A wafer defect classification method, device, and computer equipment

Abstract

This specification provides a method, apparatus, and computer device for wafer defect classification. Among them, the wafer defect classification method includes: obtaining an image set composed of wafer defect images obtained during the wafer manufacturing process by the same wafer manufacturer. Inputting the image set into a warrant generation model to generate weights for controlling the degree of attention of the classification model to different positions of the images. Finally, applying the generated weights to the classification model and inputting the image set into the classification model to obtain the defect types of each image. And for each image subset corresponding to a defect type, clustering is performed, and the clustering result is used as the wafer defect classification result of the image. Adaptive weights are generated for the wafer defect images of different manufacturers, which enables the same classification model to be applied to different manufacturers. By using the method of clustering after classification, different fine-grained levels can be selected according to different manufacturers, so that the solution can be applied to different manufacturers and the classification accuracy can be improved.

Description

Technical Field

[0001] This specification relates to the field of image processing technologies, and particularly to a method, apparatus, and computer device for classifying wafer defects. Background Art

[0002] As one of the core driving forces of modern technology, the semiconductor industry has extremely high requirements for product quality due to the complexity and precision of its production process. A wafer is the basic raw material for manufacturing semiconductor products, and the quality of the wafer affects the performance and reliability of semiconductor products. During the wafer manufacturing process, due to the influence of various factors, such as material defects, process parameter fluctuations, equipment failures, etc., various defects often appear on the wafer surface. These defects will affect the yield rate of the wafer and also affect the quality of the semiconductor materials manufactured based on the wafer.

[0003] In the semiconductor manufacturing process, defect detection is a key link to ensure product quality. Different types of wafer defects are often caused by different reasons. Classifying wafer defects helps to quickly discover problems existing in the wafer defect manufacturing process, so as to take measures as soon as possible to reduce wafer defects and ensure the quality of the finished product. However, with the continuous progress of semiconductor manufacturing processes, the types and quantities of defects are also increasing continuously, which poses a huge challenge to defect detection. Traditional defect detection methods mainly rely on manual experience and simple image processing technologies. These methods often have difficulty ensuring the accuracy and efficiency of detection when facing large-scale and diverse defect data.

[0004] It can be seen that the current wafer defect classification has the problem of inaccuracy.

[0005] In recent years, with the rapid development of artificial intelligence technologies, especially the progress of deep learning and computer vision technologies, new solutions have been provided for semiconductor defect detection. However, there are still some problems in the existing defect detection methods in practical applications.

[0006] First, the classification and recognition process of defect data lacks a unified standard, resulting in difficulty in comparing and migrating detection results between different manufacturers. Second, due to different companies' requirements for the fine-grained classification of defects, the existing methods are difficult to flexibly adapt to this diverse demand. Summary of the Invention

[0007] To overcome the problems existing in the related technologies, this specification provides a method, apparatus, and computer device for classifying wafer defects.

[0008] According to the first aspect of the embodiments of this specification, a method for classifying wafer defects is provided, and the method includes:

[0009] Obtain an image set including multiple images; the multiple images in the image set are wafer defect images obtained by the same wafer manufacturer during the wafer manufacturing process within a preset time period;

[0010] Input the image set into a weight generation model to obtain weight data corresponding to the image set; the weight data indicates the degree of attention for controlling the classification model to different positions of the image;

[0011] Apply the weight to the classification model, and input the image set into the classification model configured with the weight data. The classification model performs weighted processing on the extracted image features based on the weight data, and performs classification of the first-level defect types based on the weighted image features to obtain the first-level defect types of each image in the image set;

[0012] For each first-level defect type, according to the number of target clusters, cluster the images in the image set corresponding to this first-level defect type, and use the clustering result as the second-level wafer defect classification result of the image set.

