Method and system for enhancing wafer detection in semiconductor manufacturing process

Abstract

The present invention provides a method and system for enhancing wafer detection in a semiconductor manufacturing process, which relates to the field of wafer detection technology, including emitting dual-frequency lasers to the wafer surface to form a controllable overlapping area, collecting beat frequency signals to obtain initial height information; performing adaptive filtering on the beat frequency signals, reconstructing the three-dimensional profile of the wafer surface, and generating an initial defect distribution map; determining a secondary detection area according to the initial defect distribution map for orthogonal scanning, identifying and eliminating pseudo defect signals, extracting defect features, and determining the defect cause type. The present invention improves the accuracy and efficiency of wafer defect detection, and helps to improve the quality of semiconductor manufacturing.

Description

Technical Field

[0001] The present invention relates to wafer detection technology, and in particular to a method and system for enhancing wafer detection in a semiconductor manufacturing process. Background Art

[0002] In the semiconductor manufacturing process, wafer surface inspection and analysis are key links to ensure device performance and yield. With the continuous improvement of integrated circuit process technology, the size of wafer surface defects is shrinking, and traditional single-frequency laser detection methods are difficult to meet the needs of high-precision and high-resolution detection. At the same time, due to the complex manufacturing environment, the reflected signal on the wafer surface is often interfered by noise, making it difficult to guarantee the accuracy and reliability of the detection results. The existing technology still has shortcomings in improving signal-to-noise ratio, phase stability and spatial resolution, which limits its application in sub-micron defect detection.

[0003] Current wafer surface inspection technologies are mostly performed in a two-dimensional scanning manner, with a single inspection track and limited coverage of complex defect areas. In addition, there is a lack of effective dynamic filtering and signal processing methods to suppress noise interference, making it difficult to accurately distinguish between real defects and false defect signals, resulting in the risk of misjudgment of inspection results. For the feature extraction and cause analysis of complex defects, the existing technology has failed to form a systematic method, and cannot effectively support quality control and defect traceability in the manufacturing process, which in turn affects wafer production efficiency and product quality.

[0004] Therefore, there is an urgent need for an enhanced wafer detection method and system that can provide high resolution and real-time dynamic adjustment, so as to effectively improve detection accuracy and efficiency and fully support defect distribution analysis and cause identification. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for enhancing wafer detection in a semiconductor manufacturing process, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] A method for enhancing wafer detection in a semiconductor manufacturing process is provided, comprising:

[0008] A dual-frequency laser with different frequencies is emitted to the surface of the wafer, the first frequency beam of the dual-frequency laser is subjected to wavefront modulation to form a first detection area, the second frequency beam of the dual-frequency laser is subjected to phase compensation to form a second detection area, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect a beat signal generated in the overlapping area, initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with an optical path difference calibration curve, the dual-frequency laser is controlled by a two-dimensional scanning galvanometer to form a spiral detection track on the wafer surface, the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat signal are calculated, and initial detection quality evaluation data is generated;

[0009] Adaptively filter the beat frequency signal, obtain signal spectrum characteristics through wavelet packet decomposition, dynamically adjust the filter parameters to suppress noise based on the signal spectrum characteristics and the initial detection quality assessment data, obtain a filtered signal, extract the difference frequency component of the filtered signal and perform phase demodulation, obtain continuous phase information in combination with the initial height information, reconstruct the three-dimensional profile of the wafer surface, record the signal strength value and phase noise level of the detection position, establish a signal intensity distribution map by using a double threshold segmentation method, fuse the signal intensity distribution map with the three-dimensional profile of the wafer surface through a data fusion algorithm, and generate an initial defect distribution map of the wafer surface;

[0010] The defect density gradient is calculated according to the initial defect distribution map, and the secondary detection area is determined as the area where the defect density gradient exceeds the preset density threshold. The secondary detection area is orthogonally scanned in a direction perpendicular to the spiral detection trajectory to obtain orthogonal scanning data; the orthogonal scanning data is feature matched with the initial defect distribution map, and a defect feature discrimination criterion is established in combination with the signal strength value and the phase noise level. Based on the defect feature discrimination criterion, pseudo defect signals are identified and eliminated to obtain optimized defect distribution data, and the geometric features and morphological features of the defects are extracted from the optimized defect distribution data, and a feature vector space is established. The distribution law and clustering characteristics of the feature parameters are analyzed in the feature vector space, the cause type of the defect is determined, and a detection report containing the defect location, feature parameters and cause analysis is generated.

[0011] In an optional embodiment,

[0012] A dual-frequency laser with different frequencies is emitted to the surface of the wafer, the first frequency beam of the dual-frequency laser is subjected to wavefront modulation to form a first detection area, the second frequency beam of the dual-frequency laser is subjected to phase compensation to form a second detection area, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect a beat signal generated in the overlapping area, and initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with an optical path difference calibration curve, including:

[0013] Emitting a dual-frequency laser to the surface of the wafer, separating a first frequency light beam and a second frequency light beam through a polarization beam splitter, measuring the light intensity distribution of the first frequency light beam and the second frequency light beam, and adjusting the splitting ratio of the polarization beam splitter according to the light intensity distribution to make the light intensities of the two beams equal;

[0014] Input the light intensity distribution data of the first frequency light beam into the iterative optimization algorithm, set the target light field distribution function to calculate the phase modulation amount between adjacent pixels, obtain the phase mask data through multiple iterative optimizations and load it into the spatial light modulator, and perform wavefront modulation on the first frequency light beam to form a first detection area;

[0015] Collecting a wavefront aberration spot array image of the second frequency light beam, calculating the centroid coordinates of each spot in the spot array image, converting the deviation between the centroid coordinates and the ideal position into wavefront gradient data, calculating the wavefront gradient data through a reconstruction matrix to obtain driving voltage data of the deformable reflector, and compensating the second frequency light beam according to the driving voltage data to form a second detection area;

[0016] respectively collecting light intensity images of the first detection area and the second detection area, calculating a two-dimensional cross-correlation function of the two light intensity images, determining a relative displacement of the two detection areas according to a peak position of the cross-correlation function, inputting the relative displacement into a feedback controller, and sending control signals output by the feedback controller to the two-dimensional translation stage and the angle adjustment device, respectively, to adjust the two detection areas to form an overlapping area on the wafer surface;

[0017] A photodetector array is used to sample the overlapping area to obtain a beat frequency signal, and the amplitude and initial phase information of the beat frequency signal are extracted, and a two-dimensional phase unwrapping algorithm is used to eliminate phase jumps to generate continuous spatial phase difference distribution data;

[0018] The piezoelectric ceramic translation stage is used to move the reflector according to the preset step length, and the displacement and phase difference data of the overlapping area are synchronously recorded at each position. The mapping relationship from displacement to phase difference is established by least squares fitting, and the polynomial coefficients of the calibration curve are obtained.

[0019] The spatial phase difference distribution data is substituted into a calculation formula containing calibration polynomial coefficients to obtain initial data characterizing the height of the wafer surface. The initial data is subjected to wavelet decomposition and specific scale coefficients are selected to reconstruct the denoised height data. The denoised height data is corrected in combination with a pre-calibrated system error compensation model to finally obtain the initial height information of the wafer surface.

[0020] In an optional embodiment,

[0021] The dual-frequency laser is controlled by a two-dimensional scanning galvanometer to form a spiral detection track on the wafer surface, the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat frequency signal are calculated, and the initial detection quality evaluation data is generated, including:

[0022] Receive the radius parameter of the wafer surface, set the maximum scanning range of the spiral trajectory according to the radius parameter, obtain the maximum scanning frequency and the maximum scanning angle of the galvanometer, determine the motion constraint condition of the galvanometer according to the maximum scanning frequency and the maximum scanning angle, calculate the pitch interval between adjacent spiral turns based on the maximum scanning range and the motion constraint condition, use the pitch interval as the step increment of the spiral trajectory, calculate the rotation angle and the scanning radius of the spiral trajectory according to the step increment, and generate a complete motion trajectory of the spiral trajectory;

[0023] Convert the complete motion trajectory into a deflection angle sequence of the galvanometer in the horizontal direction and the vertical direction; smooth the deflection angle sequence to obtain a control sequence of the galvanometer, wherein the control sequence includes an angle value, an angular velocity, and an angular acceleration of the galvanometer;

[0024] Input the control sequence into the driver of the two-dimensional scanning galvanometer in time order, control the dual-frequency laser to move along the spiral detection track on the wafer surface, collect the beat frequency signal in the overlapping area in real time, construct a time window, perform Fourier transform on the beat frequency signal in the window to obtain a power spectrum, extract the signal peak power and background noise power in the power spectrum, calculate the ratio of the two to obtain the signal-to-noise ratio of the beat frequency signal, and the length of the time window is determined according to the periodic characteristics of the beat frequency signal;

[0025] Performing phase extraction on the beat frequency signal, calculating the phase difference between adjacent sampling points, calculating the standard deviation of the phase difference to obtain a phase stability parameter, combining the phase stability parameter with the wavelength parameter of the dual-frequency laser to calculate the minimum resolvable height difference, and determining the spatial resolution parameter;

[0026] The signal-to-noise ratio is normalized to obtain a first evaluation parameter, the reciprocal of the phase stability parameter is normalized to obtain a second evaluation parameter, the reciprocal of the spatial resolution parameter is normalized to obtain a third evaluation parameter, and the three evaluation parameters are weighted according to a preset weight coefficient to obtain a comprehensive evaluation index;

[0027] When the comprehensive evaluation index meets the detection quality threshold requirement, the corresponding signal-to-noise ratio, phase stability parameter and spatial resolution parameter are combined to generate detection quality evaluation data.

[0028] In an optional embodiment,

[0029] The beat frequency signal is subjected to adaptive filtering processing, signal spectrum characteristics are obtained by wavelet packet decomposition, filtering parameters are dynamically adjusted based on the signal spectrum characteristics and the initial detection quality assessment data to suppress noise, a filtered signal is obtained, a difference frequency component of the filtered signal is extracted and phase demodulated, continuous phase information is obtained in combination with the initial height information, and a three-dimensional profile of the wafer surface is reconstructed, including:

[0030] Selecting a wavelet basis function to perform wavelet packet decomposition on the beat frequency signal to obtain a plurality of frequency band signals, calculating the frequency band energy ratio according to the plurality of frequency band signals to obtain a signal spectrum feature, determining an energy threshold based on the signal spectrum feature and a signal-to-noise ratio parameter in the initial detection quality assessment data, and performing noise reduction processing on the frequency band signal;

[0031] Determine the filtering window length according to the phase stability parameter in the initial detection quality assessment data, and perform adaptive filtering on the beat frequency signal to obtain a filtered signal;

[0032] Performing Hilbert transform on the filtered signal to obtain a transformed signal, combining the filtered signal with the transformed signal to extract a difference frequency component to obtain an instantaneous phase signal, detecting a phase jump point of the instantaneous phase signal and performing phase compensation to obtain an unwrapped phase, determining a phase integration starting point using the initial height information, and integrating the unwrapped phase to obtain continuous phase information;

[0033] The initial height difference is calculated by combining the continuous phase information with the equivalent wavelength of the dual-frequency laser, and the actual height difference between adjacent sampling points is corrected by combining the spatial resolution parameter in the initial detection quality assessment data, and the initial height distribution is obtained by accumulating the actual height difference between adjacent sampling points along the spiral scanning trajectory;

[0034] The initial height distribution is divided into a plurality of overlapping local reconstruction areas, the phase standard deviation of the local reconstruction area is calculated to obtain the local phase fluctuation, the local reconstruction area that needs to be smoothed is identified according to the phase stability parameter, the smoothing factor is calculated using the spatial resolution parameter, a least squares objective function including the smoothing factor is constructed, the local reconstruction area is optimized to obtain the smoothed local height distribution, the smoothed local height distribution is continuously transitioned and spliced ​​with the height distribution of the unsmoothed area in the overlapping area, and the three-dimensional contour of the wafer surface is reconstructed.

