A method, system, electronic device and storage medium for synthesizing defect data
By using specific noise to generate the intersection of defect mask and texture images, the defective semiconductor wafer images are synthesized, which solves the problem of insufficient training data and improves the accuracy of wafer defect detection.
Patent Information
- Application Number
- CN202411944426.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-24
- Publication Date
- 2025-04-04
AI Technical Summary
Since there are few defective semiconductor wafer images, the generalization performance of the trained wafer defect detection model is poor, which affects the accuracy of wafer defect detection.
By generating a defect mask using specific noise, the defect mask and texture image intersect, and a texture local image is obtained. Based on the defect mask, texture local image, and defect-free semiconductor wafer image, the defective semiconductor wafer image is synthesized.
Through the synthetic defective image training model, the accuracy of wafer defect detection is improved, and the problem of insufficient training data in the prior art is solved.
Smart Images
Figure CN119762361A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image processing, and in particular to a method, system, electronic device and storage medium for synthesizing defect data. Background Art
[0002] The background textures of semiconductor wafer images are complex and diverse, and the distribution of different types of semiconductor wafer images is extremely uneven; generally speaking, defect-free semiconductor wafer images on the production line account for the vast majority, while defective semiconductor wafer images are rare.
[0003] When using defective semiconductor wafer images to train a wafer defect detection model, due to the small number of defective semiconductor wafer images, the generalization performance of the trained wafer defect detection model is poor, which affects the accuracy of wafer defect detection. Therefore, there is an urgent need for a method that can synthesize semiconductor wafer defect data to enrich the training data used to train the wafer defect detection model. Summary of the invention
[0004] In view of this, an embodiment of the present invention provides a method, system, electronic device and storage medium for synthesizing defect data to synthesize semiconductor wafer defect data.
[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:
[0006] A first aspect of an embodiment of the present invention discloses a method for synthesizing defect data, the method comprising:
[0007] Generate defect mask using specific noise;
[0008] Randomly select texture images from the texture dataset;
[0009] Taking the intersection of the defect mask and the texture image to obtain a local texture image;
[0010] A defective semiconductor wafer image is synthesized based on the defect mask, the texture local image, and the defect-free semiconductor wafer image.
[0011] Preferably, generating a defect mask using specific noise includes:
[0012] Perform random enhancement on specific noise;
[0013] The specific noise after random enhancement processing is binarized to generate a defect mask.
[0014] Preferably, taking the intersection of the defect mask and the texture image to obtain a local texture image includes:
[0015] Performing random enhancement processing on the texture image;
[0016] The defect mask and the texture image after random enhancement processing are intersected to obtain a local texture image.
[0017] Preferably, synthesizing a defective semiconductor wafer image based on the defect mask, the texture local image, and the defect-free semiconductor wafer image comprises:
[0018] Complementing the defect mask in a bitwise inverted manner to obtain a first image;
[0019] Obtain a second image by taking pixel-by-pixel intersection of the first image and a non-defective semiconductor wafer image;
[0020] Performing weighted fusion on the texture local image and the defect-free semiconductor wafer image to obtain a third image;
[0021] The second image and the third image are fused to obtain a defective semiconductor wafer image.
[0022] Preferably, after generating the defect mask, the method further includes:
[0023] The defect mask whose area is smaller than the area threshold is filtered out.
[0024] Preferably, after synthesizing the defective semiconductor wafer image, the method further includes:
[0025] A wafer defect detection model is trained using the defective semiconductor wafer image until the wafer defect detection model converges.
[0026] Preferably, after training the wafer defect detection model using the defective semiconductor wafer image, the method further includes:
[0027] Acquire an image of a semiconductor wafer to be inspected;
[0028] The semiconductor wafer image to be inspected is input into the converged wafer defect detection model for reasoning to obtain defect information of the semiconductor wafer image to be inspected.
[0029] A second aspect of an embodiment of the present invention discloses a system for synthesizing defect data, the system comprising:
[0030] A generating unit, used for generating a defect mask by using specific noise;
[0031] A selection unit, used for randomly selecting a texture image from a texture data set;
[0032] A processing unit, configured to take an intersection of the defect mask and the texture image to obtain a local texture image;
[0033] A synthesis unit is used to synthesize a defective semiconductor wafer image based on the defect mask, the texture local image, and a defect-free semiconductor wafer image.
