A method, device, equipment and readable medium for determining wafer defect mode

Through the fusion of machine vision and machine learning models, multi-model fusion calculation is performed using image data and information data, the problem of insufficient sample size in wafer defect pattern classification is solved, the recognition accuracy and robustness are improved, and the existing solutions are solved The problem of low accuracy in technology.

CN117437455BActive Publication Date: 2024-11-12PENGXI SEMICONDUCTOR TECHNOLOGY (BEIJING) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311218870.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-09-20
Publication Date
2024-11-12
Estimated Expiration
2043-09-20

AI Technical Summary

Technical Problem

In the prior art, machine vision deep learning models have insufficient sample size in wafer defect pattern classification, resulting in low accuracy of recognition results, long R&D cycle, and complex image data source provision, affecting classification accuracy.

Method used

The fusion method of machine vision and machine learning models is adopted, and image data and information data are obtained, respectively input into pre-trained models for classification. The fusion algorithm is used to calculate the results to improve the classification accuracy and solve the problem of small sample size and large categories.

Benefits of technology

Through multi-model fusion calculation, the accuracy and robustness of the wafer recognition process are improved, and the problem of a single model being limited by the sample data volume is improved, and the accuracy of the recognition results is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117437455B_ABST
    Figure CN117437455B_ABST
Patent Text Reader

Abstract

The present application provides a method, device, equipment and readable medium for determining a wafer defect pattern. The method includes: acquiring image data and information data of a wafer to be identified; inputting the image data into a pre-trained first data model to obtain a first classification result of the first data model for defect pattern prediction; and inputting the information data into a pre-trained second data model to obtain a second classification result of the second data model for defect pattern prediction; fusing the first classification result with the second classification result based on a predetermined fusion algorithm to obtain a fusion result; determining the defect pattern of the wafer to be identified based on the fusion result. This technical solution can get rid of the problem of relying on a single model for identification being limited by the amount of sample data, and the fusion calculation can obtain higher accuracy, which can more effectively improve the accuracy and robustness of the wafer identification process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of information technology, and in particular to a method, device, equipment and readable medium for determining a wafer defect mode. Background Art

[0002] In recent years, with the rapid development of science and technology, the manufacture and use of wafers has become one of the important foundations of scientific and technological development. In the existing technology, the classification of defect patterns on wafers is generally carried out by using a deep learning model of machine vision. The general process is wafer image acquisition, wafer image enhancement processing, and the construction of the initial network structure of the deep learning model for training and use.

[0003] However, the inventors found that there are at least the following technical problems in the related art:

[0004] If machine vision technology is used to classify defect patterns, the sample size of the image data source is small and the complexity and variability are low, while the target categories expected to be output by the model are large, which often results in low accuracy and inaccurate classification of wafer defect patterns.

[0005] However, if a large number of image data sources and manually labeled categories are used to improve the accuracy of the output classification of the deep learning model of machine vision, there will be a long R&D cycle, and the provision of image data sources requires cooperation from multiple parties. There will still be a problem of low improvement in the accuracy of the output classification. Summary of the invention

[0006] One purpose of the present application is to provide a method, device, equipment and readable medium for determining a wafer defect pattern, at least to solve the problems of insufficient samples of deep learning models of machine vision, low accuracy of recognition results, and long R&D cycle. The purpose of the present application is to provide a new method for determining defect patterns. The method adopts a fusion of two models, machine vision and machine learning, to improve classification accuracy through model fusion. Machine learning can perform well even when the sample size is small, with low-order feature construction and good feature engineering methods, and has good robustness to different data distributions and noise distributions, which can improve the memory capacity of the model, and combined with machine vision deep learning, it has a strong feature extraction capability, and can learn high-order complex features, thereby improving the generalization ability of the model. Therefore, the fusion of the two solves the problem of low accuracy caused by small sample size and many categories.

[0007] To achieve the above objectives, some embodiments of the present application provide the following aspects:

[0008] In a first aspect, some embodiments of the present application further provide a method for determining a wafer defect mode, the method comprising:

[0009] Acquire image data and information data of the wafer to be identified;

[0010] Inputting the image data into a pre-trained first data model to obtain a first classification result of defect mode prediction by the first data model; and inputting the information data into a pre-trained second data model to obtain a second classification result of defect mode prediction by the second data model;

[0011] Performing a fusion calculation on the first classification result and the second classification result based on a predetermined fusion algorithm to obtain a fusion result;

[0012] The defect mode of the wafer to be identified is determined according to the fusion result.