[0013] According to the second aspect of the embodiments of the present specification, there is provided a wafer defect classification device, and the device includes:

[0014] An image set acquisition module, configured to obtain an image set including multiple images; the multiple images in the image set are wafer defect images obtained by the same wafer manufacturer during the wafer manufacturing process within a preset time period;

[0015] A weight data acquisition module, configured to input the image set into a weight generation model to obtain weight data corresponding to the image set; the weight data indicates the degree of attention to different positions of the image;

[0016] A classification module, configured to input the image set into a classification model configured with the weight data. The classification model performs weighted processing on the extracted image features based on the weight data, and performs classification of the first-level defect types based on the weighted image features to obtain the first-level defect types of each image in the image set;

[0017] A clustering module, configured to, for each first-level defect type, according to the number of target clusters, cluster the images in the image set corresponding to this first-level defect type, and use the clustering result as the second-level defect classification result of the image set.

[0018] According to the third aspect of the embodiments of the present specification, there is provided a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the wafer defect classification method as described in the first aspect of the embodiments of the present specification.

[0019] According to a fourth aspect of the embodiments of this specification, a computer device is provided, and the computer device includes:

[0020] One or more processors;

[0021] A memory for storing one or more programs;

[0022] When the one or more programs are executed by the one or more processors, the one or more processors implement the wafer defect classification method described in the first aspect.

[0023] This specification provides a wafer defect classification method. An image set composed of wafer defect images obtained during the wafer manufacturing process by the same wafer manufacturer is acquired. The image set is input into a warrant generation model to generate weights for controlling the attention degree of the classification model to different positions of the images. Finally, the generated weights are applied to the classification model, and the image set is input into the classification model to obtain the defect types of each image. And for each image subset corresponding to a defect type, clustering is performed, and the clustering result is used as the wafer defect classification result of the image.

[0024] Since the defects existing in the wafers produced by the same wafer manufacturer within a period of time are generally caused by the same reasons, this makes the positions of the wafer defects of the same wafer manufacturer close in the images. And the defect positions in different wafer manufacturers may vary due to different defect reasons. In the above method, considering the different situations of the positions and manifestations of the defects in the wafers manufactured by different wafer manufacturers, adaptive weights are generated for the wafer defect images of different manufacturers, which enables the same classification model to be applied to different manufacturers. And using this weight data for classification can make the classification model pay more attention to the positions where the defects are located, and further make the large categories obtained by classification more accurate. Moreover, by the method of clustering after classification, different fine-grained levels can be selected according to different manufacturers, so that the solution can be applied to different manufacturers and the classification accuracy can be improved.

[0025] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing the embodiments consistent with this specification, and are used together with the specification to explain the principles of this specification.

[0027] Figure 1 is a flowchart of a wafer defect classification method shown in this specification according to an exemplary embodiment.

[0028] Figure 2 is a wafer defect image shown in this specification.

[0029] Figure 3 This is a block diagram of a wafer defect classification device shown in accordance with an exemplary embodiment of this specification.

[0030] Figure 4 This is a hardware structure diagram of a computer device shown in accordance with an exemplary embodiment of this specification. Detailed implementation manners

[0031] Here, the exemplary embodiments will be described in detail, and examples thereof are shown in the 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 implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with this specification. On the contrary, they are merely examples of devices and methods consistent with some aspects of this specification as detailed in the appended claims.

[0032] The terms used in this specification are for the sole purpose of describing specific embodiments and are not intended to limit this specification. The singular forms "a", "the", and "said" used in this specification and the appended claims are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0033] It should be understood that although the terms first, second, third, etc. may be used in this specification to describe various information, such 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 "when" or "while" or "in response to determining".

[0034] Based on the existing problems, this specification provides a wafer defect classification method. An image set composed of wafer defect images obtained by the same wafer manufacturer during the wafer manufacturing process is acquired. The image set is input into a warrant generation model to generate weights for controlling the degree of attention of the classification model to different positions of the images. Finally, the generated weights are applied to the classification model, and the image set is input into the classification model to obtain the defect types of each image. And for each image subset corresponding to a defect type, clustering is performed, and the clustering result is used as the wafer defect classification result of the image.

