A differential manifold-based integrated circuit surface defect classification method

Abstract

The application discloses a kind of integrated circuit surface defect classification methods based on differential manifold, including: microscopic imaging acquisition system, image pre-processing system and defect image classification system, first, the integrated circuit image after pre-processing is mapped to Riemann manifold space;Second, the feature descriptor of image is constructed, the characteristic vector is obtained, that is, covariance matrix can be constructed and corrected as the positive definite symmetric matrix of Lie group manifold;Then, further map the matrix to low-dimensional Lie group space, design Lie group kernel function;Finally, realize the accurate classification of integrated circuit surface diversity defect image.The application solves the problem of rapid classification of integrated circuit surface diversity defect image, improves the reliability in the process of integrated circuit manufacturing, and guarantees the quality of integrated circuit downstream application industry.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of integrated circuit manufacturing, and particularly relates to an integrated circuit surface defect classification method based on differential manifold. BACKGROUND

[0002] As a core component of modern electronic technology, integrated circuits play an important role in various fields of application. Its high integration, miniaturization and high performance characteristics make it have wide application prospects in the fields of information technology, communication, medical treatment and industrial control. SUMMARY

[0003] The purpose of the present application is to provide an integrated circuit surface defect classification method based on differential manifold to solve the problems raised in the background.

[0004] To achieve the above purpose, the present application provides the following technical scheme: an integrated circuit surface defect classification method based on differential manifold, specifically comprising the following steps:

[0005] Step (1) collecting digital images of each part of the integrated circuit surface through microscopic imaging;

[0006] Step (2) pre-processing the collected images;

[0007] Step (3) extracting features and classifying the pre-processed integrated circuit surface images to facilitate subsequent defect type judgment.

[0008] Further, the collection process of step (1) needs to use a high-power microscope. Usually, a complete integrated circuit needs to be divided into hundreds of regions for image collection, and there is always a small overlapping part between the images of each region, so it is often difficult to obtain a standard template.

[0009] Further, in step (2), since a high-power microscope is used to collect industrial images online, the complexity of the environment often leads to poor image quality, such as: introducing various unknown noises in a complex industrial environment, low image contrast, and shaking blur. No matter what causes the image quality to decline, it will affect the accuracy of subsequent defect classification. Pre-processing the collected images will help improve the accuracy of defect classification. The specific steps include:

[0010] Step (2.1) analyzing what factors affect the quality of the integrated circuit surface images collected online;

[0011] Step (2.2) denoising the integrated circuit images collected online;

[0012] Step (2.3) improves the overall brightness and performs contrast enhancement processing on the noise-reduced integrated circuit surface image.

[0013] Step (2.4) completes the preprocessing operation of the integrated circuit image.

[0014] Further, in the step (2.2), many kinds of noise introduced in the online acquisition process of the industrial environment, such as: salt and pepper noise, Gaussian noise, mixed noise. This patent performs noise reduction processing on multiple unknown noises, specifically including:

[0015] Step (2.2.1) analyzes the integrated circuit surface defect image using topology-related theory;

[0016] Step (2.2.2) selects a homotopy mapping suitable for the integrated circuit surface defect image to construct a new topological space;

[0017] Step (2.2.3) combines the Gaussian-Laplacian transform to design a filter template suitable for multiple unknown noises;

[0018] Step (2.2.4) realizes the noise reduction processing of multiple unknown noises.

[0019] Further, in the step (2.3), the integrated circuit surface image is first subjected to contrast enhancement processing, and then zero-crossing processing is performed to ensure that the intrinsic features of the defects are extracted and the precision of defect classification is improved.

[0020] Further, the step (3) specifically includes:

[0021] Step (3.1) selects a suitable mapping method for the preprocessed integrated circuit image to map the image to a Riemannian manifold space.

[0022] Step (3.2) constructs a feature descriptor of the image to obtain a feature vector L ;

[0023] Step (3.3) constructs a covariance matrix J ;

[0024] Step (3.4) corrects the covariance matrix J to a positive definite symmetric matrix of Lie group manifold X ;

[0025] Step (3.5) calculates the distance between the positive definite symmetric matrices D ;

[0026] Step (3.6) maps the positive definite symmetric matrix image descriptor to a low-dimensional Lie group space and designs a Lie group kernel function;

[0027] Step (3.7) realizes the accurate classification of the integrated circuit surface defect image, and forms a "defect type library" of the system.

[0028] Further, in step (3.1), since the metric of the Euclidean space is difficult to be used for nonlinear data, the preprocessed integrated circuit image is mapped to a Riemannian manifold space so as to more significantly represent various different features of the image, including color, edge, and texture.

