Laser hole processing method and system for printed circuit board

Abstract

The invention relates to the field of printed circuit board processing, and provides a laser hole processing method and system for a printed circuit board, and the method comprises the steps: firstly carrying out the laser browning processing of a thin core board with the thickness of 0.076 mm, then processing a laser hole with the diameter of 0.1 mm in the thin core board through a laser direct hole forming method, and then filling and leveling up the laser hole, the thickness of the thin core plate is increased to the required thickness of 0.203 mm through a pressing process to obtain a thick core plate, and finally, a laser hole with the diameter of 0.1 mm is machined in the thick core plate subjected to laser browning treatment through a laser direct hole forming method. Therefore, the quality bottlenecks of hole wall roughness, hole filling defects and the like in the traditional multi-time pressing process can be effectively broken through, the interlayer conduction reliability can be effectively guaranteed, and a reliable technical path can be provided for industrialization of HDI products of any layer.

Description

Technical Field

[0001] The present application relates to the field of printed circuit board processing, and more specifically, to a laser hole processing method and system for printed circuit boards. Background Art

[0002] As electronic products develop rapidly towards miniaturization and multifunctionality, high-density interconnection (HDI) technology has become the core development direction in the field of printed circuit board manufacturing. In particular, arbitrary-layer HDI technology can achieve full-layer interconnection through laser drilling, which can significantly increase line density and reduce substrate thickness, providing key support for cutting-edge fields such as 5G communication equipment, wearable electronics and high-performance chip packaging.

[0003] However, in the processing of ultra-thin core boards (such as 50-micron-level thickness), traditional mechanical drilling technology is prone to micro-cracks in the substrate and interlayer offset due to physical stress concentration, while conventional laser hole-forming processes are prone to increased hole wall roughness and incomplete metallization filling after multiple lamination and thickening. What is more noteworthy is that when the core board undergoes multiple lamination and thickening, the original microporous structure is easily affected by the hot pressing process and deformed, resulting in reduced interlayer conduction reliability. This defect is particularly prominent in high-frequency and high-speed circuit scenarios that require signal integrity. In addition, the copper layer processing process on the surface of the ultra-thin core board directly affects the laser energy absorption efficiency. Improper pretreatment will lead to systemic quality risks such as poor consistency of hole diameter and hole position accuracy offset, which seriously restricts the yield improvement and industrialization process of any layer HDI products.

[0004] Therefore, an optimized laser hole processing solution for printed circuit boards is desired. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present application provides a laser hole processing method and system for printed circuit boards.

[0006] According to one aspect of the present application, a laser hole processing method for a printed circuit board is provided, which includes: providing a thin core board with a thickness of 0.076 mm;

[0007] After the thin core board is laser browned, a laser direct hole forming method is used to process a laser hole with a diameter of 0.1 mm on the thin core board, including: acquiring an image of the thin core board, and determining the target offset data based on feature semantic comparison analysis between the thin core board image and the target layout template image; adjusting the laser processing parameters based on the target offset data; filling the laser hole, and thickening the thin core board to the required thickness of 0.203 mm through a lamination process to obtain a thick core board; after the thick core board is laser browned, a laser direct hole forming method is used to process a laser hole with a diameter of 0.1 mm on the thick core board.

[0008] According to another aspect of the present application, a laser hole processing system for a printed circuit board is provided, which includes: a target offset data determination module, which is used to obtain a thin core board image and determine the target offset data based on feature semantic comparison analysis between the thin core board image and the target layout template image;

[0009] The processing parameter adjustment module is used to adjust the laser processing parameters based on the target offset data.

[0010] This application has significant technical effects due to the adoption of the above technical solutions:

[0011] The laser hole processing method and system for printed circuit boards provided in the present application firstly performs laser browning treatment on a thin core board with a thickness of 0.076mm, then uses a laser direct hole forming method to process a laser hole with a diameter of 0.1mm on the thin core board, then fills the laser hole, and thickens the thin core board to the required thickness of 0.203mm through a lamination process to obtain a thick core board, and finally uses a laser direct hole forming method to process a laser hole with a diameter of 0.1mm on the thick core board after laser browning treatment. In this way, it can effectively break through the quality bottlenecks such as hole wall roughness and hole filling defects in the traditional multiple lamination process, and can effectively guarantee the reliability of interlayer conduction, and can provide a reliable technical path for the industrialization of any layer of HDI products. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 The figure is a flow chart of a laser hole processing method for a printed circuit board according to an embodiment of the present application.

[0014] Figure 2 Flow chart of step S2 in the laser hole processing method for a printed circuit board according to an embodiment of the present application.

[0015] Figure 3 Flow chart of step S21 in the laser hole processing method for a printed circuit board according to an embodiment of the present application.

[0016] Figure 4 Flow chart of step S213 in the laser hole processing method for a printed circuit board according to an embodiment of the present application. Figure 5Flow chart of step S2132 in the laser hole processing method for a printed circuit board according to an embodiment of the present application.

[0017] Figure 6 4 is a system block diagram of a laser hole processing system for a printed circuit board according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0019] With the rapid development of electronic products towards miniaturization and multifunctionality, high-density interconnection (HDI) technology has become the core direction of printed circuit board manufacturing. Among them, arbitrary-layer HDI technology realizes full-layer interconnection through laser drilling, significantly improves line density and reduces substrate thickness, providing key support for 5G communication equipment, wearable electronics, and high-performance chip packaging.

[0020] However, when processing ultra-thin core boards (such as 50-micron-thick), traditional mechanical drilling can easily cause micro-cracks in the substrate and interlayer offsets, while conventional laser hole-forming processes may cause rough hole walls and incomplete metallization filling after multiple lamination thickening. In addition, during multiple lamination thickening processes, the microporous structure is prone to reduce interlayer conduction reliability due to thermal compression deformation, especially in high-frequency and high-speed circuits, which has a significant impact on signal integrity. At the same time, the pretreatment process of the copper layer on the surface of the core board directly affects the laser energy absorption efficiency. Improper treatment will cause problems such as poor hole diameter consistency and hole position accuracy offset, which seriously restricts the yield improvement and industrialization process of any layer HDI products.