[0035] In an optional embodiment,

[0036] The signal strength value and phase noise level of the detection position are recorded, and a signal strength distribution map is established by using a double threshold segmentation method. The signal strength distribution map is fused with the three-dimensional contour of the wafer surface through a data fusion algorithm to generate an initial defect distribution map of the wafer surface, including:

[0037] Collecting signal strength values ​​at the detection position, performing time domain sampling on the signal strength values ​​to obtain a time series, constructing an envelope based on the time series, and calculating the variance within the overlapping time window of the envelope as the phase noise level at the detection position;

[0038] Selecting a defect-free area as a reference area at the detection position, extracting the signal strength value of the reference area to obtain a background noise reference value, using the background noise reference value to standardize the signal strength value at the detection position to obtain a standardized signal strength value, calculating the mean and standard deviation of the standardized signal strength value, and determining an upper threshold and a lower threshold based on the mean and standard deviation;

[0039] Using a lower limit threshold to perform threshold segmentation on the normalized signal intensity value to obtain a candidate defect area, extracting shape parameters and connectivity parameters of the candidate defect area, and screening the candidate defect area based on the shape parameters and connectivity parameters to obtain a preliminary defect area;

[0040] Extracting the area between the lower threshold and the upper threshold from the standardized signal strength value as the area to be determined, taking the ratio of the phase noise level of the area to be determined to the background noise reference value as the noise concentration, and determining the area to be determined that meets the threshold condition as the extended defect area based on the noise concentration;

[0041] Merging the preliminary defect area with the extended defect area, and performing boundary optimization processing on the merged area using a morphological algorithm to obtain a binary signal intensity distribution map;

[0042] Calculating the standard deviation of the phase sequence of the detection position to obtain a phase stability parameter, constructing a data fusion weight calculation function based on the phase stability parameter, and inputting the phase stability parameter into the data fusion weight calculation function to obtain a data fusion weight;

[0043] The signal intensity distribution diagram is numerically processed to obtain defect distribution data, the wafer surface three-dimensional contour data and the defect distribution data are normalized respectively, and the normalized wafer surface three-dimensional contour data and defect distribution data are weightedly fused using the data fusion weight, and a feature diagram characterizing the wafer surface defect distribution is generated based on the fused feature data.

[0044] In an optional embodiment,

[0045] The defect density gradient is calculated according to the initial defect distribution map, and the area where the defect density gradient exceeds the preset density threshold is determined as a secondary detection area. The secondary detection area is orthogonally scanned in a direction perpendicular to the spiral detection track to obtain orthogonal scanning data; the orthogonal scanning data is feature matched with the initial defect distribution map, and a defect feature discrimination criterion is established in combination with the signal strength value and the phase noise level. The pseudo defect signal is identified and eliminated based on the defect feature discrimination criterion, and the optimized defect distribution data is obtained, including:

[0046] Performing Gaussian smoothing on the initial defect distribution map to obtain a smoothed image, performing first-order difference and second-order difference operations on the smoothed image in orthogonal directions to obtain a first-order gradient matrix and a second-order gradient matrix, convolving the first-order gradient matrix with a Gaussian kernel to obtain a smoothed gradient, convolving the second-order gradient matrix with a Laplace kernel to obtain an edge enhancement gradient, fusing the smoothed gradient and the edge enhancement gradient to construct a morphological gradient operator, using the morphological gradient operator to calculate a defect density gradient, and determining an area where the defect density gradient exceeds a preset density threshold as a secondary detection area;

[0047] Calculating the trajectory tangent vector and curvature distribution according to the parameter equation of the spiral detection trajectory, establishing a sampling density adjustment coefficient based on the curvature distribution, calculating a normal vector perpendicular to the spiral detection trajectory according to the trajectory tangent vector, projecting the normal vector onto the wafer surface to generate an orthogonal scanning trajectory grid, locally encrypting the orthogonal scanning trajectory grid according to the sampling density adjustment coefficient, scanning the secondary detection area along the encrypted orthogonal scanning trajectory grid, and acquiring orthogonal scanning data including signal strength values ​​and phase data;

[0048] Performing wavelet decomposition on the orthogonal scanning data to obtain multi-layer decomposition coefficients, selecting several layers of decomposition coefficients according to the coefficient energy to form a feature matrix, respectively calculating the mutual correlation coefficient and mutual information entropy between the feature matrix and the initial defect distribution map, combining the mutual correlation coefficient and mutual information entropy into a matching evaluation index in a weighted manner, and establishing a spatial correspondence between the orthogonal scanning data and the initial defect distribution map based on the matching evaluation index;

[0049] Calculating the signal intensity ratio of each feature point in the orthogonal scanning data, normalizing the phase data to obtain the local phase fluctuation, calculating the spatial consistency coefficient of the neighborhood of the feature point in combination with the spatial correspondence, and establishing a defect feature discrimination criterion in combination with the signal intensity ratio, the local phase fluctuation and the spatial consistency coefficient;

[0050] Based on the defect feature discrimination criterion, pseudo-defect feature points are identified and eliminated, the defect boundary is extracted from the remaining feature points using a region growing algorithm, and the morphological parameters of the defect boundary are calculated to obtain optimized defect distribution data.

[0051] In an optional embodiment,

[0052] Extract the geometric features and morphological features of the defects from the optimized defect distribution data, establish a feature vector space, analyze the distribution law and clustering characteristics of the feature parameters in the feature vector space, determine the cause type of the defect, and generate a test report containing the defect location, feature parameters and cause analysis, including:

[0053] Perform multi-scale edge detection on the optimized defect distribution data, calculate an adaptive threshold coefficient based on the grayscale distribution of the defect distribution data, perform local binarization on the defect distribution data to obtain a binarized image, construct an edge response map based on the binarized image, use a non-maximum suppression algorithm to locate edge pixels in the edge response map to obtain an edge pixel point set, construct a dynamic connection strategy based on the intensity distribution of the edge pixel point set to repair the broken edge, and generate a continuous and complete defect boundary point set;

[0054] A polar coordinate reference system is constructed based on the defect boundary point set, a curvature distribution and a normal vector distribution of the defect boundary point set are calculated, an adaptive meshing is performed on the defect area according to the curvature distribution to obtain a mesh area set, local geometric features and morphological features are extracted in each mesh of the mesh area set, and surface feature parameters in the mesh are calculated in combination with the normal vector distribution to obtain an initial feature set;

[0055] Performing multi-layer wavelet decomposition on the initial feature set to obtain decomposition coefficients, calculating the energy distribution of the decomposition coefficients, selecting decomposition coefficients whose energy distribution is higher than a preset energy threshold to construct a feature vector space, performing feature dimension reduction in the feature vector space using an orthogonal transformation method to obtain a reduced dimension feature set, calculating a discrimination coefficient based on the reduced dimension feature set, and establishing a feature measurement criterion based on the discrimination coefficient;

[0056] In the feature vector space, the local density distribution of feature points is calculated based on the dimension reduction feature set, an adaptive search radius is determined according to the local density distribution, a density threshold function is constructed using the adaptive search radius and the feature measurement criterion, density connectivity between feature points is measured based on the density threshold function, and feature points with density connectivity are hierarchically clustered to obtain a defect type set;

[0057] A feature-cause mapping model is established according to the defect type set and feature measurement criteria, the mapping model is used to perform discriminant analysis on the reduced-dimensional feature set, the confidence of the discrimination result is calculated, the defect cause type is determined based on the discrimination result with the highest confidence, the position information corresponding to the defect boundary point set, the reduced-dimensional feature set and the defect cause type are integrated to generate a detection report, and the detection report is output to the process control system.

[0058] According to a second aspect of the embodiments of the present invention,

[0059] A system for enhancing wafer detection in a semiconductor manufacturing process is provided, comprising:

[0060] The first unit is used to emit dual-frequency lasers with different frequencies to the surface of the wafer, the first frequency beam of the dual-frequency laser forms a first detection area after wavefront modulation, and the second frequency beam of the dual-frequency laser forms a second detection area after phase compensation, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect the beat signal generated in the overlapping area, the initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with the optical path difference calibration curve, the dual-frequency laser is controlled by a two-dimensional scanning galvanometer to form a spiral detection track on the wafer surface, the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat signal are calculated, and initial detection quality evaluation data is generated;

[0061] The second unit is used to perform adaptive filtering on the beat frequency signal, obtain signal spectrum characteristics through wavelet packet decomposition, dynamically adjust the filtering parameters to suppress noise based on the signal spectrum characteristics and the initial detection quality assessment data, obtain a filtered signal, extract the difference frequency component of the filtered signal and perform phase demodulation, obtain continuous phase information in combination with the initial height information, reconstruct the three-dimensional profile of the wafer surface, record the signal strength value and phase noise level of the detection position, establish a signal intensity distribution map by a double threshold segmentation method, fuse the signal intensity distribution map with the three-dimensional profile of the wafer surface through a data fusion algorithm, and generate an initial defect distribution map of the wafer surface;

[0062] The third unit is used to calculate the defect density gradient according to the initial defect distribution map, determine the secondary detection area as the area where the defect density gradient exceeds the preset density threshold, perform orthogonal scanning on the secondary detection area in a direction perpendicular to the spiral detection trajectory, and obtain orthogonal scanning data; perform feature matching on the orthogonal scanning data and the initial defect distribution map, establish a defect feature discrimination criterion based on the signal strength value and the phase noise level, identify and eliminate pseudo-defect signals based on the defect feature discrimination criterion, obtain optimized defect distribution data, extract the geometric features and morphological features of the defects from the optimized defect distribution data, establish a feature vector space, analyze the distribution law and clustering characteristics of the feature parameters in the feature vector space, determine the cause type of the defect, and generate a test report containing the defect location, feature parameters and cause analysis.

