A method, device and readable storage medium for detecting foreign matter defects in microelectronic products
The GMC sparse noise denoising model processed the foreign object defect collision signal of microelectronics, which solved the problem of low detection accuracy, achieved efficient identification of foreign object particle collision signal, and improved the reliability of detection results.
Patent Information
- Application Number
- CN202410986735.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-23
AI Technical Summary
The detection results of existing microelectronic products have low accuracy, especially under the influence of environmental noise, which is difficult to accurately identify foreign object particle collision signals.
The GMC sparse noise denoising model is adopted to segment the collision signal of foreign matter defects, singular value decomposition and signal kurtitude calculation, and a sparse encoding matrix is constructed to remove environmental noise, enhance signal amplitude, and improve detection accuracy.
Effectively remove environmental noise, enhance signal amplitude, improve the accuracy of foreign object defect detection for microelectronics products, and ensure the reliability of detection results.
Smart Images

Figure CN118520227B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of microelectronic testing technology, and in particular to a method and device for detecting foreign matter defects in microelectronic products, and a computer-readable storage medium. Background Art
[0002] Integrated circuits are an important part of the development of science and technology and the foundation for all walks of life to achieve intelligence and digitalization. Electronic packaging, as an indispensable process in the manufacturing process of integrated circuits, has an important impact on the cost, performance and reliability of integrated circuit products. However, more than 25% of the failures in integrated circuit products are caused by packaging. This is because the electronic packaging process is relatively complex and involves a variety of materials and processes. Foreign particles such as bonding wire tails, conductive glue particles, ceramic particles, broken silicon chips, and dust flying in the environment that have not been thoroughly cleaned are easily introduced into the integrated circuit cavity. When the product is in an environment of clustered impact vibration or high-speed disguised motion, movable foreign particles are easily activated and freed, and move irregularly in the cavity, and even collide with the core cavity wall and internal chips and circuits, which may cause deformation of bonding wires, short circuits, and circuit function failures, seriously affecting the reliability of microelectronic products. Especially in the field of aerospace, the low reliability of microelectronic products will not only cause huge economic losses, but may even threaten the lives of astronauts. Therefore, how to effectively detect foreign body defects in microelectronic products after packaging is the key to ensuring product reliability.
[0003] Common foreign body detection methods include pre-capping microscopic inspection, post-capping X-ray inspection, Matera test and Particle Impact Noise Detection (PIND). The microscopic inspection method relies on manual use of a microscope to perform appearance inspection on raw material chips, and has low inspection efficiency and accuracy. In addition, this method may introduce foreign particles when capping after the inspection is completed, thereby causing foreign body defects. Although the X-ray inspection method can non-destructively inspect packaged products, its inspection accuracy is affected by the resolution of the X-ray equipment, and the X-ray inspection equipment is expensive and requires regular maintenance and calibration, which makes the inspection cost high. The Matera inspection method involves complex mechanical test conditions and data analysis processes, and requires a long inspection time, resulting in low inspection efficiency. Therefore, the high-efficiency and low-cost PIND method is often used for foreign body defect detection.
[0004] The PIND method uses a vibration table to generate a certain amount of mechanical shock and vibration. The shock loosens the foreign body defect bound in the cavity of the device under test, and the vibration causes the foreign body defect to slide and collide in the cavity, generating a collision signal. After the sensor collects the collision signal, it converts it into a voltage output and displays it to determine whether the device under test has a foreign body defect. However, with the continuous improvement of the process level, the size and quality of foreign body defects can be controlled within a smaller range, which will result in the foreign body collision signal being submerged in the environmental noise even under the best test conditions, greatly affecting the accuracy of foreign body defect detection. Summary of the invention
[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problem of low accuracy of detection results of the microelectronic product foreign body defect detection method in the prior art.
[0006] In order to solve the above technical problems, the present invention provides a method for detecting foreign matter defects in microelectronic products, comprising:
[0007] The foreign body defect collision signal of the microelectronic product to be detected is obtained, and the weights of the sparse coefficients in the penalty term of the GMC sparse denoising model of the foreign body defect collision signal are set to obtain the target penalty term, which specifically includes:
[0008] The foreign body defect collision signal is segmented to obtain Dimensional foreign body defect collision signal matrix;
[0009] Perform singular value decomposition on the foreign body defect collision signal matrix to obtain singular values; taking each column element in the foreign body defect collision signal matrix as a sub-signal to obtain m sub-signals, and calculating the signal kurtosis of each sub-signal respectively;
[0010] Calculate the The singular values and The square sum of the signal kurtosis of the sub-signals is calculated, and the reciprocal of the square sum is used as the sparse coding matrix of the foreign body defect collision signal. Line The weight of the column sparse coefficient; where, , ;
[0011] Based on the fidelity term and the target penalty term, a target GMC sparse denoising model is obtained; and the target GMC sparse denoising model is iteratively solved to obtain a sparse coding matrix of the foreign object defect collision signal;
[0012] The foreign body defect collision signal is sparsely represented by using the sparse coding matrix to obtain a target foreign body defect collision signal, thereby performing foreign body defect detection on the microelectronic product to be detected.
[0013] Preferably, the sparse coding matrix of the foreign body defect collision signal Line The calculation formula for the weight of the column sparse coefficient is:
[0014] ,
[0015] in, The sparse coding matrix representing the collision signal of foreign body defects Line The weight of the column sparse coefficient; Indicates The signal kurtosis of each sub-signal; Indicates singular values;
[0016] ,
[0017] in, Indicates The first data points; Indicates The number of data points in each sub-signal; Indicates The mean of all data points in the subsignal.
