Image correction method of digital imaging system
By establishing a mapping relationship between vector features and deviation data in a digital imaging system and using a deep learning model for intelligent correction, the deviation problem in photolithography pattern manufacturing of digital imaging systems is solved, achieving high-precision and high-efficiency imaging effects, applicable to fields such as PCB, OLED, and semiconductors.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEFEI ZHONGQUN PHOTOELECTRIC TECH CO LTD
- Filing Date
- 2026-01-06
- Publication Date
- 2026-05-01
AI Technical Summary
In the process of photolithography pattern manufacturing, existing digital imaging systems suffer from deviations between the actual image pattern and the original design pattern due to factors such as optical distortion, uneven exposure energy distribution, and fluctuations in the development process. This is especially true in high-density interconnects and micro-line processing, where traditional compensation methods lack adaptive capabilities and cannot perform targeted corrections.
By acquiring CAM vector graphics data, performing post-development graphic matching and alignment, extracting geometric features, constructing a deviation dataset, and using convolutional neural networks or graph neural networks to establish a mapping relationship between vector features and deviation data, intelligent correction is performed. Image correction is carried out using local regionalization strategies and deep learning models.
It achieves high-precision image correction, improves imaging accuracy and efficiency, is applicable to various lithography fields, has continuous learning capabilities, is compatible with existing processes, and requires no large-scale equipment modification.
Smart Images
Figure CN121961887A_ABST
Abstract
Description
An image correction method for a digital imaging system Technical Field
[0001] This invention belongs to the interdisciplinary field of optoelectronic manufacturing and intelligent image processing, and specifically relates to an image correction method for a digital imaging system. Background Technology
[0002] Digital direct imaging systems are widely used in PCB and other photolithography pattern manufacturing processes. Their principle involves transferring circuit patterns generated by computer-aided design (CAM) onto a photosensitive layer using a digital light processing (DLP) projection component. However, due to factors such as optical distortion, uneven exposure energy distribution, development process fluctuations, and differences in the properties of developing materials, there are often deviations between the actual image and the original CAM design. These deviations directly affect the accuracy and reliability of the final image.
[0003] Currently, common compensation methods mainly rely on empirical formulas or fine-tuning of process parameters, such as scaling the global dimensions, linewidth compensation, or adjusting exposure energy. However, these methods lack adaptability and cannot make targeted corrections based on complex and diverse actual deviation patterns. Especially in high-density interconnect (HDI) and micro-line fabrication, the limitations of traditional methods become increasingly apparent. Summary of the Invention
[0004] To address the aforementioned problems, this invention discloses an image correction method for a digital imaging system. The correction method includes the following steps: acquiring data of a CAM vector graphic in the digital imaging system; projecting the vector image generated by the digital imaging system onto a photosensitive substrate and developing it; acquiring the actual image after development and performing pattern matching and spatial alignment between the actual image and the CAM vector graphic; extracting geometric features from the aligned image and CAM vector graphic data, and constructing an image deviation dataset; using the CAM vector graphic data and the actual deviation data as training samples to train a model and establish a mapping relationship between vector features and deviation data; correcting the input CAM vector graphic data based on the trained model, and then importing it into the digital imaging system for imaging.
[0005] Furthermore, the data of the CAM vector graphics is in polygonal description form.
[0006] Furthermore, during model training, the actual image is segmented according to primitive categories.
[0007] Furthermore, before acquiring the actual image after development, the process includes region segmentation and region of interest identification of the actual image, and only the predetermined region of interest is acquired during acquisition.
[0008] Furthermore, the geometric features include one or more combinations of line width, edge, and spacing.
[0009] Furthermore, the CAM vector graphics data contains images of various common exposure graphic distortion features.
[0010] Furthermore, the model is a convolutional neural network, a graph neural network, or a deep learning model incorporating an attention mechanism.
[0011] This invention also discloses a model training method for image correction in a digital imaging system. The model training method includes the following steps: acquiring data of CAM vector graphics in the digital imaging system; projecting the vector image generated by the digital imaging system onto a photosensitive substrate and developing it; acquiring the actual image after development, and performing pattern matching and spatial alignment between the actual image and the CAM vector graphics; extracting geometric features from the aligned image and CAM vector graphics data, and constructing an image deviation dataset; using the CAM vector graphics data and the actual deviation data as training samples for model training, and establishing a mapping relationship between vector features and deviation data.
