Printed circuit board ultrathin prepreg laser hole processing method and system

Abstract

The invention provides a printed circuit board ultrathin prepreg laser hole processing method and system, and relates to the field of communication manufacturing, and the method comprises the steps: providing a PCB core board, manufacturing an inner layer circuit board, then carrying out a pressing technology, carrying out laser browning and laser processing drilling, forming a plurality of blind holes, and carrying out automatic optical detection. And the processing intelligence is improved, and the quality risk is controllable. According to the laser processing link, a traditional multi-step hole repairing mode is abandoned, and blind hole forming with the hole diameter smaller than or equal to 75 micrometers is achieved directly through single-time high-precision laser copper breaking. And in the quality control link, abnormal blind holes deviating from group characteristics are automatically identified based on blind hole image data acquired by an AOI system, standardized and large-scale adaptation of defect detection is realized, and manual inspection of whether the positions of the holes deviate or not, whether the hole diameters meet requirements or not and whether residues or pollutants exist in the holes or not is assisted.

Description

Technical Field

[0001] This application relates to the field of communication manufacturing. In particular, in the embodiments of this application, it relates to a method and system for laser drilling of ultra-thin prepregs for printed circuit boards. Background Art

[0002] With the rapid development of electronic products towards miniaturization and high density, the market demand for high-density interconnect (HDI) and ultra-high-end high-density interconnect (UltraHDI) printed circuit boards (PCBs) has been continuously climbing. Under this trend, the process requirements for ultra-thin prepregs (such as 1017PP, 1010PP, with a dielectric thickness ≤ 25μm after lamination) and micro-hole processing (hole diameter ≤ 75μm) in PCB manufacturing have become increasingly stringent.

[0003] Traditional laser drilling processes usually adopt a multi-step method of "1-time copper breaking and window opening + N-time hole repair" to meet the accuracy requirements of micro-holes, resulting in low processing efficiency, increased equipment wear, and the need for additional investment in high-precision hole repair equipment. In addition, ultra-thin prepregs are easily affected by thermal stress during the lamination process. If the residual glue is not removed thoroughly (such as using traditional chemical glue removal processes), it may lead to the failure of blind hole electrical connections. Although AOI (Automated Optical Inspection) systems are widely used for blind hole quality inspection, existing methods mostly rely on manual experience to judge defects, lacking an efficient automated anomaly recognition mechanism and being difficult to meet the requirements of large-scale production.

[0004] Therefore, an optimized laser hole processing solution for ultra-thin prepregs of printed circuit boards is expected to achieve the coordinated optimization of laser hole processing and quality control through a single-pass laser copper breaking forming process and an intelligent AOI defect detection algorithm. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. The embodiments of this application provide a method and system for laser drilling of ultra-thin prepregs for printed circuit boards. It provides a PCB core board and manufactures an inner layer circuit board, then performs a lamination process, laser brownification, and laser processing for drilling to form multiple blind holes and conducts an automated optical inspection to achieve an improvement in processing intelligence and controllability of quality risks. Among them, the laser processing link abandons the traditional multi-step hole repair mode and directly realizes the forming of blind holes with a hole diameter ≤ 75μm through a single-pass high-precision laser copper breaking. The quality control link, based on the blind hole image data collected by the AOI system, automatically identifies abnormal blind holes that deviate from the group characteristics, realizes the standardization and large-scale adaptation of defect detection, and assists in manually checking whether the position of the holes is offset, whether the hole diameter meets the requirements, and whether there are residues or contaminants in the holes and other defects.

[0006] According to one aspect of this application, a method for laser drilling of ultra-thin prepregs for printed circuit boards is provided, which includes:

[0007] Provide a PCB core board;

[0008] Produce an inner layer circuit pattern on the PCB core board to obtain an inner layer circuit board;

[0009] Press the inner layer circuit board, the prepreg layer and the outer copper foil layer through a pressing process to obtain a multi-layer board pressing structure; after laser brownification of the multi-layer board pressing structure, use a laser beam to drill a plurality of laser holes in the outer copper foil layer and the prepreg layer of the multi-layer board pressing structure, and the plurality of laser holes form a plurality of blind holes for realizing electrical connection between different layer circuits;

[0010] Perform automatic optical inspection on the plurality of blind holes.

[0011] According to another aspect of the present application, there is provided a laser hole processing system for a printed circuit board ultra-thin prepreg, which is used to execute the above-mentioned laser hole processing method for a printed circuit board ultra-thin prepreg, and includes:

[0012] A blind hole detection image acquisition module, which is used to acquire a plurality of blind hole detection images collected by an AOI system;

[0013] A blind hole image feature encoding module, which is used to extract image features from the plurality of blind hole detection images to obtain a plurality of blind hole image feature encoding vectors;

[0014] A blind hole image clustering module, which is used to perform blind hole image feature clustering analysis on the plurality of blind hole image feature encoding vectors to obtain a blind hole image clustering encoding vector;

[0015] A difference description coefficient calculation module, which is used to calculate the blind hole image feature difference description coefficients between each blind hole image feature encoding vector in the plurality of blind hole image feature encoding vectors and the blind hole image clustering encoding vector respectively to obtain a plurality of blind hole image feature difference description coefficients;

[0016] A defective blind hole determination module, which is used to mark the blind hole corresponding to the largest one among the plurality of blind hole image feature difference description coefficients as a suspected defective blind hole.

[0017] Compared with the prior art, a laser hole processing method and system for ultra-thin prepreg of printed circuit board provided by the present application provides a PCB core board and manufactures an inner layer circuit board, then performs a lamination process and laser brownification and laser processing for drilling to form a plurality of blind holes and performs automatic optical inspection, realizing the improvement of processing intelligence and the controllability of quality risks. Among them, in the laser processing link, the traditional multi-step hole repair mode is abandoned, and the blind hole with a hole diameter ≤ 75μm is directly formed by single-shot high-precision laser copper breaking. In the quality control link, based on the blind hole image data collected by the AOI system, abnormal blind holes deviating from the population characteristics are automatically identified, realizing the standardization and large-scale adaptation of defect detection, and assisting manual inspection of whether the position of the hole is offset, whether the hole diameter meets the requirements, and whether there are residues or contaminants and other defects in the hole. Description of the Drawings

[0018] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 It is a flowchart of the laser hole processing method for ultra-thin prepreg of printed circuit board according to an embodiment of the present application.

[0020] Figure 2 It is a schematic diagram of data flow of the laser hole processing method for ultra-thin prepreg of printed circuit board according to an embodiment of the present application.

[0021] Figure 3 It is a flowchart of automatically optically inspecting a plurality of blind holes in the laser hole processing method for ultra-thin prepreg of printed circuit board according to an embodiment of the present application.

[0022] Figure 4 It is a flowchart of performing blind hole image feature clustering analysis on a plurality of blind hole image feature coding vectors in the laser hole processing method for ultra-thin prepreg of printed circuit board according to an embodiment of the present application to obtain a blind hole image clustering coding vector.

[0023] Figure 5 It is a flowchart of performing clustering compensation on a plurality of blind hole image feature coding vectors in the laser hole processing method for ultra-thin prepreg of printed circuit board according to an embodiment of the present application to obtain a blind hole image feature linear clustering compensation component coding vector.

[0024] Figure 6 It is a system block diagram of the laser hole processing system for ultra-thin prepreg of printed circuit board according to an embodiment of the present application. Detailed Embodiments

[0025] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0026] The term "exemplary" used herein means "serving as an example, embodiment, or illustration". Any embodiment described herein as "exemplary" is not necessarily to be construed as superior to or better than other embodiments.

[0027] In addition, for a better description of the present application, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present application can also be implemented without some specific details. In some instances, methods, means, elements, and circuits well-known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0028] Furthermore, the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, "a plurality" means two or more unless otherwise specifically defined.

[0029] As electronic products tend to be miniaturized and highly integrated, the market demand for high-density interconnect (HDI) and ultra-high-end high-density interconnect (UltraHDI) printed circuit boards (PCBs) is on the rise, and the process requirements for ultra-thin prepregs and microvia processing are becoming more stringent. Traditional laser drilling uses multiple operations to meet the accuracy requirements, resulting in low efficiency, increased equipment wear, and the need for additional investment in high-precision hole repair equipment. If the thermal stress and residual glue problems in the lamination of ultra-thin prepregs are not properly handled, electrical connection failures may occur. Although AOI (Automated Optical Inspection) systems are widely used for blind hole quality inspection, they mostly rely on manual experience and lack an efficient automated anomaly recognition mechanism, making it difficult to meet the demands of large-scale production. The existing processes face challenges in terms of efficiency, cost, and adaptability.

