Wafer Defect Detection Method and Device

Abstract

This specification provides a wafer defect detection method and apparatus. The wafer defect detection method includes: acquiring an original image of a target wafer and preprocessing it to obtain a test image; generating a target anomaly probability based on the test image and a trained anomaly recognition model; determining that the target wafer has a defect when the target anomaly probability is greater than a preset first threshold; importing the test image into a preset classification model to determine the defect classification of the target wafer when the target wafer has a defect; and determining the location information of the known defect when the anomaly of the target wafer is a known defect type, which can improve the detection accuracy and detection effect of wafer defects.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of semiconductor manufacturing quality control technology, and in particular to wafer defect detection methods and apparatus. Background Technology

[0002] Advanced integrated circuit manufacturing processes typically involve hundreds of steps. Even a tiny error in any step can lead to the failure of the entire chip. In particular, as the critical dimensions of circuits continue to shrink, the requirements for process control become increasingly stringent. Therefore, in order to detect and resolve problems in a timely manner during actual production, it is necessary to equip products with highly sensitive optical defect detection equipment for online inspection.

[0003] In the existing technology, the methods for inspecting actual wafers have low accuracy and cannot meet the growing demand. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a wafer defect detection method. One or more embodiments of this specification also relate to a wafer defect detection apparatus, a computing device, a computer-readable storage medium, and a computer program, to address technical deficiencies in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a wafer defect detection method is provided, comprising:

[0006] The original image of the target wafer is acquired and preprocessed to obtain the image to be tested;

[0007] Based on the image to be tested and the trained anomaly recognition model, the probability of the target anomaly is generated.

[0008] When the probability of an anomaly of the target exceeds a preset first threshold, it is determined that the target wafer has a defect. The first threshold is obtained based on a preset first update strategy.

[0009] If the target wafer is found to have defects, the image to be tested is imported into a preset classification model to determine the defect classification of the target wafer. The defect classification includes known defect types and unknown defect types.

[0010] When the anomaly of the target wafer is of a known defect type, determine the location information of the known defect.

[0011] In some embodiments, when the anomaly of the target wafer is a known defect type, determining the location information of the known defect includes:

[0012] When the anomaly of the target wafer is a known defect type, the corresponding localization model is determined based on the known defect type;

[0013] Import the image to be tested into the localization model to determine the location information.

[0014] In some embodiments, the first update strategy includes:

[0015] When training an anomaly recognition model based on a preset training set, the error rate of each training session is calculated. The training set includes multiple training wafer images with known results.

[0016] After processing the first number of training wafer images, the average of the first number of error rates is calculated to obtain the first current error rate;

[0017] The first threshold is updated based on a preset first calculation formula, wherein the first calculation formula includes:

[0018]

[0019] in, This represents the updated first threshold. This represents the first threshold before the update. This represents the preset first calculation factor. F c1 Indicates the current error rate. F t1 This represents the preset first target error rate.

[0020] In some embodiments, the classification model includes an anomaly classification model and an unknown defect verification model; wherein, the step of importing the image to be tested into the preset classification model to determine the defect classification of the target wafer includes:

[0021] The image to be tested is imported into the anomaly classification model to generate anomaly classification confidence. The anomaly classification model is an SVM model.

[0022] If the confidence level of the anomaly classification is less than the preset second threshold, the image to be tested is imported into the unknown defect verification model to determine the defect classification of the target wafer. The second threshold is obtained based on the preset second update strategy, and the unknown defect verification model is an isolated forest model.

[0023] In some embodiments, the second update strategy includes:

[0024] When training an anomaly classification model based on a pre-set training set, the error rate of each training session is calculated. The training set includes multiple training wafer images with known results.

[0025] After processing a second number of training wafer images, the average of the second number of error rates is calculated to obtain the second current error rate;

[0026] The second threshold is updated based on a preset second calculation formula, wherein the second calculation formula includes:

[0027]

[0028] in, This represents the updated second threshold. This represents the second threshold before the update. This represents the preset second calculation factor. F c2 Indicates the second current error rate. F t2 This represents the preset second target error rate.

[0029] In some embodiments, the step of importing the image to be tested into a preset classification model to determine the defect classification of the target wafer further includes:

[0030] If the anomaly classification confidence level is not less than the latest second threshold, the defect classification of the target wafer is determined to be a known defect type.