[0021] Based on this, the present application proposes a laser hole processing method for a printed circuit board. Figure 1 FIG. 1 is a flow chart of a laser hole processing method for a printed circuit board according to an embodiment of the present application. Figure 1 As shown, the laser hole processing method for a printed circuit board according to an embodiment of the present application includes: S1, providing a thin core board with a thickness of 0.076 mm; S2, after the thin core board is laser browned, a laser direct hole forming method is used to process a laser hole with a diameter of 0.1 mm on the thin core board; S3, filling the laser hole, and thickening the thin core board to the required thickness of 0.203 mm through a lamination process to obtain a thick core board; S4, after the thick core board is laser browned, a laser direct hole forming method is used to process a laser hole with a diameter of 0.1 mm on the thick core board.

[0022] In step S1, a thin core board with a thickness of 0.076 mm is provided. It should be understood that for ultra-thin core boards, traditional mechanical drilling can easily cause microcracks in the substrate and interlayer offset due to physical stress concentration. However, the 0.076 mm thin core board is relatively moderate in thickness. The use of laser direct hole forming can effectively avoid the stress problem caused by mechanical drilling, avoid the generation of microcracks and interlayer offset from the source, and help to ensure the structural integrity of the core board.

[0023] In step S2, after the thin core board is subjected to laser browning treatment, a laser direct hole forming method is used to process a laser laser hole with a diameter of 0.1 mm on the thin core board. It should be understood that the condition of the copper layer on the surface of the thin core board has a significant effect on the absorption efficiency of the laser energy. The untreated copper layer surface may have problems such as oxidation and inconsistent roughness, which will cause the laser energy to be unevenly distributed on the surface of the copper layer. The laser browning treatment can change the microstructure and chemical properties of the copper layer surface to form a uniform browning layer with good light absorption properties. In this way, in the subsequent laser hole forming process, the laser energy can be more effectively absorbed by the copper layer, providing better conditions for precise hole forming. Accordingly, laser direct hole forming can use the high energy density of the laser to instantly melt and vaporize the material, which is a non-contact processing, does not generate physical stress, can effectively avoid damage to the thin core board, and improve the product yield. Therefore, in this application, a laser direct hole forming method is used to process a laser laser hole with a diameter of 0.1 mm.

[0024] In particular, in the laser direct hole forming process, laser browning treatment forms a roughened surface through chemical etching and oxidation reaction. Although it improves the copper layer's absorption efficiency of laser energy, this process will cause local microscopic shrinkage of the substrate or reorganization of the surface copper crystal structure, resulting in a micron-level offset between the preset drilling coordinates and the actual physical structure of the material. In addition, the thin core board (0.076mm) will produce anisotropic thermal expansion at instantaneous high temperature due to differences in heat conduction rates during laser irradiation. If the laser focus position is not dynamically corrected, the actual hole forming position will deviate from the design coordinates. Through target offset adjustment, the relative position of the laser spot and the core board microstructure can be calibrated in real time to ensure the hole forming accuracy of 0.1mm aperture, while avoiding the problem of interlayer conduction misalignment caused by cumulative errors. This plays a key role in the precise alignment of multi-layer structures in subsequent thickening and pressing processes.

[0025] The existing patent CN118175737A proposes a method for laser missed connection during laser hole processing of a printed circuit board, which first determines the target layout of the circuit board and identifies the target position, then collects the surface image of the circuit board and preprocesses it, then analyzes the preprocessed image to accurately identify the target position, and compares the identified position with the expected position. After comparison, the calibration result is mapped to the laser device coordinate system, and the laser parameters are set according to the mapped position to perform laser hole processing. Finally, after the processing is completed, the actual target position is compared with the expected position again, and the laser parameters are adjusted accordingly. However, the patented method compares the identified position with the expected position, and then performs coordinate mapping and calibration through linear transformation. This linear mapping has a certain effect on the global and linear deformation of the PCB board, but has limited compensation ability for local deformation or nonlinear deformation. In addition, although the patented method also uses image processing and automatic calculation, its offset calculation and calibration strategy are relatively simple, and it is necessary to manually pre-set or adjust parameters such as mapping coefficients, and the degree of intelligence and adaptability may be low.

[0026] Based on this, the technical concept of the present application is to use computer vision image processing and offset calculation algorithms to map the real-time collected thin core board image and the target layout template image in the background database to high-dimensional feature space respectively, perform deep feature extraction, and generate target real-time distribution feature maps representing the actual processing scene and target standard distribution feature maps under standard process conditions respectively. Afterwards, feature phase alignment and cross-domain correlation analysis are performed on the two types of feature maps, and the nonlinear deformation laws caused by microscopic shrinkage, crystal reorganization and thermal expansion caused by laser browning treatment are captured by adaptively matching feature phase differences, and then the semantic-level perception feature maps of target position offset are generated by fusion. Finally, the target offset data is parsed through feature decoding to drive dynamic compensation of laser focus position and energy density. This feature processing mechanism breaks through the limitations of traditional linear mapping through deep learning models, accurately identifies local nonlinear deformations of ultra-thin core boards caused by material modification and thermal stress accumulation during laser processing, and at the same time reduces the need for manual intervention and enhances process adaptability. Specifically, Figure 2 FIG. 1 is a flow chart of step S2 in the laser hole processing method for a printed circuit board according to an embodiment of the present application. Figure 2 As shown, step S2 includes: S21, acquiring a thin core board image, and determining target offset data based on feature semantic comparison analysis between the thin core board image and the target layout template image; S22, adjusting laser processing parameters based on the target offset data.

[0027] In step S21, a thin core board image is acquired, and target offset data is determined based on feature semantic comparison analysis between the thin core board image and the target layout template image. Specifically, Figure 3FIG. 1 is a flow chart of step S21 in the laser hole processing method for a printed circuit board according to an embodiment of the present application. Figure 3 As shown, step S21 includes: S211, acquiring a thin core board image captured by a camera; S212, extracting a target layout template image from a background database; S213, determining target offset data based on feature semantic comparison analysis between the thin core board image and the target layout template image.

[0028] In step S211, the image of the thin core plate captured by the camera is obtained. It should be understood that the image of the thin core plate directly records the position information, geometric shape information, etc. of the target on the thin core plate during the actual processing process. These targets are key identifiers for positioning in laser hole processing. Their distribution and specific coordinate positions in the image are the key basis for determining whether the target is offset and the specific offset data.

