Secondary simulation identification method and system for microwave form on surface of glass wafer

Abstract

The invention discloses a secondary simulation identification method and system for a microwave form of a glass wafer surface, and relates to the technical field of wafer simulation, and the method comprises the steps: obtaining two-dimensional matrix data of wafer surface thickness distribution, and carrying out the mask matrix processing; performing two-dimensional Fourier transform on the processed two-dimensional matrix data to obtain surface thickness frequency domain distribution, identifying PSD power spectral density of the surface thickness frequency domain distribution, and extracting a characteristic frequency region of the PSD power spectral density; constructing a self-adaptive filter, wherein the self-adaptive filter comprises a low-pass filter and a high-pass filter which are set based on the self-adaptive cut-off frequency; and performing frequency domain separation on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data, and constructing a microwave form secondary simulation surface. The technical problem that in the prior art, the surface microwave form cannot be recognized within the range of the whole glass wafer is solved, and the technical effect of accurately recognizing and reconstructing the surface microwave form of the glass wafer is achieved.

Description

Technical Field

[0001] This invention relates to the field of wafer simulation technology, specifically to a secondary simulation identification method and system for microwave morphology on the surface of a glass wafer. Background Technology

[0002] In the processing and quality inspection of glass wafers, surface microwave morphology typically manifests as minute topographic features superimposed on the overall thickness undulations, with amplitudes at the nanoscale and distributed across the entire wafer. Existing detection and analysis methods are mainly based on overall wafer thickness distribution data or surface morphology measurements of local areas. Macroscopic undulations and warping components dominate the overall thickness data, making it difficult to effectively separate microwave morphology features. Furthermore, local surface morphology analysis is limited by the measurement range and cannot reflect the microwave morphology distribution across the entire glass wafer. Therefore, it is impossible to effectively identify and simulate the surface microwave morphology across the entire glass wafer. Summary of the Invention

[0003] This application provides a secondary simulation identification method and system for microwave morphology on the surface of a glass wafer, which addresses the technical problem that existing technologies cannot identify surface microwave morphology across the entire glass wafer.

[0004] In view of the above problems, this application provides a secondary simulation identification method and system for microwave morphology on the surface of a glass wafer.

[0005] The first aspect of this application provides a secondary simulation identification method for microwave morphology on the surface of a glass wafer, the method comprising:

[0006] Two-dimensional matrix data of wafer surface thickness distribution is obtained, and the two-dimensional matrix data is processed by mask matrix processing; a two-dimensional Fourier transform is performed on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, the PSD power spectral density of the surface thickness frequency domain distribution is identified, and the characteristic frequency region of the PSD power spectral density is extracted; an adaptive filter is constructed, the adaptive filter including a low-pass filter and a high-pass filter based on an adaptive cutoff frequency, the adaptive cutoff frequency being obtained by iterative search through a defined optimization objective function with the goal of minimizing the spatial autocorrelation length; frequency domain separation is performed on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data; a microwave morphology secondary simulation surface is constructed based on the high-frequency separation data and the low-frequency separation data.

[0007] A second aspect of this application provides a secondary simulation and recognition system for microwave morphology on the surface of a glass wafer, the system comprising:

[0008] The system includes a processing module for acquiring two-dimensional matrix data of wafer surface thickness distribution and performing mask matrix processing on the two-dimensional matrix data; an identification module for performing two-dimensional Fourier transform on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, identifying the PSD power spectral density of the surface thickness frequency domain distribution, and extracting the characteristic frequency region of the PSD power spectral density; a filter construction module for constructing an adaptive filter, which includes a low-pass filter and a high-pass filter based on an adaptive cutoff frequency, wherein the adaptive cutoff frequency is obtained through iterative search using a defined optimization objective function with the goal of minimizing the spatial autocorrelation length; a frequency domain separation module for performing frequency domain separation on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data; and a surface construction module for constructing a microwave morphology secondary simulation surface based on the high-frequency separation data and the low-frequency separation data.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] This application obtains two-dimensional matrix data of wafer surface thickness distribution, performs mask matrix processing on the two-dimensional matrix data, performs two-dimensional Fourier transform on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, identifies the PSD power spectral density of the surface thickness frequency domain distribution, and extracts the characteristic frequency region of the PSD power spectral density; constructs an adaptive filter, the adaptive filter including a low-pass filter and a high-pass filter based on an adaptive cutoff frequency, the adaptive cutoff frequency being obtained through iterative search using a defined optimization objective function with the goal of minimizing the spatial autocorrelation length; performs frequency domain separation on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data; and constructs a secondary simulation surface for microwave morphology based on the high-frequency separation data and low-frequency separation data. This invention solves the technical problem of existing technologies being unable to identify surface microwave morphology across the entire glass wafer. By performing frequency domain analysis on wafer surface thickness data and implementing secondary simulation processing based on an adaptive cutoff frequency to achieve high- and low-frequency separation, it achieves the technical effect of accurate identification and reconstruction of microwave morphology on the glass wafer surface. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic flowchart of a secondary simulation and identification method for microwave morphology on a glass wafer surface provided in this application embodiment;

[0013] Figure 2 This is a schematic diagram of a secondary simulation and recognition system for microwave morphology of a glass wafer surface provided in an embodiment of this application.

[0014] Figure reference numerals: Processing module 11, Identification module 12, Filter construction module 13, Frequency domain separation module 14, Surface construction module 15. Detailed Implementation

[0015] This application provides a secondary simulation identification method and system for microwave morphology on the surface of a glass wafer. It addresses the technical problem that existing technologies cannot identify surface microwave morphology across the entire glass wafer. By performing frequency domain analysis on the wafer surface thickness data and implementing secondary simulation processing based on an adaptive cutoff frequency to separate high and low frequencies, the technical effect of accurately identifying and reconstructing the microwave morphology on the glass wafer surface is achieved.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0018] Example 1, as Figure 1 As shown, this application provides a secondary simulation and identification method for microwave morphology on the surface of a glass wafer, the method comprising:

[0019] Step S100: Obtain two-dimensional matrix data of the thickness distribution on the wafer surface, and perform mask matrix processing on the two-dimensional matrix data.

[0020] In this embodiment, a thickness scanning device is first used to scan and measure the glass wafer surface point by point. The thickness scanning device forms a regular sampling grid on the wafer surface according to a preset sampling interval, and obtains the corresponding thickness measurement value at each sampling grid node. The thickness measurement values ​​of all sampling nodes are arranged in order of their spatial position in the wafer plane coordinate system, thereby obtaining two-dimensional matrix data characterizing the thickness distribution of the wafer surface. The matrix elements in the two-dimensional matrix data correspond one-to-one with the sampling positions on the wafer surface.