[0012] The present invention also discloses a computer-readable storage medium storing a computer program, which, when executed, performs an image correction method for a digital imaging system as described in any of the preceding claims.
[0013] The present invention also discloses a computer device, including a processor and a storage medium, wherein a computer program is stored on the storage medium, and the processor reads from the storage medium and runs the computer program to perform an image correction method for a digital imaging system as described in any of the preceding claims.
[0014] Beneficial effects: 1. High accuracy: Based on model learning of actual deviation patterns, intelligent correction of line width, size and edge is achieved, which greatly improves imaging accuracy.
[0015] 2. Intelligent processing: It adopts a localized regional correction strategy to avoid computational redundancy caused by processing whole map data, resulting in higher efficiency.
[0016] 3. It has strong versatility and is not only suitable for PCB manufacturing, but also for LDI-related fields such as photolithography, OLED, and semiconductors.
[0017] 4. Iterative optimization: The model has continuous learning capabilities, and the correction effect gradually improves as data accumulates; it has good compatibility and can be directly embedded into existing CAM processing and LDI process flows without the need for large-scale equipment modification.
[0018] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 shows a flowchart of an image correction method for a digital imaging system according to an embodiment of the present invention; Figure 2 is a CAM illustration in an embodiment of the present invention; Figure 3 is the final image after exposure and development by a direct imaging system; Figure 4 is the correction of Figure 2 after model correction; Figure 5 is the effect of re-exposure of the corrected CAM image. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] To address the discrepancy between the image and the actual image during digital direct imaging, this invention discloses an image correction method for a digital imaging system, as shown in Figure 1. The correction method includes the following steps: acquiring data of a CAM vector graphic in the digital imaging system; projecting the vector image generated by the digital imaging system onto a photosensitive substrate and developing it; acquiring the developed actual image and performing pattern matching and spatial alignment between the actual image and the CAM vector graphic; extracting geometric features from the aligned image and CAM vector graphic data, and constructing an image deviation dataset; using the CAM vector graphic data and the actual deviation data as training samples to train a model and establish a mapping relationship between vector features and deviation data; correcting the input CAM vector graphic data based on the trained model, and then importing it into the digital imaging system for imaging.
[0023] Specifically, in this embodiment, a large amount of CAM vector graphics data is acquired, and then the acquired CAM vector graphics data is developed and imaged using a digital imaging system. Next, pattern matching and spatial alignment are performed between the actual image and the CAM vector graphics to obtain the graphic deviation between them. A graphic deviation dataset is constructed using a large amount of graphic deviation data, and then an artificial intelligence model is used to train the graphic deviation dataset to obtain the mapping relationship between the vector graphics data and the deviation data. Finally, the input CAM vector graphics data is corrected using the mapping relationship, and the corrected data is imported into the digital imaging system for direct imaging. In this technical solution, the correspondence between CAM vector graphics data and the actual image generated by the digital imaging system is used. An artificial intelligence model is trained using a large amount of data, and then the vector graphics are corrected before passing through the digital imaging system to obtain the final corrected image. This embodiment does not require complex adjustments to the various parameters of the digital imaging system or relies on formulas or experience, and can effectively and accurately image through the digital imaging system. Pattern matching, a technique combining image processing and computer vision, is mainly used here to position and align the CAM vector graphics and the actual exposed and developed image.
[0024] As shown in Figures 2-5, Figure 2 is the CAM vector graphic we expect to generate. However, if generated directly using a digital imaging system, due to parameter deviations and other reasons, the result might look like Figure 3. In this embodiment, the CAM vector graphic data in Figure 2 is corrected to obtain data similar to that in Figure 4. Then, the data corresponding to Figure 4 is processed by a digital imaging system to obtain the corrected Figure 5. Therefore, this embodiment can effectively improve the imaging accuracy of the digital imaging system.
[0025] Furthermore, the data of the CAM vector graphics is in polygonal description form.