[0063] According to a third aspect of the embodiments of the present invention,

[0064] An electronic device is provided, comprising:

[0065] processor;

[0066] a memory for storing processor-executable instructions;

[0067] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0068] A fourth aspect of the embodiments of the present invention is:

[0069] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0070] In this embodiment, by using dual-frequency laser and wavefront modulation technology, combined with the optical path adjustment system to form a controllable overlapping area, high-precision initial height measurement of the wafer surface is achieved. The spiral detection trajectory and two-dimensional scanning galvanometer are used to improve the detection efficiency and coverage. By calculating the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat frequency signal, initial detection quality assessment data is generated, providing a reliable basis for subsequent processing. Adaptive filtering and wavelet packet decomposition technology are used to effectively suppress noise and improve signal quality. By extracting the difference frequency component and phase demodulation, combined with the initial height information, accurate reconstruction of the three-dimensional profile of the wafer surface is achieved. Using the dual threshold segmentation method and data fusion algorithm, the initial defect distribution map of the wafer surface is generated, providing comprehensive information for defect detection. By calculating the defect density gradient and determining the secondary detection area, orthogonal scanning is used for in-depth detection, which improves the accuracy of defect detection. The defect feature discrimination criterion is established in combination with the signal intensity value and the phase noise level, and the pseudo-defect signal is effectively identified and eliminated. By analyzing the geometric and morphological features of defects and establishing a feature vector space, accurate classification and analysis of the causes of defects are achieved. Ultimately, a comprehensive inspection report is generated that includes defect locations, feature parameters, and cause analysis, providing an important basis for quality control and process optimization in the semiconductor manufacturing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A schematic flow chart of a method for enhancing wafer detection in a semiconductor manufacturing process according to an embodiment of the present invention;

[0072] Figure 2 Schematic diagram of the structure of a system for enhancing wafer detection in a semiconductor manufacturing process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0074] The technical solution of the present invention is described in detail with specific embodiments below. 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.

[0075] Figure 1 FIG. 1 is a flow chart of a method for enhancing wafer detection in a semiconductor manufacturing process according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0076] S101. Emitting dual-frequency lasers with different frequencies to the surface of the wafer, the first frequency beam of the dual-frequency laser forms a first detection area after wavefront modulation, and the second frequency beam of the dual-frequency laser forms a second detection area after phase compensation, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, using a photodetector array to collect the beat signal generated in the overlapping area, based on the phase difference of the beat signal combined with the optical path difference calibration curve to obtain the initial height information of the wafer surface, controlling the dual-frequency laser to form a spiral detection track on the wafer surface through a two-dimensional scanning galvanometer, calculating the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat signal, and generating initial detection quality evaluation data;

[0077] S102. Adaptively filter the beat frequency signal, obtain signal spectrum characteristics through wavelet packet decomposition, dynamically adjust the filter parameters to suppress noise based on the signal spectrum characteristics and the initial detection quality assessment data, obtain a filtered signal, extract the difference frequency component of the filtered signal and perform phase demodulation, obtain continuous phase information in combination with the initial height information, reconstruct the three-dimensional profile of the wafer surface, record the signal strength value and phase noise level of the detection position, establish a signal intensity distribution map using a double threshold segmentation method, fuse the signal intensity distribution map with the three-dimensional profile of the wafer surface through a data fusion algorithm, and generate an initial defect distribution map of the wafer surface;

[0078] S103. Calculate the defect density gradient according to the initial defect distribution map, determine the secondary detection area as the area where the defect density gradient exceeds the preset density threshold, perform orthogonal scanning on the secondary detection area in a direction perpendicular to the spiral detection trajectory, and obtain orthogonal scanning data; perform feature matching on the orthogonal scanning data and the initial defect distribution map, establish a defect feature discrimination criterion based on the signal strength value and the phase noise level, identify and eliminate pseudo-defect signals based on the defect feature discrimination criterion, and obtain optimized defect distribution data, extract the geometric features and morphological features of the defects from the optimized defect distribution data, establish a feature vector space, analyze the distribution law and clustering characteristics of the feature parameters in the feature vector space, determine the cause type of the defect, and generate a detection report containing the defect location, feature parameters and cause analysis.

[0079] Among them, dual-frequency laser refers to a composite light source composed of two laser beams with different frequencies. In this technical solution, dual-frequency laser is used for wafer surface detection. The detection area is formed by the first frequency beam and the second frequency beam respectively, and a beat signal is generated in the overlapping area for extracting height information. Wavefront modulation changes the phase, amplitude or waveform of the light wave so that it forms a specific wavefront structure during propagation. This technology is applied in the first frequency beam to optimize its detection performance and form a detection area. The beat signal is a modulated frequency signal generated by the interference effect when two light waves with slightly different frequencies overlap in space. The phase difference of the beat signal can be used to obtain the height information of the wafer surface.

[0080] The optical path difference calibration curve is a curve obtained through experiments or calculations, which reflects the corresponding relationship between the optical path difference and the phase difference. It is used to calibrate the measurement accuracy of the detection system and obtain the initial height information of the wafer surface. Phase stability is the stability of the signal phase over time or space. High phase stability helps to improve detection accuracy. Phase demodulation is the process of extracting phase information from the beat frequency signal. Combined with the height information, continuous phase information of the wafer surface can be obtained.

[0081] In an optional embodiment, dual-frequency lasers with different frequencies are emitted to the surface of the wafer, the first frequency beam of the dual-frequency laser is wavefront modulated to form a first detection area, the second frequency beam of the dual-frequency laser is phase compensated to form a second detection area, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect a beat signal generated in the overlapping area, and the initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with an optical path difference calibration curve, including:

[0082] Emitting a dual-frequency laser to the surface of the wafer, separating a first frequency light beam and a second frequency light beam through a polarization beam splitter, measuring the light intensity distribution of the first frequency light beam and the second frequency light beam, and adjusting the splitting ratio of the polarization beam splitter according to the light intensity distribution to make the light intensities of the two beams equal;

[0083] Input the light intensity distribution data of the first frequency light beam into the iterative optimization algorithm, set the target light field distribution function to calculate the phase modulation amount between adjacent pixels, obtain the phase mask data through multiple iterative optimizations and load it into the spatial light modulator, and perform wavefront modulation on the first frequency light beam to form a first detection area;

[0084] Collecting a wavefront aberration spot array image of the second frequency light beam, calculating the centroid coordinates of each spot in the spot array image, converting the deviation between the centroid coordinates and the ideal position into wavefront gradient data, calculating the wavefront gradient data through a reconstruction matrix to obtain driving voltage data of the deformable reflector, and compensating the second frequency light beam according to the driving voltage data to form a second detection area;

[0085] respectively collecting light intensity images of the first detection area and the second detection area, calculating a two-dimensional cross-correlation function of the two light intensity images, determining a relative displacement of the two detection areas according to a peak position of the cross-correlation function, inputting the relative displacement into a feedback controller, and sending control signals output by the feedback controller to the two-dimensional translation stage and the angle adjustment device, respectively, to adjust the two detection areas to form an overlapping area on the wafer surface;

[0086] A photodetector array is used to sample the overlapping area to obtain a beat frequency signal, and the amplitude and initial phase information of the beat frequency signal are extracted, and a two-dimensional phase unwrapping algorithm is used to eliminate phase jumps to generate continuous spatial phase difference distribution data;

[0087] The piezoelectric ceramic translation stage is used to move the reflector according to the preset step length, and the displacement and phase difference data of the overlapping area are synchronously recorded at each position. The mapping relationship from displacement to phase difference is established by least squares fitting, and the polynomial coefficients of the calibration curve are obtained.

[0088] The spatial phase difference distribution data is substituted into a calculation formula containing calibration polynomial coefficients to obtain initial data characterizing the height of the wafer surface. The initial data is subjected to wavelet decomposition and specific scale coefficients are selected to reconstruct the denoised height data. The denoised height data is corrected in combination with a pre-calibrated system error compensation model to finally obtain the initial height information of the wafer surface.

[0089] For example, the dual-frequency laser measurement system first emits a dual-frequency laser through a laser, which contains two beams of different frequencies. The laser uses a neodymium-doped yttrium aluminum garnet crystal as a gain medium, and generates a dual-frequency laser output with a frequency interval of 20MHz through an acousto-optic modulator, and the output power can be adjusted in the range of 0-2W.

[0090] The dual-frequency laser is separated by a polarization beam splitter, and the separated first-frequency beam and second-frequency beam are respectively monitored in real time by a light intensity detector. By collecting light intensity data over a certain period of time, the average light intensity ratio of the two beams is calculated. When the light intensity ratio deviates from 1.0, the controller achieves light intensity balance by adjusting the splitting ratio of the polarization beam splitter, so that the power error of the two beams is controlled within ±2%.

[0091] For the first frequency light beam, a wavefront modulation method based on iterative optimization is adopted. First, a light intensity detector is used to obtain the light intensity distribution data of the incident light beam, and the target light field is set as a Gaussian distribution function. The phase modulation amount between adjacent pixels is calculated by an iterative optimization algorithm, and the phase mask data is obtained after 50 iterative optimizations. The calculated phase data is loaded into a liquid crystal spatial light modulator with a resolution of 1920×1080 pixels to realize the wavefront modulation of the first frequency light beam, forming a first detection area with a uniformity better than 95%.

[0092] For the second frequency beam, an adaptive optical system is used for wavefront compensation. The wavefront aberration information is collected by a Shack-Hartmann wavefront sensor, and the microlens array spacing of the sensor is 400 microns and the focal length is 24 mm. The centroid coordinates of each spot are calculated by an image processing algorithm, and the deviation is obtained by comparing with the ideal position. The deviation is converted into wavefront gradient data, and the control voltage of the 37 actuators of the deformable mirror is calculated by a reconstruction algorithm. The response time of the deformable mirror is less than 1 millisecond, and the travel range is ±4 microns. The second frequency beam is compensated in real time to form a second detection area.

[0093] A high-speed camera is used to collect the light intensity distribution images of the two detection areas, with an image resolution of 1024×1024 pixels and a frame rate of 200fps. The two-dimensional cross-correlation function of the two images is calculated, and the relative displacement is determined according to the cross-correlation peak. The displacement is input into the PID feedback controller, and the controller outputs signals to drive the two-dimensional translation stage and the angle adjustment device respectively to achieve accurate overlap of the two detection areas. The area of ​​the overlapping area can reach 10mm×10mm.

[0094] A 16×16 array photodetector is used to sample the overlapping area. The detector has a responsivity of 0.5A / W and a bandwidth of 100MHz. The beat frequency signal is obtained through a synchronous sampling circuit to extract the amplitude and initial phase information of the signal. The two-dimensional phase unwrapping algorithm is combined to eliminate phase jumps and generate continuous spatial phase difference distribution data.