[0029] In step S212, the target layout template image is extracted from the background database. It should be understood that the target layout template image includes where the target should be located and the relative position relationship between them under ideal conditions (i.e., without any deformation or offset). In other words, the target layout template image provides an ideal target position and layout as a reference. By comparing and analyzing the thin core board image actually collected with the template image, it is possible to detect whether the target has been offset and to specifically analyze the actual offset data.

[0030] In step S213, the target offset data is determined based on the feature semantic comparison analysis between the thin core board image and the target layout template image. Specifically, Figure 4 FIG. 1 is a flow chart of step S213 in the laser hole processing method for a printed circuit board according to an embodiment of the present application. Figure 4 As shown, step S213 includes: S2131, performing target position feature extraction on the thin core board image and the target layout template image to obtain a target real-time distribution feature map and a target standard distribution feature map; S2132, performing target position feature adaptive phase alignment perception on the target real-time distribution feature map and the target standard distribution feature map to obtain a target position offset semantic level perception feature map; S2133, performing feature decoding on the target position offset semantic level perception feature map to obtain fixed target offset data.

[0031] In step S2131, the target position feature extraction is performed on the thin core plate image and the target layout template image to obtain the target real-time distribution feature map and the target standard distribution feature map. Specifically, in an embodiment of the present application, step S2131 includes: inputting the thin core plate image and the target layout template image into the target position sensitive twin detection network to obtain the target real-time distribution feature map and the target standard distribution feature map. Accordingly, considering the micro-contraction and crystal reorganization of the substrate caused by the laser browning treatment, as well as the anisotropic thermal expansion effect during the processing, there is a dynamically changing nonlinear deformation between the preset target coordinates and the actual physical structure of the material. The traditional linear mapping method based on the geometric model (such as the existing patent CN118175737A) can only compensate for the global uniform deformation, but cannot effectively identify the micro-depression formed by the local grain reconstruction of the copper layer or the regional distortion caused by the accumulation of thermal stress. This limitation stems from the shallow analysis of the correlation between the feature layers by conventional image processing technology, and it is difficult to extract the implicit spatial mapping relationship between the deformed target image and the standard template. Based on this, the present application obtains the target real-time distribution feature map and the target standard distribution feature map by inputting the thin core plate image and the target layout template image into the target position sensitive twin detection network. That is to say, the present application synchronously maps the real-time collected thin core plate image and the standard template image to the high-dimensional feature space through the target position sensitive twin detection network, and its core lies in constructing a dual-channel feature extraction mechanism. The network captures the surface topological changes of the copper layer caused by laser browning (such as micro-area wrinkles formed by grain boundary migration) from the deformed real-time image through a dual-branch structure with shared weights, and extracts the geometric constraint features under the ideal state from the standard template. This twin architecture can effectively decouple the inherent characteristics of the substrate from the interference of processing deformation, and strengthen the network's characterization ability of nonlinear distortions such as micro-contraction and thermal expansion through comparative learning, so that the real-time distribution feature map contains not only the explicit geometric information of the target, but also the evolution law of the microstructure after material modification. In this way, the submicron-level offset caused by the reorganization of the copper layer crystal can be encoded into a resolvable feature vector while retaining the macroscopic layout of the target.

[0032] In step S2132, the target real-time distribution feature map and the target standard distribution feature map are subjected to target position feature adaptive phase alignment perception to obtain a target position offset semantic level perception feature map. Specifically, Figure 5 FIG. 1 is a flow chart of step S2132 in the laser hole processing method for a printed circuit board according to an embodiment of the present application. Figure 5As shown, step S2132 includes: S2132-1, performing feature decoupling and phase alignment on the target real-time distribution feature map and the target standard distribution feature map to obtain a set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs; S2132-2, performing target position feature joint attention perception on each phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pair in the set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs to obtain a set of target distribution feature joint perception attention weights; S2132-3, based on the set of target distribution feature joint perception attention weights, performing attention-driven significant aggregation on the set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs to obtain a target position offset semantic-level perception feature map. It should be understood that the micro-contraction and crystal reorganization of the copper layer caused by laser browning treatment, and the anisotropic thermal expansion effect caused by superimposed laser irradiation, make the target position offset present complex nonlinear deformation characteristics. When the prior art uses linear mapping calibration, it cannot analyze the local depression formed by the reconstruction of the copper grain boundary or the regional distortion induced by the thermal stress gradient, resulting in the laser focus compensation can only correct the global translation or rotation error, but cannot adapt to the dynamic changes of the micro deformation. This limitation stems from the shallow modeling of the intrinsic correlation of the feature space by the traditional method, ignoring the coupling effect of the plastic deformation and elastic thermal expansion of the micro-topological structure of the copper layer during the material modification process. Therefore, the present application performs adaptive phase alignment perception of the target position feature on the target real-time distribution feature map and the target standard distribution feature map to obtain the semantic level perception feature map of the target position offset. Specifically, the two types of feature maps are first decoupled into a set of local feature encoding matrices, the essence of which is to decompose the macroscopic target layout into feature units at the microscopic grain scale, such as each unit corresponding to the migration trajectory of the copper grain boundary or the micro-region tensile morphology caused by thermal expansion. Through the feature phase alignment measurement, the semantic consistency of the real-time features and the standard features in the local structure is dynamically searched, such as identifying the correspondence between the grain boundary offset caused by laser browning and the ideal grain boundary distribution in the standard template. In this process, phase alignment does not rely on the rigid matching of physical coordinates, but captures the composite action law of nonlinear elastic deformation caused by thermal stress accumulation (such as local area compression or stretching) and plastic deformation caused by material modification (such as permanent displacement formed by copper crystal slip) through structural similarity judgment in feature space.