[0021] Next, a mask matrix is ​​generated based on the wafer's geometric shape. The mask matrix is ​​a two-dimensional matrix with the same dimensions as the two-dimensional matrix data. By using the wafer center as the center and the wafer's nominal diameter as the boundary in the two-dimensional matrix coordinate system, it is determined whether each matrix position is within the wafer's geometric outline. The mask matrix elements corresponding to the positions within the wafer's geometric outline are set as valid identifiers, while the mask matrix elements corresponding to the positions outside the wafer's geometric outline are set as invalid identifiers, thus forming a mask matrix based on the wafer's geometric boundary.

[0022] The mask matrix is ​​then multiplied element-wise with the two-dimensional matrix data, which masks the thickness data corresponding to the matrix positions outside the wafer geometry, leaving only the thickness data corresponding to the matrix positions inside the wafer geometry, thus obtaining the two-dimensional matrix data after mask matrix processing.

[0023] Step S200: Perform a two-dimensional Fourier transform on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, identify the PSD power spectral density of the surface thickness frequency domain distribution, and extract the characteristic frequency region of the PSD power spectral density.

[0024] In this embodiment, the processed two-dimensional matrix data is first used as input data. The two-dimensional discrete Fourier transform method is used to perform frequency domain transformation on the two-dimensional matrix data. By performing Fourier operations on the thickness data in two orthogonal spatial directions, the wafer surface thickness distribution is transformed from spatial domain representation to frequency domain representation, thereby obtaining the surface thickness frequency domain distribution that characterizes the wafer surface thickness variation. Each frequency component in the surface thickness frequency domain distribution corresponds to the variation of wafer surface thickness fluctuations at different spatial scales.

[0025] Next, the PSD power spectral density is calculated for the frequency domain distribution of the surface thickness. This is achieved by squaring the amplitude of each frequency component in the frequency domain distribution of the surface thickness.

[0026] Finally, frequency range analysis is performed on the PSD power spectral density to extract the characteristic frequency regions of the PSD power spectral density. The characteristic frequency regions include at least low-frequency characteristic regions below the first preset frequency threshold and high-frequency characteristic regions greater than or equal to the first preset frequency threshold. At the same time, the corresponding directional identifiers, frequency indices, and sampling intervals are recorded for the low-frequency characteristic regions and the high-frequency characteristic regions, respectively.

[0027] Furthermore, the method provided in the application embodiments also includes:

[0028] Extract the characteristic frequency region of the PSD power spectral density. The characteristic frequency region includes at least a low-frequency characteristic region less than a first preset frequency threshold and a high-frequency characteristic region greater than or equal to the first preset frequency threshold. Simultaneously, record the directional identifier, frequency index, and sampling interval of the low-frequency characteristic region and the high-frequency characteristic region.

[0029] In this embodiment, when extracting the characteristic frequency region, the actual physical distance between adjacent sampling points in the spatial domain is first determined based on the sampling interval used during the acquisition of the two-dimensional matrix data of the wafer surface thickness distribution. Then, combined with the number of data points in the row and column directions of the two-dimensional matrix data, the physical dimensions of the wafer surface in the corresponding directions are determined. Based on this, according to the correspondence between the spatial and frequency domains in the two-dimensional discrete Fourier transform, each frequency index in the surface thickness frequency domain distribution is mapped to a specific spatial frequency, thereby determining a specific spatial frequency for each frequency index in the PSD power spectral density.

[0030] After determining the spatial frequency, a first preset frequency threshold is set within the entire spatial frequency range. The spatial frequency magnitude corresponding to each frequency index in the PSD power spectral density is compared with the first preset frequency threshold. Frequency components with spatial frequencies less than the first preset frequency threshold are classified as low-frequency characteristic regions, and frequency components with spatial frequencies greater than or equal to the first preset frequency threshold are classified as high-frequency characteristic regions. This achieves a clear division of characteristic frequency regions in the PSD power spectral density. The low-frequency characteristic regions correspond to large-scale changes in wafer surface thickness fluctuations, while the high-frequency characteristic regions correspond to small-scale changes in wafer surface thickness fluctuations.

[0031] After dividing the low-frequency and high-frequency feature regions, parameter input processing was performed on both regions. Specifically, the positions of frequency components belonging to the low-frequency and high-frequency feature regions in the frequency domain distribution of surface thickness were recorded as frequency indices to identify the correspondence of each frequency component in the PSD power spectral density. Simultaneously, the sampling interval used during the acquisition of the two-dimensional matrix data of the wafer surface thickness distribution was also entered as a sampling interval parameter to maintain consistency between spatial domain sampling and frequency domain frequency.

[0032] After completing the input of frequency index and sampling interval, the directional identification of frequency components in low-frequency and high-frequency characteristic regions is input. By judging the offset of each frequency component relative to the zero-frequency component in the row and column directions in the frequency domain distribution of surface thickness, the corresponding main change direction is determined and recorded as the directional identification, thereby realizing the input of directional identification, frequency index and sampling interval of low-frequency and high-frequency characteristic regions.

[0033] Step S300: Construct an adaptive filter, which includes a low-pass filter and a high-pass filter based on an adaptive cutoff frequency. The adaptive cutoff frequency is obtained by iteratively searching through a defined optimization objective function with the goal of minimizing the spatial autocorrelation length.

[0034] In this embodiment, an adaptive filter is constructed based on a signal processing method for frequency domain separation. The adaptive filter includes a low-pass filter and a high-pass filter set based on the same adaptive cutoff frequency. The low-pass filter is used to retain the low-frequency components in the frequency domain distribution of surface thickness, and the high-pass filter is used to retain the high-frequency components in the frequency domain distribution of surface thickness. The adaptive cutoff frequency serves as the boundary between the two for frequency domain separation.

[0035] Before constructing the low-pass and high-pass filters, the cutoff frequency for frequency domain separation is adaptively determined by defining an optimization objective function that minimizes the spatial autocorrelation length. Specifically, candidate cutoff frequencies are selected within a preset cutoff frequency search range. Based on these candidate cutoff frequencies, corresponding low-pass and high-pass filters are constructed to perform frequency domain separation of the surface thickness frequency distribution. The separated high-frequency and low-frequency data samples are then converted to the spatial domain to calculate the spatial autocorrelation length. Subsequently, by comparing the spatial autocorrelation lengths under different candidate cutoff frequency conditions, the candidate cutoff frequencies are iteratively updated until a cutoff frequency that minimizes the spatial autocorrelation length is obtained. This cutoff frequency is then determined as the adaptive cutoff frequency.

[0036] After determining the adaptive cutoff frequency, corresponding low-pass and high-pass filters are constructed based on the adaptive cutoff frequency to obtain the adaptive filter.