[0013] In a second aspect, some embodiments of the present application further provide a device for determining a wafer defect mode, the device comprising:

[0014] A data acquisition module, used to acquire image data and information data of the wafer to be identified;

[0015] A classification result output module is used to input the image data into a pre-trained first data model to obtain a first classification result of the first data model for defect mode prediction; and input the information data into a pre-trained second data model to obtain a second classification result of the second data model for defect mode prediction;

[0016] A fusion calculation module, used for performing fusion calculation on the first classification result and the second classification result based on a predetermined fusion algorithm to obtain a fusion result;

[0017] A defect mode determination module is used to determine the defect mode of the wafer to be identified according to the fusion result.

[0018] In a third aspect, some embodiments of the present application further provide a computer device, the device comprising:

[0019] one or more processors; and

[0020] A memory storing computer program instructions, which, when executed, cause the processor to perform the method for determining a wafer defect mode as described above.

[0021] In a fourth aspect, some embodiments of the present application further provide a computer-readable medium having computer program instructions stored thereon, wherein the computer program instructions can be executed by a processor to implement the method for determining a wafer defect mode as described above.

[0022] Compared with the prior art, in the solution provided by the embodiment of the present application, by acquiring the image data and information data of the wafer to be identified, two levels of data are obtained, and the two levels of data are respectively input into the pre-trained first data model and the second data model, and the first classification result of the defect mode prediction by the first data model and the second classification result of the defect mode prediction by the second data model are used to perform fusion calculation to obtain the fusion result, and the defect mode of the wafer to be identified is determined by the fusion result. Through such a multi-model fusion calculation setting, the problem of relying on a single model for identification being limited by the amount of sample data is eliminated, and the fusion calculation obtains higher accuracy, which can more effectively improve the accuracy and robustness of the wafer identification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flowchart of a method for determining a wafer defect mode provided in Embodiment 1 of the present application;

[0024] Figure 2 It is a flowchart of a method for determining a wafer defect mode provided in Embodiment 2 of the present application;

[0025] Figure 3 is a schematic diagram of a process for determining a wafer defect mode provided in Embodiment 3 of the present application;

[0026] Figure 4 It is a structural schematic diagram of a device for determining a wafer defect mode provided in Embodiment 4 of the present application;

[0027] Figure 5 It is a structural diagram of the computer device provided in Example 5 of the present application. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] The technical solution provided by the embodiment of the present application is described in detail below through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0030] Embodiment 1

[0031] Figure 1 FIG. 1 is a flow chart of a method for determining a wafer defect mode provided in Embodiment 1 of the present application. Figure 1 As shown, the specific steps include: The process includes:

[0032] Step S101, acquiring image data and information data of a wafer to be identified;

[0033] The wafer to be identified may be a wafer that is subjected to quality inspection during or after wafer production.

[0034] Wafer refers to the silicon wafer used to make silicon semiconductor circuits, and its original material is silicon. High-purity polysilicon is dissolved and doped with silicon crystal seeds, and then slowly pulled out to form cylindrical single crystal silicon. After grinding, polishing and slicing, the silicon crystal rod is formed into a silicon wafer, that is, a wafer. Currently, the wafer production lines on the market are mainly 8 inches and 12 inches. The main processing methods for wafers are sheet processing and batch processing, that is, processing one or more wafers at the same time. As the feature size of semiconductors becomes smaller and smaller, the processing and measurement equipment becomes more and more advanced, which makes the wafer processing size also reduce.

[0035] In this solution, the wafer inspection method includes image detection after image acquisition, and the latter is to detect after acquiring corresponding information and combining corresponding information. Here, image data can be collected based on demand, for example, image acquisition is performed according to the required image resolution, and image acquisition is often taken directly above the wafer.

[0036] Among them, the information data can be a klarf file, that is, the surface defect information of the wafer to be identified can be recorded as a defect information file. In this scheme, the klarf file can be a document used to store and access defect information data, such as xrel, yrel, xindex, yindex coordinate position and other information.

[0037] Step S102, inputting the image data into a pre-trained first data model to obtain a first classification result of defect mode prediction by the first data model; and inputting the information data into a pre-trained second data model to obtain a second classification result of defect mode prediction by the second data model;

[0038] In this solution, the defect mode may be a defect generation mode of a wafer image, such as a point defect, a line defect, or a regional defect.

[0039] Point defects refer to local defects on the surface of wafer chips, which are usually caused by raw material contamination or improper manufacturing process. Point defects are usually divided into the following categories:

[0040] 1. Cave defects: usually caused by bubbles or gas remaining in the wafer chip.

[0041] 2. Bubble defects: usually caused by uneven expansion of metal materials or gas residue in the wafer chip.

