Method and system for parallel tomographic detection of internal defects of tgv via holes

Abstract

The application discloses a TGV via hole internal defect multi-light field parallel tomographic detection method and system, and relates to the technical field of semiconductor packaging. The method comprises the following steps: layer-by-layer imaging of a to-be-detected TGV via hole is performed by using a composite coherent modulation light field to obtain an original tomographic image sequence; phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence; the tomographic images are divided into fragments and distributed to multiple computing power units for parallel iterative reconstruction to obtain multiple partition three-dimensional contour point clouds; coordinate calibration and point cloud fusion are performed, and topological correction is performed to obtain a TGV via hole three-dimensional topological model; and spatial matching is performed with a standard TGV three-dimensional topological template, and TGV via hole internal defects are identified according to the spatial matching result. The technical problem of low three-dimensional detection precision of TGV via hole internal defects in the prior art is solved, and the technical effect of improving the three-dimensional detection precision of TGV via hole internal defects is achieved.

Description

Technical Field

[0001] This invention relates to the field of semiconductor packaging technology, specifically to a multi-field parallel tomography detection method and system for defects inside TGV vias. Background Technology

[0002] With the development of advanced semiconductor packaging technology, through-glass vias (TGVs), as an important structure for achieving vertical interconnects, are widely used in high-density packaging, wafer-level packaging, and 3D integration. TGVs typically have high aspect ratios, microscale structures, and complex internal wall morphologies. Defects such as wall cracks, residues, abnormal apertures, and localized defects may exist within them, affecting the quality of subsequent metal filling and the electrical performance and reliability of the package structure. Currently, defect detection in TGVs mainly employs optical microscopy, cross-sectional analysis, and 2D scanning. However, due to the narrow internal space of TGVs, the detection process is easily affected by factors such as glass substrate refraction, light propagation attenuation, and internal structural obstruction, resulting in incomplete defect information and difficulty in accurately reflecting the 3D morphological characteristics inside the via. Furthermore, existing detection methods suffer from insufficient 3D reconstruction accuracy, difficulty in defect spatial localization, and low detection efficiency when dealing with minute defects and complex internal wall structures, making it difficult to meet the requirements for accurate detection of the internal structure of TGVs in high-reliability semiconductor packaging processes. Summary of the Invention

[0003] This application provides a multi-field parallel tomographic detection method and system for internal defects in TGV through-holes, which solves the technical problem of low three-dimensional detection accuracy of internal defects in TGV through-holes in the prior art.

[0004] A first aspect of this application provides a multi-field parallel tomographic detection method for internal defects in TGV vias, the method comprising:

[0005] A composite coherent modulated light field is formed by coupling a parallel coherent backlight field with a zero-degree annular uniform light field. This composite coherent modulated light field is then used to perform layered imaging of the TGV via under test, obtaining an original tomographic image sequence. Based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the via, phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence. The standardized tomographic image sequence is then segmented according to spatial regions, and the resulting tomographic image segments are distributed to multiple computing units. Parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV via structure to obtain multiple partitioned three-dimensional contour point clouds. The coordinates of the multiple partitioned three-dimensional contour point clouds are calibrated and fused. Topological correction is then performed based on the taper characteristics and inner wall morphology characteristics of the TGV via to obtain a three-dimensional topological model of the TGV via. The three-dimensional topological model of the TGV via is spatially matched with a standard TGV three-dimensional topological template. Based on the spatial matching results, internal defects of the TGV via are identified, and the defect type, spatial location, and morphological size of the internal defects are output.

[0006] A second aspect of this application provides a multi-field parallel tomographic detection system for internal defects in TGV through-holes, the system comprising:

[0007] Layered Imaging Module: A parallel coherent backlight field and a zero-degree annular uniform light field are coupled to form a composite coherent modulated light field. This composite coherent modulated light field is then used to perform layered imaging of the TGV via under test, obtaining the original tomographic image sequence. Image Processing Module: Based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the via, phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence. Partitioned Point Cloud Generation Module: The standardized tomographic image sequence is segmented according to spatial regions, and the resulting tomographic image segments are distributed to multiple... The computing unit uses an optical imaging model adapted to the TGV through-hole structure for parallel iterative reconstruction to obtain multiple partitioned 3D contour point clouds; the topology correction module performs coordinate calibration and point cloud fusion on the multiple partitioned 3D contour point clouds, and performs topology correction based on the taper characteristics and inner wall morphology characteristics of the TGV through-hole to obtain a 3D topology model of the TGV through-hole; the defect identification module performs spatial matching between the 3D topology model of the TGV through-hole and the standard TGV 3D topology template, identifies internal defects of the TGV through-hole based on the spatial matching results, and outputs the defect type, spatial location, and morphological size of the internal defects.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] First, a composite coherent modulated light field is formed by coupling a parallel coherent backlight field with a zero-degree annular uniform light field. This composite coherent modulated light field is then used to perform layered imaging of the TGV via under test, obtaining the original tomographic image sequence. Next, based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the via, phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence. Then, the standardized tomographic image sequence is segmented according to spatial regions, and the resulting tomographic image segments are distributed to multiple computing units. Parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV via structure to obtain multiple partitioned 3D contour point clouds. Further, coordinate calibration and point cloud fusion are performed on the multiple partitioned 3D contour point clouds, and topological correction is performed based on the taper characteristics and inner wall morphology characteristics of the TGV via to obtain a 3D topological model of the TGV via. Finally, the 3D topological model of the TGV via is spatially matched with a standard TGV 3D topological template. Based on the spatial matching results, internal defects of the TGV via are identified, and the defect type, spatial location, and morphological size of the internal defects are output. This invention solves the technical problem of low accuracy in three-dimensional detection of internal defects in TGV through-holes in existing technologies, and achieves the technical effect of improving the accuracy of three-dimensional detection of internal defects in TGV through-holes. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the multi-field parallel tomographic detection method for internal defects in TGV through-holes provided in an embodiment of this application;

[0012] Figure 2 This is a schematic diagram of the multi-field parallel tomography detection system for internal defects in TGV through-holes provided in an embodiment of this application;

[0013] Figure 3 This is a schematic diagram of the detection results of the orifice on the upper surface of the TGV through hole provided in an embodiment of this application;

[0014] Figure 4 This is a schematic diagram of the detection results of the orifice on the lower surface of the TGV through hole provided in an embodiment of this application;

[0015] Figure 5 This is a schematic diagram of the TGV through-hole three-dimensional hole reconstruction result provided in an embodiment of this application.

[0016] Figure labeling: Layered imaging module 11, image processing module 12, partitioned point cloud generation module 13, topology correction module 14, defect recognition module 15. Detailed Implementation

[0017] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0018] Example 1, as Figure 1 As shown, this application provides a multi-field parallel tomographic detection method for internal defects in TGV through-holes, wherein the method includes:

[0019] A composite coherent modulated light field is formed by coupling a parallel coherent backlight field with a zero-degree annular uniform light field. The composite coherent modulated light field is then used to perform layered imaging of the TGV aperture under test to obtain the original tomographic image sequence.

