Printed circuit board solder resist method

Abstract

The application relates to a printed circuit board solder resist method, which realizes multi-dimensional physical mark embedding and automatic parameterized layout design by presetting distinguishable micro passive synchronous marks in the PCB design stage and combining optical phase coding. Before and after exposure, high-speed imaging detection and adaptive optical algorithms are used to collect and normalize the phase response of the multi-layer synchronous marks, extract multi-dimensional alignment reference features, establish a distributed alignment error model, distinguish the overall rigid displacement and local stress deformation, and based on the model, real-time correct the exposure equipment parameters to realize spatial error compensation.

Description

Technical Field

[0001] This invention relates to the fields of "printed circuit board solder mask technology, high-precision optical alignment and adaptive phase encoding technology", and more particularly to a printed circuit board solder mask method. Background Technology

[0002] In the current printed circuit board (PCB) manufacturing field, especially in the solder mask exposure and photomask alignment processes for multilayer PCBs, alignment accuracy directly affects product yield and reliability. As electronic devices evolve towards higher density and integration, multilayer PCB designs are becoming increasingly complex, with line widths and apertures shrinking, placing ever higher precision demands on the alignment of solder mask exposure between layers. Mainstream solder mask processes generally employ integrated photomask alignment and development devices. These devices use alignment targets pre-set on the substrate or mask, employing high-resolution imaging and precision mechanical motion systems to achieve the positioning and exposure of the solder mask. However, with the increase in the number of layers and the complexity of the circuitry, the process becomes significantly more sensitive to environmental disturbances.

[0003] Traditional technical routes mainly rely on high-speed mechanical alignment devices and target-based imaging recognition methods. Most devices use two-color markings, QR codes, or metal targets with recognizable shapes. By importing servo compensation algorithms through a vision system, automatic alignment of the mask and the substrate is achieved. At the same time, some high-end manufacturing lines have introduced environmental perception modules such as local temperature and humidity sensing and real-time vibration monitoring to trigger real-time fine-tuning compensation of the alignment equipment. However, existing photomask alignment processes usually have the following prominent problems: (1) Imaging recognition depends on the clarity of the markings. When there is a small amount of dust or impurities on the board surface, it is easy to misjudge the target; (2) The mechanical motion compensation response speed of the equipment is limited, making it difficult to completely eliminate the effects of instantaneous micro-vibration or thermal deformation during exposure and alignment; (3) Temperature and humidity changes can cause instantaneous expansion and contraction of materials. It is difficult to compensate consistently during mass production, which can easily lead to the spread of alignment errors between batches; (4) For multi-layer stacked structures, due to the influence of factors such as stress and local deformation of each layer, single-layer targets or simple visual alignment can no longer meet the requirements of high-precision multi-layer alignment.

[0004] In recent years, some advanced manufacturing enterprises have tried to introduce digital image correlation (DIC) analysis, multispectral recognition or passive insertion plastic markers to improve alignment robustness, but still face the following technical bottlenecks: (1) The number of active feedback and mechanical compensation modules has increased significantly, which has greatly increased the complexity and cost of the system and puts great pressure on equipment maintenance and production line scheduling; (2) The accuracy of real-time environmental parameter acquisition and control is limited, and it can often only be analyzed and compensated after the fact, making it difficult to achieve immediate and global perception and decoupling of disturbances; (3) Distributed multi-point monitoring and traditional target structures are difficult to balance the spatial resolution and anti-interference ability of high-density multi-layer areas.

[0005] Especially in high-end multilayer PCB solder mask production lines, the following technical challenges often arise: During the solder mask exposure process, even slight vibrations, temperature and humidity fluctuations, or localized dust accumulation can easily cause misalignment using traditional optical alignment methods, while the corrections from traditional real-time mechanical compensation systems are often insufficient to completely eliminate errors due to limitations in response and accuracy. On the other hand, due to the stacking differences in multilayer structures, a single physical target or marker structure cannot reflect the entire interlayer spatial state. During batch production, the environmental influence on alignment parameters often exhibits unpredictable random drift, resulting in a significant increase in the defect rate of some batches. Summary of the Invention

[0006] This application provides a solder mask method for printed circuit boards, which aims to solve one of the problems or issues of the prior art mentioned in the background section.

[0007] This application provides a method for solder mask application on printed circuit boards, specifically including:

[0008] S1: Based on PCB design data, miniature passive synchronization markers are pre-set in designated areas of each layer of the circuit board. Through optical phase encoding, distinguishable multi-layer alignment signals are formed to achieve the embedding of multi-dimensional physical identifiers.

[0009] S2: Before the solder mask is exposed, a high-speed imaging device is used to perform a global scan of the synchronous markings on the surface of the multilayer PCB and the mask, and to collect the original optical response signals that include environmental differences and vibration interference.

[0010] S3: Perform denoising and normalization preprocessing on the acquired raw optical response signal to eliminate invalid signal components caused by various environmental disturbances such as temperature, humidity, dust or minor deformation, and obtain standardized phase response data of multi-layer synchronous markers.

[0011] S4: Input the standardized phase response data into the adaptive optics processing algorithm, extract the features of the optical phase difference of each micro passive synchronization mark, and refine the multi-dimensional alignment reference feature set that reflects the working condition information such as physical offset and local deformation.

[0012] S5: Based on the above multidimensional alignment reference feature set, establish a distributed alignment error model of multilayer PCB under different environmental disturbance conditions, distinguish the spatial distribution characteristics of overall rigid displacement and local stress deformation, and realize the perception of physical differences.

[0013] S6: For the distributed alignment error model, an optical phase difference superposition algorithm is used to generate a virtual alignment plane for the current inter-layer alignment relationship column, so as to realize the difference compensation of alignment data under disturbance environment.

[0014] S7: Determine whether the current distribution alignment error model is within the preset threshold range. If it exceeds the threshold, automatically correct the alignment parameters of the exposure equipment and the mask pattern data according to the physical offset of each layer to achieve flexible parameter dynamic adaptation.

[0015] S8: During the solder mask exposure process, the actual exposure operation is completed based on the corrected alignment parameters and mask pattern data, and the current batch working condition label and environmental parameters are recorded for subsequent traceability.

[0016] S9: After exposure, the synchronous marker area is detected by re-imaging to obtain the phase response of the synchronous marker after actual exposure. The phase response is compared with the initial multidimensional alignment reference to calculate the final alignment residual and achieve dynamic feedback verification.

[0017] S10: Based on the dynamic feedback verification results, if the alignment residual does not meet the process requirements, an abnormal alarm or supplementary exposure process will be triggered. The distributed alignment error model and adaptive optics processing algorithm will be continuously optimized by recording residual information to improve the consistency and robustness of subsequent batch manufacturing.

[0018] The solder mask method for printed circuit boards provided in this application has the following beneficial effects:

[0019] (1) By embedding high-density, unique optical phase-encoded synchronization markers at the physical level, a multi-layered spatial reference distribution for alignment is effectively achieved. Unlike traditional methods that rely solely on mechanical positioning or single-point alignment markers, this invention, with its multi-dimensionally distributed optical phase signals, can accurately sense alignment error sources such as temperature and humidity fluctuations, equipment micro-vibrations, or dust particle interference leading to overall rigid movement and local stress deformation. The adaptive optics algorithm can automatically distinguish and compensate for the physical effects of various disturbances in real time. Experimental and batch data show that, compared to existing processes, this invention can reduce the standard deviation of alignment residuals by 30% to 60% and control the main error distribution within ±6μm, significantly exceeding the stability boundary achieved solely through real-time servo compensation.

[0020] (2) This invention employs a quantitative distribution error model and virtual alignment plane technology to normalize the complex working conditions under the coupling of the three factors of "plate-film-environment" into data-driven flexible alignment compensation, achieving high-consistency production without the need for frequent machine stop calibration or manual secondary calibration. The fully automated optical monitoring and compensation process reduces the average time of the alignment process by more than 20%, and reduces the batch yield fluctuation to less than 2%, significantly alleviating the batch instability problem caused by environmental changes and effectively improving the dimensional consistency and production of mass-produced products.

[0021] (3) Since the core of this system relies on optical signal recognition and distribution modeling, the requirements for traditional high-cost real-time mechanical compensation are greatly reduced; and all kinds of disturbances are reduced to the distributed alignment error model for unified processing, reducing the dependence on hardware such as equipment basic accuracy and environmental temperature control. It can achieve universal high-precision exposure in low-to-mid-end automated exposure lines and various complex layers and material systems, significantly expanding its applicability. Since error compensation is mainly achieved through data-driven and optical algorithms, the iteration and maintenance cycle of conventional equipment is extended, and the operation and maintenance costs can be expected to decrease by 10% to 30%.

[0022] (4) This invention automatically collects batch environmental tags, synchronously marks responses, and exposes results throughout the entire process, and uses residual data feedback to drive the continuous self-learning and optimization of the distributed error model and optical processing strategy, establishing a traceable big data system for the entire process-environment-product chain. For abnormal batches and specific operating conditions, the system can automatically attribute the cause, triggering local compensation or process parameter evolution, supporting adaptive generalization for subsequent similar process platforms, different layers / material systems / mask designs, effectively promoting the upgrade of the solder resist exposure field towards a self-evolving and intelligent production mode. Attached Figure Description

[0023] Appendix Figure 1 This is a main flowchart of a solder mask method for printed circuit boards.

[0024] Appendix Figure 2 This is a sub-flowchart of a solder mask method for printed circuit boards.

[0025] Appendix Figure 3 This is another sub-flowchart of a printed circuit board solder mask method. Detailed Implementation

[0026] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0027] The following disclosure provides many different embodiments or examples for implementing various structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. In addition, examples of various specific processes and materials are provided, but those skilled in the art will recognize the application of other processes and the use of other materials.

[0028] As attached Figure 1 As shown, this application provides a method for solder mask application on a printed circuit board, specifically including:

[0029] S1: Based on PCB design data, miniature passive synchronization markers are pre-set in designated areas of each layer of the circuit board. Through optical phase encoding, distinguishable multi-layer alignment signals are formed to achieve the embedding of multi-dimensional physical identifiers.

[0030] S2: Before the solder mask is exposed, a high-speed imaging device is used to perform a global scan of the synchronous markings on the surface of the multilayer PCB and the mask, and to collect the original optical response signals that include environmental differences and vibration interference.

