Processing method, system and application for sparse compression reconstruction of ldi sub-pixel image

Abstract

The application provides a processing method, system and application of sparse compression reconstruction of an LDI sub-pixel image, the method comprising: calculating, by a computer image compression algorithm, a low-resolution to-be-exposed image corresponding to each phase structure in advance; loading the calculated low-resolution to-be-exposed image to a digital micro mirror in a time sequence, wherein the display time of each pattern matches the rotation position of the corresponding phase structure; coupling the pattern of the digital micro mirror and a phase encoding mask in a frequency domain through a 4-f optical system, and reconstructing a high-resolution exposure pattern at a sub-pixel level by using a diffraction effect; projecting the high-resolution light field of the reconstructed high-resolution exposure pattern to a photoresist surface, and forming a target circuit pattern by accumulating an exposure dose; and obtaining a micro-nano structure with sub-pixel accuracy after development. The system comprises an image compression module, a pattern matching module and a pattern forming module. The application reduces the manufacturing cost of a laser direct imaging device under the condition of the same accuracy.

Description

Technical Field

[0001] This invention relates to the field of printed circuit board (PCB) exposure equipment technology, and in particular to a processing method, system, and application for sparse compression reconstruction of LDI subpixel images. Background Technology

[0002] Laser direct imaging (LDI) exposure machines are a new generation of PCB exposure equipment. They simplify the exposure process, eliminating the need for exposure masks and offering significantly better real-time performance than traditional exposure machines that require mask fabrication. Their use of digital masks is driving the upgrading of electronic manufacturing towards greener, more environmentally friendly, and more refined processes, making them a core piece of equipment in high-end PCB intelligent manufacturing. The basic principle is to directly control the laser to irradiate the photosensitive adhesive (including photoresist) onto the designed circuit pattern through digital processing, thereby forming the circuit pattern on the photosensitive adhesive. This can be achieved by directly controlling the laser beam for rapid single-point scanning or by controlling multiple pixels of a digital micromirror (DMD) for simultaneous laser direct imaging. Current technology determines the resolution based on the pixel size of the LDI system, and no approximate sub-pixel enhancement technology has been found for LDI. When using a digital micromirror as the laser direct imaging method, because the digital micromirror is a digital mirror and not a continuous imaging device, there is a fixed pixel size in the entire imaging system, and the precision of the manufactured circuit board is determined by this size.

[0003] Prior art 1, application number CN202411785211.0, discloses a data processing method, system, and related equipment for laser direct imaging, used to achieve a balance between imaging accuracy and scanning efficiency in laser scanning imaging. The method includes: acquiring each target region of the pixel row containing pixels of a preset type in the original dot matrix image; increasing the resolution of the target region in the vertical direction of the laser scanning direction, and decreasing the resolution of other regions outside the target region in the vertical direction of the laser scanning direction, to obtain a corrected image for laser scanning imaging. Although this achieves a balance between imaging accuracy and scanning efficiency in laser scanning imaging, it employs a selective resolution adjustment strategy, increasing resolution in specific target regions while sacrificing accuracy in other regions, which has limitations in regional resolution adjustment and a single scanning direction.

[0004] Prior art 2, application number CN202510034829.1, discloses a method and system for quality inspection of perforated parts. The method includes: a laser passing through a small hole in the workpiece to form a diffraction image, which appears as multiple layers of bright circular rings; projecting the diffraction image onto an imaging unit; the imaging unit transmitting the diffraction image to a processing unit; the processing unit extracting the bright spot center region of the diffraction image, performing a polar coordinate transformation on the original diffraction image based on the sub-pixel points of the bright spot center region to obtain a horizontal stripe image, extracting the midline sub-pixel coordinates of the horizontal stripe image, converting the extracted midline sub-pixel coordinates back to the original Cartesian image coordinates to obtain the trajectory midline of each ring, and calculating the actual center, actual roundness, and actual diameter of the ring based on the trajectory midline of each ring; and an analysis unit analyzing the actual center, actual roundness, and actual diameter to obtain defect information of the workpiece to be inspected. Although this method can efficiently and accurately detect minute defects, sub-pixel level defect detection through post-processing is only suitable for analyzing specific geometric features such as rings.

[0005] Existing technology three, application number: CN202411675121.6, discloses a method for measuring whole-layer atmospheric turbulence based on centroid drift and light intensity scintillation, belonging to the field of space-to-ground laser communication technology. The atmospheric turbulence parameter measurement steps include: using a high-resolution imaging lens and a high-speed visible light camera as receiving devices to acquire starlight spot images after atmospheric turbulence; then, edge detection of the star image is performed using Zernike moments to accurately analyze and identify shape features in the star image, achieving sub-pixel-level edge detection; based on this, dynamic Lanzos interpolation is performed on the star point edge regions to achieve resampling and preserve star image details; the normalized amplitude enhancement weighting method of grayscale gradient is used to improve the accuracy of centroid positioning, enhance anti-interference capability, and achieve higher positioning accuracy; finally, the whole-layer atmospheric turbulence intensity in the atmospheric channel transmission conditions during space-to-ground laser communication is obtained by calculating the light intensity scintillation variance and centroid drift variance. Although this method improves the robustness, accuracy, and convenience of whole-layer atmospheric turbulence measurement, it relies on algorithms to compensate for environmental disturbances, resulting in poor hardware-level anti-interference capability.

