A method and system for improving the defect detection sensitivity of semiconductor wafers

Abstract

The present invention discloses a method and system for improving the defect detection sensitivity of a semiconductor wafer, belonging to the technical field of wafer defect detection. The key points of its technical solution include: obtaining an undersampled Fourier coefficient vector of the semiconductor wafer to be detected based on the single-pixel sensing method; reconstructing the undersampled Fourier coefficient vector according to the compressive sensing algorithm to obtain a target Fourier coefficient vector; obtaining the spectrum of the semiconductor wafer to be detected according to the target Fourier coefficient vector; and performing defect detection on the semiconductor wafer to be detected according to the spectrum. The present invention performs defect detection based on the single-pixel sensing method. Compared with the method of using an array detection device such as a CCD to collect optical signals in the traditional optical imaging technology, this method has a wide spectral response range, a fast sampling speed, and can maintain a high detection sensitivity in weak light scenarios and long-distance scenarios. At the same time, the present invention uses spectral features to perform defect detection on the semiconductor wafer to be detected, improving the detection speed.

Description

Technical Field

[0001] The present invention relates to the technical field of wafer defect detection, and more specifically, to a method and system for improving the sensitivity of semiconductor wafer defect detection. Background Art

[0002] The technology of semiconductor wafer surface defect detection is one of the core technologies in chip production and manufacturing. Initially, the detection of wafer surface defects relied on human vision. With the development of methods and theories such as image processing and machine learning, the method of semiconductor wafer surface defect detection based on machine vision and artificial intelligence is gradually replacing the traditional technology of human eye detection. However, most of these methods based on artificial intelligence and machine vision are based on the method and mode of imaging first and then processing, that is, first use an array imaging device to image the wafer surface, and then use relevant algorithms or models to process the acquired image.

[0003] For example, the patent application with the publication number CN117423639A provides a wafer surface modeling defect detection system. This application uses a CCD camera to photograph the surface of the wafer to be detected, determines the image signal of the wafer surface, and determines the wafer surface feature data based on machine vision through the acquired signal information. Then, the wafer surface feature data is input into the wafer surface detection model to detect the defects on the wafer surface.

[0004] However, the data used by this system is large, which causes great storage and transmission bandwidth pressure on the observation system, easily causes the loss of high-dimensional information such as the frequency, phase, and polarization of optical signals, and the CCD camera has strict environmental requirements and is not suitable for collecting images in low-light scenes and long-distance scenes. Therefore, there are deficiencies in the existing technology. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a method and system for improving the sensitivity of semiconductor wafer defect detection. Based on the characteristics of the single-pixel sensing method, the present invention can quickly and accurately collect information in low-light scenes and long-distance scenes, and combine spectral features to quickly detect the defects of the semiconductor wafer to be detected, improving the detection efficiency and sensitivity.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] The present invention provides a method for improving the sensitivity of semiconductor wafer defect detection, including:

[0008] Obtaining an undersampled Fourier coefficient vector of the semiconductor wafer to be detected based on the single-pixel sensing method;

[0009] Reconstructing the undersampled Fourier coefficient vector according to the compressive sensing algorithm to obtain a target Fourier coefficient vector;

[0010] Obtaining a frequency spectrum of the semiconductor wafer to be inspected according to the target Fourier coefficient vector;

[0011] Defect detection is performed on the semiconductor wafer to be inspected according to the frequency spectrum.

[0012] As a further improvement of the present invention, the method of obtaining an undersampled Fourier coefficient vector of a semiconductor wafer to be inspected based on a single-pixel perception method includes:

[0013] Constructing color-coded patterns based on Bayer filters and Fourier transforms;

[0014] Determining sampling points on the surface of the semiconductor wafer to be inspected based on a circular sampling method and spectrum energy parameters;

[0015] Projecting the color coding pattern onto each of the sampling points using a four-step phase shift method to obtain a Fourier coefficient corresponding to each of the sampling points;

[0016] The Fourier coefficients corresponding to each of the sampling points are combined to obtain an under-sampling Fourier coefficient vector of the semiconductor wafer to be inspected.