[0030] In view of the above technical problems, the technical concept of this invention aims to construct a highly efficient and collaborative laser drilling system for ultra-thin prepregs by integrating the single-shot laser copper punching process with an intelligent defect detection algorithm. In the laser processing stage, the traditional multi-step hole repair mode is abandoned, and blind holes with a diameter ≤ 75μm are directly formed by single-shot high-precision laser copper punching, which helps to shorten the processing cycle and reduce the equipment complexity. In the quality control stage, based on the blind hole image data collected by the AOI system, unsupervised feature clustering and difference analysis algorithms are used to automatically identify abnormal blind holes that deviate from the population characteristics, replacing manual experience judgment and achieving standardized and large-scale adaptation of defect detection. This intelligent method can assist in manually checking whether the hole position is offset, whether the hole diameter meets the requirements, and whether there are residues or contaminants in the hole and other defects. Through the closed-loop linkage between the laser process and intelligent detection, it ultimately helps to achieve the comprehensive goals of improving processing intelligence and controlling quality risks.

[0031] This application proposes a method for laser drilling of ultra-thin prepregs for printed circuit boards. Figure 1 It is a flowchart of the method for laser drilling of ultra-thin prepregs for printed circuit boards according to an embodiment of this application. Figure 2 It is a schematic diagram of data flow of the method for laser drilling of ultra-thin prepregs for printed circuit boards according to an embodiment of this application. As Figure 1 and Figure 2 shown, the method for laser drilling of ultra-thin prepregs for printed circuit boards according to an embodiment of this application includes: S110, providing a PCB core board; S120, fabricating an inner layer circuit pattern on the PCB core board to obtain an inner layer circuit board; S130, laminating the inner layer circuit board, prepreg layer, and outer copper foil layer through a lamination process to obtain a multi-layer board lamination structure; S140, after laser brownification of the multi-layer board lamination structure, using a laser beam to drill a plurality of laser holes in the outer copper foil layer and prepreg layer of the multi-layer board lamination structure, and the plurality of laser holes form a plurality of blind holes for realizing electrical connection between different layer circuits; S150, performing automatic optical inspection on the plurality of blind holes.

[0032] In the above method for laser hole machining of ultra-thin prepreg for printed circuit boards, in step S110, a PCB core board is provided. It should be understood that considering the trend of miniaturization and high density of electronic products, the requirements for PCB core boards are becoming increasingly stringent. This includes not only the basic requirements in terms of mechanical strength, thermal stability, etc., but also electrical performance indicators such as dielectric constant and loss tangent. In addition, in order to meet the manufacturing requirements of HDI (High Density Interconnect) and UltraHDI (Ultra High-End High Density Interconnect) PCBs, the selected PCB core board also needs to support fine machining capabilities at the micron level, especially the special requirements for blind via forming technology. Specifically, when selecting a PCB core board, multiple factors need to be considered comprehensively. First is the material selection. Currently, common PCB core board materials include glass fiber reinforced epoxy resin (FR-4), polyimide (PI), etc., which have different physical and chemical properties. For example, FR-4 is widely used in various electronic devices due to its good mechanical properties, relatively low cost, and moderate electrical performance; while polyimide performs excellently in the manufacturing of flexible or rigid-flexible circuit boards due to its excellent high-temperature resistance and flexibility. Next are the requirements for the size and flatness of the PCB core board. To ensure consistency and accuracy in the subsequent processing, the selected core board should have high flatness. This means that there should be no obvious unevenness or warping on the entire board surface, otherwise it may lead to inaccurate lithography pattern transfer, uneven etching, and other problems. In addition, the thickness of the core board also needs to be strictly controlled to adapt to the design requirements of different hierarchical structures. Usually, the selection of the core board thickness needs to consider factors such as the mechanical strength requirements of the final product, signal transmission characteristics, and heat dissipation performance. After obtaining a PCB core board that meets the specifications, a series of pretreatment processes also need to be carried out on it to ensure that the surface quality and cleanliness meet the standards. This step mainly includes removing the surface oxide layer, stains, and other impurities, and at the same time preparing for the production of the inner layer circuit pattern. Common pretreatment methods include chemical cleaning, plasma treatment, etc. Among them, chemical cleaning removes surface contaminants by using specific solvents, while plasma treatment can improve surface activity at the molecular level and enhance the adhesion between the photoresist and the core board. After appropriate pretreatment, the surface of the core board will become smoother and more receptive to subsequent operations. In the above method for laser hole machining of ultra-thin prepreg for printed circuit boards, in step S120, an inner layer circuit pattern is made on the PCB core board to obtain an inner layer circuit board. It should be understood that specifically, the process of making the inner layer circuit pattern begins with the preparation of design documents. These design documents usually contain the specific layout of the circuit, including the position, width, spacing of the conductors, and other important electrical parameters. Based on this information, professional computer-aided design (CAD) software can be used to generate a photomask or a digital file directly for the exposure equipment.The importance of this step lies in that it ensures that every detail can be accurately transferred onto the PCB core board, thus guaranteeing the functionality and stability of the final circuit board. Next, a photosensitive material, i.e., photoresist, is coated on the surface of the PCB core board. This material is sensitive to light of a specific wavelength and can form the required circuit pattern through the processes of exposure and development. Selecting the appropriate photoresist is crucial, such as positive or negative photoresist, as it directly affects the precision and clarity of the lines. During the coating process, it is necessary to ensure that the photoresist evenly covers the entire surface of the PCB core board, avoiding the occurrence of bubbles, unevenness, or missing areas. Here, the spin coating method or the spraying method is usually adopted to achieve this goal. The former uses centrifugal force to evenly distribute the photoresist, while the latter is suitable for substrates with more complex shapes. After coating the photoresist, a high-precision exposure device is used to project the designed circuit pattern onto the photoresist. This process is called exposure, which is carried out by using ultraviolet light or other types of light passing through a pre-prepared mask or directly generated by a digital micromirror device (DMD) to cause chemical changes in the photoresist. The quality of exposure is directly related to the resolution and accuracy of the final circuit pattern. Therefore, during the entire exposure process, factors such as the energy distribution of the light source, exposure time, and environmental conditions must be strictly controlled. For example, an appropriate exposure time can ensure that the photoresist reacts fully, while too long or too short a time may result in a blurred or incomplete pattern. After exposure, the development stage is entered. During this process, the photoresist in the unexposed area or the exposed area is removed by the corresponding developer, leaving a mask consistent with the designed pattern. The developer is used to dissolve the photoresist in the unexposed area (for positive photoresist) or, conversely, dissolve the photoresist that has been exposed to light (for negative photoresist). The developer can effectively dissolve the unpolymerized photoresist without damaging the polymerized part. The ideal development result should be that a clear and complete circuit pattern appears, while retaining enough photoresist to protect the underlying copper foil area from subsequent etching. Through development, the originally transparent photoresist layer now presents a clearly visible circuit pattern, preparing for the next etching process. Subsequently, the etching process is carried out, which is one of the core steps in fabricating the inner layer circuit pattern. In this step, a chemical etchant is used to remove the copper foil layer on the surface of the substrate that is not covered by the photoresist, only retaining the part protected by the photoresist as the conduction path. Control during the etching process is very critical because any deviation may result in a line width not meeting the requirements or an open circuit phenomenon. An excessively high etching rate may lead to insufficient line width or even an open circuit, while too low may cause undercutting, affecting the fineness of the circuit. Therefore, it is necessary to precisely regulate the concentration, temperature, and flow rate of the etching solution to ensure uniform and thorough removal of the excess copper layer. In addition, it should be noted that the by-products generated during the etching process may remain on the surface of the circuit board, affecting its electrical performance. Therefore, the cleaning step cannot be ignored either.This step is usually called desmearing, aiming to expose the finally formed inner layer circuit pattern. The desmearing process also requires the use of specific chemical solvents, which can effectively dissolve the photoresist without damaging the underlying copper wires. It should be noted that over-erosion of the copper wires should be avoided during the desmearing process to prevent affecting the electrical conductivity of the circuit. After the desmearing treatment, the surface of the PCB core board presents a clear and complete circuit pattern, providing a solid foundation for subsequent multi-layer lamination and other processing steps.

