Wafer defect area image quality evaluation method, device, equipment and storage medium

Abstract

The invention discloses a wafer defect area image quality evaluation method and device, equipment and a storage medium. The method comprises the steps of obtaining a target wafer image; segmenting the target wafer image to obtain a plurality of to-be-identified images, inputting the plurality of to-be-identified images into the current feature extraction model for feature extraction to obtain a plurality of target feature maps with target marks, and inputting the plurality of target feature maps into the current prediction model for evaluation to obtain a plurality of prediction quality indexes; determining an overall quality index of the target wafer image based on the plurality of predicted quality indexes; acquiring image quality evaluation parameters; and determining a quality evaluation result of the target wafer image based on the overall quality index and the image quality evaluation parameter. According to the scheme, the defects can be positioned to the segmented image blocks, image quality evaluation is independently performed on each image block, and compared with overall quality evaluation, subtle changes and defect characteristics can be better captured, and the accuracy of image quality evaluation of the wafer defect area is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wafer quality inspection, and particularly relates to a method, device, equipment and storage medium for evaluating the image quality of defective areas of wafers. Background Art

[0002] A wafer is a circular silicon wafer and an important component structure in the process of manufacturing semiconductor integrated circuits. Its quality directly affects the performance of semiconductor integrated circuits. To control product quality, wafer defect detection is required. By collecting wafer images, defects can be detected based on the wafer images, the defect distribution and causes of the wafers can be determined, providing a basis for improving the production process. Traditional wafer image defect detection is realized manually, which is time-consuming and laborious, the detection effect is unstable, and it has high requirements for the professional knowledge and experience of the detection personnel, and is not applicable to large-scale wafer image defect detection. Applying deep learning technology to realize wafer defect detection based on wafer images can improve the detection efficiency and effect. However, at present, it is mostly limited to the detection of the overall image.

[0003] However, in the wafer image, the image quality of its abnormal area is not completely related to the image quality shown by the overall image, and the accuracy of the overall image quality evaluation is relatively low. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a method, device, equipment and storage medium for evaluating the image quality of defective areas of wafers. To solve the problem of low accuracy of wafer image quality evaluation in the prior art.

[0005] According to one aspect of the present application, a method for evaluating the image quality of defective areas of wafers is disclosed. The method includes:

[0006] Obtain a target wafer image;

[0007] Perform a segmentation process on the target wafer image to obtain a plurality of images to be recognized;

[0008] Input the plurality of images to be recognized into a feature extraction model for feature extraction to obtain a target feature map corresponding to each of the plurality of images to be recognized, and the target defective area in the target feature map has a target mark;

[0009] Input the plurality of target feature maps into the current prediction model for evaluation to obtain a prediction quality index corresponding to each of the plurality of target feature maps, and the current prediction model is formed by optimizing and iterating the previous prediction model with the current loss gradient;

[0010] Based on the plurality of prediction quality indexes, determine the overall quality index of the target wafer image;

[0011] Obtain image quality evaluation parameters;

[0012] Based on the overall quality index and the image quality evaluation parameters, determine the quality evaluation result of the target wafer image.

[0013] In some embodiments, after inputting the multiple images to be recognized into a feature extraction model for feature extraction to obtain target feature maps corresponding to the multiple images to be recognized respectively, the method further includes:

[0014] Input the target feature maps corresponding to the multiple images to be recognized respectively into a Softmax activation function for processing to obtain defect class prediction probabilities corresponding to the multiple target feature maps respectively.

[0015] In some embodiments, the method further includes:

[0016] Determine set quality indexes corresponding to the multiple images to be recognized respectively;

[0017] Based on a target loss function, the multiple set quality indexes, and the multiple predicted quality indexes, determine a target loss value;

[0018] Through a backpropagation algorithm, transmit the target loss value back to the previous prediction model to iteratively optimize the previous prediction model to obtain the current prediction model;

[0019] The target loss function is:

[0020]

[0021] Where:

[0022] Loss is the target loss function;

[0023] N is the total number of samples, M is the total number of defect classes, y ij is an indicator variable of sample i with respect to defect class j. If sample i belongs to defect class j, then y i j is 1, otherwise y i j is 0, p i j is the probability that the model predicts that sample i belongs to defect class j.

