Semiconductor chip defect detection positioning method and system

Abstract

The application provides a semiconductor chip defect detection positioning method and system, relates to the technical field of semiconductor chip detection, and comprises the following steps: acquiring an original image and electrical test data; forming a defect candidate area based on local density analysis; performing variational mode decomposition on the electrical data and marking abnormal points of electrical characteristics; mapping image features to an electrical characteristic space and calculating mutual information values to determine defect points; and calculating diffusion field distribution and paths and determining defect severity. The application realizes accurate positioning and severity evaluation of chip defects, and improves detection efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor chip detection, and in particular to a semiconductor chip defect detection and positioning method and system. BACKGROUND

[0002] With the rapid development of integrated circuit technology, the manufacturing process of semiconductor chips is becoming increasingly complex, the chip size is continuously shrinking, and the integration continues to improve. In the production process of semiconductor chips, due to process fluctuations, material defects, environmental influences and other factors, various defects often occur in the chips, which will affect the performance and reliability of the chips. Therefore, accurate detection and positioning of semiconductor chip defects are of great significance for improving chip yield, reducing production costs, and ensuring product quality.

[0003] Traditional semiconductor chip defect detection methods mainly include optical detection and electrical testing. Optical detection obtains chip surface images through scanning electron microscopes, atomic force microscopes and other equipment, and identifies defects through image processing algorithms; electrical testing measures electrical parameters such as current, voltage, resistance, etc. of the chip to determine whether the function of the chip is normal. However, these methods have some limitations in practical application. SUMMARY

[0004] Embodiments of the present application provide a semiconductor chip defect detection and positioning method and system, which can solve the problems in the prior art.

[0005] In a first aspect of the embodiments of the present application, a semiconductor chip defect detection and positioning method is provided, comprising:

[0006] Obtaining an original image of a semiconductor chip to be tested and electrical test data of different sampling points;

[0007] Calculating the local density value of each pixel point in the original image, and calculating the deviation from the regional density mean value, clustering the pixel points exceeding the deviation threshold to form a defect candidate region;

[0008] Performing variational mode decomposition on the electrical test data of the different sampling points to obtain a set of intrinsic mode functions, performing Hilbert transform on the intrinsic mode functions and calculating the instantaneous frequency, and marking the sampling points exceeding the frequency threshold as electrical characteristic abnormal points;

[0009] Mapping the image features of the defect candidate region to the electrical characteristic parameter space through nonlinear transformation, converting the electrical characteristic abnormal points to abnormal probability distribution in the image space through reverse mapping, calculating the feature mutual information value of the electrical characteristic parameter space and the image space, and determining that there is a defect point at the position when the feature mutual information value is greater than the correlation threshold;

[0010] The diffusion field distribution is calculated with the defect position as a source point, and the diffusion path of the defect is predicted by combining the time-space evolution law of the electric property abnormal point.

[0011] The local density value of each pixel point in the original image is calculated, and the deviation from the regional density average is calculated, and the pixel points exceeding the deviation threshold are clustered to form a defect candidate region, including:

[0012] The original image is divided into a plurality of local windows, the standard deviation of the gray value of the pixel points in each local window is calculated, the product of the gray value standard deviation and the adjustment coefficient is added to the unit base, and then multiplied by the first reference bandwidth parameter to obtain the adaptive bandwidth parameter; the local density value of the center pixel point is calculated by substituting the adaptive bandwidth parameter into the Gaussian kernel function;

[0013] A multi-scale feature pyramid is constructed based on the local density value, and the local density average and the local density standard deviation are calculated at each scale layer; the normalized deviation of the center pixel point is obtained by subtracting the local density average of the scale layer from the local density value of the center pixel point and dividing by the local density standard deviation; when the normalized deviation of any scale layer is greater than the deviation threshold, the position coordinates of the center pixel point are recorded as abnormal point coordinates;

[0014] The density flow between adjacent abnormal points is calculated, and the density flow is the rate of change of the local density value with the spatial position on the path connecting the two abnormal points, and when the maximum value of the density flow is less than the second reference parameter, the two abnormal points are marked as belonging to the same abnormal region, and the set of abnormal points with the same abnormal region number is output as a defect candidate region.

[0015] The variational mode decomposition is performed on the electrical test data of the different sampling points to obtain a set of intrinsic mode functions, the Hilbert transform is performed on the intrinsic mode functions and the instantaneous frequency is calculated, and the sampling points exceeding the frequency threshold are marked as electric property abnormal points, including:

[0016] The local variance of the electrical test data of the different sampling points is calculated in a sliding time window, and the variational mode decomposition is performed to obtain a set of intrinsic mode functions, the energy density of each intrinsic mode function in the set of intrinsic mode functions is calculated, and the orthogonality index between the intrinsic mode functions with energy density greater than the energy threshold is calculated, and the intrinsic mode functions with the orthogonality index less than the orthogonality threshold are determined as effective intrinsic mode functions;

[0017] The Hilbert transform is performed on the effective intrinsic mode functions to obtain an analytic signal, and the generalized zero-crossing rate is calculated based on the zero-crossing characteristics of the analytic signal, and the time derivative of the generalized zero-crossing rate is divided by the frequency normalization coefficient to obtain the instantaneous frequency;

[0018] According to the position distance of the sampling points, a spatial weight coefficient is assigned, a sum of products of the spatial weight coefficient and the instantaneous frequency is divided by a sum of the spatial weight coefficient to obtain a reference frequency, and a frequency deviation value of the instantaneous frequency of different sampling points from the reference frequency is calculated, and sampling points exceeding a frequency threshold are marked as electrical characteristic abnormal points.

[0019] Image features of a defect candidate region are mapped to an electrical characteristic parameter space through a nonlinear transformation, and electrical characteristic abnormal points are converted into an abnormal probability distribution in the image space through reverse mapping, and a feature mutual information value of the electrical characteristic parameter space and the image space is calculated, and when the feature mutual information value is greater than a correlation threshold, it is determined that a defect point exists at the position, comprising:

[0020] An image of a defect candidate region of a chip to be tested is obtained, and the defect candidate region image is decomposed into a multi-scale feature vector through discrete wavelet transform;

[0021] A local covariance matrix of the multi-scale feature vector is calculated, an eigenvalue decomposition of the local covariance matrix is performed to obtain a feature subspace, a plurality of feature vectors with a contribution rate of the top 5% in the feature subspace are extracted to construct a mapping basis, a sum of products of an inner product operation result of the mapping basis and a radial basis kernel function and an adaptive weight coefficient is determined as a nonlinear mapping function, and the multi-scale feature vector is converted to an electrical characteristic parameter space through the nonlinear mapping function to obtain a parameter mapping vector;

[0022] A position potential is calculated based on Mahalanobis distance between the parameter mapping vector and an electrical characteristic abnormal point, and a neighboring potential is calculated based on Euclidean distance between adjacent parameter mapping vectors; a conditional probability is determined according to a weighted combination of the position potential and the neighboring potential, the conditional probability is obtained through maximum likelihood estimation to obtain a reverse mapping probability, and the reverse mapping probability is converted to an image space to obtain a defect probability distribution;

[0023] A ratio of a first edge probability corresponding to the parameter mapping vector to a second edge probability corresponding to the defect probability distribution is taken as a feature mutual information value; when the feature mutual information value is greater than a correlation threshold, it is determined that a defect point exists in the defect candidate region, and spatial coordinate information of the defect point is output.

