application;

[0017] Figure 6 This diagram illustrates a flowchart of another wafer image clustering method provided in an embodiment of this application.

[0018] Figure 7 This diagram illustrates the structure of a wafer image clustering device according to an embodiment of this application.

[0019] Figure 8 This diagram illustrates the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the above and other features and advantages of this application clearer, the application is further described below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for the purpose of explanation to those skilled in the art, and are exemplary only, not restrictive.

[0021] In the following description, numerous specific details are set forth to provide a thorough understanding of this application. However, it will be apparent to those skilled in the art that the specific details are not required to practice this application. In other instances, well-known steps or operations have not been described in detail to avoid obscuring this application.

[0022] Currently, the K-means algorithm is commonly used for clustering wafer images. The K-means algorithm is an iterative clustering analysis algorithm. Its steps are as follows: The data is pre-divided into K groups; K objects are randomly selected as the initial cluster centers; then, the distance between each object and each seed cluster center is calculated, and each object is assigned to the nearest cluster center. The cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned to a cluster, the cluster centers are recalculated based on the existing objects in the cluster. This process is repeated until a termination condition is met. The termination condition may be that no (or a minimum number) objects are reassigned to different clusters, no (or a minimum number) cluster centers change, or the sum of squared errors reaches a local minimum.

[0023] However, the above clustering methods require a pre-defined K value, which represents the number of clusters in the final clustering result. The magnitude of the K value directly determines the final clustering effect. But in general, users have no way of knowing the specific K value and need to manually try and find the optimal K value, which requires a lot of time and manpower.

[0024] In view of this, the inventors propose a fast and accurate wafer image clustering method. This method maps the effective information of the wafer into a wafer image, extracts contours using image processing algorithms, and then fills in the contours to form the image clustering result. This allows for rapid clustering of wafer images without any pre-setting. Furthermore, facing the challenges of continuously increasing wafer production sizes and the growing number of defective die points, this method significantly reduces the time cost of traditional clustering algorithms while maintaining clustering effectiveness, thus improving clustering efficiency.

[0025] Meanwhile, the inventors also used the clustering results to mark the potentially risky chips on the wafer, thereby classifying them into different risk levels. This reduces sealing risks and packaging costs to some extent.

[0026] One embodiment of this application provides a method for clustering wafer images. Figure 1 This diagram illustrates a flowchart of a wafer image clustering method provided in an embodiment of this application. Figure 1 The clustering method for the wafer image may include the following steps.

[0027] S11, Based on the wafer image containing defective grains, generate a binary image corresponding to the wafer image.

[0028] Here, the wafer image can be an image with defective grains already marked. There are various ways to mark grains in a wafer image, such as eigenvalue marking and symbol marking. Eigenvalue marking refers to setting type eigenvalues ​​for grains, using different eigenvalues ​​to represent defective grains and normal grains. It should be noted that grain marking is one of the eigenvalues ​​of a grain.

[0029] The "binary image" involved in this application embodiment can be called a binarized image or a binary image, which is an image in which each pixel has only two possible values ​​or grayscale levels. Each pixel in a binary image can represent a grain. Different types of grains can be represented by different pixel values. For example, the pixel value of a defective grain is a non-zero value, while the pixel value of a normal grain is zero. To better distinguish grain types in a binary image, the non-zero value can optionally be set to 128 or higher.

[0030] In this embodiment, an image segmentation method can be used to separate the dies from the background in the wafer image. Then, the pixel value of each die is determined according to its type, thereby generating a binary image. The die type can be determined based on the die's markings.

[0031] S12, Based on the pixel information of the defective grains in the binary image, generate the topological structure information of the defective grains.

[0032] Here, pixel information can be pixel-related information, including but not limited to pixel coordinates, grain feature values, etc. Topology information can be information related to the topology in the binary image, and can include the topology itself and a list of topologies. Each topology includes at least one defective grain. The list of topologies includes pixel information of the defective grains constituting each topology.

[0033] In this embodiment, the pixel information of each defective grain is identified from the binary image, the connection relationship between defective grains is determined, and the topological structure corresponding to the defective grain is constructed based on the connection relationship to generate topological structure information. In other words, the defective grain in the binary image can be regarded as a series of points, and the topological structure is constructed and topological structure information is generated based on the connection relationship between the points.

[0034] Figure 2 This illustration shows a schematic diagram of a topology provided in an embodiment of this application, such as... Figure 2 The binary image contains only one circular topological structure. It should be noted that, to highlight the topological structure, Figure 2 Normal grains are not marked; only black areas are used to represent normal grain areas.

[0035] S13, Based on the topological information, generate the contour image corresponding to the binary image.