[0031] In some embodiments, the anomaly detection model includes a discriminator, an encoder, a first prediction model, and a second prediction model, wherein generating the target anomaly probability based on the image to be tested and the trained anomaly detection model includes:

[0032] The image to be tested is reconstructed to obtain the reconstructed image;

[0033] Calculate the L1 norm of the image under test and the reconstructed image to obtain the reconstruction error characteristics;

[0034] Import the image to be tested into the discriminator and generate discriminator features;

[0035] The image to be tested is imported into the encoder to generate latent spatial features;

[0036] The error features, discriminator features, and latent space features are concatenated and reconstructed to obtain the first concatenation vector;

[0037] A standardized vector is generated based on the mean and standard deviation of each dimension in the first concatenated vector;

[0038] The standardized vector is imported into the first prediction model to obtain the first anomaly probability, where the first prediction model is an SVM model;

[0039] The standardized vector is concatenated with the first anomaly probability to obtain the second concatenated vector;

[0040] The second concatenated vector is imported into the second prediction model to generate the second anomaly probability, wherein the second prediction model is a random forest model;

[0041] The first and second anomaly probabilities are weighted and summed based on preset weights to obtain the target anomaly probability.

[0042] According to a second aspect of the embodiments of this specification, a wafer defect detection apparatus is provided, comprising:

[0043] The acquisition module is configured to acquire the original image of the target wafer and perform preprocessing to obtain the image to be tested;

[0044] The generation module is configured to generate the target anomaly probability based on the image to be tested and the trained anomaly recognition model.

[0045] The first determination module is configured to determine that the target wafer has a defect when the target anomaly probability is greater than a preset first threshold, wherein the first threshold is obtained based on a preset first update strategy.

[0046] The second determination module is configured to import the image to be tested into a preset classification model when it is determined that there are defects in the target wafer, and determine the defect classification of the target wafer, wherein the defect classification includes known defect types and unknown defect types.

[0047] The third determination module is configured to determine the location information of known defects when the anomaly of the target wafer is a known defect type.

[0048] In some embodiments, when the anomaly of the target wafer is a known defect type, determining the location information of the known defect includes:

[0049] When the anomaly of the target wafer is a known defect type, the corresponding localization model is determined based on the known defect type;

[0050] Import the image to be tested into the localization model to determine the location information.

[0051] In some embodiments, the first update strategy includes:

[0052] When training an anomaly recognition model based on a preset training set, the error rate of each training session is calculated. The training set includes multiple training wafer images with known results.

[0053] After processing the first number of training wafer images, the average of the first number of error rates is calculated to obtain the first current error rate;

[0054] The first threshold is updated based on a preset first calculation formula, wherein the first calculation formula includes:

[0055]

[0056] in, This represents the updated first threshold. This represents the first threshold before the update. This represents the preset first calculation factor. F c1Indicates the current error rate. F t1 This represents the preset first target error rate.

[0057] In some embodiments, the classification model includes an anomaly classification model and an unknown defect verification model; wherein, the step of importing the image to be tested into the preset classification model to determine the defect classification of the target wafer includes:

[0058] The image to be tested is imported into the anomaly classification model to generate anomaly classification confidence. The anomaly classification model is an SVM model.

[0059] If the confidence level of the anomaly classification is less than the preset second threshold, the image to be tested is imported into the unknown defect verification model to determine the defect classification of the target wafer. The second threshold is obtained based on the preset second update strategy, and the unknown defect verification model is an isolated forest model.

[0060] In some embodiments, the second update strategy includes:

[0061] When training an anomaly classification model based on a pre-set training set, the error rate of each training session is calculated. The training set includes multiple training wafer images with known results.

[0062] After processing a second number of training wafer images, the average of the second number of error rates is calculated to obtain the second current error rate;

[0063] The second threshold is updated based on a preset second calculation formula, wherein the second calculation formula includes:

[0064]

[0065] in, This represents the updated second threshold. This represents the second threshold before the update. This represents the preset second calculation factor. F c2 Indicates the second current error rate. F t2 This represents the preset second target error rate.

[0066] In some embodiments, the step of importing the image to be tested into a preset classification model to determine the defect classification of the target wafer further includes:

[0067] If the anomaly classification confidence level is not less than the latest second threshold, the defect classification of the target wafer is determined to be a known defect type.

[0068] In some embodiments, the anomaly detection model includes a discriminator, an encoder, a first prediction model, and a second prediction model, wherein generating the target anomaly probability based on the image to be tested and the trained anomaly detection model includes:

[0069] The image to be tested is reconstructed to obtain the reconstructed image;

[0070] Calculate the L1 norm of the image under test and the reconstructed image to obtain the reconstruction error characteristics;

[0071] Import the image to be tested into the discriminator and generate discriminator features;

[0072] The image to be tested is imported into the encoder to generate latent spatial features;

[0073] The error features, discriminator features, and latent space features are concatenated and reconstructed to obtain the first concatenation vector;

[0074] A standardized vector is generated based on the mean and standard deviation of each dimension in the first concatenated vector;

[0075] The standardized vector is imported into the first prediction model to obtain the first anomaly probability, where the first prediction model is an SVM model;

[0076] The standardized vector is concatenated with the first anomaly probability to obtain the second concatenated vector;

[0077] The second concatenated vector is imported into the second prediction model to generate the second anomaly probability, wherein the second prediction model is a random forest model;

[0078] The first and second anomaly probabilities are weighted and summed based on preset weights to obtain the target anomaly probability.