[0042] 3. Metal particles: Metal particles are caused by the sputtering or adhesion of metal materials during the manufacturing process.

[0043] Line defects refer to local defect lines on the surface of wafer chips, which are usually caused by manufacturing processes or equipment failures. Line defects are usually divided into the following categories:

[0044] 1. Scratches: usually caused by wear and tear of equipment during the manufacturing process or impact during the process.

[0045] 2. Streaks: Usually caused by improper ingredients during the manufacturing process or uneven materials inside the wafer chip.

[0046] 3. Bumps or dents: usually caused by damage to equipment or tools during the manufacturing process.

[0047] Regional defects refer to local defects on wafer chips, which are usually caused by manufacturing process or equipment failure. Regional defects are usually divided into the following categories:

[0048] 1. Grain defect: If a grain on the wafer chip is defective or shrunk, the performance of the wafer chip will be reduced or even fail.

[0049] 2. Uneven Vickers hardness: usually caused by uneven pressure, uneven materials or equipment failure during wafer chip manufacturing.

[0050] 3. Thermal residual stress: Temperature changes during wafer chip manufacturing will cause residual stress in the wafer chip. If these residual stresses are too large, the performance of the wafer chip will be reduced or even fail.

[0051] The above are the common defect classifications and related characteristics of wafer chips. During the manufacturing and testing process, these defects need to be effectively detected and classified to ensure the quality and performance of wafer chips.

[0052] The first data model can be a model trained based on historically collected image data, and the model can obtain a prediction result of the defect mode after inputting an image. Taking the input image to identify the animal category as an example, after training, the model can output the category of the animal in the image. For example, if an image of an animal is input, the output of the animal is: the probability of a dog is 0.7, the probability of a cat is 0.2, and the probability of a cow is 0.1.

[0053] The second data model can be a model trained based on the information data of the wafer collected historically. The model can obtain the prediction result of the defect pattern after inputting the information data of the wafer. Taking the coordinate data of the defect point on the input wafer as an example, after the model is trained, it can output the defect category of the wafer. For example, a set of coordinate data with 30 defects is input, and the defect category of the wafer is output as follows: the probability of cave defects is 0.8, the probability of bubble defects is 0.1, and the probability of metal particles is 0.1. Among them, cave defects are usually caused by bubbles or gas residues in the wafer chip. Bubble defects are usually caused by uneven expansion of metal materials or gas residues in wafer chips. Metal particles are caused by sputtering or adhesion of metal materials during the manufacturing process.

[0054] As can be understood from the above examples, the first classification result and the second classification result may include multiple classification categories and probability values ​​corresponding to each category. For the first classification result and the second classification result, there may be overlapping parts or non-overlapping parts. For the non-overlapping parts, in the classification results of the model that did not predict the category, its probability value may be set to 0.

[0055] In some embodiments of the present application, the training process of the first data model includes:

[0056] Obtain historical image data of wafer samples;

[0057] Performing enhancement processing on the historical image data to obtain image sample data;

[0058] The sample data is input into an initial model for model training to obtain a first data model.

[0059] In the technical solution, in the process of training the first data model, historical image data of wafer samples can be obtained and directly input into the initial model for model training, or the data can be input into the initial model for model training after image enhancement to obtain the first data model.

[0060] Specifically, the training method can be supervised training or unsupervised training. Enhancement is the action or process of increasing the scale or number.

[0061] In deep learning, deep networks need a lot of training data to generalize well and achieve good accuracy. But in some cases, the amount of image data is not large enough. In this case, some techniques need to be used to increase the training data, for example, using techniques such as random rotation, displacement, shearing and flipping.

[0062] Image enhancement is the process of generating new images for the purpose of training deep learning models. These new images are generated using existing training images. Different image enhancement techniques can be used to enhance images using various techniques. For example: spatial enhancement, scaling, flipping, rotation, shearing, translation, pixel enhancement, brightness, contrast, saturation, hue, etc.

[0063] Data augmentation is a common practice in deep learning. Therefore, each deep learning framework has its own augmentation methods or even a whole library for it. For example, let’s see how to apply image augmentation using built-in methods in Keras, PyTorch, and Albumentations.

[0064] Keras' ImageDataGenerator class provides a quick and easy way to augment images. It provides many different augmentation techniques such as normalization, rotation, shift, flip, brightness change, and many more. The main benefit of using Keras' ImageDataGenerator class is the ability to provide real-time data augmentation. This means that it generates augmented images while the model is in the training phase. The ImageDataGenerator class ensures that the model receives new variations of the images at each epoch. But it only returns the transformed images and does not add them to the original image dataset. Another advantage of ImageDataGenerator is that it has a low memory footprint. It supports a range of image augmentation methods, and we will now focus on five main types of methods.