[0037] Furthermore, in the method provided in the application embodiments, the adaptive cutoff frequency is obtained through iterative search using a defined optimization objective function with the objective of minimizing the spatial autocorrelation length, and further includes:

[0038] Define the cutoff frequency search range and search step size; initialize the adaptive filter, select candidate cutoff frequencies within the cutoff frequency search range, and perform frequency domain separation using low-pass and high-pass filters based on the candidate cutoff frequencies to obtain high-frequency and low-frequency separated data samples; perform inverse two-dimensional Fourier transform on the high-frequency and low-frequency separated data samples to obtain high-frequency and low-frequency spatial domain data; calculate the spatial autocorrelation length of the high-frequency spatial domain data; reselect candidate cutoff frequencies based on the calculated spatial autocorrelation length until an adaptive cutoff frequency with the goal of minimizing the spatial autocorrelation length is obtained.

[0039] In this embodiment, the cutoff frequency search range and search step size are first defined. Specifically, based on the sampling interval used during the acquisition of the two-dimensional matrix data of the wafer surface thickness distribution, the physical distance between adjacent sampling points in the spatial domain is determined. Then, combined with the number of data points in the row and column directions of the two-dimensional matrix data, the physical dimensions of the wafer surface in the corresponding directions are determined. Next, based on the correspondence between the spatial domain and frequency domain in the two-dimensional discrete Fourier transform, the resolvable spatial frequency range of the wafer surface thickness data is determined, and this spatial frequency range is used as the cutoff frequency search range. Simultaneously, based on the cutoff frequency search accuracy requirements, a search step size is set within the cutoff frequency search range to limit the variation interval between adjacent candidate cutoff frequencies.

[0040] Next, the adaptive filter is initialized, and candidate cutoff frequencies are selected within the cutoff frequency search range according to the search step size. When constructing corresponding low-pass and high-pass filters based on the candidate cutoff frequencies, the candidate cutoff frequencies serve as the boundaries for frequency domain separation. By judging the spatial frequencies corresponding to each frequency component in the surface thickness frequency domain distribution, the pass state or suppression state of the frequency components is set, thus forming low-pass and high-pass filters with the candidate cutoff frequencies as boundaries. Subsequently, the low-pass and high-pass filters are applied to the surface thickness frequency domain distribution to perform frequency domain separation processing, obtaining high-frequency and low-frequency separated data samples corresponding to the candidate cutoff frequencies.

[0041] Then, the high-frequency separated data samples and the low-frequency separated data samples were subjected to inverse two-dimensional Fourier transform to convert the frequency domain separation results into spatial domain representation, thereby obtaining the corresponding high-frequency spatial domain data and low-frequency spatial domain data. The high-frequency spatial domain data represents the high-frequency thickness fluctuation information retained by high-pass filtering.

[0042] The spatial autocorrelation length of the high-frequency spatial domain data is then calculated. In this process, the high-frequency spatial domain data is first mean-reduced to eliminate the influence of the overall bias on the correlation analysis results. Then, a two-dimensional autocorrelation calculation is performed on the mean-reduced high-frequency spatial domain data to obtain a two-dimensional autocorrelation function reflecting the correlation characteristics of the high-frequency spatial domain data under different spatial delay conditions. Next, a one-dimensional autocorrelation function along the zero-delay direction is extracted from the two-dimensional autocorrelation function. Based on the decay characteristics of the one-dimensional autocorrelation function with spatial delay, the spatial delay distance corresponding to the decay of the one-dimensional autocorrelation function value to 1 / e is determined, and this spatial delay distance is defined as the spatial autocorrelation length of the high-frequency spatial domain data.

[0043] Finally, based on the calculated spatial autocorrelation length, a new candidate cutoff frequency is selected, and the steps of frequency domain separation, inverse two-dimensional Fourier transform, and spatial autocorrelation length calculation are repeated until an adaptive cutoff frequency with the goal of minimizing the spatial autocorrelation length is obtained, and this adaptive cutoff frequency is determined as the cutoff frequency parameter for constructing the adaptive filter.

[0044] Furthermore, in the method provided in the application embodiments, calculating the spatial autocorrelation length of the high-frequency spatial domain data further includes:

[0045] The high-frequency spatial domain data is subjected to mean removal processing, and two-dimensional autocorrelation is calculated on the processed high-frequency spatial domain data to obtain a two-dimensional autocorrelation function; a one-dimensional autocorrelation function along the zero-delay direction is extracted from the two-dimensional autocorrelation function; the delay distance at which the autocorrelation function value decays to 1 / e is calculated based on the one-dimensional autocorrelation function as the autocorrelation length.

[0046] In this embodiment, the high-frequency spatial domain data is first subjected to mean removal processing. The values ​​of all spatial sampling points in the high-frequency spatial domain data are statistically summed and their average value is calculated. Then, the average value is subtracted point by point from the value corresponding to each spatial sampling point, so that the processed high-frequency spatial domain data satisfies the zero mean condition as a whole, thereby eliminating the influence of overall bias or DC component on the correlation calculation results.

[0047] After the mean removal process is completed, two-dimensional autocorrelation is calculated on the processed high-frequency spatial domain data. By using the two-dimensional spatial correlation operation method, the mean-removed high-frequency spatial domain data is multiplied point by point with itself under different spatial delay conditions and then summed to obtain the distribution of the correlation degree of the high-frequency spatial domain data in two-dimensional space as the spatial delay changes, thus forming a two-dimensional autocorrelation function. The function value in the two-dimensional autocorrelation function reflects the correlation strength of the high-frequency spatial domain data under the corresponding spatial delay.

[0048] Then, a one-dimensional autocorrelation function along the zero-delay direction is extracted from the two-dimensional autocorrelation function. Specifically, the delay value in one spatial direction of the two-dimensional autocorrelation function is fixed to zero, and the corresponding autocorrelation function curve is extracted only along the other spatial direction. This transforms the correlation change of the two-dimensional autocorrelation function in the zero-delay direction into a one-dimensional autocorrelation function, enabling quantitative analysis of spatial correlation scales.

[0049] Finally, the decay characteristics of the one-dimensional autocorrelation function are analyzed. The function value of the one-dimensional autocorrelation function at the zero-delay position is used as the initial reference value. The spatial delay distance corresponding to the first decay of the function value of the one-dimensional autocorrelation function to 1 / e of the initial reference value is found along the spatial delay direction. This spatial delay distance is determined as the spatial autocorrelation length of the high-frequency spatial domain data, thus completing the calculation of the spatial autocorrelation length of the high-frequency spatial domain data.