[0024] A diffusion field distribution is calculated with the defect point position as a source point, and a diffusion path of the defect is predicted in combination with a spatiotemporal evolution law of the electrical characteristic abnormal point; a correlation coefficient of image features of each sampling point on the diffusion path is calculated to determine the severity of the defect, comprising:

[0025] obtain a local material conductivity gradient and a stress intensity of a defect position, determine a diffusion coefficient according to a ratio of the local material conductivity gradient and the stress intensity, and take a product of the diffusion coefficient and a gradient of a defect diffusion field intensity as a diffusion field intensity distribution;

[0026] collect electrical characteristic test data around the defect, extract a feature vector of the electrical characteristic test data through wavelet transform, mark a local maximum point of the feature vector as an abnormal point, calculate a propagation speed and an acceleration of the abnormal point, and determine an abnormal diffusion direction based on the propagation speed and the acceleration;

[0027] convert the diffusion field intensity distribution into an intensity probability distribution, convert the abnormal diffusion direction into a direction probability distribution, fuse the intensity probability distribution and the direction probability distribution by using a maximum entropy criterion to obtain a joint probability distribution, and determine a defect diffusion path along a direction in which the joint probability distribution is maximum;

[0028] obtain a local image gray level co-occurrence matrix of each sampling point on the defect diffusion path, map the gray level co-occurrence matrix into a gray level feature vector through sparse coding, and calculate a correlation coefficient between the gray level feature vectors; when the correlation coefficient is greater than a feature correlation threshold, it is determined that the defect has a diffusion risk, and a position coordinate of the defect diffusion path is output.

[0029] convert the diffusion field intensity distribution into an intensity probability distribution, convert the abnormal diffusion direction into a direction probability distribution, fuse the intensity probability distribution and the direction probability distribution by using a maximum entropy criterion to obtain a joint probability distribution, and determine a defect diffusion path along a direction in which the joint probability distribution is maximum, including:

[0030] divide the diffusion field intensity distribution into adaptive equal-width intensity intervals, wherein an interval width is determined according to a standard deviation of the intensity distribution, count a number of pixels in each intensity interval and divide the number of pixels by a total number of pixels to obtain an intensity probability distribution;

[0031] adaptively smooth the abnormal diffusion direction by using an anisotropic Gaussian kernel function, wherein variances of the Gaussian kernel function are respectively set along a gradient direction and a perpendicular gradient direction to obtain a direction probability distribution with direction selectivity;

[0032] calculate a product of the intensity probability distribution and a natural logarithm thereof to obtain an intensity information entropy, calculate a product of the direction probability distribution and a natural logarithm thereof to obtain a direction information entropy, multiply the intensity probability distribution by a corresponding gray level value to obtain an intensity mean value, and multiply the direction probability distribution by a corresponding direction angle to obtain a direction mean value;

[0033] Set an initial weight parameter, add the intensity information entropy and the direction information entropy by the weight parameter, iteratively update the weight parameter to make the weighted sum maximum, and ensure that the intensity mean value and the direction mean value weighted by the weight parameter meet the mean value range; stop iteration when the weight parameter difference of adjacent two iterations is less than a convergence threshold value, and obtain an optimal weight parameter; substitute the optimal weight parameter, the intensity probability and the direction probability into an exponential function to obtain a joint probability distribution;

[0034] On the joint probability distribution, start from the probability maximum point, and perform path search based on a direction continuity constraint to obtain a defect diffusion path.

[0035] In a second aspect of the embodiment of the present application, a semiconductor chip defect detection and positioning system is provided, comprising:

[0036] A first unit is configured to acquire an original image of a to-be-tested semiconductor chip and electrical test data of different sampling points;

[0037] A second unit is configured to calculate a local density value of each pixel point in the original image, and calculate a deviation from a regional density mean value, and cluster pixel points exceeding a deviation threshold value to form a defect candidate region;

[0038] A third unit is configured to perform variational mode decomposition on the electrical test data of the different sampling points to obtain a set of intrinsic mode functions, perform Hilbert transform on the intrinsic mode functions and calculate an instantaneous frequency, and mark sampling points exceeding a frequency threshold value as electrical characteristic abnormal points;

[0039] A fourth unit is configured to map image features of the defect candidate region to an electrical characteristic parameter space through a nonlinear transformation, convert the electrical characteristic abnormal points into an abnormal probability distribution in the image space through reverse mapping, and calculate a feature mutual information value of the electrical characteristic parameter space and the image space, and determine that a defect point exists at the position when the feature mutual information value is greater than a correlation threshold value;

[0040] A fifth unit is configured to calculate a diffusion field distribution with the defect point position as a source point, combine a spatiotemporal evolution law of the electrical characteristic abnormal points, and predict a diffusion path of the defect; and calculate a correlation coefficient of image features of each sampling point on the diffusion path to determine a severity of the defect.

[0041] In a third aspect of the embodiment of the present application,

[0042] An electronic device is provided, comprising:

[0043] A processor;

[0044] A memory for storing processor-executable instructions;

[0045] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0046] A fourth aspect of the embodiments of the present application,

[0047] A computer readable storage medium is provided, and computer program instructions are stored on the computer readable storage medium, and the computer program instructions are executed by a processor to implement the method described above.

[0048] The beneficial effects of the present application are as follows:

[0049] The present application combines the image features of the semiconductor chip with the electrical test data, realizes accurate detection and positioning of defects, greatly improves the detection accuracy and efficiency, reduces the missed detection rate and false detection rate, and meets the high requirements of advanced semiconductor processes on defect detection.

[0050] The present application analyzes the electrical test data by using variational mode decomposition and Hilbert transform, and establishes a mapping relationship between the image space and the electrical characteristic space through nonlinear transformation, realizes early identification of potential defects, provides an important basis for timely adjustment of process parameters, and effectively avoids the generation of batch defective products.

[0051] The present application can accurately evaluate the severity and development trend of defects through defect diffusion field analysis and spatiotemporal evolution law prediction, provides a visual representation of defect evolution, provides a scientific basis for continuous optimization of chip manufacturing process and yield improvement, and has significant practical value. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The flowchart of the semiconductor chip defect detection and positioning method of the embodiments of the present application is shown in the figure.

[0053] Figure 2 The defect candidate region nonlinear mapping and defect detection flowchart is shown in the figure. DETAILED DESCRIPTION

[0054] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described in detail below in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0055] The technical scheme of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0056] Figure 1A flowchart of a semiconductor chip defect detection and positioning method according to an embodiment of the present application is shown in FIG. 1, which includes the following steps: Figure 1

[0057] Obtaining an original image of a semiconductor chip to be tested and electrical test data of different sampling points;

[0058] Calculating a local density value of each pixel point in the original image, and calculating a deviation from a regional density mean value, and clustering pixel points exceeding a deviation threshold value to form a defect candidate region;

[0059] Performing variational mode decomposition on the electrical test data of the different sampling points to obtain a set of intrinsic mode functions, performing Hilbert transform on the intrinsic mode functions and calculating an instantaneous frequency, and marking sampling points exceeding a frequency threshold value as electrical characteristic abnormal points;

[0060] Mapping image features of the defect candidate region to an electrical characteristic parameter space through a nonlinear transformation, converting the electrical characteristic abnormal points into an abnormal probability distribution in the image space through reverse mapping, and calculating a feature mutual information value of the electrical characteristic parameter space and the image space, and determining that a defect point exists at the position when the feature mutual information value is greater than a correlation threshold value;

[0061] Calculating a diffusion field distribution with the defect point position as a source point, combining a spatiotemporal evolution law of the electrical characteristic abnormal points, and predicting a diffusion path of the defect; calculating a correlation coefficient of image features of each sampling point on the diffusion path to determine a severity of the defect.