[0017] As a further improvement of the present invention, the method of constructing a color coding pattern based on a Bayer filter and Fourier transform includes:

[0018] Decomposing the Bayer filter into three colors: red, green, and blue to obtain red, green, and blue coding templates;

[0019] generating a grayscale basis pattern according to Fourier spectrum information;

[0020] Multiplying the red, green and blue coding templates by the grayscale basis pattern respectively to obtain three single-spectrum Fourier basis patterns;

[0021] The three single-spectral Fourier basis patterns are combined to obtain the color-coded pattern.

[0022] As a further improvement of the present invention, the method of determining sampling points on the surface of the semiconductor wafer to be inspected based on the circular sampling method and the spectrum energy parameter includes:

[0023] Based on the circular sampling method, a first sampling path is determined with the surface center of the semiconductor wafer to be inspected as the starting point according to a preset sampling angle and sampling radius, wherein the first sampling path includes a plurality of sections;

[0024] For two adjacent sections in the first sampling path, update the second section according to the spectral energy parameter of the first section, and traverse the first sampling path step by step with one section as the step length to obtain the second sampling path. The first section is the section closer to the starting point among the two adjacent sections, and the second section is the section farther from the starting point among the two adjacent sections;

[0025] Determine multiple sampling points on the second sampling path based on equidistant sampling.

[0026] As a further improvement of the present invention, the updating of the second section according to the spectral energy parameter of the first section includes:

[0027] Select N seed points in the first section and calculate the spectral energy E of each seed point i , where i = 1,..., N, and N is a positive integer greater than 1;

[0028] According to the spectral energy E i , obtain the spectral energy parameter of the first section where μ is a compensation coefficient;

[0029] Compare the spectral energy parameter ξ with a preset threshold, and update the second section according to the comparison result.

[0030] As a further improvement of the present invention, the reconstruction of the undersampled Fourier coefficient vector according to the compressive sensing algorithm to obtain the target Fourier coefficient vector includes:

[0031] Select a measurement matrix and a sparse matrix according to the characteristics of the undersampled Fourier coefficient vector;

[0032] Calculate a sensing matrix according to the measurement matrix and the sparse matrix;

[0033] Reconstruct the undersampled Fourier coefficient vector based on the sensing matrix to obtain the target Fourier coefficient vector.

[0034] As a further improvement of the present invention, the reconstruction of the undersampled Fourier coefficient vector based on the sensing matrix to obtain the target Fourier coefficient vector includes:

[0035] Construct a reconstruction model according to the smoothing function F σ (u): where the parameter σ represents the approximation degree of the smoothing function to the L0 norm, φ is the sensing matrix, u is the undersampled Fourier coefficient vector, and v is the target Fourier coefficient vector;

[0036] Set the value of the parameter σ to be σ ∈ (σ1, σ2,..., σz ) where σ1 > σ2 > … > σ z , where z is the number of values of σ;

[0037] Optimize and solve the smoothing function corresponding to each said value to obtain the solution corresponding to each value;

[0038] Select the global optimal solution from the solutions corresponding to each said value, and reconstruct the undersampled Fourier coefficient vector according to the global optimal solution to obtain the target Fourier coefficient vector.

[0039] As a further improvement of the present invention, the defect detection of the to-be-detected semiconductor wafer according to the spectrum includes:

[0040] Calculate the spectrum similarity coefficient between the spectrum and a preset spectrum, and the preset spectrum is obtained according to a wafer template without defects;

[0041] Select multiple feature points in the spectrum, and calculate the correlation coefficient between each said feature point and the corresponding preset feature point, and the preset feature point is obtained according to the preset spectrum;

[0042] Perform defect detection on the to-be-detected semiconductor wafer according to the spectrum similarity coefficient and the correlation coefficient.

