Wafer defect category detection method and device
Through the combination method of autoencoder and convolutional neural network, abnormal detection is performed first and defect category detection is performed, which solves the problem that traditional methods are difficult to identify complex defect patterns, and improves the accuracy and efficiency of wafer defect detection.
Patent Information
- Application Number
- CN202411000172.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-15
AI Technical Summary
Traditional wafer defect detection methods are difficult to effectively capture the spatial distribution and complex patterns of defects, resulting in low recognition accuracy and overfitting problems.
Use the autoencoder as the wafer reconstruction model to be reconstructed. The wafer diagram to be detected is initially judged by the reconstruction error. If it is an abnormal diagram, then the defect category detection is performed through the second data set. The pre-trained convolutional neural network extracts features and performs similarity comparisons to determine the defect category.
It improves the recognition accuracy and accuracy of wafer defect detection, avoids overfitting, and can more accurately identify complex defect patterns, solving the recognition problems under the problem of data imbalance.
Smart Images

Figure CN118967601A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of semiconductor production and manufacturing, and in particular to a method and device for detecting wafer defect categories. Background Art
[0002] Wafers are key components in the semiconductor manufacturing process. They are round sheets made of silicon or other semiconductor materials and are used to manufacture integrated circuit chips (dies). Wafers need to have very high purity and flatness. However, during the semiconductor manufacturing process, defects such as scratches and voids may appear on the wafers due to various reasons. These defects may affect the performance and reliability of the final manufactured chips. Detecting and analyzing these defects is particularly important for locating problematic links and equipment in the manufacturing process.
[0003] Traditional defect detection methods mainly rely on numerical data analysis to cluster or classify the coordinates of defect points. However, these methods may not be able to effectively capture the spatial distribution and complex patterns of defects. Therefore, a method is needed to better perform wafer defect detection by capturing the spatial distribution and complex patterns of defects. Summary of the invention
[0004] The present invention describes a wafer defect category detection method and device, which performs image analysis on a wafer map drawn based on coordinate information of defect points to more efficiently and accurately detect the defect category of the wafer.
[0005] In a first aspect, a wafer defect category detection method is provided, comprising:
[0006] Obtaining a target wafer image to be inspected;
[0007] Inputting the target wafer image into a wafer reconstruction model to obtain a reconstructed wafer image; wherein the wafer reconstruction model is an autoencoder pre-trained based on a first data set including a plurality of normal wafer images;
[0008] Determining a target type of the target wafer image, including a normal wafer image and an abnormal wafer image, according to a reconstruction error between the target wafer image and the reconstructed wafer image;
[0009] When the target type is an abnormal wafer image, defect detection is performed on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image; wherein the second data set contains multiple abnormal wafer images and their respective defect categories.
[0010] In a possible implementation, obtaining a target wafer image to be inspected includes:
[0011] Obtaining defect point coordinate data of the wafer to be inspected;
[0012] The target wafer map is drawn according to the defect point coordinate data.
[0013] In a possible implementation manner, the normal wafer image is a wafer image in which defect points on the wafer image do not form a defect pattern; and the abnormal wafer image is a wafer image in which defect points on the wafer image form a defect pattern.
[0014] In a possible implementation manner, determining the target type of the target wafer image according to a reconstruction error between the target wafer image and the reconstructed wafer image includes:
[0015] When the reconstruction error is greater than a preset first threshold, determining the target type as an abnormal wafer image;
[0016] When the reconstruction error is less than or equal to the first threshold, the target type is determined as a normal wafer image.
[0017] In a possible implementation, defect detection is performed on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image, including:
[0018] Perform feature extraction on the target wafer image and multiple abnormal wafer images in the second data set respectively to obtain a target feature and multiple abnormal features;
[0019] The target feature and each abnormal feature are respectively compared for similarity, and the defect category corresponding to the abnormal feature with a higher similarity ranking is determined as the target defect category of the target wafer image.
[0020] In a possible implementation, feature extraction is performed on the target wafer image and multiple abnormal wafer images in the second data set to obtain target features and multiple abnormal features, including:
[0021] Using a pre-trained convolutional neural network, feature extraction is performed on the target wafer image and multiple abnormal wafer images in the second data set to obtain target features and multiple abnormal features.