[0050] Furthermore, in the method provided in the application embodiments, after calculating the spatial autocorrelation length of the high-frequency spatial domain data, it further includes:

[0051] Calculate the spatial autocorrelation length of the low-frequency spatial domain data; by comparing the spatial autocorrelation length of the low-frequency spatial domain data with the spatial autocorrelation length of the high-frequency spatial domain data, determine whether the spatial autocorrelation length verification result passes.

[0052] In this embodiment, the same spatial autocorrelation analysis method as that used for high-frequency spatial domain data is employed to calculate the spatial autocorrelation length of low-frequency spatial domain data. Specifically, the low-frequency spatial domain data is first mean-removed to eliminate the influence of the overall bias on the correlation analysis results. Then, a two-dimensional autocorrelation calculation is performed on the mean-removed low-frequency spatial domain data to obtain a two-dimensional autocorrelation function describing the change in the correlation degree of the low-frequency spatial domain data under different spatial delay conditions. A one-dimensional autocorrelation function along the zero-delay direction is extracted from the two-dimensional autocorrelation function. Based on the decay characteristics of the one-dimensional autocorrelation function with spatial delay, the spatial delay distance corresponding to when the value of the one-dimensional autocorrelation function decays to 1 / e of its initial value at zero delay is determined, and this spatial delay distance is determined as the spatial autocorrelation length of the low-frequency spatial domain data.

[0053] After obtaining the spatial autocorrelation length of the low-frequency spatial domain data, the spatial autocorrelation length of the low-frequency spatial domain data is numerically compared with that of the high-frequency spatial domain data. By determining whether the spatial autocorrelation length of the low-frequency spatial domain data is greater than that of the high-frequency spatial domain data, the order relationship between the two spatial domain data in terms of spatial correlation scale is determined. If the spatial autocorrelation length of the low-frequency spatial domain data is greater than that of the high-frequency spatial domain data, the spatial autocorrelation length verification result is considered to be passed; otherwise, the spatial autocorrelation length verification result is considered to be failed.

[0054] Furthermore, in the method provided in the application embodiments, re-selecting the candidate cutoff frequency based on the calculated spatial autocorrelation length further includes:

[0055] Based on the cutoff frequency search range, at least two distinct candidate cutoff frequencies are simultaneously obtained; at least two spatial autocorrelation lengths of the at least two distinct candidate cutoff frequencies are calculated, and the cutoff frequency search range is updated according to the magnitude of the at least two spatial autocorrelation lengths; this process is repeated until the cutoff frequency search range is less than a convergence threshold, and the midpoint of the output search range below the convergence threshold is extracted as the adaptive cutoff frequency.

[0056] In this embodiment, at least two distinct candidate cutoff frequencies are first obtained simultaneously based on the cutoff frequency search range. Specifically, two distinct candidate cutoff frequencies are selected between the lower and upper limits of the cutoff frequency search range, and the two candidate cutoff frequencies correspond to different value positions within the cutoff frequency search range, so as to ensure that the candidate cutoff frequencies can cover different cutoff frequency values ​​within the current cutoff frequency search range.

[0057] Next, at least two spatial autocorrelation lengths for at least two distinct candidate cutoff frequencies are calculated. Specifically, for each candidate cutoff frequency, a corresponding low-pass filter and high-pass filter are constructed. By judging the spatial frequencies corresponding to each frequency component in the surface thickness frequency domain distribution, the pass state or suppression state of the frequency component is set, thus forming a low-pass filter and a high-pass filter with the candidate cutoff frequency as the boundary. Subsequently, the low-pass filter and high-pass filter are applied to the surface thickness frequency domain distribution to perform frequency domain separation, obtaining the corresponding high-frequency separated data samples. The high-frequency separated data samples are subjected to inverse two-dimensional Fourier transform to obtain high-frequency spatial domain data. The high-frequency spatial domain data is then subjected to mean removal processing, two-dimensional autocorrelation calculation, and extraction of a one-dimensional autocorrelation function along the zero-delay direction. Then, based on the one-dimensional autocorrelation function, the delay distance at which the function value decays to 1 / e is determined as the spatial autocorrelation length, thus obtaining the spatial autocorrelation length corresponding to each candidate cutoff frequency.

[0058] The cutoff frequency search range is then updated based on the magnitude of at least two spatial autocorrelation lengths. During this process, the candidate cutoff frequency with the smaller spatial autocorrelation length is selected as the target candidate cutoff frequency, and the frequency interval between the target candidate cutoff frequency and the other candidate cutoff frequency is used as the updated cutoff frequency search range. This narrows the updated cutoff frequency search range compared to the previous one, thus limiting the selection of subsequent candidate cutoff frequencies to the frequency interval with the smaller spatial autocorrelation length.

[0059] After completing one cutoff frequency search range update, the steps of obtaining at least two different candidate cutoff frequencies, calculating the corresponding spatial autocorrelation length, and updating the cutoff frequency search range are repeated. After each update, the interval width of the updated cutoff frequency search range is calculated. When the interval width of the cutoff frequency search range is less than the convergence threshold, the cutoff frequency search range is determined as the output search range below the convergence threshold, and the midpoint of the output search range is extracted as the adaptive cutoff frequency, where the midpoint is the average of the lower and upper limits of the output search range.

[0060] Step S400: Perform frequency domain separation on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data.

[0061] In this embodiment, a low-pass filter and a high-pass filter are applied to the surface thickness frequency domain distribution to perform frequency domain separation processing. Specifically, based on the spatial frequency magnitudes corresponding to each frequency component in the surface thickness frequency domain distribution, the low-pass filter retains frequency components with spatial frequencies lower than the adaptive cutoff frequency and suppresses frequency components with spatial frequencies greater than or equal to the adaptive cutoff frequency, thereby obtaining low-frequency separated data containing only low-frequency components. Simultaneously, the high-pass filter retains frequency components with spatial frequencies greater than or equal to the adaptive cutoff frequency and suppresses frequency components with spatial frequencies lower than the adaptive cutoff frequency, thereby obtaining high-frequency separated data containing only high-frequency components.

[0062] Through the frequency domain separation steps described above, different spatial frequency components in the surface thickness frequency domain distribution are divided into high-frequency separated data and low-frequency separated data according to the adaptive cutoff frequency.

[0063] Step S500: Construct a microwave morphology secondary simulation surface based on the high-frequency separation data and the low-frequency separation data.