[0079] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising:

[0080] Memory and processor;

[0081] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the above-described wafer defect detection method.

[0082] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions that, when executed by a processor, implement the steps of the above-described wafer defect detection method.

[0083] According to a fifth aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the above-described wafer defect detection method.

[0084] At least one embodiment of the present specification obtains a test image by acquiring the original image of the target wafer and performing preprocessing; based on the test image and a trained anomaly recognition model, a target anomaly probability is generated; when the target anomaly probability is greater than a preset first threshold, it is determined that the target wafer has a defect; when it is determined that the target wafer has a defect, the test image is imported into a preset classification model to determine the defect classification of the target wafer; when the anomaly of the target wafer is a known defect type, the location information of the known defect is determined, which can improve the detection accuracy and detection effect of wafer defects. Attached Figure Description

[0085] Figure 1 This is a flowchart of some embodiments of a wafer defect detection method provided in this specification;

[0086] Figure 2 This is a flowchart of some embodiments of a wafer defect detection method provided in some embodiments of this specification;

[0087] Figure 3 This is a simplified structural diagram of a wafer defect detection device provided in some embodiments of this specification;

[0088] Figure 4 This is a structural block diagram of a computing device provided in some embodiments of this specification. Detailed Implementation

[0089] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0090] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the one or more embodiments of this specification. The singular forms “a” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. The modifications “a” and “a plurality” as used in this disclosure are illustrative and not restrictive, and those skilled in the art will understand that they should be understood as “one or more” unless the context clearly indicates otherwise.

[0091] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0092] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0093] SVM model: Support Vector Machine.

[0094] GAN model: Generative Adversarial Networks.

[0095] See Figure 1 , Figure 1 A flowchart of a wafer defect detection method according to some embodiments of this specification is shown, which specifically includes the following steps.

[0096] Step 101: Acquire the original image of the target wafer and perform preprocessing to obtain the image to be tested.

[0097] In some embodiments, the execution entity of the wafer defect detection method (such as a pre-defined computing device) can connect to the target device via a wired or wireless connection. Then, it acquires the original image of the target wafer and performs preprocessing to obtain the image to be tested. The original image can refer to image data or information obtained by photographing, scanning, or other methods targeting the wafer. For example, the image to be tested can be an image obtained by photographing, processing, scanning, or other processing methods using devices such as optical sensors, electron beam sensors (SEM), or X-ray sensors. The specific method is selected as needed and is not limited here. The image to be tested can refer to the collection of image data obtained after preprocessing the original image. As an example, preprocessing can include conventional processing methods such as tensor processing and noise reduction, which will not be elaborated further here.

[0098] It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future wireless connection methods.

[0099] Step 102: Based on the image to be tested and the trained anomaly recognition model, generate the target anomaly probability.

[0100] An anomaly detection model can refer to a computational model obtained by combining multiple computational formulas or a combination of computational models trained to identify wafer anomalies. The target anomaly probability can refer to the probability value calculated based on the anomaly detection model, which can be used to determine whether there is an anomaly in the image under test.

[0101] In some embodiments, the anomaly detection model includes a discriminator, an encoder, a first prediction model, and a second prediction model. During actual training, the anomaly detection model needs to train three models and an encoder. The three models include a GAN model, a first prediction model, and a second prediction model. After the GAN model is trained, its generator is used to generate reconstructed data, and its discriminator is used to extract feature vectors. The encoder can refer to a computational model used to generate latent spatial features. The first and second prediction models will be described in detail later. Based on this, the aforementioned execution entity can generate the target anomaly probability based on the image to be tested and the trained anomaly detection model using the following steps:

[0102] Step 1: Reconstruct the image to be tested to obtain the reconstructed image.

[0103] Reconstruction can refer to the image reconstructed by the generator of an anomaly detection model based on the image under test. The reconstructed image is the image obtained through the processing of this generator.

[0104] Step 2: Calculate the L1 norm of the image to be tested and the reconstructed image to obtain the reconstruction error features.

[0105] In image processing, the L1 norm tends to generate sparse solutions, which can automatically filter redundant information, making the matching process more focused on regions of significant difference, improving the representation of differences, thereby enhancing defect expression and increasing the accuracy of subsequent judgments.

[0106] Step 3: Import the image to be tested into the discriminator and generate discriminator features.

[0107] Discriminator features can refer to feature vectors generated by the discriminator, whose data dimension is higher than that of reconstruction error features.

[0108] Step 4: Import the image to be tested into the encoder to generate latent spatial features.