[0035] Since the defects in wafers produced by the same wafer manufacturer within a certain period of time are generally caused by the same reasons, the positions of the wafer defects of the same wafer manufacturer in the image are close. While the positions of the defects in different wafer manufacturers may vary due to different reasons for the defects. In the above method, considering the different positions and manifestations of the defects in the wafers manufactured by different wafer manufacturers in the image, adaptive weights are generated for the wafer defect images of different manufacturers, and these weights are used for classification, which can make the classification model pay more attention to the positions where the defects are located, and thus make the large categories obtained by classification more accurate. Moreover, by using the method of clustering after classification, the accuracy of the final classification result can be improved.

[0036] Next, the embodiments of this specification will be described in detail.

[0037] As Figure 1 shown, Figure 1 is a flowchart of a method shown in this specification according to an exemplary embodiment, including the following steps:

[0038] Step 101, obtain an image set including multiple images.

[0039] The multiple images in the image set are wafer defect images obtained by the same wafer manufacturer during the wafer manufacturing process within a preset time period.

[0040] Specifically, since there are differences in the defect positions in the images of different manufacturers, weights need to be generated for the images of the same wafer manufacturer in this specification. Therefore, first, the defect images of the wafers produced by the same wafer manufacturer are obtained, that is, the wafer defect images.

[0041] Among them, the wafer manufacturer is the manufacturer that produces and manufactures wafers. The wafer defect image can be: for the wafers manufactured by this wafer manufacturer, the wafer image is obtained through a scanning electron microscope (SEM), and at the same time, the wafer defect images with defects in the wafers are screened out. The scanning electron microscope can take pictures and scan the wafers and identify the existing defects. However, it can only identify the existence of defects and the positions of the defects, and cannot specifically determine the types of the defects.

[0042] The images of the wafer defects can be referred to Figure 2 . Figure 2 It can be seen in that there are defects in the form of partial under-etching of some long strip shapes in the image, and the coarse classification defect category corresponding to this defect can be called pattern fail.

[0043] For different wafer manufacturers, there are differences in the manufacturing equipment, SEM, production processes they use, and the problems existing in the production processes, etc., which may lead to different positions of defects in the wafer defect images. For example, due to problems in the settings of the wafer manufacturing equipment of a certain manufacturer, defects generally exist in the central position of the wafer; for another example, the centers of the SEMs of different manufacturers are different, resulting in offsets in different directions in the captured images. For example, the center of some manufacturers is offset to the lower right corner, and the center of some manufacturers is offset to the lower left corner. All these make the positions of defects in the wafer defect images of different manufacturers different, while the positions of defects in the wafer defect images of the same manufacturer are approximate.

[0044] In addition, for different wafer manufacturers, their production requirements are different, resulting in different patterns on the produced wafers. This makes the backgrounds of the wafer defect images obtained by different wafer manufacturers different, some with complex backgrounds and some with simple backgrounds. This may lead to the situation that in the process of wafer defect classification, the wafer defect images of some manufacturers have complex backgrounds, and the complex backgrounds affect the classification results of wafer defects. Therefore, through this method, the classification model can pay more attention to the positions where the wafer defects are located rather than the backgrounds, thereby avoiding the influence of the backgrounds on the classification results of wafer defects.

[0045] After obtaining the image set in step 101, the images can also be preprocessed. In an optional implementation manner, there can be preset image preprocessing methods corresponding to each wafer manufacturer; the method further includes: determining the preprocessing method corresponding to the image set according to the preset image preprocessing methods corresponding to each wafer manufacturer, and preprocessing the images in the image set through the determined preprocessing method.

[0046] Since the images corresponding to different wafer manufacturers are different, generally different preprocessing methods are required for processing. For example, if the images of a certain manufacturer have high noise, filtering processing can be performed; if the images of a certain manufacturer have low clarity, data enhancement methods can be used for enhancement processing, etc. Therefore, by presetting the preprocessing methods corresponding to different manufacturers, preprocessed images that better meet the requirements can be obtained.

