Wafer defect classification method and device, storage medium and program product
Through the two-layer nested classification algorithm, the combination of coarse-grained and fine-grained classifiers is used to solve the problem of inaccurate classification of wafer defects in the prior art, and achieve higher classification accuracy and production efficiency.
Patent Information
- Application Number
- CN202411006429.1
- 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 prior art, wafer defect classification methods are difficult to achieve accurate classification, resulting in inaccurate classification.
Using a two-layer nested classification algorithm, firstly classify the wafer defect data into large categories through a coarse-grained classifier, and then subdivides it through a fine-grained classifier, and classifies it using the common characteristics of multiple sub-defect types to avoid the classification results in There is a sway between multiple defect types.
It improves the accuracy of wafer defect classification and improves semiconductor production efficiency and product quality.
Smart Images

Figure CN118968143A_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, storage medium and program product. Background Art
[0002] Wafers are the basic raw materials of the semiconductor industry and play an extremely important role in the semiconductor manufacturing process. Wafer defect classification is also one of the key links in the semiconductor manufacturing process. Accurate classification of wafer defects helps improve the quality of semiconductor production.
[0003] With the continuous advancement of semiconductor technology, the types and forms of wafer defects are becoming more and more diverse. In view of the increasingly diverse and complex types of wafer defects, the wafer defect classification methods in the existing technology are difficult to meet the needs of accurate defect classification and have the problem of inaccurate classification. Summary of the invention
[0004] To overcome the problems existing in the related art, this specification provides a wafer defect classification method, device, storage medium and program product.
[0005] According to a first aspect of an embodiment of this specification, a wafer defect classification method is provided, the method comprising:
[0006] Acquire defect data to be processed; the defect data to be processed is data used to characterize a wafer defect image;
[0007] Determining a first defect type to which the defect data to be processed belongs from a first defect type set by a first classifier; each defect type in the first defect type set includes a plurality of sub-defect types having common characteristics;
[0008] The second defect type to which the defect data to be processed belongs is determined from a plurality of sub-defect types included in the first defect type by using a second classifier corresponding to the first defect type.
[0009] According to a second aspect of an embodiment of this specification, a wafer defect classification device is provided, including:
[0010] A data acquisition module, used to acquire defect data to be processed; the defect data to be processed is data used to characterize a wafer defect image;
[0011] A first classification module, configured to determine, from a first defect type set, a first defect type to which the defect data to be processed belongs, by a first classifier; each defect type in the first defect type set includes a plurality of sub-defect types having common characteristics;
[0012] The second classification module is used to determine the second defect type to which the defect data to be processed belongs from a plurality of sub-defect types included in the first defect type by using a second classifier corresponding to the first defect type.
[0013] 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. 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.
[0014] According to a fourth aspect of an embodiment of this specification, a computer device is provided, the computer device comprising:
[0015] processor;
[0016] Memory for storing computer programs;
[0017] 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.
[0018] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, including a computer program or instructions, which, when executed by a processor, implements the wafer defect classification method as described in the first aspect of the embodiments of this specification.
[0019] In one or more embodiments of the present specification, after obtaining the defect data to be processed, the defect data to be processed is classified by a coarse-grained first classifier, and the first defect type to which the defect data to be processed belongs is determined from the first defect type set, and the first defect type includes multiple sub-defect types with common characteristics, so that when the first classifier is used for classification, more attention is paid to the common characteristics of multiple sub-defect types, and the major category to which the defect data belongs is determined by the common characteristics. The classification is performed by a fine-grained second classifier, and the classification process is to determine the second defect type from the multiple sub-defect types corresponding to the first defect type, so that the classification process is classified by the detailed differences between multiple sub-defect types with common characteristics, rather than by the common characteristics of multiple sub-defect types, avoiding the classification result from fluctuating between multiple defect types, effectively improving the accuracy of wafer defect classification, and further improving semiconductor production efficiency and product quality.
[0020] 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
[0021] 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.
[0022] Figure 1A This is a schematic diagram of a wafer defect shown in this specification.
[0023] Figure 1B It is a schematic diagram of another wafer defect shown in this specification.
[0024] Figure 2 The flowchart of a wafer defect classification method is shown in this specification according to an exemplary embodiment.
[0025] Figure 3 It is a schematic diagram of defect types and sub-defect types included in a first defect type set shown in this specification.