[0034] A third aspect of an embodiment of the present invention discloses an electronic device, including a memory and a processor;
[0035] The memory is used to store computer programs;
[0036] The processor is used to execute the computer program, specifically to implement the method for synthesizing defect data disclosed in the first aspect of the embodiment of the present invention.
[0037] A fourth aspect of an embodiment of the present invention discloses a computer storage medium for storing a computer program. When the computer program is executed, it is specifically used to implement the method for synthesizing defect data disclosed in the first aspect of an embodiment of the present invention.
[0038] Based on the above-mentioned embodiment of the present invention, a method, system, electronic device and storage medium for synthesizing defect data are provided. The method is: using specific noise to generate a defect mask; randomly selecting a texture image from a texture data set; taking the intersection of the defect mask and the texture image to obtain a local texture image; based on the defect mask, the local texture image, and a defect-free semiconductor wafer image, synthesizing a defective semiconductor wafer image. In this scheme, a defect mask is generated using specific noise, and the defect mask and a texture image randomly selected from a texture data set are intersected to obtain a local texture image. Based on the defect mask, the local texture image, and a defect-free semiconductor wafer image, a defective semiconductor wafer image is synthesized, and the synthesized defective semiconductor wafer image is used for training a wafer defect detection model to improve the accuracy of wafer defect detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0040] Figure 1 A flowchart of a method for synthesizing defect data provided by an embodiment of the present invention;
[0041] Figure 2 An example diagram of defect data synthesis steps provided by an embodiment of the present invention;
[0042] Figure 3 A structural block diagram of a system for synthesizing defect data provided by an embodiment of the present invention;
[0043] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.
[0045] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising 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, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.
[0046] The background textures of semiconductor wafer images are complex and diverse, and the distribution of different categories of semiconductor wafer images is extremely uneven; generally speaking, defect-free semiconductor wafer images on the production line (also called GOOD data) account for the vast majority, while defective semiconductor wafer images (also called NG data) are rare.
[0047] When a wafer defect detection model is trained using defective semiconductor wafer images (equivalent to using deep learning for wafer defect detection), due to the small number of defective semiconductor wafer images, the generalization performance of the trained wafer defect detection model is poor, which affects the accuracy of wafer defect detection.
[0048] To solve the above problems, this solution proposes a method, system, electronic device and storage medium for synthesizing defect data, which uses specific noise to generate a defect mask, takes the intersection of the defect mask and a texture image randomly selected from a texture data set, and obtains a local texture image. Based on the defect mask, the local texture image, and the defect-free semiconductor wafer image, a defective semiconductor wafer image is synthesized, and the synthesized defective semiconductor wafer image is used for training a wafer defect detection model to improve the accuracy of wafer defect detection.
[0049] In practical applications, a large number of defect-free semiconductor wafer images (GOOD data) can be used to synthesize a large number of defective semiconductor wafer images (NG data), thereby solving the problem of too little NG data and improving the accuracy of the model in detecting wafer defects.
[0050] The present solution is described in detail below through various embodiments.
[0051] See also Figure 1 , shows a flow chart of a method for synthesizing defect data provided by an embodiment of the present invention, the method comprising:
[0052] Step S101: Generate a defect mask using specific noise.
[0053] In the specific implementation of step S101, the specific noise is randomly enhanced (which can increase the richness of the defect mask generated subsequently), and the random enhancement process is any of the following processing methods: rotation, translation, and scaling; the specific noise after the random enhancement process is binarized to generate a defect mask.
[0054] It should be noted that, considering that semiconductor wafer defect detection is sensitive to defect size, the generated defect mask needs to be filtered to a certain extent; in some embodiments, after the defect mask is generated, the defect mask with an area smaller than the area threshold is filtered out, that is, the defect mask (blob) with an area that is too small is filtered out.
[0055] In practical applications, the specific noise in this solution may be Perlin noise, or other noise having a texture similar to Perlin noise.