[0036] Here, the contour image can be an image that includes topological edge information and contains only two pixel values. Furthermore, the feature values ​​of each grain in the contour image can also include contour feature values ​​to indicate whether it lies on the contour.

[0037] In this embodiment of the application, the contour curve of each topological structure can be extracted using a contour extraction method to generate a contour image corresponding to the binary image. For example... Figure 2 As shown, the topology may include an inner contour and an outer contour.

[0038] S14, Based on the contour list corresponding to the contour curve in the contour image, fill the contour region in the contour image with pixels to generate a filled image.

[0039] Here, the contour list can include pixel information of the defective grains that constitute the contour curve, and can represent the shape and location of the contour. The contour list can be generated when extracting the contour curve. The contour region can be the area enclosed by the contour curve.

[0040] In this embodiment, pixel regions that need to be filled within the contour area are determined based on the contour list. These pixel regions can be areas containing pixels located within the contour area and marked as defective grains. The determined pixel regions are then filled to generate a filled image. Figure 3 This diagram illustrates a contour image provided in an embodiment of this application. Figure 4 This diagram illustrates a filled image provided in an embodiment of this application.

[0041] It should be noted that, in the embodiments of this application, the pixel filling method can be to fill with non-zero pixel values.

[0042] S15, label each contour region in the filled image as a cluster result.

[0043] Here, each clustering result is a cluster. In this embodiment, a unique label is assigned to each contour region in the filled image to mark the clustering result of the contour region. That is, a contour region is clustered into a cluster.

[0044] In the above embodiments, by generating a binary image corresponding to the wafer image, the grains and defects in the wafer image can be clearly distinguished, which helps to more accurately identify and locate defective grains. Based on the pixel information of the defective grains in the binary image, topological structure information of the defective grains is generated, allowing defective grains with similar characteristics to be connected together, facilitating subsequent clustering. Based on the topological structure information, a contour image corresponding to the binary image is generated, thus providing a clearer and more intuitive distinction of wafer surface defects and benefiting subsequent clustering operations. Based on the contour list corresponding to the contour curves in the contour image, the contour regions in the contour image are pixel-filled to generate a filled image, and each contour region in the filled image is marked as a clustering result. In this way, by filling the contour interiors, the clustering results of the image are formed, thereby achieving the effect of fast clustering of wafer images without any pre-setting, reducing the time cost required by the clustering algorithm, and improving clustering efficiency.

[0045] The inventors discovered that because the energy and focus during wafer production have strong continuity, the grains surrounding defective grains that can form clusters also have certain risks. Even if no problems are found in the current inspection, there are still risks for subsequent use, and the closer to the cluster, the higher the risk.

[0046] Therefore, in some embodiments, after marking each contour region in the filled image as a clustering result in S15, the clustering method of the wafer image may further include: determining the cluster type of each clustering result based on the number of defective grains in each clustering result, wherein the cluster type includes non-noise clusters; and determining the risk level of adjacent normal grains based on the distance between adjacent normal grains and non-noise clusters in the filled image.

[0047] Here, the cluster type can include non-noise clusters and noise clusters. In this application embodiment, "nearby normal grains" can be normal grains whose distance from non-noise clusters is less than a preset distance. The preset distance is not less than 1, and can be set according to user requirements.

[0048] In this embodiment, defective grains in each cluster are counted. Based on the statistical results of the number of defective grains, each cluster is then labeled as a noise cluster or a non-noise cluster. A noise cluster can be a cluster with fewer than a preset number of defective grains, while a non-noise cluster can be a cluster with at least a preset number of defective grains.

[0049] Then, each non-noise cluster is traversed, and the surrounding normal grains centered on each non-noise cluster are found and identified as neighboring normal grains. The distance from each neighboring grain to its non-noise cluster is determined, and the risk level of the neighboring normal grains is determined based on the distance.

[0050] In some embodiments, after determining the risk level, the pixel values ​​of normal grains can be set according to the risk level so that grains of different risk levels can be displayed on the filled image. Furthermore, risk feature values ​​in the pixel information of the grains can also be set according to the risk level.

[0051] In one example, a neighboring normal grain that is 1 distance away from a non-noise cluster is marked as risk 1, which is the highest risk level; a neighboring normal grain that is 2 distance away from a non-noise cluster is marked as risk 2, which is a higher risk level... and so on, until all neighboring normal grains that are at a preset distance from the non-noise cluster are found and marked. Figure 5 This diagram illustrates a filled image for marking risk levels according to an embodiment of this application. Figure 5 As shown, different pixel values ​​are used to mark normal grains with different risk levels.