[0006] Current technologies 1, 2, and 3 suffer from significant limitations on image resolution due to the diffraction limit of the optical system and mechanical precision, relying on hardware-level anti-interference systems for post-processing, and addressing the issue of sub-pixel-level lossless conversion from digital design to physical devices. Therefore, this invention provides a method and system for sparse compression reconstruction of LDI sub-pixel images. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method for sparse compression reconstruction of LDI subpixel images, comprising the following steps:

[0008] At least two phase structures are partitioned on at least one rotatable transparent glass slide, and the rotation frequency of the transparent glass slide is synchronized with the pattern refresh frequency of the digital micromirror; a low-resolution image to be exposed corresponding to each phase structure is pre-calculated using a computer image compression algorithm.

[0009] The calculated low-resolution image to be exposed is loaded onto the digital micromirror in a time sequence, and the display time of each pattern is matched with the rotation position of the corresponding phase structure. The pattern of the digital micromirror and the phase-encoded mask are coupled in the frequency domain through a 4-f optical system, and the high-resolution exposure pattern at the sub-pixel level is reconstructed by using the diffraction effect.

[0010] The high-resolution light field of the reconstructed high-resolution exposure pattern is projected onto the photoresist surface, and the target circuit pattern is formed by accumulating the exposure dose; after development, a micro-nano structure with sub-pixel precision is obtained, completing the entire process of conversion from computation to physical pattern.

[0011] Optionally, the process of pre-calculating the low-resolution image to be exposed corresponding to each phase structure includes the following steps:

[0012] The optical transfer characteristics of each phase structure on the rotating transparent glass slide are quantified, and a model of its correspondence with spatial frequency is established. Each phase structure is regarded as an individual frequency domain filter, and its rotation position forms a fixed mapping with a specific frequency band component.

[0013] Based on the frequency domain characteristics of the phase structure, a set of overcomplete basis function libraries is generated. Each basis function corresponds to the modulation mode of the phase structure on the incident light at a specific rotation angle. The combination of basis functions that best characterizes the frequency domain features of the target pattern is selected by sparse optimization algorithm.

[0014] The target high-resolution image is decomposed into multiple sub-band components in the frequency domain. Matching and tracking are performed based on the obtained basis function library. The frequency domain component that best matches the modulation characteristics of the current phase structure is extracted first. The remaining residual image is transferred to the matching process of the next phase structure. The process is repeated until all phase structures are matched or the residual is lower than the set threshold.

[0015] The frequency domain components obtained by decomposition are converted into low-resolution control instructions according to the following rules. Each component corresponds to a display frame of the digital micromirror. The frame display timing is strictly synchronized with the rotation phase of the phase structure. The frame content is converted into a realizable micromirror array state through deconvolution processing.

[0016] Optionally, the process of selecting the combination of basis functions that best characterizes the frequency domain features of the target image using a sparse optimization algorithm includes the following steps:

[0017] Using the established correspondence model, the optical transmission characteristics at each rotation angle are converted into N-dimensional feature vectors, with each dimension of the vector corresponding to the energy coupling coefficient of a specific spatial frequency; the feature vectors of all phase structures at all rotation positions constitute the original basis function space, whose dimension is consistent with the frequency domain decomposition depth of the target image.

[0018] For the frequency domain sub-band components of the target image, calculate the projection energy ratio between each basis function and the current sub-band component; set a dynamic energy threshold function, whose independent variables include: the main frequency band weight of the correspondence model corresponding to the current rotation angle of the phase structure, the cumulative frequency domain energy of the previous phase structure that has been matched, and only retain basis functions whose projection energy ratio exceeds the threshold to enter the candidate set;

[0019] Set dual constraints on candidate basis functions, including positive and negative constraints: use the selected basis functions as the initial population and perform frequency domain topology adjustment: the output basis function combination satisfies the condition that the joint frequency domain coverage of the combination reaches more than a preset percentage of the total energy of the target image.

[0020] Optional, positive constraint: the main lobe direction of the basis function must have an angle less than the maximum modulation direction of the current phase structure that is less than the set tolerance; negative constraint: the side lobe energy distribution of the basis function must not have a spectral overlap of more than 5% with the matched frequency domain components.