[0065] PyTorch is a Python based library that helps in building deep learning models and using them in various applications. But it is not just a deep learning library, it is also a scientific computing library. The main advantage of using PyTorch is that you can apply image augmentation techniques individually on selected images.

[0066] Starting with importing the images, we will define the imshow() function to visualize the actual and transformed images.

[0067] Scale: When scaling or resizing, resize the image to a given size.

[0068] Cropping: In cropping, a portion of the image is selected, e.g. in the given example, a center cropped image is returned.

[0069] Flip: When flipping, the image is flipped horizontally or vertically.

[0070] Pixel enhancement: Pixel enhancement is the process of changing the color properties of an image by changing the pixel values ​​of the image.

[0071] Albumentations is a computer vision tool that improves the performance of deep convolutional neural networks. Albumentations is a Python library for fast and flexible image augmentation. It efficiently implements a rich variety of image transformation operations that are optimized for performance while providing a concise and powerful image augmentation interface for different computer vision tasks, including object classification, segmentation, and detection.

[0072] In this embodiment, using image data augmentation when training a deep learning neural network can improve the performance and generalization ability of the model.

[0073] The technical solution provided in this embodiment can obtain more accurate prediction results by enhancing the image during the training process of the first data model.

[0074] In some embodiments of the present application, the information data includes a defect information data document;

[0075] Accordingly, the training process of the second data model includes:

[0076] Obtain defect information data files of wafer samples;

[0077] Structuring the defect information data document to obtain structured data; wherein the structured data includes derived features;

[0078] Performing feature engineering processing on the structured data to obtain training sample data;

[0079] The sample data is input into the initial model for model training to obtain a second data model.

[0080] In the process of training the second data model, klarf data of the defect pattern can be collected and organized into structured data. The klarf data contains the coordinate position information of the defect, the number of defects on each wafer map, and the generation of derived features such as standard deviation and mean of the defect appearance distribution. Feature engineering is performed on the generated structured data, such as dimensionality increase and dimensionality reduction, to construct memory features for low-order features. Feature engineering is to perform a series of engineering processes on the original data, refine it into features, and use it as input for algorithms and models. Common data types include structured data and unstructured data.

[0081] Structured data includes numerical data and categorical data. To a certain extent, numerical data can be converted into categorical data.

[0082] Unstructured data, such as text, images, audio and video. This type of data has no specific category, and the size of each piece of data varies.

[0083] Different types of data features have different dimensions. In order to eliminate the dimensional impact of different data, it is necessary to normalize the features to make different indicators comparable. For example, if you use meters (m) and kilograms (kg) as units to analyze the impact of a person's height and weight on health, the height feature will be in the range of 1.6 to 1.8m, and the weight feature will be in the range of 50 to 100kg. The results of the analysis will obviously tend to favor the weight feature with a larger numerical difference. To get more accurate results, it is necessary to perform feature normalization so that each indicator is at the same numerical level for analysis.

[0084] This solution can obtain features that more accurately represent the defect mode of the wafer through feature engineering processing of structured data.

[0085] The advantage of such a configuration of the present technical solution is that it can improve the standardization of the input data of the second data model, making the model training faster and more accurate, so as to output more accurate prediction results through the second data model.

[0086] Step S103, performing fusion calculation on the first classification result and the second classification result based on a predetermined fusion algorithm to obtain a fusion result;

[0087] In this solution, the fusion algorithm can be based on the classification results output by the two models, and the fusion calculation is performed to obtain an algorithm for uniformly outputting probability values ​​for various categories. The fusion algorithm can be determined based on the parameter data of the two models, or it can be determined based on different categories. For example, when the first classification result and the second classification result are point defects, the second classification result has a greater weight in the fusion calculation process; when the first classification result and the second classification result are line defects or regional defects, the first classification result has a greater weight in the fusion calculation process.

[0088] It can be understood that the weight of the fusion calculation can be calculated for multiple classification results output by each model, or the probability values ​​output by the two models can be weighted differently for different classification results. For example, for the first classification result, the weight is fixedly set to 0.4, then each classification in the first classification structure is multiplied by a weight of 0.4 in the fusion calculation, and at the same time, the weight of the second classification result is fixedly set to 0.6, and the corresponding calculation is performed. For other situations, for example, different calculations are performed for each classification, for example, the first classification result and the second classification result both include three categories A, B, and C. When the fusion calculation is performed for category A, the probability value of the first classification result is multiplied by a weight of 0.4, and the probability value of the second classification result is multiplied by a weight of 0.6. When the fusion calculation is performed for category B, the probability value of the first classification result is multiplied by a weight of 0.5, and the probability value of the second classification result is multiplied by a weight of 0.5. When the fusion calculation is performed for category C, the probability value of the first classification result is multiplied by a weight of 0.7, and the probability value of the second classification result is multiplied by a weight of 0.3.

