Glass wafer through hole three-dimensional shape detection system based on machine vision

Abstract

The invention relates to the technical field of semiconductor manufacturing and detection, and discloses a glass wafer through hole three-dimensional shape detection system based on machine vision, and the system comprises a multi-angle light source cooperation illumination module which is used for achieving the comprehensive illumination of a glass wafer through hole; the polarization modulation module is used for enhancing weak scattering signals and acquiring scattering field information with a high signal-to-noise ratio; the high numerical aperture optical module is used for collecting scattered field data; the Fourier diffraction region inverse scattering reconstruction module is used for realizing preliminary reconstruction of three-dimensional defect distribution; the compressed sensing iterative optimization module is used for improving the three-dimensional reconstruction precision and obtaining accurate three-dimensional defect distribution of the side wall of the through hole of the glass wafer; according to the invention, the diffraction limit of traditional optical detection is broken through, and surface and internal defects can be simultaneously detected and distinguished.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor manufacturing inspection, and more specifically, it relates to a three-dimensional topography inspection system for glass wafer through-holes based on machine vision. Background Art

[0002] With the continuous development of modern integrated circuit manufacturing processes, the quality of glass wafer through-holes has a crucial impact on the performance and reliability of semiconductor products. As a vertical interconnection structure in three-dimensional integrated circuits, the topography defects of glass wafer through-holes directly affect the signal transmission quality and packaging reliability.

[0003] Currently, the inspection of glass wafer through-hole topography mainly uses technologies such as optical microscopes, scanning electron microscopes, and confocal microscopes. These traditional inspection methods have three key technical problems in practical applications: for high aspect ratio through-holes, it is difficult for traditional lighting methods to effectively irradiate the sidewall area, resulting in detection blind spots; buried defects such as microcracks and bubbles that may exist inside the sidewalls have extremely weak scattered signals and are mixed with surface scattered signals, making it difficult to distinguish and extract; with the improvement of the process, the size of sidewall defects to be detected is getting smaller and smaller, even reaching the order of 100 nm, and the resolution of conventional optical imaging systems is restricted by the Abbe diffraction limit, making it difficult to meet the detection requirements.

[0004] The above problems lead to the inability of the existing technology to comprehensively and accurately detect the three-dimensional topography of high aspect ratio glass wafer through-holes. In particular, it is impossible to distinguish surface defects and internal defects, it is difficult to effectively detect micro-defects of the order of 100 nm, and the coverage rate of the sidewalls of high aspect ratio through-holes is low, seriously affecting the quality control and yield improvement of semiconductor products.

[0005] Therefore, a high-precision three-dimensional topography inspection method for glass wafer through-holes that can break through the above technical limitations is needed. Summary of the Invention

[0006] The present invention provides a three-dimensional topography inspection system for glass wafer through-holes based on machine vision, which solves the technical problem in the related art of being unable to comprehensively and accurately detect the three-dimensional topography of high aspect ratio glass wafer through-holes.

[0007] The present invention provides a three-dimensional topography inspection system for glass wafer through-holes based on machine vision, including: A multi-angle light source collaborative illumination module for achieving comprehensive illumination of the glass wafer through-holes; A polarization modulation module for enhancing weak scattered signals and obtaining scattered field information with high signal-to-noise ratio; A high numerical aperture optical module for collecting scattered field data; A Fourier diffraction region inverse scattering reconstruction module for achieving preliminary reconstruction of the three-dimensional defect distribution; A compressive sensing iterative optimization module, which is used to improve the 3D reconstruction accuracy and obtain the accurate 3D defect distribution on the sidewalls of the vias in the glass wafer.

[0008] In a preferred embodiment, the multi-angle light source collaborative illumination module includes: A main light source for vertical illumination, which is used to obtain the information of the top and bottom of the via; An array of auxiliary light sources with adjustable angles, which is dedicated to illuminating the sidewalls of the via; A device for precisely controlling the phase and polarization state of the light source; An optical path separation mirror group, which is used to guide the optical signals reflected at different angles to different sensors.

[0009] In a preferred embodiment, the polarization modulation module includes: A polarization modulation device, which is used to modulate the polarization state of the incident light; A polarization differential imaging device, which is used to enhance the scattering signal; A holographic interferometry device, which is used to obtain the phase information of the scattering field.

[0010] In a preferred embodiment, the high numerical aperture optical module includes: An objective lens with a numerical aperture NA>0.9 and a dark field illumination device; A scattering field information recording device; An angle scanning device, which is used to expand the spatial frequency coverage range of the system.