[0022] In a possible implementation, defect detection is performed on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image, including:
[0023] The target wafer image is input into a defect classification model to obtain a target defect category of the target wafer image; wherein the defect classification model is pre-trained based on the second data set.
[0024] In a second aspect, a wafer defect category detection device is provided, comprising:
[0025] An acquisition unit configured to acquire a target wafer image to be detected;
[0026] A reconstruction unit is configured to input the target wafer image into a wafer reconstruction model to obtain a reconstructed wafer image; wherein the wafer reconstruction model is an autoencoder, which is pre-trained based on a first data set including a plurality of normal wafer images;
[0027] A first detection unit is configured to determine a target type of the target wafer image, including a normal wafer image and an abnormal wafer image, according to a reconstruction error between the target wafer image and the reconstructed wafer image;
[0028] The second detection unit is configured to, when the target type is an abnormal wafer image, perform defect detection on the defect pattern on the target wafer image based on a second data set to obtain a target defect category of the target wafer image; wherein the second data set contains multiple abnormal wafer images and their respective defect categories.
[0029] According to a third aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed in a computer, the computer is caused to execute the method of the first aspect.
[0030] According to a fourth aspect, a computing device is provided, comprising a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method of the first aspect is implemented.
[0031] The present invention proposes a wafer defect category detection method and device. The method first reconstructs the wafer image to be detected based on a pre-trained wafer reconstruction model, and compares the wafer to be detected with the normal wafer pattern learned by the wafer reconstruction model. When the reconstruction error is greater than a preset threshold, it means that the wafer reconstruction model cannot correctly reconstruct the wafer image to be detected, and further indicates that the wafer image to be detected belongs to an abnormal wafer image, that is, it contains a defective pattern. When the wafer image to be detected belongs to an abnormal wafer image, the defect category of the defective pattern on the wafer image to be detected is detected to obtain the final defect category.
[0032] For defect classification and recognition of semiconductor wafer images, there is a serious data imbalance problem in production. The number of normal wafer images accounts for the vast majority, while the samples of abnormal wafer images containing defective patterns may be extremely rare. This method avoids the problems of overfitting and low recognition accuracy that may be encountered in direct classification or clustering by first performing anomaly detection and then performing staged detection of defect categories. Overall, the accuracy and precision of recognition are improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the technical solutions of the multiple embodiments disclosed in this specification, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only the multiple embodiments disclosed in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0034] Figure 1 A flowchart of a wafer defect category detection method disclosed in an embodiment of the present invention;
[0035] Figure 2 Schematic diagram of various types of wafer defect patterns disclosed in an embodiment of the present invention;
[0036] Figure 3 A schematic block diagram of a wafer defect category detection device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION
[0037] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0038] Figure 1 Flow chart of a wafer defect category detection method disclosed in an embodiment of the present invention. Figure 1 As shown, the method at least includes: step 101, obtaining a target wafer image to be detected; step 102, inputting the target wafer image into a wafer reconstruction model to obtain a reconstructed wafer image; wherein the wafer reconstruction model is an autoencoder, which is pre-trained based on a first data set containing multiple normal wafer images; step 103, determining the target type of the target wafer image according to a reconstruction error between the target wafer image and the reconstructed wafer image, including a normal wafer image and an abnormal wafer image; step 104, when the target type is an abnormal wafer image, performing defect detection on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image; wherein the second data set contains multiple abnormal wafer images and their respective defect categories.
[0039] The specific execution process of each of the above steps is described below.
[0040] First, in step 101, a target wafer image to be inspected is obtained.