[0089] Step S104, determining the defect mode of the wafer to be identified according to the fusion result.

[0090] In this solution, the fusion result may include multiple categories and the probability value obtained after the fusion calculation of each category. The defect mode of the wafer to be identified can be determined based on the fusion result. For example, after the fusion calculation, the probability of a cave defect is 0.75, the probability of a bubble defect is 0.15, and the probability of a metal particle is 0.1. Based on the fusion result, it can be determined that the defect mode of the wafer to be identified is a cave defect.

[0091] In some embodiments of the present application, determining the defect mode of the wafer to be identified according to the fusion result includes:

[0092] Identify whether the fusion result for the current defect mode in the fusion result is greater than a set threshold;

[0093] If yes, it is determined that the defect mode of the wafer to be identified belongs to the current defect mode;

[0094] If not, it is determined that the defect mode of the wafer to be identified does not belong to the current defect mode.

[0095] The threshold setting may be predetermined and may be adjusted as the model is continuously updated during use. Here, for example, the threshold setting may be 0.5. If the probability value of each defect mode included in the fusion result is greater than the threshold setting of 0.5, it can be determined that the current defect mode is the defect mode of the wafer to be identified. Here, it can be understood that the probability value of each defect mode can be identified by traversing to obtain the final result.

[0096] It is understandable that after the fusion calculation, the probability value of any defect mode in the fusion result may be less than the set threshold value of 0.5. In this case, a corresponding prompt message can be generated to re-acquire the image data or use manual recognition to determine the defect mode.

[0097] The advantage of setting up this scheme is that the defect pattern of the wafer can be identified based on the fusion result and the predetermined set threshold. Based on the comparison with the set threshold, the identification result of the wafer defect pattern can be obtained, and when none of the set thresholds are met, corresponding processing measures are provided, so that the scope of application of this scheme will be greatly improved.

[0098] Embodiment 2

[0099] Figure 2 FIG. 1 is a flow chart of a method for determining a wafer defect mode provided in Embodiment 2 of the present application. Figure 2 As shown, the specific steps include:

[0100] Step S201, acquiring image data and information data of a wafer to be identified;

[0101] Step S202, inputting the image data into a pre-trained first data model to obtain a first classification result of defect mode prediction by the first data model; and inputting the information data into a pre-trained second data model to obtain a second classification result of defect mode prediction by the second data model;

[0102] Step S203, obtaining a first evaluation index of the first data model and a second evaluation index of the second data model;

[0103] The first evaluation index and the second evaluation index may be indicators used to evaluate the accuracy of the first data model and the second data model. For example, after the model training is completed, a test sample may be constructed to test the trained model to obtain the accuracy of each model, and the accuracy may be determined as the first evaluation index and the second evaluation index, respectively.

[0104] In this solution, it can also be the confidence of the model. In addition, for the model, its own structure can also be used as an evaluation indicator, or as part of the evaluation indicator. For example, if the structure of the first data model is 10 layers, and the structure of the second evaluation indicator is 3 layers, the difference in the number of layers can be used as an evaluation indicator, or as part of the evaluation indicator. In addition, the evaluation indicators of each model can also be determined based on the training time of the model, the number of samples used, etc.

[0105] Step S204, determining, in the fusion algorithm, a first fusion weight of the first classification result and a second fusion weight of the second classification result according to the first evaluation index and the second evaluation index;

[0106] The fusion weight of the fusion algorithm can be determined according to the first evaluation index and the second evaluation index. For example, if the first evaluation index is 10 layers and the second evaluation index is 3 layers, the first fusion weight of the first classification result in the fusion algorithm can be determined to be 10 / 13, and the second fusion weight of the second classification result in the fusion algorithm can be determined to be 3 / 13.

[0107] In some embodiments of the present application, determining, in the fusion algorithm, a first fusion weight of the first classification result and a second fusion weight of the second classification result according to the first evaluation index and the second evaluation index includes:

[0108] The calculation is performed using the following formula:

[0109] A_weight=A_accuracy / (A_accuracy+B_accuracy);

[0110] B_weight=B_accuracy / (A_accuracy+B_accuracy);

[0111] Among them, A_weight is the first fusion weight, B_weight is the second fusion weight, A_accuracy is the first evaluation index, and B_accuracy is the second evaluation index.