[0021] Optionally, the process of setting the dynamic energy threshold function includes the following steps:

[0022] Based on the established correspondence model, the distribution of the main frequency band weights under the current phase structure rotation angle is extracted, reflecting the modulation efficiency advantage of the target frequency band energy at a specific rotation angle; the main frequency band weights serve as the basic adjustment factor of the threshold function, and in the initial screening stage, the efficient basis function that best matches the current rotation angle is preferentially retained.

[0023] The cumulative frequency domain energy distribution of the preceding phase structure is introduced as a dynamic correction term of the threshold function, and the energy ratio of the covered frequency domain sub-bands is updated in real time, so that the threshold function automatically improves the screening sensitivity of the remaining unmatched high-frequency or low-energy sub-bands as the optimization process progresses.

[0024] The ratio of the projected energy of the basis function to the current sub-band component is compared with a dynamic threshold. The main frequency band weight dominates the initial value of the threshold, and the basis function that is strongly correlated with the current rotation angle is given priority. The cumulative energy distribution dynamically lowers the threshold corresponding to the covered frequency band, forcing the algorithm to migrate to the unmatched frequency band. Only when the projected energy ratio of the basis function satisfies both constraints is it included in the candidate set.

[0025] Optionally, the process of reconstructing a sub-pixel-level high-resolution exposure pattern using diffraction effects includes the following steps:

[0026] The synchronous frequency timing of phase structure rotation and micromirror refresh is obtained. Each phase structure activates the corresponding pre-calculated low-resolution image to be exposed at a specific rotation angle. The spatial distribution characteristics of the phase structure and the micromirror pattern are conjugate on the object plane of the 4-f system, forming a space-frequency joint modulation unit.

[0027] By utilizing the Fourier transform characteristics of the 4-f system, the modulated optical field is decomposed into the frequency domain. Each phase structure rotation position corresponds to a set of characteristic basis functions, and its main lobe direction is uniquely determined by the current rotation angle. The frequency domain components of the micromirror pattern are projected and matched with the phase structure basis functions, and only basis function components whose projected energy exceeds the dynamic threshold are retained.

[0028] The filtered frequency domain components are inversely transformed to the image plane by a 4-f system. The basis function groups activated at different rotation angles form multiple low-resolution light fields with sub-pixel phase differences on the image plane. During the exposure time, they are incoherently superimposed through diffraction effects. The superposition result is equivalent to frequency domain interpolation reconstruction of the target high-resolution image.

[0029] Optionally, the discretized phase delay generated by the phase structure partitioning layout in the space-frequency joint modulation unit performs directional diffraction on the incident light field; the binary amplitude modulation of the micro-mirror pattern and the phase delay field are coupled by complex multiplication in the frequency domain.

[0030] Optionally, the process of projecting and matching the frequency domain components of the micromirror pattern with the phase structure basis functions includes the following steps:

[0031] The modulated optical field is decomposed into the frequency domain, and the main lobe direction characteristics of the phase structure rotation angle are locked. The basis functions corresponding to each rotation angle are converted into frequency domain feature vectors by the spatial distribution of the established space-frequency joint modulation unit and phase structure through a 4-f system.

[0032] The frequency domain components of the micromirror pattern are selected through a dynamic energy gate. When the phase structure rotates to a predetermined angle, the main lobe of its generated basis function scans the frequency domain components of the micromirror pattern like a probe. The projection energy threshold is used as the matching criterion: the basis function is activated only when the frequency domain energy covered by the main lobe of the basis function exceeds the set dynamic threshold.

[0033] The retained basis function components carry dual information: they contain both the rotation parameters of the phase structure and the local frequency domain characteristics of the micromirror pattern. During the inverse transform, they spontaneously reconstruct a low-resolution optical field with a specific phase difference.

[0034] This invention provides a sparse compression reconstruction system for LDI subpixel images, comprising:

[0035] The image compression module is responsible for partitioning at least two phase structures on at least one rotatable transparent glass slide, with the rotation frequency of the transparent glass slide synchronized with the pattern refresh frequency of the digital micromirror; and pre-calculating the low-resolution image to be exposed corresponding to each phase structure using a computer image compression algorithm.

[0036] The pattern matching module is responsible for loading the calculated low-resolution image to be exposed onto the digital micromirror in a time sequence. The display time of each pattern is matched with the rotation position of the corresponding phase structure. The pattern of the digital micromirror and the phase-encoded mask are coupled in the frequency domain through a 4-f optical system, and the high-resolution exposure pattern at the sub-pixel level is reconstructed by using the diffraction effect.

[0037] The patterning module is responsible for projecting the high-resolution light field of the reconstructed high-resolution exposure pattern onto the photoresist surface, forming the target circuit pattern by accumulating the exposure dose; after development, it obtains a micro-nano structure with sub-pixel precision, completing the entire process of conversion from computation to physical pattern.