[0043] As a further improvement of the present invention, the spectrum similarity coefficient where f(n) represents the spectrum, g(n) represents the preset spectrum, and n represents the frequency.

[0044] The present invention provides a system for improving the defect detection sensitivity of a semiconductor wafer, including:

[0045] Acquisition module: Obtain the undersampled Fourier coefficient vector of the to-be-detected semiconductor wafer based on the single-pixel sensing method;

[0046] Reconstruction module: Reconstruct the undersampled Fourier coefficient vector according to the compressive sensing algorithm to obtain the target Fourier coefficient vector;

[0047] Calculation module: Obtain the spectrum of the to-be-detected semiconductor wafer according to the target Fourier coefficient vector;

[0048] Detection module: Perform defect detection on the to-be-detected semiconductor wafer according to the spectrum.

[0049] Based on the single-pixel sensing method and the compressive sensing algorithm, the present invention uses fewer sampling points to obtain the spectrum of the to-be-detected semiconductor wafer, and performs defect detection on the wafer based on the spectrum characteristics. The present invention has a wide range of applicable scenarios, requires less data volume, and can quickly and accurately achieve defect detection. Description of the Drawings

[0050] Figure 1 This is a flowchart of the steps of a method for improving the defect detection sensitivity of a semiconductor wafer according to the present invention;

[0051] Figure 2 This is a schematic diagram of an existing circular sampling method;

[0052] Figure 3 This is a flowchart of the steps of section update in the present invention;

[0053] Figure 4 This is a schematic structural diagram of a system for improving the defect detection sensitivity of a semiconductor wafer according to the present invention. Detailed implementation manners

[0054] The technical solution of the present invention will be described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention.

[0055] The term "and / or" in the following text is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " generally represents an "or" relationship between the associated objects before and after.

[0056] As Figure 1 shown, an embodiment of the present application provides a method for improving the defect detection sensitivity of a semiconductor wafer, including:

[0057] Obtaining an undersampled Fourier coefficient vector of a semiconductor wafer to be detected based on a single-pixel sensing method;

[0058] Reconstructing the undersampled Fourier coefficient vector according to a compressive sensing algorithm to obtain a target Fourier coefficient vector;

[0059] Obtaining the spectrum of the semiconductor wafer to be detected according to the target Fourier coefficient vector;

[0060] Performing defect detection on the semiconductor wafer to be detected according to the spectrum.

[0061] Traditional methods based on artificial intelligence and machine vision mostly use array imaging devices (such as CCD and CMOS) to image the surface of the wafer, and then use relevant algorithms or models to process the obtained images. However, this method has high requirements for imaging quality, and the imaging quality of the array imaging device is greatly affected by the environment. When the image quality is poor, the detection sensitivity and accuracy will be greatly affected.

[0062] Therefore, compared with traditional methods, this embodiment combines the single-pixel sensing method. Data is collected by a single-pixel detector. Compared with array imaging devices, the single-pixel detector has a fast sampling speed and high sensitivity, and can be applied to various scenarios. Then, combined with the compressive sensing algorithm, the number of samplings is reduced, and finally, spectral features are used for defect detection. Compared with the prior art, this embodiment can improve the sensitivity and efficiency of semiconductor wafer defect detection in low-light scenarios or long-distance scenarios.

[0063] Furthermore, this embodiment provides a method for obtaining an undersampled Fourier coefficient vector of a semiconductor wafer to be detected based on the single-pixel sensing method, including:

[0064] Construct a color-encoded pattern based on a Bayer filter and Fourier transform;

[0065] Determine sampling points on the surface of the semiconductor wafer to be detected based on the circular sampling method and spectral energy parameters;

[0066] Use the four-step phase-shift method to project the color-encoded pattern onto each sampling point to obtain the Fourier coefficient corresponding to each sampling point;

[0067] Combine the Fourier coefficients corresponding to each sampling point to obtain the undersampled Fourier coefficient vector of the semiconductor wafer to be detected.