[0033] Specifically, in the embodiment of the present application, step S2132-1 includes: performing feature decoupling on the target real-time distribution feature map and the target standard distribution feature map to obtain a set of target real-time distribution local feature coding matrices and a set of target standard distribution local feature coding matrices. The process can be expressed by the formula:

[0034] Decouple(F1)={M 11 ,M 12 ,...,M 1i ,...,M 1n}

[0035] Decouple(F2)={M 21 ,M 22 ,...,M 2j ,...,M 2n}

[0036] Among them, F1 is the target real-time distribution feature map, F2 is the target standard distribution feature map, Decouple(·) is the feature decoupling operation, M 11 , M 12 , M 1i and M 1n are the first, second, i-th and n-th target real-time distribution local feature coding matrices in the set of target real-time distribution local feature coding matrices, M 21 , M 22 , M 2j and M 2n They are respectively the first, second, jth and nth target standard distribution local feature coding matrices in the set of target standard distribution local feature coding matrices;

[0037] Based on the feature phase alignment between any two target real-time distribution local feature coding matrices and target standard distribution local feature coding matrices in the set of target real-time distribution local feature coding matrices and the set of target standard distribution local feature coding matrices, the set of target real-time distribution local feature coding matrices and the set of target standard distribution local feature coding matrices are dynamically searched and aligned for feature phases to obtain a set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs. The process can be expressed by the formula:

[0038]

[0039] Where T is the transpose operation, ‖·‖ F To calculate the Frobenius norm, d P (M 1i ,M 2j ) is M 1i and M 2j The characteristic phase alignment between To return the j value corresponding to the maximum value, k is the position where the maximum approximate matching value in the set of local feature encoding matrices of the target standard distribution is found.

[0040] Accordingly, considering that the global feature map contains complex nonlinear deformations (such as grain boundary reconstruction and thermal stress gradient distortion), it is difficult to directly model local details. The existing linear calibration method only focuses on global translation or rotation errors and cannot analyze the dynamics of microscopic deformation, resulting in compensation failure. To this end, the target real-time distribution feature map and the target standard distribution feature map are decomposed into a set of target real-time distribution local feature encoding matrices and a set of target standard distribution local feature encoding matrices through feature decoupling. The core is to convert high-dimensional global features into feature units at the microscopic grain scale (such as copper grain boundary migration trajectory and micro-region tensile morphology), thereby separating the independent representation of the local structure. This not only reduces the computational complexity of high-dimensional global features, but also avoids semantic confusion problems in global features by focusing on local deformation features (such as the coupling effect of plastic deformation and elastic thermal expansion). It can provide operational and refined input for subsequent phase alignment, so that the model can effectively model the dynamic changes of the microscopic topological structure of the copper layer.

[0041] It should be understood that due to the failure of physical coordinate matching caused by nonlinear deformation (such as local compression / stretching, copper crystal slip) caused by laser irradiation, traditional rigid alignment methods cannot adapt to dynamic deformation. The present application dynamically searches for the structural similarity of local feature pairs in the feature space (such as the correspondence between grain boundary offset and standard distribution) through feature phase alignment measurement, rather than relying on fixed spatial position relationships. This strategy can capture the composite deformation law of thermal stress accumulation and material modification, such as identifying the superposition effect of local elastic deformation and permanent plastic displacement. That is, the dynamic search mechanism traverses the target real-time distribution local feature coding matrix set and the target standard distribution local feature coding matrix set, screens out the feature pairs with the highest semantic consistency, effectively solves the problems of spatial dislocation, deformation and occlusion, and generates a phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pair set, thereby establishing an interpretable mapping relationship between microscopic deformation dynamics and standard templates, overcoming the defect of insufficient modeling of feature space correlation in traditional methods.

[0042] In an embodiment of the present application, step S2132-2, each phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pair in the set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs is subjected to joint attention perception of target position features to obtain a set of joint perception attention weights of target distribution features. It should be understood that although the phase alignment has established a preliminary feature correspondence, the importance and interaction patterns (such as complementarity, competition or redundancy) of different feature pairs still need to be further modeled. For example, some grain boundary migration trajectories may be sensitive to deformation, and thermal expansion micro-regions may need to be jointly analyzed with plastic deformation features. By performing joint attention perception of target position features on the geometry of phase-aligned feature pairs, the model can adaptively learn the interaction weights of feature pairs, such as enhancing the compression-stretch feature contrast in key areas and suppressing noise interference (such as artifacts caused by thermal stress gradients). The attention mechanism not only quantifies the importance of feature pairs, but also mines their deep associations (such as the dynamic coupling law of the microscopic topology of the copper layer), thereby improving the robustness of feature expression. That is, information filtering and feature selection are achieved through weight distribution, providing a differentiated contribution basis for subsequent aggregation to enhance the model's ability to analyze complex deformations.

[0043] Specifically, in the embodiment of the present application, step S2132-2 includes: calculating the correlation matrix between each phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pair in the set of phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pairs to obtain a set of target real-time-standard distribution correlation matrices; performing trace measurement on each target real-time-standard distribution correlation matrix in the set of target real-time-standard distribution correlation matrices to obtain a set of target real-time-standard distribution trace measurement values; based on the set of target real-time-standard distribution trace measurement values, obtaining a set of target distribution feature joint perception attention weights. The above process can be expressed by the formula:

[0044]

[0045] Among them, A net (·) is the feature joint perception attention weight calculation, {M 1i ,M 2k} pair is the phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pair, Tr(·) is the trace value of the matrix, a i It is M 1i and M 2k The target real-time-standard distribution trace measurement value between , sigmoid is the normalization function, w i M1i and M 2k The target distribution features between them jointly perceive the attention weight.

[0046] Preferably, in another example of the present application, based on the set of target real-time-standard distribution trace measurement values, a set of target distribution feature joint perceptual attention weights is obtained, including: based on the set of phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pairs, the set of target real-time-standard distribution trace measurement values ​​is compactified based on the manifold of bidirectional connectivity to obtain a set of optimized target real-time-standard distribution trace measurement values. Based on the set of optimized target real-time-standard distribution trace measurement values, a set of target distribution feature joint perceptual attention weights is obtained. The process is expressed by the formula:

[0047] M a =M 1i T

[0048] M b =M 2k

[0049] α=count{d L1 (m ai ,m bi )<ε}

[0050] β=count{d L2 (m ai ,m bi )<ε}

[0051] M 1i T′ =exp[-(α-β+a i )⊙M 1i T ]

[0052] M 2k ′=exp[-(β-α+a i )⊙M 2k ]

[0053] a i ′=Tr(M 1i T′ M 2k ′)

[0054] w i =sigmoid(a i ′)