[0038] The present invention provides an application of a sparse compression reconstruction method for LDI subpixel images, comprising: a phase encoding design stage, a dynamic light field reconstruction stage, and a circuit pattern transfer stage.

[0039] This invention relates to a processing method for achieving higher resolution imaging in laser direct imaging (LDI) intelligent devices. The main problem it solves is enabling digital micromirrors with larger pixel sizes to achieve sub-pixel-scale imaging onto printed circuit boards within laser direct imaging devices, thereby reducing the manufacturing cost of these devices while maintaining the same level of precision. This invention provides a sparse compression reconstruction method for LDI sub-pixel images, overcoming this size limitation. This can be considered as improving manufacturing precision while maintaining existing equipment costs, or reducing the cost of LDI devices while maintaining the same manufacturing precision.

[0040] Other features and advantages of the invention will be set forth in the following description, 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 particularly pointed out in the written description and the accompanying drawings.

[0041] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0042] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0043] Figure 1This is a flowchart of the sparse compression reconstruction method for LDI subpixel images in Embodiment 1 of the present invention;

[0044] Figure 2 This is a process diagram of pre-calculating the low-resolution image to be exposed corresponding to each phase structure in Embodiment 2 of the present invention;

[0045] Figure 3 This is a process diagram of reconstructing a sub-pixel-level high-resolution exposure pattern using diffraction effect in Embodiment 5 of the present invention;

[0046] Figure 4 This is a process diagram of forming the target circuit pattern by accumulating exposure dose in Embodiment 8 of the present invention;

[0047] Figure 5 This is a block diagram of the processing system for sparse compression reconstruction of LDI subpixel images in Embodiment 9 of the present invention;

[0048] Figure 6 This is a schematic diagram showing the uneven light intensity on the flat top of the exposure lines in Embodiment 9 of the present invention. Detailed Implementation

[0049] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0050] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0051] In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application. In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0052] Example 1: As Figure 1 As shown, this embodiment of the invention provides a method for sparse compression reconstruction of LDI subpixel images, comprising the following steps:

[0053] S100: At least two phase structures are set in sections on at least one rotatable transparent glass slide, and the rotation frequency of the transparent glass slide is synchronized with the pattern refresh frequency of the digital micromirror; a low-resolution image to be exposed corresponding to each phase structure is pre-calculated using a computer image compression algorithm;

[0054] S200: The calculated low-resolution image to be exposed is loaded onto the digital micromirror in a time sequence, and the display time of each pattern is matched with the rotation position of the corresponding phase structure; the pattern of the digital micromirror and the phase-encoded mask are coupled in the frequency domain through a 4-f optical system, and the high-resolution exposure pattern at the sub-pixel level is reconstructed by using the diffraction effect.

[0055] S300: Projects the high-resolution light field of the reconstructed high-resolution exposure pattern onto the photoresist surface, and forms the target circuit pattern by accumulating the exposure dose; after development, it obtains a micro-nano structure with sub-pixel precision, completing the entire process conversion from calculation to physical pattern.

[0056] The working principle and beneficial effects of the above technical solution are as follows: Firstly, at least two phase structures are partitioned on at least one rotatable transparent glass slide, with the rotation frequency of the transparent glass slide synchronized with the pattern refresh frequency of the digital micromirror. Using a computer image compression algorithm, a low-resolution image to be exposed corresponding to each phase structure is pre-calculated. Secondly, the calculated low-resolution image to be exposed is loaded onto the digital micromirror in a time sequence, with the display time of each pattern matching the rotation position of the corresponding phase structure. A 4-f optical system couples the pattern of the digital micromirror with the phase-encoded mask in the frequency domain, reconstructing a sub-pixel-level high-resolution exposure pattern using diffraction effects. Finally, the high-resolution light field of the reconstructed high-resolution exposure pattern is projected onto the photoresist surface, and the target circuit pattern is formed by accumulating the exposure dose. After development, a sub-pixel precision micro / nano structure is obtained, completing the entire process from computation to physical pattern conversion. The above solution achieves super-diffraction-limited photolithography pattern fabrication through system-level synergy between computational optics and dynamic light field modulation; by establishing synchronous control of phase structure rotation and DMD pattern refresh, a time-controlled optical transfer function is constructed. Rotational modulation of the multi-phase structure forms an overcomplete basis in the frequency domain, breaking the Abbe diffraction limit through diffraction effects. Photolithography process compatibility optimization employs dose control using time-space joint modulation.

[0057] This embodiment achieves an increase in equivalent numerical aperture while maintaining the simplicity of the optical system through deep coupling of hardware (rotating phase plate + DMD + 4f system) and algorithm (compressed sensing reconstruction).