[0047] In step 103, input the image set into the weight generation model to obtain the weight data corresponding to the image set.

[0048] Among them, the weight data indicates the degree of attention to different positions of the image.

[0049] Specifically, in order to adapt to the different positions of wafer defects in the wafer defect images of different manufacturers, it is necessary to input the image set into the weight generation model. The weight generation model generates adaptive weights applied to the classification model according to the feature distribution characteristics of the images in the image set. This can enable the classification model to focus more on the wafer defects themselves in the images rather than other positions, avoiding the influence of other positions on the classification results of wafer defects.

[0050] For the specific form of the weight generation model, in an alternative embodiment, the weight generation model includes a first module, a second module, and a third module; step 103 specifically includes: the first module extracts the features of each image in the image set; the second module calculates the mean and variance of the features of each image according to the features of each image; the third module generates the weight data corresponding to the image set according to the mean and variance.

[0051] Among them, the extracted features can be in the form of feature vectors.

[0052] Specifically, first, the first module is an encoder for encoding the wafer defect image into the form of a feature vector. In an alternative embodiment, the first module can reuse the feature extraction module of the classification model. The second module is used to calculate the mean and variance of the feature vectors of each image. Through the mean of the feature vectors, the average representation of multiple images can be obtained, and the typical representation of the object corresponding to a group of images can also be obtained. Through the variance of the feature vectors, the degree of change of each feature in the image can be obtained. Through the mean and variance of a group of image feature vectors, the characteristics of a group of images can be described.

[0053] Furthermore, the third module can generate the weights of the image set corresponding to this group of images according to the mean and variance. The third module can be expressed as a function of the mean and variance, W = f(μ,σ), where W represents the weights of the image set, and μ and σ represent the mean and variance of the feature vectors of the image set, respectively. In one embodiment, the third module can implement the above function f through a neural network.

[0054] The generated weight matrix can be applied to the classification model, so that the classification model can focus more on the defective part in the image, making the classification results of the classification model more accurate.

[0055] Step 105, input the image set into the classification model configured with the weight data. The classification model performs weighted processing on the extracted image features based on the weight data, and performs classification of the first-level defect types based on the weighted image features to obtain the first-level defect types of each image in the image set.

[0056] Specifically, the weights obtained in step 103 can be applied to the classification model. Thus, the images in the above image set can be more accurately classified for wafer defects through the classification model. By introducing an adaptive weight mechanism, the classification model can automatically adjust the weights according to the data characteristics of different manufacturers, thereby improving the transferability and adaptability of the classification model.

[0057] The weights are equivalent to applying a mask to the image. For the mask, it can take a value of 1 at positions where some defects are more likely to exist, and 0 or a decimal between 0 and 1 at other positions. In other words, the image features can be multiplied by the weights to obtain the image data after masking processing. This can make the model pay different attention to different positions of different images and pay more attention to the positions where the defects are located.

[0058] Among them, the above classification model can be a model trained based on a training data set for classifying wafer defects. The training data set includes multiple wafer defect images, and each wafer defect image corresponds to a label.

[0059] In an optional embodiment, the classification model can be a Vision Transformer (ViT). ViT has good performance in classifying large amounts of data, and in the scenario applied in this specification, the amount of data of wafer defect images is large, and good results can be obtained through ViT. Thus, the problems of processing efficiency and accuracy in the related art for large-scale defect data are solved.

[0060] For the application method of the weights, the weights are applied to each layer of the classification model, and the weights of different positions of the image when input into each layer can be controlled in a weighted manner.

[0061] When the classification model is ViT, the generated adaptive weights can be applied to each Transformer block of the ViT model. Specifically, in each Transformer block of the ViT model, adaptive weights are introduced, and the input image features are adjusted in a weighted manner, and then the output of each Transformer block is adjusted. In this way, the model can automatically adjust the weights according to the data characteristics of different manufacturers, thereby improving the adaptability and transferability of the model.