[0052] In the above embodiments, by determining the cluster type based on the number of defective dies, the type of clustering results in the wafer image can be identified more accurately. Furthermore, by determining the risk level based on the distance between neighboring normal dies and non-noise clusters, the potential risks to normal dies in the wafer can be assessed. This is significant for predicting and preventing potential defect propagation and evaluating wafer quality, thereby reducing packaging costs to some extent. Simultaneously, the generated risk feature values ​​can provide support for subsequent decision-making and visually display the defect distribution and risk situation of the wafer, facilitating further analysis and research.

[0053] In some embodiments, the topology information includes at least one topology structure. S13, generating a contour image corresponding to the binary image based on the topology structure information may include: determining edge defect grains in each topology structure based on the topology structure information; connecting the edge defect grains in each topology structure to form a contour curve corresponding to each topology structure; setting the pixel values ​​of defect grains in the binary image other than the defect grains on the contour curve to zero, thereby generating the contour image.

[0054] Here, edge defect grains can refer to defect grains located at the edges of a topology. Topology information can include at least one topology and a list of topologies corresponding to each topology. The list of topologies includes pixel information for at least one defect grain.

[0055] In this embodiment, each topology in the topology information is analyzed, and the search begins along the edge of the topology to track edge defect grains. A contour curve corresponding to each topology is then drawn, and contour information is generated simultaneously. The contour information may include an initial contour list corresponding to each contour curve.

[0056] In addition, during the process of drawing the contour curve, it is also necessary to determine whether certain edge defect grains belong to the same contour. Specifically, this can be done by determining whether there are connection points in the neighborhood of these edge defect grains. If so, then these edge defect grains belong to the topology.

[0057] The "edge of the topology" in this application embodiment includes an inner edge and an outer edge. In this application embodiment, the topology may correspond to one outer edge. Alternatively, the topology may correspond to one outer edge and at least one inner edge. Thus, the topology may correspond to an outer contour curve. Alternatively, the topology may correspond to one outer contour curve and at least one inner contour curve, such as... Figure 2 As shown.

[0058] It should be noted that other methods, such as edge detection, can also be used to extract the edge contour of the topology, and this application does not impose any restrictions on these methods.

[0059] Because binary images can contain, for example Figure 2 The illustrated topology, a "donut" shape, may affect subsequent contour filling if only the outer contour is extracted. Therefore, in this embodiment, all edge contours of each topology need to be extracted, while preserving the hierarchical relationship between contours. For example, the inner contour is the child contour, and the outer contour is the parent contour. Furthermore, the hierarchical relationship between contours can be set in the initial contour list.

[0060] The "contour image" involved in the embodiments of this application can be an image in which the pixel value of defective grains on the contour curve is non-zero, the pixel value of normal grains is 0, and the pixel value of defective grains located outside the contour curve is also 0.

[0061] It should be noted that when setting the defective grains other than the contour curve to 0, the pixel values ​​of these defective grains need to be changed, but other feature values ​​in the pixel information remain unchanged.

[0062] In the above embodiments, by utilizing topological information, edge defect grains in the wafer image can be identified more accurately, thereby quickly generating contour curves. By setting defect grains other than those on the contour curve to 0, the contour curve can be better highlighted, facilitating subsequent rapid and accurate clustering. Simultaneously, the generated contour image can visually display the distribution of defect grains in the wafer, thus helping analysts better understand the defect distribution characteristics and quickly identify potential problem areas.

[0063] To further ensure the accuracy of subsequent clustering structures, in some embodiments, the topology information may include a first topology and a second topology. In S14, before filling the contour regions in the contour image with pixels based on the contour list corresponding to the contour curves in the contour image to generate a filled image, the clustering method for the wafer image may further include: extracting an initial contour list for each contour curve in the contour image; performing deduplication on each initial contour list to obtain an intermediate contour list corresponding to each initial contour list; merging at least two intermediate contour lists with the same defective grains corresponding to the first topology to obtain a contour list corresponding to the first topology; and using the intermediate contour list corresponding to the second topology as the contour list of the second topology.

[0064] Here, the initial contour list can be a list corresponding to the contour curve, which can include pixel information of all defective grains on the contour curve. The intermediate contour list can be the initial contour list after deduplication.

[0065] In this embodiment, to remove duplicate defect grains from each initial contour list, each initial contour list can be deduplicated to obtain an intermediate contour list corresponding to each initial contour list. The number of initial contour lists is the same as the number of intermediate contour lists; that is, one initial contour list corresponds to one intermediate contour list.

[0066] The "first topology" in the embodiments of this application can refer to a topology that includes at least two contour curves. These at least two contour curves may include an outer contour curve and at least one inner contour curve.

[0067] Due to the unique characteristics of the first topology, when extracting all contours of the first topology, at least two initial contour lists are generated, representing an outer contour curve and at least one inner contour curve of the topology, and these initial contour lists have a hierarchical relationship. For example, the initial contour list of the outer contour curve is the parent contour list of the initial contour list of the inner contour curve, and the initial contour list of the inner contour curve is the child contour list of the initial contour list of the outer contour curve.