[0058] This embodiment describes a processing method for achieving higher resolution imaging in a laser direct imaging (LDI) intelligent device. The main problem it addresses is enabling sub-pixel-sized imaging of a large-pixel-size digital micromirror (DMD) onto a printed circuit board (PCB) within a laser direct imaging device, thereby reducing the manufacturing cost of the device while maintaining the same precision. This invention provides a sparse compression reconstruction method for LDI sub-pixel images, overcoming this size limitation. It can be considered as improving manufacturing precision while maintaining existing equipment costs, or reducing the cost of LDI devices while maintaining the same manufacturing precision.

[0059] Example 2: Figure 2 As shown, based on Example 1, the process of pre-calculating the low-resolution image to be exposed corresponding to each phase structure provided by this embodiment of the invention includes the following steps:

[0060] S101: Quantify the optical transfer characteristics of each phase structure on the rotating transparent glass slide, establish its correspondence model with spatial frequency, and regard each phase structure as an individual frequency domain filter, with its rotation position forming a fixed mapping with specific frequency band components.

[0061] S102: Based on the frequency domain characteristics of the phase structure, a set of overcomplete basis function libraries is generated. Each basis function corresponds to the modulation mode of the phase structure on the incident light at a specific rotation angle. The combination of basis functions that best characterizes the frequency domain features of the target pattern is selected through a sparse optimization algorithm.

[0062] S103: Decompose the target high-resolution image into multiple sub-band components in the frequency domain, perform matching and tracking based on the obtained basis function library, prioritize extracting the frequency domain component that best matches the modulation characteristics of the current phase structure, and transfer the remaining residual image to the matching process of the next phase structure. Repeat until all phase structures are matched or the residual is lower than the set threshold.

[0063] S104: Convert the frequency domain components obtained by decomposition into low-resolution control instructions according to the following rules. Each component corresponds to a display frame of the digital micromirror. The frame display timing is strictly synchronized with the rotation phase of the phase structure. The frame content is converted into an achievable micromirror array state through deconvolution processing.

[0064] The working principle and beneficial effects of the above technical solution are as follows: First, the optical transfer characteristics of each phase structure on the rotating transparent glass slide are quantified, and a correspondence model between the phase structure and the spatial frequency is established. Each phase structure is regarded as an individual frequency domain filter, and its rotation position forms a fixed mapping with a specific frequency band component. Second, based on the frequency domain characteristics of the phase structure, a set of overcomplete basis function libraries is generated. Each basis function corresponds to the modulation mode of the phase structure on the incident light at a specific rotation angle. The basis function combination that best represents the frequency domain characteristics of the target image is selected by a sparse optimization algorithm. Then, the target high-resolution image is decomposed into multiple sub-band components in the frequency domain. Matching and tracking are performed according to the obtained basis function library. The frequency domain component that best matches the modulation characteristics of the current phase structure is extracted first. The remaining residual image is transferred to the matching process of the next phase structure. The process is repeated until all phase structures are matched or the residual is lower than a set threshold. Finally, the frequency domain components obtained by decomposition are converted into low-resolution control instructions according to the following rules. Each component corresponds to a display frame of the digital micromirror. The frame display timing is strictly synchronized with the rotation phase of the phase structure. The frame content is converted into a realizable micromirror array state through deconvolution processing. The above scheme achieves computational optical compression and physical reconstruction of high-resolution lithographic patterns through the synergistic effects of quantizing the optical properties of the phase structure, constructing a frequency domain basis function library, sparse decomposition of the target image, and spatiotemporal coding conversion. By deeply coupling the dynamic modulation characteristics of the rotating phase structure with the spatiotemporal coding of the digital micromirror, and through the fixed mapping relationship between the rotation position of the phase structure and the spatial frequency, it ensures that each low-resolution control frame accurately matches the frequency domain filtering characteristics of the current phase structure, thereby achieving sparse coding of sub-pixel information in the optical frequency domain. Based on the target image decomposition using an overcomplete basis function library, high-frequency information is adaptively allocated to the exposure timing corresponding to different phase structures. A residual iteration mechanism ensures that all frequency band components are effectively captured, equivalent to constructing an inverse system of a time-varying optical transfer function, enabling the superposition of diffraction fields from multiple low-resolution frames to overcome the physical resolution limitations of the digital micromirror. The rotation synchronization of the phase structure and the frame refresh of the digital micromirror form a closed-loop control. The decomposed frequency domain components are directly converted into exposure commands that strictly match the hardware timing, ensuring that each calculated low-resolution image satisfies the physical constraints of the subsequent 4-f system frequency domain coupling, avoiding spectral aliasing in the reconstructed light field. The sparse optimization algorithm prioritizes encoding frequency domain components with concentrated energy, significantly reducing the number of basis functions to be processed while ensuring reconstruction accuracy. Combined with residual threshold control, a controllable trade-off between computational complexity and graphic fidelity is achieved, ensuring that the method is suitable for the rapid fabrication of large-scale integrated circuit graphics.