[0112] In this solution, the first evaluation index may be the accuracy of the first data model, and the second evaluation index may be the accuracy of the second data model.

[0113] After the first evaluation index and the second evaluation index are determined, the first fusion weight of the first classification result and the second fusion weight of the second classification result can be determined. For example, the first fusion weight can be obtained by adding the accuracy of the two models and using the accuracy of the first data model as the numerator to obtain the corresponding weight value. The second fusion weight is similar.

[0114] Through such calculation, the sum of A_weight and B_weight is 1, which meets our basic requirements for weight setting.

[0115] The advantage of this setting of the scheme is that the respective fusion weights of the two models in the fusion calculation process can be determined based on the accuracy of the two models, so that a more accurate fusion result can be obtained after the fusion calculation.

[0116] Step S205, for each defect mode, based on the first classification result and the first fusion weight, and the second classification result and the second fusion weight, a fusion calculation is performed to obtain a fusion result for the current defect mode;

[0117] After the first fusion weight and the second fusion weight are determined, various defect modes included in the classification results of the first data model and the second data model may be calculated one by one to obtain a probability value of each defect mode.

[0118] In some embodiments of the present application, for each defect mode, a fusion calculation is performed based on the first classification result and the first fusion weight, and the second classification result and the second fusion weight to obtain a fusion result for the current defect mode, including:

[0119] The calculation is performed using the following formula:

[0120] P i =a i *A_weight+b i *B_weight;

[0121] Among them, P i is the fusion result of the current defect mode, a i is the first classification probability value of the current defect mode predicted by the first data model, b i is the second classification probability value of the current defect mode predicted by the second data model, A_weight is the first fusion weight, and B_weight is the second fusion weight.

[0122] In this solution, after the first fusion weight and the second fusion weight are determined, the probability value of each defect mode can be calculated respectively.

[0123] The advantage of this setting of the scheme is that the predicted probability value after the fusion of the two models can be calculated for each defect mode according to the fusion weight, which improves the robustness of the process of determining the defect mode of the wafer and has more accurate precision.

[0124] Step S206, determining the defect mode of the wafer to be identified according to the fusion result.

[0125] Based on the above embodiments, this embodiment provides a specific method for determining fusion weights and a method for fusion calculation. In actual operation, based on such a fusion method, a multi-model output result that is more in line with expectations can be obtained, thereby improving the accuracy of identifying defect patterns of wafers.

[0126] Embodiment 3

[0127] Figure 3 FIG. 1 is a schematic diagram of the process of determining a wafer defect mode provided in the third embodiment of the present application. Figure 3 As shown, the specific steps include:

[0128] S1: Machine vision construction classification model A;

[0129] Collect the image to be classified, perform image enhancement, and then build a deep learning network structure model, enter the deep learning model for training and prediction to output the classification probability a;

[0130] S2: Machine learning constructs classification model B;

[0131] 1. Collect the klarf data of the defect pattern and organize the klarf data into structured data (including the coordinate location information of the defect, the number of defects on each wafer map, and the generation of derived features such as the standard deviation and mean of the defect appearance distribution);

[0132] 2. Perform feature engineering on the generated structured data, such as dimensionality increase and dimensionality reduction, and construct memory features for low-order features;

[0133] 3. Build a machine learning model for training and predict the output classification probability b;

[0134] S3: Fusion of machine vision model and machine learning model;

[0135] The above output probabilities a and b are fused and judged, and weights are allocated according to the evaluation indicators of the model. The formula is as follows:

[0136] Weight division:

[0137] A_weight=A_accuracy / (A_accuracy+B_accuracy);

[0138] B_weight=B_accuracy / (A_accuracy+B_accuracy);

[0139] Among them, A_weight+B_weight=1;

[0140] Classification probability after fusion:

[0141] P = a*A_weight+b*B_weight;

[0142] Here, if P ≥ 0.5, it belongs to the current prediction category, and if P < 0.5, it does not belong to the current prediction category, and the process ends.

[0143] The defect pattern classification method in this patented technical solution improves deep learning and machine learning into a fusion mode. Since the improvement utilizes machine learning for low-order feature construction and good feature engineering methods, it can perform well even when the sample size is small, and is more robust to data distribution and noise, which can improve the model's memory capacity. Machine vision deep learning has powerful automatic feature extraction capabilities and can learn high-order complex features, thereby improving the model's generalization ability. Therefore, the fusion of the two solves the problem of low accuracy caused by a small sample size and many categories. Through the method of this patent, the accuracy of the model can be greatly improved, producing a good model fusion effect.