[0041] The wafer image is composed of multiple pixels, and any pixel can be a normal pixel or a defective pixel (hereinafter referred to as normal pixel and defective pixel). When there are multiple defective pixels on the wafer image, these defective pixels may form a special defect pattern in the spatial arrangement. Figure 2 Schematic diagram of various types of wafer defect patterns disclosed in an embodiment of the present invention. Figure 2 As shown in the figure, schematic diagrams of 10 common defect patterns on wafer images are shown. In each wafer image, the white part is the defect point, the dark gray part is the normal point, and the black part is the background. The defect pattern can be named according to the pattern formed by the distribution of defect points on the wafer. Among them, Figure 2 (a) can be called a ring defect, Figure 2 (b) can be called a central defect. Figure 2 (c) can be called a local defect. Figure 2 (d) can be called a local edge defect. Figure 2 (e) can be called a scratch defect. Figure 2 (f) can be called an edge ring defect, Figure 2 (g) can be called random defects, Figure 2 (h) can be called a nearly full defect. Figure 2 (i) can be called a mask defect. Figure 2 (j) can be called a stripe defect.
[0042] It should be noted that Figure 2 Only schematic diagrams of some common wafer defect patterns are shown, and those skilled in the art can define defect categories of other defect patterns according to actual needs.
[0043] In the present invention, a normal wafer image is defined as a wafer image in which defect points on the wafer image do not form a defect pattern; an abnormal wafer image is defined as a wafer image in which defect points on the wafer image form a defect pattern. That is, even if there are a few defect points on the wafer image, as long as these defect points do not form a defect pattern, it is still considered to be a normal wafer image.
[0044] In one embodiment, the data obtained from the semiconductor production machine may be the original defect point coordinate data, without direct wafer map data. In this embodiment, step 101 specifically includes: obtaining the defect point coordinate data of the wafer to be inspected; and drawing the target wafer map according to the defect point coordinate data.
[0045] Then, in step 102, the target wafer image is input into a wafer reconstruction model to obtain a reconstructed wafer image; wherein the wafer reconstruction model is an autoencoder, which is pre-trained based on a first data set containing a plurality of normal wafer images.
[0046] Various types of autoencoders can be used as wafer reconstruction models, for example, basic autoencoders (Auto Encoder), sparse autoencoders (Sparse Auto Encoder, SAE), variational autoencoders (Variational Auto Encoder, VAE), etc. can be used, without limitation here.
[0047] Among them, since the first data set only contains data of normal wafer images, the wafer reconstruction model trained by the first data set learns their potential representations by compressing and reconstructing normal wafer image samples, thereby learning the characteristics and distribution of normal wafer images.
[0048] When a new wafer image is input, the wafer reconstruction model will try to reconstruct it. Since the wafer reconstruction model has only learned the characteristics and distribution of normal wafer images, when the input new wafer image (the target wafer image to be detected) is a normal wafer image, the wafer reconstruction model can smoothly reconstruct it, and the difference (reconstruction error) between the output reconstructed wafer image and the target wafer image will be relatively small; when the input target wafer image is an abnormal wafer image, it is difficult for the wafer reconstruction model to reconstruct it correctly, and the output reconstructed wafer image will be relatively different from the target wafer image. According to the size of the reconstruction error, the type of the target wafer image to be detected can be determined.
[0049] Next, in step 103, a target type of the target wafer image is determined according to a reconstruction error between the target wafer image and the reconstructed wafer image, including a normal wafer image and an abnormal wafer image.
[0050] The reconstruction error can be calculated using a variety of methods, for example, Mean Square Error (MSE), Cross Entropy, etc., which are not limited here.
[0051] As mentioned above, the wafer reconstruction model that has only learned the normal wafer image has a smaller reconstruction error for the normal wafer image, but a larger reconstruction error for the abnormal wafer image. Based on this, in one embodiment, step 103 specifically includes: when the reconstruction error is greater than a preset first threshold, determining the target type as an abnormal wafer image; when the reconstruction error is less than or equal to the first threshold, determining the target type as a normal wafer image.
[0052] In the above steps 101 to 103 , the target wafer image to be inspected may be subjected to a first round of rough classification, and when the target wafer image is an abnormal wafer image, a second round of detailed classification in step 104 may be performed.
[0053] Finally, in step 104, when the target type is an abnormal wafer image, defect detection is performed on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image; wherein the second data set contains multiple abnormal wafer images and their respective defect categories.