[0031] S3: Perform denoising and normalization preprocessing on the acquired raw optical response signal to eliminate invalid signal components caused by various environmental disturbances such as temperature, humidity, dust or minor deformation, and obtain standardized phase response data of multi-layer synchronous markers.

[0032] S4: Input the standardized phase response data into the adaptive optics processing algorithm, extract the features of the optical phase difference of each micro passive synchronization mark, and refine the multi-dimensional alignment reference feature set that reflects the working condition information such as physical offset and local deformation.

[0033] S5: Based on the above multidimensional alignment reference feature set, establish a distributed alignment error model of multilayer PCB under different environmental disturbance conditions, distinguish the spatial distribution characteristics of overall rigid displacement and local stress deformation, and realize the perception of physical differences.

[0034] S6: For the distributed alignment error model, an optical phase difference superposition algorithm is used to generate a virtual alignment plane for the current inter-layer alignment relationship column, so as to realize the difference compensation of alignment data under disturbance environment.

[0035] S7: Determine whether the current distribution alignment error model is within the preset threshold range. If it exceeds the threshold, automatically correct the alignment parameters of the exposure equipment and the mask pattern data according to the physical offset of each layer to achieve flexible parameter dynamic adaptation.

[0036] S8: During the solder mask exposure process, the actual exposure operation is completed based on the corrected alignment parameters and mask pattern data, and the current batch working condition label and environmental parameters are recorded for subsequent traceability.

[0037] S9: After exposure, the synchronous marker area is detected by re-imaging to obtain the phase response of the synchronous marker after actual exposure. The phase response is compared with the initial multidimensional alignment reference to calculate the final alignment residual and achieve dynamic feedback verification.

[0038] S10: Based on the dynamic feedback verification results, if the alignment residual does not meet the process requirements, an abnormal alarm or supplementary exposure process will be triggered. The distributed alignment error model and adaptive optics processing algorithm will be continuously optimized by recording residual information to improve the consistency and robustness of subsequent batch manufacturing.

[0039] Step S1: Based on PCB design data, pre-place miniature passive synchronization markers in designated areas of each layer of the circuit board. Using optical phase encoding, distinguishable multi-layer alignment signals are formed to achieve the embedding of multi-dimensional physical identifiers. Specifically, this includes:

[0040] S1.1: Perform structured analysis on the PCB design data, extract the geometric stacking structure, solder mask distribution and alignment accuracy requirements of each layer in the target multilayer PCB, and output a list of preset physical identification areas.

[0041] The system takes a complete original design data file of a multilayer PCB as input and performs structured analysis on parameters such as layer stack-up information, solder mask distribution, and alignment accuracy requirements. A layer stack-up analysis algorithm (parameters: common PCB layout formats such as Gerber / ODB++) is used to automatically layer and extract geometric information from the multilayer PCB design. Furthermore, a rule-based solder mask identification method (parameters: solder mask area marking codes, design comments, and design rule checklists) is used to automatically extract the spatial distribution boundaries and usable areas of each layer's solder mask area, and outputs distributed solder mask area boundary coordinate data. Finally, using an alignment accuracy requirement library (parameters: standard tolerances, local high-precision alignment requirements, and layer stacking relationship matrix), a multilayer structure alignment accuracy grading comparison algorithm is executed to classify alignment accuracy levels for different line density areas and functional modules, and outputs grading labels. Furthermore, a spatial embeddability analysis algorithm (parameters: minimum interlayer safety spacing rule, solder mask / conductor area exclusion relationship) is employed to determine the available space for embedding physical identifiers within each layer, eliminating interference areas related to electrical and mechanical functions, and generating qualified candidate areas for physical identifiers. Through data aggregation and conflict avoidance algorithms (parameters: minimum spacing between adjacent layer identifiers, synchronization mark spacing threshold), these candidate areas are filtered to form a final list of preset physical identifier areas that meet alignment accuracy and safety specifications. Through the aforementioned structured parsing algorithm chain, PCB design data is transformed into layered, partitioned, and hierarchical physical identifier embedding base data, enabling high-precision alignment of multi-layer PCBs with underlying design data support.

[0042] For example, for a 12-layer High-Density Interconnect (HDI) PCB, its standard Gerber design file is imported, and the layer stack-up parameters n=12, minimum interlayer spacing 100μm, and solder mask detection tolerance ±30μm are set in the analytical algorithm. Through automated layer structure segmentation, the xy-dimensional boundaries and z-axis stack-up order of each layer are extracted. Solder mask coverage areas and window areas of each layer are located using solder mask identification rules, outputting a total of 108 typical solder mask coordinate blocks. A high-precision interval tolerance of 10μm and a standard interval tolerance of 30μm are set in the alignment accuracy database, and areas such as control chip BGA pins are labeled as high-precision embedding areas. During the embeddability analysis, candidate areas overlapping with high-current traces (line width greater than 0.3mm) and via areas (diameter greater than 0.2mm) are filtered out, forming 8-12 sets of usable identification embedding windows for each layer. Finally, a candidate region conflict filtering algorithm is used to eliminate overlapping candidate blocks with a cross-layer overlap of less than 200μm or a same-layer overlap of less than 300μm, outputting the final physical identifier preset region list (data entries: 65 blocks in 9 layers), laying the data foundation for subsequent embedding and phase encoding of synchronization tags. After this step, the system can quickly locate the optimal physical identifier embedding area in complex, high-density multilayer PCB structures, significantly improving the physical integrity and environmental adaptability of the alignment system.

[0043] S1.2: Based on the pre-defined list of physical identifier regions, a specific optical phase coding design algorithm is used to automatically allocate phase coding parameters for each layer of identifier regions, generating a unique set of synchronization mark parameters, and realizing the generation of multi-layer distinguishable synchronization signal modes.

[0044] For each candidate region in the pre-defined list of physical identifier regions, input the region's spatial coordinates, geometric dimensions, the distribution of the solder mask area in its layer, and the alignment accuracy level label.

[0045] An optical phase coding design algorithm is adopted (parameters: coding type selection [binary phase / multi-level phase / wavelength multiplexing], minimum distinguishable phase step Δφ=π / 8, minimum coding collision distance d). min =2, the maximum number of allocable codewords M = number of regions × number of layers), to realize the initial phase coding parameter allocation for each layer of the identifier region.

[0046] Furthermore, a physical inter-layer conflict avoidance algorithm (parameter: minimum codeword Hamming distance h) is used. min =3), combined with the sensitivity classification of the adjacent layer solder mask functional area, the phase code that may be misjudged or interfered with the signal in the codeword pool is automatically excluded, and the phase code of the adjacent layer and region is spatially mapped and judged to obtain the interlayer security codeword distribution matrix.

[0047] Furthermore, by employing multi-layer unique phase code generation strategy parameters: region center coordinates (Cx,Cy), layer number L, and encoding rule F(x,y,L), unique synchronization marker parameters are generated in batches for all physical identification regions according to inter-layer grouping and position sequence, thus forming a set of synchronization marker parameters.

[0048] A conflict detection concurrent allocation algorithm (parameters: minimum safe interval within the same layer / across layers, effective signal window width Tw) is adopted to perform global traversal and allocation consistency verification on the generated synchronization mark parameter set. It automatically identifies and compensates for phase code allocation anomalies caused by geometric proximity or functional overlap, and completes stable mapping of multi-layer distinguishable synchronization signal modes.

[0049] Through the above algorithm chain, the list of physical identifier preset areas is transformed into a unique set of synchronous marker parameters that can be directly applied to downstream synchronous marking layout design, optical integration inspection and phase decoding processes, so as to achieve a high degree of consistency between the alignment parameters of multilayer PCBs and the actual physical structure and rapid identification under environmental disturbances.

[0050] For example, in the identification embedding process of a 12-layer HDIPCB board, with 8 groups of valid physical identification areas per layer already determined, the actual number of synchronization tags to be allocated is 96. A multi-level phase coding strategy is adopted, with a φ step size of π / 8, allowing 16 basic codes to be encoded within one phase cycle. For spatially adjacent areas, a minimum Hamming distance of 4 is set. Basic codes are allocated using the spatial mapping algorithm F(x,y,L)=Mod(α·x+β·y+γ·L,16), combined with cyclical allocation using security codeword groups, resulting in 12 codeword sequences. After collision detection, for two adjacent identification areas with a geometric distance of 80μm, the algorithm automatically increases the codeword interval to avoid recognition overlap. Finally, a set of synchronization tag parameters of length 96 is output, each parameter containing a layer number, identification area coordinates, and a unique multi-level phase code. Under this parameter set, laboratory imaging recognition tests show that there is no cross-identification or omission in the encoding and decoding of all areas, and the false judgment rate is less than 0.2%, which significantly improves the uniqueness of the identification and the robustness to disturbances of the solder mask exposure alignment system under high-density multilayer PCB structures.

[0051] S1.3: Based on the set of synchronization marker parameters, select a suitable miniature passive synchronization marker structure (such as a miniature reflective structure or photochromic dot of a specific size), and generate a physical synchronization marker layout design file through parameterization using the PCB process database to realize the process link design of the marker physical structure.

[0052] S1.4: Integrate the physical synchronization mark layout design file with the original PCB stack-up layout, apply the integrated layout verification algorithm to verify design rule conflicts, spatial overlaps and signal interference issues during the synchronization mark embedding process, and output the optimal multi-layer physical synchronization mark integrated layout.

[0053] S1.5: Based on the final multi-layer physical synchronization mark integrated layout, generate a process execution work order corresponding to the solder mask exposure process. Through the automatic instruction issuance of the manufacturing process, guide the accurate embedding of the physical synchronization mark and the realization of optical phase encoding in the actual circuit board manufacturing process, and provide a unified physical base for subsequent high-speed imaging recognition and adaptive alignment algorithms.

[0054] Step S2: Before the solder mask is exposed, a high-speed imaging device is used to perform a global scan of the synchronous markings on the surface of the multilayer PCB and the mask, acquiring the original optical response signal including environmental differences and vibration interference. Specifically, this includes:

[0055] S2.1: Perform process positioning on the multilayer PCB sample to be exposed and its paired mask surface, determine the initial coordinate distribution of the micro passive synchronous marking area in three-dimensional space, and generate a spatial distribution benchmark for the synchronous marking.