[0056] For example: assuming that the specific noise is Perlin noise P; in order to increase the richness of the defect mask M, the Perlin noise P is first randomly enhanced (rotated, translated or scaled); then the Perlin noise P after the random enhancement process is binarized to generate the defect mask M; the defect mask M with an area smaller than the area threshold is filtered out.
[0057] Step S102: randomly selecting a texture image from the texture data set.
[0058] In the specific implementation of step S102 , a texture image T is randomly selected from a texture data set. The texture data set may be a DTD (Describable Textures Dataset) texture data set, or the texture data set may be constructed by collecting data according to actual needs.
[0059] Step S103: Taking the intersection of the defect mask and the texture image to obtain a local texture image.
[0060] In the process of implementing step S103, in order to increase the richness of the defect data finally generated, the texture image T (a texture image randomly selected from the texture data set) is first randomly enhanced, and the random enhancement processing is any of the following processing methods: rotation, translation, scaling, color change; then the defect mask M and the texture image T after the random enhancement processing are intersected to obtain the texture local image N.
[0061] For example: after randomly selecting a texture image T from a texture data set, the texture image T is first randomly enhanced (rotated, translated, scaled or color transformed); then the defect mask M and the texture image T after random enhancement are intersected to obtain a texture local image N.
[0062] Step S104: synthesizing a defective semiconductor wafer image based on the defect mask, the local texture image, and the defect-free semiconductor wafer image.
[0063] In the specific implementation of step S104 , the defect mask M, the texture local image N, and the defect-free semiconductor wafer image S (GOOD data) are synthesized into a defective semiconductor wafer image O (NG data) through a certain processing logic.
[0064] Specifically, the defect mask M is complemented in a bitwise inversion manner (i.e., the defect mask M is bitwise inverted) to obtain a first image M'; the first image M' and the defect-free semiconductor wafer image S are intersected pixel by pixel to obtain a second image M1; the texture local image N and the defect-free semiconductor wafer image S are weightedly fused to obtain a third image O'; the second image M1 and the third image O' are fused to obtain a defective semiconductor wafer image O.
[0065] Specifically, the specific method of weighted fusion of the texture local image N and the defect-free semiconductor wafer image S is as follows: weighted fusion of the texture local image N and the defect-free semiconductor wafer image S is performed according to formula (1) to obtain a third image O'.
[0066] O'=β*N+(1-β)*(M*S)(1)
[0067] In formula (1), N is the texture local image; M is the defect mask; S is the defect-free semiconductor wafer image (GOOD data); “*” indicates pixel-by-pixel multiplication; β is a hyperparameter, which is generally uniformly sampled from a certain interval, for example, β is generally uniformly sampled from [0.15, 1].
[0068] In some embodiments, after synthesizing a defective semiconductor wafer image, the defective semiconductor wafer image is used to train a wafer defect detection model until the wafer defect detection model converges.
[0069] Specifically, the synthesized defective semiconductor wafer image is added to the NG dataset, and then the NG dataset is used to train the wafer defect detection model until the wafer defect detection model converges.
[0070] In practical applications, a semiconductor wafer image to be inspected is obtained, and the semiconductor wafer image to be inspected is input into a converged wafer defect detection model for reasoning to obtain defect information of the semiconductor wafer image to be inspected, and the defect information at least includes the defect category and the defect location.
[0071] In an embodiment of the present invention, a defect mask is generated using specific noise, and the defect mask and a texture image randomly selected from a texture data set are intersected to obtain a local texture image. A defective semiconductor wafer image is synthesized based on the defect mask, the local texture image, and a defect-free semiconductor wafer image, and the synthesized defective semiconductor wafer image is used for training a wafer defect detection model to improve the accuracy of wafer defect detection.
[0072] To better understand the above embodiments of the present invention Figure 1 The content in Figure 2 The example diagram of the defect data synthesis step is illustrated by way of example; Figure 2 It includes steps one to three.
[0073] Step 1: Perform rotation transformation and binarization on the Perlin noise P to generate a defect mask M.
[0074] Step 2: Randomly select a texture image T from the texture dataset; perform matting based on the defect mask M and the texture image T to obtain a texture local image N (that is, Figure 2 The above-mentioned “cutout processing” is to take the intersection of the defect mask M and the texture image T.