[0068] The core of the single-pixel sensing method is to modulate the encoded pattern with the target to be detected, realize the measurement of the Fourier spectrum of the target to be detected, and analyze based on the measurement results.

[0069] The color-encoded pattern provided in this embodiment can alleviate the reflection or interference of light by the thin film structure on the surface of the semiconductor wafer to be detected compared with the traditional gray-encoded pattern, and improve the detection accuracy.

[0070] Furthermore, this embodiment provides a step for constructing a color-encoded pattern based on a Bayer filter and Fourier transform. The step includes:

[0071] Decompose the Bayer filter into three colors: red, green, and blue to obtain red, green, and blue encoding templates;

[0072] Generate a gray-scale base pattern according to Fourier spectrum information;

[0073] Multiply the red, green, and blue encoding templates with the gray-scale base pattern respectively to obtain three single-spectrum Fourier base patterns;

[0074] Combine the three single-spectrum Fourier base patterns to obtain a color-encoded pattern.

[0075] Specifically, a Bayer filter refers to adding a filter with a periodic color distribution on the surface of a monochrome image sensor, and this filter is composed of smaller sub - filters arranged. Each period of the Bayer filter consists of 2×2 sub - filters, including three - colored sub - filters, namely red, green, and blue.

[0076] If the Bayer filter is decomposed according to the three - colored sub - filters, encoding templates for red, green, and blue can be obtained, denoted as M R , M G and M B , and the expressions are:

[0077]

[0078] Then, a gray - scale base pattern containing Fourier spectrum information is generated as:

[0079]

[0080] where a represents the DC component, b represents the coefficient modulating the contrast, x and y represent two - dimensional Cartesian coordinates, f x and f y represent the frequencies in the x and y directions, represents the phase. Since the four - step phase - shifting method is adopted in this embodiment, four gray - scale base patterns are obtained.

[0081] The encoding templates for red, green, and blue are respectively multiplied point - by - point with the gray - scale base pattern to obtain three single - spectrum Fourier base patterns, and the three single - spectrum Fourier base patterns are combined to obtain a color - encoded pattern as:

[0082]

[0083] where c represents color, and R, G, B respectively represent red, green, and blue. Similarly, since there are four color - encoded patterns.

[0084] Furthermore, this embodiment provides a step of using the four - step phase - shifting method to project the color - encoded pattern onto each sampling point to obtain the Fourier coefficient corresponding to each sampling point. Specifically, it includes modulating the color - encoded pattern with the semiconductor wafer to be detected to obtain the expected light - intensity response value as:

[0085]

[0086] where S is the field of view size, i.e., the size of the projected color-coded pattern, and R(x, y) represents the reflected light intensity on the surface of the semiconductor wafer to be detected. However, considering the influence of detector noise and environmental noise, the actually collected light intensity response value should be expressed as:

[0087]

[0088] where D n represents environmental noise, and γ represents the gain of the single-pixel detector.

[0089] According to the light intensity response values at different phases, the Fourier coefficients C(f x , f y ) are calculated as:

[0090]

[0091] where j is the imaginary unit, and D0(f x , f y ), D π (f x , f y ) respectively represent the light intensity response values at

[0092] By performing the above steps at each sampling point, the Fourier coefficients corresponding to each sampling point can be obtained.

[0093] Furthermore, this embodiment provides a method for determining sampling points on the surface of a semiconductor wafer to be detected based on a circular sampling method and spectral energy parameters, which can ensure detection accuracy while reducing the number of sampling points. Specifically, it includes:

[0094] Based on the circular sampling method, starting from the center of the surface of the semiconductor wafer to be detected, a first sampling path is determined according to a preset sampling angle and sampling radius. The first sampling path includes multiple segments;

[0095] For two adjacent segments in the first sampling path, the second segment is updated according to the spectral energy parameter of the first segment. Taking one segment as the step size, the first sampling path is traversed in sequence to obtain a second sampling path. The first segment is the segment closer to the starting point among the two adjacent segments, and the second segment is the segment farther from the starting point among the two adjacent segments;

[0096] Based on equidistant sampling, multiple sampling points are determined on the second sampling path.