[0056] Using the multilayer PCB sample to be exposed and its matching mask as the process object, the input is the spatial structure data of the multilayer PCB and the physical layout information of the mask, which have been integrated with micro passive synchronous markers.

[0057] A precision process positioning platform (parameters: six-degree-of-freedom fine-tuning displacement stage, laser interferometer with an accuracy δ≤1μm) is used to achieve the initial physical positioning of multilayer PCBs and mask samples in a global reference system.

[0058] Furthermore, the PCB and mask surfaces are spatially calibrated using a two-dimensional and three-dimensional machine vision positioning system (parameters: high-resolution industrial camera, structured light / laser triangulation). The global physical coordinate data of each marker in the set of all synchronous marker parameters is extracted, and the spatial coordinates (x, y, z) of each synchronous marker area are calculated by the fusion of the positioning camera and the ranging unit.

[0059] Furthermore, a multi-layer physical identifier region spatial mapping algorithm is applied (parameter: inter-layer positioning transformation matrix (T)). l (Feature corner point matching tolerance ε=2μm), projecting the mask and the synchronization marks on each layer of the PCB onto a unified three-dimensional coordinate reference system to achieve spatial alignment of multi-layer synchronization marks.

[0060] Furthermore, the initial three-dimensional coordinate distribution of the synchronous marker area is written into the synchronous marker spatial distribution reference data structure to form a standard spatial positioning template that can be used for subsequent high-speed imaging scanning and optical data decoding.

[0061] Through the above positioning, calibration and fusion operations, the spatial positions of the micro passive synchronous marking areas on different layers and different templates are uniquely mapped to the process reference system, ensuring the physical consistency and integral alignment of the data source for subsequent imaging acquisition and phase analysis, and realizing the generation of alignment reference spatial data for complex multilayer systems before solder resist exposure.

[0062] For example, for a 12-layer HDIPCB and its custom mask, a 3D positioning system equipped with an industrial camera with a resolution of 0.5μm and adaptive structured light projection was used to place the PCB and mask on a six-axis precision process positioning stage with a travel range of 250mm×250mm×50mm. A total of 96 sets of actual synchronous markers were recorded, including their layer number and spatial center coordinates (x, y, z). With 200 frames acquired, a spatial mapping algorithm was used to complete the 3D point cloud registration of all markers, with a maximum inter-layer positioning error of 1.2μm. The x, y, z spatial coordinates and layer number of all synchronous markers were stored as a synchronous marker spatial distribution reference index table. For 10 boards tested in the same batch, the repeatability positioning error was less than 2μm, verifying that this step can provide a high-precision, stable, and consistent spatial reference for optical scanning and phase feature recognition in the context of multi-layer high-precision processes, thereby improving the spatial robustness of the alignment process in complex multi-layer PCB solder mask processes.

[0063] S2.2: Based on the spatial distribution benchmark of synchronous markers, a high-speed imaging device is used to synchronously acquire the original optical image signal of the micro passive synchronous marker area on the surface of multilayer PCB and mask, so as to obtain high-resolution original optical image data containing physical position, reflection / refraction difference and relative relationship of multilayer markers.

[0064] S2.3: The high-resolution raw optical image data is read in parallel through a multi-channel mechanism, and an automatic marker recognition algorithm and ROI (Region of Interest) extraction process are executed to accurately locate the pixel region and geometric contour of each synchronous marker and output the pixel set of the synchronous marker region.

[0065] S2.4: Using a phase encoding and decoding algorithm, the original phase response values ​​of the synchronization markers under the current conditions of light, temperature, humidity and random vibration are extracted from the pixel set of the synchronization marker region to generate the original optical phase response matrix.

[0066] S2.5: The original optical phase response matrix is ​​labeled and associated with the synchronously marked spatial distribution benchmark and environmental parameters (including real-time data from sensors such as temperature, humidity, and acceleration) to synthesize a batch operating condition optical response dataset to support subsequent denoising, normalization, and disturbance feature analysis processes.

[0067] Step S3: Perform denoising and normalization preprocessing on the acquired raw optical response signal to eliminate invalid signal components caused by various environmental disturbances such as temperature, humidity, dust, or minor deformations, and obtain standardized phase response data of multi-layer synchronization markers. Figure 2 As shown, it specifically includes:

[0068] S3.1: Perform baseline correction processing on the raw optical response signal output by the high-speed imaging device to eliminate global amplitude shift caused by imaging system gain drift or light source fluctuation, obtain a baseline correction signal, and ensure signal consistency in subsequent processing.

[0069] S3.2: Based on the spatial domain filtering algorithm, background noise suppression is performed on the baseline correction signal to filter out random noise components caused by ambient light interference, stray light from optical components, etc., and output a noise suppression signal to improve the signal-to-noise ratio of the synchronization mark response signal.

[0070] The input is the optical response signal after baseline correction processing, and the target is the baseline correction signal set obtained by the synchronous marking areas on the surface of the multilayer PCB and mask under the high-speed imaging system.

[0071] A spatial filtering algorithm (parameters: two-dimensional Gaussian filter kernel σ = 1.5, mean filter window size n = 7 × 7) is used to perform whole-frame convolution processing on the baseline correction signal, thereby achieving preliminary suppression of high-frequency random noise components in the imaging signal and effectively reducing the local noise level caused by ambient stray light and non-uniform response of imaging equipment.

[0072] Furthermore, by using a spatial correlation enhancement algorithm (parameter: correlation threshold τ = 0.85), the intensity change trend of adjacent pixel blocks within the synchronous marker ROI region is analyzed. Spatial adaptive smoothing constraints are applied to pixels with low correlation and abrupt changes, further reducing the influence of non-structural noise sources such as dust and foreign spots on the optical lens surface, and preserving the true response morphology of the synchronous marker subject.

[0073] Furthermore, a multi-scale background modeling algorithm (parameters: number of pyramid layers L=3, background template update weight λ=0.1) is applied to filter out background components in the slowly varying regions of imaging from bottom to top, distinguishing between synchronous marker response and slowly varying ambient light interference, thereby achieving effective subtraction of low-frequency background noise.

[0074] Furthermore, a local signal-to-noise ratio (SNR) discrimination algorithm is used on the filtered output signal to dynamically verify the validity of the synchronization marker signal within the ROI using the following formula:

[0075]

[0076] in, Let be the mean value of the filtered signal within the i-th synchronization marker region. This represents the variance of the neighborhood background noise.

[0077] For regions where the SNR is below a preset threshold (SNR<12dB), signal enhancement compensation techniques (such as weight boosting or local resampling) are used to improve the stability of the response intensity.

[0078] Through the above-mentioned multi-level spatial filtering and background suppression algorithms, the baseline correction signal is transformed into a noise suppression signal, and the output synchronization mark optical response data with high signal-to-noise ratio and significantly reduced background stray light components is generated. This lays a high-quality data foundation for subsequent chain preprocessing steps such as temporal denoising, spatial weighting, and normalization, and improves the ability to effectively extract multi-level alignment synchronization mark information.

[0079] For example, for 12-layer HDIPCB solder resist exposure samples, a high-speed industrial camera with a resolution of 2048×2048 pixels was used to acquire optical signal frames of the synchronous marking area. Under conditions of ±3℃ temperature fluctuation in the exposure chamber and 1700±150 lux variation in ambient illuminance (stray light), a Gaussian filter with σ=1.5 and a mean window of n=7×7 were used for spatial filtering of each synchronous marking image, reducing the background noise standard deviation from the original 18gray level to 5gray level. After introducing multi-scale background modeling, slowly varying low-frequency components caused by exposure lamp reflection and local equipment shadows were further eliminated. SNR discrimination results showed that after filtering, the mean signal intensity of each synchronous mark increased by more than 30%, and the SNR range within the ROI increased from the original 9–14 dB to 15–22 dB. For areas with SNR below 12 dB, the stability of the output signal of each synchronous mark was significantly enhanced after weighted enhancement. In batch sample testing, this method achieves a false detection rate of less than 0.6% for synchronization markers, effectively supporting high accuracy and environmental robustness in subsequent processes such as fine recognition of synchronization markers, phase decoding, and distributed alignment error modeling. The output noise-suppressed signal is directly used as the standard input for subsequent S3.3 adaptive time-domain denoising processing, ensuring that the optical feature input source of the PCB multilayer alignment reference has excellent stability and a high signal-to-noise ratio.

[0080] S3.3: Adaptive temporal denoising technology is used to detect and remove abnormal time points corresponding to disturbances such as sudden changes in environmental temperature and humidity, dust projection, and local micro-deformation in the noise suppression signal, forming a noise removal signal and enhancing robustness to occasional physical disturbances.

[0081] The input is a noise-suppressed signal after spatial filtering and background modeling. The target is high signal-to-noise ratio optical response sequence data acquired by a high-resolution imaging system, which is used to simultaneously mark the multilayer PCB and mask surfaces.

[0082] An adaptive temporal denoising algorithm (parameters: sliding window length w = 16 frames, dynamic threshold α = 3.0, outlier detection probability p = 0.01) is used to perform frame-by-frame temporal sliding analysis on the response signal sequence within each synchronization marker ROI region, thereby achieving automatic detection of temporal abrupt disturbances.

[0083] Furthermore, by using a multi-level change point detection method (parameters: first-order difference detection window Δw = 4, local extremum detection ratio β = 0.25), the short-time mean and variance of the sequence signal are dynamically monitored, and points of sudden response changes and abnormal pulses (including sudden jumps caused by sudden changes in temperature and humidity, local dust shadows, and stress deformation) in the sequence are identified and marked.

[0084] Furthermore, a robust outlier confidence calculation model is adopted to apply discriminative weights to time series points marked as outliers, as specifically implemented as follows:

[0085]

[0086] Among them, S i Let μ be the signal strength of the synchronization marker for the i-th frame. local σ local These are the mean and standard deviation within the local sliding window, respectively, P. outlier (i) represents the confidence probability of an outlier in frame i.

[0087] For confidence probability threshold P outlier For signal points (i)>p, abnormal disturbance points are removed or corrected through data interpolation, sequence smoothing or direct elimination strategies to ensure that the continuity of the signal sequence is not disrupted by occasional disturbances.

