Wafer defect classification method and device, medium, equipment and program product
Classify wafer defect images through hierarchical clustering algorithm, solve the problem of difficult to identify defect type changes in semiconductor manufacturing, realize rapid identification and flexible response, reduce R&D and debugging time, and improve product yield and interpretability .
Patent Information
- Application Number
- CN202411006431.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-15
AI Technical Summary
In the semiconductor manufacturing process, it is difficult to classify wafer defects, especially during the R&D stage, new defect types frequently appear due to equipment parameter adjustment, which makes manual labeling time-consuming and model debugging time long, making it difficult to identify new defect types in time.
The hierarchical clustering algorithm is used to cluster the image data of wafer defects, determine the cluster information through the tree map, and update the classification results based on the distance between the new defect data and the cluster, reduce the calculation amount and debugging time, and respond flexibly Defect type changes.
It realizes the rapid identification of new defect types, reduces manual annotation dependence and model debugging time, improves R&D efficiency and product speed to market, and enhances interpretability and classification accuracy.
Smart Images

Figure CN118968144A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of semiconductor manufacturing technology, and in particular to a wafer defect classification method, device, medium, equipment and program product. Background Art
[0002] Wafers are the basic raw materials for manufacturing semiconductor products and play an extremely important role in the semiconductor manufacturing process. Defects on wafers can lead to the failure of finished chips. With the continuous advancement of semiconductor technology, the types and forms of wafer defects are becoming more and more diverse. Especially in the research and development process, new types of defects will continue to emerge. In the research and development process, how to classify wafer defects is an urgent problem to be solved. Summary of the invention
[0003] To overcome the problems existing in the related art, this specification provides a wafer defect classification method, device, medium, equipment and program product.
[0004] According to a first aspect of an embodiment of this specification, a wafer defect classification method is provided, the method comprising:
[0005] Acquire an original data set; the original data set includes a plurality of data for characterizing a wafer defect image;
[0006] The original data set is clustered by a hierarchical clustering algorithm to obtain a dendrogram representing the distance hierarchy between the original data, and cluster information of multiple clusters is determined from the dendrogram according to a preset distance threshold;
[0007] Acquire new defect data; the new defect data is a wafer defect image that is not in the original data set;
[0008] The cluster information is updated according to the distance between the new defect data and each cluster and the distance threshold, thereby determining the wafer defect category.
[0009] According to a second aspect of an embodiment of this specification, there is provided a wafer defect classification device, comprising:
[0010] An original data set acquisition module is used to acquire an original data set; the original data set includes a plurality of data for characterizing a wafer defect image;
[0011] A clustering processing module, used to perform clustering processing on the original data set through a hierarchical clustering algorithm to obtain a dendrogram representing the distance hierarchy between the original data, and determine cluster information of multiple clusters from the dendrogram according to a preset distance threshold;
[0012] A new defect data acquisition module is used to acquire new defect data; the new defect data is a wafer defect image that is not in the original data set;
[0013] The defect category determination module is used to update the cluster information according to the distance between the new defect data and each cluster and the distance threshold, so as to determine the wafer defect category.
[0014] According to a third aspect of the embodiments of this specification, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the wafer defect classification method as described in the first aspect of the embodiments of this specification is implemented.
[0015] According to a fourth aspect of an embodiment of this specification, a computer device is provided, the computer device comprising:
[0016] processor;
[0017] Memory for storing computer programs;
[0018] When the computer program is executed by the processor, the processor implements the wafer defect classification method as described in the first aspect of the embodiment of this specification.
[0019] According to a fifth aspect of the embodiments of this specification, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the wafer defect classification method as described in the first aspect of the embodiments of this specification is implemented.
[0020] In one or more embodiments of the present specification, multiple original data sets for characterizing wafer defect images are obtained, and the original data sets are clustered by hierarchical clustering to obtain a dendrogram, and at least one cluster is determined from the dendrogram according to a preset distance threshold. New defect data is then obtained, and the final wafer defect classification result is determined based on the distance between the new defect data and the multiple clusters. The above method utilizes a hierarchical clustering algorithm, and there is no need to specify the number of clusters in advance. It can flexibly respond to changes in defect types during the research and development stage, and reduces debugging and model iteration training time. The above method utilizes a modified incremental hierarchical clustering algorithm to reduce the amount of calculation when new defect data is added.