[0109] The encoder can extract features and reduce the dimensionality of the input image, mapping the image to a latent space of the target dimension, thus obtaining a latent space feature vector. Generally, the data dimension of the latent space features is larger than that of the reconstruction error features, but smaller than that of the discriminator features. As an example, the data dimension of the reconstruction error features can be 4, the data dimension of the discriminator features can be 256, and the latent space feature vector can be 128. These can be set as needed; no specific restrictions are imposed here.

[0110] Step 5: Concatenate and reconstruct the error features, discriminator features, and latent space features to obtain the first concatenated vector.

[0111] When performing feature concatenation, the dimension of the concatenated vector is the sum of the dimensions of each vector (feature) being concatenated. As an example, suppose the reconstruction error feature has a dimension of 50, the discriminator feature has a dimension of 200, and the latent space feature has a dimension of 100, then the concatenated vector will be 50 + 200 + 100 = 350 dimensions.

[0112] Step 6: Generate a standardized vector based on the mean and standard deviation of each dimension in the first concatenated vector.

[0113] In some optional implementations, a standardized vector is generated based on a preset third calculation formula and the mean and standard deviation of each dimension in the first concatenated vector, wherein the third calculation formula includes:

[0114]

[0115] in, Z i This represents the data in the i-th dimension of the standardized vector. v i This represents the data in the i-th dimension of the first concatenated vector. μ i This represents the dynamic weighted value (preset value) of the i-th dimension. σ i This represents the standard deviation of the i-th dimension. This represents the importance weight of the i-th dimension. γ t This represents the time decay factor (preset value) that dynamically adjusts with time t. β i This represents the direction coefficient of the data in the i-th dimension.

[0116] Specifically, importance weights can refer to the preset weight coefficients corresponding to the reconstruction error features, discriminator features, and latent space features, respectively. For example, when considering the following... When the feature of the corresponding dimension is the reconstruction error feature, This represents the preset weights corresponding to the reconstruction error features. When... When the features of the corresponding dimension are discriminator features, This represents the preset weights corresponding to the discriminator features. When When the features of the corresponding dimension are latent space features, This represents the preset weights corresponding to the features in the latent space.

[0117] Directional coefficient β i The methods for determining include:

[0118] when v i Greater than μ i When, it represents the direction coefficient. β i It can indeed correspond. β +.

[0119] when v i Greater than μ i When, it represents the direction coefficient. β i It can indeed correspond. β -

[0120] in, β The plus sign indicates a preset processing weight for positive offsets. β- This indicates the preset processing weight for negative offsets, which can be set as needed to perform specific processing for abnormal directions.

[0121] Step 7: Import the standardized vector into the first prediction model to obtain the first anomaly probability, where the first prediction model is an SVM model.

[0122] The first prediction model is a pre-trained model, and its training steps can be performed according to various existing training methods, without much restriction here. Since the SVM model can handle high-dimensional data and maintain high classification accuracy while preventing overfitting, it is a preferred choice. The first anomaly probability, i.e., the standardized vector, is the probability data output by this first prediction model.

[0123] Step 8: Concatenate the standardized vector with the first anomaly probability to obtain the second concatenated vector.

[0124] Step 9: Import the second concatenated vector into the second prediction model to generate the second anomaly probability, where the second prediction model is a random forest model.

[0125] The second prediction model is a pre-trained model, and its training steps can be performed according to various existing training methods, without much restriction here. The random forest model possesses the ability to model non-linear relationships and can automatically learn complex interaction relationships between features through the ensemble of multiple decision trees. It is suitable for multimodal data fusion scenarios. Furthermore, the random forest model's random sampling features and sample mechanism reduce its sensitivity to noise, and it can still stably output the second anomaly probability even when the data distribution is imbalanced. Therefore, the random forest model is the preferred choice here. The second anomaly probability is the probability data output by the second prediction model, which is then imported into the standardized vector.

[0126] Step 10: Based on preset weights, the first anomaly probability and the second anomaly probability are weighted and summed to obtain the target anomaly probability.

[0127] In some optional implementations, the weight of the first anomaly probability is 0.4 and the weight of the second anomaly probability is 0.6. Their specific values ​​can be set as needed, and no specific restrictions are imposed here.

[0128] Through the above steps, in-depth calculations and comparisons can be performed on image data across various dimensions. The first judgment is based on the SVM model, followed by a second judgment based on the random forest model. Finally, a weighted summation is used to calculate the anomaly probability, which can further enhance the accuracy of defect identification.

[0129] Step 103: When the target anomaly probability is greater than a preset first threshold, it is determined that the target wafer has a defect, wherein the first threshold is obtained based on a preset first update strategy.

[0130] In some embodiments, the presence of a defect in the target wafer can be determined by comparing a preset first threshold with the target anomaly probability calculated in step 102. For example, if the target anomaly probability is greater than the preset first threshold, the target wafer is determined to have a defect. If the target anomaly probability is not greater than the preset first threshold, the target wafer is determined to have no defect.