[0088] Furthermore, by combining multi-dimensional batch environmental monitoring data (such as real-time temperature and humidity change rates and acceleration fluctuation amplitude), the occurrence time of anomalies is correlated with external disturbance events. A dynamic elimination window expansion algorithm (parameter: expansion factor γ = 1.5) is adopted to increase the noise reduction of anomalies in high-incidence disturbance areas, effectively addressing large-area signal shifts caused by occasional environmental shocks or equipment resonance during batch production.

[0089] Through continuous adaptive temporal denoising processing, the noise suppression signal is transformed into a "noise removal signal" output, providing stable and reliable synchronous marker response data that has been cleaned by multiple perturbations for subsequent spatial weighting and normalization steps, thereby enhancing the environmental robustness of the multilayer PCB solder mask alignment signal.

[0090] For example, for the optical response sequence of multilayer synchronization markers on a 12-layer HDIPCB, with 320 frames sampled, the proportion of outliers in the temporal domain increased to 4% during periods of temperature abrupt change (batch temperature rose from 24℃ to 28℃, lasting 15 frames) and equipment vibration (peak acceleration 0.045g). Under the adaptive temporal denoising algorithm, with a window length w = 16 and a dynamic threshold α = 3.0, outliers filtered out by the confidence criterion accounted for 3.8% of all data points. The sequence continuity index (SL) after interpolation and correction improved to 0.97. After batch output of the noise-removed signals, the misjudgment rate of subsequent spatial weighting processing decreased to 0.3%. Verification results show that under the background of temperature and humidity abrupt changes and equipment disturbances, this step can improve the robustness of the effective synchronization marker response signal by more than 30%, significantly suppressing the interference of occasional physical disturbances on the alignment signal data stream, and providing data assurance for the stable extraction of highly reliable multilayer optical alignment features.

[0091] S3.4: Perform multi-layer synchronization tag differentiation and weighting processing on the noise-removed signal. Based on the preset spatial distribution and physical coding characteristics of each synchronization tag, assign different weights to highlight the effective synchronization tag signal, obtain the spatial weighted signal, and realize the spatial decoupling of the synchronization tag signal.

[0092] The input is a noise-removed signal processed by an adaptive time-domain denoising algorithm, which includes high signal-to-noise ratio optical response sequence data of each synchronously marked area on the surface of the multilayer PCB and mask.

[0093] A multi-layer region differentiation weighted processing algorithm (parameters: spatial distribution index table, encoding weight parameter set (wk)) is adopted to identify the hierarchy, physical location and encoding type of each synchronization marker pixel region data, so as to realize the differentiated distinction of multi-layer synchronization marker signals.

[0094] Furthermore, by using a synchronous marker spatial distribution weight allocation method (parameters: physical coordinate index ((xk,yk,zk)), hierarchical label l, encoding method e), a differentiated weight (wk) is automatically assigned based on the spatial geometric distribution of each marker within the PCB stack-up structure and the physical signal sensing intensity, thereby enhancing the significance of the effective synchronous response signal at the physical base point.

[0095] Furthermore, a multi-dimensional physical coding fusion weighted algorithm is adopted (parameters: coding discrimination degree, interlayer spatial interference threshold). For situations where there is signal interference between high-density layers or adjacent distribution areas, specific compensation or suppression weights are assigned to suppress spatial signal aliasing caused by local coupling or noise leakage, thereby improving the independence of the synchronization marker distinguishing signals.

[0096] Furthermore, the spatially weighted signal matrix (S) is realized through the following weighting calculation formula. weighted )generate:

[0097] S weighted (k)=w k ·S clean (k)

[0098] Among them, S weighted (k) represents the weighted k-th synchronization marker response signal, w k Its spatial distribution and physical encoding related weights, S clean (k) is the corresponding noise removal synchronization marker signal.

[0099] Furthermore, a spatial decoupling discrimination algorithm is applied to perform spatial autocorrelation analysis and linear independence evaluation on the weighted signal matrix, using the following autocorrelation coefficient matrix criterion:

[0100]

[0101] Among them, R ij Let Cov(·) be the correlation coefficient between weighted signals with different synchronization markers, and σ be the covariance. i For signal

[0102] S weighted The standard deviation of (i). For R ij Pairs of labels close to zero are considered to achieve effective spatial decoupling.

[0103] Through the above-mentioned multi-layer differentiation weighting and spatial decoupling processing, the signal strength and the spatial physical attributes and coding characteristics of the synchronization mark are organically integrated to realize the spatial differentiation and decoupling processing of multi-layer signals, output spatial weighted signals, create conditions for subsequent normalization processing and multi-dimensional alignment reference construction, and realize the data independence and physical criterion consistency of multi-layer resist welding alignment signals.

[0104] For example, in a 12-layer HDIPCB solder mask exposure process scenario, the input is a multi-layer synchronous marker response signal processed by a time-domain denoising algorithm, and the spatial distribution weight is set to the highest layer w. top =1.2, intermediate layers decrease to w according to the stacking density. mid =1.0, bottom layer w bottom =0.8. For edge regions or markers with encoding interference, the compensation factor δ should be set as appropriate. e =0.85, mutual interference reduction weight η =0.75. After weighting using the above algorithm, the signal spatial decoupling autocorrelation coefficient R ij The mean value was reduced to 0.13, and the boundary layer separation was improved by 20%. The batch output spatial weighted signal had improved effective component discrimination. As the input source for subsequent normalization transformation, after randomized batch testing, the standard deviation of the positional residual distribution was reduced by 18% within the temperature and humidity fluctuation range, which significantly improved the decoupling capability and spatial discrimination accuracy of multilayer synchronous markers.

[0105] S3.5: The spatially weighted signal is subjected to intensity and phase normalization transformation using the normalization transformation method to eliminate the multi-layer synchronization mark response deviation caused by systematic errors such as uneven exposure energy and imaging gain, and outputs normalized phase response data as the sole input for subsequent feature extraction and alignment reference construction.

[0106] Step S4: The standardized phase response data is input into an adaptive optics processing algorithm. By extracting features from the optical phase difference of each micro passive synchronization marker, a multi-dimensional alignment reference feature set reflecting working condition information such as physical offset and local deformation is extracted. For example... Figure 3 As shown, it specifically includes:

[0107] S4.1: Input the standardized phase response data of the multi-layer synchronization markers into the adaptive optics processing algorithm module as a unified input basis for subsequent optical feature analysis, ensuring consistent mapping of the data structure for the synchronization marker response data of all layers.

[0108] The input is multi-layer synchronous marker standardized phase response data output after S3.5 normalization processing. The data structure is a multi-point region signal set covering all PCB layers and mask surfaces, layered organization, and after integrated amplitude and phase calibration.

[0109] A structured data loading method (parameters: multi-level index table, coded tag set) is adopted to batch import the standardized phase response data of all synchronization tags into the adaptive optics processing algorithm module, complete the collection and mapping of basic data structures, and form a unified input.

[0110] Furthermore, through a multidimensional data consistency check algorithm (parameters: data integrity threshold θ = 0.99, missing data filling strategy: interpolation mode - linear, recent mean), the missing values, duplicate values ​​and outliers of the imported data are automatically checked, structurally abnormal data is removed or local missing areas are filled, ensuring that all synchronous marker response data are structurally consistent in space, hierarchy and coding domain.

[0111] Furthermore, an input feature normalization mapping method (parameters: normalization standard μ = 0, σ = 1, selectable normalization interval [0,1]) is adopted to perform overall hierarchical normalization processing on the standardized phase response data of all labels, ensuring that the amplitude and phase distribution of the input data are restricted to a uniform scale domain, and eliminating the scale difference of signal input between different levels and different physical characteristics.

[0112] Furthermore, through the interlayer spatial index reconstruction algorithm (parameters: physical coordinate mapping matrix Λ, layer marker l), a unified mapping of each synchronization marker in its physical embedding layer, spatial coordinates and mask space is achieved, and the original input signal set is reassembled into a multi-dimensional array structure (Index: layer number x marker type x spatial position), providing an efficient data scheduling interface for subsequent optical feature analysis algorithms.

[0113] Through the above chain-like data collection and structure mapping processing, the standardized phase response data of multi-layer synchronous markers was input into the adaptive optics processing algorithm in a complete, standardized and highly consistent manner, which established a unified data foundation for subsequent technical steps such as multi-dimensional feature extraction, phase difference analysis and alignment reference construction.

[0114] For example, in a 12-layer HDIPCB solder mask process batch, the input is a normalized phase response data matrix with 12 synchronous markers per layer, totaling 144 markers, with a sampling resolution of 2048×2048 pixels. Using structured data loading, the system automatically completes data loading within 0.4ms. Through consistency checks, two markers were found to be suspicious, which were recovered by interpolation using the current batch mean. The normalization process resulted in the amplitude range of each layer being unified to [0,1], with an inter-batch phase mean μ = 0 and a standard deviation σ = 0.41. Spatial index reconstruction outputs a multi-dimensional array structure, with the index structure being [layer = 12, marker = 12, region = 12, area ...

[0115] , the full data mapping efficiency is better than 10μs / time. The final output is standardized and structured input data, which provides a standardized and efficient underlying data foundation for feature extraction, phase difference calculation and distributed alignment error modeling in S4.2 and beyond, effectively ensuring the process consistency and environmental robustness of multilayer high-precision solder resist exposure process.

[0116] S4.2: The phase difference is calculated using a phase difference analysis algorithm on the standardized phase response data of each layer of synchronization markers to obtain the optical phase difference characteristic parameters of the micro passive synchronization markers in the PCB board-mask system at different physical locations.

[0117] The input is the standardized phase response data of all synchronous marks of the multilayer PCB and paired masks, which are processed by S3.5 normalization. The data structure has hierarchical differentiation, physical space labeling and integrated amplitude-phase calibration characteristics, serving as the unified analysis basis for adaptive optics processing algorithms.

[0118] A phase difference analysis algorithm with multi-layer region grouping (parameters: layer number grouping table L, synchronous marker physical coordinate set M) is adopted to realize the pairwise traversal of all synchronous marker points of each layer and its corresponding mask surface, and construct an analysis index under the dual constraints of physical space coordinates and marker encoding.