[0011] In a preferred embodiment, the Fourier diffraction region inverse scattering reconstruction module includes a processor, and the processor is configured to: Establish a scattering model based on the Rytov approximation; Apply synthetic aperture technology to merge the scattering data at different angles; Solve the inverse problem and reconstruct the 3D defect distribution.

[0012] In a preferred embodiment, the synthetic aperture technology includes the following steps: Perform a Fourier transform on the scattering field data at each incident angle to obtain its spatial spectrum; Map the spatial spectra of each angle to the corresponding positions in the three-dimensional Fourier space; Perform a weighted average processing on the overlapping regions, and the weights are related to the signal intensity and coherence; Fill the missing regions in the spectrum space through interpolation.

[0013] In a preferred embodiment, the compressive sensing iterative optimization module includes a processor, and the processor is configured to perform the following operations: Introduce the sparsity constraint of the defect distribution; The alternating direction multiplier method is used to solve the optimization problem; Combined with the side wall light reflection model, a complete side wall development diagram is constructed.

[0014] In a preferred embodiment, the sparsity constraint is constructed as a compressed sensing optimization problem, which minimizes the objective function value in the sparse transform domain while ensuring that the error between the reconstruction result and the actual measurement data is within a preset allowable range, wherein the allowable error is related to the system noise level; The optimization problem is solved by an iterative shrinkage algorithm. In each iteration, the reconstruction result is subjected to sparse domain threshold processing to gradually improve the reconstruction accuracy.

[0015] In a preferred embodiment, the processor of the compressed sensing iterative optimization module is further configured to perform the following operations: Adaptive threshold segmentation is performed on the reconstructed 3D defect distribution to distinguish surface defects from internal defects; Calculate defect density, size distribution and depth distribution based on the spatial distribution characteristics of defects; Combined with the material optical property model, the potential impact of defects on electrical performance is analyzed; Generate a comprehensive inspection report with 3D defect visualization, sidewall expansion, and defect statistics.

[0016] In a preferred embodiment, a computer-readable storage medium is used to store computer-readable instructions, which, when read by a computer, can run a machine vision-based glass wafer through-hole three-dimensional morphology detection system.

[0017] The beneficial effects of the present invention are: The ability to simultaneously detect and distinguish defects on the surface and inside of the glass wafer through-hole sidewalls has been achieved. Through polarization modulation and holographic interferometry technology, weak scattered signals can be effectively extracted, making it possible to detect defects buried deep inside the sidewalls (such as microcracks, bubbles, etc.), providing an important basis for comprehensively evaluating through-hole quality and predicting potential failure risks.

[0018] It breaks through the diffraction limit of traditional optical imaging, improves detection resolution, and can effectively identify tiny defects in modern high-end integrated circuits.

[0019] Significantly improves the detection capability of high aspect ratio vias. For vias with a depth-to-diameter ratio of 20:1, the sidewall coverage is improved, which basically eliminates the detection blind spots in traditional methods and provides reliable guarantee for comprehensive quality assessment.

[0020] The defect detection rate has been greatly improved. The compressed sensing iterative optimization algorithm effectively suppresses noise and artifacts in the reconstruction process, improves the defect detection rate, and significantly reduces the risk of missed detection.

[0021] It provides a more comprehensive basis for quality assessment. Through three-dimensional defect tomography and sidewall expansion diagrams, it provides deeper and more comprehensive quality information than surface topography, providing strong support for the reliability monitoring and process optimization of high-end integrated circuits. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a module diagram of a three-dimensional topography detection system for glass wafer through-holes based on machine vision according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the scope of protection of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0024] In at least one embodiment of the present invention, a three-dimensional topography detection system for glass wafer through-holes based on machine vision is disclosed, as Figure 1 shown, including: A multi-angle light source collaborative illumination module for achieving comprehensive illumination of the glass wafer through-holes; Specifically, it includes the following steps: Step 1.1, configure a main light source for vertical illumination to obtain information on the top and bottom of the through-hole; The main light source uses a laser with a wavelength of 532 nm, which is vertically incident on the sample surface after collimation and beam expansion processing.

[0025] Step 1.2, set up an array of auxiliary light sources with adjustable angles for illuminating the sidewalls of the through-hole; The auxiliary light source consists of 16 independently controlled coherent light sources, which are evenly distributed around the through-hole, and the incident angle can be adjusted within the range of 20° - 70° to adapt to through-hole structures with different depth-to-width ratios.

[0026] Step 1.3, achieve precise control of the phase and polarization state of the light source; Each light source is equipped with a phase modulator and a polarization control element, enabling the system to generate an illumination light field with a specific phase relationship and polarization state, providing a basis for subsequent polarization modulation signal enhancement and holographic interference.