[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present specification. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the specification and, together with the description, serve to explain the principles of the specification.
[0023] Figure 1A This is a schematic diagram of a wafer defect shown in this specification.
[0024] Figure 1B It is a schematic diagram of another wafer defect shown in this specification.
[0025] Figure 1C It is a schematic diagram of another wafer defect shown in this specification.
[0026] Figure 2 The flowchart of a wafer defect classification method is shown in this specification according to an exemplary embodiment.
[0027] Figure 3 It is a schematic diagram of a hierarchical clustering dendrogram shown in this specification.
[0028] Figure 4 It is a block diagram of a wafer defect classification device according to an exemplary embodiment of the present specification.
[0029] Figure 5 It is a hardware structure diagram of a computer device shown in this specification according to an exemplary embodiment. DETAILED DESCRIPTION
[0030] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this specification. Instead, they are merely examples of devices and methods consistent with some aspects of this specification as detailed in the appended claims.
[0031] The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit this specification. The singular forms "a", "the" and "the" used in this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0032] It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this specification, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".
[0033] Semiconductor manufacturing is an extremely delicate and complex process, and any tiny defect may lead to the failure of the finished chip. Wafers are important raw materials used by the semiconductor industry to manufacture chips and other products. In order to ensure the yield of chips and other products, wafer defects need to be identified. At the same time, since different types of wafer defects often have different causes, wafer defects need to be classified to identify problems in the current production process and adaptively solve corresponding problems to improve the yield of wafers.
[0034] With the continuous advancement of semiconductor technology, the size of wafers is getting smaller and smaller, and the types and forms of defects are becoming more and more diverse, which makes the detection and classification of defects more difficult. Especially in the R&D stage, since parameters will be changed during the R&D stage, some new types of defects may appear, which makes defect classification more difficult. Among them, the R&D stage refers to the stage of studying the parameters of wafer production equipment. During the R&D stage, you can try a variety of different parameters of wafer production equipment, such as changing the exposure dose of the lithography machine, adjusting the gas flow of the etching equipment, etc., to find parameters with fewer wafer defects and achieve better manufacturing results.
[0035] Wafer defect diagram Figure 1A , Figure 1B and Figure 1C shown. Figure 1A The figure shows a pattern failure defect. Figure 1A In the figure, the trace of one dent is different from that of the other dents, and the dent is etched more than the shape of the other dents. Figure 1B FIG. 4 shows another pattern fail defect, in which there are some locations where long strip-shaped indentations are not etched enough. Figure 1C In the image, there are some circular pits on the wafer surface that should not exist.
[0036] In the related art, wafer defect classification is generally carried out through manual labeling and preset classification rules. Specifically, the defect type can be determined by manual analysis, and the defect type can be labeled for a training data set including multiple defect images. The labeled training data set is used to train an artificial neural network model that can be used to classify defects. In this way, the artificial neural network model can be used to classify new defects.
[0037] However, the above methods in the related art have the following problems:
[0038] First, manual labeling is time-consuming and labor-intensive, and it is difficult to cover all possible defect types.
[0039] Second, during the semiconductor R&D stage, new defect types may continue to appear due to adjustments to production equipment parameters or the application of new technologies and processes. In order to identify new defect types, classification rules need to be adjusted frequently and models need to be retrained, which not only takes a lot of time, but may also miss key defect detections due to untimely model adjustments, leading to increased R&D time.
[0040] In view of the above problems, this specification provides a wafer defect classification method, which obtains multiple original data sets for characterizing wafer defect images, clusters the original data sets through a hierarchical clustering method, obtains a dendrogram, and determines at least one cluster from the dendrogram according to a preset distance threshold. Then, new defect data is obtained, and the final wafer defect classification result is determined according to the distance between the new defect data and the multiple clusters.
[0041] First, the above method uses a hierarchical clustering algorithm to target the characteristics of the R&D stage, which can identify new defect types as quickly as possible, flexibly respond to changes in defect types, and shorten the R&D time. Specifically, the difference between the hierarchical clustering algorithm and other clustering algorithms is that the hierarchical clustering algorithm does not need to specify the categories and number of clusters in advance, which can better respond to the characteristics of new defect types that will appear in the R&D stage. Through the hierarchical clustering algorithm, even new defect types that have not been encountered during model training can be detected in time during the R&D stage, and the parameters of the production equipment can be adjusted in time, shortening the R&D time.