[0119] Furthermore, the core unit (parameter: normalized phase response vector) is calculated through the phase difference. For each pair of physical corresponding synchronization mark regions, the phase difference between the synchronization marks of each PCB layer and the mask system is quantitatively calculated according to the following formula:

[0120]

[0121] in, This represents the normalized optical phase difference between the k-th synchronization mark in the l-th layer and the PCB and mask. This represents the phase response of the k-th synchronization marker on the current PCB layer. This represents the phase response of the synchronous markers corresponding to the same area on the mask surface.

[0122] Furthermore, through the phase difference statistics module (parameters: environmental disturbance label E, region resolution threshold τ) r Batch statistics and spatial clustering were performed on the phase difference data of all physically corresponding marker points to analyze the overall offset, local anomalies, and multi-layer collaborative change patterns, forming a spatially significant phase difference feature matrix:

[0123]

[0124] Where, N l This represents the total number of identifiable synchronization markers in layer l.

[0125] Furthermore, a spatial consistency constraint detection algorithm is adopted (parameters: autocorrelation threshold γ, physical neighborhood size d). n The spatial consistency of the phase difference distribution of synchronous markers in the same or adjacent layers is checked, and markers are applied to the detected spatial abrupt changes.

[0126] An asynchronous anomaly filter is used, combined with environmental labels such as real-time temperature and humidity and stress disturbance, to physically attribute the detected local phase difference abrupt changes and label them as rigid overall displacement or local non-uniform deformation characteristics.

[0127] Through the above-mentioned multi-level phase difference analysis and spatial statistical-attribution processing, the standardized phase response data is efficiently transformed into a set of physical offset and micro-area deformation phase difference characteristic parameters of multi-layer and multi-region, realizing the structured and quantifiable characterization of physical state information in the multi-layer PCB alignment system.

[0128] For example, for a 12-layer HDIPCB solder mask process sample, the input consists of 144 synchronized marker-normalized phase response data vectors covering all batch layers, with single-point phase noise less than 0.05 radians. Region pairwise analysis is used to calculate the phase response of each marker pairwise. The intralayer phase difference distribution range was obtained in radians [-0.02, 0.13], with an outlier ratio of less than 0.8%. Spatial clustering algorithm automatically identified three concentrated local anomaly areas. Environmental label analysis showed that the temperature of this batch increased by 3.2℃ during exposure, and the phase difference in the corresponding area increased significantly, which was determined to be localized thermal expansion and contraction non-rigid deformation. After processing, a multi-dimensional layered phase difference matrix was output as the standard input for physical offset and deformation perception, effectively supporting subsequent physical offset modeling and alignment reference feature extraction. The above processing features high efficiency, low false positives, and accurate physical attribution in the batch automated optical inspection process, improving the process adaptability and stability of solder resist alignment.

[0129] S4.3: Based on the acquired optical phase difference feature parameters, a multi-dimensional feature extraction algorithm is used to generate feature vectors, thereby achieving a preliminary distinction between physical offset (overall rigid body movement) and local deformation (such as stress-induced micro-area bending and non-rigid body torsion caused by thermal expansion and contraction).

[0130] The input conditions are the normalized phase response data of all synchronous marks on the multilayer PCB and mask surface after normalization. The data structure has hierarchical distinction and spatial physical label attributes.

[0131] A multi-dimensional feature extraction algorithm (parameters: phase difference feature matrix, hierarchical space mapping parameter table) is adopted to realize the function of generating feature vectors for all synchronization tags in hierarchical, spatial and encoding dimensions.

[0132] Furthermore, by using the principal component extraction method (parameter: cumulative contribution rate threshold 98%), feature components representing the main differences are extracted from the full phase difference feature parameter space to generate an initial principal component feature vector group, and its mapping weight in the overall phase difference feature space is obtained, which facilitates subsequent feature sensitivity analysis for different physical states.

[0133] Furthermore, a rigid transformation parameter estimation method (parameter: multi-layer rigid affine model) is adopted. Based on the initial feature vector group, the overall rigid body movement feature components are separated, the average translation and rotation angle of the multi-layer region synchronous markers in physical space are calculated, the main components of the overall rigid displacement are quantified, and a feature subset representing the physical offset (overall movement) is formed.

[0134] Furthermore, a local residual analysis algorithm (parameters: local neighborhood range, non-rigid residual threshold = 0.02) is used to perform spatial residual decomposition on the remaining features after the rigid components have been removed, extracting nonlinear deformation features manifested as local stress, micro-area bending and thermal expansion and contraction, generating a subset of local deformation features, and achieving accurate separation and preliminary attribution of various non-rigid disturbances.

[0135] Furthermore, a feature normalization fusion algorithm (parameters: amplitude normalization domain [-1,1], normalization weight set) is adopted to normalize and concatenate rigid feature subsets and non-rigid feature subsets into structured multidimensional feature vectors, thereby achieving efficient and unified encoding of all physical offsets (overall rigid body movement) and local deformations (micro-area bending, non-rigid torsion, stress response, etc.) information.

[0136] Through the above-mentioned multi-level feature extraction, decomposition and fusion chain, the phase difference feature matrix is ​​systematically transformed into a multi-dimensional feature vector reflecting the physical offset and local deformation state of multi-layer PCBs, realizing a structured, quantifiable and hierarchical representation of physical working condition information, and providing a technical foundation for the construction of alignment reference and the distribution of alignment error model input.

[0137] For example, for a batch of 12-layer HDIPCB solder mask exposure process, the input is a normalized phase difference feature matrix covering 144 synchronous markers across all layers. Principal component extraction is set with a cumulative contribution rate threshold of 98%, extracting the first 5 principal components with weights of 0.42, 0.23, 0.17, 0.09, and 0.07, respectively. Rigid affine transformation parameter estimation yields an overall translation of [18.2, 14.7] and a rotation angle of (0.042) degrees. In local residual analysis, the maximum non-rigid residual value within the neighborhood is 0.019, meeting the parameter threshold. Local deformation features are located in three high-temperature sensitive regions (each approximately 220 μm in length). Finally, all feature components are normalized and output to form a multi-dimensional feature vector of length 8, which serves as the standard input for subsequent multi-layer alignment error modeling. In actual operation, the principal components of this multi-dimensional feature vector exhibit high sensitivity to various physical disturbances, accurate physical attribution, and batch-to-batch repeatability and consistency exceeding 98%.

[0138] S4.4: Aggregate the phase difference multidimensional feature vectors and use the heterogeneous marker feature fusion model to weight the working condition related parameters, so as to highlight the alignment reference features of physical offset and local deformation reflected by each micro passive synchronous marker under different spatial distribution positions and sensing levels.

[0139] S4.5: Output the multi-dimensional alignment benchmark feature set after comprehensive weighted fusion, which serves as the standard input for the subsequent steps of establishing the distributed alignment error model, thereby improving the ability to perceive physical differences under the influence of environmental disturbances and efficiently transmitting alignment compensation data.

[0140] Step S5: Based on the above multidimensional alignment reference feature set, establish a distributed alignment error model of multilayer PCB under different environmental disturbance conditions, distinguish the spatial distribution characteristics of overall rigid displacement and local stress deformation, and realize the perception of physical differences.

[0141] Specifically, it includes:

[0142] S5.1: Using a multidimensional alignment reference feature set as input, the spatial feature analysis algorithm is called to reconstruct the spatial distribution of the standardized phase response data of each layer of micro passive synchronization markers, generating a spatial phase distribution map to obtain a feature base containing physical offset and deformation information.

[0143] S5.2: Based on the spatial phase distribution map, the rigid affine transformation analysis method is applied to separate the rigid offset parameters that reflect the overall rigid displacement of the multilayer PCB board, extract the total translation and rotation between each layer, and form an overall rigid displacement parameter set to characterize the overall board displacement caused by container-level environmental disturbances.

[0144] The input data is a spatial phase distribution map generated by step S5.1. This map has completed the spatial location mapping and physical state normalization of the synchronization marks based on the multidimensional alignment reference feature set, and has a standardized phase response array structure covering all synchronization mark areas of the multi-layer PCB.

[0145] The rigid affine transformation analysis method (parameters: affine transformation matrix A1, target layer number l) is adopted to realize the overall modeling of the physical coordinates and phase response of all synchronization markers in the spatial phase distribution map of each layer, and the spatial transformation of each layer is classified as rigid movement (including translation and rotation).

[0146] Using the least squares fitting algorithm (parameter: labeled coordinate pair {X... i ,Y i} and the coordinate pair of the reference frame {X i,ref ,Y i,ref This involves estimating the rigid affine parameters of the synchronously labeled point cloud, specifically solving for the global translation vector and rotation matrix of the target layer relative to the reference layer (or mask layer). The rigid transformation satisfies the following formula:

[0147]

[0148] Among them, R l Let t be the rigid rotation matrix of the l-th layer. l For a rigid translation vector, (x) i ,y i (x) represents the spatial coordinates of the i-th synchronization marker in this layer. i ′,y i ′) is its coordinate after being mapped to the reference frame.

[0149] Furthermore, the optimal rigidity matching of the covariance matrix is ​​solved using the singular value decomposition (SVD) method, and the global rotation angle θ between each layer is quantified. l Translation component (ΔX) l ,ΔY l To obtain the overall rigid displacement parameter set:

[0150] θ l ,ΔX l ,ΔY l

[0151] This parameter set describes the overall rotation and translation state of each layer of a multilayer PCB.

[0152] Furthermore, spatial consistency is checked using rigid affine parameters (parameter: threshold ∈ rigid This process verifies the fitting residual distribution of the overall rigid displacement parameters, screens for large-scale overall mismatch or deformation trends of the foundation plate, and ensures the effectiveness and representativeness of the rigid displacement parameters.

[0153] Through the above rigid affine transformation analysis, the phase distribution map reconstructed by S5.1 is transformed into rotation-translation parameters that clearly characterize the overall rigid motion of the multilayer plate. This establishes a physical quantitative model for the multilayer synchronous displacement caused by container-level environmental disturbances (such as overall temperature changes and the movement of the equivalent rigid body of the equipment), and effectively decouples the rigid displacement characteristics from subsequent local deformation modeling (S5.3), error decomposition, and dynamic compensation.