[0020] Furthermore, a composite coherent modulated light field is formed by coupling a parallel coherent backlight field with a zero-degree annular uniform light field, including:

[0021] The parallel coherent backlight field is controlled to transmit through the glass substrate and the interior of the TGV via to be detected along the axial direction, forming a transmitted light field characterizing the internal structure of the via. The zero-degree annular uniform light field is controlled to be uniformly projected along the circumference of the TGV via to be detected, forming an annular illumination light field characterizing the via wall profile and internal scattering characteristics. The parallel coherent backlight field and the zero-degree annular uniform light field are synchronously modulated according to a unified acquisition clock, so that the transmitted light field and the annular illumination light field work together at each tomographic acquisition position to form a composite coherent modulated light field.

[0022] First, the beam output from the coherent light source is processed by a collimation and beam expansion assembly and a parallel light shaping assembly to form a parallel coherent backlight field propagating along a fixed propagation direction. The incident direction of the parallel coherent backlight field is adjusted according to the central axis of the TGV via to be detected, so that the parallel coherent backlight field passes through the glass substrate and the internal region of the via along the TGV via axis. During the propagation of the parallel coherent backlight field, the difference in optical parameters between the glass substrate and the internal structure of the via is used to generate phase and intensity changes in the transmitted light field corresponding to the internal structure of the via, thus obtaining a transmitted light field carrying information about the axial depth of the via. At the same time, a ring-shaped incident light is generated by a ring light source, and the incident light is homogenized by a ring homogenizing assembly, so that the formed zero-degree ring homogenizing light field is uniformly projected onto the via wall region along the circumference of the via with the central axis of the TGV via as a reference. The zero-degree ring homogenizing light field causes reflection, scattering, and interference responses on the surface of the via wall and in the internal defect region, so that the ring illumination light field carries information about the via wall contour and the internal scattering characteristics of the via. Furthermore, a unified time reference is established between the parallel coherent backlight field and the zero-degree annular uniform light field through the acquisition control unit. The on-time, modulation frequency, and acquisition trigger signal of the two light fields are synchronously controlled so that the transmitted light field and the annular illumination light field act simultaneously on the TGV aperture to be detected at the same tomographic acquisition depth and the same spatial position. By adjusting the light intensity ratio, phase relationship, and spatial overlap area of ​​the two light fields, the axial structural response information in the transmitted light field and the aperture wall contour response information in the annular illumination light field are optically coupled to form a composite coherent modulated light field that simultaneously contains the axial structural information, radial contour information, and internal scattering anomaly information of the aperture, providing a composite optical response signal for subsequent tomographic image acquisition.

[0023] Preferably, the TGV via to be tested is fixed on a precision displacement platform, and a tomographic scanning coordinate system is established according to the axial direction of the TGV via. The composite coherent modulation light field is controlled to be incident on the TGV via to be tested, so that the axial transmission response information and circumferential scattering response information contained in the composite coherent modulation light field act on different structural regions inside the via. By setting the axial scanning step size of the imaging acquisition component, the imaging focal plane is moved layer by layer along the depth direction of the TGV via. Each time it moves to a preset tomographic acquisition position, the optical response signal corresponding to the current focal plane is acquired, and the optical response signal is captured by an image sensor. The response signal is converted into a two-dimensional tomographic image. During the single-layer image acquisition process, the intensity distribution, phase change, and scattering response of the transmitted light after the composite coherent modulated light field passes through the TGV via are recorded synchronously, so that the obtained two-dimensional tomographic image simultaneously contains the cross-sectional structure information of the via and the defect response information corresponding to the current depth position. According to the axial coordinates corresponding to each tomographic acquisition position, the two-dimensional tomographic images acquired at multiple depth positions are spatially sorted and calibrated to establish the correspondence between the tomographic image and the internal spatial position of the TGV via, forming an original tomographic image sequence containing structural information of the via at different axial depths.

[0024] Based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the via, phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence.

[0025] Furthermore, phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence, including:

[0026] The refractive index, medium dispersion parameters, glass propagation thickness, and imaging wavelength of the glass substrate corresponding to the TGV via to be detected are obtained. Based on the refractive index, medium dispersion parameters, glass propagation thickness, and imaging wavelength, the medium phase offset corresponding to each tomographic acquisition position is calculated. A glass medium phase compensation operator is constructed based on the medium phase offset. The phase component in each original tomographic image is reverse-corrected using the glass medium phase compensation operator. The consistency of the reverse correction result is constrained according to the phase continuity relationship between adjacent tomographic acquisition positions to obtain a phase-compensated tomographic image sequence. Local gray-level abrupt change features, edge gradient features, and texture discontinuity features are extracted from the phase-compensated tomographic image sequence. Candidate defect pixels are screened based on the continuity of response at the same spatial location in adjacent tomographic images. The defect response weight is determined based on the gray-level deviation, edge response intensity, and cross-layer response continuity of the candidate defect pixels. The candidate defect pixels are weighted and enhanced according to the defect response weight. The weighted enhanced tomographic images are then subjected to gray-level normalization and image size unification processing to obtain a standardized tomographic image sequence.

[0027] The refractive index, medium dispersion parameters, glass propagation thickness, and imaging light wavelength of the glass substrate corresponding to the TGV through-hole to be detected are obtained. Specifically, an ellipsometer is used to measure the refractive index variation curve of the glass substrate within the imaging light wavelength range to obtain the refractive index parameters corresponding to different wavelengths. The Sellmeier dispersion formula is used to fit the refractive index variation to obtain the medium dispersion parameters. Based on the refractive index, medium dispersion parameters, glass propagation thickness, and imaging light wavelength, the phase delay of the composite coherent modulated light field after passing through the glass substrate is calculated. Specifically, according to the first... The phase shift of the medium is calculated based on the light propagation path length corresponding to the tomographic acquisition position and the refractive index of the glass medium. ,in, This represents the phase offset of the medium corresponding to the i-th tomographic acquisition position; Indicates the wavelength of the imaging light; Indicates the wavelength of the glass substrate The refractive index below; Indicates the refractive index of air; This represents the light propagation thickness corresponding to the i-th tomographic position.

[0028] Furthermore, a glass medium phase compensation operator is constructed based on the medium phase shift corresponding to each chromatography acquisition position: ,in, This represents the phase compensation operator corresponding to the i-th tomographic image; Represents the image's frequency domain coordinates; This represents the phase shift at the corresponding spatial frequency position. A two-dimensional Fourier transform is performed on each tomographic image to obtain the frequency domain image: ,in, This represents the original tomographic image of the i-th layer; This represents the corresponding frequency domain image. The phase inversion correction is performed on the frequency domain image using the phase compensation operator described above: The phase-compensated tomographic image was obtained by inverse Fourier transform: in, This represents the tomographic image of the i-th layer after medium phase compensation.

[0029] Furthermore, to reduce the phase jump caused by propagation errors between adjacent tomographic positions, phase continuity is calculated for corresponding spatial positions in adjacent tomographic images. Taking the adjacent... Layer and first Same coordinate points in layer image The phase value is calculated using differential calculation: ,in, This indicates the amount of phase change between adjacent layers; Indicates the first The phase value after layer compensation. When the phase change is greater than 0.15 rad, the corresponding region is identified as an abnormal phase region, and the average phase value of the three adjacent layers is used for correction. This yields a tomographic image sequence after phase continuity correction.