[0068] It should be noted that, from an image perspective, the multiple initial contour lists corresponding to the first topological structure represent the same image pattern. Therefore, the parent contour list and child contour list corresponding to the first topological structure have duplicate elements.

[0069] Since one intermediate contour list corresponds to one initial contour list, the first topology corresponds to at least two intermediate contour lists, and the hierarchical relationship between these intermediate contour lists is consistent with the hierarchical relationship between their corresponding initial contour curves.

[0070] In this embodiment, at least two intermediate contour lists corresponding to the first topology are merged, and duplicate defective grains are removed to obtain a complete contour list of the first topology. This ensures the integrity of the contour list of the first topology and guarantees the accuracy of subsequent clustering results.

[0071] The "second topology" involved in this application embodiment can be a solid topology, that is, a topology that includes a contour curve. This contour curve is an outer contour curve. Therefore, the list of intermediate contours corresponding to the second topology can be used as the final contour list of the second topology.

[0072] It should be noted that the first and second topologies do not refer to quantity, but rather to different types of topologies.

[0073] In the above embodiments, the deduplication process ensures that the defect grains in each initial contour list appear only once, and the two contour lists with the same topology are merged to remove duplicate defect grains, resulting in a complete contour list. This ensures the accuracy of subsequent clustering results.

[0074] The inventors discovered that isolated points can affect the running speed of clustering algorithms, especially for wafer images generated from defect detection results, where the number of isolated points is relatively large, which can have a significant impact on the algorithm's running speed.

[0075] Therefore, in some embodiments, before performing pixel filling on the contour region in the contour image based on the contour list corresponding to the contour curve in the contour image in S14 to generate a filled image, the clustering method of the wafer image may further include: filtering out isolated points in the contour image based on the contour list corresponding to the contour curve in the contour image to obtain a filtered contour image.

[0076] Here, an isolated point in the contour image can be a contour curve containing only one defective grain. In this embodiment, by detecting the number of defective grains in all contour lists in the contour image, when the contour list contains only one defective grain, the defective grain can be marked as an isolated point, that is, an isolated feature value in the pixel information of the defective grain is set to "yes". The isolated feature value is used to characterize whether it is an isolated point.

[0077] Additionally, the pixel values ​​of defective pixels marked as isolated points are set to 0, thus filtering out isolated points and obtaining the filtered contour image. Simultaneously, the contour list corresponding to the isolated points is deleted.

[0078] In the above embodiments, filtering out isolated points in the contour image can reduce unnecessary subsequent calculations, thereby improving the overall running speed of the algorithm.

[0079] In some embodiments, S14, based on the contour list corresponding to the contour curve in the contour image, pixel filling is performed on the contour region in the contour image to generate a filled image. This may include: sorting all contour lists in the filtered contour image according to the number of defective grains; and sequentially performing pixel filling on the contour region corresponding to each contour list in ascending order to generate a filled image.

[0080] Here, the sorting can be either ascending or descending. Ascending order means arranging items from fewest to most, while descending order means arranging items from most to fewest.

[0081] The "contour region" mentioned in the embodiments of this application can refer to the region contained by the contour curve, including the region where the contour curve is located. The contour region can be determined by a topology list.

[0082] A filled image is an image created by filling a contoured region with pixels. Compared to... Figure 3 , Figure 4 The outline area has been filled, and Figure 4 Only non-noise clusters are shown.

[0083] In this embodiment, the filtered contour images are sorted according to the number of defective grains in each contour list. If the sorting is ascending, the first contour list has the fewest defective grains, the second list has the second fewest, and so on, with the last contour list having the most. If the sorting is descending, the first list has the most defective grains, the second list has the second fewest, and so on, with the last contour list having the most.

[0084] Then, in order of increasing number of defective grains, the contour regions corresponding to each contour list are filled with pixels in sequence. That is, the contour regions corresponding to the contour list with the fewest defects are filled with pixels first, then the contour regions corresponding to the second-to-last fewest defects are filled with pixels, and so on. Finally, the contour regions corresponding to the contour list with the most defects are filled with pixels, thereby obtaining the filled image.

[0085] In the above embodiments, by sorting all the contour lists, pixel filling of the contour regions can be performed in an orderly manner. Furthermore, filling the smaller contour regions first reduces the number of filling steps for the larger contour regions, thereby simplifying the filling process and enabling faster completion of the pixel filling task, reducing computational complexity and time costs.