[0055] Among them, M a It is M 1i The transposed matrix, M b It is M 2k, that is, M 1i The corresponding phase-aligned target standard distribution local feature encoding matrix, m ai It is M a The eigenvalue of the i-th position in bi It is M b The eigenvalue of the i-th position in m ai ∈M a ,m bi ∈M b , d L1 (m ai ,m bi ) is used to calculate m ai and m bi The L1 distance between L2 (m ai ,m bi ) is used to calculate m ai and m bi , ε is the preset threshold, count{·} is the number of calculated conditions, α is the number of L1 distance connections, β is the number of L2 distance connections, exp is the exponential function value with the natural constant e as the base, ⊙ is the point multiplication by position, M 1i T′ It is M 1i T The optimized matrix, M 2k ′ is M 2k The optimized matrix, a i ′ is a i The optimized target real-time-standard distribution trace metric value after optimization, w i M 1u and M 2k The target distribution features between them jointly perceive the attention weight.

[0056] In particular, for any two target real-time distribution local feature coding matrices and target standard distribution local feature coding matrices, it is necessary to accurately quantify the topological correlation strength through matrix trace measurement. Specifically, when the initial feature coding matrix is ​​given, the system first counts the number of bidirectional connections based on the characteristic spectrum spacing, and identifies the topological connection areas reflecting the local deformation structure (such as grain boundary slip trajectory, thermal stress gradient distribution) by analyzing the dynamic interval threshold between adjacent eigenvalues ​​in the spectral distribution, and constructs a mapping relationship between the number of connections and the topological steady-state measure. This measure converts the geometric deformation at the grain scale (such as compression-tensile strain field, slip band dislocation density) into a resolvable numerical indicator by calculating the spectral density coupling of the two sets of matrices in the discrete shape structure representation space, thereby evaluating the structural similarity of phase-synchronized feature pairs under dynamic deformation. In order to optimize the measurement accuracy, the system introduces a manifold compactification mechanism under topological constraints, and iteratively tunes the trace metric operator through implicit space mapping: in each round of iteration, the manifold embedding parameters of the initial matrix are adjusted based on the current measurement value through back propagation, while constraining the robustness invariance of long-range correlation features (such as grain boundary network connectivity and micro-region thermal expansion synergy effects), gradually compressing redundant spectral components and enhancing the weight distribution of key deformation modes (such as plastic deformation-dominated areas). This process continues until the compaction degree of the manifold space meets the preset convergence conditions, and finally forms an optimized trace metric model that can accurately characterize the dynamic correlation of the microscopic topology of the copper layer, which can effectively support the multi-scale modeling of complex nonlinear deformation coupling laws.

[0057] Finally, based on the set of target distribution features and perceptual attention weights, the set of phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pairs is subjected to attention-driven significant aggregation to obtain the semantic-level perceptual feature map of target position offset. The above process can be expressed as:

[0058]

[0059] F f =[M′1,M′2,...,M′ i ,...,M′ n ]

[0060] Wherein, W1 and W2 are the first joint perception weight matrix and the second joint perception weight matrix, It is subtracted by position point. It is added by position point, M′ i It is M 1i and M 2k The combined target real-time-standard distribution feature joint perception matrix, that is, the i-th target real-time-standard distribution feature joint perception matrix in the set of target real-time-standard distribution feature joint perception matrices, M′1, M′2 and M′ nare the first, second and nth target real-time-standard distribution feature joint perception matrices in the set of target real-time-standard distribution feature joint perception matrices, F f It is a semantic-level perceptual feature map of target position offset. It should be understood that simple feature splicing will blur key deformation information (such as local depression or regional distortion), while significant aggregation is guided by attention weights to highlight the expression of high-contribution phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pairs. For example, higher weights are given to the distortion area induced by thermal stress gradient, making it dominant in the final feature map. That is, by focusing on important feature interaction areas (such as the coupling effect of crystal reorganization and thermal expansion), the scattered local deformation features are integrated into a globally coherent semantic-level perceptual feature map. Significant aggregation not only retains the dynamic details of microscopic deformation (such as the nonlinear offset of copper grain boundary migration trajectory), but also enhances the comprehensiveness of feature expression through information complementarity. The target position offset feature map finally generated clearly represents the composite effect of elastic / plastic deformation, which can provide high-precision dynamic deformation feedback for laser focus compensation, and helps to break through the limitations of traditional linear calibration.

[0061] In step S2133, the semantic-level perceptual feature map of the target position offset is feature decoded to obtain the target offset data. Specifically, in an embodiment of the present application, step S21313 includes: inputting the semantic-level perceptual feature map of the target position offset into a decoder-based target offset meter to obtain the target offset data. That is, the semantic-level perceptual feature map of the target position offset obtained by aligning the target real-time distribution feature map and the target standard distribution feature map is decoded to accurately obtain the target offset data. In particular, the decoder-based target offset meter can accurately measure the offset of the target in each dimension, including position, angle and other information, through precise calculation and analysis. It can take into account various subtle changes and complex relationships in the feature map, thereby providing high-precision target offset data.

[0062] The following is a detailed description of a specific implementation process of "inputting the target position offset semantic level perceptual feature map into the decoder-based target offset metric to obtain target offset data":

[0063] After obtaining the semantic-level perceptual feature map of target position offset, it is input into the target offset metric based on the decoder. The decoder is based on a neural network architecture, and its internal structure is optimized through a large amount of data training. When the semantic-level perceptual feature map of target position offset enters the decoder, it will undergo a series of complex processing processes.

[0064] The hidden layer of the neural network begins to parse the input feature map. In this process, neurons first filter and process the information in the feature map to identify key features related to target offset. As the network level deepens, features at different levels are gradually integrated. The connection weights between neurons determine the importance and combination of different features, combining low-level basic features into higher-level and more representative features that can more accurately reflect the characteristic patterns of target offset.

[0065] The nonlinear activation function in the neural network is used to perform a nonlinear transformation on the integrated features. This step is crucial because the target offset caused by factors such as laser browning and thermal expansion presents complex nonlinear features. Nonlinear transformation can enhance the neural network's ability to express these complex features, allowing the decoder to better capture the nonlinear relationship in the target offset feature map.