[0075] Step 3: complement the defect mask M (i.e., invert it bit by bit) to obtain a first image M'; based on the first image M', the defect-free semiconductor wafer image S, and the texture local image N, synthesize the defective semiconductor wafer image O.
[0076] The above is the relevant description of this plan.
[0077] In practical applications, this solution can be used to generate a large amount of NG data (defective semiconductor wafer images), which can then be used to train a model for detecting wafer defects. Based on the above, one of the practical application scenarios of this solution is as follows:
[0078] A1. Collect a large amount of data about semiconductor wafers on the production line, which includes a large amount of GOOD data and a small amount of NG data; clean up the collected data.
[0079] A2. Collect texture datasets according to actual needs.
[0080] A3. Using the above-mentioned embodiment of the present invention Figure 1 A large amount of defect data (defective semiconductor wafer images) is synthesized by the various steps in the process, and the synthesized defect data is added to the NG data set.
[0081] A4. Build a wafer defect detection model according to actual needs and set the initial parameters of the wafer defect detection model.
[0082] A5. Train the wafer defect detection model using the NG data set until the wafer defect detection model converges.
[0083] A6. After the wafer defect detection model is trained (after the wafer defect detection model is trained to convergence), the image to be tested (the semiconductor wafer image to be tested) is input into the wafer defect detection model for inference. The wafer defect detection model outputs defect information of the image to be tested, and the defect information at least includes the defect category and the defect location.
[0084] It can be seen from the above practical application scenarios that this solution can use a large amount of GOOD data to synthesize a large amount of NG data, thereby solving the problem of too little NG data and further improving the accuracy of the model in detecting wafer defects.
[0085] Corresponding to the method for synthesizing defect data provided by the above embodiment of the present invention, see Figure 3 , the embodiment of the present invention also provides a structural block diagram of a system for synthesizing defect data, the system comprising: a generating unit 301, a selecting unit 302, a processing unit 303, and a synthesizing unit 304;
[0086] The generating unit 301 is configured to generate a defect mask using specific noise.
[0087] In a specific implementation, the generating unit 301 is specifically used to: perform random enhancement processing on the specific noise; perform binarization processing on the specific noise after the random enhancement processing to generate a defect mask.
[0088] The selection unit 302 is used to randomly select a texture image from the texture data set.
[0089] The processing unit 303 is used to obtain the intersection of the defect mask and the texture image to obtain a local texture image.
[0090] In a specific implementation, the processing unit 303 is specifically used for: performing random enhancement processing on the texture image; taking the intersection of the defect mask and the texture image after random enhancement processing to obtain a local texture image.
[0091] The synthesis unit 304 is used to synthesize the defective semiconductor wafer image based on the defect mask, the texture local image, and the defect-free semiconductor wafer image.
[0092] In an embodiment of the present invention, a defect mask is generated using specific noise, and the defect mask and a texture image randomly selected from a texture data set are intersected to obtain a local texture image. A defective semiconductor wafer image is synthesized based on the defect mask, the local texture image, and a defect-free semiconductor wafer image, and the synthesized defective semiconductor wafer image is used for training a wafer defect detection model to improve the accuracy of wafer defect detection.
[0093] Preferably, combined Figure 3 As shown in the figure, the synthesis unit 304 includes a first processing module, a second processing module, a fusion module, and a third processing module; the execution principle of each module is as follows:
[0094] The first processing module is used to complement the defect mask in a bitwise inversion manner to obtain a first image.
[0095] The second processing module is used to obtain a second image by taking the intersection of the first image and the image of the semiconductor wafer without defects pixel by pixel.
[0096] The fusion module is used for weighted fusion of the texture local image and the defect-free semiconductor wafer image to obtain a third image.
[0097] The third processing module is used to fuse the second image and the third image to obtain a defective semiconductor wafer image.
[0098] Preferably, combined Figure 3 The system further comprises:
[0099] The filtering unit is used to filter out defect masks whose areas are smaller than an area threshold.