[0086] For example, for a "donut" shaped contour curve, since this contour curve is generally generated around the outer edge of the wafer image, the number of defective grains in the contour list extracted from the contour curve is relatively large. Filling the contour area according to the number of defective grains from few to many can ensure that the defective grains within the upper and lower limits of the "donut" shaped contour area are filled into the contour area before the area is filled. In this way, the filling time of this area can be shortened and the overall filling rate can be improved.

[0087] In some embodiments, sequentially filling the contour regions corresponding to each contour list with pixels in ascending order to generate a filled image may include: determining the target defect grains in the contour region corresponding to the Nth contour list in ascending order; obtaining the filling feature value of each target defect grain; determining the target defect grain as the grain to be filled when the filling feature value is not selected, and updating the filling feature value to selected; updating the pixel value of the grain to be filled to a preset pixel value to fill the corresponding contour region; and generating a filled image based on the filled contour region corresponding to each contour list.

[0088] Here, the target defective grain can be a defective grain with a pixel value of zero. The fill feature value can be a feature value belonging to the pixel information, used to indicate whether it is selected for filling. The grain to be filled is the pixel that needs to be filled. N is an integer not less than 1. The preset pixel value can be a non-zero pixel value.

[0089] In this embodiment, following the order of increasing number of defective grains, the following filling operations are performed sequentially on each contour region: Defective grains with a pixel value of zero in the corresponding contour region of the contour list are identified and designated as target defective grains. Filling feature values ​​are extracted from the pixel information of each target defective grain, and it is detected whether the filling feature value of each target defective grain is unselected. When the filling feature value is unselected, the target defective grain can be considered a region that needs to be filled; that is, target defective grains with unselected filling feature values ​​meet the filling conditions and need to be filled. The pixel value of the target defective grains with unselected filling feature values ​​is updated from 0 to a preset pixel value, and the filling feature value is updated to selected. When the filling feature value is selected, no operation is performed on the corresponding target defective grain, and this target defective grain is skipped. This process continues until all target defective grains in the contour region have been detected and updated, resulting in the filled contour region.

[0090] By analogy, the filled contour regions corresponding to all contour regions can be obtained, thus obtaining the filled image.

[0091] It should be noted that to find defective grains with a pixel value of zero in the contour region corresponding to the contour list, you can either search according to the topology list corresponding to the contour list, or determine that the defective grain belongs to the contour region by judging that a certain contour curve has limit values ​​in both the horizontal and vertical coordinate directions of a defective grain.

[0092] In addition, the default setting for the defect grain filling characteristic value is not selected.

[0093] In the above embodiments, the filling characteristic value of defective grains is used to indicate whether pixel filling is to be performed, which can quickly fill the contour area, improve the filling rate, and thus speed up the clustering speed.

[0094] Alternatively, you can first identify the target defect grains in all contour regions, then update the filling feature value of the target defect grains corresponding to each contour region to be selected, and finally update all filling feature values ​​to the pixel value of the selected target defect grains and set it to the preset pixel value to obtain the filled image.

[0095] In one example, during contour filling, the feature value of each defect grain is evaluated. If the fill feature value of the defect grain is positive, the defect grain is skipped. If the defect grain is determined to belong to the contour, it is added to the contour by updating its pixel value from 0 to a non-zero pixel value and updating its fill feature value from negative to positive. The result after contour filling is the clustering result for that cluster.

[0096] In some embodiments, S11, generating a binary image corresponding to the wafer image based on the wafer image containing defective grains, may include: obtaining a first limit value of the grains in the wafer image in the horizontal direction and a second limit value in the vertical direction; obtaining the size of the binary image corresponding to the wafer image based on the first limit value and the second limit value; determining the pixel information of each grain in the wafer image in the binary image based on the first limit value and the second limit value; and determining the pixel value of each grain in the binary image based on the type of each grain in the wafer image.

[0097] Here, the first limit value can include the upper and lower limits in the horizontal direction, and the second limit value can include the upper and lower limits in the vertical direction. The grain type can include normal grains and defective grains.

[0098] In this embodiment, the indices of the spatial positions of all dies on the wafer image are checked to confirm the maximum and minimum values ​​of the dies in the horizontal and vertical directions. Then, the length of the binary image is set to the difference between the maximum and minimum values ​​in the horizontal direction, and the width is set to the difference between the maximum and minimum values ​​in the vertical direction. Next, the difference between the current die's horizontal position and its minimum value, and the difference between its vertical position and its maximum value are calculated. The result of these two directional differences is the pixel coordinate of the die in the binary image. This process is repeated to calculate the pixel coordinates of all dies in the binary image. Based on the dies' characteristic values, normal dies and defective dies are distinguished. The pixel values ​​of defective dies are set to non-zero pixel values, and those of normal dies are set to 0, thus obtaining the binary image.