[0095] A piezoelectric ceramic stage with a resolution of 2nm was used to move the reflector in steps of 100nm, and data from 100 positions were collected in the range of 0-10 microns. The displacement and phase difference data of the overlapping area were recorded simultaneously, and the polynomial coefficients of the calibration curve were obtained by least squares fitting. The nonlinear error of the calibration curve was less than λ / 100.

[0096] The spatial phase difference distribution data is combined with the calibration curve coefficients to calculate and obtain the initial data characterizing the wafer surface height. The data is decomposed into 4 layers using the db4 wavelet basis, and the 3-4 layers of scale coefficients are selected for reconstruction to obtain the denoised height data. Combined with the pre-calibrated system error compensation model for correction, the wafer surface profile data with a measurement range of 10mm×10mm and a height resolution better than 5nm is finally obtained.

[0097] In this embodiment, by separating and balancing the dual-frequency laser, combined with wavefront modulation and adaptive optical compensation technology, a large uniform detection area is constructed, and the spatial resolution and signal-to-noise ratio of the measurement system are improved. The cross-correlation-based automatic alignment method and feedback control system are adopted to achieve accurate overlap of the detection area, and improve the measurement stability and reliability of the system. Combined with high-precision calibration and multi-scale data processing methods, accurate measurement of the wafer surface height is achieved, and the system has the characteristics of high measurement accuracy and strong anti-interference ability.

[0098] In an optional embodiment, controlling the dual-frequency laser to form a spiral detection track on the wafer surface through a two-dimensional scanning galvanometer, calculating the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat frequency signal, and generating initial detection quality evaluation data includes:

[0099] Receive the radius parameter of the wafer surface, set the maximum scanning range of the spiral trajectory according to the radius parameter, obtain the maximum scanning frequency and the maximum scanning angle of the galvanometer, determine the motion constraint condition of the galvanometer according to the maximum scanning frequency and the maximum scanning angle, calculate the pitch interval between adjacent spiral turns based on the maximum scanning range and the motion constraint condition, use the pitch interval as the step increment of the spiral trajectory, calculate the rotation angle and the scanning radius of the spiral trajectory according to the step increment, and generate a complete motion trajectory of the spiral trajectory;

[0100] Convert the complete motion trajectory into a deflection angle sequence of the galvanometer in the horizontal direction and the vertical direction; smooth the deflection angle sequence to obtain a control sequence of the galvanometer, wherein the control sequence includes an angle value, an angular velocity, and an angular acceleration of the galvanometer;

[0101] Input the control sequence into the driver of the two-dimensional scanning galvanometer in time order, control the dual-frequency laser to move along the spiral detection track on the wafer surface, collect the beat frequency signal in the overlapping area in real time, construct a time window, perform Fourier transform on the beat frequency signal in the window to obtain a power spectrum, extract the signal peak power and background noise power in the power spectrum, calculate the ratio of the two to obtain the signal-to-noise ratio of the beat frequency signal, and the length of the time window is determined according to the periodic characteristics of the beat frequency signal;

[0102] Performing phase extraction on the beat frequency signal, calculating the phase difference between adjacent sampling points, calculating the standard deviation of the phase difference to obtain a phase stability parameter, combining the phase stability parameter with the wavelength parameter of the dual-frequency laser to calculate the minimum resolvable height difference, and determining the spatial resolution parameter;

[0103] The signal-to-noise ratio is normalized to obtain a first evaluation parameter, the reciprocal of the phase stability parameter is normalized to obtain a second evaluation parameter, the reciprocal of the spatial resolution parameter is normalized to obtain a third evaluation parameter, and the three evaluation parameters are weighted according to a preset weight coefficient to obtain a comprehensive evaluation index;

[0104] When the comprehensive evaluation index meets the detection quality threshold requirement, the corresponding signal-to-noise ratio, phase stability parameter and spatial resolution parameter are combined to generate detection quality evaluation data.

[0105] For example, the radius parameter of the wafer needs to be obtained first. For example, a wafer with a diameter of 200 mm has a radius of 100 mm. According to the radius parameter, the maximum scanning range of the spiral trajectory is set to 110 mm, leaving a margin of 10 mm to ensure complete scanning coverage.

[0106] The galvanometer system uses a high-precision two-dimensional scanning galvanometer with a maximum scanning frequency of 500 Hz and a maximum mechanical scanning angle of plus or minus 15 degrees. Considering that the magnification of the optical system is 2 times, the actual maximum optical scanning angle is plus or minus 30 degrees. Based on these parameters, the motion constraints of the galvanometer can be determined: the angular velocity does not exceed 3140 radians per second, and the angular acceleration does not exceed 314,000 radians per square second.

[0107] After determining the motion constraints, calculate the pitch interval between adjacent spiral turns. Taking the actual application as an example, set the pitch interval to 0.1 mm as the step increment of the spiral trajectory. Starting from the center of the wafer, the scanning radius increases by 0.1 mm for each rotation. When the scanning radius reaches the maximum scanning range of 110 mm, a total of 1100 complete spiral trajectories are required.

[0108] When converting the spiral trajectory into a galvanometer control sequence, the spatial coordinates need to be converted into the horizontal and vertical deflection angles of the galvanometer. When converting, the equivalent focal length of the optical system is considered to be 500 mm. The converted angle sequence is smoothed using a fifth-order Bezier curve to generate a control sequence containing angle values, angular velocity, and angular acceleration. Smoothing can avoid sudden changes in the galvanometer motion and improve scanning stability.

[0109] The control sequence is executed by the galvanometer driver, and the sampling frequency is set to 10 kHz. During the scanning process, the beat frequency signal of the overlapping area is collected in real time. Taking the wavelength difference of the dual-frequency laser as 10 microns as an example, the time window length is set to 1 millisecond, and the beat frequency signal in the window is fast Fourier transformed. The ratio of the signal peak power and the background noise power is extracted from the power spectrum to obtain the signal-to-noise ratio of the beat frequency signal.

[0110] When extracting the phase of the beat frequency signal, the Hilbert transform method is used. The phase difference between adjacent sampling points is calculated, and the phase difference of 1000 sampling points is statistically analyzed to obtain the standard deviation of the phase difference as the phase stability parameter. Combined with the wavelength parameters of the dual-frequency laser, the minimum resolvable height difference of the system, that is, the spatial resolution parameter, can be determined.

[0111] The obtained signal-to-noise ratio, phase stability parameter, and spatial resolution parameter were normalized. The signal-to-noise ratio was normalized by dividing by the maximum signal-to-noise ratio value of 100 under ideal conditions; the phase stability parameter was normalized by taking the reciprocal and dividing by 0.01 radians; and the spatial resolution parameter was normalized by taking the reciprocal and dividing by 1 nanometer. The weight coefficients of the three parameters were set to 0.4, 0.3, and 0.3, respectively, and the comprehensive evaluation index was calculated.

[0112] When the comprehensive evaluation index is greater than 0.8, it is determined that the detection quality threshold requirement is met, and the corresponding parameter combination is used to generate detection quality evaluation data.

[0113] In this embodiment, a spiral detection trajectory can be used to achieve continuous scanning of the wafer surface, avoiding the acceleration and deceleration process in traditional raster scanning, and improving detection efficiency and stability. By reasonably setting the pitch interval, it is ensured that the spatial resolution of the detection meets the requirements. The time window analysis method is introduced, and a comprehensive detection quality evaluation system is established by combining the three key parameters of signal-to-noise ratio, phase stability and spatial resolution. Through parameter normalization and weighted calculation, objective comprehensive evaluation indicators are obtained. The real-time evaluation mechanism can be used to detect abnormalities in detection quality in a timely manner and ensure the reliability of the detection results. By adjusting the scanning parameters and evaluation thresholds, it can adapt to the detection needs of wafers of different specifications, and has good adaptability and scalability.

[0114] In an optional implementation, the beat frequency signal is adaptively filtered, the signal spectrum characteristics are obtained by wavelet packet decomposition, the filtering parameters are dynamically adjusted based on the signal spectrum characteristics and the initial detection quality assessment data to suppress noise, a filtered signal is obtained, the difference frequency component of the filtered signal is extracted and phase demodulated, continuous phase information is obtained in combination with the initial height information, and the three-dimensional profile of the wafer surface is reconstructed, including:

[0115] Selecting a wavelet basis function to perform wavelet packet decomposition on the beat frequency signal to obtain a plurality of frequency band signals, calculating the frequency band energy ratio according to the plurality of frequency band signals to obtain a signal spectrum feature, determining an energy threshold based on the signal spectrum feature and a signal-to-noise ratio parameter in the initial detection quality assessment data, and performing noise reduction processing on the frequency band signal;

[0116] Determine the filtering window length according to the phase stability parameter in the initial detection quality assessment data, and perform adaptive filtering on the beat frequency signal to obtain a filtered signal;

[0117] Performing Hilbert transform on the filtered signal to obtain a transformed signal, combining the filtered signal with the transformed signal to extract a difference frequency component to obtain an instantaneous phase signal, detecting a phase jump point of the instantaneous phase signal and performing phase compensation to obtain an unwrapped phase, determining a phase integration starting point using the initial height information, and integrating the unwrapped phase to obtain continuous phase information;

[0118] The initial height difference is calculated by combining the continuous phase information with the equivalent wavelength of the dual-frequency laser, and the actual height difference between adjacent sampling points is corrected by combining the spatial resolution parameter in the initial detection quality assessment data, and the initial height distribution is obtained by accumulating the actual height difference between adjacent sampling points along the spiral scanning trajectory;

[0119] The initial height distribution is divided into a plurality of overlapping local reconstruction areas, the phase standard deviation of the local reconstruction area is calculated to obtain the local phase fluctuation, the local reconstruction area that needs to be smoothed is identified according to the phase stability parameter, the smoothing factor is calculated using the spatial resolution parameter, a least squares objective function including the smoothing factor is constructed, the local reconstruction area is optimized to obtain the smoothed local height distribution, the smoothed local height distribution is continuously transitioned and spliced ​​with the height distribution of the unsmoothed area in the overlapping area, and the three-dimensional contour of the wafer surface is reconstructed.

[0120] Exemplarily, when processing the beat frequency signal, the wavelet packet decomposition method is first used to perform multi-scale analysis on the signal. Specifically, the db4 wavelet basis function is selected to perform 8-layer wavelet packet decomposition on the beat frequency signal to obtain 256 frequency band signals. The energy value is calculated for each frequency band signal, and the ratio of each frequency band energy to the total energy is used as the spectrum feature. Combined with the signal-to-noise ratio parameter obtained by the initial detection, the energy threshold is set to the inverse of the signal-to-noise ratio, and the frequency band signals below the threshold are subjected to soft threshold noise reduction processing. For example, when the signal-to-noise ratio is 10, the energy threshold is set to 0.1, and the frequency band signals with an energy ratio lower than 0.1 will be suppressed.