[0024] In some embodiments, the inputting the multiple images to be recognized into a feature extraction model for feature extraction to obtain target feature maps corresponding to the multiple images to be recognized respectively includes:

[0025] Perform image processing on each image to be recognized to obtain multiple transformed images with the same target pixel size;

[0026] Perform data partitioning on the multiple transformed images to obtain multiple transformed training images;

[0027] Input multiple transformed training images into the feature extraction model for feature extraction to obtain the target feature map of one-dimensional vectors.

[0028] In some embodiments, inputting multiple transformed training images into the feature extraction model for feature extraction to obtain the target feature map of one-dimensional vectors includes:

[0029] Input multiple transformed training images into a CNN convolutional neural network for feature extraction. The CNN convolutional neural network includes four convolutional layers, one pooling layer, and one dimensionality transformation layer. The transformed training images pass through the four convolutional layers and one pooling layer in sequence to form an intermediate feature map of three-dimensional vectors, and the intermediate feature map of three-dimensional vectors is reduced to a target feature map of one-dimensional vectors through the dimensionality transformation layer.

[0030] In some embodiments, determining the overall quality metric of the target wafer image based on multiple prediction quality metrics includes:

[0031] Determine the weight corresponding to each target feature map, where the weight corresponding to each target feature map is determined based on the target marking situation on each target feature map;

[0032] Based on the prediction quality metric of each target feature map and the weight corresponding to each target feature map, perform weighted summation to determine the overall quality metric of the target wafer image.

[0033] In some embodiments, the method further includes:

[0034] Obtain an inspection target image, where the inspection target image has a target marking of a target defect area;

[0035] Determine a target anomaly detection algorithm;

[0036] Based on the target anomaly detection algorithm, perform defect area detection on each inspection target image to obtain a defect prediction area;

[0037] Extract the true defect area of each target image to obtain a defect true area;

[0038] Determine the intersection over union between the defect prediction area and the defect true area of each target image, where the intersection over union represents the ratio of the intersection area to the union area between the defect prediction area and the defect true area;

[0039] Determine the accuracy Acc of each target image, where the accuracy is used to represent the ratio of the number of correctly classified samples to the total number of samples;

[0040] Based on the intersection over union and the accuracy Acc, assign a quality metric to each wafer image to generate the assigned quality metric for each of the target images.

[0041] Obtain the preset quality assessment metrics.

[0042] Based on the preset quality assessment metrics and the assigned quality metric, determine the target training images that meet the quality requirements.

[0043] Construct an image quality assessment dataset based on the target training images, where the image quality assessment dataset is used to optimize the training set of the feature extraction model.

[0044] According to another aspect of the present application, there is also disclosed a wafer defect area image quality assessment device, characterized in that the device includes:

[0045] A target wafer image acquisition module for acquiring a target wafer image;

[0046] A to-be-recognized image determination module for performing a segmentation process on the target wafer image to obtain a plurality of to-be-recognized images;

[0047] A target feature map determination module for inputting the plurality of to-be-recognized images into a feature extraction model for feature extraction to obtain a target feature map corresponding to each of the plurality of to-be-recognized images, where the target defect area in the target feature map has a target mark;

[0048] A predicted quality metric determination module for inputting the plurality of target feature maps into the current prediction model for evaluation to obtain a predicted quality metric corresponding to each of the plurality of target feature maps, where the current prediction model is formed by optimizing and iterating the previous prediction model with the current loss gradient;

[0049] An overall quality metric determination module for determining the overall quality metric of the target wafer image based on the plurality of predicted quality metrics;

[0050] A quality assessment parameter acquisition module for acquiring image quality assessment parameters;

[0051] A quality assessment result determination module for determining the quality assessment result of the target wafer image based on the overall quality metric and the image quality assessment parameters.

[0052] According to another aspect of the present application, there is also disclosed an electronic device, the electronic device includes a memory and at least one processor, and instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the electronic device executes each step of the wafer defect area image quality assessment method as described in any one of the above.

[0053] According to another aspect of the present application, a computer-readable storage medium is also disclosed. Instructions are stored on the computer-readable storage medium, and characterized in that when the instructions are executed by a processor, each step of the method for evaluating the image quality of a wafer defect area as described in any one of the above is implemented.