[0062] In an optional embodiment, calculating a local density value of each pixel point in the original image, and calculating a deviation from a regional density mean value, and clustering pixel points exceeding a deviation threshold value to form a defect candidate region includes:

[0063] Dividing the original image into a plurality of local windows, calculating a standard deviation of gray values of pixel points in each local window, adding a product of the standard deviation of gray values and an adjustment coefficient to a unit base, and then multiplying the result by a first reference bandwidth parameter to obtain an adaptive bandwidth parameter; and substituting the adaptive bandwidth parameter into a Gaussian kernel function to calculate a local density value of a center pixel point;

[0064] Constructing a multi-scale feature pyramid based on the local density value, calculating a local density mean value and a local density standard deviation at each scale layer; subtracting the local density mean value of the center pixel point at each scale layer from the local density value of the center pixel point and dividing the result by the local density standard deviation to obtain a standardized deviation; and recording a position coordinate of the center pixel point as an abnormal point coordinate when the standardized deviation of any scale layer is greater than a deviation threshold value;

[0065] ​A density flow between adjacent abnormal points is calculated, the density flow being a rate of change of a local density value with a spatial position on a path connecting the two abnormal points, when a maximum value of the density flow is less than a second reference parameter, the two abnormal points are marked as belonging to a same abnormal region, and a set of abnormal points having a same abnormal region number is output as a defect candidate region.

[0066] In the embodiment, to more accurately analyze defect features in the image, a local density value needs to be calculated for each pixel point of the original image. The system divides the original image into a plurality of local windows of MxM size, M can be set as 15, so that each local window contains 225 pixel points. For edge regions of the image, a mirror filling method is used to ensure that all pixel points can be completely processed.

[0067] For each pixel point (x, y) in the image, a local window is established with the point as the center, and a standard deviation of gray scale values of all pixel points in the window is calculated. In actual application, for a region containing obvious edge texture, the standard deviation of gray scale in the local window is 25.6; and for a relatively smooth region, the standard deviation of gray scale is only 3.2. Based on the calculated standard deviation σ, an adaptive bandwidth parameter h is calculated in combination with an adjustment coefficient α (set as 0.5) and a unit base c (set as 1.0), and the calculation method is h = (c + α × σ) × h0, where h0 is a first reference bandwidth parameter, and is set as 5.0. For a texture-rich region, h is 14.8; and for a smooth region, h is 6.6, thereby achieving adaptive adjustment of the bandwidth parameter.

[0068] The adaptive bandwidth parameter h calculated is substituted into a Gaussian kernel function to calculate a local density value D(x, y) of the center pixel point (x, y). The Gaussian kernel function calculates weights according to spatial distances and gray scale value differences between pixel points, and the local density value is obtained after the weights are normalized. For example, for an industrial part surface image of 512x512 pixels, the local density value of a pixel point in a normal region is 0.85, and the local density value of a pixel point in a defect region is reduced to 0.35.

[0069] A multi-scale feature pyramid is constructed based on the calculated local density value, containing L scale layers (set as L = 3). The first layer uses an original resolution, and each subsequent layer is obtained by downsampling the previous layer, and the sampling ratio is set as 0.5. For an original 512x512 image, the resolutions of the three-layer pyramid are 512x512, 256x256 and 128x128, respectively.

[0070] On the scale layer i, a mean value μi of local density values of all pixel points on the layer is calculated i and a standard deviation σi of the local density values is calculated iFor example, in the first layer (original resolution), μ1 is 0.78, σ1 is 0.12; in the second layer, μ2 is 0.76, σ2 is 0.15; in the third layer, μ3 is 0.74, σ3 is 0.18. For each pixel point (x, y) on the scale layer i, its normalized deviation Z(x, y, i) = (D(x, y, i) - μ i ) / σ i is calculated, where D(x, y, i) represents the local density value of point (x, y) in the i-th layer.

[0071] A deviation threshold T is set, which can be T = 2.5, indicating that when the local density value of a pixel point deviates from the mean value by more than 2.5 times the standard deviation, the point is determined as an abnormal point. All pixel points in the feature pyramid are traversed, and if the normalized deviation Z(x, y, i) of a point on any scale layer i is greater than the threshold T, the coordinates (x, y) of the point are recorded as abnormal point coordinates. For example, for a surface pit defect, the normalized deviation of a point in the first layer is 2.8, which exceeds the set threshold 2.5, so it is marked as an abnormal point.

[0072] For all marked abnormal points, the density flow between them is calculated to determine whether they belong to the same abnormal region. For example, two adjacent abnormal points (2, 3) and (5, 8) are randomly selected, 10 points are uniformly sampled along the straight line path between the two points to obtain the sampling point coordinate sequence: (2, 3), (2.3, 3.5), (2.6, 4), (2.9, 4.5), (3.2, 5), (3.5, 5.5), (3.8, 6), (4.1, 6.5), (4.4, 7), (4.7, 7.5), (5, 8); the local density values of adjacent points in the sampling point sequence are calculated to obtain the local density sequence: 0.82, 0.78, 0.75, 0.71, 0.68, 0.65, 0.61, 0.58, 0.55, 0.52; the difference between adjacent density values in the local density sequence is calculated to obtain the density difference sequence: 0.04, 0.03, 0.04, 0.03, 0.03, 0.04, 0.03, 0.03, 0.03; the maximum value 0.04 is selected from the density difference sequence as the maximum density flow F, and the maximum density flow F is compared with the second reference parameter 0.3; since 0.04 is less than 0.3, it is determined that the two abnormal points belong to the same abnormal region.

[0073] Through the above density flow analysis, the system clusters all the abnormal points, and outputs the abnormal point set with the same abnormal region number as a defect candidate region. For example, in a 512x512 image, three defect candidate regions can be detected: region 1 contains 23 abnormal points, the center position is at (125, 240), and the area is about 78 square pixels; region 2 contains 18 abnormal points, the center position is at (320, 180), and the area is about 65 square pixels; and region 3 contains 32 abnormal points, the center position is at (410, 380), and the area is about 105 square pixels. These defect candidate regions can be further used for subsequent defect classification and evaluation.

[0074] In an alternative embodiment, the electrical test data of the different sampling points is subjected to variational mode decomposition to obtain a set of intrinsic mode functions, Hilbert transform is performed on the intrinsic mode functions, and the instantaneous frequency is calculated, and the sampling points exceeding the frequency threshold are marked as electrical property abnormal points, including:

[0075] The local variance of the electrical test data of the different sampling points is calculated in the sliding time window, the local variance is taken as a constraint condition to perform variational mode decomposition to obtain a set of intrinsic mode functions, the energy density of each intrinsic mode function in the set of intrinsic mode functions is calculated, the orthogonality index between the intrinsic mode functions with energy density greater than the energy threshold is calculated, and the intrinsic mode functions with the orthogonality index less than the orthogonality threshold are determined as effective intrinsic mode functions;

[0076] Hilbert transform is performed on the effective intrinsic mode functions to obtain an analytical signal, the generalized zero-crossing rate is calculated based on the zero-crossing characteristics of the analytical signal, and the time derivative of the generalized zero-crossing rate is divided by the frequency normalization coefficient to obtain the instantaneous frequency;

[0077] A spatial weight coefficient is assigned according to the position distance of the sampling points, the sum of the product of the spatial weight coefficient and the instantaneous frequency is divided by the sum of the spatial weight coefficients to obtain a reference frequency, and the frequency deviation value of the instantaneous frequency of different sampling points from the reference frequency is calculated, and the sampling points exceeding the frequency threshold are marked as electrical property abnormal points.