[0097] As Figure 2As shown, the existing circular sampling method expands outward in a spiral trajectory in the direction of the arrow according to a certain radius increment, gradually increasing the sampling radius. This spiral method ensures progressive coverage from low frequency to high frequency and only samples the points on the spiral path. However, this method usually performs sampling step by step based on a preset radius increment, which has a large degree of subjectivity. If the radius increment is large, it will lead to too few sampling points and affect the detection result. If the radius increment is small, it will lead to too many sampling points and a large amount of data needs to be processed. Therefore, this embodiment further provides a method that can adaptively adjust the radius increment based on the spectral energy parameter to ensure the sampling efficiency.

[0098] Specifically, first determine the first sampling path based on the circular sampling method, that is, Figure 2 the path shown in, and call each ring in the first sampling path a section. Then, perform the update step, as Figure 3 shown, this step includes:

[0099] Select N seed points in the first section and calculate the spectral energy E i of each seed point, where i = 1,..., N, and N is a positive integer greater than 1;

[0100] According to the spectral energy E i , obtain the spectral energy parameter of the first section, where μ is a compensation coefficient;

[0101] Compare the spectral energy parameter ξ with a preset threshold, and update the second section according to the comparison result.

[0102] The larger the value of the spectral energy parameter ξ, the more concentrated the energy distribution of this section. μ is used to compensate for the sampling error.

[0103] Exemplarily, assume that the preset threshold is k1, the preset radius increment is r1, and the ratio obtained by comparing the spectral energy parameter ξ with the preset threshold is The adjusted radius increment is Then, update the second section according to the adjusted radius increment.

[0104] This embodiment determines the sampling points based on the spectral energy parameter. That is, if the value of the spectral energy parameter ξ of the first path is relatively high, the radius increment of the second path is reduced. If the value of the spectral energy parameter ξ of the first path is relatively low, the radius increment of the second path is increased, and sampling points are evenly distributed at a preset sampling rate in the second sampling path. Through the above method, the acquisition efficiency of effective information can be ensured under the premise of reducing the data processing volume.

[0105] Further, after combining the Fourier coefficients corresponding to each sampling point to obtain an undersampled Fourier coefficient vector, this embodiment provides a method for reconstructing the undersampled Fourier coefficient vector, including:

[0106] Select a measurement matrix and a sparse matrix according to the characteristics of the undersampled Fourier coefficient vector;

[0107] Calculate a sensing matrix according to the measurement matrix and the sparse matrix;

[0108] Reconstruct the undersampled Fourier coefficient vector based on the sensing matrix to obtain a target Fourier coefficient vector.

[0109] Further, the step of reconstructing the undersampled Fourier coefficient vector based on the sensing matrix to obtain a target Fourier coefficient vector includes:

[0110] Construct a reconstruction model according to the smoothing function F σ (u): where the parameter σ represents the approximation degree of the smoothing function to the L0 norm, φ is the sensing matrix, u is the undersampled Fourier coefficient vector, and v is the target Fourier coefficient vector;

[0111] Set the value of the parameter σ as σ ∈ (σ1, σ2,..., σ z ), where σ1 > σ2 >... > σ z , where z is the number of values of σ;

[0112] Optimize and solve the smoothing function corresponding to each value to obtain the solution corresponding to each value;

[0113] Select the global optimal solution from the solutions corresponding to each value, and reconstruct the undersampled Fourier coefficient vector according to the global optimal solution to obtain a target Fourier coefficient vector.