[0027] Step 1.4, guide the light signals reflected at different angles to different sensors through an optical path separation mirror group; The optical path separation mirror group consists of a dichroic mirror, a polarization beam splitter, and a reflector, and can separate signals according to the wavelength, polarization state, and incident angle of light and guide them to corresponding detectors.

[0028] The output result of this module is to form an illumination system that can comprehensively illuminate all parts of the through hole, solving the problem of the detection blind area on the side wall of the through hole in the prior art and supporting the technical feature of "multi-angle light source collaborative illumination technology" in the claims of this application.

[0029] The polarization modulation module is used to enhance weak scattering signals and obtain scattering field information with high signal-to-noise ratio; Specifically, it includes the following steps: Step 2.1, construct a polarization modulation system to modulate the polarization state of the incident light; This system consists of a polarization state generator and an electro-optic modulator, and can generate light fields with different polarization states and perform fast modulation in time.

[0030] According to the optical characteristics of the defect, set the optimal polarization state sequence to enhance the scattering efficiency of specific types of defects.

[0031] Step 2.2, use polarization difference imaging technology to enhance the scattering signal; Collect scattering images of the same area under different polarization states , where, 、 、 respectively represent the scattering images collected under the 、 、 th polarization states; represents the total number of different polarization state images collected; obtain the enhanced difference image through a specific linear combination: ; where, represents the spatial coordinates used to locate the position of each pixel in the image; represents the total number of different polarization state images collected; is the weight coefficient that determines the contribution ratio of each polarization state image in the difference result and is determined by an optimization algorithm to maximize the contrast between the target signal and the background noise; is the finally obtained enhanced difference image with a higher signal-to-noise ratio.

[0032] In addition, this method can effectively suppress background noise and enhance the target scattering signal, and is particularly suitable for detecting weak defect scattering signals in the through holes of glass wafers.

[0033] Step 2.3, implement holographic interferometry to obtain the phase information of the scattering field; Using off-axis digital holography technology, the interference pattern of the reference light and the scattered light is recorded, and the interference intensity distribution is expressed as: ; where, represents the interference intensity distribution at the spatial position ; is the reference light field, representing the complex amplitude of the reference light electric field at the spatial position ; is the scattered light field, representing the complex amplitude of the scattered light electric field at the spatial position ; represents the reference light field intensity; represents the scattered light field intensity; and respectively represent the complex conjugates of the reference light field and the scattered light field; represents the complex conjugate operation.

[0034] By performing Fourier transform and spatial filtering on the interference pattern, the term is extracted, and then the complex amplitude of the scattered field is restored.

[0035] The output result of this module is to obtain the complex amplitude distribution of the scattered field with high signal-to-noise ratio, including the amplitude and phase information of the scattered field, solving the problem that weak scattered signals are difficult to detect in the prior art, and supporting the technical features regarding "polarization modulation signal enhancement technology" in the claims of this application.

[0036] A high numerical aperture optical module for collecting scattered field data; Specifically, it includes the following steps: Step 3.1, constructing a high numerical aperture optical system; An objective lens with a numerical aperture NA > 0.9 and a dark field illumination device are used, and the theoretical resolution is better than 300 nm.

[0037] The system includes a main objective lens, an auxiliary objective lens, and a diaphragm system. The main objective lens is responsible for signal acquisition, and the auxiliary objective lens and the diaphragm system together form a dark field illumination system to suppress the interference of direct reflected light.

[0038] Step 3.2, recording the defect scattered field information; Utilizing the scattering characteristics of the defect on the incident light, the scattered field data is collected.

[0039] The scattered field can be expressed as: ; where, represents the scattered field distribution at the observation point ; is the Green's function, describing the position from the scatterer To the observation point Wave propagation; is the wave number, equal to , is the optical wave wavelength; is the refractive index distribution function of the defect region, describing the optical properties at the defect; is the refractive index of the background medium, i.e., the refractive index of the defect-free region; is the total electric field at the defect position, including the incident field and the scattered field; represents the integration region, i.e., the entire volume of the scatterer; represents at the position the volume element.

[0040] Step 3.3, adopt the angular scanning technique to expand the spatial frequency coverage range of the system; By changing the incident light angle or rotating the sample, multi-angle scattering data are obtained.

[0041] For each incident angle, the system can collect the spatial frequency components within a specific range. By synthesizing the scattering data of multiple angles, the expansion of the spatial frequency coverage range is achieved.