[0078] In this embodiment, an electrical property abnormal point detection method based on variational mode decomposition and Hilbert transform is proposed. The method performs variational mode decomposition on the electrical test data of different sampling points, performs Hilbert transform on the obtained intrinsic mode functions and calculates the instantaneous frequency, and marks the sampling points exceeding the frequency threshold as electrical property abnormal points.

[0079] When processing the electrical test data of different sampling points, the local variance in the sliding time window is calculated as a constraint condition. Taking the collected voltage data as an example, for a certain sampling point, the local variance is calculated every 5 ms within a 30 ms time window. At t = 100 ms, the local variance of the voltage data of this sampling point in the previous 30 ms is 0.0025 V 2 The local variance value is used as a constraint condition to perform variational mode decomposition.

[0080] In the variational mode decomposition process, the number of modes K is set to 5, the penalty factor a is set to 2000, the bandwidth control parameter t is set to 0, the tolerance is set to 10 -7 , and the maximum number of iterations is set to 500. Through these parameter settings, the original electrical test data is decomposed to obtain 5 intrinsic mode functions. Then the energy density of each intrinsic mode function is calculated, for example, the energy densities of the 5 intrinsic mode functions obtained are 0.85, 0.76, 0.45, 0.12 and 0.09 respectively. Set the energy threshold to 0.2, then the intrinsic mode functions with energy density greater than the threshold are the first three, that is, 0.85, 0.76 and 0.45.

[0081] For the three intrinsic mode functions with larger energy density, the orthogonality index between them is calculated. The orthogonality index calculation method is to divide the inner product of two intrinsic mode functions by the product of their norms. For example, the orthogonality index of the first and second intrinsic mode functions is 0.08, the orthogonality index of the first and third intrinsic mode functions is 0.15, and the orthogonality index of the second and third intrinsic mode functions is 0.22. Set the orthogonality threshold to 0.2, then the orthogonality index of the first and second intrinsic mode functions is less than the threshold, the orthogonality index of the first and third intrinsic mode functions is also less than the threshold, and the orthogonality index of the second and third intrinsic mode functions is greater than the threshold. Therefore, the first and second intrinsic mode functions are determined as effective intrinsic mode functions, and the third intrinsic mode function does not meet the orthogonality requirement and is discarded.

[0082] Perform Hilbert transform on the determined effective intrinsic mode functions. Taking the first effective intrinsic mode function as an example, it is represented as a real part signal, and its imaginary part signal is obtained by Hilbert transform, and the two form an analytic signal. According to the phase change characteristics of the analytic signal, the distribution of its zero-crossing points is calculated. In specific implementation, the time when the signal crosses zero from negative to positive is recorded, for example, zero-crossing occurs at t = 105 ms, 115 ms and 125 ms, then the generalized zero-crossing rate in this time period is 0.1 times / ms.

[0083] The time derivative of the generalized zero-crossing rate can be approximated by the difference between adjacent time points. For example, the generalized zero-crossing rate at t = 110 ms is 0.09 times / ms, and at t = 120 ms, it is 0.11 times / ms. Therefore, the time derivative of the zero-crossing rate at t = 115 ms is approximately 0.002 times / ms. 2 Assuming the frequency normalization coefficient is 0.004, the instantaneous frequency obtained is 0.5 Hz.

[0084] Considering the spatial distribution of the sampling points, spatial weight coefficients are assigned according to the distance between the sampling points. For example, for a 5x5 sampling point array, the center point has coordinates (3, 3), and other points have weights calculated according to the Euclidean distance from the center point. At a certain time, the sampling point at position (1, 1) has an instantaneous frequency of 52.3 Hz and a spatial weight of 0.2; the sampling point at position (2, 2) has an instantaneous frequency of 50.1 Hz and a spatial weight of 0.6; and the sampling point at position (3, 3) has an instantaneous frequency of 49.8 Hz and a spatial weight of 1.0. The sum of the product of the spatial weight and the instantaneous frequency is divided by the sum of the spatial weights to obtain the reference frequency, which is (52.3x0.2+50.1x0.6+49.8x1.0) / (0.2+0.6+1.0) = 50.22 Hz.

[0085] The frequency deviation value of each sampling point is calculated from the instantaneous frequency and the reference frequency. For example, the frequency deviation of position (1, 1) is |52.3-50.22| = 2.08 Hz, the frequency deviation of position (2, 2) is |50.1-50.22| = 0.12 Hz, and the frequency deviation of position (3, 3) is |49.8-50.22| = 0.42 Hz. Set the frequency threshold to 1.5 Hz, then the frequency deviation of position (1, 1) exceeds the threshold, and it is marked as an abnormal point, while the frequency deviations of positions (2, 2) and (3, 3) do not exceed the threshold and are considered normal points.

[0086] Through the above method, the abnormal points of electrical characteristics on the circuit board can be effectively identified, providing a basis for subsequent defect positioning and quality control. In practical applications, the parameter settings of variational mode decomposition, energy threshold, orthogonal threshold, and frequency threshold can be adjusted according to different types of electrical test data and application scenarios to obtain the best abnormal detection effect. In addition, the allocation method of spatial weight can also be optimized according to the layout characteristics of the actual circuit board to improve the accuracy of abnormal point detection.

[0087] Figure 2Fig. 1 is a schematic diagram of a nonlinear mapping of a defect candidate region and a defect detection process. In an optional embodiment, image features of a defect candidate region are mapped to an electrical characteristic parameter space through a nonlinear transformation, and electrical characteristic abnormal points are converted to an abnormal probability distribution in the image space through reverse mapping. A feature mutual information value of the electrical characteristic parameter space and the image space is calculated. When the feature mutual information value is greater than a correlation threshold, it is determined that a defect point exists at the position, including:

[0088] An image of a defect candidate region of a chip to be tested is obtained. The defect candidate region image is decomposed through discrete wavelet transformation to obtain a multi-scale feature vector.

[0089] A local covariance matrix of the multi-scale feature vector is calculated. Eigenvalue decomposition is performed on the local covariance matrix to obtain a feature subspace. A plurality of feature vectors with a contribution rate of the top 5% in the feature subspace are extracted to construct a mapping basis. A sum of a product of an inner product operation result of the mapping basis and a radial basis kernel function and an adaptive weight coefficient is determined as a nonlinear mapping function. The multi-scale feature vector is converted to an electrical characteristic parameter space through the nonlinear mapping function to obtain a parameter mapping vector.

[0090] A position potential is calculated based on a Mahalanobis distance between the parameter mapping vector and an electrical characteristic abnormal point. A neighboring potential is calculated based on an Euclidean distance between adjacent parameter mapping vectors. A conditional probability is determined according to a weighted combination of the position potential and the neighboring potential. A reverse mapping probability is obtained through maximum likelihood estimation of the conditional probability. The reverse mapping probability is converted to an image space to obtain a defect probability distribution.

[0091] A ratio of a first edge probability corresponding to the parameter mapping vector to a second edge probability corresponding to the defect probability distribution is taken as a feature mutual information value. When the feature mutual information value is greater than a correlation threshold, it is determined that a defect point exists in the defect candidate region. Spatial coordinate information of the defect point is output.

[0092] In the embodiment, a method of mapping image features of a defect candidate region to an electrical characteristic parameter space through a nonlinear transformation and determining a defect point through reverse mapping is described in detail.