[0144] This solution achieves higher accuracy in defect pattern classification by integrating machine learning and deep learning models. By integrating machine learning models, it can achieve better performance even when the sample size is small and is more robust to data distribution and noise. The model's memory and generalization capabilities are comprehensively improved by the feature engineering method of low-order features and the automatic extraction of high-order features.

[0145] Embodiment 4

[0146] Figure 4 Schematic diagram of the structure of the device for determining the wafer defect mode provided in the fourth embodiment of the present application. Figure 4 As shown, specifically including the following:

[0147] The data acquisition module 410 is used to acquire image data and information data of the wafer to be identified;

[0148] The classification result output module 420 is used to input the image data into a pre-trained first data model to obtain a first classification result of the first data model for defect mode prediction; and input the information data into a pre-trained second data model to obtain a second classification result of the second data model for defect mode prediction;

[0149] A fusion calculation module 430 is used to perform fusion calculation on the first classification result and the second classification result based on a predetermined fusion algorithm to obtain a fusion result;

[0150] The defect mode determination module 440 is used to determine the defect mode of the wafer to be identified according to the fusion result.

[0151] The wafer defect mode determination device in the embodiment of the present application can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc., and the non-mobile electronic device can be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present application.

[0152] The wafer defect mode determination device in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0153] The wafer defect mode determination device provided in the embodiment of the present application can implement each process implemented in the above method embodiment, and will not be described again here to avoid repetition.

[0154] Embodiment 5

[0155] In addition, the embodiment of the present application also provides a computer device, Figure 5 Schematic diagram of the structure of the computer device provided in Example 5 of the present application. Figure 5 As shown, the device includes a memory 51 for storing computer-readable instructions and a processor 52 for executing the computer-readable instructions, wherein when the computer-readable instructions are executed by the processor, the processor is triggered to execute the method described.

[0156] The methods and / or embodiments in the embodiments of the present application may be implemented as computer software programs. For example, the embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program code for executing the method shown in the flowchart. When the computer program is executed by the processing unit, the above functions defined in the method of the present application are executed.

[0157] It should be noted that the computer-readable medium described in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device.

[0158] In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0159] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0160] The flow chart or block diagram in the accompanying drawings shows the possible architecture, function and operation of the equipment, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated system for hardware that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0161] As another aspect, the embodiments of the present application further provide a computer-readable medium, which may be included in the device described in the above embodiments; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more computer-readable instructions, which may be executed by a processor to implement the steps of the methods and / or technical solutions of the above multiple embodiments of the present application.

[0162] In a typical configuration of the present application, the terminal and the equipment of the service network each include one or more processors (CPU), input / output interface, network interface and memory.

[0163] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0164] Computer readable media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, modules of programs 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, read-only compact disk (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0165] In addition, an embodiment of the present application further provides a computer program, which is stored in a computer device, so that the computer device executes the method for controlling code execution.

[0166] It should be noted that the present application can be implemented in software and / or a combination of software and hardware, for example, can be implemented using an application specific integrated circuit (ASIC), a general purpose computer or any other similar hardware device. In certain embodiments, the software program of the present application can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including related data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and similar devices. In addition, some steps or functions of the present application can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform each step or function.

[0167] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or basic features of the present application. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present application is limited by the attached claims rather than the above description, so it is intended to include all changes that fall within the meaning and scope of the equivalent elements of the claims in the present application. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