[0054] The present invention includes but is not limited to the following beneficial effects: (1) This solution can locate defects to the segmented image blocks, and perform image quality evaluation on each image block separately. Compared with the overall quality evaluation, it can better capture subtle changes and defect features, improving the accuracy of image quality evaluation; (2) This solution can optimize and iterate the previous prediction model through the current loss gradient, enabling dynamic optimization of the model, so that the model can continuously improve its prediction performance according to new data; (3) This solution can dynamically adjust model parameters by optimizing the loss value, making it better adapt to new data and requirements, thereby improving the performance of the prediction model; (4) This solution can score the authenticity of the defect area, and construct an image quality evaluation data set based on the target training image obtained after scoring, thereby realizing the optimization of the feature extraction model and improving the extraction accuracy of the feature extraction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for description in the embodiments or the prior art.

[0056] Figure 1 is a flowchart of a method for evaluating the image quality of a wafer defect area according to an embodiment of the present application;

[0057] Figure 2 is another flowchart of a method for evaluating the image quality of a wafer defect area according to an embodiment of the present application;

[0058] Figure 3 is another flowchart of a method for evaluating the image quality of a wafer defect area according to an embodiment of the present application;

[0059] Figure 4 is another flowchart of a method for evaluating the image quality of a wafer defect area according to an embodiment of the present application;

[0060] Figure 5 is another flowchart of a method for evaluating the image quality of a wafer defect area according to an embodiment of the present application;

[0061] Figure 6 is a structural block diagram of a device for evaluating the image quality of a wafer defect area according to an embodiment of the present application;

[0062] Figure 7 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention;

[0063] Figure 8 It is a schematic diagram of defects in the image of the defective area of the wafer in an embodiment of the present invention;

[0064] Figure 9 It is a schematic diagram of the target feature map of the image of the defective area of the wafer in an embodiment of the present invention;

[0065] Figure 10 It is a schematic diagram of the principle of quality assessment of the image of the defective area of the wafer in an embodiment of the present invention. Detailed implementation manners

[0066] An embodiment of the present invention provides a method for evaluating the quality of an image of a defective area of a wafer. The method includes: obtaining a target wafer image; performing a segmentation process on the target wafer image to obtain a plurality of images to be recognized; inputting the plurality of images to be recognized into a feature extraction model for feature extraction to obtain a target feature map corresponding to each of the plurality of images to be recognized, where the target defective area in the target feature map has a target mark; inputting the plurality of target feature maps into a current prediction model for evaluation to obtain a prediction quality index corresponding to each of the plurality of target feature maps, where the current prediction model is formed by optimizing and iterating the previous prediction model with the current loss gradient; determining an overall quality index of the target wafer image based on the plurality of prediction quality indexes; obtaining image quality evaluation parameters; and determining a quality evaluation result of the target wafer image based on the overall quality index and the image quality evaluation parameters. This solution can locate defects to the segmented image blocks, and perform image quality evaluation on each image block separately. Compared with the overall quality evaluation, it can better capture subtle changes and defect features, and improve the accuracy of image quality evaluation.

[0067] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order different from that shown or described herein. In addition, the terms "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0068] For ease of understanding, the specific process of an embodiment of the present invention will be described below. Specifically, Figure 1 It is a flowchart of a method for evaluating the quality of an image of a defective area of a wafer, as Figure 1 shown, and includes the following steps:

[0069] S100. Obtain a target wafer image.

[0070] Specifically, in some examples, a wafer image can be obtained based on a camera equipped with high resolution. A wafer image can also be obtained based on computed tomography (CT) or X-ray imaging: by converting the three-dimensional image inside the obtained wafer into a two-dimensional image, the converted two-dimensional image is used as the target wafer image. Further, the target wafer image can also be directly obtained from an image database, where the image database pre-stores multiple wafer images.

[0071] S102. Perform a segmentation process on the target wafer image to obtain multiple images to be recognized.

[0072] In one example, the target wafer image can be input into a data processing model for segmentation processing to obtain multiple images to be recognized.

[0073] S104. Input the multiple images to be recognized into a feature extraction model for feature extraction to obtain the target feature maps corresponding to the multiple images to be recognized respectively.

[0074] In one example, Figure 2 is another flowchart of the wafer defect area image quality evaluation method. This flowchart gives an exemplary illustration of step S104. Refer to Figure 2 , including the following steps:

[0075] S200. Perform image processing on each image to be recognized to obtain multiple converted images with the same target pixel size.