[0093] To obtain candidate defect regions for the chip under test, high-resolution images can be acquired using a defect detection system during chip manufacturing. Taking a specific detection scenario as an example, the original image is 2048×2048 pixels with a grayscale value range of 0-255. This candidate defect region image is then subjected to multi-scale decomposition using discrete wavelet transform and a three-level decomposition using Daubechies wavelet basis functions, resulting in 10 sub-band images: LL3, LH3, HL3, HH3, LH2, HL2, HH2, LH1, HL1, and HH1. These 10 sub-band images are then rearranged and organized into a multi-scale feature vector F with dimensions of 2048×2048×10.

[0094] Calculate the local covariance matrix C for the multi-scale feature vector F. Specifically, for each pixel (i,j), extract the feature vector within its 7×7 neighborhood, construct a 10×49 local feature matrix, and calculate the covariance of this matrix to obtain a 10×10 local covariance matrix C(i,j). Perform eigenvalue decomposition on each local covariance matrix C(i,j) to obtain the eigenvalues ​​λ1≥λ2≥...≥λ 10 and their corresponding feature vectors v1, v2, ..., v 10 Calculate the contribution rate λ for each eigenvalue. k / Σλ i The mapping basis B is constructed by selecting the feature vectors with the top 5% cumulative contribution rates. In actual calculations, selecting the first 2-3 feature vectors is usually sufficient to meet the 5% contribution rate requirement. In this example, the first two feature vectors, v1 and v2, are selected to form the mapping basis B.

[0095] The nonlinear mapping function M is determined by multiplying the inner product of the mapping basis B and the radial basis function kernel function with the adaptive weight coefficients. The radial basis function kernel function is a Gaussian kernel function K(x,y) = exp(-||xy||). 2 / 2σ 2 ), where σ is the kernel width parameter, set to 1.5 times the standard deviation of the eigenvectors; in this example, σ = 0.12. The adaptive weight coefficients are obtained by minimizing the mapping error, with initial values ​​set to uniformly distributed random numbers. The final weight values ​​w = [0.35, 0.42, 0.23] are obtained through iterative optimization using gradient descent. The nonlinear mapping function M(F) = Σw i ·K(B,F i ), where F i Let be the i-th component of the eigenvector F. The multi-scale eigenvector F is transformed into the electrical characteristic parameter space through this nonlinear mapping function, resulting in a parameter mapping vector P with dimensions 2048×2048×3, corresponding to the three electrical characteristic parameters: leakage current, threshold voltage, and saturation current.

[0096] The positional potential energy U is calculated based on the Mahalanobis distance between the parameter mapping vector P and the pre-calibrated electrical characteristic anomaly point E. p The Mahalanobis distance takes into account the correlation between parameters and is calculated as d(P,E)=√((PE)). T S -1 (PE)), where S is the covariance matrix of the electrical characteristic parameters. In this example, the standard value of the outlier E is [2.5nA, 0.65V, 120μA], and the location potential U p =exp(-d 2 / 2). Simultaneously, the adjacent potential energy U is calculated based on the Euclidean distance between adjacent parameter mapping vectors. n The Euclidean distance is defined as d(P) i ,P j )=||P i -P j ||, adjacent potential energy U n =exp(-d 2 / 2). By weighted combination of position potential energy U p and adjacent potential energy U n Determine the conditional probability P c =αU p +(1-α)U n Where α is the weighting coefficient, set to 0.7. The conditional probability P... c The reverse mapping probability P is obtained through maximum likelihood estimation. r The reverse mapping uses Bayes' theorem, P r (F|P)=P c (P|F)·P(F) / P(P), where P(F) is the prior probability and P(P) is the marginal probability, obtained through histogram statistics. The inverse mapping probability Pr is transformed into the image space to obtain the defect probability distribution D, with a dimension of 2048×2048.

[0097] The first marginal probability H(P) corresponding to the parameter mapping vector P is obtained by calculating the entropy of P, H(P) = -Σp(p i )·log(p(p i )), where p(p i ) represents the value p in the parameter mapping vector P. i The probability of defect. The second marginal probability H(D) corresponding to the defect probability distribution D is obtained by calculating the entropy of D, H(D) = -Σp(d i )·log(p(d i )), where p(d i ) represents the median d of the defect probability distribution D. iThe feature mutual information value I(P;D) = H(P) + H(D) - H(P, D) is calculated, where H(P, D) is the joint entropy of P and D. In this example, H(P) = 3.45, H(D) = 2.87, and H(P, D) = 4.76, so I(P;D) = 1.56. Comparing the feature mutual information value with a preset correlation threshold 1.2, since 1.56 > 1.2, it is determined that the defect candidate region has a defect point. The spatial coordinate information of the defect point is output, which is (1024, 768) in this example, corresponding to the position below and to the right of the center of the image.

[0098] By the above method, a nonlinear mapping relationship between image features and electrical characteristic parameters is established, the defect position is determined by mutual information measurement, and the accuracy and reliability of chip defect detection are effectively improved. The method is suitable for defect detection in various chip manufacturing processes, and has good detection effect on small defects that are difficult to identify by traditional image processing methods.

[0099] In an optional embodiment, a diffusion field distribution is calculated with the defect point position as a source point, and the diffusion path of the defect is predicted in combination with the spatio-temporal evolution law of the electrical characteristic abnormal point. A correlation coefficient of the image features of each sampling point on the diffusion path is calculated to determine the severity of the defect, including:

[0100] The local material conductivity gradient and stress intensity of the defect point position are obtained, the diffusion coefficient is determined according to the ratio of the local material conductivity gradient to the stress intensity, and the product of the diffusion coefficient and the gradient of the defect point diffusion field intensity is taken as the diffusion field intensity distribution;

[0101] Electrical characteristic test data around the defect point are collected, a feature vector of the electrical characteristic test data is extracted by wavelet transform, and a local maximum point of the feature vector is marked as an abnormal point. The propagation speed and acceleration of the abnormal point are calculated, and the abnormal diffusion direction is determined based on the propagation speed and the acceleration;

[0102] The diffusion field intensity distribution is converted into an intensity probability distribution, the abnormal diffusion direction is converted into a direction probability distribution, the intensity probability distribution and the direction probability distribution are fused by using the maximum entropy criterion to obtain a joint probability distribution, and the diffusion path of the defect is determined along the direction with the maximum joint probability distribution;

[0103] A local image gray level co-occurrence matrix of each sampling point on the defect diffusion path is obtained, the gray level co-occurrence matrix is mapped into a gray level feature vector by sparse coding, and a correlation coefficient between the gray level feature vectors is calculated. When the correlation coefficient is greater than a feature correlation threshold, it is determined that the defect has a diffusion risk, and the position coordinates of the defect diffusion path are output.

[0104] The embodiment relates to a defect diffusion prediction method, which predicts a diffusion path of a defect and evaluates severity of the defect by analyzing electrical characteristics and diffusion characteristics of a defect point in a material. A local material conductivity gradient and stress intensity of a defect point position are obtained. A microelectrode array is used to measure the conductivity of a region around the defect point, and the measurement range is a 10mm*10mm region around the defect point, and the measurement interval is 0.1mm. For example, after a defect point is detected on the surface of a certain metal material, the microelectrode array is used to measure the conductivity of the defect point, and the conductivity is 2.3*10 6 S / m. The conductivity of a region 0.5mm away from the defect point is 2.5*10 6 S / m, and thus the conductivity gradient is calculated as 0.4*10 6 S / m*mm. Meanwhile, the stress intensity of the defect point is measured by a stress sensor, and the stress intensity is 120MPa. The diffusion coefficient is determined according to the ratio of the local material conductivity gradient to the stress intensity, and in this example, the ratio is 3.33*10 3 S / (m*MPa*mm). The diffusion coefficient D is calculated as 0.083mm 2 / s by multiplying the ratio by a material characteristic coefficient 0.025. The product of the diffusion coefficient and the gradient of the diffusion field intensity of the defect point is taken as the diffusion field intensity distribution. For example, when the diffusion field intensity gradient of the defect point is 0.5 / mm 2 , the diffusion field intensity distribution of the region is 0.0415 / mm 2 * s.