[0018] Preferably, the target GMC sparse denoising model is expressed as:
[0019] ,
[0020] in, Represents the target GMC sparse denoising model; For the fidelity item; represents the foreign body defect collision signal matrix; Represents the Laplace wavelet analysis dictionary; The sparse coding matrix representing the foreign body defect collision signal is given by OK Column sparse coefficient composition; represents the regularization parameter; is the target penalty item; The sparse coding matrix representing the collision signal of foreign body defects Line The weight of the column sparse coefficient; The sparse coding matrix representing the collision signal of foreign body defects Line Column sparse coefficient; Represents the Hadamard product.
[0021] Preferably, the construction process of the fidelity item includes:
[0022] Establishing a Laplace wavelet analysis dictionary based on the foreign body defect collision signal and the foreign body defect collision signal matrix;
[0023] Based on the sparse coding matrix of the foreign body defect collision signal and the Laplace wavelet analysis dictionary, a target defect collision signal matrix is obtained;
[0024] A fidelity term is constructed based on the foreign object defect collision signal matrix and the target defect collision signal matrix.
[0025] Preferably, establishing a Laplace wavelet analysis dictionary based on the foreign body defect collision signal and the foreign body defect collision signal matrix comprises:
[0026] Setting a value range of the oscillation frequency, a value range of the time shift parameter, and a value range of the viscous damping ratio; taking values of the oscillation frequency, the time shift parameter, and the viscous damping ratio within the value range of the oscillation frequency, the value range of the time shift parameter, and the value range of the viscous damping ratio, respectively, constructing a Laplace wavelet atom based on each value, and obtaining a Laplace wavelet atom set based on all value combinations;
[0027] respectively calculating the similarity between each Laplace wavelet atom in the Laplace wavelet atom set and the foreign body defect collision signal;
[0028] The Laplace wavelet analytic dictionary is constructed based on the oscillation frequency, time shift parameter, viscous damping ratio and Laplace wavelet atomic basis function corresponding to the Laplace wavelet atom with the greatest similarity.
[0029] Preferably, the Laplace wavelet atom is represented as:
[0030] ,
[0031] in, represents the Laplace wavelet atom; represents the oscillation frequency; represents the viscous damping ratio; represents the time shift parameter; Indicates time; represents the wavelet support interval;
[0032] The calculation formula of the similarity between the Laplace wavelet atom and the foreign body defect collision signal is:
[0033] ,
[0034] in, express and similarity; represents the Laplace wavelet atom; Indicates the collision signal of foreign body defect; represents the Euclidean norm.
[0035] Preferably, the target GMC sparse denoising model is iteratively solved to obtain a sparse coding matrix of the foreign object defect collision signal, including:
[0036] Step 1: Introduce proxy variables and Lagrange multipliers to update the target GMC sparse denoising model, and decompose the updated target GMC sparse denoising model into a proxy variable constraint subproblem and a sparse coding matrix constraint subproblem;
[0037] Step 2: setting the initial sparse coding matrix, the initial Lagrange multiplier and the weight of each sparse coefficient in the initial sparse coding matrix, and calculating the initial value of the target GMC sparse denoising model based on the initial sparse coding matrix and the weight of each sparse coefficient in the initial sparse coding matrix;
[0038] Step 3: Substitute the initial sparse coding matrix, the weight of each sparse coefficient in the initial sparse coding matrix, and the initial Lagrange multiplier into the proxy variable constraint subproblem to solve the target proxy variable;
[0039] Step 4: Substitute the target proxy variable and the initial Lagrange multiplier into the sparse coding matrix constraint subproblem to solve the target sparse coding matrix;
[0040] Step 5: Use the target sparse coding matrix to sparse the foreign body defect collision signal, and perform singular value decomposition on the sparse foreign body defect collision signal to obtain target singular values, thereby calculating the weight of each sparse coefficient in the target sparse coding matrix;
[0041] Step 6: Solving the target Lagrangian multiplier based on the target proxy variable, the target sparse coding matrix and the initial Lagrangian multiplier;
[0042] Step 7: Calculate the target value of the GMC sparse denoising model based on the target sparse coding matrix and the weight of each sparse coefficient in the target sparse coding matrix, and calculate the difference between the initial value of the target GMC sparse denoising model and the target value of the target GMC sparse denoising model;
[0043] Step 8: Determine the difference and the preset convergence error. If the difference is less than or equal to the preset convergence error, use the target sparse coding matrix as the sparse coding matrix of the foreign object defect collision signal.
[0044] Step 9: If the difference is greater than the preset convergence error, the target value of the GMC sparse denoising model is used as the initial value of the GMC sparse denoising model, the target sparse coding matrix is used as the initial sparse coding matrix, the weight of each sparse coefficient in the target sparse coding matrix is used as the weight of each sparse coefficient in the initial sparse coding matrix, the target Lagrangian multiplier is used as the initial Lagrangian multiplier, and return to execute step 3 until the difference between the initial value of the GMC sparse denoising model and the target value of the GMC sparse denoising model is less than or equal to the preset convergence error.