Claims

1. A method for determining a wafer defect mode, characterized in that: The method comprises: Acquire image data and information data of a wafer to be identified; wherein the information data is surface defect information of the wafer to be identified; Inputting the image data into a pre-trained first data model to obtain a first classification result of defect mode prediction by the first data model; and inputting the information data into a pre-trained second data model to obtain a second classification result of defect mode prediction by the second data model; Performing a fusion calculation on the first classification result and the second classification result based on a predetermined fusion algorithm to obtain a fusion result; Determining a defect mode of the wafer to be identified according to the fusion result; The first classification result and the second classification result are fused based on a predetermined fusion algorithm to obtain a fusion result, including: Obtaining a first evaluation index of the first data model and a second evaluation index of the second data model; Determine, in the fusion algorithm, a first fusion weight of the first classification result and a second fusion weight of the second classification result according to the first evaluation index and the second evaluation index; For each defect mode, a fusion calculation is performed based on the first classification result and the first fusion weight, and the second classification result and the second fusion weight to obtain a fusion result for the current defect mode; Wherein, determining, in the fusion algorithm, a first fusion weight of the first classification result and a second fusion weight of the second classification result according to the first evaluation index and the second evaluation index includes: The calculation is performed using the following formula: A_weight=A_accuracy / (A_accuracy+B_accuracy); B_weight=B_accuracy / (A_accuracy+B_accuracy); Among them, A_weight is the first fusion weight, B_weight is the second fusion weight, A_accuracy is the first evaluation index, and B_accuracy is the second evaluation index; Wherein, based on the first classification result and the first fusion weight, and the second classification result and the second fusion weight, a fusion calculation is performed to obtain a fusion result for the current defect mode, including: The calculation is performed using the following formula: P i =a i *A_weight+b i *B_weight; Among them, P i is the fusion result of the current defect mode, a i is the first classification probability value of the current defect mode predicted by the first data model, b i is the second classification probability value of the current defect mode predicted by the second data model, A_weight is the first fusion weight, and B_weight is the second fusion weight; Wherein, determining the defect mode of the wafer to be identified according to the fusion result includes: Identify whether the fusion result for the current defect mode in the fusion result is greater than a set threshold; If yes, it is determined that the defect mode of the wafer to be identified belongs to the current defect mode; If not, it is determined that the defect mode of the wafer to be identified does not belong to the current defect mode.

2. The method according to claim 1, characterized in that The training process of the first data model includes: Obtain historical image data of wafer samples; Performing enhancement processing on the historical image data to obtain image sample data; The sample data is input into an initial model for model training to obtain a first data model.

3. The method according to claim 1, characterized in that The information data includes defect information data files; Accordingly, the training process of the second data model includes: Obtain defect information data files of wafer samples; Structuring the defect information data document to obtain structured data; wherein the structured data includes derived features; Performing feature engineering processing on the structured data to obtain training sample data; The sample data is input into the initial model for model training to obtain a second data model.

4. A device for determining a wafer defect mode, characterized in that: The device comprises: A data acquisition module, used to acquire image data and information data of the wafer to be identified; wherein the information data is surface defect information of the wafer to be identified; A classification result output module is used to input the image data into a pre-trained first data model to obtain a first classification result of the first data model for defect mode prediction; and input the information data into a pre-trained second data model to obtain a second classification result of the second data model for defect mode prediction; A fusion calculation module, used for performing fusion calculation on the first classification result and the second classification result based on a predetermined fusion algorithm to obtain a fusion result; A defect mode determination module, used to determine the defect mode of the wafer to be identified according to the fusion result; The first classification result and the second classification result are fused based on a predetermined fusion algorithm to obtain a fusion result, including: Obtaining a first evaluation index of the first data model and a second evaluation index of the second data model; Determine, in the fusion algorithm, a first fusion weight of the first classification result and a second fusion weight of the second classification result according to the first evaluation index and the second evaluation index; For each defect mode, a fusion calculation is performed based on the first classification result and the first fusion weight, and the second classification result and the second fusion weight to obtain a fusion result for the current defect mode; Wherein, determining, in the fusion algorithm, a first fusion weight of the first classification result and a second fusion weight of the second classification result according to the first evaluation index and the second evaluation index includes: The calculation is performed using the following formula: A_weight=A_accuracy / (A_accuracy+B_accuracy); B_weight=B_accuracy / (A_accuracy+B_accuracy); Among them, A_weight is the first fusion weight, B_weight is the second fusion weight, A_accuracy is the first evaluation index, and B_accuracy is the second evaluation index; Wherein, based on the first classification result and the first fusion weight, and the second classification result and the second fusion weight, a fusion calculation is performed to obtain a fusion result for the current defect mode, including: The calculation is performed using the following formula: P i =a i *A_weight+b i *B_weight; Among them, P i is the fusion result of the current defect mode, a i is the first classification probability value of the current defect mode predicted by the first data model, b i is the second classification probability value of the current defect mode predicted by the second data model, A_weight is the first fusion weight, and B_weight is the second fusion weight; Wherein, determining the defect mode of the wafer to be identified according to the fusion result includes: Identify whether the fusion result for the current defect mode in the fusion result is greater than a set threshold; If yes, it is determined that the defect mode of the wafer to be identified belongs to the current defect mode; If not, it is determined that the defect mode of the wafer to be identified does not belong to the current defect mode.

5. A computer device, characterized in that: The device comprises: one or more processors; and A memory storing computer program instructions, wherein when the computer program instructions are executed, the processor executes the method for determining a wafer defect mode according to any one of claims 1 to 3.

6. A computer-readable medium, characterized in that Computer program instructions are stored thereon, and the computer program instructions can be executed by a processor to implement the method for determining a wafer defect mode as described in any one of claims 1-3.