[0062] In other words, the classification model includes multiple Transformer blocks for self-attention processing, and a weight matrix corresponding to the weight data is configured in each Transformer block. Step 103 can include: each Transformer block of the classification model performs self-attention processing on the input image features to obtain an attention matrix or an image feature matrix as an intermediate matrix, and multiplies the intermediate matrix bit by bit with the weight matrix to obtain the image features output by the Transformer block.

[0063] Among them, the first-level defect type in step 105 refers to the rough classification. After the rough classification is completed, in step 107, for each separated defect type, a fine classification within this defect type can be performed to obtain the second-level defect type, so that the classification result can be more accurate.

[0064] In the related art, if a classification model of machine learning is directly used for fine classification, due to different classification criteria of different manufacturers, it may be impossible to use one model to classify the wafer defect images of all manufacturers. Moreover, if a new wafer defect type appears, the classification model needs to be adjusted. These may lead to inaccurate classification results. By the method of this specification, first, rough classification is performed through a classification model. For rough classification, the types of rough classification of different manufacturers are generally the same, but there are differences in fine classification. For example, there will be a defect category of pattern fail, but under this type, there are different fine classification categories for different manufacturers. The rough classification is generally determined based on the fixed standards of the semiconductor industry, and generally no new categories will be added. By this method, the classification accuracy can be improved and it can be applied to different manufacturers.

[0065] In addition, in an optional implementation manner, the weight generation model and the classification model are jointly trained according to a training data set. The training data set includes wafer defect images of multiple different wafer manufacturers, and classification labels of each wafer defect image. That is, the training process is similar to the above application process. First, the image set is input into the weight generation model to obtain weight data, and then the weight data is applied to the classification model, and the classification model is used to classify the image set. After classification, according to the labels and classification results, the parameters of the weight generation model and the classification model are adjusted by using a preset loss function.

[0066] In other words, the weight generation model and the classification model are trained in the following manner: Input the wafer defect image of any wafer manufacturer in the training data set into the weight generation model to obtain the first weight data generated by the weight generation model; the training data set includes wafer defect images of multiple different wafer manufacturers, and the first-level classification labels of each wafer defect image; input the wafer defect image of this wafer manufacturer into the classification model configured with the first weight data, and the classification model performs weighted processing on the extracted image features based on the first weight data, and performs classification of the first-level defect type based on the weighted image features; according to the first-level classification labels of each wafer image and the first-level defect types of each image output by the classification model, adjust the parameters of the weight generation model and the classification model.

[0067] Among them, during the joint training process, the parameters of the model can be optimized by minimizing the classification error and the weight generation error. During the training process, the cross-validation method is used to ensure the generalization ability of the model on data from different manufacturers.

[0068] This can make the generated weights better applied to the classification model, making the classification results more accurate.

[0069] In addition, the classification model can be evaluated after training is completed. Specifically, by performing classification tasks on data from multiple manufacturers, the classification accuracy and transferability of the model are evaluated. The evaluation results show that the improved ViT algorithm performs well on data from different manufacturers and has high classification accuracy and transferability.

[0070] Step 107, for each first-level defect type, according to the number of target clusters, cluster the images in the image set corresponding to this first-level defect type, and use the clustering result as the second-level wafer defect classification result of the image set.

[0071] As described above, coarse classification can be performed first, and then for each defect type obtained by coarse classification, clustering is performed within the images corresponding to this defect type, so as to obtain the wafer defect classification result of each image.

[0072] For the specific clustering algorithm, it can be the K-Means clustering algorithm. Of course, this example does not represent a limitation to this specification.

[0073] For the number of target clusters of each defect type, in an alternative embodiment, the number of target clusters is preset for the wafer manufacturing manufacturer corresponding to the image set and each first-level defect type.