[0045] Preferably, the updated target GMC sparse denoising model is expressed as:
[0046] ,
[0047] in, Represents the updated target GMC sparse denoising model; represents a proxy variable; represents the Lagrange multiplier; The weight matrix composed of the weights of all sparse coefficients in the sparse coding matrix representing the foreign body defect collision signal No. List; For the fidelity item; represents the foreign body defect collision signal matrix; Represents the Laplace wavelet analysis dictionary; A sparse coding matrix representing a foreign body defect collision signal; represents the regularization parameter; is the target penalty item; The sparse coding matrix representing the collision signal of foreign body defects Line The weight of the column sparse coefficient; The sparse coding matrix representing the collision signal of foreign body defects Line Column sparse coefficient; represents the Hadamard product; represents the penalty parameter;
[0048] The proxy variable constraint subproblem is expressed as:
[0049] ,
[0050] in, represents the solution vector of the proxy variables, The first Line The element value of the column, represents the Lagrange multiplier Line The element value of the column;
[0051] The sparse coding matrix constraint subproblem is expressed as:
[0052] ,
[0053] in, represents the solution vector of the sparse coding matrix;
[0054] The calculation formula of the target Lagrange multiplier is:
[0055] ,
[0056] in, represents the target Lagrange multiplier, represents the Lagrange multiplier, represents the relaxation factor, represents the target sparse coding matrix, Represents the target proxy variable.
[0057] The present invention also provides a microelectronic product foreign body defect detection device, comprising:
[0058] The target penalty term construction module is used to obtain the foreign body defect collision signal of the microelectronic product to be detected, set weights for each sparse coefficient in the penalty term of the GMC sparse denoising model of the foreign body defect collision signal, and obtain the target penalty term, which specifically includes:
[0059] The signal segmentation submodule is used to segment the foreign body defect collision signal to obtain Dimensional foreign body defect collision signal matrix;
[0060] The singular value and signal kurtosis calculation submodule is used to perform singular value decomposition on the foreign body defect collision signal matrix to obtain singular values; taking each column element in the foreign body defect collision signal matrix as a sub-signal to obtain m sub-signals, and calculating the signal kurtosis of each sub-signal respectively;
[0061] The sparse coefficient weight allocation submodule is used to calculate the The singular values and The square sum of the signal kurtosis of the sub-signals is calculated, and the reciprocal of the square sum is used as the sparse coding matrix of the foreign body defect collision signal. Line The weight of the column sparse coefficient; where, , ;
[0062] A sparse coding matrix acquisition module is used to obtain a target GMC sparse denoising model based on a fidelity term and the target penalty term; and iteratively solve the target GMC sparse denoising model to obtain a sparse coding matrix of a foreign object defect collision signal;
[0063] The foreign body defect detection module is used to sparsely represent the foreign body defect collision signal using the sparse coding matrix to obtain a target foreign body defect collision signal, thereby performing foreign body defect detection on the microelectronic product to be detected.
[0064] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for detecting foreign matter defects in microelectronic products are implemented.
[0065] The microelectronic product foreign body defect detection method provided in the present application uses the GMC penalty function for sparse denoising of the collision signal of the foreign body defect of the microelectronic product. The sparse coding matrix is obtained by solving the GMC sparse denoising model to remove the environmental noise in the collision signal of the foreign body defect and enhance the signal amplitude. Considering the kurtosis characteristics of the signal, it can effectively reflect the collision of foreign body particles submerged in the environmental noise. The greater the signal kurtosis, the greater the possibility that the signal is generated by the collision of foreign body defects. The collision information of foreign body particles is mainly concentrated in the larger singular values. Therefore, when constructing the penalty term of the GMC sparse denoising model, the present application divides the collision signal of the foreign body defect into segments. dimensional foreign body defect collision signal matrix, thereby obtaining m sub-signals, and calculating the signal kurtosis of each sub-signal respectively, and performing singular value decomposition on the foreign body defect collision signal to obtain singular values, and the weight of the sparse coefficient of the jth row and the i-th column in the sparse coding matrix is constructed based on the inverse of the square and arithmetic square root of the signal kurtosis of the jth singular value and the i-th sub-signal. The larger the signal kurtosis and the singular value, the smaller the weight of the sparse coefficient, so that the solved sparse coding matrix can attenuate the sub-signal with a larger kurtosis and the larger singular value component to a smaller extent, retain the collision signal generated by the sliding and collision of the foreign body defect in the cavity as much as possible, remove the environmental noise in the collision signal of the foreign body defect, and thus improve the accuracy of the foreign body defect detection result. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:
[0067] Figure 1 A flow chart of the method for detecting foreign matter defects in microelectronic products provided in this application;
[0068] Figure 2 A foreign body defect collision signal provided in an embodiment of the present application;
[0069] Figure 3 The signal processing method provided in the embodiment of the present application is used for Figure 2 The signal shown is a foreign body defect collision signal after denoising; wherein, Figure 3 (a) is the wavelet packet denoising method. Figure 2 The signal shown is the foreign body defect collision signal after denoising. Figure 3 (b) is the VMD denoising method. Figure 2 The signal shown is the foreign body defect collision signal after denoising. Figure 3 (c) in the figure is the L1 sparse denoising method. Figure 2 The signal shown is the foreign body defect collision signal after denoising. Figure 3 (d) is to use the method provided by this application to Figure 2 The signal shown is a foreign body defect collision signal after denoising;
[0070] Figure 4 This is a schematic diagram of the structure of the microelectronic product foreign body defect detection device provided in this application. DETAILED DESCRIPTION
[0071] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.