[0105] The electrical characteristic test data around the defect point are collected. A high-precision resistivity measuring instrument is used to collect the resistivity data of the region around the defect point at a 0.1mm grid interval, and a 100*100 size data matrix is formed. A feature vector of the electrical characteristic test data is extracted through wavelet transform. Specifically, Db4 wavelet is used for 3-layer decomposition, and the wavelet coefficients of the third layer are retained to form the feature vector. Local maximum points of the feature vector are marked as abnormal points by setting a threshold (for example, the feature value is greater than twice the average value). For example, three abnormal points with coordinates (2.3mm, 1.7mm), (2.5mm, 2.1mm) and (2.8mm, 2.4mm) are detected in the collected data. The propagation speed and acceleration of the abnormal points are obtained by calculating the position changes of the abnormal points at different time points (for example, every 5 minutes). For example, the abnormal point moves from the position (2.3mm, 1.7mm) to (2.5mm, 2.1mm) in 5 minutes, and the propagation speed is calculated as 0.0894mm / min and the direction is 58.0 degrees. The abnormal point moves to (2.8mm, 2.4mm) in 5 minutes, and the propagation speed is calculated as 0.0949mm / min and the direction is 45.0 degrees, and the acceleration is 0.0011mm / min 2The abnormal diffusion direction is determined based on the propagation speed and acceleration, and is about 51.5 degrees.

[0106] The diffusion field intensity distribution is converted into an intensity probability distribution, and the sum of the diffusion field intensity values of all points is 1 through normalization processing. For example, in a 5mm*5mm area around the defect, there are a total of 2500 sampling points, and the normalized intensity value of each point is between 0.0001 and 0.0015. The abnormal diffusion direction is converted into a direction probability distribution, and a von Mises distribution is used to generate a direction probability density function within a 360-degree range, with the measured abnormal diffusion direction as the center and a standard deviation of 15 degrees. The intensity probability distribution and the direction probability distribution are fused using the maximum entropy criterion to obtain a joint probability distribution. In a specific implementation, for each spatial point, the joint probability is the weighted product of the intensity probability and the direction probability corresponding to the point, and the weights are 0.6 and 0.4, respectively. Along the direction with the maximum joint probability, the next maximum probability point is iteratively found from the defect point at a step size of 0.1mm, forming a defect diffusion path. For example, starting from the defect point (0, 0), the calculated diffusion path points are (0.1mm, 0.08mm), (0.2mm, 0.17mm), (0.3mm, 0.25mm), etc.

[0107] The local image gray level co-occurrence matrix of each sampling point on the defect diffusion path is obtained. A high-resolution scanning electron microscope is used to obtain an image of a 1mm*1mm area around each sampling point, with a resolution of 0.01mm / pixel. The image gray level value is quantized to 16 levels, and the gray level co-occurrence matrix of each sampling point image is calculated, with a matrix size of 16*16. The gray level co-occurrence matrix is mapped to a gray level feature vector through sparse coding, and a K-SVD algorithm is used to generate a 32-dimensional feature vector with a dictionary size of 64 and a sparsity of 5. The correlation coefficient between the gray level feature vectors of adjacent sampling points is calculated using the Pearson correlation coefficient method. For example, the correlation coefficient between the gray level feature vectors of the first and second points on the diffusion path is 0.92, and the correlation coefficient between the second and third points is 0.88. When the correlation coefficient is greater than a feature correlation threshold (set to 0.85), it is determined that the defect has a diffusion risk, and the position coordinates of the defect diffusion path are output, such as [(0, 0), (0.1mm, 0.08mm), (0.2mm, 0.17mm), (0.3mm, 0.25mm)...].

[0108] Through the above method, the diffusion path of the defect in the material can be accurately predicted, and the severity can be evaluated, providing an important basis for material quality control and failure prevention. In practical applications, the prediction accuracy reaches more than 90%, and the average error of the diffusion path prediction is less than 0.2mm.

[0109] In an alternative embodiment, the diffusion field intensity distribution is converted into an intensity probability distribution, the abnormal diffusion direction is converted into a direction probability distribution, the intensity probability distribution and the direction probability distribution are fused by a maximum entropy criterion to obtain a joint probability distribution, and the defect diffusion path is determined along the direction in which the joint probability distribution is maximum, comprising:

[0110] The diffusion field intensity distribution is divided into adaptive equal-width intensity intervals, wherein the interval width is determined according to the standard deviation of the intensity distribution, the number of pixels in each intensity interval is counted and divided by the total number of pixels to obtain an intensity probability distribution;

[0111] An anisotropic Gaussian kernel function is used to adaptively smooth the abnormal diffusion direction, wherein the variance of the Gaussian kernel function is set along the gradient direction and the perpendicular gradient direction respectively to obtain a direction probability distribution with direction selectivity;

[0112] The product of the intensity probability distribution and its natural logarithm is calculated to obtain an intensity information entropy, the product of the direction probability distribution and its natural logarithm is calculated to obtain a direction information entropy, the intensity probability distribution is multiplied by the corresponding gray value to obtain an intensity mean value, and the direction probability distribution is multiplied by the corresponding direction angle to obtain a direction mean value;

[0113] An initial weight parameter is set, the intensity information entropy and the direction information entropy are weighted and added according to the weight parameter, the weight parameter is iteratively updated to maximize the weighted sum, while ensuring that the intensity mean value and the direction mean value weighted according to the weight parameter meet the mean value range; the iteration is stopped when the difference between the weight parameters of two adjacent iterations is less than a convergence threshold to obtain an optimal weight parameter; the optimal weight parameter and the intensity probability and the direction probability are substituted into an exponential function to obtain a joint probability distribution;

[0114] On the joint probability distribution, path searching is performed based on the direction continuity constraint to obtain the defect diffusion path.

[0115] In this embodiment, the method of how to convert the diffusion field intensity distribution and the abnormal diffusion direction into a probability distribution and fuse to obtain a joint probability distribution, and then determine the defect diffusion path will be described in detail.

[0116] The process of converting the intensity distribution of the diffusion field into the intensity probability distribution requires determining the appropriate interval width. The system determines the interval width based on the standard deviation σ of the intensity distribution of the diffusion field, and the value is 0.5σ. For example, when the standard deviation of the intensity distribution is 20, the interval width is 10. Assuming that the intensity value ranges from 0 to 255, it can be divided into 26 equal-width intervals [0, 9], [10, 19], etc. For each interval, the system counts the number of pixels falling into the interval, such as 1000 pixels for the interval [0, 9] and 2500 pixels for the interval [10, 19], and the total number of pixels is 100000, so the probabilities of the two intervals are 0.01 and 0.025, respectively. In this way, the mapping relationship from the intensity value to the probability, i.e., the intensity probability distribution, is obtained.

[0117] When the abnormal diffusion direction is converted into the direction probability distribution, an anisotropic Gaussian kernel function is used for smoothing processing. The system sets the variance parameter of the Gaussian kernel function according to the local gradient information. The variance along the gradient direction is set to a small value such as 3, and the variance perpendicular to the gradient direction is set to a large value such as 9, so that the kernel function constructed in this way has selectivity to the direction. In actual operation, for each point (x, y) in the image, the system calculates the gradient direction thereof, constructs a rotated Gaussian kernel, and the kernel has a small variance in the gradient direction and a large variance perpendicular to the gradient direction. The original direction data is convolved by applying this kernel function to obtain the smoothed direction field. For example, for a region with an original direction of 45°, after smoothing by applying the adaptive Gaussian kernel, the direction values of the region and its adjacent regions will be adjusted according to the kernel function weight, the direction consistency is enhanced, and the influence of noise is suppressed.