[0030] Subsequently, defect pixel detection is performed on the phase-compensated tomographic image sequence. For each pixel in the tomographic image, the local grayscale change value is calculated: ,in, Indicates pixel grayscale deviation; Represents the pixel centered at the current pixel. The average gray value of the neighborhood. The Sobel operator is used to calculate the edge gradient intensity corresponding to the pixel position: ,in, and These represent the gradient values ​​in the horizontal and vertical directions, respectively. Simultaneously, the response changes of the same spatial coordinates in successive tomographic images are statistically analyzed to calculate the cross-layer response continuity: ,in, This indicates the number of layers in a continuous sequence that satisfy the grayscale variation range; N represents the total number of layers involved in the analysis. In this embodiment, five consecutive layers of images are used for statistical analysis. When there is an abnormal response at the same location in three or more consecutive layers, that location is identified as a candidate defect pixel.

[0031] Furthermore, the defect response weights are calculated based on grayscale deviation, edge gradient intensity, and cross-layer response continuity: ,in, The value represents the defect response weight corresponding to the pixel location; 0.4, 0.3 and 0.3 represent the fusion weights for grayscale difference, edge features and cross-layer continuity, respectively.

[0032] Enhancement processing is performed on candidate defect pixels based on the defect response weights: ,in, This represents the enhanced pixel grayscale value. After enhancement, grayscale normalization is performed on each tomographic image: ,in, This represents the normalized grayscale value of the image; and These represent the maximum and minimum grayscale values ​​in the enhanced image, respectively. Finally, all tomographic images were uniformly adjusted to a resolution of 2048×2048 pixels and spatially calibrated according to a tomographic acquisition interval of 0.5μm, resulting in a standardized tomographic image sequence with a uniform grayscale range, uniform spatial scale, and uniform interlayer positional relationship.

[0033] The standardized tomographic image sequence is segmented according to spatial regions, and the resulting tomographic image segments are distributed to multiple computing units. Parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV through-hole structure to obtain multiple partitioned three-dimensional contour point clouds.

[0034] Furthermore, parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV through-hole structure, including:

[0035] Based on the spatial location of each tomographic image slice, spatial coordinate identifiers and adjacent overlapping areas are set for each tomographic image slice, and each tomographic image slice is assigned to a corresponding computing unit. Each computing unit uses the optical imaging model adapted to the TGV through-hole structure to perform forward imaging on the initialized three-dimensional voxel distribution, calculates the imaging residual between the forward imaging result and the corresponding tomographic image slice, and iteratively corrects the three-dimensional voxel distribution based on the imaging residual. When the imaging residual meets the preset iteration termination condition, the through-hole contour points are extracted from the corrected three-dimensional voxel distribution to obtain the corresponding partitioned three-dimensional contour point cloud.

[0036] First, based on the spatial coordinate information corresponding to each tomographic image in the standardized tomographic image sequence, the two-dimensional tomographic images are divided according to the lateral spatial region of the TGV aperture. A single image slice size of 512×512 pixels is used, and a single-layer tomographic image of 2048×2048 pixels is divided into multiple spatial sub-regions. A corresponding spatial coordinate identifier is established for each spatial sub-region, including the slice number, the range of two-dimensional pixel coordinates, and the corresponding axial tomographic depth position. Simultaneously, a 64-pixel wide overlap region is set between adjacent image slices to ensure that adjacent slices contain some of the same structural information, guaranteeing continuous spatial constraints during subsequent point cloud fusion. Based on the spatial coordinate identifiers corresponding to each image slice, the tomographic image slices of different spatial regions are allocated to multiple parallel computing units through a task scheduling unit. Each computing unit independently performs the 3D reconstruction calculation for its corresponding region. Furthermore, after receiving the corresponding tomographic image slices, each computing unit establishes an initial three-dimensional voxel space based on the optical imaging model adapted to the TGV through-hole structure. Specifically, a three-dimensional voxel grid is established along the axial direction of the TGV through-hole, and an initial optical response value is set for each voxel position, forming an initial three-dimensional voxel distribution matrix. ,in, Indicates the spatial position when the initial iteration number is 0. The corresponding voxel response value.

[0037] Using the forward imaging operator in the optical imaging model, the three-dimensional voxel distribution in the current iteration state is projected and calculated to obtain the theoretical tomographic image: ,in, This represents the forward imaging result corresponding to the k-th image slice in the m-th iteration; This represents the optical projection operator corresponding to the k-th spatial region; This represents the three-dimensional voxel distribution at the m-th iteration.

[0038] The difference between the forward imaging result and the actual acquired tomographic image is calculated by segmenting the images to obtain the imaging residual: ,in, This represents the imaging residual corresponding to the k-th slice in the m-th iteration; This represents the k-th tomographic image slice actually acquired.

[0039] Furthermore, the three-dimensional voxel distribution is iteratively updated based on the imaging residuals. A residual back-projection method is used to map the two-dimensional image residual information to the corresponding three-dimensional voxel space, and corrections are made according to the voxel positions. ,in, This represents the three-dimensional voxel distribution after the (m+1)th iteration; The back projection operator represents the forward imaging operator; This represents the voxel update step size, which is set to 0.1 in this embodiment.

[0040] After each iteration, the residual mean square error is calculated for the current reconstruction result: ,in, The m-th iteration represents the reconstruction error; N represents the number of pixels in the current tomographic image slice. This represents the actual pixel value collected; This represents the corresponding pixel value generated in the m-th iteration. When the change in reconstruction error is less than 0.001 during 5 consecutive iterations, or when the number of iterations reaches 200, the iterative calculation of the current computing unit is stopped, and the converged three-dimensional voxel distribution is obtained.

[0041] Furthermore, TGV via contour points are extracted from the converged 3D voxel distribution. Specifically, the 3D voxel space is thresholded based on the voxel response values, and regions with voxel response values ​​greater than 0.6 are identified as via structure regions. A 3D edge extraction algorithm is used to extract the boundary voxels of the via inner wall. The coordinates of the boundary points are calculated based on the spatial connection relationship between adjacent boundary voxels, forming a set of 3D contour points corresponding to the current spatial segment. ,in, This represents the 3D contour point cloud corresponding to the k-th spatial piece; This represents the 3D coordinates of the i-th contour point. Finally, each computing unit outputs the corresponding 3D contour point cloud for the spatial region, providing a data foundation for subsequent point cloud coordinate calibration, fusion, and the construction of the TGV via 3D topology model.

[0042] Furthermore, the construction of the optical imaging model adapted to the TGV through-hole structure includes:

[0043] A three-dimensional voxel mesh is established based on the axial tomographic position and aspect ratio of the TGV via to be detected, and the spatial projection relationship between the three-dimensional voxel mesh and each tomographic image segment is determined according to the axial extension direction of the via. Based on the light transmission characteristics of the TGV glass substrate, the imaging wavelength, and the diffraction resolution of the imaging system, the light propagation attenuation parameters and diffraction diffusion parameters corresponding to each axial tomographic position are determined. These light propagation attenuation parameters and diffraction diffusion parameters are loaded onto the spatial projection relationship to construct a forward imaging operator for mapping the three-dimensional voxel distribution to a tomographic image. Through-hole structural constraints are established based on the continuity of the TGV via axis, the continuity of the via wall profile, and the axial variation characteristics of the via diameter. These through-hole structural constraints are combined with the forward imaging operator to obtain the optical imaging model adapted to the TGV via structure.