[0072] See also Figure 1 , Figure 1 A flow chart of a method for detecting foreign matter defects in microelectronic products provided in an embodiment of the present application, the method specifically includes:
[0073] S10: Obtain a foreign body defect collision signal of a microelectronic product to be detected, set a weight for each sparse coefficient in a penalty term of a GMC sparse denoising model of the foreign body defect collision signal, and obtain a target penalty term, which specifically includes:
[0074] S100: Segment the foreign body defect collision signal to obtain Dimensional foreign body defect collision signal matrix;
[0075] For example, the acquired foreign body defect collision signal of length N is divided into blocks to obtain The foreign body defect collision signal matrix of dimension, the block operator is:
[0076] ,
[0077] in, Indicates foreign body defect collision signal; Indicates the window length of the block operator; Represents the window shift length, which is used to eliminate the end effect of the block operator; based on The foreign body defect collision signal matrix of dimension is obtained Sub-signals, ;
[0078] S101: Perform singular value decomposition on the foreign body defect collision signal matrix to obtain singular values; take each column element in the foreign body defect collision signal matrix as a sub-signal, obtain m sub-signals, and calculate the signal kurtosis of each sub-signal respectively;
[0079] S102: Calculate the The singular values and The square sum of the signal kurtosis of the sub-signals is calculated and the inverse of the square sum is used as the inverse of the square root of the foreign body defect collision signal in the sparse coding matrix. Line The weight of the column sparse coefficient; where, , ;
[0080] S20: based on the fidelity term and the target penalty term, a target GMC sparse denoising model is obtained; and the target GMC sparse denoising model is iteratively solved to obtain a sparse coding matrix of the foreign object defect collision signal;
[0081] S30: sparsely represent the foreign body defect collision signal using a sparse coding matrix to obtain a target foreign body defect collision signal, thereby performing foreign body defect detection on the microelectronic product to be detected.
[0082] The microelectronic product foreign body defect detection method provided in the present application uses the GMC penalty function for sparse denoising of the collision signal of the foreign body defect of the microelectronic product, thereby removing the environmental noise in the collision signal of the foreign body defect and enhancing the signal amplitude. At the same time, considering the kurtosis characteristics of the signal can effectively reflect the collision of foreign body particles submerged in the environmental noise. The greater the signal kurtosis, the greater the possibility that the signal is generated by the collision of foreign body defects. The collision information of foreign body particles is mainly concentrated in the larger singular values. Therefore, when constructing the penalty term of the GMC sparse denoising model, the present application divides the collision signal of the foreign body defect into segments. dimensional foreign body defect collision signal matrix, thereby obtaining m sub-signals, and calculating the signal kurtosis of each sub-signal respectively, and performing singular value decomposition on the foreign body defect collision signal to obtain singular values, and the weight of the sparse coefficient of the jth row and the i-th column in the sparse coding matrix is constructed based on the inverse of the square and arithmetic square root of the signal kurtosis of the jth singular value and the i-th sub-signal. The larger the signal kurtosis and the singular value, the smaller the weight of the sparse coefficient, so that the solved sparse coding matrix can attenuate the sub-signal with a larger kurtosis and the larger singular value component to a smaller extent, retain the collision signal generated by the sliding and collision of the foreign body defect in the cavity as much as possible, remove the environmental noise in the collision signal of the foreign body defect, and thus improve the accuracy of the foreign body defect detection result.
[0083] Specifically, in some embodiments of the present application, the sparse coding matrix of the foreign body defect collision signal in step S102 is Line The calculation formula for the weight of the column sparse coefficient is:
[0084] ,
[0085] in, The sparse coding matrix representing the collision signal of foreign body defects Line The weight of the column sparse coefficient; Indicates The signal kurtosis of each sub-signal; Indicates singular values;
[0086] ,
[0087] in, Indicates The first data points; Indicates The number of data points in each sub-signal; Indicates The mean of all data points in the subsignal.
[0088] Specifically, the construction process of the fidelity item in step S20 includes:
[0089] Based on the foreign body defect collision signal and the foreign body defect collision signal matrix, a Laplace wavelet analysis dictionary is established, which specifically includes:
[0090] The value range of the oscillation frequency, the value range of the time shift parameter and the value range of the viscous damping ratio are set, and the oscillation frequency, the time shift parameter and the viscous damping ratio are respectively taken within the value range of the oscillation frequency, the value range of the time shift parameter and the value range of the viscous damping ratio. A Laplace wavelet atom is constructed based on each value, and a Laplace wavelet atom set is obtained based on all value combinations; specifically, the Laplace wavelet atom is expressed as:
[0091] ,
[0092] in, represents the Laplace wavelet atom; represents the oscillation frequency; represents the viscous damping ratio; represents the time shift parameter; Indicates time; represents the wavelet support interval;
[0093] The similarity between each Laplace wavelet atom in the Laplace wavelet atom set and the collision signal of the foreign body defect is calculated respectively; specifically, the calculation formula of the similarity between the Laplace wavelet atom and the collision signal of the foreign body defect is:
[0094] ,
[0095] in, express and similarity; represents the Laplace wavelet atom; Indicates the collision signal of foreign body defect; represents the Euclidean norm;
[0096] Based on the oscillation frequency, time shift parameter, viscous damping ratio and Laplace wavelet atomic basis function corresponding to the Laplace wavelet atom with the greatest similarity, a Laplace wavelet analytic dictionary is constructed;
[0097] Based on the sparse coding matrix of the foreign body defect collision signal and the Laplace wavelet analysis dictionary, the target defect signal matrix is obtained;
[0098] The fidelity term is constructed based on the foreign body defect collision signal matrix and the target defect signal matrix.