[0064] In this embodiment, when constructing the microwave morphology secondary simulation surface, firstly, deviation calculation is performed on the two-dimensional matrix data of the wafer surface thickness distribution. Matrix data with deviations greater than a preset deviation threshold are marked, thereby dividing the two-dimensional matrix data into two-dimensional abnormal matrix data and two-dimensional normal matrix data. Subsequently, clustering processing is performed on the two-dimensional abnormal matrix data to obtain multiple two-dimensional abnormal matrix data corresponding to multiple defect types. Then, for each of the multiple two-dimensional abnormal matrix data, a corresponding adaptive cutoff frequency is configured, and frequency domain separation is performed on the two-dimensional abnormal matrix data based on the adaptive cutoff frequency to obtain multiple sets of high-frequency separation data and multiple sets of low-frequency separation data. At the same time, high-frequency separation data and low-frequency separation data are correspondingly obtained for the two-dimensional normal matrix data. Finally, based on the high-frequency and low-frequency separation data corresponding to the two-dimensional normal matrix data, and the multiple sets of high-frequency and low-frequency separation data corresponding to the two-dimensional abnormal matrix data, the microwave morphology secondary simulation surface is constructed.

[0065] Furthermore, in the method provided in the application embodiments, after obtaining the two-dimensional matrix data of the wafer surface thickness distribution, it further includes:

[0066] Deviation calculation is performed on the two-dimensional matrix data, and matrix data with deviations greater than a preset deviation threshold are marked to obtain two-dimensional abnormal matrix data and two-dimensional normal matrix data. The two-dimensional abnormal matrix data is clustered to obtain multiple two-dimensional abnormal matrix data corresponding to multiple defect types. Multiple adaptive cutoff frequencies are configured for the multiple two-dimensional abnormal matrix data to obtain multiple sets of high-frequency separation data and multiple sets of low-frequency separation data. Based on the high-frequency separation data and low-frequency separation data corresponding to the two-dimensional normal matrix data, and the multiple sets of high-frequency separation data and multiple sets of low-frequency separation data corresponding to the two-dimensional abnormal matrix data, a microwave morphology secondary simulation surface is constructed.

[0067] In this embodiment, when calculating the deviation of the two-dimensional matrix data, a point-by-point difference calculation method is adopted. First, a preset reference thickness value is determined, and the thickness value of each matrix position in the two-dimensional matrix data is subtracted from the preset reference thickness value to obtain the deviation value. Then, the deviation value is compared point by point with a preset deviation threshold. Matrix positions with deviations greater than the preset deviation threshold are marked, and the data set corresponding to the marked matrix positions is determined as two-dimensional abnormal matrix data, while the data set corresponding to the unmarked matrix positions is determined as two-dimensional normal matrix data, thereby obtaining two-dimensional abnormal matrix data and two-dimensional normal matrix data.

[0068] Next, clustering is performed on the two-dimensional anomaly matrix data. In this process, a similarity measure based on Euclidean distance is used. The spatial location parameters and thickness fluctuation feature parameters corresponding to each anomaly matrix position in the two-dimensional anomaly matrix data are constructed into feature representations for similarity calculation. The spatial location parameters are obtained by converting the row and column indices of the anomaly matrix position in the two-dimensional matrix data with the sampling interval, while the thickness fluctuation feature parameters are determined by the deviation values ​​corresponding to the anomaly matrix positions. For any two anomaly matrix positions, the difference between their spatial location parameters and the difference between their thickness fluctuation feature parameters are calculated, and these differences are used to form a difference vector. The Euclidean distance is calculated by summing the squares of the components of the difference vector and then taking the square root. The Euclidean distance is used as the feature similarity measure between the two anomaly matrix positions, where a smaller Euclidean distance indicates a higher feature similarity. After obtaining the feature similarity between the anomaly matrix positions, hierarchical clustering is used to cluster the two-dimensional anomaly matrix data. Specifically, each anomaly matrix location is initially treated as a cluster set. Based on feature similarity metrics, the two cluster sets with the smallest Euclidean distance are repeatedly selected and merged. The Euclidean distance between cluster sets is updated after each merge until the Euclidean distance between all cluster sets is greater than a preset clustering threshold, resulting in multiple cluster sets. Each cluster set is then identified as a two-dimensional anomaly matrix data point corresponding to a defect type, thus obtaining multiple two-dimensional anomaly matrix data points corresponding to multiple defect types. These defect types include sharp defect areas, texture anomaly areas, or local fluctuation anomaly areas. For each cluster set's corresponding two-dimensional anomaly matrix data point, its defect type is determined based on the amplitude characteristics of its thickness fluctuation feature parameters and the distribution characteristics of its spatial location parameters. When the absolute value of the deviation value within a cluster is generally large and the anomaly matrix positions show a locally concentrated distribution in terms of spatial location parameters, and the deviation values ​​of adjacent anomaly matrix positions vary significantly, the cluster is identified as a sharp defect region. When the anomaly matrix positions within a cluster show a continuously extended distribution in terms of spatial location parameters and the deviation values ​​show a recurring trend along the spatial direction, the cluster is identified as a texture anomaly region. When the anomaly matrix positions within a cluster show a discrete distribution in terms of spatial location parameters and the deviation values ​​show irregular fluctuations in a local range but do not have overall abrupt change characteristics, the cluster is identified as a local fluctuation anomaly region. This enables type labeling of multiple two-dimensional anomaly matrix data corresponding to multiple defect types.

[0069] Subsequently, multiple adaptive cutoff frequencies were configured for each of the multiple two-dimensional anomaly matrix datasets. This process first involves constructing several constraint features based on the multiple two-dimensional anomaly matrix datasets. These constraint features describe the limitations of the anomaly regions in terms of spatial scale, fluctuation amplitude, and distribution pattern. Then, the dominant frequency distribution of the multiple two-dimensional anomaly matrix datasets is analyzed. This dominant frequency distribution characterizes the main frequency range of thickness fluctuation energy concentration in each anomaly region. Finally, by combining the dominant frequency distribution and multiple constraint features, multiple adaptive cutoff frequencies are configured for each of the multiple two-dimensional anomaly matrix datasets.

[0070] After obtaining multiple adaptive cutoff frequencies, low-pass and high-pass filters are constructed based on the corresponding adaptive cutoff frequencies for each of the multiple two-dimensional anomaly matrix data. These filters are then applied to the corresponding data for frequency domain separation, resulting in multiple sets of high-frequency and low-frequency separated data corresponding to the multiple two-dimensional anomaly matrix data. Simultaneously, corresponding frequency domain separation processing is performed on the two-dimensional normal matrix data to obtain high-frequency and low-frequency separated data corresponding to the two-dimensional normal matrix data.