[0076] Specifically, in this step, pixel processing can be performed on each image to be recognized to process it into an image of 2048×2048 pixels. Further, each image to be recognized is cropped to obtain an image block with a size of 512*512. Further, data enhancement is performed on the multiple images to be recognized respectively through image processing operations such as brightness distortion, image rotation and translation, so as to obtain multiple converted images.

[0077] S202. Perform data partitioning on the multiple converted images to obtain multiple converted training images.

[0078] Exemplarily, the multiple converted images can be partitioned in a ratio of 8:2, that is, 80% is used as the converted training images.

[0079] S204. Input the multiple converted training images into a feature extraction model for feature extraction to obtain the target feature maps of one-dimensional vectors.

[0080] Specifically, the feature extraction model can be a CNN convolutional neural network, that is, multiple transformed training images are input into the CNN convolutional neural network for feature extraction. Among them, as Figure 10 shown, the CNN convolutional neural network includes four convolutional layers, one pooling layer, and one dimensionality transformation layer. The transformed training images pass through the four convolutional layers and one pooling layer in sequence to form an intermediate feature map of three-dimensional vectors. The intermediate feature map of three-dimensional vectors is reduced to a target feature map of one-dimensional vectors after passing through the dimensionality transformation layer. Exemplarily, the target feature map is as Figure 9 shown. It can be understood that during the feature extraction process, relevant information of the defect regions extracted more will be marked, such as marking information such as the border and center point of the defect region, that is, the target defect region in the target feature map has a target mark, and the target feature map marked with the target mark represents the target feature map with the target defect region.

[0081] Furthermore, in this example, the feature extraction model can be optimized based on the Figure 3 steps shown. Refer to Figure 3 , including the following steps:

[0082] S300. Obtain a test target image, and the test target image has a target mark of the target defect region.

[0083] Specifically, obtain the image with the target mark from multiple target feature images as the test target image.

[0084] S302. Determine the target anomaly detection algorithm.

[0085] In one example, the anomaly detection algorithm can be YOLO or Faster R-CNN, or Mask R-CNN, etc.

[0086] S304. Detect the defect region of each test target image based on the target anomaly detection algorithm to obtain a defect prediction region.

[0087] Specifically, directly input the test target image into the anomaly detection algorithm model to obtain the test target image with a defect prediction region identifier predicted based on the anomaly detection algorithm.

[0088] S306. Extract the true defect region of each target image to obtain a defect true region.

[0089] Specifically, extract the true defect region information from the target mark region of the test target image to obtain the defect true region. The defect true region can be a bounding box or key points.

[0090] S308. Determine the intersection over union between the defect prediction region and the defect true region of each target image.

[0091] Specifically, the intersection over union (IoU) represents the ratio of the intersection area to the union area between the defect prediction region and the actual defect region. Based on the value of the IoU, a threshold can be set (e.g., 0.5). If the value of the IoU is greater than this threshold, it is considered that the defect prediction region determined by the target anomaly detection algorithm is accurate; otherwise, the prediction result is considered inaccurate.

[0092] S310. Determine the accuracy Acc of each target image.

[0093] Among them, the accuracy is used to represent the ratio of the number of correctly classified samples to the total number of samples.

[0094] S312. Assign a quality index to each wafer image based on the IoU and the accuracy Acc, and generate the assigned quality index of each target image.

[0095] Specifically, based on the correct IoU result and the accuracy Acc value, a scoring standard can be determined. Generally, a higher Acc value corresponds to a higher quality score. Further, based on this scoring standard, the assigned quality index of each target image is determined. Exemplarily, the quality score corresponding to the assigned quality index is greater than this scoring standard.

[0096] S314. Obtain the preset quality assessment index.

[0097] This preset quality assessment index can be a preset quality score.

[0098] S316. Determine the target training images that meet the quality requirements based on the preset quality assessment index and the assigned quality index.

[0099] Specifically, the assigned quality index and the preset quality assessment index can be compared. When the assigned quality index is greater than the preset quality assessment index, it is considered that this target image is the target training image that meets the quality requirements.

[0100] S318. Construct an image quality assessment dataset based on the target training images. The image quality assessment dataset is used to optimize the training set of the feature extraction model.

[0101] It can be understood that this solution can score the authenticity of the defect region, and construct an image quality assessment dataset based on the target training images obtained after scoring, so as to realize the optimization of the feature extraction model and improve the extraction accuracy of the feature extraction model.

[0102] S106. Input multiple target feature maps into the current prediction model for evaluation to obtain the corresponding prediction quality indexes of the multiple target feature maps.