[0044] Preferably, a three-dimensional voxel mesh is established based on the axial tomographic position, depth, diameter, and aspect ratio parameters of the TGV via to be detected. The axial extension direction of the TGV via is used as the Z-axis of the three-dimensional coordinate system, and the radial direction of the via is used as the X and Y axes, respectively. The space to be reconstructed is divided into multiple regular three-dimensional voxel units, and each three-dimensional voxel unit is assigned corresponding spatial coordinates and initial optical response values. Based on the acquisition depth, magnification, and focal plane position of each tomographic image slice, a spatial projection relationship between the three-dimensional voxel coordinates and the pixel coordinates of the two-dimensional tomographic image is established, so that each three-dimensional voxel unit can be mapped to the pixel position in the corresponding tomographic image slice. The spatial projection relationship is expressed as follows: ,in, Represents three-dimensional voxel space coordinates; Represents the pixel coordinates of a two-dimensional tomographic image; This represents the projection matrix from three-dimensional space to two-dimensional image space.

[0045] Furthermore, based on the light transmission characteristics of the TGV glass substrate, the imaging light wavelength, and the diffraction resolution parameters of the imaging system, the light propagation attenuation parameters and diffraction diffusion parameters corresponding to each axial tomographic position are determined. Specifically, based on the absorption and scattering effects of the glass substrate during light propagation, the light propagation attenuation coefficients corresponding to different propagation depths are calculated: ,in, This represents the attenuation coefficient of light after it has traveled to a depth of z. denoted by , which represents the light propagation attenuation parameter corresponding to the TGV glass substrate; z represents the light propagation distance.

[0046] Simultaneously, the diffraction diffusion range of the optical system is determined based on the imaging light wavelength and numerical aperture, and the spot diffusion characteristics at different axial positions are described using the point spread function: ,in, Represents the three-dimensional point spread function; This represents the diffraction and diffusion parameters determined by the imaging wavelength, numerical aperture, and axial defocus.

[0047] The light propagation attenuation parameter and diffraction diffusion parameter are loaded into the spatial projection relationship to construct a forward imaging operator for mapping the three-dimensional voxel distribution to a tomographic image. Specifically, based on the three-dimensional voxel response value, the light propagation attenuation coefficient, and the diffraction diffusion response relationship, the two-dimensional pixel response corresponding to each voxel is accumulated and calculated to obtain the simulated tomographic image. ,in, This represents a two-dimensional tomographic image obtained through forward imaging; Represents the three-dimensional voxel response value; This represents the light propagation attenuation coefficient corresponding to the propagation depth; This represents the diffraction and diffusion response of a voxel after it passes through an optical system.

[0048] Furthermore, through-hole structural constraints are established based on the continuity of the TGV through-hole axis, the continuity of the through-hole wall profile, and the axial variation characteristics of the through-hole diameter. Specifically, based on the structural characteristic of the TGV through-hole penetrating the glass substrate axially, multiple axial sections are extracted along the axial direction of the through-hole in the established three-dimensional voxel mesh, and the corresponding through-hole center position is determined according to the voxel distribution of the through-hole region in each axial section; the offset of the through-hole center position between adjacent axial sections is calculated, and the offset of the center position characterizes the degree of change of the through-hole axis, establishing a through-hole axis continuity constraint to ensure that the through-hole center at adjacent axial positions remains continuously changing during the three-dimensional voxel reconstruction process. The through-hole axis continuity constraint is expressed as: ,in, Indicates the continuity constraint value of the hole shaft; The coordinates of the through hole center corresponding to the i-th axial section are given; N represents the number of axial sections.

[0049] Furthermore, based on the morphological characteristics of the continuously varying inner wall of the TGV through-hole along the depth direction, hole wall contour points are extracted in each axial section. Then, based on the spatial correspondence of hole wall contour points between adjacent axial sections, the positional change between corresponding hole wall points in adjacent sections is calculated to characterize the degree of continuous variation of the hole wall contour along the axial direction. This establishes a hole wall contour continuity constraint, ensuring that the reconstructed hole wall surface remains smooth and continuous. The hole wall contour continuity constraint is expressed as follows: ,in, This represents the constraint value for the continuity of the hole wall profile; This represents the coordinates of the j-th hole wall profile point in the i-th axial section; M represents the number of hole wall profile points in a single axial section.

[0050] Furthermore, based on the tapered structural characteristics formed during the TGV through-hole machining process, the radial distances of the hole wall contour points corresponding to different axial positions are calculated to obtain the equivalent hole diameter of each axial section, and the hole diameter variation gradient between adjacent axial sections is calculated. Based on the deviation between the hole diameter variation gradient and the designed tapered diameter of the TGV through-hole, an axial variation constraint on the hole diameter is established, ensuring that the reconstructed through-hole diameter varies along the axial direction according to the actual tapered pattern formed during machining. The axial variation constraint on the hole diameter is expressed as follows: ,in, This indicates the constraint value for axial variation of the aperture; This represents the equivalent aperture of the i-th axial section; This represents the theoretical diameter variation of adjacent axial sections, determined by the taper of the TGV through-hole machining.

[0051] The hole axis continuity constraint, hole wall contour continuity constraint, and hole diameter axial variation constraint are fused to form a through-hole structure constraint term suitable for the structural characteristics of TGV through holes. Specifically, the hole axis continuity constraint value, hole wall contour continuity constraint value, and hole diameter axial variation constraint value are weighted and fused to obtain a comprehensive structural constraint term: ,in, This represents the comprehensive structural constraint term corresponding to the three-dimensional voxel distribution; Indicates the continuity constraint value of the hole shaft; This represents the constraint value for the continuity of the hole wall profile; This indicates the constraint value for axial variation of the aperture; , , These represent the weighting coefficients of the corresponding structural constraints. The weighting coefficients are determined based on the degree of influence of different structural constraints on the reconstruction accuracy of TGV vias. In this embodiment, they are set to 0.4, 0.35, and 0.25, respectively.

[0052] Finally, the integrated structural constraint term is introduced into the 3D reconstruction objective function corresponding to the forward imaging operator, so that the forward imaging operator, when solving for the 3D voxel distribution, simultaneously satisfies the actual optical propagation law and the TGV aperture structure constraint, forming an optical imaging model adapted to the TGV aperture structure: Where V represents the three-dimensional voxel distribution to be solved; I represents the actual acquired tomographic image; This represents a forward imaging operator constructed based on spatial projection relations, light propagation attenuation parameters, and diffraction diffusion parameters; Represents the comprehensive structural constraint terms; This represents the structural constraint adjustment coefficient.

[0053] Coordinate calibration and point cloud fusion are performed on the three-dimensional contour point clouds of the multiple partitions, and topological correction is performed based on the taper characteristics and inner wall morphology characteristics of the TGV through hole to obtain the three-dimensional topological model of the TGV through hole.