[0154] For example, in a 12-layer high-end HDIPCB solder mask alignment batch, the spatial phase distribution map input covers 12 synchronously marked physical coordinates and standardized phase responses for each layer. Using rigid affine transformation analysis, the rotation angle range between each layer and the design reference layer or mask reference plane is fitted to [-0.03, 0.07] degrees, and the global translation component range is ΔX = [-12μm, +15μm], ΔY = [-10μm, +13μm], with SVD residuals all better than 2μm. For a certain layer in the batch where the ambient temperature was 2.5℃ higher than the average and overall warping occurred, the rigid rotation parameter of that layer was significantly increased, and the translation residual after fitting increased. The rigid offset parameter set obtained from this analysis serves as a key input for subsequent physical error decomposition (S5.3-S5.5) and real-time parameter compensation (S7), significantly improving the ability to perceive, discriminate, and compensate for disturbances across the entire PCB board and multi-layer container levels.

[0155] S5.3: Based on the overall rigid displacement parameter set, a local stress field modeling algorithm is used to perform secondary residual analysis on the spatial phase distribution map, obtain the small nonlinear offset between each synchronization mark, refine the stress distribution pattern at each local location, and output the local stress deformation parameter set to achieve quantitative characterization of physical stress disturbance.

[0156] Using the overall rigid displacement parameter set as input, local stress field modeling and analysis are performed on the spatial phase distribution pattern of multilayer PCBs.

[0157] The local residual calculation method is adopted (parameter: spatial phase distribution map (S)). φ), overall rigid displacement parameter set (T rigid ), local neighborhood radius (r) loc This allows for residual analysis of the phase space after the removal of the overall rigid components, and identification of non-rigid response components between each synchronization marker.

[0158] Furthermore, through the spatial local residual mapping algorithm (parameters: spatial index ((xi,yi)), the local window size (d) win ), defining the spatial neighborhood (N) around each synchronization tag. i ), calculate its quadratic residual ε nl,i Used to characterize adjacent points

[0159] Small nonlinear offsets between:

[0160]

[0161] in, The normalized phase response value of the synchronization marker i, This represents the theoretical phase response in the neighborhood predicted based on the overall rigid transformation model.

[0162] Furthermore, the local stress pattern clustering algorithm (parameters: residual distribution threshold, feature vector extraction window (wf)) is used to cluster and identify patterns of all synchronously labeled quadratic residual values, and iteratively extract stress deformation patterns with significant spatial distribution characteristics, including in-plane bending, local warping, thermal expansion and contraction, etc., and encode each group of patterns into an independent physical deformation feature vector.

[0163] Furthermore, through the local stress quantization operator (parameter: weight allocation set (λ) j The spatial propagation model (G(x,y)) performs weighted and normalized processing on all local deformation feature vectors to generate a spatially balanced set of local stress deformation parameters, thereby enabling a quantitative evaluation of the intensity of local stress disturbance.

[0164] Through the above chain algorithm processing, the spatial phase distribution map of multilayer PCBs, after comprehensively eliminating the overall rigidity offset factor, further realizes the refined modeling and physical attribution of various nonlinear disturbances such as local stress, thermal bending, and non-uniform micro-area deformation, providing key technical support for establishing a highly sensitive and high-resolution working condition sensing channel.

[0165] For example, for a batch of 12-layer HDIPCB solder mask exposure process, given that the overall rigid displacement parameter set is known as translation (t=[14.3,12.1],μm) and rotation angle (△=0.035°), the spatial phase distribution map is acquired at a resolution of 2048×2048 pixels, and the local neighborhood radius (r) is set. loc=120, μm), and the quadratic residuals of all synchronous markers were analyzed. Local residual statistics showed that the residuals of 87 synchronous markers exceeded 0.018 radians, mainly concentrated in the thermally sensitive area and the area with high processing stress. The local stress clustering algorithm automatically clustered four main stress modes: in-plane tension, thermal warping, copper foil directional stress, and point stress concentration. Using spatial distribution weighting, the local stress deformation parameter set was normalized and output. The residual intensity of the distribution location corresponding to the maximum stress component was 0.023 radians, and the bending mode weight was 0.41. The above parameter set was verified by subsequent multi-layer alignment error modeling. It can accurately locate and quantify various non-rigid deformation regions caused by physical disturbances, support high-precision spatial distribution correction under flexible alignment compensation process, and effectively improve the robustness and dimensional consistency of multi-layer PCB solder mask process environment.

[0166] S5.4: Input the overall rigid displacement parameter set and the local stress deformation parameter set into the distributed alignment error modeling framework. Through the hierarchical error decomposition technique, construct the distributed alignment error model of multilayer PCB under specific environmental disturbance conditions, and output the distributed alignment error parameter matrix containing spatial multi-point error factors and their weight distribution.

[0167] S5.5: Utilizing the distributed alignment error parameter matrix, anomaly data identification and feature attribution algorithms are executed to locate and classify abnormal error regions caused by environmental disturbances, and output a physical difference perception report, providing accurate spatial error information input for subsequent virtual alignment plane construction and automatic alignment parameter compensation.

[0168] Step S6: For the distributed alignment error model, an optical phase difference superposition algorithm is used to generate a virtual alignment plane for the current inter-layer alignment relationship column, realizing compensation for the differences in alignment data under disturbance conditions. Specifically, this includes:

[0169] S6.1: Feature screening is performed on the multi-layer synchronous marker physical offset matrix in the distributed alignment error model. Principal component analysis algorithm is used to extract key spatial alignment features that characterize the overall rigid displacement and local stress deformation, forming a set of physical offset features to clarify the input basis for subsequent optical phase processing.

[0170] S6.2: Based on the above set of physical offset features, using the optical phase response parameters as input, the phase difference distribution map of each synchronization mark is extracted through the phase difference calculation module to obtain the multi-layer phase difference feature matrix after superimposing environmental disturbance factors, thereby realizing the mapping between physical offset features and optical phase response.

[0171] For sub-step S6.2, the input condition is the set of physical offset features obtained by filtering in step S6.1. The set contains the overall rigid displacement parameters and local stress deformation parameters selected by the principal component analysis algorithm. This set represents the actual physical state of the multilayer PCB under specific environmental disturbances in the form of standardized physical feature vectors.

[0172] An optical phase difference extraction method (parameters: physical offset feature set, standardized synchronization marker phase response data, layer number set (L), synchronization marker index set (Ml)) is used to achieve fine-grained hierarchical analysis of phase information between multiple synchronization markers.

[0173] Using the phase difference calculation module (algorithm: layer-by-layer relative phase method), the instantaneous phase difference between each synchronization mark in each layer (l) and the mark at the same position in the corresponding reference layer (such as the design layer or mask layer) is calculated:

[0174]

[0175] in, The phase difference of the (i)th synchronization mark in layer (l) is... This refers to the phase response of the index-synchronized marker on the base layer.

[0176] Furthermore, through a multi-layer phase difference superposition analysis algorithm (parameters: physical offset feature set, spatial coordinate matrix ({X...)...) i,l ,Y i,l The phase difference distribution of each layer of synchronous marker points is classified, global trends and local anomalies are screened, and based on physical parameter attributes, a clear mapping between physical offset characteristics (such as rigid translation, rotation, and local stress distortion) and optical phase behavior is achieved.

[0177] Furthermore, by using phase difference clustering and modeling methods (parameters: spatial distribution threshold, maximum number of target groups), the phase difference results of all synchronously labeled data are spatially adaptively grouped to establish phase response subsets corresponding to different physical disturbances, thereby distinguishing the overall movement area and stress anomaly area and improving the regional disturbance resolution capability.

[0178] Furthermore, through anomaly correction and defect elimination algorithms (parameters: anomaly identification threshold, weight allocation coefficient), dynamic correction and weighted elimination are performed on the synchronous labeling region data that have mismatch or abnormal deviation after clustering, thereby enhancing the anti-interference ability of the multi-layer phase difference feature matrix and improving the physical consistency and process adaptability of the simulation results.

[0179] Through the above chain processing method, a multi-layer phase difference feature matrix after superimposed environmental disturbance factors correction is obtained, realizing the mapping and dynamic adaptation between physical offset features and multi-dimensional data based on optical phase difference.

[0180] For example, for the exposure alignment of solder resist on a batch of 12-layer high-density interconnect printed circuit boards, the physical offset feature set extracted after principal component analysis is 36-dimensional, including the overall translation and rotation parameters of each layer and multiple sets of local stress components. The input normalized phase response data matrix is ​​[12 layers × 12 synchronization markers]. Using the layer-by-layer relative phase method, the phase difference of each of the 12 synchronization markers is obtained sequentially, and the global phase mean and standard deviation of each layer are calculated. An adaptive clustering algorithm is used to distribute the phase difference according to (δ... nlu Based on a radian classification of λ = 0.09, significant alignment shifts were found in 8 layers, and local anomalies were observed in 4 layers. This was determined based on the weighting coefficient λ. i By weighted elimination of abnormal values, a multi-layer phase difference feature matrix is ​​finally output, with its global noise distribution mean reduced to 0.011 radians and the residual rate in abnormal areas decreasing from 3.6% to 0.7%. The phase difference feature matrix output in this step serves as the standardized input for downstream virtual alignment plane reconstruction, fully ensuring the accurate response and flexible compensation capability of the multi-layer PCB solder mask process to complex disturbance factors.

[0181] S6.3: Perform a mirror reconstruction algorithm on the multi-layer phase difference feature matrix. Based on the spatial transformation relationship in the distributed alignment error model, map the optical phase difference of each layer to a unified virtual alignment plane, realize the spatial normalization representation of multi-layer data, and lay the mathematical foundation for the generation of the virtual alignment plane.

[0182] S6.4: Using the virtual alignment plane generation module, the spatially normalized multi-layer phase difference characteristics are fitted twice to dynamically construct a virtual alignment reference plane that can adapt to the current disturbance environment, thereby generating a compensation datum that explicitly describes the inter-layer relationship and alignment reference.

[0183] S6.5: Perform residual analysis on the generated virtual alignment reference plane and the initial design alignment plane, and use the difference compensation algorithm to output the corrected alignment data compensation amount for the alignment error distribution of each layer, so as to realize the difference compensation of the virtual alignment plane to the actual alignment data.