[0065] In summary, this embodiment expands the effective degrees of freedom of the optical system through dynamic modulation of the rotating phase structure, transforms the resolution bottleneck of traditional spatial light modulators into a time-frequency joint optimization problem, and ultimately achieves a pattern generation capability on the photoresist surface that surpasses the limitations of hardware pixels.

[0066] Example 3: Based on Example 2, the process of selecting the basis function combination that best represents the frequency domain features of the target image using a sparse optimization algorithm, provided in this embodiment of the invention, includes the following steps:

[0067] S1011: Using the established correspondence model, the optical transmission characteristics at each rotation angle are converted into N-dimensional feature vectors, with each dimension of the vector corresponding to the energy coupling coefficient of a specific spatial frequency; the feature vectors of all phase structures at all rotation positions constitute the original basis function space, the dimension of which is consistent with the frequency domain decomposition depth of the target image.

[0068] S1012: For the frequency domain sub-band components of the target image, calculate the projection energy ratio of each basis function with the current sub-band component; set a dynamic energy threshold function, whose independent variables include: the main frequency band weight of the correspondence model corresponding to the current rotation angle of the phase structure, the cumulative frequency domain energy of the previous phase structure that has been matched, and only retain basis functions whose projection energy ratio exceeds the threshold to enter the candidate set.

[0069] S1013: Set dual constraints on candidate basis functions, including positive and negative constraints: use the selected basis functions as the initial population and perform frequency domain topology adjustment: the output basis function combination satisfies the condition that the joint frequency domain coverage of the combination reaches more than the preset percentage of the total energy of the target image.

[0070] For the positive constraint, the main lobe direction of the basis function must have an angle less than the maximum modulation direction of the current phase structure that is less than the set tolerance; for the negative constraint, the sidelobe energy distribution of the basis function must not have a spectral overlap of more than 5% with the matched frequency domain components.

[0071] The working principle and beneficial effects of the above technical solution are as follows: Firstly, this embodiment utilizes the established correspondence model to convert the optical transmission characteristics at each rotation angle into an N-dimensional feature vector, where each dimension of the vector corresponds to the energy coupling coefficient of a specific spatial frequency. The feature vectors of all phase structures at all rotation positions constitute the original basis function space, whose dimension is consistent with the frequency domain decomposition depth of the target image. Secondly, for the frequency domain sub-band components of the target image, the projection energy ratio of each basis function to the current sub-band component is calculated. A dynamic energy threshold function is set, whose independent variables include: the main frequency band weight of the correspondence model corresponding to the current rotation angle of the phase structure. The algorithm first selects basis functions whose cumulative frequency domain energy already matched by the preceding phase structure and whose projected energy ratio exceeds a threshold, then applies dual constraints—positive and negative—to the candidate basis functions. The selected basis functions are used as the initial population, and frequency domain topology adjustment is performed. The output basis function combinations satisfy the condition that the combined frequency domain coverage reaches a preset percentage of the total energy of the target image. For the positive constraint, the main lobe direction of the basis function must have an angle less than a set tolerance with the maximum modulation direction of the current phase structure. For the negative constraint, the sidelobe energy distribution of the basis function must not have a spectral overlap greater than 5% with the matched frequency domain components. This scheme is based on a rotation angle-optical transfer characteristic correspondence model, quantifying the spatial modulation capability of the phase structure into multi-dimensional frequency domain coupling coefficients to form the original basis function space. Through dual selection using projected energy ratio and dynamic energy threshold, the algorithm prioritizes basis functions strongly correlated with the current sub-band energy distribution of the target image, achieving adaptive focusing of frequency domain energy. The dynamic threshold function, through main band weights and cumulative energy feedback, ensures that the selection process always iteratively converges towards the high-energy region of the unmatched frequency domain components. The dual constraint mechanism ensures the physical realizability of the basis function combination from a time-frequency joint dimension. The positive constraint forces the main lobe direction of the basis function to align with the maximum modulation direction of the phase structure, avoiding the loss of modulation efficiency in the actual system due to rotation angle deviation. The negative constraint eliminates the energy redundancy of the matched frequency domain components by suppressing sidelobe spectral overlap. The synergy of the two ensures that the final output basis function combination can meet the joint frequency domain coverage requirement with the minimum cardinality, achieving the essential goal of sparse optimization. The entire process embeds the physical constraints of the optical system (such as the rotational adjustable range of the phase structure and modulation directionality) into the mathematical optimization process. The spatial dimension of the original basis function is strictly consistent with the frequency domain decomposition depth of the target image, ensuring that the basis function combination output by the algorithm can directly drive the actual optical modulation device. The joint coverage characteristics of the basis functions output by frequency domain topology adjustment essentially reflect the maximum realizable reconstruction fidelity of the optical system for the frequency domain features of the target image.