[0054] Furthermore, coordinate calibration and point cloud fusion are performed on the multiple partitioned 3D contour point clouds, and topological correction is performed based on the taper characteristics and inner wall morphology characteristics of the TGV through-hole to obtain a 3D topological model of the TGV through-hole, including:

[0055] Based on the spatial coordinate identifier, the three-dimensional contour point clouds of each partition are transformed to a unified three-dimensional coordinate system. Corresponding feature points of the three-dimensional contour point clouds of adjacent partitions are extracted in the adjacent overlapping area, and the point cloud coordinate transformation parameters are determined based on the positional deviation between the corresponding feature points. The point cloud coordinate transformation parameters are used to register the three-dimensional contour point clouds of adjacent partitions, and the registered overlapping point clouds are fused to obtain a complete through-hole contour point cloud. Multiple hole wall sections are extracted from the complete through-hole contour point cloud along the axial direction of the TGV through-hole. The taper deviation and inner wall morphology deviation are determined based on the center position and equivalent hole diameter change of each hole wall section, and the complete through-hole contour point cloud is corrected based on the taper deviation and inner wall morphology deviation to obtain a three-dimensional topological model of the TGV through-hole.

[0056] After obtaining the 3D contour point clouds of multiple partitions, the point clouds of each partition are transformed into a unified 3D coordinate system based on the spatial coordinate identifiers corresponding to the 3D contour point clouds of each partition. Specifically, based on the spatial positional relationships established during the tomographic image segmentation process, the spatial transformation relationship between the local coordinate system and the global coordinate system corresponding to the point clouds of each partition is obtained, and the coordinate transformation is completed using rotation matrices and translation vectors. The transformation relationship is expressed as follows: ,in, This represents the transformed global coordinate point; Let R represent the local coordinates of a point in the partitioned point cloud; R represents the spatial rotation matrix; and T represents the spatial translation vector. After coordinate transformation, corresponding feature points are extracted within the overlapping region between adjacent partitioned 3D contour point clouds. Specifically, this is achieved by calculating the normal vector and curvature features of the neighborhood of each point within the overlapping region, selecting points with similar local geometric features as matching point pairs, and solving for the point cloud registration parameters based on the positional deviation between the matching point pairs. For the , The registration error of a set of corresponding feature point pairs is expressed as follows: ,in, This indicates the point cloud registration error; Represents the first point cloud in the first partition. One feature point; represents the corresponding feature points in adjacent partition point clouds; N represents the number of corresponding feature points. By solving for the rotation matrix and translation vector that minimizes the registration error, the coordinate transformation parameters between adjacent partition point clouds are obtained, and the coordinate transformation parameters are used to perform spatial registration of the 3D contour point clouds of each partition.

[0057] After completing the registration of adjacent partition point clouds, duplicate points within the overlapping area are fused. Specifically, the spatial distance between the registered overlapping point clouds is used to determine whether they belong to the same contour point. When the distance between two point cloud points is less than 0.1 μm, the two points are considered to be at the same spatial location, and a fusion point is generated using a position averaging method. ,in, Indicates the coordinates of the merged contour points; The first to participate in the integration The coordinates of each corresponding point are given; M represents the number of points involved in the fusion. For point cloud data within non-overlapping areas, the original spatial coordinates are retained, thus obtaining a complete through-hole contour point cloud containing complete TGV through-hole inner wall contour information.

[0058] Further, the complete through-hole profile point cloud is analyzed along the axial direction of the TGV through-hole. Specifically, the complete through-hole profile point cloud is sliced ​​at axial intervals of 0.5 μm to obtain multiple hole wall cross-section point sets, and each hole wall cross-section is subjected to circle fitting to obtain the center position and equivalent aperture of the corresponding cross-section. For the k-th hole wall cross-section, its equivalent aperture is calculated as follows: ,in, This represents the equivalent aperture of the k-th hole wall section; This indicates the location of the fitting center of the cross section; Indicates the coordinates of the i-th hole wall point; This indicates the number of hole wall points contained in the cross-section. For example... Figure 3 As shown, the aperture boundary is extracted from the set of cross-sectional points corresponding to the upper surface of the TGV through-hole to be inspected. A circle is fitted to the extracted aperture boundary points, and the aperture diameter of the upper surface is determined based on the diameter of the fitted circle. The roundness of the upper surface is determined based on the radial distance deviation from each aperture boundary point to the center of the fitted circle. In this embodiment, the aperture diameter of the upper surface of the TGV through-hole to be inspected is 98.426 μm, and the roundness is 0.888, thereby obtaining the aperture size and contour morphology of the upper surface of the TGV through-hole to be inspected.

[0059] like Figure 4As shown, the aperture boundary is extracted and a circle is fitted to the point set of cross-sectional points corresponding to the lower surface of the TGV through hole to be tested. The aperture diameter and roundness of the lower surface are determined based on the fitting results. In this embodiment, the aperture diameter of the lower surface of the TGV through hole to be tested is 100.957 μm, and the roundness is 0.786. By associating the aperture diameter of the upper surface and the aperture diameter of the lower surface with their corresponding axial positions, the basic data of the aperture diameter variation along the axial direction of the TGV through hole are obtained for subsequent taper variation feature calculation.

[0060] Based on the equivalent aperture variation relationship between adjacent axial sections, the taper variation characteristics of the TGV through hole are calculated: ,in, This represents the aperture variation gradient in the k-th axial interval; and These represent the equivalent apertures of two adjacent cross sections; and This indicates the axial position of the corresponding cross-section. The taper deviation region of the through hole is determined based on the deviation between the aperture variation gradient and the standard taper curve.

[0061] Simultaneously, based on the variation of radial distance from the hole wall point to the cross-sectional fitting center, the inner wall morphology deviation is calculated: ,in, This represents the deviation in the inner wall morphology of the k-th section; This represents the actual radial distance corresponding to the i-th hole wall point; This represents the average radius of the cross-section. Based on the taper deviation and inner wall morphology deviation, the complete through-hole contour point cloud is locally corrected. Specifically, when the aperture change gradient of multiple consecutive axial cross-sections deviates from the standard taper range by more than 5%, the radial coordinates of the contour points in the corresponding region are proportionally adjusted. When the inner wall morphology deviation exceeds 0.2 μm, the abnormal points are smoothed and corrected based on the surface fitting results of adjacent hole wall points, ensuring that the corrected point cloud maintains the continuity of the hole axis and hole wall. Finally, the contour point cloud after coordinate calibration, point cloud fusion, and topology correction is reconstructed to generate a TGV through-hole 3D topology model containing information on hole axis position, hole wall surface, aperture change, and local morphology anomalies.

[0062] like Figure 5 As shown, the complete through-hole contour point cloud after coordinate calibration, point cloud fusion and topology correction is connected by surface according to the corresponding three-dimensional coordinates to form a three-dimensional hole reconstruction result characterizing the internal hole wall structure of the TGV through-hole. The three-dimensional hole reconstruction result retains the hole diameter variation, hole wall surface undulation and local morphological changes of the TGV through-hole along the axial direction, and is used to construct the three-dimensional topology model of the TGV through-hole, and to provide three-dimensional structural data for subsequent spatial matching with the standard TGV three-dimensional topology template.

[0063] The TGV through-hole 3D topology model is spatially matched with the standard TGV 3D topology template. Based on the spatial matching result, internal defects of the TGV through-hole are identified, and the defect type, spatial location, and shape size of the internal defects are output.