[0099] It should be noted that, compared with the binary image, the only difference between the wafer image and the image is the change in pixel coordinates and pixel values; the other feature values ​​in the pixel information of the die remain unchanged.

[0100] In the above embodiments, the pixel information of each die in the wafer image in the binary image can be accurately determined by the first limit value and the second limit value, thereby accurately locating the position of the die in the binary image and generating binary images of wafer images of different sizes and types, improving flexibility and adaptability, and obtaining binary images without complex processing, thereby simplifying the binary image generation process.

[0101] To further understand the wafer image clustering method provided in the embodiments of this application, another wafer image clustering method is also provided in another aspect of the embodiments of this application. Figure 6 This document illustrates a flowchart of another wafer image clustering method provided in an embodiment of this application, as shown below. Figure 6 As shown, the clustering method for this wafer image may include the following steps.

[0102] S61 generates a binary image of the wafer image.

[0103] S62, extract the pattern contour (i.e., contour curve) of defect grains in the binary image.

[0104] S63, remove duplicate values ​​from each outline list.

[0105] S64, merge lists of outlines with the same elements.

[0106] S65, determine if the number of pixels (i.e., defective grains) in the contour list is 1. If not, jump to S66; if yes, jump to S67.

[0107] S66, filter a list of outlines with a filter quantity of 1.

[0108] S67, sort by the number of pixels in the outline list, then jump to S68.

[0109] S68, iterate through the contour regions corresponding to each contour list.

[0110] S69, determine whether the defective grains meet the filling conditions (i.e., belong to a certain contour region and are not marked). If yes, jump to S610. If no, jump to S611.

[0111] Here, "mark" refers to filling a feature value of "yes".

[0112] S610, fill the defective grain into the outline and mark the defective grain, then jump to S611.

[0113] Here, filling the contour with defective grains can be understood as updating the pixel values ​​of the defective grains to non-zero pixel values.

[0114] S611 forms clusters (i.e., clustering results) and filters them based on the number of defective grains contained within.

[0115] S612, find normal grains whose spatial location from the filtered clusters meets the distance requirement.

[0116] S613, based on the search results, classify different risk levels according to distance.

[0117] In the above embodiments, clustering of dies on a wafer image does not require any pre-setting. By processing at the image level, it can minimize the time cost of traditional clustering algorithms while ensuring clustering effectiveness, especially given the challenges of increasing wafer production sizes and the growing number of defective dies. Furthermore, by using the clustering results to mark potentially risky dies on the wafer and classify them into different risk levels, packaging costs can be reduced to some extent.

[0118] Another aspect of this application provides a clustering apparatus for wafer images. Figure 7 This diagram illustrates the structure of a wafer image clustering device according to an embodiment of this application. Figure 7 As shown, the wafer image clustering device 70 may include the following modules.

[0119] Binary image generation module 71 is used to generate a binary image corresponding to a wafer image based on a wafer image containing defective grains.

[0120] The structural information generation module 72 is used to generate the topological structure information of the defective grains based on the pixel information of the defective grains in the binary image.

[0121] The contour image generation module 73 is used to generate a contour image corresponding to a binary image based on topological information.

[0122] The filled image generation module 74 is used to fill the contour regions in the contour image with pixels based on the contour list corresponding to the contour curves in the contour image, so as to generate a filled image.

[0123] The labeling module 75 is used to label each contour region in the filled image as a clustering result.

[0124] In the above embodiments, by generating a binary image corresponding to the wafer image, the grains and defects in the wafer image can be clearly distinguished, which helps to more accurately identify and locate defective grains. Based on the pixel information of the defective grains in the binary image, topological structure information of the defective grains is generated, allowing defective grains with similar characteristics to be connected together, facilitating subsequent clustering. Based on the topological structure information, a contour image corresponding to the binary image is generated, thus providing a clearer and more intuitive distinction of wafer surface defects and benefiting subsequent clustering operations. Based on the contour list corresponding to the contour curves in the contour image, the contour regions in the contour image are pixel-filled to generate a filled image, and each contour region in the filled image is marked as a clustering result. In this way, by filling the contour interiors, the clustering results of the image are formed, thereby achieving the effect of fast clustering of wafer images without any pre-setting, reducing the time cost required by the clustering algorithm, and improving clustering efficiency.

[0125] In some embodiments, the wafer image clustering device 70 may further include the following modules.

[0126] The cluster type determination module is used to determine the cluster type of each cluster result based on the number of defective grains in each cluster result after each contour region in the filled image is labeled as a cluster result, wherein the cluster type includes non-noise clusters.

[0127] The risk determination module is used to determine the risk level of neighboring normal grains based on the distance between neighboring normal grains and non-noise clusters in the filled image.