[0071] Finally, when constructing the microwave morphology secondary simulation surface based on the high-frequency and low-frequency separation data corresponding to the two-dimensional normal matrix data, and the multiple sets of high-frequency and low-frequency separation data corresponding to the two-dimensional abnormal matrix data, the spatial domain reconstruction method is adopted. The inverse two-dimensional Fourier transform is performed on the high-frequency and low-frequency separation data to obtain the corresponding high-frequency spatial domain data and low-frequency spatial domain data. Under the premise of maintaining the consistency of spatial coordinates, the high-frequency spatial domain data and low-frequency spatial domain data corresponding to the two-dimensional normal matrix data are combined. At the same time, the multiple sets of high-frequency spatial domain data and multiple sets of low-frequency spatial domain data corresponding to the multiple two-dimensional abnormal matrix data are superimposed on the corresponding regions according to their spatial positions, thereby constructing the microwave morphology secondary simulation surface.

[0072] Furthermore, in the method provided in the application embodiments, configuring multiple adaptive cutoff frequencies for the multiple two-dimensional anomaly matrix data respectively further includes:

[0073] Construct multiple constraint features based on the multiple two-dimensional anomaly matrix data; analyze the dominant frequency distribution of the multiple two-dimensional anomaly matrix data; configure multiple adaptive cutoff frequencies for the multiple two-dimensional anomaly matrix data respectively based on the dominant frequency distribution and the multiple constraint features.

[0074] In this embodiment, multiple constraint features are first constructed based on multiple two-dimensional anomaly matrix data. Specifically, for each two-dimensional anomaly matrix data, its region type is first determined as a sharp defect region, a texture anomaly region, or a local fluctuation anomaly region, and corresponding constraint features are extracted accordingly. When the region type is a sharp defect region, the gradient components are calculated by performing first-order difference calculations in the row and column directions of the two-dimensional anomaly matrix data, and the gradient components are synthesized to obtain the gradient magnitude as a constraint feature. When the region type is a texture anomaly region, the curvature is calculated by performing second-order difference calculations in the row and column directions of the two-dimensional anomaly matrix data, and synthesized as a constraint feature. When the region type is a local fluctuation anomaly region, a fixed-size sliding window is used to move point by point on the two-dimensional anomaly matrix data, and the dispersion of the thickness value is calculated within each window and the dispersion is used as a local fluctuation constraint feature, thereby constructing multiple constraint features that correspond one-to-one with the three region types.

[0075] After constructing multiple constraint features, the dominant frequency distributions of multiple two-dimensional anomaly matrix data are analyzed. Specifically, a two-dimensional discrete Fourier transform is performed on each two-dimensional anomaly matrix data to obtain the surface thickness frequency domain distribution, and the PSD power spectral density is calculated on the surface thickness frequency domain distribution. Subsequently, the frequency position with the highest energy is searched in the PSD power spectral density, and the frequency corresponding to this frequency position is determined as the dominant frequency distribution of the two-dimensional anomaly matrix data, thereby obtaining the dominant frequency distributions corresponding to each of the multiple two-dimensional anomaly matrix data.

[0076] Finally, multiple adaptive cutoff frequencies are configured for multiple two-dimensional anomaly matrix data based on the dominant frequency distribution and multiple constraint features. In this process, firstly, multiple two-dimensional neighborhood matrix data are obtained based on the multiple two-dimensional anomaly matrix data. These neighborhood matrix data are formed by neighborhood windows centered on the anomaly matrix positions, used to characterize the thickness fluctuations of the anomaly region within a local spatial range. Subsequently, the dominant frequency distribution and multiple constraint features are weighted according to the multiple two-dimensional neighborhood matrix data to obtain multiple adaptive cutoff frequencies corresponding to the multiple two-dimensional anomaly matrix data.

[0077] Furthermore, in the method provided in the application embodiments, configuring multiple adaptive cutoff frequencies for the multiple two-dimensional anomaly matrix data based on the dominant frequency distribution and the multiple constraint features, further includes:

[0078] The multiple two-dimensional anomaly matrix data are obtained based on multiple two-dimensional neighborhood matrix data using a neighborhood window; the dominant frequency distribution and the multiple constraint features are weighted according to the multiple two-dimensional neighborhood matrix data to obtain multiple adaptive cutoff frequencies corresponding to the multiple two-dimensional anomaly matrix data.

[0079] In this embodiment, multiple two-dimensional anomaly matrix data are first obtained based on a neighborhood window, forming multiple two-dimensional neighborhood matrix data. Specifically, a neighborhood window truncation method is used, with each anomaly matrix position in the two-dimensional anomaly matrix data as the center position. A fixed number of matrix positions are extended along both the row and column directions according to a preset neighborhood window size in the two-dimensional matrix data, truncating a local matrix region containing the anomaly matrix position and its surrounding adjacent matrix positions. This local matrix region is then determined as the corresponding two-dimensional neighborhood matrix data, thereby forming a corresponding two-dimensional neighborhood matrix data for each anomaly matrix position.

[0080] After obtaining multiple two-dimensional neighborhood matrix data, the dominant frequency distribution is weighted according to these data. Specifically, for each two-dimensional neighborhood matrix data, the difference between the maximum and minimum thickness values ​​within the neighborhood, as well as the mean square error of the thickness values ​​within the neighborhood, are calculated. These differences and mean square errors are used as quantitative indicators reflecting the intensity of thickness fluctuations within the neighborhood. Subsequently, based on the calculation results of the differences and mean square errors, initial weight values ​​are assigned to the dominant frequency distribution, and these initial weight values ​​are normalized to ensure that the weights corresponding to the dominant frequency distribution and the weights corresponding to subsequent constraint features meet a unified proportional constraint condition, thus completing the weight configuration of the dominant frequency distribution.

[0081] After configuring the weights of the dominant frequency distribution, multiple constraint features are weighted according to multiple two-dimensional neighborhood matrix data. Specifically, in each two-dimensional neighborhood matrix data, the average absolute value of the thickness value difference between adjacent matrix positions is calculated as the gradient change index, the average absolute value of the second-order difference of the thickness value between adjacent matrix positions is calculated as the curvature change index, and the mean square error of the thickness value within the neighborhood is calculated as the local fluctuation index. Subsequently, the gradient change index, curvature change index, and local fluctuation index are compared with their corresponding preset thresholds, and initial weight values ​​are assigned to multiple constraint features based on the comparison results. Then, the initial weight values ​​of multiple constraint features are normalized so that the weights corresponding to multiple constraint features and the weights corresponding to the dominant frequency distribution jointly satisfy the constraint condition of consistent weight sum, thereby completing the weight configuration of multiple constraint features.

[0082] After configuring the weights of the dominant frequency distribution and multiple constraint features, the frequency values ​​corresponding to the dominant frequency distribution and the weights corresponding to the multiple constraint features are weighted and calculated. The frequency values ​​corresponding to the dominant frequency distribution are multiplied by their corresponding weights, and the weights corresponding to the multiple constraint features are used in the combination calculation of the frequency values. This yields an adaptive cutoff frequency that corresponds one-to-one with each two-dimensional anomaly matrix data, and finally, multiple adaptive cutoff frequencies corresponding to multiple two-dimensional anomaly matrix data are obtained.