[0064] Furthermore, the TGV via 3D topology model is spatially matched with a standard TGV 3D topology template, and internal defects of the TGV via are identified based on the spatial matching results, including:

[0065] The TGV through-hole 3D topology model is aligned with the standard TGV 3D topology template in terms of orifice center alignment, through-hole axis alignment, and axial dimension alignment. The local spatial deviation between the aligned TGV through-hole 3D topology model and the standard TGV 3D topology template is calculated, and the region where the local spatial deviation continuously exceeds the preset defect deviation threshold is identified as a candidate defect region. Based on the spatial position, connectivity, extension direction, and morphological size of the candidate defect region within the TGV through-hole, microcracks in the hole wall, foreign objects inside the hole, residue at the bottom of the hole, internal bubbles, or through-hole blockage are identified, and the defect type, 3D coordinate range, and morphological size of the corresponding internal defect are output.

[0066] After obtaining the 3D topological model of the TGV through-hole, spatial matching is performed between the TGV through-hole 3D topological model and the standard TGV 3D topological template to identify abnormal structures inside the through-hole. Specifically, firstly, the spatial reference of the TGV through-hole 3D topological model and the standard TGV 3D topological template is unified. The center position of the orifice is calculated based on the set of boundary points corresponding to the two models, and the orifice centers of the two models are made to coincide through translation transformation. Next, the axial point sets of the orifice walls of the two models are extracted, and the axial direction of the through-hole is determined using principal component analysis. The principal axis directions of the two models are rotated and calibrated to make the 3D topological model of the TGV through-hole to be detected consistent with the through-hole axis direction of the standard template. Simultaneously, the scale factor is calculated based on the corresponding axial lengths of the two models, and the axial scale of the model to be detected is normalized to ensure that both are at the same spatial scale. The spatial scale transformation relationship is expressed as follows: ,in, This represents the coordinates of the points in the spatially aligned topology model to be detected. R represents the coordinates of the original topology model points to be tested; T represents the axis alignment rotation matrix; S represents the orifice center alignment translation vector; and S represents the axial scale correction coefficient.

[0067] After spatial alignment, the point-by-point deviation is calculated between the corresponding spatial positions of the 3D topology model of the TGV via to be inspected and the standard TGV 3D topology template. Specifically, the relationship between corresponding points of the two 3D models is established according to the nearest neighbor matching method, and the Euclidean distance deviation between corresponding points is calculated: ,in, This represents the local spatial deviation of the point corresponding to the i-th spatial location; This represents the coordinates of the i-th point in the 3D topological model of the TGV via to be inspected; This indicates the coordinates of the corresponding point in the standard TGV 3D topology template.

[0068] Furthermore, continuous region analysis was performed on the local spatial deviations of all corresponding points. Regions with deviations greater than 0.5 μm and continuously distributed along the through-hole axis for more than three adjacent sections were identified as candidate defect regions. For each candidate defect region, its spatial location, connectivity, and three-dimensional morphology parameters were extracted. The spatial location was determined by the point cloud coordinate range of the defect region, the connectivity was determined by the spatial distance between adjacent points, and the morphology dimensions were calculated using the maximum length, maximum width, and volume of the defect region. Specifically, the three-dimensional dimensional parameters of the candidate defect region are expressed as follows: , , ,in, Indicates the length of the defect area; Indicates the width of the defect area; Indicates the axial height of the defect area; , , , , , These represent the boundary coordinates of the point cloud of the defect region in three spatial directions.

[0069] Furthermore, the defect type is determined based on the spatial location, connectivity, extension direction, and three-dimensional dimensional characteristics of the candidate defect region within the TGV via. When the candidate defect region is elongated and continuous along the via wall, and the angle between its length and the axial or circumferential direction of the via is less than 15°, it is determined to be a microcrack in the via wall. When the candidate defect region is located inside the cavity and discontinuously connected to the via wall, and is distributed in an independent blocky shape, it is determined to be a foreign object inside the via based on its volume size. When the abnormal region is concentrated at the bottom of the via and its axial height is less than 5 μm, it is determined to be a residue at the bottom of the via. When the abnormal region is approximately spherically distributed and has continuous internal curvature, it is determined to be an internal bubble. When the candidate defect region is distributed throughout the via along the axial direction and causes a continuous decrease in the effective cross-sectional area of ​​the via, it is determined to be a via blockage.

[0070] After the defect type identification is completed, the three-dimensional coordinate range of the defect is generated according to the point cloud coordinates corresponding to the candidate defect area, and the defect type, spatial location and shape size of the corresponding internal defect are output, so as to realize the three-dimensional positioning and quantitative characterization of internal defects of TGV through holes.

[0071] In summary, the embodiments of this application have at least the following technical effects:

[0072] First, a composite coherent modulated light field is formed by coupling a parallel coherent backlight field with a zero-degree annular uniform light field. This composite coherent modulated light field is then used to perform layered imaging of the TGV via under test, obtaining the original tomographic image sequence. Next, based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the via, phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence. Then, the standardized tomographic image sequence is segmented according to spatial regions, and the resulting tomographic image segments are distributed to multiple computing units. Parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV via structure to obtain multiple partitioned 3D contour point clouds. Further, coordinate calibration and point cloud fusion are performed on the multiple partitioned 3D contour point clouds, and topological correction is performed based on the taper characteristics and inner wall morphology characteristics of the TGV via to obtain a 3D topological model of the TGV via. Finally, the 3D topological model of the TGV via is spatially matched with a standard TGV 3D topological template. Based on the spatial matching results, internal defects of the TGV via are identified, and the defect type, spatial location, and morphological size of the internal defects are output. This invention solves the technical problem of low accuracy in three-dimensional detection of internal defects in TGV through-holes in existing technologies, and achieves the technical effect of improving the accuracy of three-dimensional detection of internal defects in TGV through-holes.

[0073] Example 2, based on the same inventive concept as the multi-field parallel tomographic detection method for internal defects of TGV through-holes in the previous examples, such as... Figure 2 As shown, this application provides a multi-field parallel tomographic detection system for internal defects in TGV through-holes, wherein the system includes:

[0074] Layered imaging module 11: Couples a parallel coherent backlight field with a zero-degree annular uniform light field to form a composite coherent modulated light field, and uses this composite coherent modulated light field to perform layered imaging of the TGV via under test, obtaining the original tomographic image sequence; Image processing module 12: Based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the via, performs phase compensation and defect pixel enhancement on the original tomographic image sequence to obtain a standardized tomographic image sequence; Partitioned point cloud generation module 13: Divides the standardized tomographic image sequence into segments according to spatial regions, and distributes the resulting tomographic image segments to multiple Each computing unit uses an optical imaging model adapted to the TGV through-hole structure for parallel iterative reconstruction to obtain multiple partitioned three-dimensional contour point clouds; Topology correction module 14: performs coordinate calibration and point cloud fusion on the multiple partitioned three-dimensional contour point clouds, and performs topology correction based on the taper characteristics and inner wall morphology characteristics of the TGV through-hole to obtain a TGV through-hole three-dimensional topology model; Defect identification module 15: performs spatial matching between the TGV through-hole three-dimensional topology model and a standard TGV three-dimensional topology template, identifies internal defects of the TGV through-hole based on the spatial matching results, and outputs the defect type, spatial location, and morphological size of the internal defects.