[0121] Adaptive filtering is performed based on the phase stability parameter in the initial detection. The higher the phase stability, the shorter the filter window is selected to retain the detail information; when the phase stability is low, the window length is increased to enhance the filtering effect. Taking the phase stability of 0.95 as an example, a 15-point Hanning window is selected for filtering; when the phase stability drops to 0.8, the window length is increased to 31 points.

[0122] The filtered signal is Hilbert transformed to obtain the analytical signal, and the original signal is combined with the transformed signal to extract the difference frequency component. The phase unwrapping is achieved by detecting the mutation point of the phase signal to perform 2π phase compensation. The initial height information is used to determine the starting point of the integration, and the phase is accumulated along the time axis to obtain continuous phase information.

[0123] The initial value of the height difference is obtained by multiplying the continuous phase information with the equivalent wavelength of the dual-frequency laser, and then corrected by combining the spatial resolution parameter. For example, when the spatial resolution is 0.1 micron, the calculated height difference between adjacent points is multiplied by the correction factor 1.05 to obtain the actual height difference. The corrected height difference is accumulated along the spiral scanning path to reconstruct the initial height distribution.

[0124] The obtained height distribution is divided into local areas with an overlap rate of 50%, and each area is 64×64 points. The standard deviation of the phase value in each area is calculated. When the standard deviation exceeds 2 times the phase stability parameter, the area is smoothed. The smoothing factor is positively correlated with the spatial resolution. The higher the resolution, the smaller the smoothing factor. The local least squares method is used to optimize the area that needs to be smoothed, and a smooth transition is achieved by weighted averaging in the overlapping area, and finally the accurate three-dimensional profile of the wafer surface is reconstructed.

[0125] In this embodiment, by combining wavelet packet decomposition and adaptive filtering, accurate identification and effective suppression of noise in the beat frequency signal are achieved, and the accuracy and reliability of phase extraction are improved. The processing parameters are dynamically adjusted based on the signal spectrum characteristics and the detection quality evaluation data, so that the noise suppression process has strong adaptability. The strategy of local reconstruction and smooth transition is adopted to effectively solve the problem of error accumulation in the process of surface profile reconstruction of large-size wafers. By optimizing the overlapping area, the continuity and smoothness between adjacent areas are ensured, and the overall quality of the reconstruction result is improved. The spatial resolution parameter is introduced to correct the height difference and smoothing factor, which improves the quantitative accuracy of the reconstruction result. The local area identification and processing combined with the phase stability parameter ensure that the reconstruction result has a good noise suppression effect while maintaining the detail features. The entire processing process makes full use of the initial detection quality evaluation data and realizes the optimal configuration of the processing parameters.

[0126] In an optional implementation, recording the signal strength value and phase noise level of the detection position, using a double threshold segmentation method to establish a signal strength distribution map, fusing the signal strength distribution map with the three-dimensional contour of the wafer surface through a data fusion algorithm, and generating an initial defect distribution map of the wafer surface includes:

[0127] Collecting signal strength values ​​at the detection position, performing time domain sampling on the signal strength values ​​to obtain a time series, constructing an envelope based on the time series, and calculating the variance within the overlapping time window of the envelope as the phase noise level at the detection position;

[0128] Selecting a defect-free area as a reference area at the detection position, extracting the signal strength value of the reference area to obtain a background noise reference value, using the background noise reference value to standardize the signal strength value at the detection position to obtain a standardized signal strength value, calculating the mean and standard deviation of the standardized signal strength value, and determining an upper threshold and a lower threshold based on the mean and standard deviation;

[0129] Using a lower limit threshold to perform threshold segmentation on the normalized signal intensity value to obtain a candidate defect area, extracting shape parameters and connectivity parameters of the candidate defect area, and screening the candidate defect area based on the shape parameters and connectivity parameters to obtain a preliminary defect area;

[0130] Extracting the area between the lower threshold and the upper threshold from the standardized signal strength value as the area to be determined, taking the ratio of the phase noise level of the area to be determined to the background noise reference value as the noise concentration, and determining the area to be determined that meets the threshold condition as the extended defect area based on the noise concentration;

[0131] Merging the preliminary defect area with the extended defect area, and performing boundary optimization processing on the merged area using a morphological algorithm to obtain a binary signal intensity distribution map;

[0132] Calculating the standard deviation of the phase sequence of the detection position to obtain a phase stability parameter, constructing a data fusion weight calculation function based on the phase stability parameter, and inputting the phase stability parameter into the data fusion weight calculation function to obtain a data fusion weight;

[0133] The signal intensity distribution diagram is numerically processed to obtain defect distribution data, the wafer surface three-dimensional contour data and the defect distribution data are normalized respectively, and the normalized wafer surface three-dimensional contour data and defect distribution data are weightedly fused using the data fusion weight, and a feature diagram characterizing the wafer surface defect distribution is generated based on the fused feature data.

[0134] Exemplarily, wafer surface defect detection first needs to record the signal strength value of the detection position. Scan the wafer surface with a high-precision sensor to collect signal strength data for each detection point. The collected signal strength values ​​are sampled in the time domain, the sampling frequency is set to 1kHz, the sampling time is 1 second, and a time series containing 1000 data points is obtained. Based on the time series, the signal envelope is constructed using the Hilbert transform. An overlapping time window of length 100ms is set on the envelope, and the window overlap rate is 50%. The signal variance value in each window is calculated as the phase noise level of the detection position.

[0135] In order to accurately identify defective areas, it is necessary to establish a reliable background noise baseline. Select a defect-free area of ​​1 square millimeter on the wafer surface as the reference area, extract the signal strength values ​​of all detection points in the area, and calculate the background noise baseline value. With this baseline value as a reference, the signal strength values ​​of all detection positions are standardized. Specifically, the signal strength value of each detection point is divided by the background noise baseline value to obtain the standardized signal strength value. Calculate the mean and standard deviation of the standardized signal strength value, and use the mean plus or minus three times the standard deviation as the upper and lower thresholds, respectively.

[0136] The lower threshold is used to segment the standardized signal intensity value to obtain preliminary candidate defect areas. Shape parameters are extracted for each candidate area, including area, perimeter, circularity and other features. At the same time, the connectivity parameters of the region are calculated, including the number of neighborhood pixels, boundary gradient and other features. The area threshold is set to 0.1 square millimeters, the circularity threshold is set to 0.8, and the neighborhood connectivity threshold is set to 6 pixels. The candidate defect areas are screened to obtain preliminary defect areas.

[0137] In the standardized signal strength value, the area between the lower threshold and the upper threshold is extracted as the area to be determined. The ratio of the phase noise level of each detection point in the area to be determined to the background noise reference value is calculated to obtain the noise concentration index. The noise concentration threshold is set to 1.5, and the area to be determined with a noise concentration greater than the threshold is determined as the extended defect area.

[0138] The preliminary defect area is merged with the extended defect area. The merged area is subjected to morphological processing by opening and closing operations, with the kernel size of the opening operation set to 3×3 pixels and the kernel size of the closing operation set to 5×5 pixels, and the region boundary is optimized to obtain a binary signal intensity distribution map.

[0139] The standard deviation of the phase sequence of the detection position is calculated to obtain the phase stability parameter. Based on this parameter, a data fusion weight calculation function is constructed, and a sigmoid function is used to map the phase stability parameter to a weight value between 0 and 1.

[0140] The signal intensity distribution map is digitized and the binary map is converted into a grayscale distribution. The wafer surface three-dimensional profile data and defect distribution data are normalized respectively, and the numerical range is uniformly mapped to between 0 and 1. The data fusion weights obtained in the previous steps are used to perform weighted fusion operations on the normalized two types of data to generate the final wafer surface defect distribution feature map.

[0141] In this embodiment, by introducing a dual threshold segmentation method and combining the two features of signal strength value and phase noise level, the defective area on the wafer surface can be identified more accurately, effectively reducing the false detection rate and missed detection rate. The data fusion algorithm is used to fuse the signal intensity distribution with the three-dimensional contour of the wafer surface, making full use of multi-source information, improving the reliability and accuracy of defect detection, and making the detection results more comprehensive and reliable. Based on the phase stability parameter, the data fusion weight is adaptively adjusted, so that the detection system can dynamically optimize the fusion strategy according to the actual detection environment and signal quality, improving the adaptability and robustness of the detection method.

[0142] In an optional implementation, the defect density gradient is calculated according to the initial defect distribution map, and the secondary detection area is determined as the area where the defect density gradient exceeds the preset density threshold, and the secondary detection area is orthogonally scanned in a direction perpendicular to the spiral detection trajectory to obtain orthogonal scanning data; the orthogonal scanning data is feature matched with the initial defect distribution map, and a defect feature discrimination criterion is established in combination with the signal strength value and the phase noise level, and the pseudo defect signal is identified and eliminated based on the defect feature discrimination criterion, and the optimized defect distribution data is obtained, including:

[0143] Performing Gaussian smoothing on the initial defect distribution map to obtain a smoothed image, performing first-order difference and second-order difference operations on the smoothed image in orthogonal directions to obtain a first-order gradient matrix and a second-order gradient matrix, convolving the first-order gradient matrix with a Gaussian kernel to obtain a smoothed gradient, convolving the second-order gradient matrix with a Laplace kernel to obtain an edge enhancement gradient, fusing the smoothed gradient and the edge enhancement gradient to construct a morphological gradient operator, using the morphological gradient operator to calculate a defect density gradient, and determining an area where the defect density gradient exceeds a preset density threshold as a secondary detection area;

[0144] Calculating the trajectory tangent vector and curvature distribution according to the parameter equation of the spiral detection trajectory, establishing a sampling density adjustment coefficient based on the curvature distribution, calculating a normal vector perpendicular to the spiral detection trajectory according to the trajectory tangent vector, projecting the normal vector onto the wafer surface to generate an orthogonal scanning trajectory grid, locally encrypting the orthogonal scanning trajectory grid according to the sampling density adjustment coefficient, scanning the secondary detection area along the encrypted orthogonal scanning trajectory grid, and acquiring orthogonal scanning data including signal strength values ​​and phase data;

[0145] Performing wavelet decomposition on the orthogonal scanning data to obtain multi-layer decomposition coefficients, selecting several layers of decomposition coefficients according to the coefficient energy to form a feature matrix, respectively calculating the mutual correlation coefficient and mutual information entropy between the feature matrix and the initial defect distribution map, combining the mutual correlation coefficient and mutual information entropy into a matching evaluation index in a weighted manner, and establishing a spatial correspondence between the orthogonal scanning data and the initial defect distribution map based on the matching evaluation index;

[0146] Calculating the signal intensity ratio of each feature point in the orthogonal scanning data, normalizing the phase data to obtain the local phase fluctuation, calculating the spatial consistency coefficient of the neighborhood of the feature point in combination with the spatial correspondence, and establishing a defect feature discrimination criterion in combination with the signal intensity ratio, the local phase fluctuation and the spatial consistency coefficient;

[0147] Based on the defect feature discrimination criterion, pseudo-defect feature points are identified and eliminated, the defect boundary is extracted from the remaining feature points using a region growing algorithm, and the morphological parameters of the defect boundary are calculated to obtain optimized defect distribution data.