[0114] Specifically, sparsely represent the target Fourier coefficient vector v as s = ψv, where s is the sparsely represented target Fourier coefficient vector and ψ is the sparse matrix. Design a matrix Φ that is not related to the sparse matrix as the measurement matrix. After reducing the dimension of v, the undersampled Fourier coefficient vector u can be obtained, denoted as u = Φs = Φψv. Let Φψ = φ, where φ is the sensing matrix, and obtain u = φv.

[0115] Since the dimension of u is less than the dimension of v, when directly solving v with u, a definite solution cannot be obtained. Therefore, optimization theory is usually used for reconstruction and solution:

[0116]

[0117] Where ||v||0 represents the L0 norm of v. However, the sparse solution based on the L0 norm is a non-convex optimization problem, which is difficult to solve. Moreover, since the L0 norm is a non-smooth function, the algorithm directly based on the L0 norm requires combinatorial optimization, and the computational complexity increases exponentially. Therefore, in this embodiment, a smoothing function is used to solve the problem, that is, the smoothing function is made to approach the L0 norm, and the problem of solving the L0 norm is transformed into the optimization problem of the smoothing function, which can accelerate the operation rate and reduce the computational complexity.

[0118] Specifically, since the L0 norm of v is the non-zero coefficients in v, a function h(v k ) is defined:

[0119]

[0120] where v k is the k-th element in the target Fourier coefficient vector v;

[0121] Then the L0 norm of v can be expressed as: where K is the dimension of v. It is found that the discontinuous function h(v k ) will lead to the discontinuity of the L0 norm of v. Therefore, an approximate smoothing function f σ (v k ) is used to replace h(v k ) to achieve the smooth estimation of the L0 norm.

[0122] If the continuous function satisfies:

[0123]

[0124] Then this function can replace h(v k ) to achieve the purpose of approximating the L0 norm. At the same time, combined with h(v k ), we can get:

[0125]

[0126] Let the function Then:

[0127]

[0128] According to the definition of the L0 norm, the L0 norm of v can be expressed as:

[0129]

[0130] Then the solution of can be transformed into the solution of

[0131] Since the smaller the σ, the better the approximation of the corresponding function to the L0 norm. However, at the same time, as σ decreases, the function curve becomes less and less smooth. Therefore, in order to obtain the global optimal solution, a descending sequence (σ1, σ2,..., σ z ) can be set for σ. By solving the solutions corresponding to each value and selecting a global optimal solution from all the solutions as the final solution, it can satisfy both function smoothness and σ approaching 0 to the greatest extent.

[0132] In this embodiment, by setting a smoothing function, the calculation speed is accelerated while the calculation difficulty is reduced, and finally the reconstructed target Fourier coefficient vector is obtained, that is, all the Fourier coefficients of the semiconductor wafer to be detected are obtained.

[0133] Furthermore, according to all the Fourier coefficients, the complete spectrum T(f x , f y ) of the semiconductor wafer to be detected is:

[0134]

[0135] where F -1 represents the inverse Fourier transform.

[0136] Furthermore, this embodiment provides a method for detecting wafer defects according to the above spectrum, including:

[0137] Calculating the spectrum similarity coefficient between the spectrum and a preset spectrum, where the preset spectrum is obtained according to a defect-free wafer template;

[0138] Selecting multiple feature points in the spectrum and calculating the correlation coefficient between each feature point and the corresponding preset feature point, where the preset feature points are obtained according to the preset spectrum;

[0139] Performing defect detection on the semiconductor wafer to be detected according to the spectrum similarity coefficient and the correlation coefficient.

[0140] where the spectrum similarity coefficient where f(n) represents the spectrum, g(n) represents the preset spectrum, and n represents the frequency.

[0141] When there are defects on the surface of the semiconductor wafer to be detected, its corresponding spectrum will also change. Therefore, according to the similarity coefficient between the spectrum and the preset spectrum, it can be preliminarily judged whether the wafer contains defects.