[0042] Compared with the methods of related technologies, the above method does not require re-data annotation and model debugging for new defect types, can automatically adapt to new defect types, reduces debugging and iteration time, and speeds up the deployment and update of defect detection systems.
[0043] Secondly, the above method can reduce the amount of calculation. Unlike the hierarchical clustering algorithm in the related art, the hierarchical clustering algorithm in the related art generally does not support incremental learning, that is, when new data points are added, since the new data points may change the distance or similarity between the original data points and affect the clustering results, it is usually necessary to re-run the hierarchical clustering algorithm, perform clustering processing on all data including the new data points, recalculate the distance matrix, and rebuild the clustering hierarchy according to the new distance matrix. The method provided in this specification improves the hierarchical clustering algorithm. For new data points, the original cluster structure is not changed, but the data point is determined to be added to a new cluster or as a new cluster based on the distance between the data point and other clusters, thereby reducing the amount of calculation and speeding up the wafer defect classification process.
[0044] Thirdly, the above method can reduce the reliance on manual labeling. The clustering algorithm is used to reduce the need for manual labeling, thus saving a lot of time and human resources. This is crucial for the rapidly developing semiconductor industry and can speed up the process of product development to market.
[0045] Finally, the above method can enhance interpretability. The clustering results of the hierarchical clustering algorithm can be represented by a tree diagram, which can help researchers and engineers analyze the differences and connections between various defect types, better understand the nature and causes of defects, and provide a basis for further quality improvement.
[0046] Next, the embodiments of this specification are described in detail.
[0047] like Figure 2 As shown, Figure 2 This is a flowchart of a wafer defect classification method according to an exemplary embodiment of the present specification, comprising the following steps:
[0048] Step 201, obtaining an original data set.
[0049] The original data set includes a plurality of data for characterizing wafer defect images.
[0050] First of all, it should be noted that the wafer defect classification method is proposed based on the characteristics of the research and development process and can be applied in the research and development process. In addition, the above method can also be applied to the actual production process of wafers, so as to timely adjust the problems in wafer production according to the identified defect types to reduce defective wafers.
[0051] The original data set may include multiple wafer defect images, and may also include multiple features extracted from the multiple wafer defect images.
[0052] The wafer defect image may be a wafer image obtained by a scanning electron microscope (SEM). The scanning electron microscope can take pictures and scan the wafer, but it can only identify the existence of defects and the location of the defects, and cannot specifically determine the type of defects. The wafer defect image may be a partial image of the wafer, such as an image that only includes the location where the wafer defects exist. Of course, the wafer defect image may also be obtained by taking pictures with other equipment that can identify wafer defects, and this specification does not limit this.
[0053] Wherein, in the case where the original data set includes multiple features extracted from multiple wafer defect images, step 201 may be, obtaining an original image set; the original image set includes multiple wafer defect images; extracting features of the images in the original image set through a feature extraction network to obtain an original data set. In this way, the image can be processed into a format that is easily recognized by the hierarchical clustering algorithm, thereby improving the classification accuracy of subsequent processes.
[0054] In an optional embodiment, the network used for feature extraction may be a convolutional neural network (CNN). The hierarchical structure of CNN can capture the hierarchical features of the image, which are crucial for distinguishing different types of defects.
[0055] Among them, the features that CNN can extract include low-level features, mid-level features, and high-level features.
[0056] In the primary layers of CNN, the model tends to extract some basic image features, which are low-level features such as edges, corners, colors, and textures. These features help distinguish the basic shapes and structures in the image.
[0057] As the network goes deeper, CNNs start to combine low-level features to form more complex patterns, resulting in mid-level features such as specific shapes and parts of objects. In the context of wafer defects, this might involve identifying specific defect shapes or patterns, such as scratches or stains.
[0058] At deeper levels of the CNN, the network is able to identify highly abstract concepts and derive high-level features, which are often related to the overall structure of the image and the specific type of defect. For example, the network may be able to distinguish between different kinds of defects, such as microcracks on a wafer and particle contamination.