[0099] Further, based on the above fidelity term and target penalty term, the target GMC sparse denoising model in the embodiment of the present application is expressed as:
[0100] ,
[0101] in, Represents the target GMC sparse denoising model; For the fidelity item; represents the foreign body defect collision signal matrix; Represents the Laplace wavelet analysis dictionary; The sparse coding matrix representing the foreign body defect collision signal is given by OK Column sparse coefficient composition; represents the regularization parameter; is the target penalty item; The sparse coding matrix representing the collision signal of foreign body defects Line The weight of the column sparse coefficient; The sparse coding matrix representing the collision signal of foreign body defects Line Column sparse coefficient; Represents the Hadamard product.
[0102] Furthermore, in step S20, solving the target GMC sparse denoising model specifically includes:
[0103] Step 1: Introduce proxy variables and Lagrange multipliers to update the target GMC sparse denoising model, and decompose the updated target GMC sparse denoising model into a proxy variable constraint subproblem and a sparse coding matrix constraint subproblem;
[0104] Specifically, the updated GMC sparse denoising model is expressed as:
[0105] ,
[0106] in, Represents the updated target GMC sparse denoising model; represents a proxy variable; represents the Lagrange multiplier; The weight matrix composed of the weights of all sparse coefficients in the sparse coding matrix representing the foreign body defect collision signal No. List; For the fidelity item; represents the foreign body defect collision signal matrix; Represents the Laplace wavelet analysis dictionary; A sparse coding matrix representing a foreign body defect collision signal; represents the regularization parameter; is the target penalty item; The sparse coding matrix representing the collision signal of foreign body defects Line The weight of the column sparse coefficient; The sparse coding matrix representing the collision signal of foreign body defects Line Column sparse coefficient; represents the Hadamard product; represents the penalty parameter;
[0107] The proxy variable constraint subproblem is expressed as:
[0108] ,
[0109] in, represents the solution vector of the proxy variables, The first Line The element value of the column, represents the Lagrange multiplier Line The element value of the column; specifically, the target proxy variable obtained by solving , The closed-form solution is:
[0110] ,
[0111] in, Indicates the target proxy variable Column vector, Represents the first Column sparse coefficient;
[0112] in, Represents the threshold function of the target penalty term, which is specifically expressed as:
[0113] ,
[0114] in, represents the non-convex parameter, satisfying ;
[0115] The sparse coding matrix constraint subproblem is:
[0116] ,
[0117] in, represents the solution vector of the sparse coding matrix; further, , represents the identity matrix;
[0118] Step 2: setting the initial sparse coding matrix, the initial Lagrange multiplier and the weight of each sparse coefficient in the initial sparse coding matrix, and calculating the initial value of the target GMC sparse denoising model based on the initial sparse coding matrix and the weight of each sparse coefficient in the initial sparse coding matrix;
[0119] Step 3: Substitute the initial sparse coding matrix, the weight of each sparse coefficient in the initial sparse coding matrix, and the initial Lagrange multiplier into the proxy variable constraint subproblem to solve the target proxy variable;
[0120] Step 4: Substitute the target proxy variable and the initial Lagrange multiplier into the sparse coding matrix constraint subproblem to solve the target sparse coding matrix;
[0121] Step 5: Use the target sparse coding matrix to sparse the foreign body defect collision signal, and perform singular value decomposition on the sparse foreign body defect collision signal to obtain target singular values, thereby calculating the weight of each sparse coefficient in the target sparse coding matrix;
[0122] Step 6: Solving the target Lagrangian multiplier based on the target proxy variable, the target sparse coding matrix and the initial Lagrangian multiplier;
[0123] Specifically, the calculation formula of the target Lagrange multiplier is:
[0124] ,
[0125] in, represents the target Lagrange multiplier, represents the Lagrange multiplier, represents the relaxation factor, represents the target sparse coding matrix, represents the target proxy variable;
[0126] Step 7: Calculate the target value of the GMC sparse denoising model based on the target sparse coding matrix and the weight of each sparse coefficient in the target sparse coding matrix, and calculate the difference between the initial value of the target GMC sparse denoising model and the target value of the target GMC sparse denoising model;
[0127] Step 8: Determine the size of the difference and a preset convergence error. If the difference is less than or equal to the preset convergence error, use the target sparse coding matrix as the sparse coding matrix of the foreign object defect collision signal.
[0128] Step 9: If the difference is greater than the preset convergence error, the target value of the GMC sparse denoising model is used as the initial value of the GMC sparse denoising model, the target sparse coding matrix is used as the initial sparse coding matrix, the weight of each sparse coefficient in the target sparse coding matrix is used as the weight of each sparse coefficient in the initial sparse coding matrix, the target Lagrangian multiplier is used as the initial Lagrangian multiplier, and return to execute step 3 until the difference between the initial value of the GMC sparse denoising model and the target value of the GMC sparse denoising model is less than or equal to the preset convergence error.