[0033] In the above method for laser drilling of ultra-thin prepreg for printed circuit boards, in step S130, the inner layer circuit board, the prepreg layer, and the outer layer copper foil layer are laminated through a lamination process to obtain a multi-layer board laminated structure. In the embodiments of the present application, the model of the prepreg layer is 1017PP or 1010PP. In the multi-layer board laminated structure, the thickness of the prepreg layer after lamination is 18μm - 25μm. It should be understood that in order to integrate more functions in a limited space, multi-layer boards came into being. A multi-layer board is composed of multiple individual layers, and each layer can contain different circuit patterns. These layers are bonded together by a specific insulating material - prepreg, and electrical connections between the layers are achieved through drilling and electroplating. Therefore, the lamination process has become a key link in constructing such a complex structure. Selecting the appropriate prepreg is crucial for the successful implementation of lamination. A prepreg is a material composed of epoxy resin or other thermosetting resins and fiberglass cloth. It is in a partially cured state at room temperature, but can melt and flow under the action of high temperature and pressure, filling the gaps between the layers, thereby achieving a firm bonding effect. Different models of prepregs have different thicknesses and dielectric constants, which makes them suitable for various different application scenarios. For example, prepregs of model 1017PP or 1010PP can provide ideal electrical insulation performance after lamination, while maintaining a relatively thin thickness (usually between 18μm - 25μm), which is particularly important for achieving micron-level laser drilling. In actual operation, the inner layer circuit board, the prepreg layer, and the outer layer copper foil layer are carefully arranged and placed in a specially designed press. Usually, the inner layer circuit board is placed at the bottom, followed by laying one or more layers of prepregs in sequence, and finally covered with the outer layer copper foil. The alignment accuracy between each layer is crucial, because even a tiny offset may cause serious electrical problems or mechanical defects in the final product. Therefore, in actual operation, precision alignment equipment is often required to ensure that each layer can be accurately stacked together. To further improve the alignment accuracy, some positioning marks can also be pre-set on the inner layer circuit board, and these marks can be used for calibration and inspection in subsequent processes. The press can work under precisely controlled temperature and pressure conditions to ensure that the entire lamination process is uniform and consistent. During the lamination process, the resin in the prepreg begins to melt and flow, gradually filling the microscopic uneven areas on the surface of the inner layer circuit board and the gaps between the layers. At the same time, the applied pressure prompts the resin to fully infiltrate the surface of the copper foil, forming a tight bond. This process not only enhances the mechanical strength of the overall structure but also effectively prevents delamination caused by air or other impurities. It should be noted that the quality of the lamination process directly affects the reliability and service life of the final product. If the lamination is improper, problems such as interlayer separation and bubble residue may occur, which will seriously affect the function and stability of the circuit board. Therefore, strict pretreatment of each material is required before lamination.For example, the surface of the inner-layer circuit board must be clean and dust-free to ensure good bonding effect; the prepreg should be protected from moisture, because moisture can cause bubbles to form in the resin during heating, which will in turn affect the interlayer bonding strength; while the outer-layer copper foil needs to have appropriate roughness to better bond with the prepreg. In addition, the influence of environmental factors such as temperature and humidity changes also needs to be considered, as these will all affect the lamination quality. Besides the physical bonding, the lamination process also plays an important role in electrical performance. As an excellent insulating material, the prepreg can effectively isolate signal interference between different layers, ensuring the purity and integrity of signal transmission. At the same time, it also has certain heat conduction performance, which helps with heat dissipation, and this is particularly important for high-power electronic devices. However, to achieve the best electrical performance, the thickness and dielectric constant of the prepreg must be strictly controlled. An overly thick prepreg will increase the distance of the signal transmission path, resulting in signal delay and loss; while an overly thin one may not provide sufficient insulation protection, easily leading to the risk of short circuit.

[0034] In the above laser hole processing method for ultra-thin prepreg of printed circuit boards, in step S140, after laser brownification of the multi-layer board lamination structure, a plurality of laser holes are drilled in the outer copper foil layer and the prepreg layer of the multi-layer board lamination structure, and the plurality of laser holes form a plurality of blind holes for realizing electrical connection between different layer circuits. In the embodiments of the present application, the aperture of the laser holes is less than or equal to 75 μm. It should be understood that laser brownification generates a microscopically rough oxide layer on the surface of the multi-layer board, which not only increases the surface roughness, thereby improving the subsequent interlayer bonding force, but also enhances the laser absorption rate of the material. This is particularly important for the laser drilling process because a higher laser absorption rate means that the copper foil and prepreg can be more effectively penetrated to form precise and high-quality blind holes. In addition, laser brownification can also improve the chemical properties of the material surface and reduce factors that may affect the drilling quality, such as the presence of residues or contaminants. Therefore, performing laser brownification treatment before laser drilling can significantly improve the accuracy and reliability of drilling, laying a solid foundation for subsequent processes. Next is the process of laser drilling. With the development of electronic technology, especially the increasing demand for high-density interconnect (HDI) and ultra-high-end high-density interconnect (UltraHDI) printed circuit boards, traditional mechanical drilling methods can no longer meet the drilling requirements of micron-level or even smaller sizes. In contrast, laser drilling technology has become an ideal choice due to its high precision, high speed, and non-contact characteristics. By using a laser beam with a high energy density, the copper foil and prepreg materials can be locally heated and evaporated in a very short time to form the required micro-holes. This method can not only achieve micro-hole processing with a diameter less than 75 μm, but also ensure smooth hole walls, reducing the burr problems that may be caused by traditional mechanical drilling. An important feature of laser drilling is its ability to achieve single-time high-precision copper breaking and forming, without the need for a multi-step method such as "1-time copper breaking and window opening + N-time hole repair" as in traditional laser drilling. This way of directly completing the penetration and hole forming of the hole layer and the dielectric layer through a single laser pulse greatly simplifies the process flow, improves production efficiency, and reduces equipment losses. In addition, since there is no need to invest in high-precision hole repair equipment additionally, the overall manufacturing cost is also reduced. More importantly, the single-time laser copper breaking and forming technology can better control the aperture size and its consistency, which is crucial for ensuring the electrical performance of the final product. In addition to improving drilling accuracy and efficiency, laser drilling can also significantly enhance the functional density of the circuit board without increasing its size. In modern electronic devices, with the continuous increase in functional requirements, how to integrate more circuit components in a limited space has become an urgent problem to be solved. By drilling a plurality of blind holes in the multi-layer board and converting them into through-holes using processes such as electroplating, electrical connection between different layer circuits can be achieved, thereby expanding its wiring ability without affecting the overall size of the circuit board.This method not only helps to reduce the volume of the circuit board, but also helps to improve the signal transmission speed and reduce signal interference, further enhancing the overall performance of the electronic device. It should be noted that during the laser drilling process, parameters such as the laser energy, pulse width, and focusing accuracy must be strictly controlled to ensure the quality of the drilled holes. For example, excessive laser energy may cause damage to the hole wall or an oversized hole diameter, while insufficient energy may not be able to completely penetrate the material, resulting in drilling failure. In addition, the selection of the laser pulse width is equally important, as it determines the size of the heat-affected zone and the smoothness of the hole wall. Additionally, after laser drilling, a desmearing process is required because some resin smears may be generated during the drilling process, and these smears will affect the quality of subsequent hole metallization. Specifically, the desmearing process uses a chemical micro-etchant to slightly corrode the copper layer on the hole wall, remove contaminants on the copper surface, and prepare for subsequent chemical treatment. Specifically, in a specific embodiment of the present application, the desmearing process uses a chemical micro-etchant to slightly corrode the copper layer on the hole wall, remove contaminants on the copper surface, and prepare for subsequent chemical treatment. In another embodiment of the present application, potassium permanganate, a strong oxidant, is used to remove the resin smears generated by laser drilling.