[0148] Exemplarily, first, the initial defect distribution map is processed to determine the secondary detection area. The initial defect distribution map is smoothed using a Gaussian smoothing kernel, the size of the smoothing kernel can be set to 5x5 pixels, and the standard deviation is 1.2. On the smoothed image, first-order difference operations are performed in the horizontal and vertical directions to obtain a first-order gradient matrix, and second-order differences are calculated to obtain a second-order gradient matrix. A Gaussian kernel of 3x3 pixels is selected, and a convolution operation is performed on the first-order gradient matrix to obtain a smooth gradient; a Laplace kernel of 3x3 pixels is selected, and a convolution operation is performed on the second-order gradient matrix to obtain an edge enhancement gradient. The smooth gradient and the edge enhancement gradient are weightedly fused according to a weight ratio of seven to three to construct a morphological gradient operator. The defect density gradient is calculated using this operator, and when the gradient value exceeds a preset threshold, the area is marked as a secondary detection area. The preset threshold can be adjusted according to actual application requirements, and the typical value is 1.5 times the average gradient.

[0149] Next, an orthogonal scanning track grid is generated based on the spiral detection track. According to the parametric equation of the spiral detection track, the tangent vector of each point on the track is calculated, and the direction of the tangent vector is the direction of probe movement. At the same time, the track curvature distribution is calculated, and the sampling density is appropriately increased in areas with larger curvature. The sampling density adjustment coefficient is proportional to the curvature. When the curvature reaches the maximum value, the sampling density is increased to twice the original density. The normal vector perpendicular to the track is calculated based on the tangent vector, and the normal vector is projected onto the wafer surface to generate an orthogonal scanning track grid. In the secondary detection area, the grid is locally encrypted according to the sampling density adjustment coefficient, and the encrypted grid spacing can reach half of the original spacing. Scan along the encrypted orthogonal scanning track grid and record the signal strength value and phase data of each sampling point.

[0150] Then, feature matching is performed on the acquired orthogonal scanning data. Wavelet transform is used to perform multi-scale decomposition on the orthogonal scanning data to obtain decomposition coefficients of different scales. Several layers of coefficients whose energy accounts for more than 80% of the total energy are selected to form a feature matrix. The mutual correlation coefficient and mutual information entropy between the feature matrix and the initial defect distribution map are calculated to reflect the degree of matching between the two in terms of linear correlation and information redundancy, respectively. The mutual correlation coefficient and mutual information entropy are combined as matching evaluation indicators with a weight ratio of six to four, and the spatial correspondence between the orthogonal scanning data and the initial defect distribution map is established.

[0151] Finally, defect feature identification and boundary extraction are performed. The ratio of the signal intensity of each feature point in the orthogonal scanning data to the background signal is calculated. The ratio should usually be greater than 1.3. The phase data is normalized, and the standard deviation of the phase value in the local area is calculated as the phase fluctuation index. Combined with the established spatial correspondence, the consistency coefficient in the neighborhood of the feature point is calculated. This coefficient reflects the degree of similarity between the feature point and the surrounding area. The signal intensity ratio, phase fluctuation and spatial consistency coefficient are comprehensively considered to establish the defect feature identification criterion. When the three indicators meet the threshold requirements at the same time, they are judged as real defect feature points. The regional growing algorithm is used to extract the defect boundary, and the morphological parameters such as the perimeter and area of ​​the boundary are calculated to finally obtain the optimized defect distribution data.

[0152] In this embodiment, by introducing morphological gradient operators and multi-level defect feature discrimination mechanisms, the accuracy of defect detection is significantly improved, the false detection rate and missed detection rate are effectively reduced, and the detection results are more reliable. The curvature-based adaptive sampling strategy and orthogonal scanning method are adopted to enhance the detection capability of complex morphological defects, improve the adaptability of the detection system to different types of defects, and expand the detection range. Through multi-feature fusion and spatial consistency analysis, effective identification and elimination of pseudo-defect signals are achieved, the credibility of the detection results is improved, and more accurate data support is provided for subsequent defect classification and process optimization.

[0153] In an optional implementation, the geometric features and morphological features of the defects are extracted from the optimized defect distribution data, a feature vector space is established, the distribution law and clustering characteristics of the feature parameters are analyzed in the feature vector space, the cause type of the defect is determined, and a test report containing the defect location, feature parameters and cause analysis is generated, including:

[0154] Perform multi-scale edge detection on the optimized defect distribution data, calculate an adaptive threshold coefficient based on the grayscale distribution of the defect distribution data, perform local binarization on the defect distribution data to obtain a binarized image, construct an edge response map based on the binarized image, use a non-maximum suppression algorithm to locate edge pixels in the edge response map to obtain an edge pixel point set, construct a dynamic connection strategy based on the intensity distribution of the edge pixel point set to repair the broken edge, and generate a continuous and complete defect boundary point set;

[0155] A polar coordinate reference system is constructed based on the defect boundary point set, a curvature distribution and a normal vector distribution of the defect boundary point set are calculated, an adaptive meshing is performed on the defect area according to the curvature distribution to obtain a mesh area set, local geometric features and morphological features are extracted in each mesh of the mesh area set, and surface feature parameters in the mesh are calculated in combination with the normal vector distribution to obtain an initial feature set;

[0156] Performing multi-layer wavelet decomposition on the initial feature set to obtain decomposition coefficients, calculating the energy distribution of the decomposition coefficients, selecting decomposition coefficients whose energy distribution is higher than a preset energy threshold to construct a feature vector space, performing feature dimension reduction in the feature vector space using an orthogonal transformation method to obtain a reduced dimension feature set, calculating a discrimination coefficient based on the reduced dimension feature set, and establishing a feature measurement criterion based on the discrimination coefficient;

[0157] In the feature vector space, the local density distribution of feature points is calculated based on the dimension reduction feature set, an adaptive search radius is determined according to the local density distribution, a density threshold function is constructed using the adaptive search radius and the feature measurement criterion, density connectivity between feature points is measured based on the density threshold function, and feature points with density connectivity are hierarchically clustered to obtain a defect type set;

[0158] A feature-cause mapping model is established according to the defect type set and feature measurement criteria, the mapping model is used to perform discriminant analysis on the reduced-dimensional feature set, the confidence of the discrimination result is calculated, the defect cause type is determined based on the discrimination result with the highest confidence, the position information corresponding to the defect boundary point set, the reduced-dimensional feature set and the defect cause type are integrated to generate a detection report, and the detection report is output to the process control system.

[0159] Exemplarily, the defect distribution data is firstly subjected to edge detection and boundary extraction. A multi-scale edge detection method is used to smooth the image at different scales through a Gaussian filter group, and the image gradient amplitude and direction are calculated at each scale. Based on the image grayscale histogram features, an iterative method is used to calculate the adaptive threshold coefficient, which is dynamically adjusted with the grayscale distribution characteristics of the local area of ​​the image. When the image is locally binarized, each pixel is compared with the average grayscale value in its neighborhood, and when the difference is greater than the adaptive threshold, it is determined to be an edge pixel. An edge response map is constructed with the defect area as the center, and the gradient amplitude and direction information of each pixel is recorded. Practice has shown that when the image resolution is 1024×1024 pixels, it is more appropriate to select a Gaussian kernel size of 3×3 to 9×9 for multi-scale decomposition.

[0160] After obtaining the edge response map, the non-maximum suppression algorithm is used to accurately locate the edge pixels. The edge response is interpolated along the gradient direction, and only the local maximum point is retained as the edge candidate point. According to the intensity distribution characteristics of the edge pixels, a dynamic connection strategy is established. When the intensity difference of adjacent edge pixels is less than the preset threshold, they are judged to belong to the same edge. In this way, the edge breaks caused by noise or occlusion can be effectively repaired, and finally a continuous and complete set of defect boundary points can be obtained.

[0161] Next, a polar coordinate reference system is constructed on the defect boundary point set, with the defect center as the pole, and the polar coordinate representation of the boundary point is calculated. On this basis, the curvature distribution of the boundary curve is calculated, and the sliding window method is used to estimate the curvature value at each boundary point. At the same time, the normal vector distribution of the boundary point is calculated, and the direction of the normal vector is perpendicular to the direction of the boundary tangent. The defect area is adaptively meshed according to the curvature distribution characteristics, using denser grids in areas where the curvature changes dramatically, and sparser grids in areas where the curvature changes gently. Local geometric features are extracted in each grid, including parameters such as area, circumference, roundness, and morphological features such as texture directionality, roughness, and other parameters. The surface feature parameters in the grid are calculated in combination with the normal vector distribution to obtain the initial set of features.

[0162] Perform multi-layer wavelet decomposition on the initial feature set, use the discrete wavelet transform method, and select appropriate wavelet basis functions for feature decomposition. Calculate the energy distribution of decomposition coefficients at each scale, and select decomposition coefficients whose energy share exceeds the preset threshold to construct the feature vector space. Use the principal component analysis method to reduce the feature dimension in the feature vector space, and retain the principal components whose cumulative contribution rate reaches the preset threshold as the reduced dimension feature set. Calculate the discrimination coefficient between features based on the reduced dimension feature set, and establish a feature measurement criterion based on the Mahalanobis distance.

[0163] In the feature vector space, the local density distribution of each feature point is calculated. The kernel density estimation method is used, and the Gaussian kernel function is selected to calculate the density value of the feature point. The adaptive search radius is determined according to the density distribution characteristics, and a smaller search radius is used in high-density areas and a larger search radius is used in low-density areas. A density-based threshold function is constructed, and when the density connectivity between two feature points is greater than the threshold, they are divided into the same category. The defect type set is obtained through the density clustering method.

[0164] Finally, a feature-cause mapping model is established, and a multi-class classifier is constructed using the support vector machine method. The reduced-dimensional feature set is discriminated and analyzed, and the discriminant probability of each class is calculated as the confidence index. The class with the highest confidence is selected as the defect cause type. The location information of the defect boundary point set, the reduced-dimensional feature set, and the cause type are integrated to generate a test report, which is output to the process control system for subsequent processing.

[0165] In this embodiment, through multi-scale edge detection and adaptive threshold segmentation methods, the defect boundary can be accurately extracted, the influence of image noise and uneven illumination is effectively overcome, and the edge positioning accuracy is improved. The dynamic connection strategy can effectively repair the broken edge and ensure the continuity and integrity of the boundary. The adaptive grid division and multi-layer wavelet decomposition method are adopted to realize the multi-scale expression of defect features, and the extracted features have good scale invariance and rotation invariance. Through feature dimensionality reduction and selection, the redundancy of the feature space is reduced, and the compactness and discriminability of the feature expression are improved. The density-based adaptive clustering method can effectively discover feature clusters of different shapes and sizes, and has strong anti-noise ability. The feature-cause mapping model adopts a probabilistic output method, which not only gives the defect type determination result, but also provides a reliable confidence assessment, providing a reliable decision-making basis for process optimization.