[0042] The output result of this module is a set of high-quality multi-angle scattering field data, containing rich spatial frequency information, providing a data basis for subsequent high-resolution three-dimensional reconstruction, solving the problem of limited resolution of traditional optical systems, and supporting the technical features regarding "high numerical aperture optical system" in the claims of this application.

[0043] The Fourier diffraction region inverse scattering reconstruction module is used to achieve the preliminary reconstruction of the three-dimensional defect distribution; Specifically, it includes the following steps: Step 4.1, establish a scattering model based on the Rytov approximation; This approximation is applicable to weak scattering conditions and represents the three-dimensional defect distribution as: ; where, represents the scattering potential distribution (related to the refractive index distribution), describing the scattering characteristics at the spatial position ; is the three-dimensional space coordinate vector; is the Fourier transform of the scattering potential, representing the representation of the scattering potential in the frequency domain; is the scattering vector, defined as the difference between the incident wave vector and the scattered wave vector, representing the momentum transfer during the scattering process; is the complex exponential function, representing the transformation kernel from the frequency domain to the spatial domain; is the spatial frequency range that the system can obtain, determined by the numerical aperture of the optical system and the measurement angle range; Represents a volume element in the frequency domain space, used for integral calculation; Represents the spatial frequency range that can be obtained by the system for integration within.

[0044] Step 4.2, Apply synthetic aperture technology to merge scattered data at different angles; According to Fourier diffraction theory, each scattering angle corresponds to a slice of the "Ewald sphere" in the Fourier space. Therefore, by merging data from multiple angles, the spatial frequency coverage of the system is extended, and the reconstruction resolution is improved.

[0045] The specific algorithm is as follows: Perform Fourier transform on the scattered field data for each incident angle to obtain its spatial spectrum; Map the spatial spectra of each angle to the corresponding positions in the three-dimensional Fourier space; Perform weighted average processing on the overlapping regions, where the weights are related to the signal intensity and coherence; Fill the missing regions in the spectral space through interpolation.

[0046] Step 4.3, Solve the inverse problem to reconstruct the three-dimensional defect distribution; Construct a system of linear equations: ; where, is the measurement data vector, representing the set of scattered field data collected from multiple angles, and each element corresponds to the scattered field value at a measurement point; is the system matrix (determined by the scattering model), describing the linear relationship between the scattering medium and the scattered field; is the defect distribution vector to be solved, representing the scattering potential distribution at each point in the three-dimensional space, and each element corresponds to the scattering characteristics of a voxel point in the space, which is directly related to the refractive index deviation at that point; is the measurement noise vector, including measurement uncertainties caused by various factors such as system noise, environmental interference, and quantization errors.

[0047] Solve this system of equations through the back-projection algorithm combined with conjugate gradient iteration to obtain a preliminary reconstruction result.

[0048] The back-projection algorithm first calculates the product of the adjoint matrix of the system matrix and the measurement data to provide an initial solution, while the conjugate gradient method gradually minimizes the reconstruction error through iteration to improve the reconstruction accuracy.

[0049] The output result of this module is the preliminary reconstruction result of the three-dimensional defect distribution on the sidewall of the through-hole of the glass wafer, with a resolution of 150 nm, which is better than the diffraction limit of the system and can distinguish surface defects and internal defects.

[0050] However, artifacts and noise may exist in the preliminary reconstruction results and further optimization is required.

[0051] This module implements the preliminary three-dimensional reconstruction of the scattered field data and supports the technical features regarding the "inverse scattering reconstruction algorithm in the Fourier diffraction region" in the claims of this application.

[0052] The compressive sensing iterative optimization module is used to improve the three-dimensional reconstruction accuracy and obtain the accurate three-dimensional defect distribution on the sidewall of the through-hole of the glass wafer; Specifically, it includes the following steps: Step 5.1, introducing the sparsity constraint of the defect distribution; In the morphology detection of the through-hole of the glass wafer, defects usually exhibit local distribution characteristics and have sparse representations in appropriate transform domains (such as the wavelet domain, gradient domain, etc.).

[0053] Based on this characteristic, a compressive sensing optimization problem is constructed: ; where, is the three-dimensional defect distribution vector to be solved, representing the scattering potential distribution at each point in space; is the sparse transform operator (such as wavelet transform, curvelet transform, etc.), which transforms the defect distribution into the sparse representation domain; represents the norm of the defect distribution in the sparse domain, used to measure the sparsity, and the smaller the value, the better the sparsity; represents norm, that is, the sum of the absolute values of the elements of the vector, used to promote the sparsity of the solution; represents norm, that is, the Euclidean norm of the vector, used to measure the data fitting error; is the forward scattering operator, describing the linear mapping relationship between the scattering medium and the scattered field; is the vector of the actually measured scattered field data, containing the scattered information collected at multiple angles; is the tolerance error threshold related to the system noise level, used to control the consistency between the reconstruction result and the measurement data; represents "subject to" (constrained by), indicating the constraint conditions of the optimization problem.