[0118] Calculating the information entropy and the mean value is a key step for fusing the probability distribution. The system processes the intensity probability distribution p i , calculates -∑p i ×ln(p i ) as the intensity information entropy. Similarly, the direction probability distribution q j is calculated, and -∑q j ×ln(q j ) is calculated as the direction information entropy. Taking the intensity probability distribution as an example, assuming that there are three interval probabilities of 0.2, 0.5, and 0.3, respectively, the intensity information entropy is -(0.2×ln(0.2)+0.5×ln(0.5)+0.3×ln(0.3)) = 1.03. The intensity mean value is calculated as ∑p i ×Z i , where Z i is the interval median value. Taking the above three intervals as an example, assuming that the interval median values are 5, 15, and 25, respectively, the intensity mean value is 0.2×5+0.5×15+0.3×25 = 16. The direction mean value is calculated in a similar manner, but the periodicity of the angle needs to be considered.

[0119] The weight optimization process of the maximum entropy criterion adopts an iterative manner. The system sets initial weights α = 0.5 and β = 0.5, and calculates a weighted information entropy H = α × H 强度 + β × H 方向 . At the same time, the system checks whether the weighted mean value is within a preset range, i.e. α × μ 强度 + β × μ 方向 needs to meet a certain range constraint, such as [10, 30]. If not, the weights are adjusted. In the adjustment, a gradient ascent method is adopted, and the weight update amount is proportional to the gradient of the entropy, and the step size is 0.01. For example, if the current α = 0.5 and β = 0.5, the gradient direction of the entropy indicates that it is more beneficial to increase α and decrease β, and then α = 0.51 and β = 0.49 are updated. In the iterative process, when the weight change of adjacent two iterations is less than 0.001, the optimization is stopped, and the optimal weights α* and β* are obtained. The final joint probability distribution is calculated as exp(α* × ln(p i + β* × ln(q j )), i.e. p i α* × q j β* .

[0120] The determination of the defect diffusion path is based on the joint probability distribution. The point with the maximum joint probability is selected as the starting point, such as the point (100, 150) with a probability value of 0.08. In the 8-neighborhood or a larger range around the point, a point with a probability value less than the maximum and a direction deviating from the current path direction by no more than 30° is searched as the next path point. The direction continuity constraint ensures the smoothness of the path. The search process may encounter multiple candidate points, and the system selects the point with the maximum joint probability and satisfying the direction constraint. For example, if the current point direction is 60°, the direction of the next point should be within the range of [30°, 90°]. The system continues this process until no new point satisfying the condition is found or a preset path length limit such as 200 points is reached. The final point sequence is the defect diffusion path.

[0121] In actual application, the above method can effectively handle noise and uncertainty in the diffusion field, reasonably fuse intensity and direction information by using the maximum entropy criterion, and accurately identify the defect diffusion path. For example, in a detection, the system successfully tracked a defect path with a length of about 180 pixels, the average joint probability of the path was 0.065, and the maximum deflection angle was 28°, which indicated that the path had high reliability and smoothness.

[0122] The semiconductor chip defect detection positioning system of the embodiment of the application comprises:

[0123] A first unit is configured to acquire an original image of a to-be-tested semiconductor chip and electrical test data of different sampling points.

[0124] The second unit is configured to calculate a local density value of each pixel point in the original image, calculate a deviation from a regional density average, and cluster pixel points exceeding a deviation threshold to form a defect candidate region.

[0125] The third unit is configured to perform variational mode decomposition on the electrical test data of the different sampling points to obtain a set of eigenmode functions, perform Hilbert transform on the eigenmode functions and calculate an instantaneous frequency, and mark sampling points exceeding a frequency threshold as electrical characteristic abnormal points.

[0126] The fourth unit is configured to map image features of the defect candidate region to an electrical characteristic parameter space through a nonlinear transformation, convert the electrical characteristic abnormal points to an abnormal probability distribution in the image space through reverse mapping, and calculate a feature mutual information value of the electrical characteristic parameter space and the image space, and determine that a defect point exists at the position when the feature mutual information value is greater than a correlation threshold.

[0127] The fifth unit is configured to calculate a diffusion field distribution with the defect point position as a source point, combine a spatiotemporal evolution rule of the electrical characteristic abnormal points to predict a diffusion path of the defect, and calculate a correlation coefficient of image features of each sampling point on the diffusion path to determine a severity of the defect.

[0128] In a third aspect, an electronic device is provided, including:

[0129] a processor;

[0130] a memory for storing processor-executable instructions;

[0131] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0132] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0133] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to perform various aspects of the present application.

[0134] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting and locating defects in a semiconductor chip, characterized in that, The method comprises the following steps: Obtaining the original image of the semiconductor chip to be tested and the electrical test data of different sampling points; Calculating the local density value of each pixel point in the original image, and calculating the deviation from the regional density average, and clustering the pixel points exceeding the deviation threshold to form a defect candidate region; Performing variational mode decomposition on the electrical test data of different sampling points to obtain a set of intrinsic mode functions, performing Hilbert transform on the intrinsic mode functions and calculating the instantaneous frequency, and marking the sampling points exceeding the frequency threshold as electrical characteristic abnormal points, comprising: Calculating the local variance of the electrical test data of different sampling points in a sliding time window, and performing variational mode decomposition on the local variance as a constraint condition to obtain a set of intrinsic mode functions, calculating the energy density of each intrinsic mode function in the set of intrinsic mode functions, and calculating the orthogonality index between the intrinsic mode functions with energy density greater than the energy threshold, and determining the intrinsic mode functions with the orthogonality index less than the orthogonality threshold as effective intrinsic mode functions; Performing Hilbert transform on the effective intrinsic mode functions to obtain an analytical signal, calculating the generalized zero-crossing rate based on the zero-crossing characteristics of the analytical signal, and obtaining the instantaneous frequency by dividing the time derivative of the generalized zero-crossing rate by the frequency normalization coefficient; According to the position distance of the sampling points, a spatial weight coefficient is assigned, the sum of the product of the spatial weight coefficient and the instantaneous frequency is divided by the sum of the spatial weight coefficients to obtain a reference frequency; and the frequency deviation value of the instantaneous frequency of different sampling points and the reference frequency is calculated, and the sampling points exceeding the frequency threshold are marked as electrical characteristic abnormal points; Mapping the image features of the defect candidate region to the electrical characteristic parameter space through nonlinear transformation, converting the electrical characteristic abnormal points to abnormal probability distribution in the image space through reverse mapping, calculating the feature mutual information value of the electrical characteristic parameter space and the image space, and determining that there is a defect point at the position when the feature mutual information value is greater than the correlation threshold; Taking the defect point position as a source point, the diffusion field distribution is calculated, the diffusion path of the defect is predicted by combining the spatio-temporal evolution law of the electrical characteristic abnormal points, the correlation coefficient of the image features of each sampling point on the diffusion path is calculated, and the severity of the defect is determined.