[0166] Figure 2 FIG. 1 is a schematic diagram of a structure of a system for enhancing wafer detection in a semiconductor manufacturing process according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0167] The first unit is used to emit dual-frequency lasers with different frequencies to the surface of the wafer, the first frequency beam of the dual-frequency laser forms a first detection area after wavefront modulation, and the second frequency beam of the dual-frequency laser forms a second detection area after phase compensation, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect the beat signal generated in the overlapping area, the initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with the optical path difference calibration curve, the dual-frequency laser is controlled by a two-dimensional scanning galvanometer to form a spiral detection track on the wafer surface, the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat signal are calculated, and initial detection quality evaluation data is generated;

[0168] The second unit is used to perform adaptive filtering on the beat frequency signal, obtain signal spectrum characteristics through wavelet packet decomposition, dynamically adjust the filtering parameters to suppress noise based on the signal spectrum characteristics and the initial detection quality assessment data, obtain a filtered signal, extract the difference frequency component of the filtered signal and perform phase demodulation, obtain continuous phase information in combination with the initial height information, reconstruct the three-dimensional profile of the wafer surface, record the signal strength value and phase noise level of the detection position, establish a signal intensity distribution map by a double threshold segmentation method, fuse the signal intensity distribution map with the three-dimensional profile of the wafer surface through a data fusion algorithm, and generate an initial defect distribution map of the wafer surface;

[0169] The third unit is used to calculate the defect density gradient according to the initial defect distribution map, determine the secondary detection area as the area where the defect density gradient exceeds the preset density threshold, perform orthogonal scanning on the secondary detection area in a direction perpendicular to the spiral detection trajectory, and obtain orthogonal scanning data; perform feature matching on the orthogonal scanning data and the initial defect distribution map, establish a defect feature discrimination criterion based on the signal strength value and the phase noise level, identify and eliminate pseudo-defect signals based on the defect feature discrimination criterion, obtain optimized defect distribution data, extract the geometric features and morphological features of the defects from the optimized defect distribution data, establish a feature vector space, analyze the distribution law and clustering characteristics of the feature parameters in the feature vector space, determine the cause type of the defect, and generate a test report containing the defect location, feature parameters and cause analysis.

[0170] According to a third aspect of the embodiments of the present invention,

[0171] An electronic device is provided, comprising:

[0172] processor;

[0173] a memory for storing processor-executable instructions;

[0174] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0175] A fourth aspect of the embodiments of the present invention is:

[0176] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0177] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for enhancing wafer detection in a semiconductor manufacturing process, characterized in that: include: A dual-frequency laser with different frequencies is emitted to the surface of the wafer, the first frequency beam of the dual-frequency laser is subjected to wavefront modulation to form a first detection area, the second frequency beam of the dual-frequency laser is subjected to phase compensation to form a second detection area, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect a beat signal generated in the overlapping area, initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with an optical path difference calibration curve, the dual-frequency laser is controlled by a two-dimensional scanning galvanometer to form a spiral detection track on the wafer surface, the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat signal are calculated, and initial detection quality evaluation data is generated; Adaptively filter the beat frequency signal, obtain signal spectrum characteristics through wavelet packet decomposition, dynamically adjust the filter parameters to suppress noise based on the signal spectrum characteristics and the initial detection quality assessment data, obtain a filtered signal, extract the difference frequency component of the filtered signal and perform phase demodulation, obtain continuous phase information in combination with the initial height information, reconstruct the three-dimensional profile of the wafer surface, record the signal strength value and phase noise level of the detection position, establish a signal intensity distribution map by using a double threshold segmentation method, fuse the signal intensity distribution map with the three-dimensional profile of the wafer surface through a data fusion algorithm, and generate an initial defect distribution map of the wafer surface; The defect density gradient is calculated according to the initial defect distribution map, and the area where the defect density gradient exceeds the preset density threshold is determined as the secondary detection area. The secondary detection area is orthogonally scanned in a direction perpendicular to the spiral detection trajectory to obtain orthogonal scanning data; the orthogonal scanning data is feature matched with the initial defect distribution map, and a defect feature discrimination criterion is established in combination with the signal strength value and the phase noise level. Based on the defect feature discrimination criterion, pseudo defect signals are identified and eliminated to obtain optimized defect distribution data, and the geometric features and morphological features of the defects are extracted from the optimized defect distribution data, and a feature vector space is established. The distribution law and clustering characteristics of the feature parameters are analyzed in the feature vector space, the cause type of the defect is determined, and a detection report containing the defect location, feature parameters and cause analysis is generated.

2. The method according to claim 1, characterized in that A dual-frequency laser with different frequencies is emitted to the surface of the wafer, the first frequency beam of the dual-frequency laser is subjected to wavefront modulation to form a first detection area, the second frequency beam of the dual-frequency laser is subjected to phase compensation to form a second detection area, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect a beat signal generated in the overlapping area, and initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with an optical path difference calibration curve, including: Emitting a dual-frequency laser to the surface of the wafer, separating a first frequency light beam and a second frequency light beam through a polarization beam splitter, measuring the light intensity distribution of the first frequency light beam and the second frequency light beam, and adjusting the splitting ratio of the polarization beam splitter according to the light intensity distribution to make the light intensities of the two beams equal; Input the light intensity distribution data of the first frequency light beam into the iterative optimization algorithm, set the target light field distribution function to calculate the phase modulation amount between adjacent pixels, obtain the phase mask data through multiple iterative optimizations and load it into the spatial light modulator, and perform wavefront modulation on the first frequency light beam to form a first detection area; Collecting a wavefront aberration spot array image of the second frequency light beam, calculating the centroid coordinates of each spot in the spot array image, converting the deviation between the centroid coordinates and the ideal position into wavefront gradient data, calculating the wavefront gradient data through a reconstruction matrix to obtain driving voltage data of the deformable reflector, and compensating the second frequency light beam according to the driving voltage data to form a second detection area; respectively collecting light intensity images of the first detection area and the second detection area, calculating a two-dimensional cross-correlation function of the two light intensity images, determining a relative displacement of the two detection areas according to a peak position of the cross-correlation function, inputting the relative displacement into a feedback controller, and sending control signals output by the feedback controller to the two-dimensional translation stage and the angle adjustment device, respectively, to adjust the two detection areas to form an overlapping area on the wafer surface; A photodetector array is used to sample the overlapping area to obtain a beat frequency signal, and the amplitude and initial phase information of the beat frequency signal are extracted, and a two-dimensional phase unwrapping algorithm is used to eliminate phase jumps to generate continuous spatial phase difference distribution data; The piezoelectric ceramic translation stage is used to move the reflector according to the preset step length, and the displacement and phase difference data of the overlapping area are synchronously recorded at each position. The mapping relationship from displacement to phase difference is established by least squares fitting, and the polynomial coefficients of the calibration curve are obtained. The spatial phase difference distribution data is substituted into a calculation formula containing calibration polynomial coefficients to obtain initial data characterizing the height of the wafer surface. The initial data is subjected to wavelet decomposition and specific scale coefficients are selected to reconstruct the denoised height data. The denoised height data is corrected in combination with a pre-calibrated system error compensation model to finally obtain the initial height information of the wafer surface.

3. The method according to claim 1, characterized in that The dual-frequency laser is controlled by a two-dimensional scanning galvanometer to form a spiral detection track on the wafer surface, the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat frequency signal are calculated, and the initial detection quality evaluation data is generated, including: Receive the radius parameter of the wafer surface, set the maximum scanning range of the spiral trajectory according to the radius parameter, obtain the maximum scanning frequency and the maximum scanning angle of the galvanometer, determine the motion constraint condition of the galvanometer according to the maximum scanning frequency and the maximum scanning angle, calculate the pitch interval between adjacent spiral turns based on the maximum scanning range and the motion constraint condition, use the pitch interval as the step increment of the spiral trajectory, calculate the rotation angle and the scanning radius of the spiral trajectory according to the step increment, and generate a complete motion trajectory of the spiral trajectory; Convert the complete motion trajectory into a deflection angle sequence of the galvanometer in the horizontal direction and the vertical direction; smooth the deflection angle sequence to obtain a control sequence of the galvanometer, wherein the control sequence includes an angle value, an angular velocity, and an angular acceleration of the galvanometer; Input the control sequence into the driver of the two-dimensional scanning galvanometer in time order, control the dual-frequency laser to move along the spiral detection track on the wafer surface, collect the beat frequency signal in the overlapping area in real time, construct a time window, perform Fourier transform on the beat frequency signal in the window to obtain a power spectrum, extract the signal peak power and background noise power in the power spectrum, calculate the ratio of the two to obtain the signal-to-noise ratio of the beat frequency signal, and the length of the time window is determined according to the periodic characteristics of the beat frequency signal; Performing phase extraction on the beat frequency signal, calculating the phase difference between adjacent sampling points, calculating the standard deviation of the phase difference to obtain a phase stability parameter, combining the phase stability parameter with the wavelength parameter of the dual-frequency laser to calculate the minimum resolvable height difference, and determining the spatial resolution parameter; The signal-to-noise ratio is normalized to obtain a first evaluation parameter, the reciprocal of the phase stability parameter is normalized to obtain a second evaluation parameter, the reciprocal of the spatial resolution parameter is normalized to obtain a third evaluation parameter, and the three evaluation parameters are weighted according to a preset weight coefficient to obtain a comprehensive evaluation index; When the comprehensive evaluation index meets the detection quality threshold requirement, the corresponding signal-to-noise ratio, phase stability parameter and spatial resolution parameter are combined to generate detection quality evaluation data.