[0026] Figure 4 It is a block diagram of a wafer defect classification device according to an exemplary embodiment of the present specification.
[0027] Figure 5 It is a hardware structure diagram of a computer device shown in this specification according to an exemplary embodiment. DETAILED DESCRIPTION
[0028] 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.
[0029] 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.
[0030] 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".
[0031] Wafers are important raw materials for manufacturing semiconductor devices. Defects on wafers can cause chips and other products manufactured using wafers to fail to function properly. In order to improve production efficiency and quality, the probability of wafer defects must be reduced as much as possible.
[0032] The same type of wafer defects are generally caused by the same reasons. Classifying wafer defects helps to analyze and discover problems in the current production process, such as inappropriate parameter settings of production equipment. After the problem is found, it can be solved to improve semiconductor production efficiency and product quality. It can be seen that the classification of wafer defects is crucial for semiconductor production.
[0033] In the related art, wafer defect classification is generally performed based on manual labeling and preset rules, that is, multiple defect types are preset, and training data is labeled to train an artificial neural network for identifying the types of wafer defects.
[0034] However, the wafer defect classification method in the related art has the problem of low classification accuracy.
[0035] In order to solve the above problems, this specification provides a wafer defect classification method. After obtaining the wafer defect data, coarse-grained classification is first performed through a first classifier to obtain a first defect type. The first defect type includes multiple sub-defect types with common characteristics. Then, fine-grained classification is performed through a second classifier corresponding to the first defect type to determine which sub-defect type of the first defect type the defect belongs to.
[0036] In the related art, an artificial neural network model is used to identify all defect types. During the training process of the artificial neural network model, all defects are considered to be mutually exclusive. However, in some cases, some defects have some common characteristics, such as Figure 1A The wafer defect shown is that part of the shape is not etched enough. Figure 1B The wafer defect shown is that some irregular shapes are over-etched. The causes of the two wafer defects and the specific forms of the defects are different, but they have a common feature, that is, both belong to the situation of pattern failure.
[0037] However, in the training process of the artificial neural network model in the related technology, all defects are considered mutually exclusive, which makes it impossible for the artificial neural network to learn the commonalities between multiple defect categories with common characteristics, and it may regard such commonalities as the unique characteristics of a certain defect type. As a result, when the artificial neural network classifies wafer defects, after identifying a certain characteristic, the recognition results will fluctuate between multiple defect types because multiple defect types have this characteristic, and the recognition accuracy is low.
[0038] In the method shown in this specification, classification is performed using a coarse-grained first classifier to obtain a first defect type, and the first defect type includes multiple sub-defect types with common characteristics. This allows more attention to be paid to the common characteristics of multiple sub-defect types when classification is performed using the first classifier, and the major category to which the defect data belongs is determined based on the common characteristics.
[0039] Classification is performed through a fine-grained second classifier, and the classification process is to determine the second defect type from multiple sub-defect types corresponding to the first defect type. This allows the classification process to be performed through detailed differences between multiple sub-defect types with common characteristics, rather than through common characteristics of multiple sub-defect types. This avoids the classification result from fluctuating between multiple defect types, effectively improving the accuracy of wafer defect classification and further improving semiconductor production efficiency and product quality.
[0040] Next, the embodiments of this specification are described in detail.
[0041] 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:
[0042] Step 201, obtaining defect data to be processed.
[0043] The defect data to be processed is data used to characterize a wafer defect image.
[0044] The method provided in this specification is used to classify wafer defects. This method can be applied in the research and development process or in the actual production process of wafers. This specification does not limit the application scenario of this method.
[0045] In addition, the method can be executed in real time during the wafer production process, or wafers can be produced first to obtain a large amount of defect data to be processed, and a unified wafer defect classification process can be performed on the large amount of defect data to be processed.
[0046] The defect data to be processed may specifically be a wafer defect image, or may be features extracted from the wafer defect image (hereinafter referred to as wafer defect features). This specification does not limit the specific form of the defect data to be processed.
[0047] In the case where the defect data to be processed is a wafer defect image, the first classifier and the second classifier may be an artificial neural network classifier capable of classifying images, and the classifier may have an encoder built in for feature extraction. In the case where the defect data to be processed is a wafer defect feature, the first classifier and the second classifier may be an artificial neural network capable of classifying the input wafer defect feature.