[0059] The advantage of using these features for cluster analysis is that these features contain multiple levels of information from the most basic to the most abstract, which can fully describe the characteristics of wafer defects. These detailed feature descriptions enable cluster analysis to more accurately classify defects with similar characteristics into one category, even if these defects may appear to be very similar or completely different to the naked eye.
[0060] Of course, the feature extraction network mentioned above can also utilize other networks in the related technology that can extract features, and this specification does not limit this.
[0061] In addition, data preprocessing, such as denoising and standardization, can be performed before feature extraction to improve the accuracy of subsequent processing.
[0062] Step 203, clustering the original data set by a hierarchical clustering algorithm to obtain a dendrogram representing the distance hierarchy between the original data, and determining cluster information of multiple clusters from the dendrogram according to a preset distance threshold.
[0063] Through step 201 and step 203 , hierarchical clustering processing may be performed first according to the existing wafer defect images.
[0064] First, we introduce the hierarchical clustering algorithm, which is an unsupervised learning algorithm used to organize objects in a data set into a hierarchical structure, namely a dendrogram. This algorithm does not specify the number of clusters in advance, but builds a clustering hierarchy by gradually merging or splitting data points.
[0065] Hierarchical clustering algorithms mainly include the following two clustering algorithms:
[0066] First, Agglomerative Hierarchical Clustering. The specific processing process of Agglomerative Hierarchical Clustering includes: first treat each data point as a separate cluster, calculate the distance between all clusters, and merge each cluster with the cluster with the closest distance. After merging, recalculate the distance between all clusters (i.e., the distance matrix), and repeat the above merging steps until the stopping condition is reached (such as reaching the preset number of clusters or the distance between clusters exceeds a certain threshold), so that a tree diagram consisting of all data points can be obtained.
[0067] The tree diagram is as follows Figure 3 shown. Figure 3 The horizontal axis is the sample number, and the vertical axis represents the distance between clusters. Different distance thresholds can produce different classification results, such as Figure 3 The distance threshold at the dotted line position shown in the defect can obtain 7 classification results. From the left, the first category includes samples 1, 29 and 26, the second category includes samples 2, 22, 21, 3 and 4, and the other categories are similar and will not be repeated.
[0068] After obtaining the dendrogram, multiple clusters can be determined from the dendrogram according to the preset distance threshold. The distance threshold indicates the shortest distance between two clusters. If the distance between two clusters is greater than or equal to the distance threshold, it can be determined that the two clusters are two independent clusters; if the distance between two clusters is less than the distance threshold, the two clusters should be regarded as the same cluster.
[0069] Second, divisive hierarchical clustering. Agglomerative hierarchical clustering is a bottom-up clustering algorithm. In contrast to agglomerative hierarchical clustering, divisive hierarchical clustering is a top-down clustering algorithm. Specifically, all data points are first treated as a cluster, and then the cluster is gradually divided into smaller sub-clusters based on the splitting criteria (such as maximizing the intra-class variance and maximizing the inter-class distance) until a certain stopping condition is met.
[0070] After stopping, multiple clusters can be determined from the dendrogram according to a preset distance threshold. The method for determining multiple clusters is the same as above and will not be described again.
[0071] By using any of the above two hierarchical clustering algorithms, at least one cluster and a dendrogram that can represent the hierarchical clustering structure can be obtained.
[0072] After obtaining at least one cluster, in order to facilitate subsequent calculations, cluster information of each cluster may also be stored, such as the centroid or center point of the cluster, and the distance matrix or distance information between clusters, which may include the distance between every two clusters.
[0073] The distance between clusters can be the distance between the center points of two clusters. The distance can be represented by Euclidean distance, Manhattan distance, etc. The distance here can also be the similarity between two clusters. This specification does not limit the specific form of the distance between clusters.
[0074] Use a hierarchical clustering algorithm to perform cluster analysis on features. Hierarchical clustering allows you to observe the distribution of data at different levels, so as to better understand the similarities and differences between defects. The hierarchical clustering algorithm selected in the method provided in this manual has the following advantages:
[0075] First, the number of clusters does not need to be specified in advance. Specifically, hierarchical clustering does not require the number of clusters to be determined in advance, unlike the k-means clustering algorithm (K-means). This is particularly important for wafer defect detection, because the type and number of defects may vary greatly depending on the production process. Especially in the R&D process, it is impossible to predict in advance the type and number of defects caused by each production parameter.