[0074] Since different manufacturers have different requirements for the fine-grained classification of defects and different criteria for the fine classification of defects, it is difficult to transfer the detection results of different manufacturers. In the above method, the number of clustering clusters is preset according to the needs of each manufacturer for clustering, so that this classification method can be applied to the data of each manufacturer.

[0075] In another alternative embodiment, some manufacturers do not have a perfect classification standard for the time being. Specifically, the classification standard is generally determined according to past production experience. According to the defect types existing in the past production process, specific fine classifications are determined. For some new wafer manufacturers, they do not have the experience of wafer classification and may not be sure how many classifications each defect type needs to be specifically divided into.

[0076] In the above case, the number of clustering clusters corresponding to each defect type can be determined according to the characteristics of the image set. Specifically, for each defect type, and for each first-level defect type, the images in the image set corresponding to the first-level defect type are input into a clustering number generation model to obtain the target number of clustering clusters; the target number of clustering clusters is applied to the clustering process of the images corresponding to the first-level defect type.

[0077] In an alternative embodiment, the clustering number generation model can be similar to the weight generation model and also includes three modules. For the images corresponding to a certain defect type, first, the first module is used to obtain the feature vectors of the images corresponding to the defect type. Specifically, the feature vectors extracted in the previous steps are reused here. The second module is used to calculate the mean and variance between the obtained feature vectors for the feature vectors. The third module is used to generate the clustering number according to the mean and variance. The third module can be specifically expressed as K = h(μ,σ), where μ and σ respectively represent the mean and variance of the feature distribution, and K represents the generated adaptive clustering number.

[0078] This can help new manufacturers who don't know how to classify specifically to classify, and through the clustering number generation model, this clustering method can be applied to various manufacturers, improving flexibility.

[0079] This specification proposes a semiconductor defect classification method that combines an improved ViT algorithm and an optimized clustering algorithm. Through a two-step process of classification first and then clustering, this method can not only improve the accuracy and efficiency of defect detection, but also flexibly adapt to the requirements of different manufacturers for the defect subdivision granularity. By introducing the improved ViT for preliminary classification and utilizing its advantages in large-scale data processing to ensure the accuracy of the classification results and being able to migrate well between different manufacturers; then using the clustering algorithm within each large category to adapt to the different subdivision granularities of different companies. This method not only improves the generality and flexibility of defect detection, but also provides an efficient and migratable solution for defect classification in the semiconductor industry.

[0080] Corresponding to the embodiments of the foregoing method, this specification also provides embodiments of a device and a computer device to which the device is applied.

[0081] As Figure 3 shown, Figure 3 is a block diagram of a wafer defect classification device shown in this specification according to an exemplary embodiment. The device includes:

[0082] An image set acquisition module 310, configured to acquire an image set including multiple images; the multiple images in the image set are wafer defect images obtained by the same wafer manufacturer during the wafer manufacturing process within a preset time period;

[0083] A weight data acquisition module 320, configured to input the image set into a weight generation model to obtain weight data corresponding to the image set; the weight data indicates the degree of attention to different positions of an image.

[0084] A classification module 330, configured to input the image set into a classification model configured with the weight data, where the classification model performs weighted processing on the extracted image features based on the weight data, and performs classification of the first-level defect types based on the weighted image features to obtain the first-level defect types of each image in the image set.

[0085] A clustering module 340, configured to, for each first-level defect type, perform clustering processing on the images in the image set corresponding to the first-level defect type according to the target number of clusters, and use the clustering result as the second-level defect classification result of the image set.

[0086] In an optional implementation manner, the apparatus further includes a cluster number generation module 350 (not shown in the figure), configured to, for each first-level defect type, input the images in the image set corresponding to the first-level defect type into a clustering number generation model to obtain the target number of clusters.

[0087] In an optional implementation manner, the target number of clusters is preset for the wafer manufacturer corresponding to the image set and each first-level defect type.

[0088] In an optional implementation manner, the classification model is a Vision Transformer (ViT).