[0048] In the above case, to obtain the defect data to be processed, the wafer defect image must first be obtained. Next, the method for obtaining the wafer defect image will be described.
[0049] The wafer defect image may be a wafer image obtained by a scanning electron microscope (SEM), which can take pictures and scan the wafer to identify defects on the wafer. However, 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.
[0050] Step 203: Determine the first defect type to which the defect data to be processed belongs from the first defect type set by using a first classifier.
[0051] Each defect type in the first defect type set includes a plurality of sub-defect types having common characteristics.
[0052] Step 203 is the step of performing coarse-grained classification on the defect data to be processed mentioned above. Through the coarse-grained classification, it is possible to identify which major category of wafer defects the defect data to be processed belongs to.
[0053] The first defect type set may include multiple defect types, each of the multiple defect types includes multiple sub-defect types, and for the multiple sub-defect types included in any defect type, the multiple sub-defect types have common characteristics, and the common characteristics of the sub-defect types corresponding to different defect types are different.
[0054] In other words, the first defect type set includes multiple major categories, which are clearly distinguished from each other, each major category includes multiple minor categories, each major category includes multiple minor categories which have common characteristics, and there are differences in details between the multiple minor categories. Step 203 is to select the major category to which the defect data to be processed belongs from the multiple major categories, which is referred to as the first defect type here.
[0055] like Figure 3As shown, the first defect type set may include three defect types: defect type A, defect type B, and defect type C. The determined first defect type may be any one of defect type A, defect type B, and defect type C. Defect type A may include sub-defect type A1, sub-defect type A2, and sub-defect type A3, which have common characteristics. Defect type B may include sub-defect type B1 and sub-defect type B2, which also have common characteristics, which are different from the common characteristics of multiple sub-defect types of defect type A. Defect type C may include sub-defect type C1 and sub-defect type C2, which also have common characteristics, which are different from those of defect type A or defect type B.
[0056] The common features mentioned above may be common characteristics in terms of the shape or pattern of the defects, for example, the shape of the dents of multiple subtypes does not match the preset shape, etc. The common features may also be common characteristics in terms of the causes of the defects.
[0057] For ease of understanding, a specific example is used here to introduce the specific form of the first defect type set. The first defect type set may include two defect types: pattern fail and scratch. Pattern fail includes: Figure 1A The sub-defect types shown and Figure 1B The sub-defect types shown in the figure have a common feature that the patterns of both sub-defect types do not match the preset patterns; scratches include a sub-defect type of long scratches and a sub-defect type of deep scratches, and the common feature is that both are scratches, but the length and depth of the scratches are different. The above examples do not limit this specification, and the actual first defect type set may include more defect types.
[0058] Among them, multiple defect types in the first defect type set can be classified based on the pattern of wafer defects. For example, according to the general pattern of defects, the first defect type set can include defects such as pattern fail, scratches, and small holes.
[0059] In an optional implementation, each defect type in the first defect type set may be determined by an engineer based on observation, analysis, and past experience.
[0060] After the first defect type is described in detail, the first classifier and the specific implementation of step 203 will be described next.
[0061] The first classifier may be a neural network model obtained through training. The input of the classifier may be the defect data to be processed, and the output may be the defect type to which the defect data to be processed belongs among the various defect types included in the first defect type set.
[0062] In an optional implementation, the first classifier may be a neural network model trained with a large amount of training data. Since the first defect type includes multiple large categories, each category has a large amount of defect data, and thus the neural network model may be used to conveniently and quickly achieve a more accurate classification effect.
[0063] Among them, the vision transformer (ViT) can be selected as the first classifier, so that a better classification effect can be achieved. When the first classifier is ViT, the defect data to be processed can be wafer defect image data. ViT is an image classification model based on the attention mechanism, which can effectively process local and global information in the image. And the amount of image data contained in each defect type in the first defect type is relatively large, which is more suitable for processing using the ViT algorithm. By inputting the defect image on the wafer into the ViT algorithm model, all defects can be coarse-grainedly classified, and the first defect type can be determined from several large categories that do not affect each other (that is, the multiple defect types included in the first defect type set).
[0064] In an optional implementation, the first classifier may also be trained using defect data included in each defect type in the first defect type set and a label of each defect data.