[0072] In summary, this embodiment generates a combination of basis functions that combines high representation efficiency (low number of basis functions) and high reconstruction accuracy (meeting frequency domain coverage) through frequency domain energy-guided sparse screening and dual-constraint optimization, providing the optimal frequency domain operation basis for optical imaging or computational imaging tasks.

[0073] Example 4: Based on Example 3, the process of setting the dynamic energy threshold function provided in this embodiment of the invention includes the following steps:

[0074] S10121: Based on the established correspondence model, the distribution of the main frequency band weight under the current phase structure rotation angle is extracted to reflect the modulation efficiency advantage of the target frequency band energy at a specific rotation angle; the main frequency band weight serves as the basic adjustment factor of the threshold function, and in the initial screening stage, the efficient basis function that best matches the current rotation angle is preferentially retained.

[0075] S10122: The cumulative frequency domain energy distribution of the matched preceding phase structure is introduced as a dynamic correction term of the threshold function, and the energy ratio of the covered frequency domain sub-bands is updated in real time, so that the threshold function automatically improves the screening sensitivity of the remaining unmatched high-frequency or low-energy sub-bands as the optimization process progresses.

[0076] S10123: Compare the ratio of the projected energy of the basis function to the current sub-band component with the dynamic threshold. The main frequency band weight dominates the initial value of the threshold, and the basis function that is strongly correlated with the current rotation angle is given priority. The cumulative energy distribution dynamically lowers the threshold corresponding to the covered frequency band, forcing the algorithm to migrate to the unmatched frequency band. Only when the projected energy ratio of the basis function satisfies both constraints can it enter the candidate set.

[0077] The working principle and beneficial effects of the above technical solution are as follows: First, based on the established correspondence model, the main frequency band weight distribution under the current phase structure rotation angle is extracted to reflect the modulation efficiency advantage of the target frequency band energy at a specific rotation angle. The main frequency band weight serves as the basic adjustment factor of the threshold function, prioritizing the retention of the efficient basis function that best matches the current rotation angle in the initial screening stage. Second, the cumulative frequency domain energy distribution of the preceding phase structure that has been matched is introduced as a dynamic correction term of the threshold function, updating the energy proportion of the covered frequency sub-bands in real time, so that the threshold function automatically improves the screening sensitivity of the remaining unmatched high-frequency or low-energy sub-bands as the optimization process progresses. Finally, the projection energy ratio of the basis function and the current sub-band component is compared with the dynamic threshold, with the main frequency band weight dominating the initial value of the threshold, and the basis function that is strongly correlated with the current rotation angle passing first. The cumulative energy distribution dynamically lowers the threshold corresponding to the covered frequency band, forcing the algorithm to migrate to the unmatched frequency band. Only when the projection energy ratio of the basis function satisfies both constraints is it entered into the candidate set. The design process of the dynamic energy threshold function in the above scheme achieves adaptive optimization of frequency domain energy screening through multi-step collaboration. By combining the main frequency band weight distribution and the cumulative frequency domain energy feedback, the algorithm retains the optimal modulation efficiency basis function under the current rotation angle while dynamically adjusting the screening direction according to the matched energy. This ensures that the basis function candidate set always contains the basis functions with the greatest representational potential for the frequency bands not covered by the target image, avoiding frequency domain energy distribution imbalance caused by a fixed threshold. The dynamic threshold comparison of the projected energy ratio combines the local optimization characteristics of the main frequency band weight with the global convergence characteristics of the cumulative energy. The main frequency band weight ensures that the high-energy basis functions that match the current physical state (rotation angle) of the phase structure are captured first in the initial screening stage. The cumulative energy distribution forces the algorithm to gradually shift to the energy supplementation of low-frequency or high-resolution sub-bands by suppressing the threshold of the saturated frequency band in real time. The two work together to drive the basis function candidate set to approach the complete frequency domain coverage requirement of the target image with the minimum number of iterations. The entire process is coupled with a feedforward-feedback mechanism based on the energy threshold. The feedforward path uses the known correspondence between the rotation angle and the modulation efficiency to quickly narrow the search range. The feedback path (S10122) avoids repeated calculations of the optimized frequency band by using historical matching data. The final output candidate set satisfies both the physical feasibility of real-time control of the optical system (guaranteed by the main frequency band weight) and the integrity requirement of the frequency domain energy distribution (guided by the accumulated energy), providing a high-purity input population for subsequent dual-constraint optimization.

[0078] In summary, this embodiment achieves dynamic adaptation of the basis function selection process, the physical characteristics of the optical system, and the frequency domain features of the target image by controlling the rotation angle correlation and historical energy distribution in a two-dimensional manner, thereby significantly reducing computational complexity while ensuring the accuracy of frequency domain reconstruction.