4. The method according to claim 1, characterized in that: The beat frequency signal is subjected to adaptive filtering processing, signal spectrum characteristics are obtained by wavelet packet decomposition, filtering parameters are dynamically adjusted based on the signal spectrum characteristics and the initial detection quality assessment data to suppress noise, a filtered signal is obtained, a difference frequency component of the filtered signal is extracted and phase demodulated, continuous phase information is obtained in combination with the initial height information, and a three-dimensional profile of the wafer surface is reconstructed, including: Selecting a wavelet basis function to perform wavelet packet decomposition on the beat frequency signal to obtain a plurality of frequency band signals, calculating the frequency band energy ratio according to the plurality of frequency band signals to obtain a signal spectrum feature, determining an energy threshold based on the signal spectrum feature and a signal-to-noise ratio parameter in the initial detection quality assessment data, and performing noise reduction processing on the frequency band signal; Determine the filtering window length according to the phase stability parameter in the initial detection quality assessment data, and perform adaptive filtering on the beat frequency signal to obtain a filtered signal; Performing Hilbert transform on the filtered signal to obtain a transformed signal, combining the filtered signal with the transformed signal to extract a difference frequency component to obtain an instantaneous phase signal, detecting a phase jump point of the instantaneous phase signal and performing phase compensation to obtain an unwrapped phase, determining a phase integration starting point using the initial height information, and integrating the unwrapped phase to obtain continuous phase information; The initial height difference is calculated by combining the continuous phase information with the equivalent wavelength of the dual-frequency laser, and the actual height difference between adjacent sampling points is corrected by combining the spatial resolution parameter in the initial detection quality assessment data, and the initial height distribution is obtained by accumulating the actual height difference between adjacent sampling points along the spiral scanning trajectory; The initial height distribution is divided into a plurality of overlapping local reconstruction areas, the phase standard deviation of the local reconstruction area is calculated to obtain the local phase fluctuation, the local reconstruction area that needs to be smoothed is identified according to the phase stability parameter, the smoothing factor is calculated using the spatial resolution parameter, a least squares objective function including the smoothing factor is constructed, the local reconstruction area is optimized to obtain the smoothed local height distribution, the smoothed local height distribution is continuously transitioned and spliced ​​with the height distribution of the unsmoothed area in the overlapping area, and the three-dimensional contour of the wafer surface is reconstructed.

5. The method according to claim 1, characterized in that The signal strength value and phase noise level of the detection position are recorded, and a signal strength distribution map is established by using a double threshold segmentation method. The signal strength distribution map is fused with the three-dimensional contour of the wafer surface through a data fusion algorithm to generate an initial defect distribution map of the wafer surface, including: Collecting signal strength values ​​at the detection position, performing time domain sampling on the signal strength values ​​to obtain a time series, constructing an envelope based on the time series, and calculating the variance within the overlapping time window of the envelope as the phase noise level at the detection position; Selecting a defect-free area as a reference area at the detection position, extracting the signal strength value of the reference area to obtain a background noise reference value, using the background noise reference value to standardize the signal strength value at the detection position to obtain a standardized signal strength value, calculating the mean and standard deviation of the standardized signal strength value, and determining an upper threshold and a lower threshold based on the mean and standard deviation; Using a lower limit threshold to perform threshold segmentation on the normalized signal intensity value to obtain a candidate defect area, extracting shape parameters and connectivity parameters of the candidate defect area, and screening the candidate defect area based on the shape parameters and connectivity parameters to obtain a preliminary defect area; Extracting the area between the lower threshold and the upper threshold from the standardized signal strength value as the area to be determined, taking the ratio of the phase noise level of the area to be determined to the background noise reference value as the noise concentration, and determining the area to be determined that meets the threshold condition as the extended defect area based on the noise concentration; Merging the preliminary defect area with the extended defect area, and performing boundary optimization processing on the merged area using a morphological algorithm to obtain a binary signal intensity distribution map; Calculating the standard deviation of the phase sequence of the detection position to obtain a phase stability parameter, constructing a data fusion weight calculation function based on the phase stability parameter, and inputting the phase stability parameter into the data fusion weight calculation function to obtain a data fusion weight; The signal intensity distribution diagram is numerically processed to obtain defect distribution data, the three-dimensional contour data of the wafer surface and the defect distribution data are normalized respectively, and the normalized three-dimensional contour data of the wafer surface and the defect distribution data are weightedly fused using the data fusion weight, and a feature diagram characterizing the defect distribution on the wafer surface is generated based on the fused feature data.

6. The method according to claim 1, characterized in that The defect density gradient is calculated according to the initial defect distribution map, and the area where the defect density gradient exceeds the preset density threshold is determined as the secondary detection area. The secondary detection area is orthogonally scanned in a direction perpendicular to the spiral detection track to obtain orthogonal scanning data; the orthogonal scanning data is feature matched with the initial defect distribution map, and a defect feature discrimination criterion is established in combination with the signal strength value and the phase noise level. The pseudo defect signal is identified and eliminated based on the defect feature discrimination criterion, and the optimized defect distribution data is obtained, including: Performing Gaussian smoothing on the initial defect distribution map to obtain a smoothed image, performing first-order difference and second-order difference operations on the smoothed image in orthogonal directions to obtain a first-order gradient matrix and a second-order gradient matrix, convolving the first-order gradient matrix with a Gaussian kernel to obtain a smoothed gradient, convolving the second-order gradient matrix with a Laplace kernel to obtain an edge enhancement gradient, fusing the smoothed gradient and the edge enhancement gradient to construct a morphological gradient operator, using the morphological gradient operator to calculate a defect density gradient, and determining an area where the defect density gradient exceeds a preset density threshold as a secondary detection area; Calculating the trajectory tangent vector and curvature distribution according to the parameter equation of the spiral detection trajectory, establishing a sampling density adjustment coefficient based on the curvature distribution, calculating a normal vector perpendicular to the spiral detection trajectory according to the trajectory tangent vector, projecting the normal vector onto the wafer surface to generate an orthogonal scanning trajectory grid, locally encrypting the orthogonal scanning trajectory grid according to the sampling density adjustment coefficient, scanning the secondary detection area along the encrypted orthogonal scanning trajectory grid, and acquiring orthogonal scanning data including signal strength values ​​and phase data; Performing wavelet decomposition on the orthogonal scanning data to obtain multi-layer decomposition coefficients, selecting several layers of decomposition coefficients according to the coefficient energy to form a feature matrix, respectively calculating the mutual correlation coefficient and mutual information entropy between the feature matrix and the initial defect distribution map, combining the mutual correlation coefficient and mutual information entropy into a matching evaluation index in a weighted manner, and establishing a spatial correspondence between the orthogonal scanning data and the initial defect distribution map based on the matching evaluation index; Calculating the signal intensity ratio of each feature point in the orthogonal scanning data, normalizing the phase data to obtain the local phase fluctuation, calculating the spatial consistency coefficient of the neighborhood of the feature point in combination with the spatial correspondence, and establishing a defect feature discrimination criterion in combination with the signal intensity ratio, the local phase fluctuation and the spatial consistency coefficient; Based on the defect feature discrimination criterion, pseudo-defect feature points are identified and eliminated, the defect boundary is extracted from the remaining feature points using a region growing algorithm, and the morphological parameters of the defect boundary are calculated to obtain optimized defect distribution data.

7. The method according to claim 1, characterized in that Extract the geometric features and morphological features of the defects from the optimized defect distribution data, establish a feature vector space, analyze the distribution law and clustering characteristics of the feature parameters in the feature vector space, determine the cause type of the defect, and generate a test report containing the defect location, feature parameters and cause analysis, including: Perform multi-scale edge detection on the optimized defect distribution data, calculate an adaptive threshold coefficient based on the grayscale distribution of the defect distribution data, perform local binarization on the defect distribution data to obtain a binarized image, construct an edge response map based on the binarized image, use a non-maximum suppression algorithm to locate edge pixels in the edge response map to obtain an edge pixel point set, construct a dynamic connection strategy based on the intensity distribution of the edge pixel point set to repair the broken edge, and generate a continuous and complete defect boundary point set; A polar coordinate reference system is constructed based on the defect boundary point set, a curvature distribution and a normal vector distribution of the defect boundary point set are calculated, an adaptive meshing is performed on the defect area according to the curvature distribution to obtain a mesh area set, local geometric features and morphological features are extracted in each mesh of the mesh area set, and surface feature parameters in the mesh are calculated in combination with the normal vector distribution to obtain an initial feature set; Performing multi-layer wavelet decomposition on the initial feature set to obtain decomposition coefficients, calculating the energy distribution of the decomposition coefficients, selecting decomposition coefficients whose energy distribution is higher than a preset energy threshold to construct a feature vector space, performing feature dimension reduction in the feature vector space using an orthogonal transformation method to obtain a reduced dimension feature set, calculating a discrimination coefficient based on the reduced dimension feature set, and establishing a feature measurement criterion based on the discrimination coefficient; In the feature vector space, the local density distribution of feature points is calculated based on the dimension reduction feature set, an adaptive search radius is determined according to the local density distribution, a density threshold function is constructed using the adaptive search radius and the feature measurement criterion, density connectivity between feature points is measured based on the density threshold function, and feature points with density connectivity are hierarchically clustered to obtain a defect type set; A feature-cause mapping model is established according to the defect type set and feature measurement criteria, the mapping model is used to perform discriminant analysis on the reduced-dimensional feature set, the confidence of the discrimination result is calculated, the defect cause type is determined based on the discrimination result with the highest confidence, the position information corresponding to the defect boundary point set, the reduced-dimensional feature set and the defect cause type are integrated to generate a detection report, and the detection report is output to the process control system.

8. A system for enhancing wafer detection in a semiconductor manufacturing process, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to emit dual-frequency lasers with different frequencies to the surface of the wafer, the first frequency beam of the dual-frequency laser forms a first detection area after wavefront modulation, and the second frequency beam of the dual-frequency laser forms a second detection area after phase compensation, the first detection area and the second detection area form a controllable overlapping area through an optical path adjustment system, a photodetector array is used to collect the beat signal generated in the overlapping area, the initial height information of the wafer surface is obtained based on the phase difference of the beat signal combined with the optical path difference calibration curve, the dual-frequency laser is controlled by a two-dimensional scanning galvanometer to form a spiral detection track on the wafer surface, the signal-to-noise ratio, phase stability and spatial resolution parameters of the beat signal are calculated, and initial detection quality evaluation data is generated; The second unit is used to perform adaptive filtering on the beat frequency signal, obtain signal spectrum characteristics through wavelet packet decomposition, dynamically adjust the filtering parameters to suppress noise based on the signal spectrum characteristics and the initial detection quality assessment data, obtain a filtered signal, extract the difference frequency component of the filtered signal and perform phase demodulation, obtain continuous phase information in combination with the initial height information, reconstruct the three-dimensional profile of the wafer surface, record the signal strength value and phase noise level of the detection position, establish a signal intensity distribution map by a double threshold segmentation method, fuse the signal intensity distribution map with the three-dimensional profile of the wafer surface through a data fusion algorithm, and generate an initial defect distribution map of the wafer surface; The third unit is used to calculate the defect density gradient according to the initial defect distribution map, determine the area where the defect density gradient exceeds the preset density threshold as the secondary detection area, perform orthogonal scanning on the secondary detection area in a direction perpendicular to the spiral detection trajectory, and obtain orthogonal scanning data; perform feature matching on the orthogonal scanning data and the initial defect distribution map, establish a defect feature discrimination criterion based on the signal strength value and the phase noise level, identify and eliminate pseudo defect signals based on the defect feature discrimination criterion, obtain optimized defect distribution data, extract the geometric features and morphological features of the defects from the optimized defect distribution data, establish a feature vector space, analyze the distribution law and clustering characteristics of the feature parameters in the feature vector space, determine the cause type of the defect, and generate a detection report containing the defect location, feature parameters and cause analysis.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images