[0065] Specifically, it is also possible to pre-: obtain a first training data set; the first training data set includes multiple defect data, and a first label for each defect data in the multiple defect data, the first label is used to characterize the defect type to which the defect data belongs in the first defect type set; use the defect data in the first training data set as input, and use the first label of the defect data in the first training data set as output, to train the first classifier.
[0066] That is, a plurality of defect data are acquired in advance, and a label is given to each defect data to form a first training data set, and the first classifier is trained by using the first training data set.
[0067] It should be noted that taking the first label as the output here means substituting the first label and the output of the model into the loss function, adjusting the model label according to the loss function, and training the first classifier.
[0068] The defect data included in the training data set is similar to the defect data to be processed, which can be a wafer defect image or a feature extracted from a wafer defect image. The first label can be used to identify the major category to which the defect data belongs, and the major category is a defect type in the first defect type, that is, the first label can identify the defect type to which the defect data belongs in the first defect type set.
[0069] Step 205 : determining, by a second classifier corresponding to the first defect type, from a plurality of sub-defect types included in the first defect type, a second defect type to which the defect data to be processed belongs.
[0070] Through step 205, fine-grained classification can be performed, and the second classifier is used to determine the sub-category (ie, the second defect type) to which the defect data to be processed belongs from the multiple sub-defect types included in the first defect type.
[0071] Step 203 is a coarse-grained classification, and step 205 is a fine-grained classification. The coarse-grained classification can obtain the large category to which the defect data to be processed belongs, and the fine-grained classification can obtain the small category to which the defect data to be processed belongs in the large category, thereby achieving the purpose of accurate classification.
[0072] For example, the pattern fail category mentioned in the example above includes Figure 1A The defects shown and Figure 1B If the first defect type determined in step 203 is pattern fail, step 205 uses the second classifier to classify the defect. Figure 1A The defects shown and Figure 1B Among the defects shown, determine the defect type to which the defect data to be processed belongs. For example, it can be determined that the defect data to be processed is Figure 1A The defect shown, i.e. the second defect type, is Figure 1A Defects shown.
[0073] The second classifier may be a trained artificial neural network, whose input is the defect data to be processed, and whose output is the sub-defect type to which the defect data to be processed belongs among the sub-defect types included in the first defect type.
[0074] In addition, the second classifier is the classifier corresponding to the first defect type, and different defect types in the first defect type set can correspond to different classifiers. For example, in the example mentioned above, there are two defect types, pattern fail and scratch, in the first defect type set. Then the pattern fail defect type corresponds to classifier 1, and the scratch defect type corresponds to classifier 2. The structures of these two classifiers can be the same or different, and the training data sets of the two classifiers are different. If the first defect type of the defect data to be processed is pattern fail, then classifier 1 can be used as the second classifier, and the classification result of the defect data to be processed can be determined by classifier 1.
[0075] In an optional implementation, the second classifier is a classifier that can be trained based on training data whose number is less than a preset threshold.
[0076] In other words, the second classifier is more suitable for processing classifiers of small magnitude. Some classification algorithms, such as the ViT algorithm, have certain requirements on the amount of training data. However, the amount of data within a large category (i.e., any defect type included in the first defect type) is uneven, and the overall magnitude is relatively small. If the ViT algorithm is used, it may not be possible to train a classifier with good results. Therefore, the second classifier with the above characteristics can be used to process it.
[0077] In an optional embodiment, the second classifier with the above characteristics can be a deep residual network (ResNet), which is a deep convolutional neural network widely used in image classification tasks. Within each large category, a ResNet network is used for subdivision. By inputting the defect data within the large category (which can be a wafer defect image) into the ResNet network, the defects can be further divided into various small categories to achieve fine classification of the defects.
[0078] When the first classifier is ViT and the second classifier is ResNet, ResNet and ViT, as a combination of outer and inner classification algorithms, can make full use of their respective advantages and improve the accuracy and efficiency of classification.
[0079] After describing the structure selection method of the second classifier, the training process of the second classifier will be briefly described next.
[0080] Similar to the first classifier, the second classifier can also be trained by the defect data corresponding to the first defect type and the second label. Specifically, the second classifier can be trained by the following method: obtaining a second training data set; the second training data set includes multiple defect data and a second label for each defect data in the multiple defect data, and the second label is used to characterize the sub-defect type to which the defect data belongs in the first defect type; using the defect data in the second training data set as input and the second label of the defect data in the second training data set as output to train the second classifier.