[0066] After the previous processing, the decoder calculates the target offset data based on the learned feature pattern. The position offset is an important part of the target offset data. The decoder determines the displacement value of the target relative to the standard position in the horizontal (x-axis) and vertical (y-axis) directions by analyzing the features. For example, if the feature map shows that the target has moved a certain pixel distance to the right in the x-axis direction, then this distance is the position offset in the x-axis direction; similarly, the corresponding offset in the y-axis direction can also be calculated. In addition to the position offset, the target may rotate during the actual processing process, and the angle offset is the parameter used to measure this change. With the target center as the rotation center and a reference direction specified (such as horizontal right is the 0° direction), the decoder calculates the angle difference between the actual target and the standard direction by analyzing the target shape, direction and other features in the feature map. This difference is the angle offset.

[0067] Finally, the decoder outputs the target offset data including position offset and angle offset. The output target offset data can provide a comprehensive and accurate basis for the subsequent laser processing parameter adjustment.

[0068] In step S22, the laser processing parameters are adjusted based on the target offset data. It should be understood that the printed circuit board has extremely high requirements for processing accuracy, and a slight offset of the target may affect the electrical performance and overall quality of the circuit board. In order to meet the requirements of high-precision processing, the laser processing parameters, such as the laser focus position and energy density, must be adjusted in real time according to the target offset data to compensate for the error caused by the target offset and ensure the accuracy of the processing. In this way, by adjusting the laser processing parameters, it is ensured that the laser can accurately process laser holes that meet the design requirements on the thin core board. Ensure that the position accuracy, aperture size and other parameters of the laser hole meet the standards, thereby improving the quality of the printed circuit board and avoiding problems such as poor line connection and electrical performance degradation caused by processing errors.

[0069] The following is a detailed description of the specific implementation process of "adjusting laser processing parameters based on target offset data": After obtaining the target offset data, it is necessary to deeply analyze the mapping adjustment relationship between the offset data and the laser processing parameters. For position offset, it is directly related to the laser focus position. When the target has an offset of Δx in the x-axis direction and an offset of Δy in the y-axis direction, the laser focusing system must make corresponding reverse adjustments based on these offset data. This is because the accuracy of the laser focus position directly determines the position where the laser energy acts on the target. If the focus position is inaccurate, the position accuracy of the laser hole cannot be guaranteed. For example, if Δx is a positive value, it means that the target is offset to the right in the x-axis direction. In order to make the laser energy accurately act on the target position, the laser focus should be moved to the left by a distance of |Δx|; similarly, when Δy is a positive value, the laser focus should be moved downward by a distance of |Δy|. Through such adjustments, it can ensure that the laser beam is accurately focused on the target position, laying a solid foundation for subsequent processing.

[0070] Angle offset also has an important impact on laser processing. It mainly changes the relative angle between the laser beam and the target, which in turn affects the shape and quality of the laser hole. When the angle offset θ is detected, the laser control system needs to adjust the laser scanning path according to this angle value. Taking the common circular laser hole processing as an example, under normal circumstances, the laser beam performs a circular scan around the center of the target to form a standard circular laser hole. However, once there is an angle offset, the laser beam can no longer scan along the conventional path, otherwise it will cause deviations in the shape of the laser hole. At this time, the laser control system will accurately calculate the corrected scanning path based on the value of θ, and compensate for the influence of the target rotation by changing the emission angle and scanning order of the laser beam, so that the final processed laser hole remains circular and accurately positioned. In addition to the influence of position offset and angle offset on the laser focus position and scanning path, they will also change the distribution of laser energy on the target. In order to ensure that the aperture size of the laser hole meets the design requirements (0.1mm in this application), the laser energy density must be adjusted accordingly. When the target is offset, if the laser energy is dispersed, it is necessary to increase the laser energy density appropriately to ensure that it can penetrate the material and form a laser hole of the specified size; on the contrary, if the energy is too concentrated, in order to avoid excessive processing resulting in an excessively large aperture, it is necessary to reduce the energy density. The specific adjustment range is not determined arbitrarily, but it is necessary to comprehensively consider the offset size and the absorption characteristics of the material to the laser energy. Usually, the relationship between the energy density adjustment coefficient k and the offset is fitted through a large amount of experimental data, that is, k = f(|Δx|,|Δy|,θ). According to this coefficient, the laser energy density E is adjusted using the adjustment formula E′ = k×E, where E′ is the adjusted laser energy density and E is the original laser energy density. Such a calculation and adjustment process can ensure that the laser energy density matches the target offset, thereby ensuring the aperture accuracy of the laser hole.

[0071] After clarifying the mapping adjustment relationship between the offset data and the laser processing parameters, the next step is to actually adjust the parameters of the laser processing equipment. The first step is to adjust the laser focusing system and send precise instructions to the laser focusing system based on the calculated position offset. Advanced laser processing equipment is equipped with a high-precision focusing system that can receive external control signals and drive the focusing lens to move precisely along the x-axis and y-axis through a motor, thereby achieving precise adjustment of the laser focus position. In this way, it is ensured that the laser beam can be accurately focused on the target position for processing, effectively avoiding processing errors caused by focus position deviation.

[0072] At the same time, the laser scanning control program needs to be modified to account for the effect of angle offset on the laser scanning path. The adjusted laser scanning path information is accurately input into the laser scanning control program, which is responsible for accurately controlling the scanning trajectory of the laser beam. It will accurately control the emission angle and position of the laser beam at different times based on the new path information, so that the laser beam scans the target strictly according to the corrected path, thereby ensuring that the shape and accuracy of the laser hole meet the design requirements.

[0073] In addition, it is also necessary to adjust the laser energy output device, and make corresponding adjustments to the laser energy output device according to the result of the energy density adjustment formula E′=k×E calculated above. The laser energy output device can generally adjust the laser energy density by changing the power supply power, pulse frequency, etc. For example, when the calculation results show that the energy density needs to be increased, the laser pulse frequency can be increased or the power supply output power can be increased; conversely, if the energy density needs to be reduced, the opposite operation is taken to ensure that the aperture of the laser hole meets the design standards.