[0131] In some optional implementations, the first update strategy includes: calculating the error rate for each training iteration when training an anomaly recognition model based on a preset training set, wherein the training set includes multiple training wafer images with known results; after processing a first number of training wafer images, calculating the average of the first number of error rates to obtain a first current error rate; and updating a first threshold based on a preset first calculation formula, wherein the first calculation formula includes:

[0132]

[0133] in, This represents the updated first threshold. This represents the first threshold before the update. This represents the preset first calculation factor. F c1 Indicates the current error rate. F t1 This represents the preset first target error rate.

[0134] It can be seen that the update of the first threshold is related to the training process of the anomaly detection model; that is, the first threshold is updated once after the anomaly detection model has been trained a first number of times. As an example, this first number can be 100.

[0135] It should be noted that training anomaly detection models typically requires a very large number of training iterations (usually tens of thousands). Therefore, the value of the first number needs to consider both that the first threshold can be updated when the anomaly detection model is completed and that the computational power required to update the first threshold should be minimized. In practice, the first number is usually between 100 and 500.

[0136] Step 104: If it is determined that there are defects in the target wafer, the image to be tested is imported into the preset classification model to determine the defect classification of the target wafer. The defect classification includes known defect types and unknown defect types.

[0137] In some embodiments, once a defect is determined in the target wafer, its classification (i.e., known defect or unknown defect) needs to be determined. A known defect can refer to a defect whose cause, manifestation, or other relevant information has been determined. An unknown defect can refer to a defect whose cause, manifestation, or other relevant information cannot be determined at present. Obviously, for known defects, the location information can be quickly located based on the corresponding defect manifestation information, data structure, etc., while for unknown defects, this is not possible; they usually require secondary manual determination and in-depth analysis.

[0138] In some optional implementations, the classification model includes an anomaly classification model and an unknown defect verification model. The step of importing the image to be tested into a preset classification model to determine the defect classification of the target wafer includes: importing the image to be tested into the anomaly classification model to generate anomaly classification confidence, wherein the anomaly classification model is an SVM model; if the anomaly classification confidence is less than a preset second threshold, importing the image to be tested into the unknown defect verification model to determine the defect classification of the target wafer, wherein the second threshold is obtained based on a preset second update strategy, and the unknown defect verification model is an isolated forest model.

[0139] An anomaly classification model can refer to an SVM model that generates anomaly classification confidence scores based on input image information. Anomaly classification confidence score represents the probability of classifying anomalies in the input image information as either known or unknown defects. When the anomaly classification confidence score falls below a second threshold, the image to be tested is further imported into an unknown defect verification model trained with an isolated forest model to re-classify the defect in the target wafer. This secondary verification increases the accuracy of determining whether the image to be tested is a known defect type. The second threshold can refer to a preset value used for comparison with the anomaly classification confidence score.

[0140] In some optional implementations, the second update strategy includes: calculating the error rate for each training iteration when training an anomaly classification model based on a preset training set, wherein the training set includes multiple training wafer images with known results; after processing a second number of training wafer images, calculating the average of the second number of error rates to obtain a second current error rate; and updating a second threshold based on a preset second calculation formula, wherein the second calculation formula includes:

[0141]

[0142] in, This represents the updated second threshold. This represents the second threshold before the update. This represents the preset second calculation factor. F c2 Indicates the second current error rate. F t2 This represents the preset second target error rate.

[0143] Similar to the update strategy for the first threshold, the update of the second threshold is related to the training process of the anomaly classification model. That is, the second threshold is updated once after the anomaly classification model has been trained a second number of times. As an example, this first number can be 1000.

[0144] Similarly, training anomaly classification models typically requires a large number of training iterations (usually tens of thousands). Therefore, the value of the second threshold needs to consider both that the second threshold can be updated by the time the anomaly classification model is completed and that the computational power required to update the second threshold should be minimized. In practice, the second threshold is usually between 800 and 2000.

[0145] Step 105: If the anomaly of the target wafer is a known defect type, determine the location information of the known defects.

[0146] In some embodiments, when the anomaly of the target wafer is determined to be a known defect type, the location of the defect can be determined based on the defect type to facilitate subsequent defect processing.

[0147] In some optional implementations, when the anomaly of the target wafer is a known defect type, the location information of the known defect is determined, including: when the anomaly of the target wafer is a known defect type, determining the corresponding positioning model according to the defect type of the known defect; importing the image to be tested into the positioning model to determine the location information.

[0148] A localization model can refer to a model used to calculate location information for various specific types of defects. Different specific types of defects can correspond to different localization models. As an example, for defects with clear edges, an edge detection algorithm can be selected; for defects with a specific shape, a template matching algorithm can be selected. Location information can refer to the region, coordinates, boundary distance, or other locatable information of the defect in the image to be tested on the target wafer.