[0100] Preferably, combined Figure 3 The system further comprises:
[0101] The training unit is used to train the wafer defect detection model using defective semiconductor wafer images until the wafer defect detection model converges. The inference unit is used to obtain the semiconductor wafer image to be detected; input the semiconductor wafer image to be detected into the converged wafer defect detection model for inference to obtain the defect information of the semiconductor wafer image to be detected.
[0102] The embodiment of the present invention provides a system for synthesizing defect data. The specific working principle of the system can be found in the above method embodiment and will not be described in detail here.
[0103] The embodiment of the present invention further provides a computer storage medium for storing a computer program. When the computer program is executed, it is specifically used to implement the method for synthesizing defect data provided by the above method embodiment.
[0104] The present invention also provides an electronic device, see Figure 4 , Figure 4 is a schematic diagram of the structure of the electronic device, which may include a memory 401 and a processor 402;
[0105] The memory 401 is used to store computer programs;
[0106] The processor 402 is used to execute a computer program, specifically to implement the method for synthesizing defect data provided by the above method embodiment.
[0107] In summary, the embodiments of the present invention provide a method, system, electronic device and storage medium for synthesizing defect data, which uses specific noise to generate a defect mask, takes the intersection of the defect mask and a texture image randomly selected from a texture data set, and obtains a local texture image. Based on the defect mask, the local texture image, and a defect-free semiconductor wafer image, a defective semiconductor wafer image is synthesized, and the synthesized defective semiconductor wafer image is used for training a wafer defect detection model to improve the accuracy of wafer defect detection.
[0108] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.
[0109] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0110] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for synthesizing defect data, characterized in that: The method comprises: Generate defect mask using specific noise; Randomly select texture images from the texture dataset; Taking the intersection of the defect mask and the texture image to obtain a local texture image; A defective semiconductor wafer image is synthesized based on the defect mask, the texture local image, and the defect-free semiconductor wafer image.
2. The method according to claim 1, characterized in that: Generate defect masks using specific noise, including: Perform random enhancement on specific noise; The specific noise after random enhancement processing is binarized to generate a defect mask.
3. The method according to claim 1, characterized in that Taking the intersection of the defect mask and the texture image to obtain a local texture image includes: Performing random enhancement processing on the texture image; The defect mask and the texture image after random enhancement processing are intersected to obtain a local texture image.
4. The method according to any one of claims 1 to 3, characterized in that: Based on the defect mask, the texture local image, and the defect-free semiconductor wafer image, synthesizing a defective semiconductor wafer image includes: Complementing the defect mask in a bitwise inverted manner to obtain a first image; Obtain a second image by taking pixel-by-pixel intersection of the first image and a non-defective semiconductor wafer image; Performing weighted fusion on the texture local image and the defect-free semiconductor wafer image to obtain a third image; The second image and the third image are fused to obtain a defective semiconductor wafer image.
5. The method according to any one of claims 1 to 3, characterized in that: After the defect mask is generated, it also includes: The defect mask whose area is smaller than the area threshold is filtered out.
6. The method according to any one of claims 1 to 3, characterized in that: After synthesizing defective semiconductor wafer images, it also includes: A wafer defect detection model is trained using the defective semiconductor wafer image until the wafer defect detection model converges.
7. The method according to claim 6, characterized in that After training the wafer defect detection model using the defective semiconductor wafer image, the method further includes: Acquire an image of a semiconductor wafer to be inspected; The semiconductor wafer image to be inspected is input into the converged wafer defect detection model for reasoning to obtain defect information of the semiconductor wafer image to be inspected.
8. A system for synthesizing defect data, characterized in that: The system comprises: A generating unit, used for generating a defect mask by using specific noise; A selection unit, used for randomly selecting a texture image from a texture data set; A processing unit, configured to take an intersection of the defect mask and the texture image to obtain a local texture image; A synthesis unit is used to synthesize a defective semiconductor wafer image based on the defect mask, the texture local image, and the defect-free semiconductor wafer image.
9. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to execute the computer program, specifically to implement the method for synthesizing defect data as claimed in any one of claims 1 to 7.
10. A computer storage medium, characterized in that: Used to store a computer program, which, when executed, is specifically used to implement the method for synthesizing defect data as claimed in any one of claims 1 to 7.