[0076] Second, flexibility. Hierarchical clustering provides a tree-structured clustering result. Researchers can choose different levels of clustering granularity as needed. For example, they can determine the appropriate clustering granularity by setting a reasonable distance threshold, thereby finding the best balance between detailed classification and large category classification.
[0077] Third, adaptability to complex data structures. Hierarchical clustering is particularly effective in capturing natural hierarchies and structures in data, which is particularly important for data such as wafer defects that may have complex structures and similarity patterns.
[0078] Fourth, interpretability. The dendrogram (hierarchical clustering tree) generated by hierarchical clustering can intuitively show the similarity relationship between data points, which is of great value for understanding the connection and differences between different defect categories.
[0079] Step 205: Obtain new defect data.
[0080] The new defect data is data that is not included in the original data set and is used to characterize the wafer defect image.
[0081] Step 207 : updating the cluster information according to the distance between the new defect data and each cluster and the distance threshold, thereby determining the wafer defect category.
[0082] Step 201 and step 203 can be regarded as the initialization stage, and step 205 and step 207 are the incremental processing stage. Compared with the method of the related art, the method provided in this specification provides an incremental processing method of hierarchical clustering, which enables the method provided in this specification to add newly acquired defect data to the cluster in real time, identify new types of defects more quickly, and further accelerate the research and development process.
[0083] Regarding the execution timing of the above steps, the initialization phase consisting of step 201 and step 203 can be executed once, and the incremental processing phase consisting of step 205 and step 207 can be executed cyclically to continuously update the clustering results according to multiple new defect data. For example, step 205 and step 207 can be executed cyclically. When new defect data is newly generated, step 205 and step 207 are executed once to update the cluster information in real time, so as to facilitate the rapid detection of problems in production or research and development.
[0084] The new defect data may be a wafer defect image obtained in the real-time production process, or may be a feature obtained by extracting features from the wafer defect image obtained in the real-time production process through a feature extraction network. Before feature extraction, the wafer defect image may also be preprocessed as necessary. The specific feature extraction and preprocessing methods are described above and will not be repeated here.
[0085] After the new defect data is acquired in step 205, the cluster information can be updated in step 207 according to the distance between the new defect data and the original cluster, that is, it is determined whether the new defect data should be added to an original cluster or as a new cluster.
[0086] An important parameter in the hierarchical clustering algorithm is the preset distance threshold, which, as mentioned above, can be used to characterize the minimum distance between two clusters. Therefore, in step 207, the wafer defect classification result can be obtained according to the distance threshold, that is, the cluster information is updated according to the distance threshold, and the defect category corresponding to the cluster to which the new defect data belongs is used as its wafer defect category to complete the classification of wafer defects.
[0087] In addition to outputting the wafer defect category to which the new defect data belongs, step 207 can further output the category to which the remaining defect data belongs, as well as the number of defect types in the data set consisting of the original data set and the new defect data. In addition, in order to facilitate users to obtain information intuitively, after step 207, a dendrogram corresponding to the above data set can also be further output, that is, the overall defect type distribution corresponding to the above data set can be included.
[0088] Next, a specific implementation method of how to update the cluster information in step 207 will be described. It should be noted that the specific implementation method described below does not limit step 207.
[0089] For the scenario of adding new defect data to the original cluster, step 207 may include: determining the cluster closest to the new defect data based on the distance between the new defect data and each cluster; when the distance between the new defect data and the closest cluster is less than the distance threshold, adding the new defect data to the closest cluster, and determining the wafer defect category as the defect category corresponding to the closest cluster; updating the information of the closest cluster, and updating the tree diagram.
[0090] Specifically, the distance between the new defect data and each cluster in the original cluster is first determined, and the cluster with the smallest distance to the new defect data is determined. If the distance between the cluster and the new defect data is less than the preset distance threshold, the new defect data can be added to the cluster. The tree diagram and the cluster information are updated, and the defect category corresponding to the cluster is used as the wafer defect category of the new defect data.
[0091] In addition, if the distance between the cluster with the smallest distance and the new defect data is greater than a preset distance threshold, the new defect data may also be used as a new cluster.
[0092] For the scenario of taking the new defect data as a new cluster, step 207 includes: when the distance between the new defect data and any cluster is greater than the distance threshold, taking the new defect data as a new cluster and creating a new defect category for it.