[0129] In order to verify the effectiveness of the microelectronic product foreign body defect detection method provided in the present application, the present application embodiment also provides a comparative example of using different sparse denoising methods to perform sparse denoising on the foreign body defect collision signal:
[0130] like Figure 2 The figure shows the foreign body defect collision signal extracted by using the SD4511 particle collision noise detection device to detect the ceramic package tube shell provided in the embodiment of the present application. The sampling frequency is 500KHz, and the signal sampling length is 51200 sampling points. It can be seen from the figure that the low-amplitude foreign body defect collision signal is submerged in the environmental noise and is difficult to distinguish, which can easily cause the missed detection or wrong detection of micro foreign body defects;
[0131] Figure 3 The following are the results of different signal processing methods. Figure 2 Schematic diagram of the result of denoising the signal in , specifically, Figure 3 (a) is the wavelet packet denoising method. Figure 2 Schematic diagram of foreign body defect collision signal obtained after denoising the signal in, Figure 3 (b) is the VMD method. Figure 2 Schematic diagram of foreign body defect collision signal obtained after denoising the signal in, Figure 3 (c) in the figure is the L1 sparse denoising method. Figure 2 Schematic diagram of foreign body defect collision signal obtained after sparse denoising of the signal in, Figure 3 (d) is the method provided by this application for Figure 2 Schematic diagram of the foreign body defect collision signal obtained after denoising the signal in. It can be seen from the figure that compared with the wavelet packet denoising method and the VMD denoising method, the signal obtained after denoising by the method provided in the present application effectively removes the interference of environmental noise. Compared with the L1 sparse denoising method, the signal amplitude loss obtained after denoising by the method provided in the present application is smaller, making the impact characteristics of the signal more prominent.
[0132] Based on the microelectronic product foreign body defect detection method provided in the above embodiment, the present application embodiment also provides a microelectronic product foreign body defect detection device, such as Figure 4As shown, the device specifically includes:
[0133] The target penalty term building module 10 is used to obtain the foreign body defect collision signal of the microelectronic product to be detected, set weights for each sparse coefficient in the penalty term of the GMC sparse denoising model of the foreign body defect collision signal, and obtain the target penalty term, which specifically includes:
[0134] The signal segmentation submodule is used to segment the foreign body defect collision signal to obtain Dimensional foreign body defect collision signal matrix;
[0135] The singular value and signal kurtosis calculation submodule is used to perform singular value decomposition on the foreign body defect collision signal matrix to obtain singular values; taking each column element in the foreign body defect collision signal matrix as a sub-signal to obtain m sub-signals, and calculating the signal kurtosis of each sub-signal respectively;
[0136] The sparse coefficient weight allocation submodule is used to calculate the The singular values and The square sum of the signal kurtosis of the sub-signals is calculated, and the reciprocal of the square sum is used as the sparse coding matrix of the foreign body defect collision signal. Line The weight of the column sparse coefficient; where, , ;
[0137] A sparse coding matrix acquisition module 20 is used to obtain a target GMC sparse denoising model based on a fidelity term and the target penalty term; and iteratively solve the target GMC sparse denoising model to obtain a sparse coding matrix of a foreign object defect collision signal;
[0138] The foreign body defect detection module 30 is used to perform sparse representation on the foreign body defect collision signal by using the sparse coding matrix to obtain a target foreign body defect collision signal, thereby performing foreign body defect detection on the microelectronic product to be detected.
[0139] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for detecting foreign matter defects in microelectronic products are implemented.
[0140] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt 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.) containing computer-usable program codes.
[0141] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0142] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0143] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0144] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.
Claims
1. A method for detecting foreign matter defects in microelectronic products, It is characterized in that include: Obtain the foreign body defect collision signal of the microelectronic product to be detected, set weights for each sparse coefficient in the penalty term of the GMC sparse denoising model of the foreign body defect collision signal, and obtain the target penalty term, which specifically includes: Segmenting the foreign body defect collision signal to obtain an n×m dimensional foreign body defect collision signal matrix; Performing singular value decomposition on the foreign body defect collision signal matrix to obtain n singular values; taking each column element in the foreign body defect collision signal matrix as a sub-signal to obtain m sub-signals, and calculating the signal kurtosis of each sub-signal respectively; Calculate the square sum and arithmetic square root of the j-th singular value and the signal kurtosis of the i-th sub-signal, and use the reciprocal of the square sum and arithmetic square root as the weight of the sparse coefficient of the j-th row and i-th column in the sparse coding matrix of the foreign body defect collision signal; wherein i∈[1,m], j∈[1,n]; Based on the fidelity term and the target penalty term, a target GMC sparse denoising model is obtained; and the target GMC sparse denoising model is iteratively solved to obtain a sparse coding matrix of the foreign body defect collision signal; which specifically includes: Step 1: Introduce proxy variables and Lagrange multipliers to update the target GMC sparse denoising model, and decompose the updated target GMC sparse denoising model into a proxy variable constraint subproblem and a sparse coding matrix constraint subproblem; Step 2: setting the initial sparse coding matrix, the initial Lagrange multiplier and the weight of each sparse coefficient in the initial sparse coding matrix, and calculating the initial value of the target GMC sparse denoising model based on the initial sparse coding matrix and the weight of each sparse coefficient in the initial sparse coding matrix; Step 3: Substitute the initial sparse coding matrix, the weight of each sparse coefficient in the initial sparse coding matrix, and the initial Lagrange multiplier into the proxy variable constraint subproblem to solve the target proxy variable; Step 4: Substitute the target proxy variable and the initial Lagrange multiplier into the sparse coding matrix constraint subproblem to solve the target sparse coding matrix; Step 5: sparse the foreign object defect collision signal using the target sparse coding matrix, and perform singular value decomposition on the sparse foreign object defect collision signal to obtain n target singular values, thereby calculating the weight of each sparse coefficient in the target sparse coding matrix; Step 6: Solving the target Lagrangian multiplier based on the target proxy variable, the target sparse coding matrix and the initial Lagrangian multiplier; Step 7: Calculate the target value of the GMC sparse denoising model based on the target sparse coding matrix and the weight of each sparse coefficient in the target sparse coding matrix, and calculate the difference between the initial value of the target GMC sparse denoising model and the target value of the target GMC sparse denoising model; Step 8: Determine the difference and the preset convergence error. If the difference is less than or equal to the preset convergence error, use the target sparse coding matrix as the sparse coding matrix of the foreign object defect collision signal. Step 9: If the difference is greater than a preset convergence error, the target value of the GMC sparse denoising model is used as the initial value of the GMC sparse denoising model, the target sparse coding matrix is used as the initial sparse coding matrix, the weight of each sparse coefficient in the target sparse coding matrix is used as the weight of each sparse coefficient in the initial sparse coding matrix, the target Lagrangian multiplier is used as the initial Lagrangian multiplier, and the process returns to step 3 until the difference between the initial value of the GMC sparse denoising model and the target value of the GMC sparse denoising model is less than or equal to the preset convergence error; The foreign body defect collision signal is sparsely represented by using the sparse coding matrix to obtain a target foreign body defect collision signal, thereby performing foreign body defect detection on the microelectronic product to be detected.