[0035] Figure 3 It is a flowchart for automatically optically inspecting multiple blind holes in the laser hole processing method of the ultra-thin prepreg for printed circuit boards according to the embodiments of the present application. As Figure 3 shown, in the embodiment of the present application, in step S150, automatically optically inspecting multiple blind holes includes: S151, acquiring multiple blind hole detection images collected by the AOI system; S152, extracting image features from the multiple blind hole detection images to obtain multiple blind hole image feature coding vectors; S153, performing blind hole image feature clustering analysis on the multiple blind hole image feature coding vectors to obtain a blind hole image clustering coding vector; S154, respectively calculating the blind hole image feature difference description coefficients between each blind hole image feature coding vector in the multiple blind hole image feature coding vectors and the blind hole image clustering coding vector to obtain multiple blind hole image feature difference description coefficients; S155, marking the blind hole corresponding to the largest one among the multiple blind hole image feature difference description coefficients as a suspected defective blind hole. It should be understood that AOI, that is, Automated Optical Inspection (automatic optical inspection). Blind hole AOI is to automatically optically inspect the blind holes drilled by laser to ensure that the quality and quantity of the holes meet the requirements. Specifically, the AOI device will take images of the PCB board and compare them with preset standard data to detect whether there are defects in the holes. Among them, the detection content includes checking whether there are missed drilled holes, checking whether the position of the holes is offset, checking whether the hole diameter meets the requirements, and checking whether there are residues or contaminants in the holes. Through AOI detection, problems in the production process can be discovered and corrected in a timely manner to ensure product quality.

[0036] Specifically, in step S151, multiple blind hole detection images collected by the AOI system are obtained. It should be understood that the multiple blind hole detection images collected contain rich information, including but not limited to the position, size, shape, and surface quality of the blind holes, etc. By analyzing these images, potential defects can be identified, such as aperture offset, inconsistent size, rough hole walls, or the presence of residues, etc. To obtain clear and detailed blind hole images, the AOI system is usually equipped with high-resolution industrial cameras and precision optical lenses. These devices can capture details at the micron level, which is particularly important for detecting blind holes with a diameter less than 75μm. Usually, these cameras are equipped with high-sensitivity sensors and advanced imaging technologies, which can provide clear and sharp images. To further improve the image quality, an appropriate lens also needs to be selected to ensure that every detail within the entire field of view can be accurately captured. Common lens types include fixed-focus lenses and zoom lenses. In addition, good lighting conditions are crucial for obtaining high-quality detection images. Common lighting methods include coaxial lighting, backlighting, and ring lighting, etc., and each method has its specific application scenarios. For example, coaxial lighting is suitable for detecting fine features on flat surfaces, while backlighting is more suitable for transparent materials or occasions where the outline needs to be highlighted. By selecting the appropriate lighting method, the contrast between the target area and the background can be maximally enhanced, thereby improving the accuracy of defect detection. In actual operation, the AOI system will automatically adjust the position and focal length of the camera according to preset parameters to ensure that each image can cover all the blind holes to be detected. This not only improves the detection efficiency but also reduces the possibility of human intervention, further enhancing the consistency and reliability of the detection results. The AOI system is usually equipped with a precision motion control system that can move precisely in the X, Y, and Z directions. This means that the camera can automatically adjust its position according to needs to ensure that each blind hole can be clearly photographed. In addition, to cope with the possible differences between different batches of products, the AOI system can also adjust its parameter settings in real time, such as exposure time, gain, etc., to adapt to different detection requirements. This flexibility enables the AOI system to provide highly accurate detection results while maintaining high efficiency. Next is the specific process of image acquisition. At this stage, the AOI system will scan each blind hole in sequence along a predetermined path and trigger the camera to take pictures at appropriate positions. To ensure that each blind hole can be completely and clearly captured, multiple shooting points are usually set above each blind hole to obtain images from different angles. This method can not only improve the comprehensiveness of detection but also effectively reduce misjudgments caused by perspective problems. For example, some tiny cracks or residues may be difficult to detect from one angle but can be clearly visible from another angle. Therefore, by shooting from multiple angles, potential problems can be maximally discovered, improving the accuracy of detection.During the acquisition process, the AOI system also utilizes various advanced imaging techniques to improve the quality of images. For example, the use of high-dynamic range (HDR) imaging technology can effectively address the issue of image brightness differences caused by uneven illumination. HDR imaging synthesizes multiple images through multiple exposures, which can retain details while avoiding overexposure or underexposure. In addition, the use of polarized light imaging technology can eliminate the influence of surface reflections, especially when detecting metal surfaces, this technology is particularly effective. By using a polarizer to filter out unnecessary reflected light, the image can be made clearer, which helps to more accurately identify subtle features in blind holes.

[0037] In an embodiment of the present application, step S152, extracting image features from multiple blind hole detection images to obtain multiple blind hole image feature coding vectors, includes: passing each of the multiple blind hole detection images through a blind hole state feature extractor based on a dilated convolutional neural network model to obtain multiple blind hole image feature coding vectors. It should be understood that considering that during the laser hole processing of ultra-thin prepregs, due to the extremely high precision requirements for blind holes with a diameter ≤ 75μm, traditional manual experience judgment is easily interfered by subjective factors and is difficult to handle the massive image data in large-scale production. Therefore, after the AOI system acquires blind hole detection images, the visual information needs to be converted into structured data that can be quantitatively analyzed. Specifically, in the technical solution of the present application, image features are extracted from multiple blind hole detection images to obtain multiple blind hole image feature coding vectors. In particular, a dilated convolutional neural network model can be used to extract features from the blind hole detection images, which can strip out the key attributes directly related to the quality of the blind holes (such as edge contours, aperture sizes, hole wall smoothness, etc.) from complex optical images and compress them into low-dimensional feature coding vectors, providing a unified data interface for subsequent algorithm processing. The purpose of this step is to eliminate the interference of redundant image information, establish a mathematical representation system for blind hole quality, and shift anomaly detection from "experience-driven" to "data-driven". Through the generation of blind hole image feature coding vectors, the system can quickly match the group feature patterns of blind holes, providing a basis for automatically identifying abnormal individuals (such as position deviation, aperture out-of-tolerance, or residual glue residue) that deviate from the clustering center, thus facilitating the improvement of the standardization level and intelligence degree of defect detection. Its implementation effects are reflected in two aspects: on the one hand, automated feature extraction replaces manual visual inspection, reducing misjudgments and missed detections caused by fatigue or lack of experience; on the other hand, the feature coding vectors serve as reference data for defect analysis, providing quantitative feedback basis for process parameter optimization (such as laser energy adjustment, debonding process improvement), and ultimately forming a closed-loop optimization link for processing and detection to ensure the efficient and high-quality production of laser holes in ultra-thin prepregs.