[0093] Since the new defect data is far away from all clusters, it can be determined that the cluster belongs to a new cluster, so a new defect category is created for it. In this case, updating the cluster information means adding the information of the cluster, such as the center point of the cluster and the distance between it and other clusters.
[0094] Through the above method, clustering results can be obtained without re-running the clustering algorithm, which ensures the real-time processing and improves the processing efficiency.
[0095] After determining the wafer defect category, the basic information of the cluster can also be updated. For example, when new defect data is added to an existing cluster, the centroid or center point of the cluster is updated. For newly added clusters or clusters with changed centroids, the distances between these clusters and other clusters can be calculated and stored.
[0096] After the clustering results are determined, some other processing may be performed to make the clustering results more accurate. Some optional processing will be described below.
[0097] In an optional embodiment, after determining the wafer defect category, you can further: update the distance between each two clusters in the cluster information; when the distance between two clusters is less than the distance threshold, merge the two clusters whose distance is less than the preset distance threshold into one cluster, and update the tree diagram.
[0098] Specifically, if the new defect data is added to the existing clusters to obtain the clustering results, it may affect the distance between the cluster to which the new defect data is added and other clusters, which may in turn affect the clustering results. Therefore, if the distance between the cluster and another cluster is closer after the new defect data is added to the cluster, the two clusters can be merged. This can ensure that the final clustering results are more accurate.
[0099] As mentioned above, after updating the cluster information, the tree diagram can also be updated. Specifically, if according to the above processing, there is a cluster update, a new cluster, or a cluster merger, the updated cluster information in the above situation can be further used to update the tree diagram. For example, if the new defect data is added to the existing cluster, resulting in the merger of clusters, the merger result can be updated to the tree diagram accordingly. If the new defect data is used as a new cluster, the new cluster can be added to the tree diagram. In this way, the complete tree diagram can be used for subsequent analysis and utilization.
[0100] In an optional implementation, if the addition of multiple new defect data leads to the merging of multiple clusters, and the number of merged clusters exceeds a certain threshold, step 203 can be re-executed, that is, a complete hierarchical clustering process is performed on all existing data (including the original data set and the added multiple new defect data). Alternatively, in the above case, a split hierarchical clustering process can be used to split a cluster according to the splitting criterion to obtain multiple sub-clusters.
[0101] If multiple clusters are merged, it may result in a cluster covering a wider range. If the distance between the two farthest data points in a cluster is far, then the current cluster may inappropriately summarize multiple defect types. Therefore, for the above situation, the clusters can be reasonably split, or the hierarchical clustering process can be re-performed to make the final clustering result more accurate.
[0102] In an optional embodiment, it is also possible to: determine a new distance threshold that is different from the distance threshold; and adjust the clusters according to the new distance threshold to determine a new clustering result.
[0103] Specifically, in some cases, with the addition of new defect data, some new defect types may be generated, so it may be necessary to further refine the defect types and change the defect classification rules accordingly. Or it may be necessary to change the defect classification rules during the analysis process to classify from a more macro or detailed perspective. In this case, researchers may change the preset distance threshold.
[0104] After determining that the distance threshold has been changed, a new distance threshold can be obtained, and the clusters can be split or merged using the new distance threshold, thereby changing the number of classification categories to determine a new clustering result.
[0105] The specific adjustment of the clusters may be: if the new distance threshold is greater than the preset distance threshold, it may be possible to determine whether there are clusters that need to be merged (for example, the distance between two clusters is less than the new distance threshold) based on the stored distance information between the clusters, and merge the clusters.
[0106] If the new distance threshold is less than the preset distance threshold, a cluster can be split according to the splitting criterion to obtain multiple sub-clusters. The cluster splitting can also be completed according to the stored dendrogram.
[0107] In the context of a specific example, at the beginning, the classification rules only need to identify the defect types at the pattern fail granularity. As the development progresses, some new defects may appear. For example, there are many different types of failures in pattern fail, such as Figure 1AAs shown in the figure, one pattern has an extra irregular part etched, and some are as follows Figure 1B As shown in the figure, some patterns are not successfully etched. Although the two defects are similar in type, their causes may be different. Therefore, in the above case, the distance threshold can be further narrowed to distinguish the two types of defects, which is more convenient for research and development.