2. The method for detecting foreign matter defects in microelectronic products according to claim 1, It is characterized in that The calculation formula of the weight of the sparse coefficient in the jth row and the ith column in the sparse coding matrix of the foreign body defect collision signal is: Among them, W j,i K represents the weight of the sparse coefficient in the jth row and ith column of the sparse coding matrix of the foreign body defect collision signal; i represents the signal kurtosis of the i-th sub-signal; ∑ j represents the jth singular value; Among them, Y il represents the lth data point in the i-th sub-signal; k represents the number of data points in the i-th sub-signal; Represents the mean of all data points in the i-th sub-signal.
3. The method for detecting foreign matter defects in microelectronic products according to claim 1, It is characterized in that The target GMC sparse denoising model is expressed as: Where J represents the target GMC sparse denoising model; is the fidelity term; Y represents the matrix of foreign body defect collision signal; D represents the Laplace wavelet analysis dictionary; s represents the sparse coding matrix of foreign body defect collision signal, which consists of n rows and m columns of sparse coefficients; λ represents the regularization parameter; ∑ j ∑ i P GMC (W j,i ⊙s j,i ) is the target penalty term; W j,i The weight of the sparse coefficient in the jth row and the ith column in the sparse coding matrix of the foreign body defect collision signal; s j,i represents the sparse coefficient of the j-th row and the i-th column in the sparse coding matrix of the foreign body defect collision signal; ⊙ represents the Hadamard product.
4. The method for detecting foreign matter defects in microelectronic products according to claim 1, It is characterized in that The construction process of the fidelity term includes: Establishing a Laplace wavelet analysis dictionary based on the foreign body defect collision signal and the foreign body defect collision signal matrix; Based on the sparse coding matrix of the foreign body defect collision signal and the Laplace wavelet analysis dictionary, a target defect collision signal matrix is obtained; A fidelity term is constructed based on the foreign object defect collision signal matrix and the target defect collision signal matrix.
5. The method for detecting foreign matter defects in microelectronic products according to claim 4, It is characterized in that Based on the foreign body defect collision signal and the foreign body defect collision signal matrix, establishing a Laplace wavelet analysis dictionary includes: Setting a value range of the oscillation frequency, a value range of the time shift parameter, and a value range of the viscous damping ratio; taking values of the oscillation frequency, the time shift parameter, and the viscous damping ratio within the value range of the oscillation frequency, the value range of the time shift parameter, and the value range of the viscous damping ratio, respectively, constructing a Laplace wavelet atom based on each value, and obtaining a Laplace wavelet atom set based on all value combinations; respectively calculating the similarity between each Laplace wavelet atom in the Laplace wavelet atom set and the foreign body defect collision signal; The Laplace wavelet analytic dictionary is constructed based on the oscillation frequency, time shift parameter, viscous damping ratio and Laplace wavelet atomic basis function corresponding to the Laplace wavelet atom with the greatest similarity.
6. The method for detecting foreign matter defects in microelectronic products according to claim 5, It is characterized in that The Laplace wavelet atom is expressed as: Among them, ψ f,ζ,τ (t) represents the Laplace wavelet atom; f represents the oscillation frequency; ζ represents the viscous damping ratio; τ represents the time shift parameter; t represents time; W represents the wavelet support interval; The calculation formula of the similarity between the Laplace wavelet atom and the foreign body defect collision signal is: Among them, C f,ζ,τ (t) represents ψ f,ζ,τ The similarity between (t) and y(t); ψ f,ζ,τ (t) represents the Laplace wavelet atom; y(t) represents the foreign body defect collision signal; ‖·‖ 2 represents the Euclidean norm.