[0026] Specifically, the "polygonal description format" is the basic data representation format for CAM vector graphics. It defines the vector description of the graphic outline through the coordinate information of discrete vertices (two-dimensional planar coordinates), the arrangement order of the vertices, and the connection relationships between the vertices. For example, as shown in Figure 2, this CAM vector graphic is described using the polygonal description format: The CAM vector graphic consists of two rectangles. Rectangle 1 has its vertex at point a, a length of 1cm, and a width of 10cm. Rectangle 2 has its vertex at point b, a length of 3cm, and a width of 0.5cm.
[0027] In traditional CAM graphics, polygon descriptions are used only as "data storage format" and are not deeply integrated with "feature extraction and classification correction" for deviation correction. In this embodiment, by clarifying the specific connotation and classification method of polygon description form, as well as the collaborative logic with geometric feature extraction and local correction, the "format characteristics" are transformed into "technical means to improve correction accuracy".
[0028] Furthermore, during model training, the actual image is segmented according to primitive categories.
[0029] Specifically, a "primitive" refers to the smallest independent geometric unit that constitutes a polygonal CAM vector graphic. Examples include linear primitives, curved primitives, polygonal primitives, and composite primitives. Different primitives exhibit different deviation characteristics during digital imaging. For instance, linear primitives are prone to line width uniformity deviations, local positional shifts, and inconsistent thickness at both ends; curved primitives are prone to curvature distortion, edge burrs, local depressions / protrusions, and arc length deviations; polygonal primitives are prone to interior angle distortion, side length deviations, corner rounding, and uneven contours; composite primitive deviations are the superposition of basic primitive deviations and are prone to transition deviations at the junctions of different primitives.
[0030] For CAM vector graphics with polygonal descriptions, based on the geometric features of the primitives and the differences in process deviation patterns, the entire CAM vector graphic is first decomposed into sub-graphics of a single primitive category. Then, the vector features and corresponding actual deviation data of each sub-graphic are extracted to construct a deviation subset dataset for each category, which is finally used for targeted training of AI models.
[0031] Upgrading the "graphic primitive" from a "basic display unit" to a "core technology carrier for improving correction accuracy and efficiency" solves the pain points of blurred mapping relationships and insufficient correction targeting caused by traditional "whole image hybrid training".
[0032] Furthermore, before acquiring the actual image after development, the process includes region segmentation and region of interest identification of the actual image, and only the predetermined region of interest is acquired during acquisition.
[0033] Camera image acquisition takes time. To improve acquisition efficiency, some CAM vector images (such as large white areas or standard, isolated circles) are not significantly different from the actual exposed images. Therefore, we are only interested in regions where the CAM vector image differs from the actual exposed image, and only these regions are used for image acquisition and data training. The predetermined regions of interest can be automatically identified by the program or manually labeled.
[0034] Specifically, when acquiring actual imaging images, instead of using a "whole-image indiscriminate acquisition" approach, we pre-identify and lock local regions (ROIs) that are "of core value to model training" based on the deviation patterns of the manufacturing process and the functional importance of the images. We then capture high-resolution image data of these regions using an image acquisition device, while selectively acquiring simplified data (or not acquiring) of non-ROI regions. This reduces the amount of invalid data, improves the relevance of training samples, and ultimately enhances the efficiency of model training.
[0035] The regions of interest include "high-frequency deviation areas" and "critical functional areas," specifically including: densely packed lines, graphic corner areas, irregularly shaped structure areas, pad areas, signal transmission line areas, etc. The deviation characteristics of these areas are more typical and have a greater impact on the correction effect. Collecting data from these areas can provide "high-value samples" and avoid invalid data redundancy caused by full-map collection.
[0036] Preferably, different acquisition precision methods can be used for regions of interest and non-interest areas. For example, "high-definition acquisition" (pixel precision ≥ 1000 dpi) can be used for regions of interest to ensure that subtle deviations such as line width and edge burrs can be identified, while acquisition precision can be reduced or only contour data can be acquired for regions of interest. This strategy allows for focused resource investment in core areas, reducing the time costs of data storage, transmission, and preprocessing, and improving the efficiency of the training process.
[0037] Furthermore, the geometric features include one or more combinations of line width, edge, and spacing.
[0038] Specifically, by combining the characteristics of primitives, different geometric features are extracted from different primitives. Using different geometric features for correction can greatly improve the efficiency of graphic correction.