[0184] Step S7: Determine whether the current distribution alignment error model is within a preset threshold range. If it exceeds the threshold, automatically correct the alignment parameters of the exposure equipment and the mask pattern data based on the physical offset of each layer to achieve flexible parameter dynamic adaptation. Specifically, this includes:

[0185] S7.1: Perform threshold discrimination processing on the input distribution alignment error model, and use the preset threshold parameter set to perform difference analysis on the rigid displacement and local stress deformation in the multidimensional alignment reference feature set to identify whether it exceeds the allowable process deviation range and output the discrimination result.

[0186] S7.2: Based on the discrimination results of the distributed alignment error model, physical offset parameters of each layer are extracted from the multidimensional alignment reference feature set. The error decomposition algorithm is used to accurately separate the physical offset parameters into the overall displacement component and the local deformation component to obtain a multi-layer physical offset vector group.

[0187] S7.3: For the obtained multi-layer physical offset vector group, the flexible parameter mapping algorithm is called to map each physical offset vector to the alignment servo axis adjustment parameters of the exposure equipment, the mask pattern space transformation parameters, and the exposure reference plane calibration parameters, respectively, so as to realize multi-dimensional quantization of parameters.

[0188] The input is based on the physical offset vector group of each layer of a multilayer PCB. This vector group is obtained by an error decomposition algorithm and includes the overall displacement component and local deformation component of each layer.

[0189] A flexible parameter mapping algorithm (with parameters set based on process tolerance zone, equipment response characteristics, and mask pattern mismatch model) is adopted to input the physical offset vector of each layer into the multi-dimensional parameter mapping module to achieve a precise correspondence between physical space error and equipment control parameters.

[0190] Furthermore, for the overall rigid displacement vector components, the adjustment parameters of the positioning servo axis of the exposure equipment are derived using a linear space mapping method, specifically including the X / Y axis translation and rotation angle correction. In this step, the physical offset data is converted into mechanical servo indicators using an affine transformation matrix, the formula of which is:

[0191]

[0192] Where ΔX and ΔY are the physical offsets of the overall X-axis and Y-axis, respectively, and θ is the overall rotational offset.

[0193] Furthermore, for local deformation components, nonlinear spatial transformation and differential vector divergence algorithms are used to achieve parameterized output of the spatial deformation of the mask pattern, specifically including regional scaling factors and local distortion compensation amounts. This processing step utilizes a polynomial spatial fitting model to establish nonlinear registration between each physical offset sampling point and the mask pattern point matrix. Its spatial transformation can be expressed as:

[0194] X′=a0+a1X+a2Y+a3X 2 +a4XY+a5Y 2

[0195] Y′=b0+b1X+b2Y+b3X 2 +b4XY+b5Y 2

[0196] Among them, a i b iThe parameters are the polynomial parameters obtained by fitting using the least squares method.

[0197] Furthermore, the optical exposure reference plane is comprehensively calibrated for the overall rigid displacement parameters and local deformation spatial transformation parameters of all layers. Through a three-dimensional spatial rotation mapping algorithm, the exposure reference plane attitude adjustment parameters under the current process state are output, including the reference plane normal vector direction and spatial offset.

[0198] The exposure equipment alignment servo axis adjustment parameters, mask pattern spatial transformation parameters, and exposure reference plane calibration parameters generated through the above steps are aggregated into a set of multi-dimensional quantitative process parameters, which are marked as the flexible process compensation control data for this batch.

[0199] By using a flexible parameter mapping algorithm, a dynamic mapping relationship is established between each physical offset vector and the equipment control parameters, enabling synchronous adaptive calibration of the exposure equipment and mask data under environmental disturbances, thereby improving the alignment accuracy and production flexibility of the solder resist process.

[0200] For example, in the solder mask exposure process of a batch of 6-layer high-density multilayer PCBs, the online process inspection system detected that the physical offset of the 3rd layer was (ΔX3,ΔY3,θ3)=(12μm,-18μm,0.022°). After local deformation was sampled by a dot matrix, 50 sets of offset data were generated. The spatial deformation parameter set (a) was obtained by polynomial fitting. i = (0.98, 0.02, ..., -0.001). After the flexible parameter mapping algorithm is executed, the translation of the alignment servo axis of the third layer is adjusted to +12μm (X direction), -18μm (Y direction), and +0.022° (rotation); the local region of the third layer of the mask pattern is deformed according to the fitted function;

[0201] The normal of the exposure reference plane can be finely adjusted from (0,0,1) to (0.001,-0.002,0.999)±25μm, and mapping and compensation can be completed automatically. The overall alignment accuracy is improved to ≤6μm. Under 20 batches of working conditions, the yield fluctuation is less than 2%.

[0202] S7.4: Using the multidimensional quantized exposure equipment, the position servo axis adjustment parameters and mask pattern spatial transformation parameters are used to perform automatic parameter dynamic refresh operations on the exposure control system and mask data management system to generate a flexible exposure process parameter set based on the physical state of the current batch.

[0203] S7.5: For the generated flexible exposure process parameter set, execute the process safety verification algorithm to determine whether the corrected alignment parameters and mask pattern data meet the product design requirements and process tolerances; if not, perform adaptive optimization iteration on the physical offset vector group and flexible parameter mapping process, and output the final corrected parameter results.

[0204] Step S8: During the solder mask exposure process, the actual exposure operation is completed based on the corrected alignment parameters and mask pattern data, and the current batch condition label and environmental parameters are recorded for subsequent traceability. Specifically, this includes:

[0205] S8.1: Receive and load the alignment parameters and mask pattern data corrected by the adaptive optics processing algorithm as input to the exposure control module during the exposure process, so as to ensure that the physical exposure is controlled based on the current optimal alignment state.

[0206] S8.2: The exposure parameter is sent to the solder mask exposure control module. The motion control system adjusts the light source projection path and mask positioning device according to the corrected alignment parameters to achieve precise alignment and pattern correction at the physical level, and ensures the physical consistency between the mask pattern data and the synchronous marking position of the board layer.

[0207] S8.3: Before the actual execution of the exposure process, environmental parameters (such as temperature, humidity, vibration amplitude and other physical variables) of this batch are collected based on the operating condition monitoring sensor. The data acquisition module is called to structure and package the current state before exposure and save it to obtain the batch operating condition label for process traceability.

[0208] Based on the alignment parameters and mask pattern data corrected by adaptive optics processing, the input data includes the exposure control parameter set that has been calibrated by the process and the batch identifier of the multilayer PCB to be exposed.

[0209] A multi-channel operating condition monitoring sensor array (parameters: high-precision temperature and humidity sensor, triaxial acceleration and vibration sensor, air dust concentration sensor, etc.) is used to achieve real-time high sampling rate acquisition of physical variables of the exposure environment.

[0210] Furthermore, through the integrated sensor data acquisition module (with parameters set according to the sensor sampling frequency, resolution, and calibration curve), the environmental state data before each batch of exposure is periodically and synchronously acquired, forming a raw data stream of time-series environmental variables.

[0211] Furthermore, data preprocessing algorithms (such as drift correction, outlier removal, and time series normalization) are used to clean and standardize the original environmental data stream, thereby eliminating sporadic sensor noise and correcting baseline drift, and obtaining standardized environmental parameters with process discrimination value.

[0212] Furthermore, through a structured data packaging algorithm (parameters include: unique batch ID, collection time, spatial location identifier, environmental multi-parameter vector, and equipment operating status identifier), the collected and standardized data is tagged and bound to the corresponding batch and equipment operating status, and batch operating condition tag data packages are generated for process lifecycle traceability.

[0213] The batch condition label storage and distribution module archives the tagged data to the manufacturing execution system database in real time, enabling batch-level full-process data traceability and anomaly analysis capabilities based on pre-exposure conditions.

[0214] Through the chain-like data flow of acquisition, preprocessing, packaging, and archiving described above, the environmental physical state of the batch before exposure is precisely coupled with the process control parameters, which forms the technical basis for intelligent traceability and anomaly correlation analysis of the entire solder resist process.

[0215] For example, before the solder mask exposure of a batch of 10-layer high-density PCBs, the system calls a temperature and humidity sensor (accuracy 0.1℃ / 1%RH, update rate 10Hz) to collect real-time data: temperature 24.6℃, humidity 47.3%, vibration sensor (resolution 1mg, triaxial acceleration) to collect X = 6.5mg, Y = 2.8mg, Z = 0.9mg, and air dust concentration sensor to measure particle concentration as 8.2μg / m³. 3 The data collection timestamp is 20240618T092307, and the spatial location number is Area-C6. The data acquisition module automatically cleaned a set of standardized environmental parameter vectors [24.6, 47.3, 6.5, 2.8, 0.9, 8.2], which, after structured packaging, generated the following batch condition labels:

[0216] {"BatchID":"A6B0923D","Timestamp":"20240618T092307","Location":"Area-C6","EnvironParam":[24.6,47.3,6.5,2.8,0.9,8.2],"DeviceState":"Ready","ParamSetID":"EXP-C123F"}. This condition label is archived in the MES database in real time and indexed and associated with the exposure parameter set and alignment correction results of this batch. The system can subsequently use this to trace back abnormal batches, locate the environmental context in which process deviations occurred, and achieve multi-batch anomaly tracing, data mining, and parameter evolution optimization, greatly improving the scientific nature of batch difference control and manufacturing consistency.

[0217] S8.4: Drive the exposure control module to perform solder resist exposure operation, monitor the exposure energy, spectral distribution and board layer synchronous marking response curve in real time during the exposure process, and use the exposure process monitoring data stream to verify the timing integrity and parameter landing of the actual exposure execution, so as to ensure that the graphic transfer is accurate and meets the multi-dimensional alignment reference requirements.

[0218] S8.5: After exposure is completed, key information such as exposure equipment operation logs, batch condition labels and environmental parameters, and synchronous marker response data generated during the exposure process of this batch of solder resist film is structured and archived through the data recording module, and a unique traceability identifier is assigned to provide full-chain traceable data support for subsequent alignment residual verification and process consistency big data analysis.

[0219] Step S9: After exposure, the synchronization mark area is detected by re-imaging to obtain the phase response of the synchronization mark after actual exposure. This is compared with the initial multidimensional alignment reference to calculate the final alignment residual, achieving dynamic feedback verification. Specifically, this includes:

[0220] S9.1: Perform process positioning on the multilayer PCB after the solder mask exposure is completed. Based on the spatial distribution benchmark of the synchronous mark generated before exposure, use a precision re-imaging device to obtain high-resolution re-imaging image data of the actual synchronous mark area to ensure that the imaging object maintains spatial consistency with the data before exposure.