[0083] In summary, the embodiments of this application have at least the following technical effects:

[0084] This application obtains two-dimensional matrix data of wafer surface thickness distribution, performs mask matrix processing on the two-dimensional matrix data, performs two-dimensional Fourier transform on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, identifies the PSD power spectral density of the surface thickness frequency domain distribution, and extracts the characteristic frequency region of the PSD power spectral density; constructs an adaptive filter, the adaptive filter including a low-pass filter and a high-pass filter based on an adaptive cutoff frequency, the adaptive cutoff frequency being obtained through iterative search using a defined optimization objective function with the goal of minimizing the spatial autocorrelation length; performs frequency domain separation on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data; and constructs a secondary simulation surface for microwave morphology based on the high-frequency separation data and low-frequency separation data. This invention solves the technical problem of existing technologies being unable to identify surface microwave morphology across the entire glass wafer. By performing frequency domain analysis on wafer surface thickness data and implementing secondary simulation processing based on an adaptive cutoff frequency to achieve high- and low-frequency separation, it achieves the technical effect of accurate identification and reconstruction of microwave morphology on the glass wafer surface.

[0085] Example 2, based on the same inventive concept as the secondary simulation recognition method for microwave morphology of a glass wafer surface in the foregoing examples, such as... Figure 2 As shown, this application provides a secondary simulation and recognition system for the microwave morphology of a glass wafer surface. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0086] Processing module 11 is used to acquire two-dimensional matrix data of wafer surface thickness distribution and perform mask matrix processing on the two-dimensional matrix data; recognition module 12 is used to perform two-dimensional Fourier transform on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, identify the PSD power spectral density of the surface thickness frequency domain distribution, and extract the characteristic frequency region of the PSD power spectral density; filter construction module 13 is used to construct an adaptive filter, the adaptive filter including a low-pass filter and a high-pass filter based on an adaptive cutoff frequency, the adaptive cutoff frequency being obtained by iterative search through a defined optimization objective function with the goal of minimizing the spatial autocorrelation length; frequency domain separation module 14 is used to perform frequency domain separation on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data; surface construction module 15 is used to construct a microwave morphology secondary simulation surface based on the high-frequency separation data and the low-frequency separation data.

[0087] Furthermore, the system is also used to implement the following functions:

[0088] Extract the characteristic frequency region of the PSD power spectral density. The characteristic frequency region includes at least a low-frequency characteristic region less than a first preset frequency threshold and a high-frequency characteristic region greater than or equal to the first preset frequency threshold. Simultaneously, record the directional identifier, frequency index, and sampling interval of the low-frequency characteristic region and the high-frequency characteristic region.

[0089] Furthermore, the system is also used to implement the following functions:

[0090] Define the cutoff frequency search range and search step size; initialize the adaptive filter, select candidate cutoff frequencies within the cutoff frequency search range, and perform frequency domain separation using low-pass and high-pass filters based on the candidate cutoff frequencies to obtain high-frequency and low-frequency separated data samples; perform inverse two-dimensional Fourier transform on the high-frequency and low-frequency separated data samples to obtain high-frequency and low-frequency spatial domain data; calculate the spatial autocorrelation length of the high-frequency spatial domain data; reselect candidate cutoff frequencies based on the calculated spatial autocorrelation length until an adaptive cutoff frequency with the goal of minimizing the spatial autocorrelation length is obtained.

[0091] Furthermore, the system is also used to implement the following functions:

[0092] The high-frequency spatial domain data is subjected to mean removal processing, and two-dimensional autocorrelation is calculated on the processed high-frequency spatial domain data to obtain a two-dimensional autocorrelation function; a one-dimensional autocorrelation function along the zero-delay direction is extracted from the two-dimensional autocorrelation function; the delay distance at which the autocorrelation function value decays to 1 / e is calculated based on the one-dimensional autocorrelation function as the autocorrelation length.

[0093] Furthermore, the system is also used to implement the following functions:

[0094] Calculate the spatial autocorrelation length of the low-frequency spatial domain data; by comparing the spatial autocorrelation length of the low-frequency spatial domain data with the spatial autocorrelation length of the high-frequency spatial domain data, determine whether the spatial autocorrelation length verification result passes.

[0095] Furthermore, the system is also used to implement the following functions:

[0096] Based on the cutoff frequency search range, at least two distinct candidate cutoff frequencies are simultaneously obtained; at least two spatial autocorrelation lengths of the at least two distinct candidate cutoff frequencies are calculated, and the cutoff frequency search range is updated according to the magnitude of the at least two spatial autocorrelation lengths; this process is repeated until the cutoff frequency search range is less than a convergence threshold, and the midpoint of the output search range below the convergence threshold is extracted as the adaptive cutoff frequency.

[0097] Furthermore, the system is also used to implement the following functions:

[0098] Deviation calculation is performed on the two-dimensional matrix data, and matrix data with deviations greater than a preset deviation threshold are marked to obtain two-dimensional abnormal matrix data and two-dimensional normal matrix data. The two-dimensional abnormal matrix data is clustered to obtain multiple two-dimensional abnormal matrix data corresponding to multiple defect types. Multiple adaptive cutoff frequencies are configured for the multiple two-dimensional abnormal matrix data to obtain multiple sets of high-frequency separation data and multiple sets of low-frequency separation data. Based on the high-frequency separation data and low-frequency separation data corresponding to the two-dimensional normal matrix data, and the multiple sets of high-frequency separation data and multiple sets of low-frequency separation data corresponding to the two-dimensional abnormal matrix data, a microwave morphology secondary simulation surface is constructed.

[0099] Furthermore, the system is also used to implement the following functions:

[0100] Construct multiple constraint features based on the multiple two-dimensional anomaly matrix data; analyze the dominant frequency distribution of the multiple two-dimensional anomaly matrix data; configure multiple adaptive cutoff frequencies for the multiple two-dimensional anomaly matrix data respectively based on the dominant frequency distribution and the multiple constraint features.

[0101] Furthermore, the system is also used to implement the following functions:

[0102] The multiple two-dimensional anomaly matrix data are obtained based on multiple two-dimensional neighborhood matrix data using a neighborhood window; the dominant frequency distribution and the multiple constraint features are weighted according to the multiple two-dimensional neighborhood matrix data to obtain multiple adaptive cutoff frequencies corresponding to the multiple two-dimensional anomaly matrix data.