[0054] The second data set is a data set of abnormal wafer images, which contains multiple abnormal wafer images and their corresponding defect categories as labels. Based on the second data set, defect detection can be performed on the defect pattern of the target wafer image. For example, when the amount of data in the second data set is small, the target wafer image can be compared with each abnormal wafer image in the second data set for similarity, and the defect category of the abnormal wafer image with the highest similarity ranking can be used as the target defect category of the target wafer image. When the amount of data in the second data set is large, a multi-classification model can be trained based on the second data set first, and each defect category in the second data set can be used as each category of the multi-classification model. Then, the multi-classification model is used to determine the target defect category of the defect pattern of the target wafer image.
[0055] Based on the above ideas, in one embodiment, step 104 specifically includes: performing feature extraction on the target wafer image and multiple abnormal wafer images in the second data set, respectively, to obtain target features and multiple abnormal features; performing similarity comparison between the target features and each abnormal feature, respectively, and determining the defect category corresponding to the abnormal feature with the highest similarity ranking as the target defect category of the target wafer image.
[0056] In a more specific embodiment, a pre-trained convolutional neural network is used to extract features from the target wafer image and multiple abnormal wafer images in the second data set, respectively, to obtain target features and multiple abnormal features.
[0057] A variety of pre-trained convolutional neural networks can be used, such as VGG (Visual Geometry Group), ResNet (Deep Residual Network), etc., without limitation here.
[0058] A variety of methods can be used to calculate the similarity between two features, for example, cosine similarity, Euclidean distance, etc. can be used, which are not limited here.
[0059] In other embodiments, other methods may be used to extract features from the wafer image, for example, a SIFT (Scale-invariant feature transform) method may be used.
[0060] In another embodiment, step 104 specifically includes: inputting the target wafer image into a defect classification model to obtain a target defect category of the target wafer image; wherein the defect classification model is pre-trained based on the second data set.
[0061] The defect classification model can be any multi-classification model based on a discriminant model. For example, it can be a multi-classification model for images based on a combination of a convolutional neural network and a multi-layer neural network, or ViT (Vision Transformer, visual Transformer model), etc., which is not limited here.
[0062] Through the above steps 101 to 104, the defect category detection of the wafer is divided into two stages. In the first stage, the deep learning model is trained in an unsupervised learning mode to identify normal wafer images, thereby realizing the detection and identification of abnormal wafer images. In this stage, a large number of normal wafer images will be detected and stopped from entering the second stage to improve efficiency, while a small number of abnormal wafer images will continue to enter the second stage. Then, in the second stage, by matching the detected abnormal patterns with the special pattern library (second data set), the method can effectively identify and classify various defect types in the wafer image.
[0063] Since there is a serious data imbalance problem in production for the classification and recognition of semiconductor wafer defect images, the samples of semiconductor wafer defect patterns may be extremely rare, that is, the number of samples in the first data set is much larger than the number of samples in the second data set. This method avoids the problems of overfitting and low recognition accuracy that may be encountered in direct classification or clustering by first performing anomaly detection and then performing defect category detection in a two-stage process.
[0064] At the same time, this method converts the coordinate data of semiconductor wafer defects into image data, so that the deep learning model can effectively utilize the spatial distribution information of defects. Compared with the method based only on numerical values, it solves the problem that it is difficult to identify anomalies based on numerical features such as density, and can more accurately identify complex defect patterns.
[0065] According to another embodiment, a wafer defect classification detection device is also provided. Figure 3 This is a schematic block diagram of a wafer defect category detection device disclosed in an embodiment of the present invention. The device can be deployed in any device, platform or device cluster with computing and processing capabilities. Figure 3 As shown, the device 300 includes:
[0066] The acquisition unit 301 is configured to acquire a target wafer image to be detected;
[0067] The reconstruction unit 302 is configured to input the target wafer image into a wafer reconstruction model to obtain a reconstructed wafer image; wherein the wafer reconstruction model is an autoencoder, which is pre-trained based on a first data set including a plurality of normal wafer images;
[0068] A first detection unit 303 is configured to determine a target type of the target wafer image, including a normal wafer image and an abnormal wafer image, according to a reconstruction error between the target wafer image and the reconstructed wafer image;
[0069] The second detection unit 304 is configured to, when the target type is an abnormal wafer image, perform defect detection on the defect pattern on the target wafer image based on a second data set to obtain a target defect category of the target wafer image; wherein the second data set contains multiple abnormal wafer images and their respective defect categories.