[0081] Specifically, the defect data corresponding to each sub-defect type can be obtained and marked with the corresponding sub-defect type (i.e., the second label). For each major category (i.e., each defect type included in the first defect type set), the defect data corresponding to the sub-defect type included in the major category and the second label are used as the second training data set to train the classifier corresponding to each major category. The classifier corresponding to the above-mentioned first defect type is the second classifier.
[0082] It should be noted that taking the second label as output means substituting the model output and the second label into the loss function, adjusting the model parameters according to the value of the loss function, and training the second classifier.
[0083] In addition, the training data sets for training the first classifier and the second classifier may be the same or different, and this specification does not limit this.
[0084] In the case where the training data sets for training the first classifier and the second classifier are the same, each defect data in the training data set may be annotated with two labels, namely a first label and a second label.
[0085] In addition, when the training data sets for training the first classifier and the second classifier are the same, each defect data in the training data set can also be labeled with only the second label, and the attribution relationship between the sub-defect type labeled by the second label and the defect type in the first defect type set can be preset, and then the first label corresponding to each defect data can be determined based on the attribution relationship and the second label corresponding to the data.
[0086] In other words, the first label is determined based on the manually annotated second label and the defect attribution relationship, wherein the defect attribution relationship shows the attribution relationship between the defect types and their sub-defect types in the first defect type set, and the second label is used to characterize the sub-defect types to which the defect data belongs.
[0087] For example, Figure 3As shown, the first defect type set includes three defect types: defect type A, defect type B and defect type C. Defect type A includes sub-defect type A1, sub-defect type A2 and sub-defect type A3. Figure 3 The relationship shown is the defect attribution relationship. The second label of a defect data in the training data set indicates that the defect data belongs to the sub-defect type A1. Then, according to the attribution relationship, the first label of the defect data can be determined to be the label with defect type A.
[0088] In this way, the labeling cost can be reduced. For example, the data labeled for the classifier in the related technology can be used. These data are often labeled with sub-defect types. Only the attribution relationship needs to be preset to train the first classifier and the second classifier, without the need for secondary labeling, thus reducing the labor cost of labeling.
[0089] Training the classifier corresponding to each major category can make the classification result more accurate. Through the above steps, accurate classification of wafer defects can be achieved, and effective support can be provided for subsequent production and improvement.
[0090] After completing the above step 205, the classification results may be verified and adjusted to ensure the classification accuracy and robustness, including reclassifying the incorrectly classified defect data or adjusting parameters to improve the accuracy and stability of the classification.
[0091] The embodiment shown in this specification provides a wafer defect classification method based on a two-layer nested classification algorithm to solve the problem of inaccurate classification in traditional methods. First, the wafer defects are divided into several large categories with obvious distinguishing features between these large categories, and then the large categories are classified using the first layer classification algorithm. Subsequently, within each large category, the defects are further subdivided and identified using the second layer classification algorithm. Through this method, the classification accuracy of wafer defects can be effectively improved and production efficiency can be improved.
[0092] Traditional classification methods usually directly classify all defects, lacking the distinction and understanding of defects of different levels. The combination of two classification algorithms used in the embodiments of this specification has the following advantages:
[0093] First, the classification accuracy is improved. Dividing defects into large categories and small categories helps reduce the complexity of classification and makes the classification at each level more accurate. The division of large categories can quickly classify defects into the most obvious categories, while the subdivision of small categories can more finely distinguish each type of defect.
[0094] Second, identify complex defects: When faced with complex and diverse defects, the direct classification method may face difficulties in dealing with complex and diverse defects, because a single classification algorithm often cannot cover all situations. By dividing the classification process into two stages, various types of defects can be better understood and identified, thereby improving the robustness and applicability of the classification.
[0095] Third, guide subsequent processing: Direct classification methods may not provide sufficiently accurate classification results, resulting in low efficiency of subsequent processing and improvement. The present invention can provide more effective guidance for subsequent processing and improvement by providing refined classification results, facilitate timely adjustment of production equipment parameters, and reduce the generation of wafer defects. Different types of defects may require different processing methods, so accurate classification results can improve the efficiency of subsequent processing and cost control.
[0096] 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:
[0097] The data acquisition module 410 is used to acquire defect data to be processed; the defect data to be processed is data used to characterize a wafer defect image.