[0038] Figure 4It is a flowchart for performing blind hole image feature clustering analysis on multiple blind hole image feature coding vectors to obtain blind hole image clustering coding vectors in the laser hole processing method of an ultra-thin prepreg for printed circuit boards according to an embodiment of the present application. As Figure 4As shown, in the embodiment of the present application, in step S153, blind hole image feature clustering analysis is performed on multiple blind hole image feature coding vectors to obtain a blind hole image clustering coding vector, including: S1531, performing linear clustering analysis on multiple blind hole image feature coding vectors to obtain an initial linear clustering center coding vector of blind hole image features; S1532, performing clustering compensation on multiple blind hole image feature coding vectors to obtain a linear clustering compensation component coding vector of blind hole image features; S1533, fusing the linear clustering compensation component coding vector of blind hole image features and the initial linear clustering center coding vector of blind hole image features to obtain a blind hole image clustering coding vector. It should be understood that although each blind hole image feature coding vector in the multiple blind hole image feature coding vectors has extracted the key attributes of the blind hole (such as shape, size, surface state), its distribution implies group commonality and individual differences, and the internal structure of the data needs to be mined through clustering. Traditional linear clustering can only capture macroscopic linear features, while blind hole defects in ultra-thin prepregs (such as micron-level offsets, residual glue residues) often show non-linear anomalies, and the clustering results need to be dynamically corrected by combining deep models to adapt to complex data distributions. That is to say, in the laser hole processing of ultra-thin prepregs, blind hole quality inspection needs to process a large amount of high-precision image data, and it is difficult for traditional manual or simple algorithms to efficiently identify complex defect patterns. Based on this, in the technical solution of the present application, blind hole image feature clustering analysis is further performed on multiple blind hole image feature coding vectors to obtain a blind hole image clustering coding vector. Specifically, the clustering analysis of the blind hole image feature coding vector is realized through a self-learning enhanced feature gain aggregation analysis method. Specifically, the principal component distribution of the blind hole features in the multiple blind hole image feature coding vectors is extracted through initial linear clustering analysis (such as K-means) to form a "group feature skeleton"; then the non-linear interaction relationship between the feature vector and the clustering center is captured by neural network modeling of the established deep collaborative implicit coding vector to capture local subtle deviations (such as burrs on the hole wall, thermal stress deformation); further, a feature clustering compensation increment operator is constructed to further correct the limitations of linear clustering. For example, the fractal non-integer dimensional scattering of the clustering space is balanced through an energy relaxation mechanism to suppress noise interference. Finally, the linear clustering center and the compensation component are fused to generate a blind hole image clustering coding vector, which not only retains global consistency but also incorporates local non-linear corrections to form a robust defect determination scale.The execution effect is reflected in three aspects: First, through the dynamic optimization of the clustering coding vector, the system can adapt to the blind hole feature distributions of different batches, avoiding misjudgment caused by process fluctuations; Second, the depth cooperation and compensation mechanism significantly improves the sensitivity to micro defects (such as a 0.5μm oversize aperture), overcoming the bottleneck of missed detection of non-linear anomalies by traditional AOI algorithms; Third, the clustering coding vector serves as a quality benchmark to drive the closed-loop optimization of laser processing parameters (such as pulse energy, focusing accuracy). For example, when the clustering shows a concentrated trend of hole position offset, the system can automatically feedback and adjust the laser positioning calibration to achieve two-way linkage between "detection - process", optimizing the laser hole processing technology of ultra-thin prepreg.

[0039] Specifically, in step S1531, linear clustering analysis is performed on multiple blind hole image feature coding vectors to obtain the initial linear clustering center coding vector of the blind hole image features, which is expressed by the blind hole linear clustering analysis formula as:

[0040] X = {x1, x2,..., x i ,..., x n}

[0041]

[0042] where X is multiple blind hole image feature coding vectors, and x1, x2, x i and x n are respectively the 1st, 2nd, ith, and nth blind hole image feature coding vectors among multiple blind hole image feature coding vectors, n is the number of vectors in X, and x cIt is the initial linear clustering center coding vector of blind hole image features. It should be understood that multiple blind hole image feature coding vectors contain high-dimensional non-linear data structures, but actual process defects (such as micro-hole offset, residual glue residue) often have potential linear separability. Despite the complex distribution of blind hole population features, there are still linear principal components (such as the aperture consistency and hole position uniformity of normal blind holes) that can characterize process stability at the macroscopic level. Through the linear clustering method under the unsupervised learning framework, the linear structure pattern reflecting the commonality of the processing system can be stripped from the high-dimensional feature space, providing a baseline reference for subsequent non-linear feature gain. This processing method stems from the stable output characteristics of process parameters in actual engineering (such as the uniformity of laser energy, the consistency of lamination stress), making the normal blind hole features show an aggregated distribution law in the low-rank subspace. By performing low-rank approximation on the blind hole image feature coding vectors through a linear clustering algorithm (such as K-means), the most significant linear principal components in the dataset are extracted to form the initial linear clustering center coding vector of blind hole image features. This coding vector is essentially a mathematical abstraction of the normal blind hole population features, and can characterize the blind hole morphology distribution under the process stable state by establishing a global feature skeleton. Moreover, it can decouple the linear and non-linear feature components, decomposing the high-dimensional feature space into a coarse-grained representation dominated by linear principal components (clustering centers) and non-linear residual components (subsequent compensation objects). In addition, it can also provide a reference system for defect detection, revealing potential anomalies by quantifying the deviation degree between individual blind hole features and the clustering center. Through iterative optimization of minimizing the sum of squared distances within the cluster, random noise in the high-dimensional features (such as AOI imaging interference) can be eliminated, and the linear principal components reflecting the essential laws of the processing system are retained. And within the subspace defined by the linear clustering center, the non-linear feature residuals of abnormal blind holes (such as hole wall burrs, thermal deformation) are explicitly separated, providing a directional guidance for the subsequent compensation and correction of the deep collaborative implicit coding vector. The linear clustering center output by this step serves both as the initial reference for quality determination and as the targeted input for non-linear gain operations, realizing the transition from process stability analysis to defect-sensitive detection.

[0043] Figure 5 It is a flowchart for clustering and compensating multiple blind hole image feature coding vectors to obtain the linear clustering compensation component coding vector of blind hole image features in the laser hole processing method of the ultra-thin prepreg for printed circuit boards according to the embodiments of the present application. As Figure 5As shown, in the embodiment of the present application, step S1532, clustering compensation is performed on multiple blind hole image feature encoding vectors to obtain a blind hole image feature linear clustering compensation component encoding vector, including: S1532-1, constructing a blind hole image feature depth collaborative implicit encoding vector between each blind hole image feature encoding vector in the multiple blind hole image feature encoding vectors and the blind hole image feature initial linear clustering center encoding vector; S1532-2, based on the blind hole image feature depth collaborative implicit encoding vector, calculating a feature clustering compensation increment operator of each blind hole image feature encoding vector in the multiple blind hole image feature encoding vectors relative to the blind hole image feature initial linear clustering center encoding vector to obtain multiple blind hole image clustering compensation increment operators; S1532-3, based on the multiple blind hole image clustering compensation increment operators, calculating the blind hole image feature linear clustering compensation component encoding vector of the multiple blind hole image feature encoding vectors.

[0044] Specifically, in step S1532-1, a blind hole image feature depth collaborative implicit encoding vector is constructed between each blind hole image feature encoding vector in the multiple blind hole image feature encoding vectors and the blind hole image feature initial linear clustering center encoding vector. It should be understood that although linear clustering analysis can extract the macroscopic principal components of blind hole population features, it cannot effectively represent the non-linear anomalies caused by process fluctuations or material property differences (such as the deformation of the hole wall caused by thermal stress and the aperture deviation caused by uneven laser energy). Traditional linear methods have limited ability to capture the complex interaction relationships hidden in the high-dimensional feature space, and actual defects often show dynamic non-linear associations between individual blind hole features and the global linear clustering structure (such as the coupling effect between residual glue distribution and laser penetration depth). Constructing a blind hole image feature depth collaborative implicit encoding vector can decouple the deep association between blind hole features and the linear clustering center through a deep learning model, solve the local non-linear deviation that is difficult to represent in the linear framework (such as the gradient change of micron-level hole position offset), and thus provide a data basis for the refined modeling of process defects. By modeling the non-linear collaborative relationship between the blind hole image feature encoding vector and the initial linear clustering center through a deep neural network, a composite feature representation containing process context information can be generated. Its core goal is to perform multi-scale fusion of the global distribution law provided by the linear clustering center (such as the aperture mean value and hole position uniformity of normal blind holes) and the local non-linear features of individual blind holes (such as the molten state of copper foil and the interlayer stress distribution of prepreg), and construct a dynamic collaborative encoding that can reflect both process benchmarks and individual anomaly potentials. This encoding extracts the implicit association between blind hole features and the clustering structure through non-linear transformation (such as the mapping relationship between laser pulse energy and hole wall roughness), forms a higher-order semantic expression oriented to defect sensitivity, and provides input data with higher discrimination for subsequent feature compensation based on the energy relaxation mechanism. The constructed blind hole image feature depth collaborative implicit encoding vector essentially realizes the quantitative unity of process stability and anomaly features.