[0142] For wafers with a high probability of containing defects, characteristic points can be further selected in the spectrum for detection. The low-frequency part of the spectrum corresponds to characteristic information such as contours and edges, and the high-frequency part corresponds to detailed information. Characteristic points can be set in different parts for analysis according to the detection needs, and the correlation coefficient can be selected as the Pearson correlation coefficient, etc. According to the results obtained from the spectrum similarity coefficient and the correlation coefficient, the general type of wafer defects can be further analyzed.

[0143] A method for improving the defect detection sensitivity of semiconductor wafers provided by an embodiment of the present application expands the detection scenario and enhances the detection sensitivity through the single-pixel sensing method, combines the adaptive sampling method and the compressive sensing algorithm to select sampling points, reconstructs the complete data with less sampling information, optimizes the sampling steps, improves the acquisition efficiency, and finally performs defect detection through spectral features. Compared with the detection method based on image features, the detection speed is accelerated.

[0144] Further, as Figure 4 shown, an embodiment of the present application provides a system for improving the defect detection sensitivity of semiconductor wafers, including:

[0145] Acquisition module: Obtain the undersampled Fourier coefficient vector of the semiconductor wafer to be detected based on the single-pixel sensing method;

[0146] Reconstruction module: Reconstruct the undersampled Fourier coefficient vector according to the compressive sensing algorithm to obtain the target Fourier coefficient vector;

[0147] Calculation module: Obtain the spectrum of the semiconductor wafer to be detected according to the target Fourier coefficient vector;

[0148] Detection module: Perform defect detection on the semiconductor wafer to be detected according to the spectrum.

[0149] Among them, the acquisition module is specifically a single-pixel detector, the reconstruction module, the calculation module and the detection module are located in the data processing device, and the single-pixel detector is communicatively connected to the data processing device.

[0150] A method and system for improving the defect detection sensitivity of semiconductor wafers provided by an embodiment of the present application combine the adaptive sampling and compressive sensing steps on the basis of the single-pixel sensing method, and combine spectral features to detect wafer defects, with a wide range of applicable scenarios, fast detection speed and high sensitivity.

[0151] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0152] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0153] These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured product including instruction means, and the instruction means implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in [[ID=]18] Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0155] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements should also be regarded as within the protection scope of the present invention.

Claims

1. A method for improving the sensitivity of semiconductor wafer defect detection, characterized in that: include: Obtaining an undersampled Fourier coefficient vector of the semiconductor wafer to be inspected based on a single-pixel perception method; reconstructing the undersampled Fourier coefficient vector according to a compressed sensing algorithm to obtain a target Fourier coefficient vector; Obtaining a frequency spectrum of the semiconductor wafer to be inspected according to the target Fourier coefficient vector; performing defect detection on the semiconductor wafer to be inspected according to the frequency spectrum; The method of obtaining the undersampled Fourier coefficient vector of the semiconductor wafer to be inspected based on the single-pixel perception method includes: Constructing color-coded patterns based on Bayer filters and Fourier transforms; Determining sampling points on the surface of the semiconductor wafer to be inspected based on a circular sampling method and spectrum energy parameters; Projecting the color coding pattern onto each of the sampling points using a four-step phase shift method to obtain a Fourier coefficient corresponding to each of the sampling points; Combining the Fourier coefficients corresponding to each of the sampling points to obtain an under-sampling Fourier coefficient vector of the semiconductor wafer to be inspected; The step of reconstructing the undersampled Fourier coefficient vector according to a compressed sensing algorithm to obtain a target Fourier coefficient vector includes: Selecting a measurement matrix and a sparse matrix based on the characteristics of the undersampled Fourier coefficient vector; Calculating a perception matrix based on the measurement matrix and the sparse matrix; Reconstructing the undersampled Fourier coefficient vector based on the perception matrix to obtain the target Fourier coefficient vector; The reconstructing the undersampled Fourier coefficient vector based on the perception matrix to obtain the target Fourier coefficient vector includes: According to the smoothing function F σ (u) Constructing the reconstruction model: Wherein, parameter σ represents the approximation degree of the smoothing function to the L0 norm, φ is the perception matrix, u is the undersampled Fourier coefficient vector, and v is the target Fourier coefficient vector; Set the value of the parameter σ to σ∈(σ1,σ2,…,σ z ), where σ1>σ2>…>σ z , where z is the number of values of σ; Optimizing and solving the smooth function corresponding to each of the values to obtain a solution corresponding to each value; Selecting a global optimal solution from the solutions corresponding to each value, and reconstructing the undersampled Fourier coefficient vector according to the global optimal solution to obtain the target Fourier coefficient vector; The performing defect detection on the semiconductor wafer to be inspected according to the frequency spectrum includes: Calculating a spectrum similarity coefficient between the spectrum and a preset spectrum, the preset spectrum being obtained based on a defect-free wafer template; Selecting a plurality of feature points in the frequency spectrum and calculating a correlation coefficient between each feature point and a corresponding preset feature point, wherein the preset feature point is obtained according to the preset frequency spectrum; Defect detection is performed on the semiconductor wafer to be inspected according to the frequency spectrum similarity coefficient and the correlation coefficient.