[0029] Further, in step (3.2), since the integrated circuit surface defect has diversity, a single feature descriptor cannot effectively represent the diversity defect features, so the feature descriptor of the patent is composed of curvature, function gradient, divergence of vector field, position spectrum, spectral value, spectral gradient, arc length, and angle feature. For any position of the image, x , y a feature vector L ( x , y ) can be defined. The curvature of the surface includes many kinds, including normal curvature, Gauss curvature, sectional curvature, mean curvature, and geodesic curvature. The commonly used curvature and sectional curvature formulas of the patent are as follows:

[0030] Let M , g be a Riemannian manifold, and ▽ be a Riemannian connection, then the curvature tensor R is: , (1)

[0031] Let be a two-dimensional subspace of the tangent space , and take a set of bases of it as X , Y , define the sectional curvature of as: , (2)

[0032] wherein, the definition of the sectional curvature is independent of the selection of the bases of .

[0033] The defect area of the integrated circuit generated in the process manufacturing process often accounts for a small proportion compared to the background area, therefore, the patent first performs block processing on the image, constructs the feature descriptor of the area containing the defect, and regards the remaining part not containing the defect as an invalid area. This can not only reduce the time consumption, but also effectively represent the defect characteristics of various types.

[0034] Further, in step (3.3), a feature vector L ( x , y is constructed by a plurality of feature descriptors.), and then the covariance between each feature vector and other feature vectors is calculated, i.e. the covariance matrix is constructed, which is designed as: , (3)

[0035] wherein, , the covariance matrix J reflects the statistical characteristics and spatial structure characteristics of the image, and the order of the feature descriptors is not determined when the covariance matrix is constructed, so the rotation and scaling of the image have good robustness. Finally, the constructed covariance matrix J is normalized.

[0036] Further, in the step (3.4), the sample size of the integrated circuit defect image is small; on the other hand, although the online collected image has been denoised, it may still have a small amount of noise, which will lead to the possibility of non-positive definite of the covariance matrix, thereby leading to unstable prediction results of the learning algorithm, so the non-positive definite problem of the covariance matrix is very important. That is, after the covariance matrix is constructed, it is necessary to determine whether the covariance matrix has a non-positive definite situation, if it exists, analyze the reason, and then use the corresponding method of adding a small positive value, regularization, principal component analysis (PCA) to effectively correct the covariance matrix to a positive definite symmetric matrix.

[0037] Further, in the step (3.5), the weight distance between two images I i and I j is: , (4)

[0038] wherein K represents the total number of image regions, is the distance of the corresponding I i th region of the image I j , and k represents the weight of the k th region.

[0039] Further, in the step (3.6), in the low-dimensional Lie group space, based on the geodesic distance, the Lie group kernel function between two images can be designed as: , (5)

[0040] wherein, the parameter is the scaling scale of the direct space.

[0041] Preferably, in the step (3.7), the Lie group kernel can be used in the classification method of k - nearest neighbor algorithm, support vector machine.

[0042] Compared with the prior art, the technical effects and advantages of the application are that the integrated circuit surface defect classification method based on differential manifold,

[0043] 1、 The application is helpful to improve the classification accuracy by pre-processing the integrated circuit image before classification;

[0044] 2、 The application maps the integrated circuit surface image to the Riemann manifold space, and the feature descriptor is composed of curvature, function gradient, divergence of vector field, position spectrum, spectral value, spectral gradient, arc length and angle feature, so as to better represent the intrinsic features of diversity defects;

[0045] 3、 The application adopts the idea of first dividing into blocks and then constructing the feature descriptor, which can effectively remove background interference;

[0046] 4、 The application effectively classifies various types of defects on the integrated circuit surface, which is helpful to more accurately judge whether the defects can be put into subsequent production;

[0047] 5、 The application has an important breakthrough in quality control of integrated circuits, improves the reliability in the integrated circuit manufacturing process, and guarantees the quality of the integrated circuit downstream application industry. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 It is a rapid classification system flowchart of the integrated circuit surface defect detection method of the application.

[0049] Figure 2 It is an image preprocessing system flowchart in the detection method of the application.

[0050] Figure 3 It is a diversity defect classification flowchart in the detection method of the application.

[0051] Figure 4 It is a working flowchart of the integrated circuit surface defect classification method based on differential manifold of the application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0053] Please refer to Figures 1-4 , the application provides a technical solution: an integrated circuit surface defect classification method based on differential manifold, specifically comprising the following steps:

[0054] Step (1) collect digital images of each part of the integrated circuit surface by microscopic imaging;

[0055] Step (2) pre-process the collected images;

[0056] Step (3) extract features and classify the pre-processed integrated circuit surface images to determine the defect type.

[0057] The collection process of step (1) needs the help of a high-power microscope. Usually, a complete integrated circuit needs to be divided into hundreds of regions for image collection, and there is always a small overlapping part between the images of each region. Therefore, it is often difficult to obtain a standard template.