[0045] In an embodiment of the present application, in step S1532-2, based on the blind hole image feature depth collaborative implicit coding vector, calculating the feature clustering compensation increment operator of each blind hole image feature coding vector in multiple blind hole image feature coding vectors relative to the blind hole image feature initial linear clustering center coding vector to obtain multiple blind hole image feature clustering compensation increment operators, includes: S1532-21, performing activation processing on the blind hole image feature initial linear clustering center coding vector to obtain the blind hole image feature initial linear clustering center activation coding vector; S1532-22, performing compensation based on the eigenvalue granularity on the blind hole image feature initial linear clustering center activation coding vector and the blind hole image feature depth collaborative implicit coding vector corresponding to the blind hole image feature coding vector at a predetermined position to obtain the blind hole image feature clustering intermediate compensation variable; S1532-23, performing non-linear resonance coupling correction on the blind hole image feature clustering intermediate compensation variable to obtain the blind hole image feature clustering intermediate correction compensation variable; S1532-24, performing normalization processing on the blind hole image feature clustering intermediate correction compensation variable to obtain the blind hole image feature clustering compensation increment operator corresponding to the blind hole image feature coding vector at a predetermined position.

[0046] Specifically, steps S1532-1 and S1532-2 are represented by the blind hole image feature optimization formula as:

[0047]

[0048] Among them, concat(·;·) is the concatenation operation, W i is the i-th learnable collaborative weight matrix among multiple learnable collaborative weight matrices, b i is the i-th collaborative bias vector among multiple collaborative bias vectors, Sigmoid is the activation function, r i is the blind hole image feature depth collaborative implicit coding vector between x i and x c , f b (x i ) is the feature clustering compensation increment operator for calculating x i , v c is the blind hole image feature initial linear clustering center activation coding vector, is the k-th eigenvalue in r i , log2 is the logarithmic function value with base 2, is the k-th eigenvalue in v c , D is the vector length of r i and v c , and r i and v c have the same length, λ i is xi The corresponding intermediate compensation variable for blind hole image feature clustering, <,> represents the inner product of vectors, Σ i is x i The corresponding self-energy term for blind hole image feature clustering, σ(x i -r i ) 2 represents the variance of the vector (x i -r i ), L is the length of the vector (x i -r i ), ∈ i is x i The corresponding self-energy scattering factor for blind hole image features, Γ i is x i The corresponding clustering relaxation rate for blind hole image features, λ i ′ is x i The corresponding intermediate correction compensation variable for blind hole image feature clustering, softmax is the normalization function, ε i is x iThe corresponding blind hole image feature clustering compensation increment operator. It should be understood that constructing the blind hole image feature clustering compensation increment operator can perform a non-linear response domain mapping on the linear clustering center through an activation function (such as Sigmoid), release its potential correlation, and combine eigenvalue granularity compensation (such as inner product operation) to quantify the interaction strength between the blind hole image feature depth collaborative implicit coding vector and the activated clustering center, establishing an energy relaxation relationship between the process benchmark and individual anomalies; finally, through non-linear resonance coupling to correct the energy scattering effect of the simulated process parameter fluctuations in the clustering space, solve the non-equilibrium problem between the linear clustering space and the real process deviation, and realize the refined modeling of defect features. Generating a dynamic blind hole image feature clustering compensation increment operator through multi-stage non-linear operations can achieve the process self-adaptation optimization of the feature space. Specifically, mapping the linear clustering center to the non-linear response domain through activation processing (such as the Sigmoid function) can enhance its sensitivity to local features while suppressing the over-smoothing effect, improving the detection sensitivity of micro-crack gradients. And, based on eigenvalue granularity compensation to quantify the interaction strength between the collaborative implicit coding vector and the activated clustering center, capturing the microscopic gradient changes of defects such as aperture deviation (±0.8μm) and hole position offset (≥3μm). In addition, introducing non-linear resonance coupling to correct the simulated laser energy scattering effect enables the compensation increment to adaptively adjust the weight distribution of different defect types, reducing the misjudgment rate of residual glue residue defects. Additionally, through Softmax normalization processing to constrain the numerical range of the compensation operator, avoiding the instability of model convergence caused by feature scale differences, and improving the training convergence speed in batch detection. The finally constructed blind hole image feature clustering compensation increment operator has both global process stability constraints and local defect sensitivity, providing a highly robust input with energy relaxation characteristics for subsequent aggregation analysis.

[0049] When calculating the blind hole image feature clustering compensation increment operator, since in addition to the representation of the blind hole image feature coding vector x i , the incremental non-linear information of the initial linear clustering center coding vector x c of the blind hole image feature is added, this results in the addition of clustering elements within the clustering system, thus causing a systematic non-equilibrium state of the clustering space distribution.

[0050] Based on this, first use the inner product <x i , x c > of the blind hole image feature coding vector x i and the initial linear clustering center coding vector x c of the blind hole image feature to construct the blind hole image feature clustering self-energy term Σ i , and then calculate the blind hole image feature clustering relaxation rate Γ i :

[0051]

[0052] where σ(x i -x c ) 2 represents the variance of the vector (x i -x c ), L is the length of the eigenvector, and ∈ i is the self-energy scattering factor of the blind hole image feature related to the blind hole image feature clustering self-energy term Σ i . The blind hole image feature clustering relaxation rate Γ i is used to represent the scattering relaxation effect of the fractal structure in the clustering space on the spatial fractal non-integer dimension.

[0053] Finally, considering the energy relaxation effect of the blind hole image feature clustering relaxation rate Γ i related to the blind hole image feature clustering self-energy term Σ i under the scattering relation, through to perform resonance coupling based on energy relaxation on the fractal structure in the clustering space, so as to achieve the resonance coupling enhancement of the blind hole image feature clustering compensation increment operator under the representation of the clustering space fractal structure, thereby obtaining the non-linear increment coupling type clustering equilibrium state correction effect.

[0054] Specifically, in step S1532-3, based on multiple blind hole image feature clustering compensation increment operators, calculate the blind hole image feature linear clustering compensation component coding vectors of multiple blind hole image feature coding vectors, which is expressed by the blind hole image feature linear clustering compensation formula as:

[0055]

[0056] where, x bIt is the encoding vector of the linear clustering compensation component of the blind hole image features. It should be understood that although the clustering compensation increment operator of the blind hole image features can characterize local non-linear deviations (such as micro-offset of hole positions and abnormal distribution of residual glue), its independent effect is vulnerable to random noise (such as AOI imaging noise and reflection interference at the lamination interface), resulting in fragmented characteristics of the compensation results. Directly using discrete individual compensation operators for defect determination will lead to the risk of overfitting (such as misjudging local process fluctuations as systematic defects). Therefore, it is necessary to eliminate the accidental noise in individual compensation through global aggregation, extract the non-linear compensation mode reflecting the essential laws of the process, solve the problems of dispersion and redundancy of compensation signals in the high-dimensional feature space, and ensure the robustness of compensation correction. By using a weighted aggregation mechanism to elevate the individual clustering compensation increment operator of the blind hole image features to a global compensation component, a feature representation characterizing the commonality of process non-linear deviations can be constructed. Specifically, by fusing multiple clustering compensation increment operators of the blind hole image features through linear combination operations (such as weighted summation), the interference of random noise on the compensation results can be suppressed, and at the same time, the characterization ability of systematic process defects (such as hole group offset caused by laser focusing misalignment and deterioration of aperture consistency caused by uneven thickness of prepreg) can be strengthened. The generated encoding vector of the linear clustering compensation component of the blind hole image features is essentially a quantization index that converges the scattered local non-linear correction information (such as the compensation value of residual glue in a single hole) into a global process deviation (such as the probability field of residual glue distribution), forming a comprehensive feature base that can inherit the stability of the linear clustering center and characterize non-linear population deviations, providing data support for the subsequent reconstruction of the clustering space. The encoding vector of the linear clustering compensation component of the blind hole image features generated after executing this step realizes the mathematical abstraction of process deviations from local anomalies to global laws, eliminates the random noise in the individual clustering compensation increment operator of the blind hole image features through weighted aggregation, and quantitatively reflects the drift trend of process parameters.