[0075] Furthermore, the layered imaging module 11 is used to perform the following methods:

[0076] The parallel coherent backlight field is controlled to transmit through the glass substrate and the interior of the TGV via to be detected along the axial direction, forming a transmitted light field characterizing the internal structure of the via. The zero-degree annular uniform light field is controlled to be uniformly projected along the circumference of the TGV via to be detected, forming an annular illumination light field characterizing the via wall profile and internal scattering characteristics. The parallel coherent backlight field and the zero-degree annular uniform light field are synchronously modulated according to a unified acquisition clock, so that the transmitted light field and the annular illumination light field work together at each tomographic acquisition position to form a composite coherent modulated light field.

[0077] Furthermore, the image processing module 12 is used to perform the following methods:

[0078] The refractive index, medium dispersion parameters, glass propagation thickness, and imaging wavelength of the glass substrate corresponding to the TGV via to be detected are obtained. Based on the refractive index, medium dispersion parameters, glass propagation thickness, and imaging wavelength, the medium phase offset corresponding to each tomographic acquisition position is calculated. A glass medium phase compensation operator is constructed based on the medium phase offset. The phase component in each original tomographic image is reverse-corrected using the glass medium phase compensation operator. The consistency of the reverse correction result is constrained according to the phase continuity relationship between adjacent tomographic acquisition positions to obtain a phase-compensated tomographic image sequence. Local gray-level abrupt change features, edge gradient features, and texture discontinuity features are extracted from the phase-compensated tomographic image sequence. Candidate defect pixels are screened based on the continuity of response at the same spatial location in adjacent tomographic images. The defect response weight is determined based on the gray-level deviation, edge response intensity, and cross-layer response continuity of the candidate defect pixels. The candidate defect pixels are weighted and enhanced according to the defect response weight. The weighted enhanced tomographic images are then subjected to gray-level normalization and image size unification processing to obtain a standardized tomographic image sequence.

[0079] Furthermore, the partitioned point cloud generation module 13 is used to perform the following method:

[0080] Based on the spatial location of each tomographic image slice, spatial coordinate identifiers and adjacent overlapping areas are set for each tomographic image slice, and each tomographic image slice is assigned to a corresponding computing unit. Each computing unit uses the optical imaging model adapted to the TGV through-hole structure to perform forward imaging on the initialized three-dimensional voxel distribution, calculates the imaging residual between the forward imaging result and the corresponding tomographic image slice, and iteratively corrects the three-dimensional voxel distribution based on the imaging residual. When the imaging residual meets the preset iteration termination condition, the through-hole contour points are extracted from the corrected three-dimensional voxel distribution to obtain the corresponding partitioned three-dimensional contour point cloud.

[0081] Furthermore, the partitioned point cloud generation module 13 is used to perform the following method:

[0082] A three-dimensional voxel mesh is established based on the axial tomographic position and aspect ratio of the TGV via to be detected, and the spatial projection relationship between the three-dimensional voxel mesh and each tomographic image segment is determined according to the axial extension direction of the via. Based on the light transmission characteristics of the TGV glass substrate, the imaging wavelength, and the diffraction resolution of the imaging system, the light propagation attenuation parameters and diffraction diffusion parameters corresponding to each axial tomographic position are determined. These light propagation attenuation parameters and diffraction diffusion parameters are loaded onto the spatial projection relationship to construct a forward imaging operator for mapping the three-dimensional voxel distribution to a tomographic image. Through-hole structural constraints are established based on the continuity of the TGV via axis, the continuity of the via wall profile, and the axial variation characteristics of the via diameter. These through-hole structural constraints are combined with the forward imaging operator to obtain the optical imaging model adapted to the TGV via structure.

[0083] Furthermore, the topology correction module 14 is used to perform the following method:

[0084] Based on the spatial coordinate identifier, the three-dimensional contour point clouds of each partition are transformed to a unified three-dimensional coordinate system. Corresponding feature points of the three-dimensional contour point clouds of adjacent partitions are extracted in the adjacent overlapping area, and the point cloud coordinate transformation parameters are determined based on the positional deviation between the corresponding feature points. The point cloud coordinate transformation parameters are used to register the three-dimensional contour point clouds of adjacent partitions, and the registered overlapping point clouds are fused to obtain a complete through-hole contour point cloud. Multiple hole wall sections are extracted from the complete through-hole contour point cloud along the axial direction of the TGV through-hole. The taper deviation and inner wall morphology deviation are determined based on the center position and equivalent hole diameter change of each hole wall section, and the complete through-hole contour point cloud is corrected based on the taper deviation and inner wall morphology deviation to obtain a three-dimensional topological model of the TGV through-hole.

[0085] Furthermore, the defect identification module 15 is used to perform the following method:

[0086] The TGV through-hole 3D topology model is aligned with the standard TGV 3D topology template in terms of orifice center alignment, through-hole axis alignment, and axial dimension alignment. The local spatial deviation between the aligned TGV through-hole 3D topology model and the standard TGV 3D topology template is calculated, and the region where the local spatial deviation continuously exceeds the preset defect deviation threshold is identified as a candidate defect region. Based on the spatial position, connectivity, extension direction, and morphological size of the candidate defect region within the TGV through-hole, microcracks in the hole wall, foreign objects inside the hole, residue at the bottom of the hole, internal bubbles, or through-hole blockage are identified, and the defect type, 3D coordinate range, and morphological size of the corresponding internal defect are output.

[0087] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-field parallel tomographic method for detecting internal defects in TGV through-holes, characterized in that, The method includes: A composite coherent modulation light field is formed by coupling a parallel coherent backlight field with a zero-degree annular uniform light field. The composite coherent modulation light field is then used to perform layered imaging of the TGV aperture to be detected, thereby obtaining the original tomographic image sequence. Based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the via, phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a standardized tomographic image sequence. The standardized tomographic image sequence is segmented according to spatial regions, and the resulting tomographic image segments are allocated to multiple computing units. Parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV through-hole structure to obtain multiple partitioned three-dimensional contour point clouds. Coordinate calibration and point cloud fusion are performed on the three-dimensional contour point clouds of the multiple partitions, and topology correction is performed based on the taper characteristics and inner wall morphology characteristics of the TGV through hole to obtain the three-dimensional topology model of the TGV through hole. The TGV through-hole 3D topology model is spatially matched with the standard TGV 3D topology template. Based on the spatial matching result, internal defects of the TGV through-hole are identified, and the defect type, spatial location, and shape size of the internal defects are output.

2. The multi-field parallel tomographic detection method for internal defects in TGV through-holes as described in claim 1, characterized in that, A composite coherent modulated optical field is formed by coupling a parallel coherent backlight field with a zero-degree annular uniform light field, including: The parallel coherent backlight field is controlled to transmit through the glass substrate and the interior of the TGV through hole along the axial direction of the through hole to form a transmitted light field characterizing the internal structure of the through hole. The zero-degree annular uniform light field is controlled to be uniformly projected along the circumference of the TGV through hole to be detected, forming an annular illumination light field that characterizes the hole wall profile and the scattering characteristics inside the hole. The parallel coherent backlight field and the zero-degree annular uniform light field are synchronously modulated according to a unified acquisition clock, so that the transmitted light field and the annular illumination light field work together at each tomographic acquisition position to form a composite coherent modulated light field.