2. The method of claim 1, wherein, Calculating the local density value of each pixel point in the original image, and calculating the deviation from the regional density average, and clustering the pixel points exceeding the deviation threshold to form a defect candidate region, comprising: The original image is divided into a plurality of local windows, the standard deviation of the gray value of the pixel points in each local window is calculated, the product of the standard deviation of the gray value and the adjustment coefficient is added to the unit base, and then multiplied by the first reference bandwidth parameter to obtain an adaptive bandwidth parameter; the adaptive bandwidth parameter is substituted into the Gaussian kernel function to calculate the local density value of the center pixel point; Based on the local density value, a multi-scale feature pyramid is constructed, the local density average and the local density standard deviation are calculated at each scale layer; at each scale layer, the local density value of the center pixel point is subtracted from the local density average of the scale layer and then divided by the local density standard deviation to obtain a standardized deviation; when the standardized deviation of any scale layer is greater than the deviation threshold, the position coordinates of the center pixel point are recorded as abnormal point coordinates; Calculate the density flow between adjacent abnormal points, the density flow being the rate of change of local density value with spatial position on the path connecting the two abnormal points, and when the maximum value of the density flow is less than a second reference parameter, mark the two abnormal points as belonging to the same abnormal region, and output the set of abnormal points with the same abnormal region number as a defect candidate region.

3. The method of claim 1, wherein, Map image features of the defect candidate region to an electrical characteristic parameter space through a nonlinear transformation, convert electrical characteristic abnormal points to an abnormal probability distribution in the image space through reverse mapping, and calculate the feature mutual information value of the electrical characteristic parameter space and the image space, and when the feature mutual information value is greater than a correlation threshold, determine that a defect point exists at the position, including: Obtain a defect candidate region image of a chip to be tested, and decompose the defect candidate region image through discrete wavelet transform to obtain a multi-scale feature vector; Calculate the local covariance matrix of the multi-scale feature vector, perform eigenvalue decomposition on the local covariance matrix to obtain a feature subspace, extract a plurality of feature vectors with a contribution rate of the top 5% in the feature subspace to construct a mapping basis, determine the sum of the product of the inner product operation result of the mapping basis and the radial basis kernel function and an adaptive weight coefficient as a nonlinear mapping function, and convert the multi-scale feature vector to an electrical characteristic parameter space through the nonlinear mapping function to obtain a parameter mapping vector; Calculate the position potential based on the Mahalanobis distance between the parameter mapping vector and the electrical characteristic abnormal point, and calculate the adjacent potential based on the Euclidean distance between adjacent parameter mapping vectors; determine the conditional probability according to the weighted combination of the position potential and the adjacent potential, obtain the reverse mapping probability through maximum likelihood estimation of the conditional probability, and convert the reverse mapping probability to the image space to obtain a defect probability distribution; Determine the ratio of the first edge probability corresponding to the parameter mapping vector to the second edge probability corresponding to the defect probability distribution as the feature mutual information value; when the feature mutual information value is greater than a correlation threshold, determine that a defect point exists in the defect candidate region, and output the spatial coordinate information of the defect point.

4. The method of claim 1, wherein, Calculate the diffusion field distribution with the defect point position as a source point, and predict the diffusion path of the defect in combination with the spatiotemporal evolution law of the electrical characteristic abnormal point. Calculate the correlation coefficient of the image features of each sampling point on the diffusion path to determine the severity of the defect, including: Obtain the local material conductivity gradient and stress intensity of the defect point position, determine the diffusion coefficient according to the ratio of the local material conductivity gradient to the stress intensity, and take the product of the diffusion coefficient and the gradient of the defect point diffusion field intensity as the diffusion field intensity distribution; Collect electrical characteristic test data around the defect point, extract a feature vector of the electrical characteristic test data through wavelet transform, and mark the local maximum point of the feature vector as an abnormal point; calculate the propagation speed and acceleration of the abnormal point, and determine the abnormal diffusion direction based on the propagation speed and the acceleration; The diffusion field intensity distribution is converted into an intensity probability distribution, the abnormal diffusion direction is converted into a direction probability distribution, the intensity probability distribution and the direction probability distribution are fused by using a maximum entropy criterion to obtain a joint probability distribution, and a defect diffusion path is determined along a direction in which the joint probability distribution is maximum; A local image gray co-occurrence matrix of each sampling point on the defect diffusion path is obtained, the gray co-occurrence matrix is mapped into a gray feature vector through sparse coding, and a correlation coefficient between the gray feature vectors is calculated; when the correlation coefficient is greater than a feature correlation threshold value, it is determined that the defect has a diffusion risk, and a position coordinate of the defect diffusion path is output.

5. The method of claim 4, wherein, The diffusion field intensity distribution is converted into an intensity probability distribution, the abnormal diffusion direction is converted into a direction probability distribution, the intensity probability distribution and the direction probability distribution are fused by using a maximum entropy criterion to obtain a joint probability distribution, and a defect diffusion path is determined along a direction in which the joint probability distribution is maximum, comprising: The diffusion field intensity distribution is divided into adaptive equal-width intensity intervals, wherein the interval width is determined according to the standard deviation of the intensity distribution, the number of pixels in each intensity interval is counted and divided by the total number of pixels to obtain an intensity probability distribution; An anisotropic Gaussian kernel function is used to adaptively smooth the abnormal diffusion direction, wherein the variance of the Gaussian kernel function is set along the gradient direction and the perpendicular gradient direction respectively, and a direction probability distribution with direction selectivity is obtained; The product of the intensity probability distribution and its natural logarithm is calculated to obtain an intensity information entropy; the product of the direction probability distribution and its natural logarithm is calculated to obtain a direction information entropy; the intensity probability distribution is multiplied by the corresponding gray value to obtain an intensity mean value, and the direction probability distribution is multiplied by the corresponding direction angle to obtain a direction mean value; An initial weight parameter is set, the intensity information entropy and the direction information entropy are weighted and added according to the weight parameter, the weight parameter is iteratively updated to maximize the weighted sum, and the intensity mean value and the direction mean value weighted according to the weight parameter are ensured to meet a mean value range; the iteration is stopped when the difference between the weight parameters of two adjacent iterations is less than a convergence threshold value, and an optimal weight parameter is obtained; the optimal weight parameter, the intensity probability and the direction probability are substituted into an exponential function to obtain a joint probability distribution; On the joint probability distribution, a defect diffusion path is obtained through path searching based on a direction continuity constraint from a probability maximum point.

6. A semiconductor chip defect detection positioning system for implementing the method according to any one of claims 1 to 5, characterized in that Comprising: A first unit is configured to obtain an original image of a semiconductor chip to be tested and electrical test data of different sampling points; A second unit is configured to calculate a local density value of each pixel point in the original image, and calculate a deviation from a regional density mean value, and cluster pixel points exceeding a deviation threshold value to form a defect candidate region; A third unit is configured to perform variational mode decomposition on the electrical test data of the different sampling points to obtain a set of eigenmode functions, perform Hilbert transform on the eigenmode functions and calculate an instantaneous frequency, and mark sampling points exceeding a frequency threshold value as electrical characteristic abnormal points; The fourth unit is configured to map the image features of the defect candidate region to an electrical characteristic parameter space through a nonlinear transformation, convert the electrical characteristic abnormal points into an abnormal probability distribution in the image space through reverse mapping, and calculate a feature mutual information value of the electrical characteristic parameter space and the image space, so as to determine that the position exists a defect point when the feature mutual information value is greater than a correlation threshold value; The fifth unit is configured to calculate a diffusion field distribution with the defect point position as a source point, and predict a diffusion path of the defect by combining a spatio-temporal evolution rule of the electrical characteristic abnormal points; Calculate a correlation coefficient of the image features of each sampling point on the diffusion path, and determine the severity of the defect.

7. An electronic device, comprising: The method comprises: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored in the memory to execute the method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by the processor, implement the method of any one of claims 1 to 5.

Images