[0089] In an optional implementation manner, the classification model includes a plurality of transformer blocks for performing self-attention processing, and a weight matrix corresponding to the weight data is configured in each transformer block; the classification module 330 is specifically configured to perform self-attention processing on the image features input into each transformer block of the classification model to obtain an attention matrix or an image feature matrix as an intermediate matrix, and perform element-wise multiplication of the intermediate matrix and the weight matrix to obtain the image features output by the transformer block.

[0090] In an optional implementation manner, the weight data acquisition module 320 includes a first module, a second module, and a third module; the first module extracts the features of each image in the image set; the second module calculates the mean and variance of the features of each image according to the features of each image; the third module generates the weight data corresponding to the image set according to the mean and variance.

[0091] In an alternative embodiment, the device further includes a preprocessing module 300 (not shown in the figure), which is configured to determine a preprocessing method corresponding to the image set according to a preset image preprocessing method corresponding to each wafer manufacturer, and preprocess the images in the image set by the determined preprocessing method.

[0092] In an alternative embodiment, the device includes a training module 360 (not shown in the figure), which is configured to input a wafer defect image of any wafer manufacturer in a training data set into a weight generation model to obtain first weight data generated by the weight generation model; the training data set includes wafer defect images of multiple different wafer manufacturers and first-level classification labels of each wafer defect image; input the wafer defect image of this wafer manufacturer into a classification model configured with the first weight data, and the classification model performs weighted processing on the extracted image features based on the first weight data and performs classification of the defect types at the first level based on the weighted image features; adjust the parameters of the weight generation model and the classification model according to the first-level classification labels of each wafer image and the first-level defect types of each image output by the classification model.

[0093] The implementation processes of the functions and roles of each module in the above device are specifically described in detail in the implementation processes of the corresponding steps in the above method, and will not be elaborated here.

[0094] 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 embodiments described above are only illustrative. The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution in this specification. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0095] As Figure 4 shown, Figure 4 FIG. shows a hardware structure diagram of a computer device where the wafer defect classification device in the embodiment is located. The device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. Among them, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are communicatively connected to each other inside the device through the bus 1050.

[0096] The processor 1010 can be implemented in the form of a general - purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0097] The memory 1020 can be implemented in forms such as ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 1020 can store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.

[0098] The input / output interface 1030 is used to connect to the input / output module to achieve information input and output. The input / output module can be configured as a component in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Among them, the input devices can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices can include a display, a speaker, a vibrator, an indicator light, etc.

[0099] The communication interface 1040 is used to connect to a communication module (not shown in the figure) to achieve communication and interaction between this device and other devices. Among them, the communication module can achieve communication through a wired method (such as USB, network cable, etc.) or through a wireless method (such as a mobile network, WIFI, Bluetooth, etc.).

[0100] The bus 1050 includes a path for transmitting information between various components of the device (such as the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0101] It should be noted that although the above - mentioned device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in the specific implementation process, this device may also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above - mentioned device may also only include the components necessary to implement the solution of the embodiments of this specification, and does not necessarily include all the components shown in the figure.

[0102] An embodiment of this specification also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the wafer defect classification method as described in the first aspect of the embodiments of this specification.

[0103] Computer-readable media include both permanent and non-permanent, removable and non-removable media and can implement the storage of information by any method or technology. The 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 technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media do not include transitory media such as modulated data signals and carrier waves.

[0104] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the acts or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0105] Those skilled in the art will readily conceive of other embodiments of this specification after considering the specification and practicing the invention herein. This specification is intended to cover any variations, uses, or adaptations of this specification, which follow the general principles of this specification and include common general knowledge or conventional technical means in the technical field not claimed in this specification. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this specification are pointed out by the following claims.

[0106] It should be understood that this specification is not limited to the exact structures described above and shown in the figures, and various modifications and changes can be made without departing from its scope. The scope of this specification is limited only by the appended claims.

[0107] The above are only the preferred embodiments of this specification and are not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this specification shall be included within the scope of protection of this specification.