[0108] It can be seen that the use of hierarchical clustering algorithm can also adapt to changes in defect classification rules by adjusting the distance threshold of clusters without retraining the model, thereby improving the overall R&D efficiency and responding to changes in defect classification rules more flexibly.
[0109] In an optional implementation, the classification effect of the model can also be evaluated based on the clustering results. The performance of the model on unseen data can be evaluated using techniques such as cross-validation, and parameters can be adjusted to optimize performance, so as to obtain better wafer defect classification results.
[0110] Through the above method, the classification results of wafer defects can be obtained. In the R&D and production stages, since different types of defects are often caused by different reasons, by classifying wafer defects, it is possible to analyze and determine the problems in the current production process, and make adaptive adjustments to the problems to reduce wafer defects.
[0111] In an optional implementation, the clustering results can also be displayed through visualization tools, such as using a tree diagram to display the clustering results to researchers and engineers. In addition, in the process of displaying through visualization tools, the wafer defect images corresponding to each type of defect and the number of wafer defect images included in each type of defect can be added. Displaying the classification results through visualization tools can enable researchers and engineers to intuitively understand the defect classification and distribution during the research and development process, thereby accelerating the research and development process.
[0112] Corresponding to the embodiments of the aforementioned method, this specification also provides embodiments of a device and a terminal to which it is applied.
[0113] like Figure 4 As shown, Figure 4 1 is a block diagram of a wafer defect classification device according to an exemplary embodiment of the present specification, the device comprising:
[0114] The original data set acquisition module 410 is used to acquire an original data set; the original data set includes a plurality of data for characterizing a wafer defect image.
[0115] The clustering processing module 420 is used to perform clustering processing on the original data set through a hierarchical clustering algorithm to obtain a dendrogram representing the distance hierarchy between the original data, and determine cluster information of multiple clusters from the dendrogram according to a preset distance threshold.
[0116] The new defect data acquisition module 430 is used to acquire new defect data; the new defect data is a wafer defect image that is not in the original data set.
[0117] The defect category determination module 440 is used to update the cluster information according to the distance between the new defect data and each cluster and the distance threshold, so as to determine the wafer defect category.
[0118] In an optional embodiment, the defect category determination module 440 is specifically used to determine the cluster closest to the new defect data based on the distance between the new defect data and each cluster; when the distance between the new defect data and the closest cluster is less than the distance threshold, the new defect data is added to the closest cluster, and the wafer defect category is determined as the defect category corresponding to the closest cluster; the information of the closest cluster is updated, and the tree diagram is updated.
[0119] In an optional implementation manner, the defect category determination module 440 is specifically configured to, when the distance between the new defect data and any cluster is greater than the distance threshold, use the new defect data as a new cluster and create a new defect category for it.
[0120] In an optional implementation, the original data set acquisition module 410 is specifically used to acquire an original image set; the original image set includes multiple wafer defect images; and features of images in the original image set are extracted through a feature extraction network to obtain an original data set.
[0121] In an optional embodiment, the above-mentioned device also includes a cluster merging module (not shown in the figure), which is used to update the distance between each two clusters in the cluster information; when the distance between two clusters is less than the distance threshold, the two clusters whose distance is less than the preset distance threshold are merged into one cluster, and the tree diagram is updated.
[0122] In an optional embodiment, the above-mentioned device also includes a cluster adjustment module (not shown in the figure), which is used to determine a new distance threshold different from the distance threshold; through the new distance threshold, the clusters in the clustering result are adjusted to determine a new clustering result.
[0123] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.
[0124] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this specification. A person of ordinary skill in the art can understand and implement it without paying creative labor.
[0125] like Figure 5 As shown, Figure 5 A hardware structure diagram of a computer device in which a wafer defect classification device according to an embodiment is located is shown, and the device may include: a processor 2010, a memory 1020 for storing computer programs, an input / output interface 2030, a communication interface 1040, and a bus 2050. The processor 2010, the memory 1020, the input / output interface 2030, and the communication interface 1040 are connected to each other in communication within the device through the bus 2050.
[0126] The processor 2010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification. When the computer program is executed by the processor, the processor implements the above-mentioned wafer defect classification method.
[0127] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 2010.