3. The multi-field parallel tomographic detection method for internal defects in TGV through-holes according to claim 1, characterized in that, Phase compensation and defect pixel enhancement are performed on the original tomographic image sequence to obtain a normalized tomographic image sequence, including: The refractive index, medium dispersion parameters, glass propagation thickness, and imaging wavelength of the glass substrate corresponding to the TGV through-hole to be detected are obtained. Based on the refractive index, medium dispersion parameters, glass propagation thickness, and imaging wavelength, the medium phase offset corresponding to each tomographic acquisition position is calculated. A glass medium phase compensation operator is constructed based on the medium phase offset. The phase components in each original tomographic image are reverse-corrected using the glass medium phase compensation operator. The consistency of the reverse correction results is constrained according to the phase continuity relationship between adjacent tomographic acquisition positions to obtain a phase-compensated tomographic image sequence. Local gray-level abrupt features, edge gradient features, and texture discontinuity features are extracted from the phase-compensated tomographic image sequence, and candidate defective pixels are screened based on the continuity of the response at the same spatial location in adjacent tomographic images. The defect response weights are determined based on the grayscale deviation, edge response intensity, and cross-layer response continuity of the candidate defect pixels. The candidate defect pixels are then weighted and enhanced according to the defect response weights. Finally, the grayscale normalization and image size unification are performed on each tomographic image after weighted enhancement to obtain a standardized tomographic image sequence.

4. The multi-field parallel tomographic detection method for internal defects in TGV through-holes according to claim 1, characterized in that, Parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV through-aperture structure, including: Based on the spatial location of each tomographic image slice, spatial coordinate identifiers and adjacent overlapping areas are set for each tomographic image slice, and each tomographic image slice is assigned to the corresponding computing unit. Each computing unit uses the optical imaging model adapted to the TGV through-hole structure to perform forward imaging on the initialized three-dimensional voxel distribution, calculates the imaging residual between the forward imaging result and the corresponding tomographic image slice, and iteratively corrects the three-dimensional voxel distribution based on the imaging residual. When the imaging residual meets the preset iteration termination condition, the through hole contour points are extracted from the corrected three-dimensional voxel distribution to obtain the corresponding partitioned three-dimensional contour point cloud.

5. The multi-field parallel tomographic detection method for internal defects in TGV through-holes according to claim 4, characterized in that, The construction of the optical imaging model adapted to the TGV through-hole structure includes: A three-dimensional voxel grid is established based on the axial tomographic position and aspect ratio of the TGV through-hole to be detected, and the spatial projection relationship between the three-dimensional voxel grid and each tomographic image slice is determined based on the axial extension direction of the through-hole. Based on the light transmission characteristics of the TGV glass substrate, the imaging light wavelength, and the diffraction resolution of the imaging system, the light propagation attenuation parameters and diffraction diffusion parameters corresponding to each axial tomographic position are determined. The light propagation attenuation parameter and diffraction diffusion parameter are loaded into the spatial projection relationship to construct a forward imaging operator for mapping a three-dimensional voxel distribution into a tomographic image; Based on the continuity of the hole axis, the continuity of the hole wall profile, and the axial variation characteristics of the hole diameter of the TGV through hole, a through hole structure constraint is established. The through hole structure constraint is combined with the forward imaging operator to obtain the optical imaging model adapted to the TGV through hole structure.

6. The multi-field parallel tomographic detection method for internal defects in TGV through-holes according to claim 4, characterized in that, Coordinate calibration and point cloud fusion are performed on the multiple partitioned 3D contour point clouds, and topology correction is performed based on the taper characteristics and inner wall morphology characteristics of the TGV through-hole to obtain a 3D topological model of the TGV through-hole, including: Based on the spatial coordinate identifier, the three-dimensional contour point cloud of each partition is transformed to a unified three-dimensional coordinate system. The corresponding feature points of the three-dimensional contour point clouds of adjacent partitions in the adjacent overlapping area are extracted, and the point cloud coordinate transformation parameters are determined based on the positional deviation between the corresponding feature points. The point cloud coordinate transformation parameters are used to register the three-dimensional contour point clouds of adjacent partitions, and the registered overlapping point clouds are fused to obtain the complete through hole contour point cloud. Multiple hole wall sections are extracted from the complete through-hole contour point cloud along the axial direction of the TGV through-hole. The taper deviation and inner wall morphology deviation are determined based on the center position of each hole wall section and the equivalent hole diameter change. The complete through-hole contour point cloud is then corrected based on the taper deviation and inner wall morphology deviation to obtain a three-dimensional topological model of the TGV through-hole.

7. The multi-field parallel tomographic detection method for internal defects in TGV through-holes according to claim 1, characterized in that, The TGV via 3D topology model is spatially matched with a standard TGV 3D topology template. Based on the spatial matching results, internal defects of the TGV via are identified, including: The TGV through-hole 3D topology model and the standard TGV 3D topology template are aligned in terms of orifice center, through-hole axis, and axial dimension. Calculate the local spatial deviation between the aligned TGV via 3D topology model and the standard TGV 3D topology template, and determine the region where the local spatial deviation continuously exceeds the preset defect deviation threshold as a candidate defect region. Based on the spatial location, connectivity, extension direction, and morphological dimensions of the candidate defect regions within the TGV via, the system identifies microcracks in the via wall, foreign objects within the via, residues at the bottom of the via, internal bubbles, or via blockage, and outputs the corresponding defect type, three-dimensional coordinate range, and morphological dimensions.

8. A multi-field parallel tomographic detection system for internal defects in TGV through-holes, characterized in that, The system for implementing the multi-field parallel tomographic detection method for internal defects in TGV through-holes according to any one of claims 1-7 comprises: Layered imaging module: The parallel coherent backlight field and the zero-degree ring uniform light field are coupled to form a composite coherent modulated light field, and the composite coherent modulated light field is used to perform layered imaging of the TGV aperture to be detected to obtain the original tomographic image sequence. Image processing module: Based on the dielectric dispersion characteristics of the TGV glass substrate and the light propagation characteristics inside the through-hole, the original tomographic image sequence is subjected to phase compensation and defect pixel enhancement to obtain a standardized tomographic image sequence; Partitioned point cloud generation module: The standardized tomographic image sequence is segmented according to spatial regions, and the resulting tomographic image segments are distributed to multiple computing units. Parallel iterative reconstruction is performed using an optical imaging model adapted to the TGV through-hole structure to obtain multiple partitioned three-dimensional contour point clouds. Topology correction module: performs coordinate calibration and point cloud fusion on the multiple partitioned 3D contour point clouds, and performs topology correction based on the taper characteristics and inner wall morphology characteristics of the TGV through hole to obtain the 3D topology model of the TGV through hole; Defect identification module: Spatially matches the TGV through-hole 3D topology model with the standard TGV 3D topology template, identifies internal defects of the TGV through-hole based on the spatial matching results, and outputs the defect type, spatial location, and shape size of the internal defects.

Images