[0103] In one example, the prediction model can be a no-reference image quality assessment model NR-IQA. After inputting multiple target feature maps into the current prediction model, the current prediction model will output the corresponding prediction quality metrics for each target feature map, and the prediction quality metric can refer to a prediction score. It can be understood that the current prediction model is formed after optimizing and iterating the previous prediction model with the current loss gradient.

[0104] Specifically, as Figure 4 shown, it is another flowchart of the wafer defect area image quality assessment method according to the embodiment of the present application, and this flowchart is used to introduce the optimization of the current prediction model. Specifically, referring to Figure 4 , it includes the following steps:

[0105] S400. Determine the set quality metrics corresponding to each of the multiple images to be recognized.

[0106] Specifically, the set quality metric can be a pre-set quality score, and each image to be recognized corresponds to a set quality score. This set value can be set as a natural value based on actual needs, and no specific limitation is made here.

[0107] S402. Determine the target loss value based on the target loss function, multiple set quality metrics, and multiple prediction quality metrics.

[0108] Among them, the target loss function is:

[0109]

[0110] Among them:

[0111] Loss is the target loss function;

[0112] N is the total number of samples, M is the total number of defect categories, y ij is the indicator variable of sample i with respect to defect category j. If sample i belongs to defect category j, then y ij is 1, otherwise y ij is 0, and p ij is the probability that the model predicts that sample i belongs to defect category j.

[0113] In this example, the total number of samples N can be only the total number of images that require feature extraction. In this example, it can refer to 80% of the converted training images. The total number of defect categories M is the total number of defect categories including point defects, line defects, surface defects, etc. The probability p ij of defect category j can be the prediction probability output after processing the target feature map corresponding to the image to be recognized through the Softmax activation function.

[0114] S404. Using the backpropagation algorithm, pass the target loss value back to the previous prediction model to iteratively optimize the previous prediction model and obtain the current prediction model.

[0115] S108. Based on multiple prediction quality metrics, determine the overall quality metric of the target wafer image.

[0116] Specifically, Figure 5 is another flowchart of the wafer defect area image quality evaluation method. This process is an exemplary illustration of step S108 for determining the overall quality metric of the target wafer image based on multiple prediction quality metrics. Refer to Figure 5 , including the following steps:

[0117] S500. Determine the weight corresponding to each target feature map.

[0118] Among them, the weight corresponding to each target feature map is determined based on the target marking situation on each target feature map. It can be understood that a relatively large weight can be given to the target feature map with target markings, and a relatively small weight can be given to the target feature map without target markings, that is, the target feature map without defect areas. The specific weight value can be set based on the actual situation, and the total weight of multiple target feature maps is 1.

[0119] S502. Based on the prediction quality metric of each target feature map and the weight corresponding to each target feature map, perform weighted summation to determine the overall quality metric of the target wafer image.

[0120] S110. Obtain the image quality evaluation parameter.

[0121] Specifically, the image quality evaluation parameter can be preset. For example, it can be a value such as 0.5 or 0.6 or 0.7, etc.

[0122] S112. Based on the overall quality metric and the image quality evaluation parameter, determine the quality evaluation result of the target wafer image.

[0123] Specifically, the overall quality metric can be compared with the image quality evaluation parameter. If the quality score corresponding to the overall quality metric is greater than the quality score corresponding to the image quality parameter, it is considered that the quality of the target wafer image is qualified; otherwise, it is unqualified.

[0124] It can be understood that in the prior art, when evaluating the quality of multiple overall images, as Figure 8 shown, when performing overall quality evaluation, Figure 8 the noise in Figure 8 a is significantly more than Figure 8 the noise in Figure 8the quality of b, but ignores the impact of regional defects on the overall quality. Based on the method of this technical solution, the defects are located in the segmented image blocks, and the image quality of each image block is evaluated separately. At this time, as Figure 8 shown Figure 8 the defects in the elliptical area of b are significantly larger than Figure 8 the defects in the elliptical area of a. The degree of blurriness of this defect has a relatively serious impact on the overall image. Therefore, at this time, it is considered that Figure 8 the image quality of a is greater than Figure 8 the image quality of b. Furthermore, the individual comparison of the image blocks in this technical solution can better capture subtle changes and defect features compared to the overall quality assessment, improving the accuracy of image quality assessment; further, this solution can achieve dynamic optimization of the model through the optimization iteration of the previous prediction model by the current loss gradient, enabling the model to continuously improve its prediction performance according to new data; further, this solution can dynamically adjust the model parameters by optimizing the loss value, making it better adapt to new data and requirements, thereby improving the performance of the prediction model.