[0058] In step (2), since the high-power microscope is used for online collection of industrial images, the complexity of the environment often leads to poor image quality, such as: introduction of various unknown noises in complex industrial environments, low image contrast, and shaking blur. No matter what causes the image quality to decline, it will affect the accuracy of subsequent defect classification. Pre-processing the collected images will help improve the accuracy of defect classification. The specific steps include:

[0059] Step (2.1) analyze what factors affect the quality of the collected integrated circuit surface images;

[0060] Step (2.2) denoise the collected integrated circuit images;

[0061] Step (2.3) improve the overall brightness of the denoised integrated circuit surface images and perform contrast enhancement processing.

[0062] Step (2.4) complete the pre-processing of the integrated circuit images.

[0063] In step (2.2), there are many types of noise introduced during online collection in industrial environments, such as: salt and pepper noise, Gaussian noise, and mixed noise. This patent performs noise reduction processing on various unknown noises, which includes:

[0064] Step (2.2.1) analyze the integrated circuit surface defect images using topology-related theories;

[0065] Step (2.2.2) select a homotopy mapping suitable for integrated circuit surface defect images to construct a new topological space;

[0066] Step (2.2.3) combine the Gaussian-Laplacian transform to design a filter template suitable for various unknown noises;

[0067] Step (2.2.4) achieves noise reduction processing for various unknown noises.

[0068] In step (2.3), the surface image of the integrated circuit is first subjected to contrast enhancement processing, and then zero-crossing processing is performed to ensure that the intrinsic features of the defects are extracted and to improve the accuracy of defect classification.

[0069] Step (3) specifically includes:

[0070] Step (3.1) Select a suitable mapping method for the preprocessed integrated circuit image and map the image to the Riemannian manifold space.

[0071] Step (3.2) Construct the feature descriptor of the image and obtain the feature vector. L ;

[0072] Step (3.3) constructs the covariance matrix using eigenvectors. J ;

[0073] Step (3.4) Correct the covariance matrix J A positive definite symmetric matrix of a Lie group manifold X ;

[0074] Step (3.5) Calculate the distance between positive definite symmetric matrices. D ;

[0075] Step (3.6) maps the positive definite symmetric matrix image descriptor to a low-dimensional Lie group space and designs the Lie group kernel function;

[0076] Step (3.7) achieves accurate classification of surface defect images of integrated circuits, forming a system's "defect type library".

[0077] In step (3.1), since the Euclidean space metric is difficult to use for nonlinear data, the preprocessed integrated circuit image is mapped to the Riemannian manifold space in order to more significantly represent the various features of the image, including color, edges, and texture.

[0078] In step (3.2), because the surface defects of integrated circuits are diverse, a single feature descriptor cannot effectively represent the diverse defect characteristics. Therefore, the feature descriptor is composed of curvature, function gradient, vector domain divergence, position spectrum, spectral value, spectral gradient, arc length, and angular features. For any position in the image ( x , y A feature vector can be defined. L ( x , y There are many types of surface curvature, including normal curvature, Gaussian curvature, cross-sectional curvature, mean curvature, and geodesic curvature. Commonly used formulas for curvature and cross-sectional curvature are as follows:

[0079] set up( M , g Let ∠ be a Riemannian manifold and ∠ be a Riemannian connection, then the curvature tensor... R for: (1)

[0080] set up For tangent space Let a two-dimensional subspace be a set of its basis. X , Y ,definition The curvature of the cross section is: (2)

[0081] in, The definition of cross-sectional curvature and The choice of the base is irrelevant.

[0082] In the manufacturing process of integrated circuits, the defect area is often relatively small compared to the background area. Therefore, this patent first divides the image into blocks, constructs feature descriptors for the defective areas, and treats the remaining non-defective parts as invalid areas. This reduces time consumption and effectively represents the characteristics of various types of defects.

[0083] In step (3.3), a feature vector is constructed using multiple feature descriptors. L ( x , y Then, by calculating the covariance between each eigenvector and other eigenvectors, the covariance matrix can be constructed, designed as follows: (3)

[0084] in, covariance matrix J This reflects the statistical and spatial structural characteristics of the image. Furthermore, since the order of feature descriptors is not determined when constructing the covariance matrix, it exhibits good robustness to image rotation and scaling. Finally, the constructed covariance matrix... J Perform a normalization operation.

[0085] In the step (3.4), due to the small sample size of the integrated circuit defect image, on the other hand, the image collected online may still have a small amount of noise after noise reduction processing, which will lead to the possibility of non-positive definite covariance matrix, resulting in unstable prediction results of learning algorithm, so the non-positive definite problem of the covariance matrix is very important. That is, after constructing the covariance matrix, it is necessary to judge whether the covariance matrix is non-positive definite, if it is, analyze the reason, and then use the corresponding method of adding a small positive value, regularization, principal component analysis (PCA) to effectively correct the covariance matrix to a positive definite symmetric matrix.