[0149] The beneficial effects of one of the embodiments in this specification include at least the following: generating a target anomaly probability based on the image to be tested and a trained anomaly recognition model; determining that the target wafer has a defect when the target anomaly probability is greater than a preset first threshold; importing the image to be tested into a preset classification model to determine the defect classification of the target wafer when the target wafer anomaly is determined to be a known defect type; and determining the location information of the known defect when the anomaly of the target wafer is a known defect type, which can improve the detection accuracy and detection effect of wafer defects.

[0150] The following is in conjunction with the appendix Figure 2 This section further explains the wafer defect detection method. Figure 2 The present specification shows a process flowchart of a wafer defect detection method provided in some other embodiments, which specifically includes the following steps.

[0151] Step 201: Acquire the original image of the target wafer and perform preprocessing to obtain the image to be tested.

[0152] Step 202: Reconstruct the image to be tested to obtain the reconstructed image.

[0153] Step 203: Calculate the L1 norm of the image to be tested and the reconstructed image to obtain the reconstruction error features.

[0154] Step 204: Import the image to be tested into the discriminator and generate discriminator features.

[0155] Step 205: Import the image to be tested into the encoder to generate latent spatial features.

[0156] Step 206: Concatenate and reconstruct the error features, discriminator features, and latent space features to obtain the first concatenated vector.

[0157] Step 207: Generate a standardized vector based on the mean and standard deviation of each dimension in the first concatenated vector.

[0158] Step 208: Import the standardized vector into the first prediction model to obtain the first anomaly probability, wherein the first prediction model is an SVM model.

[0159] Step 209: Concatenate the standardized vector with the first anomaly probability to obtain the second concatenated vector.

[0160] Step 210: Import the second concatenated vector into the second prediction model to generate the second anomaly probability, wherein the second prediction model is a random forest model.

[0161] Step 211: The first anomaly probability and the second anomaly probability are weighted and summed based on preset weights to obtain the target anomaly probability.

[0162] Step 212: When the target anomaly probability is greater than a preset first threshold, it is determined that the target wafer has a defect, wherein the first threshold is obtained based on a preset first update strategy.

[0163] Step 213: If it is determined that there are defects in the target wafer, the image to be tested is imported into the preset classification model to determine the defect classification of the target wafer. The defect classification includes known defect types and unknown defect types.

[0164] Step 214: If the anomaly of the target wafer is a known defect type, determine the location information of the known defect.

[0165] In some embodiments, steps 201-214 are related to Figure 1 The specific implementation of the corresponding steps in those embodiments and the resulting technical effects can be found in the following references. Figure 1 The steps involved will not be elaborated upon here.

[0166] Corresponding to the above method embodiments, this specification also provides embodiments of a wafer defect detection device. Figure 3 A schematic diagram of the structure of a wafer defect detection device provided in some embodiments of this specification is shown. For example... Figure 3 As shown, the device includes:

[0167] The acquisition module 301 is configured to acquire the original image of the target wafer and perform preprocessing to obtain the image to be tested.

[0168] The generation module 302 is configured to generate the target anomaly probability based on the image to be tested and the trained anomaly recognition model.

[0169] The first determining module 303 is configured to determine that the target wafer has a defect when the target anomaly probability is greater than a preset first threshold, wherein the first threshold is obtained based on a preset first update strategy.

[0170] The second determining module 304 is configured to import the image to be tested into a preset classification model to determine the defect classification of the target wafer when it is determined that there is a defect in the target wafer. The defect classification includes known defect types and unknown defect types.

[0171] The third determination module 305 is configured to determine the location information of the known defect when the anomaly of the target wafer is a known defect type.

[0172] In some embodiments, when the anomaly of the target wafer is a known defect type, determining the location information of the known defect includes:

[0173] When the anomaly of the target wafer is a known defect type, the corresponding localization model is determined based on the known defect type;

[0174] Import the image to be tested into the localization model to determine the location information.

[0175] In some embodiments, the first update strategy includes:

[0176] When training an anomaly recognition model based on a preset training set, the error rate of each training session is calculated. The training set includes multiple training wafer images with known results.

[0177] After processing the first number of training wafer images, the average of the first number of error rates is calculated to obtain the first current error rate;

[0178] The first threshold is updated based on a preset first calculation formula, wherein the first calculation formula includes:

[0179]

[0180] in, This represents the updated first threshold. This represents the first threshold before the update. This represents the preset first calculation factor. F c1 Indicates the current error rate. F t1 This represents the preset first target error rate.

[0181] In some embodiments, the classification model includes an anomaly classification model and an unknown defect verification model; wherein, the step of importing the image to be tested into the preset classification model to determine the defect classification of the target wafer includes:

[0182] The image to be tested is imported into the anomaly classification model to generate anomaly classification confidence. The anomaly classification model is an SVM model.