[0098] The first classification module 420 is used to determine the first defect type to which the defect data to be processed belongs from a first defect type set by using a first classifier; each defect type in the first defect type set includes a plurality of sub-defect types having common characteristics.
[0099] The second classification module 430 is configured to determine, by using a second classifier corresponding to the first defect type, a second defect type to which the defect data to be processed belongs from a plurality of sub-defect types included in the first defect type.
[0100] In an optional implementation, the second classifier is a classifier that can be trained based on training data whose number is less than a preset threshold.
[0101] In an optional implementation, the second classifier is a deep residual network ResNet.
[0102] In an optional implementation, the first classifier is a visual transformer ViT.
[0103] In an optional embodiment, the device also includes a first classifier training module (not shown in the figure), which is used to: obtain a first training data set; the first training data set includes multiple defect data, and a first label for each defect data in the multiple defect data, and the first label is used to characterize the defect type to which the defect data belongs in the first defect type set; use the defect data in the first training data set as input, and use the first label of the defect data in the first training data set as output to train the first classifier.
[0104] In an optional embodiment, the device also includes a second classifier training module (not shown in the figure), which is used to: obtain a second training data set; the second training data set includes multiple defect data, and a second label for each defect data in the multiple defect data, and the second label is used to characterize the sub-defect type to which the defect data belongs in the first defect type; use the defect data in the second training data set as input, and use the second label of the defect data in the second training data set as output to train the second classifier.
[0105] In an optional embodiment, the first label is determined based on a manually annotated second label and a defect attribution relationship, wherein the defect attribution relationship shows the attribution relationship between defect types in the first defect type set and their sub-defect types, and the second label is used to characterize the sub-defect types to which the defect data belongs.
[0106] 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.
[0107] 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.
[0108] like Figure 5 As shown, Figure 5A hardware structure diagram of a computer device in which a wafer defect classification device of an embodiment is located is shown, and the device may include: a processor 1010, a memory 1020 for storing computer programs, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other in communication within the device through the bus 1050.
[0109] The processor 1010 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.
[0110] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may 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 1010.
[0111] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (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.
[0112] 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.).
[0113] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).
[0114] It should be noted that, although the above 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, 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 the present specification, and does not necessarily include all the components shown in the figure.
[0115] 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.
[0116] 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.
[0117] 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.
[0118] 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.
[0119] 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.
[0120] 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.
[0121] 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 defect data to be processed; the defect data to be processed is data used to characterize a wafer defect image; Determining a first defect type to which the defect data to be processed belongs from a first defect type set by a first classifier; each defect type in the first defect type set includes a plurality of sub-defect types having common characteristics; The second defect type to which the defect data to be processed belongs is determined from a plurality of sub-defect types included in the first defect type by using a second classifier corresponding to the first defect type.
2. The method according to claim 1, characterized in that: The second classifier is a classifier that can be trained based on training data whose quantity is less than a preset threshold.
3. The method according to claim 2, characterized in that The second classifier is a deep residual network ResNet.
4. The method according to claim 1, characterized in that: The first classifier is a visual transformer ViT.
5. The method according to claim 1, characterized in that The method further comprises: Acquire a first training data set; the first training data set includes a plurality of defect data and a first label of each defect data in the plurality of defect data, the first label being used to characterize the defect type to which the defect data belongs in the first defect type set; The first classifier is trained by taking the defect data in the first training data set as input and taking the first label of the defect data in the first training data set as output.
6. The method according to claim 1, characterized in that The method further comprises: Acquire a second training data set; the second training data set includes a plurality of defect data and a second label of each defect data in the plurality of defect data, the second label being used to characterize the sub-defect type to which the defect data belongs in the first defect type; The second classifier is trained by taking the defect data in the second training data set as input and taking the second label of the defect data in the second training data set as output.
7. The method according to claim 5, characterized in that: The first label is determined based on the manually annotated second label and the defect attribution relationship, wherein the defect attribution relationship shows the attribution relationship between the defect types in the first defect type set and their sub-defect types, and the second label is used to characterize the sub-defect types to which the defect data belongs.
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 7 is implemented.
9. A computer device, characterized in that: The computer device comprises: processor; Memory, for 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 7.
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 7.
Citation Information
Cited By
Wafer defect classification method and device and computer equipment
CN119832351A
A wafer defect classification method, device, and computer equipment
CN119832351B
Wafer defect classification method and device, storage medium and electronic equipment
CN120823450A