[0039] Furthermore, the corrected vector graphics can be adaptively adjusted according to different process parameters to suit different optical systems and developing environments.
[0040] In essence, adaptive adjustment based on different process parameters involves switching between different models in the background. These different models are trained based on the different process parameters. Specifically, in different optical systems and developing environments, the mapping relationship between vector features and deviation data differs due to parameter variations. This embodiment allows for adaptive modification of the mapping relationship to suit different optical systems and developing environments.
[0041] Furthermore, the model is a convolutional neural network, a graph neural network, or a deep learning model incorporating an attention mechanism.
[0042] In another embodiment of the present invention, a model training method for image correction in a digital imaging system is disclosed. The model training method includes the following steps: acquiring data of CAM vector graphics in the digital imaging system; projecting the vector image generated by the digital imaging system onto a photosensitive substrate and developing it; acquiring the actual image after development, and performing pattern matching and spatial alignment between the actual image and the CAM vector graphics; extracting geometric features from the aligned image and the CAM vector graphics data, and constructing an image deviation dataset; using the CAM vector graphics data and the actual deviation data as training samples for model training, and establishing a mapping relationship between vector features and deviation data.
[0043] In another embodiment of the present invention, a computer-readable storage medium is disclosed, on which a computer program is stored, and the computer program, when run, executes an image correction method for a digital imaging system as described in any of the above embodiments.
[0044] In another embodiment of the present invention, a computer device is also disclosed, including a processor and a storage medium, wherein a computer program is stored on the storage medium, and the processor reads from the storage medium and runs the computer program to perform an image correction method for a digital imaging system as described in any of the foregoing embodiments.
[0045] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An image correction method for a digital imaging system, characterized in that, The correction method includes the following steps: acquiring data of CAM vector graphics in a digital imaging system; projecting the vector image generated by the digital imaging system onto a photosensitive substrate and developing it; acquiring the actual image after development, and performing pattern matching and spatial alignment between the actual image and the CAM vector graphics; extracting geometric features from the aligned image and CAM vector graphics data, and constructing a graphic deviation dataset; using the CAM vector graphics data and the actual deviation data as training samples to train a model and establish a mapping relationship between vector features and deviation data; correcting the input CAM vector graphics data according to the trained model, and then importing it into the digital imaging system for imaging.
2. The image correction method for a digital imaging system according to claim 1, characterized in that, The data for the CAM vector graphics is in polygonal description format.
3. The image correction method for a digital imaging system according to claim 1, characterized in that, During model training, the actual image is segmented according to primitive categories.
4. The image correction method for a digital imaging system according to claim 1, characterized in that, Before acquiring the actual image after development, the process also includes region segmentation and region of interest identification of the actual image, and only the predetermined region of interest is acquired during acquisition.
5. The image correction method for a digital imaging system according to claim 1, characterized in that, The geometric features include one or more combinations of line width, edge, and spacing.
6. The image correction method for a digital imaging system according to claim 1, characterized in that, The CAM vector graphics data contains images of various common exposure graphic distortion features.
7. The image correction method for a digital imaging system according to claim 1, characterized in that, The model is a convolutional neural network, a graph neural network, or a deep learning model that incorporates an attention mechanism.
8. A model training method for image correction in a digital imaging system, characterized in that, The model training method includes the following steps: acquiring data of CAM vector graphics in a digital imaging system; projecting the vector image generated by the digital imaging system onto a photosensitive substrate and developing it; acquiring the actual image after development, and performing pattern matching and spatial alignment between the actual image and the CAM vector graphics; extracting geometric features from the aligned image and CAM vector graphics data, and constructing a graphic deviation dataset; using the CAM vector graphics data and the actual deviation data as training samples for model training, and establishing a mapping relationship between vector features and deviation data.
9. A computer-readable storage medium, characterized in that, The medium contains a computer program, which, when run, executes an image correction method for a digital imaging system as described in any one of claims 1 to 7.
10. A computer device, characterized in that, It includes a processor and a storage medium, on which a computer program is stored, and the processor reads from the storage medium and runs the computer program to perform an image correction method for a digital imaging system as described in any one of claims 1 to 7.