[0057] Specifically, in step S1533, the encoding vector of the linear clustering compensation component of the blind hole image features and the initial encoding vector of the linear clustering center of the blind hole image features are fused to obtain the clustering encoding vector of the blind hole image, which is expressed by the clustering encoding formula of the blind hole image as:

[0058] v f =α·x b +β·x c

[0059] where α and β are the trainable weighted hyperparameters corresponding to X b and X c respectively, and V fIt is the clustering coding vector of blind hole images. It should be understood that although the linear clustering compensation component coding vector of blind hole image features can characterize the characteristics of the blind hole population under stable process conditions (such as the average aperture and the main components of the hole position distribution), it cannot effectively capture the non-linear anomalies caused by laser parameter drift or material property fluctuations (such as the micron-level hole wall deformation gradient and the spatial heterogeneity of the residual glue distribution). Although the initial linear clustering center coding vector of blind hole image features extracts local defect features through non-linear modeling, its independent effect is easily interfered by random noise and lacks the constraint of the global process benchmark. Therefore, fusing the linear clustering compensation component coding vector of blind hole image features and the initial linear clustering center coding vector of blind hole image features can solve the limitations of single feature representation, balance the contradiction between global process stability (characterized by the linear clustering center) and local defect sensitivity (characterized by the compensation component), and construct a comprehensive feature base that can not only reflect the common laws of the processing system but also accurately locate individual anomalies, thus overcoming the problem of insufficient generalization ability of traditional detection methods in complex process scenarios. By integrating linear and non-linear feature components through a weighted fusion mechanism, a robust clustering coding vector for defect detection is generated, which can dynamically adjust the contribution weights of the linear clustering center and the compensation component through trainable hyperparameters (α, β) to achieve enhanced abnormal feature under process benchmark constraints. By retaining the macroscopic process laws carried by the linear clustering center (such as the standard aperture range and the hole group distribution density), it can provide a stability anchor for defect determination. Moreover, by incorporating the non-linear deviation patterns characterized by the compensation component (such as the hole position offset vector field caused by thermal stress and the aperture gradient trend caused by laser defocusing), the detection sensitivity to composite defects (coexistence of position offset + residues) can be improved. In addition, by constructing a dynamic decision boundary that adapts to process fluctuations through linear-nonlinear collaborative mapping in the feature space, the model can distinguish systematic parameter drift (requiring equipment calibration) from random processing defects (requiring hole position repair), providing an interpretable decision basis for quality control. In the embodiment of the present application, in step S154, the blind hole image feature difference description coefficients between each blind hole image feature coding vector in the multiple blind hole image feature coding vectors and the blind hole image clustering coding vector are calculated respectively to obtain multiple blind hole image feature difference description coefficients, including: calculating the cosine distance between each blind hole image feature coding vector in the multiple blind hole image feature coding vectors and the blind hole image clustering coding vector as the blind hole image feature difference description coefficient. It should be understood that in the laser drilling of ultra-thin prepregs, blind hole defects (such as micron-level position offset, aperture out-of-tolerance, or residues) often appear as subtle non-linear anomalies in the population characteristics. Although the blind hole image feature coding vector extracts global commonalities (such as the morphological distribution of normal blind holes) through clustering analysis, traditional linear clustering methods are difficult to capture local non-linear deviations (such as hole wall deformation or uneven residual glue distribution caused by thermal stress).Therefore, it is necessary to calculate the difference description coefficient between each blind hole image feature coding vector and the blind hole image clustering coding vector to convert the abstract feature difference into a quantifiable numerical index. That is to say, since the determination of blind hole quality depends on the refined comparison of "group commonality" and "individual deviation", and simple clustering only provides a macroscopic framework and cannot directly locate the microscopic deviation degree of specific defects. Therefore, a dynamic difference criterion is constructed by calculating the difference description coefficient to achieve the precise positioning of defects. Specifically, the blind hole image clustering coding vector integrates the linear clustering center (representing the distribution of the main components of the group) and the non-linear compensation component (such as the depth collaborative implicit coding vector for modeling local complex relationships) to form a robust reference feature. By calculating the difference coefficient between each blind hole image feature coding vector and this reference, the system can quantify the degree of its deviation from the group characteristics. For example, using the feature clustering compensation increment operator combined with the energy relaxation mechanism to balance the linear and non-linear components, the difference description coefficient can not only reflect geometric deviations (such as hole position offset), but also sense process fluctuations (such as differences in hole wall roughness caused by uneven laser energy). This process is essentially an assessment of the "relative abnormality" of blind hole quality, avoiding misjudgment caused by fixed thresholds.

[0060] Specifically, in step S155, the blind hole corresponding to the largest one among the multiple blind hole image feature difference description coefficients is marked as a suspected defective blind hole. It should be understood that in the laser hole processing of ultra-thin prepregs, the abnormality degree of defective blind holes (such as micron-level position offset or residues) is often reflected by their deviation amount from the group characteristics. Although the blind hole image feature difference description coefficient has quantified the feature difference of each blind hole, the traditional threshold determination method is easily interfered by noise or process fluctuations, resulting in mislabeling or missing labeling. Therefore, the blind hole corresponding to the largest one among the multiple blind hole image feature difference description coefficients is marked as a suspected defective blind hole. In this way, the dynamic extreme value determination can be used to replace the fixed threshold to achieve the adaptability and precision of defect recognition. Specifically, based on the difference description coefficient generated by the feature gain aggregation analysis network, the system does not need to preset an artificial threshold, but automatically identifies the abnormal points that deviate most from the group commonality through the non-integer dimensional scattering characteristics of the fractal structure in the clustering space. For example, when the compensation increment operator makes a non-linear correction to the linear clustering result, the one with the largest difference coefficient often corresponds to the comprehensive manifestation of composite defects such as hole position offset, aperture out-of-tolerance, or residual glue residue. This determination method not only avoids the over-reliance of traditional methods on a single parameter (such as aperture), but also can capture the complex defect patterns of multi-dimensional feature collaborative anomalies, such as invisible hole wall cracks caused by the interlayer thermal stress coupling between copper foil and prepreg during laser drilling.

[0061] In summary, the laser hole processing method for the ultra-thin prepreg of printed circuit boards according to the embodiments of the present application is elucidated. It provides a PCB core board and manufactures an inner-layer circuit board, then performs a lamination process and laser brownification and laser processing for drilling to form multiple blind holes and conducts an automatic optical inspection, achieving an improvement in processing intelligence and controllability of quality risks. Among them, the laser processing step abandons the traditional multi-step hole repair mode and directly realizes the formation of blind holes with a diameter ≤ 75 μm through single-time high-precision laser copper breaking. The quality control step is based on the blind hole image data collected by the AOI system, automatically identifies abnormal blind holes deviating from the group characteristics, realizes the standardization and large-scale adaptation of defect detection, and assists manual inspection of whether the hole position is offset, whether the hole diameter meets the requirements, and whether there are residues or contaminants and other defects in the hole.

[0062] Figure 6 FIG. is a system block diagram of a laser hole processing system for an ultra-thin prepreg of printed circuit boards according to an embodiment of the present application. Among them, the laser hole processing system for an ultra-thin prepreg of printed circuit boards is used to execute the laser hole processing method for the ultra-thin prepreg of printed circuit boards as described above. Specifically, as Figure 6 shown, the laser hole processing system 100 for an ultra-thin prepreg of printed circuit boards according to an embodiment of the present application includes: a blind hole detection image acquisition module 110, configured to acquire a plurality of blind hole detection images collected by an AOI system; a blind hole image feature encoding module 120, configured to perform image feature extraction on the plurality of blind hole detection images to obtain a plurality of blind hole image feature encoding vectors; a blind hole image clustering module 130, configured to perform blind hole image feature clustering analysis on the plurality of blind hole image feature encoding vectors to obtain a blind hole image clustering encoding vector; a difference description coefficient calculation module 140, configured to calculate blind hole image feature difference description coefficients between each blind hole image feature encoding vector in the plurality of blind hole image feature encoding vectors and the blind hole image clustering encoding vector respectively to obtain a plurality of blind hole image feature difference description coefficients; and a defective blind hole determination module 150, configured to mark the blind hole corresponding to the largest one among the plurality of blind hole image feature difference description coefficients as a suspected defective blind hole.

[0063] Here, those skilled in the art can understand that the specific operations of each step in the above laser hole processing system for an ultra-thin prepreg of printed circuit boards have been introduced in detail in the description of the laser hole processing method for the ultra-thin prepreg of printed circuit boards above, and therefore, the repeated description thereof will be omitted. Figures 1 to 5 of the laser hole processing method for the ultra-thin prepreg of printed circuit boards, and thus, the repeated description thereof will be omitted.