[0183] If the confidence level of the anomaly classification is less than the preset second threshold, the image to be tested is imported into the unknown defect verification model to determine the defect classification of the target wafer. The second threshold is obtained based on the preset second update strategy, and the unknown defect verification model is an isolated forest model.

[0184] In some embodiments, the second update strategy includes:

[0185] When training an anomaly classification model based on a pre-set training set, the error rate of each training session is calculated. The training set includes multiple training wafer images with known results.

[0186] After processing a second number of training wafer images, the average of the second number of error rates is calculated to obtain the second current error rate;

[0187] The second threshold is updated based on a preset second calculation formula, wherein the second calculation formula includes:

[0188]

[0189] in, This represents the updated second threshold. This represents the second threshold before the update. This represents the preset second calculation factor. F c2 Indicates the second current error rate. F t2 This represents the preset second target error rate.

[0190] In some embodiments, the step of importing the image to be tested into a preset classification model to determine the defect classification of the target wafer further includes:

[0191] If the anomaly classification confidence level is not less than the latest second threshold, the defect classification of the target wafer is determined to be a known defect type.

[0192] In some embodiments, the anomaly detection model includes a discriminator, an encoder, a first prediction model, and a second prediction model, wherein generating the target anomaly probability based on the image to be tested and the trained anomaly detection model includes:

[0193] The image to be tested is reconstructed to obtain the reconstructed image;

[0194] Calculate the L1 norm of the image under test and the reconstructed image to obtain the reconstruction error characteristics;

[0195] Import the image to be tested into the discriminator to generate discriminator features;

[0196] The image to be tested is imported into the encoder to generate latent spatial features;

[0197] The error features, discriminator features, and latent space features are concatenated and reconstructed to obtain the first concatenation vector;

[0198] A standardized vector is generated based on the mean and standard deviation of each dimension in the first concatenated vector;

[0199] The standardized vector is imported into the first prediction model to obtain the first anomaly probability, where the first prediction model is an SVM model;

[0200] The standardized vector is concatenated with the first anomaly probability to obtain the second concatenated vector;

[0201] The second concatenated vector is imported into the second prediction model to generate the second anomaly probability, wherein the second prediction model is a random forest model;

[0202] The first and second anomaly probabilities are weighted and summed based on preset weights to obtain the target anomaly probability.

[0203] The above is a schematic scheme of a wafer defect detection device according to this embodiment. It should be noted that the technical solution of this wafer defect detection device and the technical solution of the wafer defect detection method described above belong to the same concept. For details not described in detail in the technical solution of the wafer defect detection device, please refer to the description of the technical solution of the wafer defect detection method described above.

[0204] Figure 4 A structural block diagram of a computing device 400 according to some embodiments of this specification is shown. The components of the computing device 400 include, but are not limited to, a memory 401 and a processor 402. The processor 402 is connected to the memory 401 via a bus 403, and a database 405 is used to store data.

[0205] The computing device 400 also includes an access device 404 that enables the computing device 400 to communicate via one or more networks 406. Examples of such networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 404 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0206] In one embodiment of this specification, the aforementioned components of the computing device 400 and Figure 4 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 4 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0207] The computing device 400 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 400 can also be a mobile or stationary server.

[0208] The processor 402 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the aforementioned wafer defect detection method. The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the aforementioned wafer defect detection method belong to the same concept. Details not described in detail in the technical solution of the computing device can be found in the description of the technical solution of the aforementioned wafer defect detection method.

[0209] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described wafer defect detection method.

[0210] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the wafer defect detection method described above. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the wafer defect detection method described above.

[0211] An embodiment of this specification also provides a computer program, wherein when the computer program is executed in a computer, it causes the computer to perform the steps of the above-described wafer defect detection method.

[0212] The above is an illustrative example of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the technical solution of the aforementioned wafer defect detection method. Details not described in detail in the computer program's technical solution can be found in the description of the aforementioned wafer defect detection method's technical solution.

[0213] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0214] Computer instructions include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in computer-readable media can be appropriately added to or removed according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0215] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0216] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0217] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the embodiments described in this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for detecting wafer defects, characterized in that, include: The original image of the target wafer is acquired and preprocessed to obtain the image to be tested; Based on the image to be tested and the trained anomaly recognition model, a target anomaly probability is generated. The anomaly recognition model includes a discriminator, an encoder, a first prediction model, and a second prediction model. Generating the target anomaly probability based on the image to be tested and the trained anomaly recognition model includes: The image to be tested is reconstructed to obtain a reconstructed image; Calculate the L1 norm of the image to be tested and the reconstructed image to obtain the reconstruction error features; The image to be tested is imported into the discriminator to generate discriminator features; The image to be tested is imported into the encoder to generate latent spatial features; By concatenating the reconstructed error features, discriminator features, and latent space features, a first concatenation vector is obtained. Calculate the mean and standard deviation of each dimension in the first concatenated vector to obtain the standardized vector; The standardized vector is imported into the first prediction model to obtain the first anomaly probability, wherein the first prediction model is an SVM model; The standardized vector is concatenated with the first anomaly probability to obtain a second concatenated vector; The second concatenated vector is imported into the second prediction model to generate a second anomaly probability, wherein the second prediction model is a random forest model; The first anomaly probability and the second anomaly probability are weighted and summed based on preset weights to obtain the target anomaly probability; When the probability of the target anomaly is greater than a preset first threshold, it is determined that the target wafer has a defect, wherein the first threshold is obtained based on a preset first update strategy; If it is determined that the target wafer has defects, the image to be tested is imported into a preset classification model to determine the defect classification of the target wafer, wherein the defect classification includes known defect types and unknown defect types; If the anomaly of the target wafer is a known defect type, determine the location information of the known defect.