[0125] Furthermore Figure 6 is the structural block diagram of the image quality assessment device for the wafer defect area. As Figure 6 shown, the device includes:

[0126] A target wafer image acquisition module for acquiring a target wafer image;

[0127] A to-be-recognized image determination module for performing segmentation processing on the target wafer image to obtain multiple to-be-recognized images;

[0128] A target feature map determination module for inputting multiple to-be-recognized images into a feature extraction model for feature extraction to obtain target feature maps corresponding to the multiple to-be-recognized images respectively, and the target defect area in the target feature map has a target mark;

[0129] A predicted quality index determination module for inputting multiple target feature maps into the current prediction model for evaluation to obtain predicted quality indexes corresponding to the multiple target feature maps respectively, and the current prediction model is formed by optimizing and iterating the previous prediction model with the current loss gradient;

[0130] An overall quality index determination module for determining the overall quality index of the target wafer image based on multiple predicted quality indexes;

[0131] A quality assessment parameter acquisition module for acquiring image quality assessment parameters;

[0132] A quality assessment result determination module for determining the quality assessment result of the target wafer image based on the overall quality index and the image quality assessment parameters.

[0133] For the application introduction of the relevant modules of the device in this example, reference can be made to the relevant introduction of the above method principle, which will not be elaborated here.

[0134] According to another aspect of the present application, the present application also discloses an electronic device, which includes a memory and at least one processor. Instructions are stored in the memory; the at least one processor invokes the instructions in the memory to enable the electronic device to execute each step of the above wafer defect area image quality evaluation method.

[0135] The above figures describe in detail the wafer defect area image quality evaluation device in the embodiments of the present invention from the perspective of modular functional entities. The following will describe in detail the electronic device in the embodiments of the present invention from the perspective of hardware processing.

[0136] Figure 7 FIG. is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device 700 may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPUs) 710 (for example, one or more processors) and a memory 720, and one or more storage media 730 for storing application programs 733 or data 732 (for example, one or more mass storage devices). Among them, the memory 720 and the storage media 730 may be transient storage or persistent storage. The program stored in the storage media 730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the electronic device 700. Further, the processor 710 may be configured to communicate with the storage media 730 and execute a series of instruction operations in the storage media 730 on the electronic device 700.

[0137] The electronic device 700 may further include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input / output interfaces 760, and / or one or more operating systems 731, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, and so on. Those skilled in the art can understand that Figure 7 The shown structure of the electronic device does not constitute a limitation on the electronic device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0138] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are run on a computer, the computer is caused to execute the steps of the method for evaluating the image quality of a defective area on a wafer.

[0139] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, device, or unit can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0140] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0141] The above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A wafer defect area image quality assessment method, characterized in that: The method comprises: Acquire a target wafer image; Slicing the target wafer image to obtain a plurality of images to be identified; Inputting the plurality of images to be identified into a feature extraction model for feature extraction, so as to obtain target feature maps corresponding to the plurality of images to be identified, wherein the target defect area in the target feature map has a target mark; Inputting the plurality of target feature maps into a current prediction model for evaluation to obtain prediction quality indicators corresponding to the plurality of target feature maps, wherein the current prediction model is formed after optimizing and iterating the previous prediction model with the current loss gradient; Determining an overall quality indicator of the target wafer image based on the plurality of predicted quality indicators; Obtain image quality assessment parameters; Based on the overall quality index and the image quality assessment parameter, a quality assessment result of the target wafer image is determined.

2. The wafer defect area image quality assessment method according to claim 1, characterized in that: After inputting the plurality of images to be identified into the feature extraction model for feature extraction to obtain target feature maps corresponding to the plurality of images to be identified, the method further includes: The target feature maps corresponding to each of the multiple images to be identified are input into the Softmax activation function for processing to obtain the defect category prediction probabilities corresponding to each of the multiple target feature maps.