[0054] Step 5.2, using the Alternating Direction Method of Multipliers (ADMM) to solve the optimization problem; This algorithm transforms the original problem into: ; where, is the three-dimensional defect distribution vector to be solved, representing the scattering potential distribution at each point in space; is the sparse transform operator that transforms the defect distribution into the sparse representation domain; represents the norm of the defect distribution in the sparse domain, which is used to promote the sparsity of the solution; is the regularization parameter that controls the balance between data fidelity and sparsity; is the forward scattering operator that describes the linear mapping relationship between the scattering medium and the scattering field; is the vector of actually measured scattering field data; represents the mean square error between the reconstruction result and the measured data, which measures the data fitting degree; represents the norm, that is, the sum of the absolute values of the elements of the vector; represents the square norm, that is, the sum of the squares of the elements of the vector.

[0055] The ADMM algorithm decomposes the problem into multiple sub-problems for iterative solution by introducing auxiliary variables and Lagrange multipliers: Data fidelity sub-problem: ; where, represents the defect distribution vector obtained in the th iteration; represents finding the variable value that minimizes the objective function; represents the mean square error between the reconstruction result and the measured data, which measures the data fitting degree; is the ADMM algorithm parameter that controls the weight of the augmented Lagrangian term; is the auxiliary variable in the th iteration, which is used to separate the optimization problem; is the normalized Lagrange multiplier in the th iteration, which is used to coordinate the primary variable and the auxiliary variable; is the augmented Lagrangian term that ensures the consistency of the primary variable and the auxiliary variable during the iterative process. is the normalized Lagrange multiplier in the th iteration, which is used to coordinate the primary variable and the auxiliary variable; is the auxiliary variable in the th iteration, which is used to separate the optimization problem; is the ADMM algorithm parameter that controls the weight of the augmented Lagrangian term; Sparsity sub-problem: ; Among them, represents the auxiliary variable vector obtained in the th iteration, which is used to promote the sparsity of the solution; represents finding the variable that minimizes the objective function value; represents the variable after sparse transformation norm, which is used to measure the sparsity degree of the solution in the transformed domain; is the sparse transformation operator that converts the signal to a domain where it is easier to exhibit sparse characteristics; is the penalty parameter that controls the consistency strength between the auxiliary variable and the original variable ; represents the square of the difference between the auxiliary variable and the original variable norm, which ensures that the two variables tend to be consistent during the iteration process; is the updated defect distribution vector in the current iteration; is the Lagrange multiplier vector in the current iteration, which is used to coordinate the consistency between the original variable and the auxiliary variable.

[0056] Dual variable update: ; Among them, represents the normalized Lagrange multiplier vector updated in the th iteration; represents the normalized Lagrange multiplier vector in the th iteration; represents the defect distribution vector obtained in the th iteration; represents the auxiliary variable vector obtained in the th iteration; This update formula adjusts the Lagrange multiplier by accumulating the difference between the original variable and the auxiliary variable, promoting the algorithm to converge to a solution that satisfies the constraint conditions.

[0057] Step 5.3, combine the sidewall light reflection model to construct a complete sidewall unfolding diagram; Based on the illumination conditions and geometric information, establish a sidewall light reflection model to describe the reflection and scattering characteristics at different positions.

[0058] In addition, use this model to correct the reconstruction result and convert it to the sidewall unfolding coordinate system to generate an intuitive sidewall unfolding diagram, which is convenient for defect analysis and evaluation.

[0059] The output result of this module is a high-precision reconstruction result of the three-dimensional defect distribution on the sidewall of the glass wafer through hole. In addition, the defect distribution is visually presented through the sidewall unfolding diagram, providing a reliable basis for subsequent analysis and quality control.

[0060] This module is the final processing link of the entire detection process and supports the technical features regarding the "compressed sensing iterative optimization algorithm" in the claims of this application.

[0061] The following details the technical solutions of the embodiments of this application through specific application examples: Application example of TSV via detection for high-end processor glass wafers: In this application example, the detection method of this application is applied to the detection of TSV (Through-Silicon Via) vias of glass wafers for high-end processors produced by a certain semiconductor manufacturer.