[0103] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0104] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A secondary simulation and identification method for microwave morphology on the surface of a glass wafer, characterized in that, The method includes: Two-dimensional matrix data of wafer surface thickness distribution is obtained, and the two-dimensional matrix data is processed by mask matrix. A two-dimensional Fourier transform is performed on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, the PSD power spectral density of the surface thickness frequency domain distribution is identified, and the characteristic frequency region of the PSD power spectral density is extracted. An adaptive filter is constructed, which includes a low-pass filter and a high-pass filter based on an adaptive cutoff frequency. The adaptive cutoff frequency is obtained by iteratively searching through a defined optimization objective function with the goal of minimizing the spatial autocorrelation length. Frequency domain separation is performed on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data; A microwave morphology secondary simulation surface is constructed based on the high-frequency separation data and the low-frequency separation data.

2. The method as described in claim 1, characterized in that, Extract the characteristic frequency region of the PSD power spectral density, wherein the characteristic frequency region includes at least a low-frequency characteristic region less than a first preset frequency threshold and a high-frequency characteristic region greater than or equal to the first preset frequency threshold; Simultaneously, the directional identifiers, frequency indexes, and sampling intervals of the low-frequency feature regions and the high-frequency feature regions are recorded.

3. The method as described in claim 1, characterized in that, The adaptive cutoff frequency is obtained through iterative search using a defined optimization objective function, with the goal of minimizing the spatial autocorrelation length. The method includes: Define the cutoff frequency search range and search step size; Initialize the adaptive filter, select a candidate cutoff frequency within the cutoff frequency search range, and perform frequency domain separation based on the low-pass and high-pass filters of the candidate cutoff frequencies to obtain high-frequency separated data samples and low-frequency separated data samples. The high-frequency separated data samples and the low-frequency separated data samples are subjected to inverse two-dimensional Fourier transform to obtain high-frequency spatial domain data and low-frequency spatial domain data; Calculate the spatial autocorrelation length of the high-frequency spatial domain data; Candidate cutoff frequencies are reselected based on the calculated spatial autocorrelation length until an adaptive cutoff frequency is obtained that minimizes the spatial autocorrelation length.

4. The method as described in claim 3, characterized in that, The method for calculating the spatial autocorrelation length of the high-frequency spatial domain data includes: The high-frequency spatial domain data is subjected to mean removal processing, and the processed high-frequency spatial domain data is subjected to two-dimensional autocorrelation calculation to obtain a two-dimensional autocorrelation function; Extract a one-dimensional autocorrelation function along the zero-delay direction from the two-dimensional autocorrelation function; The time delay required for the autocorrelation function value to decay to 1 / e based on the one-dimensional autocorrelation function is used as the autocorrelation length.

5. The method as described in claim 3, characterized in that, After calculating the spatial autocorrelation length of the high-frequency spatial domain data, the method further includes: Calculate the spatial autocorrelation length of the low-frequency spatial domain data; The spatial autocorrelation length of the low-frequency spatial domain data is compared with that of the high-frequency spatial domain data to determine whether the spatial autocorrelation length verification result passes.

6. The method as described in claim 3, characterized in that, The candidate cutoff frequency is reselected based on the calculated spatial autocorrelation length, and the method includes: Based on the cutoff frequency search range, at least two different candidate cutoff frequencies are obtained simultaneously. Calculate at least two spatial autocorrelation lengths for the at least two distinct candidate cutoff frequencies, and update the cutoff frequency search range based on the magnitude of the at least two spatial autocorrelation lengths; Repeat until the cutoff frequency search range is less than the convergence threshold, and extract the midpoint of the output search range below the convergence threshold as the adaptive cutoff frequency.

7. The method as described in claim 1, characterized in that, After obtaining the two-dimensional matrix data of the wafer surface thickness distribution, the method also includes: The deviation of the two-dimensional matrix data is calculated, and the matrix data with a deviation greater than a preset deviation threshold is marked to obtain two-dimensional abnormal matrix data and two-dimensional normal matrix data. Cluster the two-dimensional anomaly matrix data to obtain multiple two-dimensional anomaly matrix data corresponding to multiple defect types; Multiple adaptive cutoff frequencies are configured for the multiple two-dimensional anomaly matrix data respectively to obtain multiple sets of high-frequency separation data and multiple sets of low-frequency separation data; Based on the high-frequency and low-frequency separation data corresponding to the two-dimensional normal matrix data, and the multiple sets of high-frequency and low-frequency separation data corresponding to the two-dimensional abnormal matrix data, a microwave morphology secondary simulation surface is constructed.

8. The method as described in claim 7, characterized in that, The method includes configuring multiple adaptive cutoff frequencies for the multiple two-dimensional anomaly matrix data respectively, including: Construct multiple constraint features based on the multiple two-dimensional anomaly matrix data; Analyze the dominant frequency distribution of the multiple two-dimensional anomaly matrix data; Multiple adaptive cutoff frequencies are configured for the multiple two-dimensional anomaly matrix data based on the dominant frequency distribution and the multiple constraint features.

9. The method as described in claim 8, characterized in that, The method further includes configuring multiple adaptive cutoff frequencies for the multiple two-dimensional anomaly matrix data based on the dominant frequency distribution and the multiple constraint features, respectively: The multiple two-dimensional anomaly matrix data are obtained based on multiple two-dimensional neighborhood matrix data of the neighborhood window; The dominant frequency distribution and the multiple constraint features are weighted according to the multiple two-dimensional neighborhood matrix data to obtain multiple adaptive cutoff frequencies corresponding to the multiple two-dimensional anomaly matrix data.

10. A secondary simulation and recognition system for microwave morphology of a glass wafer surface, characterized in that, The system is used to execute a secondary simulation and identification method for microwave morphology of a glass wafer surface as described in any one of claims 1-9, and the system comprises: The processing module is used to acquire two-dimensional matrix data of the thickness distribution on the wafer surface and to perform mask matrix processing on the two-dimensional matrix data. The identification module is used to perform a two-dimensional Fourier transform on the processed two-dimensional matrix data to obtain the surface thickness frequency domain distribution, identify the PSD power spectral density of the surface thickness frequency domain distribution, and extract the characteristic frequency region of the PSD power spectral density. A filter construction module is used to construct an adaptive filter, which includes a low-pass filter and a high-pass filter based on an adaptive cutoff frequency. The adaptive cutoff frequency is obtained by iteratively searching through a defined optimization objective function with the goal of minimizing the spatial autocorrelation length. The frequency domain separation module is used to perform frequency domain separation on the low-pass filter and the high-pass filter to obtain high-frequency separation data and low-frequency separation data. A surface construction module is used to construct a microwave morphology secondary simulation surface based on the high-frequency separation data and the low-frequency separation data.

Images