2. The method according to claim 1, characterized in that, If the anomaly on the target wafer is a known defect type, determine the location information of the known defect, including: If the anomaly of the target wafer is a known defect type, the corresponding localization model is determined according to the defect type of the known defect; The image to be tested is imported into the positioning model to determine the location information.

3. The method according to claim 1, characterized in that, The first update strategy includes: When training the anomaly recognition model according to a preset training set, the error rate of each training session is calculated, wherein the training set includes multiple training wafer images with known results; After processing the first number of training wafer images, the average of the first number of error rates is calculated to obtain the first current error rate; The first threshold is updated based on a preset first calculation formula, wherein the first calculation formula includes: in, This represents the updated first threshold. This represents the first threshold before the update. This represents the preset first calculation factor. F c1 This represents the first current error rate. F t1 This represents the preset first target error rate.

4. The method according to claim 1, characterized in that, The classification model includes an anomaly classification model and an unknown defect verification model; wherein, the step of importing the image to be tested into the preset classification model to determine the defect classification of the target wafer includes: The image to be tested is imported into the anomaly classification model to generate anomaly classification confidence, wherein the anomaly classification model is an SVM model; If the confidence level of the anomaly classification is less than a preset second threshold, the image to be tested is imported into the unknown defect verification model to determine the defect classification of the target wafer. The second threshold is obtained based on a preset second update strategy, and the unknown defect verification model is an isolated forest model.

5. The method according to claim 4, characterized in that, The second update strategy includes: When training the anomaly classification model according to a preset training set, the error rate of each training session is calculated, wherein the training set includes multiple training wafer images with known results; After processing a second number of training wafer images, the average of the second number of error rates is calculated to obtain the second current error rate; The second threshold is updated based on a preset second calculation formula, wherein the second calculation formula includes: in, This represents the updated second threshold. This represents the second threshold before the update. This represents the preset second calculation factor. F c2 This indicates the second current error rate. F t2 This represents the preset second target error rate.

6. The method according to claim 4, characterized in that, Also includes: If the anomaly classification confidence level is not less than the latest second threshold, the defect classification of the target wafer is determined to be a known defect type.

7. A wafer defect detection device, characterized in that, include: The acquisition module is configured to acquire the original image of the target wafer and perform preprocessing to obtain the image to be tested; The generation module is configured to generate a target anomaly probability based on the image to be tested and a trained anomaly recognition model. The anomaly recognition model includes a discriminator, an encoder, a first prediction model, and a second prediction model. Generating the target anomaly probability based on the image to be tested and the trained anomaly recognition model includes: reconstructing the image to be tested to obtain a reconstructed image; calculating the L1 norm of the image to be tested and the reconstructed image to obtain reconstruction error features; importing the image to be tested into the discriminator to generate discriminator features; importing the image to be tested into the encoder to generate latent space features; and concatenating the obtained reconstruction error features. The discriminator features and latent space features are used to obtain a first concatenated vector; the mean and standard deviation of each dimension in the first concatenated vector are calculated to obtain a standardized vector; the standardized vector is imported into the first prediction model to obtain a first anomaly probability, wherein the first prediction model is an SVM model; the standardized vector is concatenated with the first anomaly probability to obtain a second concatenated vector; the second concatenated vector is imported into the second prediction model to generate a second anomaly probability, wherein the second prediction model is a random forest model; the first anomaly probability and the second anomaly probability are weighted and summed based on preset weights to obtain the target anomaly probability; The first determining module is configured to determine that the target wafer has a defect when the target anomaly probability is greater than a preset first threshold, wherein the first threshold is obtained based on a preset first update strategy. The second determining module is configured to, when it is determined that the target wafer has defects, import the image to be tested into a preset classification model to determine the defect classification of the target wafer, wherein the defect classification includes known defect types and unknown defect types; The third determining module is configured to determine the location information of the known defect when the anomaly of the target wafer is a known defect type.

8. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the wafer defect detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed by a processor, they implement the steps of the wafer defect detection method according to any one of claims 1 to 6.

Images