[0128] In some embodiments, the topology information includes at least one topological structure. The contour image generation module 73 may include the following sub-modules.

[0129] The edge grain determination submodule is used to determine the edge defect grains in each topology based on topology information.

[0130] The contour curve forming submodule is used to connect the edge defect grains in each topology to form the contour curve corresponding to each topology.

[0131] The contour image generation submodule is used to set the pixel values ​​of defect grains in the binary image (excluding defect grains on the contour curve) to zero, thereby generating a contour image.

[0132] In some embodiments, the topology information includes a first topology and a second topology. The wafer image clustering device 70 may also include the following modules.

[0133] The initial list extraction module is used to extract the initial contour list of each contour curve in the contour image before generating the filled image by filling the contour region in the contour image with pixels based on the contour list corresponding to the contour curve in the contour image.

[0134] The intermediate list acquisition module is used to perform deduplication processing on each of the initial contour lists to obtain an intermediate contour list corresponding to each of the initial contour lists.

[0135] The list merging module is used to merge at least two intermediate contour lists with the same defective grains corresponding to the first topology to obtain the contour list corresponding to the first topology.

[0136] The list is used as a module to take the list of intermediate contours corresponding to the second topology as the contour list of the second topology.

[0137] In some embodiments, the wafer image clustering device 70 may further include the following modules.

[0138] The filtering module is used to fill the contour regions in the contour image with pixels based on the contour list corresponding to the contour curve in the contour image, so as to filter the isolated points in the contour image before generating the filled image.

[0139] In some embodiments, the image filling generation module 74 may further include the following sub-modules.

[0140] The sorting submodule is used to sort all the contour lists in the filtered contour image according to the number of defective grains.

[0141] The fill submodule is used to fill the corresponding contour regions of each contour list with pixels in ascending order of number of pixels to generate a filled image.

[0142] In some embodiments, the filling submodule may include the following units.

[0143] The target grain determination unit is used to determine the target defect grains in the contour region corresponding to the Nth contour list in ascending order, wherein the target defect grains are defect grains with a pixel value of zero.

[0144] The feature value acquisition module is used to acquire the filling feature value of each target defect grain.

[0145] The update unit is used to update the pixel value of the target defect grain to a preset pixel value to fill the corresponding contour area when the fill feature value is not selected, and to update the fill feature value to be selected.

[0146] The generation unit is used to generate a filled image based on the filled contour region corresponding to each contour list.

[0147] In some embodiments, the binary image generation module may include the following sub-modules.

[0148] The limit value acquisition submodule is used to acquire the first limit value of the grain in the horizontal direction and the second limit value in the vertical direction in the wafer image;

[0149] The size acquisition submodule is used to obtain the size of the binary image corresponding to the wafer image based on the first limit value and the second limit value;

[0150] The information determination submodule is used to determine the pixel information of each grain in the wafer image in the binary image based on the first limit value and the second limit value.

[0151] The pixel value determination submodule is used to determine the pixel value of each die in the binary image based on the type of each die in the wafer image.

[0152] It should be understood that the specific features, operations, and details described herein with respect to the methods of this application can also be similarly applied to the apparatus and system of this application, or vice versa. Furthermore, each step of the methods of this application described above can be performed by a corresponding component or unit of the apparatus or system of this application.

[0153] It should be understood that the various modules / units of the device of this application can be implemented wholly or partially through software, hardware, firmware, or a combination thereof. Each module / unit can be embedded in the processor of the electronic device in hardware or firmware form or independent of the processor, or it can be stored in the memory of the electronic device in software form for the processor to call to execute the operation of each module / unit. Each module / unit can be implemented as an independent component or module, or two or more modules / units can be implemented as a single component or module.

[0154] Another aspect of this application provides an electronic device. Figure 8This diagram illustrates the structure of an electronic device according to an embodiment of this application. Figure 8 As shown, the electronic device 80 includes a processor 81 and a memory 82 storing computer program instructions. The processor 81 executes the computer program instructions to implement the steps of the aforementioned wafer image clustering method. This electronic device 80 can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities.

[0155] In one embodiment, the electronic device 80 may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the electronic device 80 can be used to provide necessary computing, processing, and / or control capabilities. The memory of the electronic device 80 may include non-volatile storage media and internal memory. The non-volatile storage media may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface and communication interface of the electronic device 80 can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the wafer image clustering method of this application.

[0156] This application provides a computer-readable storage medium storing a clustering program for wafer images, wherein when the clustering program for wafer images is executed by a processor, it can implement the steps of any of the wafer image clustering methods provided in the examples of this application.