[0074] In the entire laser processing process, in order to ensure high precision and stability of processing, real-time monitoring and feedback adjustment are indispensable links. Use sensors to continuously monitor the target position. Once it is found that the actual processing situation does not meet expectations, it is necessary to repeat the above analysis and adjustment steps to further optimize the laser processing parameters. Through such a real-time monitoring and feedback adjustment mechanism, problems that occur during the processing can be discovered and corrected in a timely manner to ensure that the entire processing process is always in the best state.

[0075] In summary, step S2 is explained clearly, which uses computer vision-based image processing and offset calculation algorithms to map the real-time collected thin core plate image and the target layout template image in the background database to high-dimensional feature space for deep feature extraction, so as to generate the target real-time distribution feature map representing the actual processing scene and the target standard distribution feature map under standard process conditions, respectively. Then, the two types of feature maps are subjected to feature phase alignment cross-domain correlation analysis to capture the nonlinear deformation law caused by micro-contraction, crystal reorganization and thermal expansion caused by laser browning treatment through adaptive matching of feature phase differences, and then fuse to generate the semantic-level perception feature map of target position offset. Finally, the target offset data is parsed through feature decoding to drive the dynamic compensation of laser focus position and energy density. In this way, it is possible to break through the limitations of traditional linear mapping, accurately identify the local nonlinear deformation caused by material modification and thermal stress accumulation of ultra-thin core plates during laser processing, and at the same time reduce the need for manual intervention and enhance process adaptability.

[0076] In step S3, the laser holes are filled, and the thin core board is thickened to the required thickness of 0.203mm through a lamination process to obtain a thick core board. It should be understood that in the processing of printed circuit boards, the initial goal is to process a 0.203mm thick core board layer, but there are technical difficulties in directly laser-processing a 0.203mm thick core board, such as the need to invest in a specific laser laser machine and replace the high-priced hole-filling solution model, which is costly. First, a 0.076mm thin core board is laser processed, and then it is thickened to 0.203mm through a lamination process, which can effectively avoid the technical and cost issues of directly processing a thick core board and meet the thickness requirements of the design specifications. And by filling the 0.1mm laser holes obtained by the initial laser processing, it can effectively prevent the laser holes from being deformed or collapsed due to factors such as pressure during the lamination process, ensuring the quality and reliability of the laser holes during subsequent line connections. Moreover, after filling the holes, the surface of the thin core board can become smooth, providing a good foundation for subsequent laser browning and laser direct hole forming processing, facilitating precise positioning and processing, and is conducive to the smooth advancement of the entire printed circuit board laser hole processing technology.

[0077] In step S4, after the thick core board is subjected to laser browning treatment, a laser direct hole forming method is used to process a laser hole with a diameter of 0.1 mm on the thick core board. It should be understood that printed circuit boards usually require multi-layer circuit interconnection to achieve complex functions. After the first round of laser laser holes is completed on the thin core board, the thick core board is obtained by lamination and thickening. At this time, it is necessary to process the laser laser holes again on the thick core board to establish new interlayer connections to meet the design requirements of multi-layer interconnection of the circuit board and ensure that the signal can be stably transmitted between different layers. Specifically, the copper layer on the surface of the thick core board may change during the lamination process, affecting the effect of subsequent laser hole forming. Laser browning treatment can form a uniform browning layer on the surface of the copper layer, enhance the copper layer's absorption efficiency of laser energy, and increase the bonding force between the copper layer and subsequent materials, creating conditions for accurate and high-quality laser hole forming. By performing laser direct hole forming on the thick core board after laser browning treatment, laser holes with a diameter of precisely 0.1 mm can be processed. This method can meet the high-precision requirements of printed circuit boards for micropores, avoid aperture deviations that may be caused by other processing methods, and ensure the electrical performance and signal transmission quality of the circuit boards.

[0078] In general, by using laser direct hole-forming technology instead of traditional mechanical drilling, the risk of substrate microcracks and interlayer offset caused by physical stress is fundamentally avoided. Laser browning treatment is first implemented at the ultra-thin core board (0.076mm) stage to optimize the laser absorption characteristics of the copper layer surface, effectively improve the consistency of hole diameter and hole position accuracy, and overcome the influence of the pretreatment process on energy absorption efficiency. Through the innovative process of processing first and then thickening, after the thin core board is processed and filled with microholes, it is pressed and thickened to 0.203mm for secondary processing, which significantly reduces the cumulative deformation effect of multiple hot pressing on the existing hole structure and ensures the reliability of interlayer conduction. In particular, the synergistic effect of two independent laser browning and hole-forming stages not only solves the surface treatment problem during thin plate processing, but also ensures high-precision secondary processing of the thickened board layer, forming a precise interconnection structure through the thickness, meeting the stringent requirements of high-frequency and high-speed circuits for signal integrity. This phased manufacturing strategy systematically breaks through the quality bottlenecks such as hole wall roughness and hole filling defects in the traditional multiple pressing process through optimized reconstruction of the process sequence, providing a reliable technical path for the industrialization of arbitrary-layer HDI products.

[0079] In summary, the laser hole processing method for printed circuit boards based on the embodiment of the present application is explained, which first performs laser browning treatment on a thin core board with a thickness of 0.076mm, and then uses a laser direct hole forming method to process a laser hole with a diameter of 0.1mm on the thin core board, and then fills the laser hole, and thickens the thin core board to the required thickness of 0.203mm through a lamination process to obtain a thick core board, and finally uses a laser direct hole forming method to process a laser hole with a diameter of 0.1mm on the thick core board after laser browning. In this way, it can effectively break through the quality bottlenecks such as hole wall roughness and hole filling defects in the traditional multiple lamination process, and can effectively guarantee the reliability of interlayer conduction, and can provide a reliable technical path for the industrialization of any layer of HDI products.

[0080] Figure 6 FIG. 1 is a system block diagram of a laser hole processing system for a printed circuit board according to an embodiment of the present application. Figure 6 As shown, a laser hole processing system 100 for a printed circuit board according to an embodiment of the present application includes: a target offset data determination module 110, which is used to acquire a thin core board image and determine target offset data based on feature semantic comparison analysis between the thin core board image and the target layout template image; and a processing parameter adjustment module 120, which is used to adjust laser processing parameters based on the target offset data.

[0081] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the laser hole processing system 100 for printed circuit boards have been described above with reference to Figures 1 to 5The laser hole processing method for a printed wiring board has been described in detail, and therefore, its repeated description will be omitted.