[0062] The diameter of the TSV via is 10 μm, the depth is 200 μm, the aspect ratio is 20:1, and the minimum defect size to be detected is 100 nm.

[0063] Traditional detection methods face the following challenges: The via depth is relatively large, and it is difficult to obtain sufficient illumination in the sidewall area; The high aspect ratio results in weak scattered signals, especially for defects in the deep area; The 100-nm defect size is close to or below the diffraction limit of conventional optical systems, making it difficult to effectively resolve.

[0064] Detection system configuration: According to the application requirements, the following detection system is configured: Configuration of multi-angle light source collaborative illumination system: Main light source: An Nd:YAG laser with a wavelength of 532 nm, an output power of 100 mW, a beam diameter of 8 mm, and a divergence angle <1 mrad; Auxiliary light source array: 16 laser diode light sources with a wavelength of 532 nm, each with a power of 20 mW, and the incident angles are set at six gears of 20°, 30°, 40°, 50°, 60°, and 70°; Light source phase control: A mirror driven by a piezoelectric ceramic, with a phase adjustment accuracy of λ / 100 - Polarization control: A liquid crystal polarization modulator with a modulation frequency up to 1 kHz; Optical path separation system: Composed of 3 dichroic mirrors (transmittance >95%) and 2 polarization beam splitters (extinction ratio >1000:1).

[0065] Configuration of polarization modulation signal enhancement system: Polarization state generator: Composed of a quarter-wave plate, a half-wave plate, and a linear polarizer, capable of generating any polarization state; Electro-optic modulator: A KDP crystal electro-optic modulator with a modulation frequency of 10 kHz and a half-wave voltage of 3.5 kV; Holographic Interferometry System: An off-axis digital holographic device is adopted, with the angle between the reference light and the object light being 3°, and the phase recovery accuracy < λ / 20.

[0066] High Numerical Aperture Optical System Configuration: Main Objective Lens: Numerical Aperture NA = 0.95, working distance 0.3 mm, magnification 100×; Auxiliary Objective Lens: Numerical Aperture NA = 0.8, used for dark-field illumination; Camera: Scientific-grade CMOS camera, pixel size 3.45 μm, resolution 4096×3000, dynamic range 16 bit; Angle Scanning System: Six-axis precision sample stage, position accuracy ±0.1 μm, angle accuracy ±0.01°.

[0067] Data Processing System Configuration: Processor: Intel Xeon Gold 6258R, 28 cores and 56 threads, main frequency 2.7 GHz; GPU Acceleration: NVIDIA Tesla A100, 40 GB video memory; Memory: 512 GB DDR4-3200; Storage: 8 TB NVMe SSD array, read and write speed > 7 GB / s.

[0068] Detection Process and Parameter Settings: Sample Pretreatment: Sample Cleaning: Use an ultrasonic cleaning system for three-level cleaning with acetone, isopropyl alcohol, and deionized water, 3 minutes for each level; Surface Treatment: After drying, use oil-free compressed air to blow off the surface dust; Sample Fixing: Use a special vacuum adsorption sample stage to ensure the fixing accuracy < ±0.5 μm.

[0069] System Calibration: Light Source Calibration: Use a power meter to measure the power of each light source to ensure the stability fluctuation < ±1%; Polarization Calibration: Use a polarization analyzer to calibrate the polarization state, with an accuracy > 99.5%; Spatial Calibration: Use a standard grating sample (period 500 nm) to calibrate the spatial resolution; Phase Calibration: Use a standard phase step sample to calibrate the phase response.

[0070] Through-hole Illumination Optimization: Set the incident angle of the auxiliary light source according to the through-hole depth-to-width ratio (20:1): The deep area is mainly illuminated by light sources at angles of 50° - 70°, the middle area is illuminated by light sources at 30° - 50°, and the shallow area is illuminated by light sources at 20° - 30°; Light source phase difference setting: The phase difference between each light source is set to 120°, forming incoherent superposition to avoid interference fringes; Light source power distribution: The power of the deep illumination light source is increased by 50% compared to the shallow part to compensate for the intensity attenuation caused by the optical path difference.

[0071] Signal acquisition and enhancement: Polarization differential acquisition: For each illumination angle, scatter images in three polarization states of horizontal polarization (p polarization), vertical polarization (s polarization), and 45° polarization are acquired; Weight coefficient optimization: The optimal linear combination weights are determined through statistical analysis, p:s:45° = 0.5:0.3:0.2; Holographic interference parameters: The intensity of the reference light is 1.5 times the average intensity of the object light, and the spatial carrier frequency is 10 pixels / cycle; Exposure parameters: High dynamic range imaging technology is adopted, and the exposure times are three gears of 10ms, 50ms, and 200ms to synthesize a 16-bit dynamic range image.