[0157] Those skilled in the art will understand that the method steps of this application can be performed by a computer program instructing related hardware, such as electronic device 80 or a processor. The computer program can be stored in a non-transitory computer-readable storage medium, and its execution causes the steps of this application to be performed. Depending on the context, any reference herein to memory, storage, or other media may include non-volatile or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0158] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A clustering method for wafer images, characterized in that, include: Generate a binary image corresponding to the wafer image based on the wafer image containing defective grains; Based on the pixel information of defective grains in the binary image, the topological structure information of the defective grains is generated. Based on the topological information, a contour image corresponding to the binary image is generated; Based on the contour list corresponding to the contour curve in the contour image, the contour region in the contour image is filled with pixels to generate a filled image. The contour list includes pixel information of the defect grains that constitute the contour curve, representing the shape and position of the contour. Each contour region in the filled image is labeled as a clustering result; The topology information includes at least one topology, and generating a contour image corresponding to the binary image based on the topology information includes: Based on the topology information, the edge defect grains in each topology are determined; The edge defect grains in each of the topologies are connected to form the contour curve corresponding to each topology; The pixel values ​​of defect grains in the binary image, except for those on the contour curve, are set to zero to generate the contour image. Before generating a filled image by filling the contour regions in the contour image with pixels based on the contour list corresponding to the contour curves in the contour image, the method further includes: Based on the contour list corresponding to the contour curve in the contour image, isolated points in the contour image are filtered to obtain a filtered contour image. The step of filling the contour regions in the contour image with pixels based on the contour list corresponding to the contour curves in the contour image to generate a filled image includes: Sort all the contour lists in the filtered contour image according to the number of defective grains; In order from fewest to most, the contour regions corresponding to each of the contour lists are filled with pixels in sequence to generate a filled image; The step of sequentially filling the contour regions corresponding to each of the contour lists with pixels in ascending order to generate a filled image includes: In ascending order, the target defect grains in the contour region corresponding to the Nth contour list are determined, wherein the target defect grains are defect grains with a pixel value of zero. Obtain the filling characteristic value of each target defect grain; If the filling feature value is not selected, the pixel value of the target defect grain is updated to a preset pixel value to fill the corresponding contour area, and the filling feature value is updated to be selected. A filled image is generated based on the filled contour region corresponding to each of the contour lists.

2. The wafer image clustering method according to claim 1, characterized in that, After labeling each contour region in the filled image as a clustering result, the method further includes: Based on the number of defective grains in each clustering result, the cluster type of each clustering result is determined, wherein the cluster type includes non-noise clusters; The risk level of the neighboring normal grains is determined based on the distance between the neighboring normal grains and the non-noise clusters in the filled image.

3. The wafer image clustering method according to claim 1, characterized in that, The topology information includes a first topology and a second topology. The first topology refers to a topology that includes at least two contour curves, and the second topology includes a topology that includes one contour curve. Before generating a filled image by filling the contour regions in the contour image with pixels based on the contour list corresponding to the contour curves in the contour image, the method further includes: Extract an initial contour list for each contour curve in the contour image; Each initial contour list is deduplicated to obtain an intermediate contour list corresponding to each initial contour list; Merge at least two intermediate contour lists with the same defective grains corresponding to the first topology to obtain the contour list corresponding to the first topology. Use the list of intermediate contours corresponding to the second topology as the contour list of the second topology.

4. The wafer image clustering method according to claim 1, characterized in that, The step of generating a binary image corresponding to a wafer image containing defective grains includes: Obtain the first limit value of the grain in the horizontal direction and the second limit value in the vertical direction in the wafer image; Based on the first limit value and the second limit value, the size of the binary image corresponding to the wafer image is obtained; Based on the first limit value and the second limit value, determine the pixel information of each grain in the wafer image in the binary image; Based on the type of each grain in the wafer image, the pixel value of each grain in the binary image is determined.

5. A clustering device for wafer images, characterized in that, The apparatus for implementing the clustering method for wafer images as described in any one of claims 1-4, the apparatus comprising: The binary image generation module is used to generate a binary image corresponding to a wafer image based on a wafer image containing defective grains. The structural information generation module is used to generate the topological structure information of defective grains based on the pixel information of defective grains in the binary image. The contour image generation module is used to generate a contour image corresponding to the binary image based on the topological structure information. The filled image generation module is used to fill the contour regions in the contour image with pixels based on the contour list corresponding to the contour curves in the contour image, so as to generate a filled image. The labeling module is used to label each contour region in the filled image as a clustering result.

6. An electronic device, characterized in that, The electronic device includes a processor and a memory storing computer program instructions, wherein when the processor executes the computer program instructions, it implements the steps of the clustering method for wafer images as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a clustering program for wafer images, wherein when the clustering program for wafer images is executed by a processor, it implements the steps of the clustering method for wafer images as described in any one of claims 1 to 4.

Images