[0128] The input / output interface 2030 is used to connect the input / output module to realize information input and output. The input / output module can be configured as a component in the device (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.
[0129] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).
[0130] The bus 2050 includes a path that transmits information between the various components of the device (eg, the processor 2010, the memory 1020, the input / output interface 2030, and the communication interface 1040).
[0131] It should be noted that, although the above device only shows the processor 2010, the memory 1020, the input / output interface 2030, the communication interface 1040 and the bus 2050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of this specification, and does not necessarily include all the components shown in the figure.
[0132] The embodiments of the present specification also provide a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned wafer defect classification method is implemented.
[0133] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0134] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0135] The present specification also provides a computer program product, including a computer program or instructions, which implement the above-mentioned wafer defect classification method when executed by a processor.
[0136] Those skilled in the art will readily appreciate other embodiments of the specification after considering the specification and practicing the invention claimed herein. The specification is intended to cover any variations, uses or adaptations of the specification that follow the general principles of the specification and include common knowledge or customary techniques in the art that are not claimed in the specification. The specification and examples are to be considered exemplary only, and the true scope and spirit of the specification are indicated by the following claims.
[0137] It should be understood that the present description is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present description is limited only by the appended claims.
[0138] The above description is only a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this specification should be included in the scope of protection of this specification.
Claims
1. A wafer defect classification method, characterized in that: The method comprises: Acquire an original data set; the original data set includes a plurality of data for characterizing a wafer defect image; The original data set is clustered by a hierarchical clustering algorithm to obtain a dendrogram representing the distance hierarchy between the original data, and cluster information of multiple clusters is determined from the dendrogram according to a preset distance threshold; Acquire new defect data; the new defect data is a wafer defect image that is not in the original data set; The cluster information is updated according to the distance between the new defect data and each cluster and the distance threshold, thereby determining the wafer defect category.
2. The method according to claim 1, characterized in that Determining a wafer defect category according to the distance between the new defect data and each cluster and the distance threshold includes: Determine the cluster closest to the new defect data according to the distance between the new defect data and each cluster; When the distance between the new defect data and the closest cluster is less than the distance threshold, the new defect data is added to the closest cluster, and the wafer defect category is determined as the defect category corresponding to the closest cluster; The information of the cluster with the closest distance is updated, and the dendrogram is updated.
3. The method according to claim 1, characterized in that Determining a wafer defect category according to the distance between the new defect data and each cluster and the distance threshold includes: When the distance between the new defect data and any cluster is greater than the distance threshold, the new defect data is used as a new cluster, and a new defect category is created for the new cluster.
4. The method according to claim 1, characterized in that The obtaining of the original data set comprises: Acquire an original image set; the original image set includes a plurality of wafer defect images; The features of the images in the original image set are extracted through a feature extraction network to obtain an original data set.
5. The method according to claim 1, characterized in that After determining the wafer defect category, the method further includes: The distance between every two clusters in the updated cluster information; When the distance between two clusters is less than the distance threshold, the two clusters whose distance is less than the preset distance threshold are merged into one cluster, and the dendrogram is updated.
6. The method according to claim 1, characterized in that The method further comprises: determining a new distance threshold different from the distance threshold; The clusters are adjusted through the new distance threshold to determine the new clustering results.
7. A wafer defect classification device, characterized in that: include: The original data set acquisition module is used to acquire the original data set; The original data set includes a plurality of data for characterizing a wafer defect image; A clustering processing module, used to perform clustering processing on the original data set through a hierarchical clustering algorithm to obtain a dendrogram representing the distance hierarchy between the original data, and determine cluster information of multiple clusters from the dendrogram according to a preset distance threshold; A new defect data acquisition module is used to acquire new defect data; the new defect data is a wafer defect image that is not in the original data set; The defect category determination module is used to update the cluster information according to the distance between the new defect data and each cluster and the distance threshold, so as to determine the wafer defect category.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the wafer defect classification method according to any one of claims 1 to 6 is implemented.
9. A computer device, characterized in that: The computer device comprises: processor; Memory for storing computer programs; When the computer program is executed by the processor, the processor implements the wafer defect classification method according to any one of claims 1 to 6.
10. A computer program product, characterized in that The method comprises a computer program or an instruction, which, when executed by a processor, implements the wafer defect classification method according to any one of claims 1 to 6.