[0072] Multi-angle scatter field acquisition: Sample rotation angles: Eight angles of 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°; 36 scatter field images are acquired at each angle (3 polarization states × 3 exposure times × 4 repetitions); Single scatter field image acquisition time: <300ms; Total time for all data acquisition: <120s / per through hole.

[0073] Inverse scattering reconstruction in the Fourier domain: Rytov approximation parameter setting: Assume the refractive index perturbation <0.01, meeting the weak scattering condition; Spatial frequency coverage range: Extended to 2.5k0 (k0 is the free space wave number) through multi-angle synthesis, equivalent to a theoretical resolution of λ / 5 ≈ 106nm; Voxel size for 3D reconstruction: 50nm × 50nm × 100nm (resolution in the xy plane is higher than in the z direction); Conjugate gradient iteration: The maximum number of iterations is 100, the convergence threshold is 1e-6, and the regularization parameter α = 0.01.

[0074] Compressive sensing optimization: Sparse transform selection: Curvelet transform is used for the sidewall expansion diagram, and 3D wavelet transform is used for the 3D reconstruction result; Regularization parameter μ: Adaptively set, with an initial value of 0.1, decreasing to 0.01 with iteration; ADMM algorithm parameters: ρ = 1.5, maximum number of iterations 500, number of inner iterations 20, convergence threshold 1e-8; Sidewall expansion coordinate transformation: Using the mapping from polar coordinates to Cartesian coordinates, with a resolution maintained at 50 nm.

[0075] Detection results and analysis: In this application example, a batch of glass wafers containing 1000 TSV vias were detected, and the following results were obtained: Detection efficiency: Complete detection time for a single via: <180 s (including 120 s for data acquisition and 60 s for data processing); System stability: No obvious performance degradation after continuous operation for 24 hours; Automation level: Fully automated throughout the process except for sample loading, and the training time for operators is <4 hours.

[0076] Defect classification statistics: Total number of detected defects: 1872, including 1256 surface defects and 616 internal defects; Defect size distribution: 100 - 200 nm accounts for 37%, 200 - 500 nm accounts for 45%, >500 nm accounts for 18%; Defect type distribution: Cracks account for 31%, pits account for 25%, particulate impurities account for 22%, bubbles account for 14%, and others account for 8%.

[0077] Detection performance indicators: Resolution: The actual test resolution reaches 105 nm, which is consistent with the theoretical expectation; Sidewall coverage rate: For vias with a depth - to - width ratio of 20:1, an effective coverage of 98.2% is achieved; Defect detection rate: Through manual sampling verification, the detection rate reaches 92.7%; False positive rate: <3%, mainly from system noise and residual contamination on the sample surface; Repeatability: Repeatedly measure the same position 10 times, with a position accuracy of <±50 nm and a size accuracy of <±10 nm.

[0078] Comparison with traditional methods: Resolution improvement: 2.5 times higher than that of ordinary optical microscopes and 1.8 times higher than that of confocal microscopes; Detection rate improvement: 52% higher than traditional optical methods and 33% higher than confocal methods; Sidewall coverage rate: 87% higher than traditional methods; Internal defect detection ability: Traditional methods can basically not detect internal defects, while this method realizes effective detection of internal defects up to 5 times the via diameter in depth.

[0079] Special case analysis: The following are several typical special cases found during the detection process and the handling methods: Case 1: Tiny cracks (about 120 nm in size) at a depth exceeding 180 μm; Detection challenge: Located in the bottom area of the through-hole, the scattering signal is extremely weak, and the signal-to-noise ratio < 3 dB; Solution: Increase the power of the 70° incident light source to 150%, and use 10 times of averaging for noise reduction; Result: The crack was successfully detected and located, with a position accuracy of ±75 nm, providing an important basis for subsequent process improvement.

[0080] Case 2: Internal air bubbles (about 150 nm in diameter and about 5 μm in depth) in the smooth area of the through-hole sidewall; Detection challenge: The surface is smooth, the reflection is strong, and the internal scattering signal is submerged; Solution: Enhance the polarization differential weight, adjust p:s:45° to 0.3:0.6:0.1, highlighting the sensitivity of the s polarization to internal defects; Result: The internal scattering signal was successfully separated, the air bubbles were detected, and their depth was accurately measured (error < ±0.5 μm).