2. The method for improving semiconductor wafer defect detection sensitivity according to claim 1, wherein: The method of constructing a color coding pattern based on a Bayer filter and a Fourier transform includes: Decomposing the Bayer filter into three colors: red, green, and blue to obtain red, green, and blue coding templates; generating a grayscale basis pattern according to Fourier spectrum information; Multiplying the red, green and blue coding templates by the grayscale basis pattern respectively to obtain three single-spectrum Fourier basis patterns; The three single-spectral Fourier basis patterns are combined to obtain the color-coded pattern.

3. The method for improving semiconductor wafer defect detection sensitivity according to claim 1, wherein: The method of determining sampling points on the surface of the semiconductor wafer to be inspected based on the circular sampling method and the spectrum energy parameter includes: Based on the circular sampling method, a first sampling path is determined with the surface center of the semiconductor wafer to be inspected as the starting point according to a preset sampling angle and sampling radius, wherein the first sampling path includes a plurality of sections; For two adjacent road sections in the first sampling path, updating the second road section according to the spectrum energy parameter of the first road section, traversing the first sampling path in sequence with a step size of one road section to obtain a second sampling path, where the first road section is the road section closer to the starting point of the two adjacent road sections, and the second road section is the road section farther from the starting point of the two adjacent road sections; A plurality of sampling points are determined on the second sampling path based on equidistant sampling.

4. The method for improving semiconductor wafer defect detection sensitivity according to claim 3, wherein: The updating of the second section according to the spectrum energy parameter of the first section includes: Select N seed points in the first section and calculate the spectrum energy E of each seed point i , i=1,…,N, N is a positive integer greater than 1; According to the spectrum energy E i , obtain the spectrum energy parameter of the first section Where μ is the compensation coefficient; The spectrum energy parameter ξ is compared with a preset threshold, and the second road section is updated according to the comparison result.

5. The method for improving semiconductor wafer defect detection sensitivity according to claim 1, wherein: The spectrum similarity coefficient Wherein f(n) represents the frequency spectrum, g(n) represents the preset frequency spectrum, and n represents the frequency.

6. A system for improving semiconductor wafer defect detection sensitivity, used to implement the method for improving semiconductor wafer defect detection sensitivity according to any one of claims 1 to 5, characterized in that: include: Acquisition module: Acquires the undersampled Fourier coefficient vector of the semiconductor wafer to be inspected based on the single-pixel perception method; Reconstruction module: reconstruct the undersampled Fourier coefficient vector according to the compressed sensing algorithm to obtain the target Fourier coefficient vector; A calculation module: obtaining a frequency spectrum of the semiconductor wafer to be inspected according to the target Fourier coefficient vector; Detection module: performs defect detection on the semiconductor wafer to be inspected according to the frequency spectrum.

Images