Claims

1. A wafer defect classification method, characterized in that, The method includes: Obtaining an image set including multiple images; the multiple images in the image set are wafer defect images obtained by the same wafer manufacturer during the wafer manufacturing process within a preset time period; Inputting the image set into a weight generation model to obtain weight data corresponding to the image set; the weight data indicates the degree of attention of the classification model to different positions of the image; Inputting the image set into a classification model configured with the weight data, and the classification model performs weighted processing on the extracted image features based on the weight data, and performs classification of the first-level defect types based on the weighted image features to obtain the first-level defect types of each image in the image set; For each first-level defect type, according to the target number of clusters, clustering the images in the image set corresponding to the first-level defect type, and using the clustering result as the second-level defect classification result of the image set; The target number of clusters is preset for the wafer manufacturer corresponding to the image set and each first-level defect type; or the target number of clusters is obtained by inputting the images in the image set corresponding to each first-level defect type into a clustering number generation model for each first-level defect type.

2. The method according to claim 1, wherein The classification model is a Vision Transformer (ViT).

3. The method according to claim 1, characterized in that The classification model includes multiple Transformer blocks for self-attention processing, and each Transformer block is configured with a weight matrix corresponding to the weight data; The classification model performs weighted processing on the extracted image features based on the weight data, including: Each Transformer block of the classification model performs self-attention processing on the input image features to obtain an attention matrix or an image feature matrix as an intermediate matrix, and performs element-wise multiplication of the intermediate matrix and the weight matrix to obtain the image features output by the Transformer block.

4. The method according to claim 1, characterized in that The weight generation model includes a first module, a second module, and a third module; The step of inputting the image set into the weight generation model to obtain the weight data corresponding to the image set includes: The first module extracts the features of each image in the image set; The second module calculates the mean and variance of the features of each image according to the features of each image; The third module generates the weight data corresponding to the image set according to the mean and variance.

5. The method according to claim 1, wherein It further includes: Determining the preprocessing method corresponding to the image set according to the pre-set image preprocessing methods corresponding to each wafer manufacturer, and preprocessing the images in the image set by the determined preprocessing method.

6. The method according to claim 1, wherein The weight generation model and the classification model are trained in the following manner: Inputting the wafer defect images of any wafer manufacturer in the training data set into the weight generation model to obtain the first weight data generated by the weight generation model; the training data set includes wafer defect images of multiple different wafer manufacturers and the first-level classification labels of each wafer defect image. Input the wafer defect images of the wafer manufacturer into a classification model configured with first weight data. The classification model performs weighted processing on the extracted image features based on the first weight data, and classifies the defect types at the first level based on the weighted image features. Adjust the parameters of the weight generation model and the classification model according to the classification labels at the first level of each wafer image and the defect types at the first level of each image output by the classification model.

7. A wafer defect classification device, characterized in that, The device includes: An image set acquisition module, configured to acquire an image set including multiple images; the multiple images in the image set are wafer defect images obtained by the same wafer manufacturer during wafer manufacturing within a preset time period. A weight data acquisition module, configured to input the image set into a weight generation model to obtain the weight data corresponding to the image set; the weight data indicates the degree of attention of the classification model to different positions of the image. A classification module, configured to input the image set into a classification model configured with the weight data. The classification model performs weighted processing on the extracted image features based on the weight data, and classifies the defect types at the first level based on the weighted image features to obtain the defect types at the first level of each image in the image set. A clustering module, configured to, for each defect type at the first level, perform clustering processing on the images in the image set corresponding to the defect type at the first level according to the target number of clusters, and use the clustering result as the defect classification result at the second level of the image set. The target number of clusters is preset for the wafer manufacturer corresponding to the image set and each defect type at the first level; or the target number of clusters is obtained by inputting the images in the image set corresponding to each defect type at the first level into a clustering number generation model.

8. A computer device, characterized in that, The computer device includes: One or more processors; A memory, configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

Images