[0070] According to another aspect of the embodiment, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed in a computer, the computer is caused to execute the method described in any of the above embodiments.
[0071] According to yet another embodiment, a computing device is provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method described in any one of the above embodiments is implemented.
[0072] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0073] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0074] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0075] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.
[0076] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A wafer defect category detection method, comprising: Obtaining a target wafer image to be inspected; Inputting the target wafer image into a wafer reconstruction model to obtain a reconstructed wafer image; wherein the wafer reconstruction model is an autoencoder pre-trained based on a first data set including a plurality of normal wafer images; Determining a target type of the target wafer image, including a normal wafer image and an abnormal wafer image, according to a reconstruction error between the target wafer image and the reconstructed wafer image; When the target type is an abnormal wafer image, defect detection is performed on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image; wherein the second data set contains multiple abnormal wafer images and their respective defect categories.
2. The method according to claim 1, characterized in that Obtain the target wafer image to be inspected, including: Obtaining defect point coordinate data of the wafer to be inspected; The target wafer map is drawn according to the defect point coordinate data.
3. The method according to claim 1, characterized in that The normal wafer image is a wafer image in which defect points on the wafer image do not form a defect pattern; the abnormal wafer image is a wafer image in which defect points on the wafer image form a defect pattern.
4. The method according to claim 1, characterized in that: Determining a target type of the target wafer image according to a reconstruction error between the target wafer image and the reconstructed wafer image includes: When the reconstruction error is greater than a preset first threshold, determining the target type as an abnormal wafer image; When the reconstruction error is less than or equal to the first threshold, the target type is determined as a normal wafer image.
5. The method according to claim 1, characterized in that Performing defect detection on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image includes: Perform feature extraction on the target wafer image and multiple abnormal wafer images in the second data set respectively to obtain a target feature and multiple abnormal features; The target feature and each abnormal feature are respectively compared for similarity, and the defect category corresponding to the abnormal feature with a higher similarity ranking is determined as the target defect category of the target wafer image.
6. The method according to claim 5, characterized in that Feature extraction is performed on the target wafer image and multiple abnormal wafer images in the second data set respectively to obtain target features and multiple abnormal features, including: Using a pre-trained convolutional neural network, feature extraction is performed on the target wafer image and multiple abnormal wafer images in the second data set to obtain target features and multiple abnormal features.
7. The method according to claim 1, characterized in that Performing defect detection on the defect pattern on the target wafer image based on the second data set to obtain a target defect category of the target wafer image includes: The target wafer image is input into a defect classification model to obtain a target defect category of the target wafer image; wherein the defect classification model is pre-trained based on the second data set.
8. A wafer defect type detection device, comprising: An acquisition unit configured to acquire a target wafer image to be detected; A reconstruction unit is configured to input the target wafer image into a wafer reconstruction model to obtain a reconstructed wafer image; wherein the wafer reconstruction model is an autoencoder, which is pre-trained based on a first data set including a plurality of normal wafer images; A first detection unit is configured to determine a target type of the target wafer image, including a normal wafer image and an abnormal wafer image, according to a reconstruction error between the target wafer image and the reconstructed wafer image; The second detection unit is configured to, when the target type is an abnormal wafer image, perform defect detection on the defect pattern on the target wafer image based on a second data set to obtain a target defect category of the target wafer image; wherein the second data set contains multiple abnormal wafer images and their respective defect categories.
9. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 7.
10. A computing device comprising a memory and a processor, wherein: The memory stores executable codes, and when the processor executes the executable codes, the method according to any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Defect detection and classification method and device, electronic equipment and storage medium
CN114764768A
Wafer failure model establishment method, wafer failure prediction method, wafer failure prediction device and related equipment
CN116012673A
Wafer defect identification method and device, electronic equipment and storage medium
CN116128839A
Method, apparatus and program for detecting anomalies in process equipment by integrating multiple anomaly detection techniques
KR102665637B1
Classification method, classification device for defect detection, electronic device and storage media
TW202227808A
Cited By
Image analysis method and system for semiconductor wafer defect detection
CN122617848A