[0064] As described above, the laser hole processing system 100 for ultra-thin prepreg of printed circuit board according to the embodiments of the present application can be implemented in various terminal devices. In one example, the laser hole processing system 100 for ultra-thin prepreg of printed circuit board can be integrated into the terminal device as a software module and / or a hardware module. For example, the laser hole processing system 100 for ultra-thin prepreg of printed circuit board can be a software module in the operating system of the terminal device, or can be an application program developed for the terminal device; of course, the laser hole processing system 100 for ultra-thin prepreg of printed circuit board can also be one of the many hardware modules of the terminal device.

[0065] Alternatively, in another example, the laser hole processing system 100 for ultra-thin prepreg of printed circuit board and the terminal device can also be separate devices, and the laser hole processing system 100 for ultra-thin prepreg of printed circuit board can be connected to the terminal device through a wired and / or wireless network, and transmit and interact information in accordance with a predefined data format.

[0066] In summary, the laser hole processing system for ultra-thin prepreg of printed circuit board based on the embodiments of the present application is described. It provides a PCB core board and manufactures an inner layer circuit board, then performs a lamination process, laser brownification, and laser processing for drilling to form multiple blind holes and performs automatic optical inspection, achieving an improvement in processing intelligence and controllability of quality risks. Among them, the laser processing link abandons the traditional multi-step hole repair mode and directly forms blind holes with a diameter ≤ 75μm through single-shot high-precision laser copper breaking. The quality control link is based on the blind hole image data collected by the AOI system, automatically identifies abnormal blind holes that deviate from the population characteristics, realizes the standardization and large-scale adaptation of defect detection, and assists manual inspection of whether the position of the hole is offset, whether the hole diameter meets the requirements, and whether there are residues or contaminants in the hole and other defects.

Claims

1. A method for laser hole processing of an ultra-thin prepreg for printed circuit boards, characterized in that, Including: Providing a PCB core board; Fabricating an inner layer circuit pattern on the PCB core board to obtain an inner layer circuit board; Pressing the inner layer circuit board, a prepreg layer, and an outer copper foil layer through a pressing process to obtain a multi-layer board pressing structure; After laser browning the multi-layer board pressing structure, using a laser beam to drill a plurality of laser holes in the outer copper foil layer and the prepreg layer of the multi-layer board pressing structure, and the plurality of laser holes form a plurality of blind holes for realizing electrical connection between different layer circuits; Performing automatic optical inspection on the plurality of blind holes.

2. The laser hole processing method for the ultra-thin prepreg of a printed circuit board according to claim 1, characterized in that The model of the prepreg layer is 1017PP or 1010PP, and in the multi-layer board pressing structure, the thickness of the prepreg layer after pressing is 18μm - 25μm.

3. The method for laser drilling of an ultra-thin prepreg for printed circuit boards according to claim 2, characterized in that, The aperture of the laser hole is less than or equal to 75μm.

4. The laser hole processing method for the ultra-thin prepreg of a printed circuit board according to claim 1, characterized in that, Performing automatic optical inspection on the plurality of blind holes includes: Obtaining a plurality of blind hole detection images collected by an AOI system; Performing image feature extraction on the plurality of blind hole detection images to obtain a plurality of blind hole image feature coding vectors; Performing blind hole image feature clustering analysis on the plurality of blind hole image feature coding vectors to obtain a blind hole image clustering coding vector; Calculating the blind hole image feature difference description coefficients between each blind hole image feature coding vector in the plurality of blind hole image feature coding vectors and the blind hole image clustering coding vector respectively to obtain a plurality of blind hole image feature difference description coefficients; Marking the blind hole corresponding to the largest one among the plurality of blind hole image feature difference description coefficients as a suspected defective blind hole.

5. The laser hole processing method for the ultra-thin prepreg of a printed circuit board according to claim 4, characterized in that Performing image feature extraction on the plurality of blind hole detection images to obtain a plurality of blind hole image feature coding vectors, including: respectively passing each blind hole detection image in the plurality of blind hole detection images through a blind hole state feature extractor based on a dilated convolutional neural network model to obtain the plurality of blind hole image feature coding vectors.

6. The method for laser hole machining of an ultra-thin prepreg for printed circuit boards according to claim 5, characterized in that, Performing blind hole image feature clustering analysis on the plurality of blind hole image feature coding vectors to obtain a blind hole image clustering coding vector, including: Performing linear clustering analysis on the plurality of blind hole image feature coding vectors to obtain a blind hole image feature initial linear clustering center coding vector; Performing clustering compensation on the plurality of blind hole image feature coding vectors to obtain a blind hole image feature linear clustering compensation component coding vector; Fusing the blind hole image feature linear clustering compensation component coding vector and the blind hole image feature initial linear clustering center coding vector to obtain the blind hole image clustering coding vector.

7. The method for laser hole machining of an ultra-thin prepreg for a printed circuit board according to claim 6, wherein Performing clustering compensation on the plurality of blind hole image feature coding vectors to obtain a blind hole image feature linear clustering compensation component coding vector, including: Constructing a blind hole image feature depth collaborative implicit coding vector between each blind hole image feature coding vector in the plurality of blind hole image feature coding vectors and the blind hole image feature initial linear clustering center coding vector; Based on the depth collaborative implicit coding vector of the blind hole image features, calculate the feature clustering compensation increment operator of each blind hole image feature coding vector in the multiple blind hole image feature coding vectors relative to the initial linear clustering center coding vector of the blind hole image features to obtain multiple blind hole image feature clustering compensation increment operators; Based on the multiple blind hole image feature clustering compensation increment operators, calculate the blind hole image feature linear clustering compensation component coding vectors of the multiple blind hole image feature coding vectors.

8. The laser hole processing method for the ultra-thin prepreg of a printed circuit board according to claim 7, wherein, Based on the depth collaborative implicit coding vector of the blind hole image features, calculating the feature clustering compensation increment operator of each blind hole image feature coding vector in the multiple blind hole image feature coding vectors relative to the initial linear clustering center coding vector of the blind hole image features to obtain multiple blind hole image feature clustering compensation increment operators includes: Perform activation processing on the initial linear clustering center coding vector of the blind hole image features to obtain the initial linear clustering center activation coding vector of the blind hole image features; Perform compensation based on the eigenvalue granularity on the initial linear clustering center activation coding vector of the blind hole image features and the depth collaborative implicit coding vector corresponding to the blind hole image feature coding vector at a predetermined position to obtain the intermediate compensation variable for blind hole image feature clustering; Perform non-linear resonance coupling correction on the intermediate compensation variable for blind hole image feature clustering to obtain the intermediate corrected compensation variable for blind hole image feature clustering; Perform normalization processing on the intermediate corrected compensation variable for blind hole image feature clustering to obtain the blind hole image feature clustering compensation increment operator corresponding to the blind hole image feature coding vector at a predetermined position.

9. The method for laser hole machining of an ultra-thin prepreg for printed circuit boards according to claim 8, wherein, Calculate the blind hole image feature difference description coefficients between each blind hole image feature coding vector in the multiple blind hole image feature coding vectors and the blind hole image clustering coding vector respectively to obtain multiple blind hole image feature difference description coefficients, including: calculating the cosine distance between each blind hole image feature coding vector in the multiple blind hole image feature coding vectors and the blind hole image clustering coding vector as the blind hole image feature difference description coefficient.

10. A laser hole processing system for ultra-thin prepreg of printed circuit board, which is used to execute the laser hole processing method for ultra-thin prepreg of printed circuit board as described in claim 1, characterized in that, The laser hole processing system for printed circuit board ultra-thin prepreg includes: A blind hole detection image acquisition module for acquiring a plurality of blind hole detection images collected by an AOI system; A blind hole image feature coding module for extracting image features from the plurality of blind hole detection images to obtain a plurality of blind hole image feature coding vectors; A blind hole image clustering module for performing blind hole image feature clustering analysis on the plurality of blind hole image feature coding vectors to obtain a blind hole image clustering coding vector; A difference description coefficient calculation module for calculating the blind hole image feature difference description coefficients between each blind hole image feature coding vector in the plurality of blind hole image feature coding vectors and the blind hole image clustering coding vector respectively to obtain a plurality of blind hole image feature difference description coefficients; A defective blind hole determination module for marking the blind hole corresponding to the largest one among the plurality of blind hole image feature difference description coefficients as a suspected defective blind hole.

Images