[0221] S9.2: The automatic marker recognition and ROI extraction algorithm is used for the re-imaging image data. Based on the previously generated synchronous marker region pixel set, pixel-level recognition and contour matching are performed in the micro passive synchronous marker regions of each layer of the multi-layer PCB to output the re-imaging pixel dataset of the synchronous marker region.

[0222] S9.3: Apply a phase encoding and decoding algorithm to the re-imaging pixel dataset, combine it with the original optical phase response matrix obtained before exposure, and obtain the standardized phase response value of the synchronous marker after actual exposure, so as to align the optical phase response data of re-imaging with the first imaging.

[0223] S9.4: Based on the initial multidimensional alignment reference feature set, the standardized phase response value of the synchronization mark after exposure is used as input. The phase response of the multi-layer synchronization mark is matched one by one through the feature correspondence algorithm. The spatial offset between the actual physical position of each mark point and the reference physical position is calculated to form the multi-layer alignment spatial residual matrix after exposure.

[0224] S9.5: Normalize the multi-layer alignment space residual matrix after exposure, comprehensively consider batch operating condition labels such as exposure parameters, ambient temperature and humidity, and stress disturbance, and use dynamic feedback verification algorithm to output the final alignment residual evaluation result, providing criteria and feedback data for subsequent abnormal alarms, compensation correction, process adaptive learning and consistency traceability processes.

[0225] Step S10: Based on the dynamic feedback verification results, if the alignment residual does not meet the process requirements, an abnormal alarm or supplementary exposure process is triggered. The distributed alignment error model and adaptive optics processing algorithm are continuously optimized by recording residual information to improve the consistency and robustness of subsequent batch manufacturing. Specifically, this includes:

[0226] S10.1: Based on the actual post-exposure synchronous marker phase response data obtained from re-imaging detection, compare it with the initial multidimensional alignment reference feature set, calculate the actual alignment residual, and obtain the batch actual alignment accuracy evaluation index.

[0227] S10.2: Apply a threshold discrimination algorithm to the actual alignment accuracy evaluation index of the batch to determine whether the alignment residual falls within the preset process standard range, and output an error exceeding the limit mark or a qualified judgment signal.

[0228] S10.3: If the error exceeding the limit marker is activated, the abnormal alarm process will be automatically triggered based on the error spatial distribution characteristics, and the process label, residual data and environmental parameters of the batch will be pushed to the upper control system in real time to form an abnormal handling data link.

[0229] S10.4: Based on the abnormal handling data link, a step-by-step supplementary exposure strategy or a fine alignment correction strategy is adopted to perform local secondary exposure compensation on the local synchronous marking area with the largest residual, so as to correct the alignment error of the process defect area in real time and realize the closed-loop correction of the process.

[0230] S10.5: The historical residual information is fused with multi-source data such as process labels and environmental parameters, and then input into the distributed alignment error model and adaptive optics processing algorithm. The model weights are iteratively optimized through machine learning to improve the model's ability to identify and compensate for errors under complex disturbance environments.

[0231] S10.6: Backtest the optimization results of the adaptive optics processing algorithm and the distributed alignment error model. If the alignment residuals of the same batch or subsequent batches are effectively converged after the model update, the process adjustment data will be automatically archived to provide a data-driven basis for parameter recommendations and robustness improvement in the batch manufacturing process.

[0232] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0233] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0234] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. Solder mask methods for printed circuit boards, specifically including: S1: Acquire PCB design data and pre-place miniature passive synchronization markers in a designated area of ​​each layer of the circuit board to form distinguishable multi-layer alignment signals; S2: Before the solder mask is exposed, a global scan is performed on the synchronous markings on the surface of the multilayer PCB and the mask to collect the original optical response signals of environmental differences and vibration interference. S3: Preprocess the original optical response signal to obtain standardized phase response data of the multilayer synchronization marker; S4: Input the standardized phase response data into the adaptive optics processing algorithm to extract features of the optical phase difference of each micro passive synchronization mark, and obtain a multi-dimensional alignment reference feature set; S5: Based on the multidimensional alignment reference feature set, establish a distributed alignment error model for multilayer PCBs under different environmental disturbance conditions; S6: Generate a virtual alignment plane for the current inter-layer alignment relationship column based on the aforementioned distribution alignment error model; S7: If the distribution alignment error model exceeds the preset threshold range, the alignment parameters of the exposure equipment and the mask pattern data are automatically corrected according to the physical offset of each layer. S8: During the solder mask exposure process, the actual exposure operation is completed based on the corrected alignment parameters and mask pattern data. S9: After exposure, the synchronization mark area is detected by re-imaging to obtain the phase response of the synchronization mark after actual exposure and to calculate the final alignment residual; S10: Based on the dynamic feedback verification results, if the alignment residual does not meet the process requirements, an abnormal alarm or supplementary exposure process will be triggered, and the distributed alignment error model and adaptive optics processing algorithm will be continuously optimized.

2. The solder mask method for printed circuit boards according to claim 1, characterized in that, Step S1 specifically includes: The PCB design data is structured and analyzed to extract the geometric stack-up structure, solder mask distribution, and alignment accuracy requirements of each layer in the target multilayer PCB. Based on a list of pre-defined physical identifier regions, a specific optical phase coding design algorithm is used to automatically allocate phase coding parameters for each layer of identifier regions, generating a unique set of synchronization marker parameters. Based on the set of synchronization marker parameters, a suitable miniature passive synchronization marker structure is selected, and a physical synchronization marker layout design file is generated through parameterization using the PCB process database. The physical synchronization mark layout design file is integrated with the original PCB stack-up layout to verify design rule conflicts, spatial overlaps and signal interference issues during the synchronization mark embedding process, and output the optimal multi-layer physical synchronization mark integrated layout. Based on the final multi-layer physical synchronization mark integrated layout, a process execution work order corresponding to the solder mask exposure process is generated. Through the automatic instruction issuance of the manufacturing process, the physical synchronization mark is guided to be accurately embedded and optically phase-encoded in the actual circuit board manufacturing process.

3. The solder mask method for printed circuit boards according to claim 1, characterized in that, Step S2 specifically includes: Process positioning was performed on the multilayer PCB sample to be exposed and its paired mask surface to determine the initial coordinate distribution of the micro passive synchronous marking area in three-dimensional space. Based on the spatial distribution benchmark of synchronous markers, the original optical image signals of the micro passive synchronous marker areas on the surface of multilayer PCBs and masks are acquired synchronously to obtain high-resolution original optical image data containing physical location, reflection / refraction differences and the relative relationship of multilayer markers; The high-resolution raw optical image data is processed using a multi-channel parallel reading mechanism, an automatic marker recognition algorithm and ROI extraction processing are executed, and a set of pixels in the synchronously marked region is output. The original phase response values ​​of the synchronization marker under the current illumination, temperature, humidity and random vibration disturbance conditions are extracted by using a phase encoding and decoding algorithm on the pixel set of the synchronization marker region, and the original optical phase response matrix is ​​generated. The original optical phase response matrix is ​​labeled and associated with the synchronously labeled spatial distribution benchmark and environmental parameters to synthesize a batch operating condition optical response dataset.

4. The solder mask method for printed circuit boards according to claim 1, characterized in that, Step S3 specifically includes: The raw optical response signal output by the high-speed imaging device is subjected to baseline correction processing to obtain a baseline correction signal, and background noise suppression is performed on the baseline correction signal to output a noise suppression signal. Adaptive temporal denoising technology is used to detect and remove abnormal time points corresponding to sudden changes in environmental temperature and humidity, dust projections, and local micro-deformation disturbances from the noise suppression signal, thus forming a noise removal signal. The noise-removed signal is subjected to multi-layer synchronization tag differentiation and weighting processing. Based on the preset spatial distribution and physical coding characteristics of each synchronization tag, different weights are assigned to obtain a spatially weighted signal. The spatially weighted signal is subjected to intensity and phase normalization transformation to output normalized phase response data.

5. The solder mask method for printed circuit boards according to claim 1, characterized in that, Step S4 specifically includes: The standardized phase response data of the multi-layer synchronous markers are input into the adaptive optics processing algorithm module. The phase difference is calculated using the phase difference analysis algorithm on the standardized phase response data of each layer of synchronous markers to obtain the optical phase difference characteristic parameters of the miniature passive markers at different physical locations and marked in the PCB board-mask system. Based on the acquired optical phase difference feature parameters, an extraction algorithm is used to generate feature vectors. The multidimensional phase difference feature vectors are aggregated, and the working condition-related parameters are weighted using a heterogeneous marker feature fusion model to output a multidimensional alignment reference feature set.

6. The solder mask method for printed circuit boards according to claim 1, characterized in that: In step S1, the structured analysis of PCB design data includes automatic identification of solder mask area boundaries, hierarchical alignment accuracy labels and spatial embeddable area determination, and screening of the final preset synchronization mark area through conflict avoidance and safety interval algorithms.

7. The solder mask method for printed circuit boards according to claim 1, characterized in that: In step S3, spatial filtering includes two-dimensional Gaussian filtering and multi-scale background modeling. Temporal denoising uses a sliding window and change point detection, and signal enhancement compensation is performed for local low SNR regions by weight boosting or data resampling.

8. The solder mask method for printed circuit boards according to claim 1, characterized in that: In step S4, multidimensional feature extraction includes principal component analysis and rigid affine model decomposition, which respectively generate the overall rigid displacement vector and local nonlinear deformation features, resulting in a structured representation of multi-layer physical offset and deformation.

9. The solder mask method for printed circuit boards according to claim 1, characterized in that: In step S5, based on the overall rigid displacement parameter set and the local stress deformation parameter set, the stress characteristics of multilayer PCBs are quantitatively analyzed and classified by local residual mapping and spatial clustering algorithms.

10. The solder mask method for printed circuit boards according to claim 1, characterized in that: In step S6, regional perturbation resolution and reconstruction of virtual alignment reference planes between multi-level synchronization markers are obtained through principal component analysis, phase difference calculation and multi-level spatial normalization, and abnormal regions are dynamically corrected and weighted elimination is performed.

Images