[0081] Case 3: Detection of the edge area of a through-hole with a high aspect ratio of 20:1; Detection challenge: The geometric occlusion effect causes the edge signal to be missing; Solution: Add four additional acquisition angles of 315°, 337.5°, 22.5°, and 45° to supplement the edge area information; Result: Achieved a 99.1% coverage rate of the edge area and detected edge cracks that were originally easily missed.

[0082] The embodiments of the present invention have been described above, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A three-dimensional topography detection system for glass wafer through-holes based on machine vision, characterized in that, Including: A multi-angle light source collaborative illumination module for achieving comprehensive illumination of the through-holes of a glass wafer; A polarization modulation module for enhancing weak scattering signals and obtaining scattering field information with a high signal-to-noise ratio; A high numerical aperture optical module for collecting scattering field data; A Fourier diffraction region inverse scattering reconstruction module for achieving a preliminary reconstruction of the three-dimensional defect distribution; A compressive sensing iterative optimization module for improving the three-dimensional reconstruction accuracy and obtaining an accurate three-dimensional defect distribution on the sidewalls of the through-holes of the glass wafer.

2. The three-dimensional topography detection system for glass wafer through-holes based on machine vision according to claim 1, wherein, The multi-angle light source collaborative illumination module includes: A main light source for vertical illumination for obtaining information on the top and bottom of the through-hole; An array of multiple auxiliary light sources with adjustable angles dedicated to illuminating the sidewalls of the through-hole; A device for precisely controlling the phase and polarization state of the light source; An optical path separation mirror group for guiding light signals reflected at different angles to different sensors.

3. The three-dimensional topography detection system for through-holes of glass wafers based on machine vision according to claim 1, characterized in that, The polarization modulation module includes: A polarization modulation device for modulating the polarization state of the incident light; A polarization difference imaging device for enhancing the scattering signal; A holographic interferometry device for obtaining the phase information of the scattering field.

4. A three-dimensional topography detection system for glass wafer through-holes based on machine vision according to claim 1, wherein The high numerical aperture optical module includes: An objective lens with a numerical aperture NA > 0.9 and a dark field illumination device; A scattering field information recording device; An angle scanning device for expanding the spatial frequency coverage range of the system.

5. The three-dimensional topography detection system for glass wafer through-holes based on machine vision according to claim 1, wherein The Fourier diffraction region inverse scattering reconstruction module includes a processor configured to: Establish a scattering model based on the Rytov approximation; Apply synthetic aperture technology to merge scattering data at different angles; Solve the inverse problem and reconstruct the three-dimensional defect distribution.

6. The three-dimensional topography detection system for glass wafer through-holes based on machine vision according to claim 5, wherein, The synthetic aperture technology includes the following steps: Perform a Fourier transform on the scattering field data for each incident angle to obtain its spatial spectrum; Map the spatial spectra of each angle to corresponding positions in the three-dimensional Fourier space; Perform a weighted average process on the overlapping regions, where the weights are related to the signal intensity and coherence; Fill the missing regions in the spectral space through interpolation.

7. A three-dimensional topography detection system for glass wafer through-holes based on machine vision according to claim 1, characterized in that, The compressive sensing iterative optimization module includes a processor configured to perform the following operations: Introduce a sparsity constraint on the defect distribution; Use the alternating direction method of multipliers to solve the optimization problem; Combine the sidewall light reflection model to construct a complete sidewall unfolded view.

8. A three-dimensional topography detection system for glass wafer through-holes based on machine vision according to claim 7, characterized in that, The sparsity constraint is constructed as a compressive sensing optimization problem, which minimizes the objective function value in the sparse transform domain while ensuring that the error between the reconstruction result and the actual measurement data is within a preset tolerance range, where the allowable error is related to the system noise level; The optimization problem is solved using an iterative shrinkage algorithm, and the reconstruction result is subjected to a sparse domain threshold process in each iteration to gradually improve the reconstruction accuracy.

9. The three-dimensional topography detection system for glass wafer through holes based on machine vision according to claim 1, wherein, The processor of the compressive sensing iterative optimization module is further configured to perform the following operations: Perform adaptive threshold segmentation on the reconstructed three-dimensional defect distribution to distinguish surface defects and internal defects; Calculate the defect density, size distribution, and depth distribution based on the spatial distribution characteristics of the defects; Analyze the potential impact of the defects on the electrical performance in combination with the material optical property model; Generate a comprehensive detection report including three-dimensional visualization of the defects, sidewall unfolded view, and defect statistical information.

10. A computer-readable storage medium, characterized in that, It is used to store computer-readable instructions that, when read by a computer, can run a three-dimensional topography detection system for glass wafer through-holes based on machine vision as described in any one of claims 1-9.

Images