7. The method for detecting foreign matter defects in microelectronic products according to claim 1, It is characterized in that The updated target GMC sparse denoising model is expressed as: Among them, L μ (s,z,β,W i ) represents the updated target GMC sparse denoising model; z represents the proxy variable; β represents the Lagrange multiplier; W i The i-th column of the weight matrix W composed of the weights of all sparse coefficients in the sparse coding matrix representing the foreign body defect collision signal; is the fidelity term; Y represents the foreign body defect collision signal matrix; D represents the Laplace wavelet analysis dictionary; s represents the sparse coding matrix of the foreign body defect collision signal; λ represents the regularization parameter; ∑ j ∑ i P GMC (W j,i ⊙s j,i ) is the target penalty term; W j,i The weight of the sparse coefficient in the jth row and ith column in the sparse coding matrix of the foreign body defect collision signal; s j,i represents the sparse coefficient of the jth row and ith column in the sparse coding matrix of the foreign body defect collision signal; ⊙ represents the Hadamard product; μ represents the penalty parameter; The proxy variable constraint subproblem is expressed as: in, represents the solution vector of the proxy variables, z j,i represents the value of the element in the jth row and ith column of the solution vector of the proxy variable, β j,i represents the element value of the jth row and ith column in the Lagrange multiplier; The sparse coding matrix constraint subproblem is expressed as: in, represents the solution vector of the sparse coding matrix; The calculation formula of the target Lagrange multiplier is: β'=β+γμ(s'-z'), Among them, β′ represents the target Lagrange multiplier, β represents the Lagrange multiplier, γ represents the relaxation factor, s′ represents the target sparse coding matrix, and z′ represents the target proxy variable.
8. A device for detecting foreign matter defects in microelectronic products, It is characterized in that include: The target penalty term construction module is used to obtain the foreign body defect collision signal of the microelectronic product to be detected, set weights for each sparse coefficient in the penalty term of the GMC sparse denoising model of the foreign body defect collision signal, and obtain the target penalty term, which specifically includes: A signal segmentation submodule, used to segment the foreign body defect collision signal to obtain an n×m dimensional foreign body defect collision signal matrix; A singular value and signal kurtosis calculation submodule is used to perform singular value decomposition on the foreign body defect collision signal matrix to obtain n singular values; take each column element in the foreign body defect collision signal matrix as a sub-signal to obtain m sub-signals, and calculate the signal kurtosis of each sub-signal respectively; A sparse coefficient weight allocation submodule is used to calculate the square sum arithmetic square root of the j-th singular value and the signal kurtosis of the i-th sub-signal, and use the reciprocal of the square sum arithmetic square root as the weight of the sparse coefficient of the j-th row and i-th column in the sparse coding matrix of the foreign body defect collision signal; wherein, i∈[1,m], j∈[1,n]; A sparse coding matrix acquisition module is used to obtain a target GMC sparse denoising model based on a fidelity term and the target penalty term; and iteratively solve the target GMC sparse denoising model to obtain a sparse coding matrix of a foreign object defect collision signal; which specifically includes: Step 1: Introduce proxy variables and Lagrange multipliers to update the target GMC sparse denoising model, and decompose the updated target GMC sparse denoising model into a proxy variable constraint subproblem and a sparse coding matrix constraint subproblem; Step 2: Set the initial sparse coding matrix, the initial Lagrange multiplier and the weight of each sparse coefficient in the initial sparse coding matrix, and calculate the initial value of the target GMC sparse denoising model based on the initial sparse coding matrix and the weight of each sparse coefficient in the initial sparse coding matrix; Step 3: Substitute the initial sparse coding matrix, the weight of each sparse coefficient in the initial sparse coding matrix and the initial Lagrange multiplier into the proxy variable constraint subproblem to solve the target proxy variable; Step 4: Substitute the target proxy variable and the initial Lagrange multiplier into the sparse coding matrix constraint subproblem to solve the target sparse coding matrix; Step 5: sparse the foreign object defect collision signal using the target sparse coding matrix, and perform singular value decomposition on the sparse foreign object defect collision signal to obtain n target singular values, thereby calculating the weight of each sparse coefficient in the target sparse coding matrix; Step 6: Solving the target Lagrangian multiplier based on the target proxy variable, the target sparse coding matrix and the initial Lagrangian multiplier; Step 7: Calculate the target value of the GMC sparse denoising model based on the target sparse coding matrix and the weight of each sparse coefficient in the target sparse coding matrix, and calculate the difference between the initial value of the target GMC sparse denoising model and the target value of the target GMC sparse denoising model; Step 8: Determine the difference and the preset convergence error. If the difference is less than or equal to the preset convergence error, use the target sparse coding matrix as the sparse coding matrix of the foreign object defect collision signal. Step 9: If the difference is greater than a preset convergence error, the target value of the GMC sparse denoising model is used as the initial value of the GMC sparse denoising model, the target sparse coding matrix is used as the initial sparse coding matrix, the weight of each sparse coefficient in the target sparse coding matrix is used as the weight of each sparse coefficient in the initial sparse coding matrix, the target Lagrangian multiplier is used as the initial Lagrangian multiplier, and the process returns to step 3 until the difference between the initial value of the GMC sparse denoising model and the target value of the GMC sparse denoising model is less than or equal to the preset convergence error; The foreign body defect detection module is used to sparsely represent the foreign body defect collision signal using the sparse coding matrix to obtain a target foreign body defect collision signal, thereby performing foreign body defect detection on the microelectronic product to be detected.
9. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for detecting foreign matter defects in microelectronic products according to any one of claims 1 to 7 are implemented.