3. The wafer defect area image quality assessment method according to claim 2, characterized in that: The method further comprises: Determining a set quality index corresponding to each of the plurality of images to be identified; Determining a target loss value based on a target loss function, a plurality of set quality indicators, and a plurality of predicted quality indicators; The target loss value is transmitted back to the previous prediction model through a back propagation algorithm to iteratively optimize the previous prediction model to obtain the current prediction model; The objective loss function is: in: Loss is the target loss function; N is the total number of samples, M is the total number of defect categories, and y ij is the indicator variable for sample i relative to defect category j. If sample i belongs to defect category j, then y ij is 1, otherwise y ij =0,p ij is the probability that the model predicts that sample i belongs to defect category j.

4. The wafer defect area image quality assessment method according to claim 1, characterized in that: The step of inputting the plurality of images to be identified into a feature extraction model for feature extraction to obtain target feature maps corresponding to the plurality of images to be identified comprises: Performing image processing on each of the to-be-recognized images to obtain a plurality of transformed images having the same target pixel size; Performing data division on the plurality of the transformed images to obtain a plurality of transformed training images; A plurality of transformed training images are input into the feature extraction model for feature extraction to obtain the target feature map of a one-dimensional vector.

5. The wafer defect area image quality assessment method according to claim 1, characterized in that: Inputting a plurality of transformed training images into the feature extraction model for feature extraction to obtain the target feature map of a one-dimensional vector includes: Multiple converted training images are input into a CNN convolutional neural network for feature extraction. The CNN convolutional neural network includes four convolutional layers, one pooling layer and one dimensional conversion layer. The converted training images pass through the four convolutional layers and one pooling layer in sequence to form an intermediate feature map of a three-dimensional vector. The intermediate feature map of the three-dimensional vector is reduced to a target feature map of a one-dimensional vector after passing through the dimensional conversion layer.

6. The wafer defect area image quality assessment method according to claim 1, characterized in that: Determining the overall quality index of the target wafer image based on the plurality of predicted quality indexes includes: Determine a weight corresponding to each target feature map, wherein the weight corresponding to each target feature map is determined based on a target marking condition on each target feature map; Based on the predicted quality index of each target feature map and the weight corresponding to each target feature map, a weighted sum is performed to determine the overall quality index of the target wafer image.

7. The wafer defect area image quality assessment method according to claim 1, characterized in that: The method further comprises: Acquire an inspection target image, wherein the inspection target image has a target mark of a target defect area; Determine the target anomaly detection algorithm; Perform defect area detection on each of the inspection target images based on the target anomaly detection algorithm to obtain a defect prediction area; Extracting the real defect area of ​​each target image to obtain the real defect area; Determine an intersection-and-union ratio between a defect prediction region and a defect real region of each of the target images, wherein the intersection-and-union ratio represents a ratio of an intersection area to a union area between the defect prediction region and the defect real region; Determine the accuracy Acc of each target image, wherein the accuracy is used to represent the ratio of the number of correctly classified samples to the total number of samples; Based on the intersection-over-union ratio and the accuracy Acc, a quality index is assigned to each wafer image to generate an assigned quality index for each target image; Obtain preset indicators for quality assessment; Based on the preset quality assessment index and the assigned quality index, determining a target training image that meets the quality requirements; An image quality assessment dataset is constructed based on the target training image, and the image quality assessment dataset is used to optimize the training set of the feature extraction model.

8. A wafer defect area image quality assessment device, characterized in that: The device comprises: A target wafer image acquisition module, used for acquiring a target wafer image; A module for determining an image to be identified, used for segmenting the target wafer image to obtain a plurality of images to be identified; A target feature map determination module is used to input the plurality of images to be identified into a feature extraction model for feature extraction, so as to obtain target feature maps corresponding to the plurality of images to be identified, wherein the target defect area in the target feature map has a target mark; A prediction quality index determination module, used for inputting the plurality of target feature graphs into a current prediction model for evaluation to obtain prediction quality indicators corresponding to the plurality of target feature graphs, wherein the current prediction model is formed after optimizing and iterating the previous prediction model with the current loss gradient; An overall quality index determination module, used to determine an overall quality index of the target wafer image based on the plurality of predicted quality indexes; A quality assessment parameter acquisition module is used to obtain image quality assessment parameters; The quality assessment result determination module is used to determine the quality assessment result of the target wafer image based on the overall quality index and the image quality assessment parameter.

9. An electronic device, characterized in that: The electronic device includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes each step of the wafer defect area image quality assessment method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the wafer defect area image quality assessment method as described in any one of claims 1 to 7 are implemented.

Images