[0086] In the step (3.5), the weight distance between the two images I i and I j is: , (4)

[0087] wherein K represents the total number of image regions, is the distance of the corresponding first I i and I j region, and represents the weight of the first k region. k

[0088] In the step (3.6), in the low-dimensional Lie group space, based on the geodesic distance, the Lie group kernel function between the two images can be designed as: , (5)

[0089] wherein, the parameter is the scaling scale of the direct space.

[0090] In the step (3.7), the Lie group kernel can be used in the classification method of k -nearest neighbor algorithm, support vector machine.

[0091] The above embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the embodiments of the present application. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments are not required to be exhausted. Any modification, substitution and improvement made within the spirit and principle of the present application shall be included in the protection scope of the claims of the present application.​

Claims

1. A method for classifying surface defects of integrated circuits based on differential manifolds, characterized in that: Specifically, the following steps are included: Step (1). Acquire digital images of various parts of the integrated circuit surface using microscopic imaging; Step (2). Perform preprocessing operations on the acquired images; Step (3). Extract and classify features from the preprocessed integrated circuit surface image; The specific steps of step (3) include: Step (3.1) Select a suitable mapping method for the preprocessed integrated circuit image and map the image to the Riemannian manifold space; Step (3.2) Construct the feature descriptor of the image and obtain the feature vector. L ; Step (3.3) constructs the covariance matrix using eigenvectors. J ; Step (3.4) Correct the covariance matrix J A positive definite symmetric matrix of a Lie group manifold X ; Step (3.5) Calculate the distance between positive definite symmetric matrices. D ; Step (3.6) maps the positive definite symmetric matrix image descriptor to a low-dimensional Lie group space and designs the Lie group kernel function; Step (3.7) achieves accurate classification of surface defect images of integrated circuits, forming a system's "defect type library"; The feature descriptor in step (3.2) consists of curvature, function gradient, vector domain divergence, position spectrum, spectral value, spectral gradient, arc length and angle features. In step (3.2), the image is first divided into blocks, and feature descriptors are constructed for the regions containing defects. The remaining parts without defects are regarded as invalid regions, representing various types of defect characteristics. In step (3.3), eigenvectors are constructed using multiple feature descriptors, and then the covariance between each eigenvector and other eigenvectors is calculated, thus constructing a covariance matrix. In step (3.4), if the covariance matrix is ​​not positive definite, the reasons for the non-positive definiteness of the covariance matrix are first analyzed, and then corresponding methods such as adding small positive values, regularization, and principal component analysis (PCA) are used to correct the covariance matrix into a positive definite symmetric matrix of a Lie group manifold.

2. The method for classifying surface defects of integrated circuits based on differential manifolds according to claim 1, characterized in that: The acquisition process in step (1) requires the use of a high-powered microscope. A complete integrated circuit needs to be divided into hundreds of regions for image acquisition, and there are always slight overlaps between the images of each region.

3. The method for classifying surface defects of integrated circuits based on differential manifolds according to claim 1, characterized in that: The preprocessing process in step (2) addresses various unknown noises and low image contrast introduced during the online acquisition process in step (1), thus improving the accuracy of defect classification. Specific steps include: Step (2.1) analyzes the factors affecting the quality of surface images acquired online by integrated circuits; Step (2.2) involves noise reduction processing of the online acquired integrated circuit images; Step (2.3) improves the overall brightness and contrast enhancement of the noise-reduced integrated circuit surface image; Step (2.4) completes the preprocessing operation of the integrated circuit image.

4. The method for classifying surface defects of integrated circuits based on differential manifolds according to claim 3, characterized in that: In step (2.2), the types of noise introduced during the online acquisition of industrial environment data include the following specific steps: Step (2.2.1) Analyze the surface defect images of integrated circuits using relevant topological theories; Step (2.2.2) Select a homotopy mapping suitable for the surface defect image of the integrated circuit and construct a new topological space; Step (2.2.3) combines Gaussian-Laplace transform to design a filter template suitable for various unknown noises; Step (2.2.4) performs noise reduction processing on various unknown noises.

5. The method for classifying surface defects of integrated circuits based on differential manifolds according to claim 3, characterized in that: In step (2.3), the surface image of the integrated circuit is first subjected to contrast enhancement processing and then zero-crossing processing is performed to extract the intrinsic features of the defects.

6. The method for classifying surface defects of integrated circuits based on differential manifolds according to claim 1, characterized in that: In step (3.5), the weighted distance between the two images uses a Riemannian metric. In step (3.6), a Lie group kernel function is designed between the two images based on geodesic distance in a low-dimensional Lie group space. In step (3.7), the Lie group kernel is used for classification using the following methods: k -Nearest neighbor algorithm, support vector machine.

Images