[0082] In summary, the laser hole processing system 100 for printed circuit boards based on the embodiment of the present application is explained, which first performs laser browning treatment on a thin core board with a thickness of 0.076mm, and then uses a laser direct hole forming method to process a laser hole with a diameter of 0.1mm on the thin core board, and then fills the laser hole, and thickens the thin core board to the required thickness of 0.203mm through a lamination process to obtain a thick core board, and finally uses a laser direct hole forming method to process a laser hole with a diameter of 0.1mm on the thick core board after laser browning. In this way, it can effectively break through the quality bottlenecks such as hole wall roughness and hole filling defects in the traditional multiple lamination process, and can effectively guarantee the reliability of interlayer conduction, and can provide a reliable technical path for the industrialization of any layer of HDI products.

Claims

1. A laser hole processing method for a printed circuit board, characterized in that: include: Provide thin core board with thickness of 0.076mm; After the thin core board is subjected to laser browning treatment, a laser hole with a diameter of 0.1 mm is processed on the thin core board by a laser direct hole forming method, including: acquiring an image of the thin core board, and determining target offset data based on feature semantic comparison analysis between the image of the thin core board and a target layout template image; adjusting laser processing parameters based on the target offset data; Filling the laser holes, and thickening the thin core board to the required thickness of 0.203 mm by a lamination process to obtain a thick core board; After the thick core board is subjected to laser browning treatment, a laser hole with a diameter of 0.1 mm is processed on the thick core board by using a laser direct hole forming method.

2. The laser hole processing method for a printed circuit board according to claim 1, characterized in that: Acquiring a thin core board image captured by a camera, and determining target offset data based on feature semantic comparison analysis between the thin core board image and the target layout template image, including: Acquiring an image of the thin core board captured by a camera; Extracting the target layout template image from a backend database; The target offset data is determined based on feature semantic comparison analysis between the thin core board image and the target layout template image.

3. The laser hole processing method for a printed circuit board according to claim 2, characterized in that: Determining target offset data based on feature semantic comparison analysis between the thin core plate image and the target layout template image includes: Extracting target position features from the thin core plate image and the target layout template image to obtain a target real-time distribution feature map and a target standard distribution feature map; Performing target position feature adaptive phase alignment perception on the target real-time distribution feature map and the target standard distribution feature map to obtain a target position offset semantic level perception feature map; Feature decoding is performed on the target position offset semantic level perception feature map to obtain the targeting offset data.

4. The laser hole processing method for a printed circuit board according to claim 3, characterized in that: Target position feature extraction is performed on the thin core plate image and the target layout template image to obtain a target real-time distribution feature map and a target standard distribution feature map, including: inputting the thin core plate image and the target layout template image into a target position sensitive twin detection network to obtain the target real-time distribution feature map and the target standard distribution feature map.

5. The laser hole processing method for a printed circuit board according to claim 4, characterized in that: The target real-time distribution feature map and the target standard distribution feature map are subjected to target position feature adaptive phase alignment perception to obtain a target position offset semantic level perception feature map, including: Performing feature decoupling and phase alignment on the target real-time distribution feature map and the target standard distribution feature map to obtain a set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs; Performing target position feature joint attention perception on each phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pair in the set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs to obtain a set of target distribution feature joint perception attention weights; Based on the set of target distribution features and joint perceptual attention weights, the set of phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pairs is subjected to attention-driven significant aggregation to obtain the target position offset semantic-level perceptual feature map.

6. The laser hole processing method for a printed circuit board according to claim 5, characterized in that: The target real-time distribution feature map and the target standard distribution feature map are subjected to feature decoupling and phase alignment to obtain a set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs, including: Performing feature decoupling on the target real-time distribution feature map and the target standard distribution feature map to obtain a set of target real-time distribution local feature coding matrices and a set of target standard distribution local feature coding matrices; Based on the feature phase alignment between any two target real-time distributed local feature coding matrices and target standard distributed local feature coding matrices in the set of target real-time distributed local feature coding matrices and the set of target standard distributed local feature coding matrices, the set of target real-time distributed local feature coding matrices and the set of target standard distributed local feature coding matrices are subjected to feature phase dynamic search and alignment to obtain a set of phase-aligned {target real-time distributed local feature coding matrix, target standard distributed local feature coding matrix} feature pairs.

7. The laser hole processing method for a printed circuit board according to claim 6, characterized in that: Performing target position feature joint attention perception on each phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pair in the set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs to obtain a set of target distribution feature joint perception attention weights, including: Calculating the correlation matrix between each phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pair in the set of phase-aligned {target real-time distribution local feature coding matrix, target standard distribution local feature coding matrix} feature pairs to obtain a set of target real-time-standard distribution correlation matrices; Performing a trace measurement on each target real-time-standard distribution association matrix in the set of target real-time-standard distribution association matrices to obtain a set of target real-time-standard distribution trace measurement values; Based on the set of target real-time-standard distribution trace measurement values, a set of target distribution feature joint perceptual attention weights is obtained.

8. The laser hole processing method for a printed circuit board according to claim 7, characterized in that: Based on the set of target real-time-standard distribution trace measurement values, a set of target distribution feature joint perception attention weights is obtained, including: Based on the set of phase-aligned {target real-time distribution local feature encoding matrix, target standard distribution local feature encoding matrix} feature pairs, the set of target real-time-standard distribution trace measurement values ​​is compactified based on the manifold of bidirectional connectivity to obtain a set of optimized target real-time-standard distribution trace measurement values; Based on the set of optimized target real-time-standard distribution trace measurement values, a set of target distribution feature joint perceptual attention weights is obtained.

9. The laser hole processing method for a printed circuit board according to claim 8, characterized in that: Feature decoding is performed on the target position offset semantic level perceptual feature map to obtain the targeting offset data, including: inputting the target position offset semantic level perceptual feature map into a decoder-based target offset metric to obtain the target offset data.

10. A laser hole processing system for printed circuit boards, characterized in that: include: A target offset data determination module is used to obtain a thin core board image and determine target offset data based on feature semantic comparison analysis between the thin core board image and the target layout template image; The processing parameter adjustment module is used to adjust the laser processing parameters based on the target offset data.

Images