[0079] Example 5: Figure 3As shown, based on Example 1, the process of reconstructing a sub-pixel-level high-resolution exposure pattern using diffraction effect provided by this embodiment of the invention includes the following steps:

[0080] S201: Obtain the synchronization frequency timing of phase structure rotation and micro-mirror refresh. Each phase structure activates the corresponding pre-calculated low-resolution image to be exposed at a specific rotation angle. The spatial distribution characteristics of the phase structure and the micro-mirror pattern are conjugate on the 4-f system object plane to form a space-frequency joint modulation unit.

[0081] The discretized phase delay generated by the phase structure partitioning layout enables directional diffraction of the incident light field; the binary amplitude modulation of the micro-mirror pattern and the phase delay field are coupled by complex multiplication in the frequency domain.

[0082] S202: Through the Fourier transform characteristics of the 4-f system, the modulated light field is decomposed into the frequency domain. Each phase structure rotation position corresponds to a set of characteristic basis functions, and its main lobe direction is uniquely determined by the current rotation angle. The frequency domain components of the micro-mirror pattern are projected and matched with the phase structure basis functions, and only the basis function components whose projected energy exceeds the dynamic threshold are retained.

[0083] S203: The filtered frequency domain components are inversely transformed to the image plane by the 4-f system. The basis function groups activated at different rotation angles form multiple low-resolution light fields with sub-pixel phase differences on the image plane. During the exposure time, they are incoherently superimposed through diffraction effect. The superposition result is equivalent to frequency domain interpolation reconstruction of the target high-resolution image.

[0084] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first obtains the synchronous frequency timing of the phase structure rotation and the micromirror refresh. Each phase structure activates the corresponding pre-calculated low-resolution image to be exposed at a specific rotation angle. The spatial distribution characteristics of the phase structure and the micromirror pattern are conjugate on the object plane of the 4-f system, forming a space-frequency joint modulation unit. The discretized phase delay generated by the partitioned layout of the phase structure performs directional diffraction on the incident light field. The binary amplitude modulation of the micromirror pattern and the phase delay field achieve complex multiplication coupling in the frequency domain. Secondly, through the Fourier transform characteristics of the 4-f system, the... The modulated light field is decomposed into the frequency domain. Each phase structure rotation position corresponds to a set of characteristic basis functions, and its main lobe direction is uniquely determined by the current rotation angle. The frequency domain components of the micromirror pattern are projected and matched with the phase structure basis functions, retaining only the basis function components whose projected energy exceeds the dynamic threshold. Finally, the filtered frequency domain components are inversely transformed to the image plane by a 4-f system. The basis function sets activated at different rotation angles form multiple low-resolution light fields with sub-pixel phase differences on the image plane. During the exposure time, they undergo incoherent superposition through diffraction effects, and the superposition result is equivalent to frequency domain interpolation reconstruction of the target high-resolution image. Through the synergistic effect of steps S201-S203, the above scheme achieves sub-pixel-level lithographic pattern reconstruction based on space-frequency joint modulation and dynamic diffraction synthesis. Through precise synchronization of phase structure rotation and micromirror refresh, the pre-calculated low-resolution image sequence is transformed into a time-varying diffraction modulation field. This process leverages the frequency domain operation characteristics of the 4-f system to achieve a breakthrough in optical field reconstruction equivalent to high-frequency interpolation, within the limitations of optical hardware physical resolution, through frequency domain screening and synthesis of multi-angle basis functions. Essentially, it maps the frequency domain optimization results of digital computation to the actual optical field distribution through diffraction physics. The combined effect of phase structure partitioning and micromirror binary modulation concentrates the incident light energy to the diffraction order corresponding to the target frequency band. Combined with dynamic threshold screening to suppress energy in non-critical frequency bands, the system retains only the diffraction components effective for sub-pixel reconstruction. Finally, through incoherent superposition on the image plane, it achieves exposure pattern generation with maximized energy utilization, significantly reducing energy loss caused by frequency domain redundancy in traditional methods. The time-varying diffraction field brought about by the rotating phase structure and the spatially distributed micromirror pattern together constitute a spatiotemporal dual encoding. In the frequency domain decomposition stage, the system characteristics effectively suppress static optical noise; in the image plane reconstruction stage, multi-frame incoherent superposition further eliminates random phase errors, making the reconstruction process inherently robust to environmental disturbances and ensuring the stable achievement of sub-pixel level accuracy.

[0085] In summary, this embodiment achieves hardware-independent super-resolution exposure capability by unifying the forward prediction of computational optics and the reverse constraint of diffraction physics into a single optical path through the spatial-frequency synergy of programmable phase modulation and binary amplitude modulation.

[0086] Example 6: Based on Example 5, the process of projecting and matching the frequency domain components of the micromirror pattern with the phase structure basis function provided in this embodiment of the invention includes the following steps:

[0087] S2021: Obtain the optical field decomposed to the frequency domain, the main lobe direction characteristics of the phase structure rotation angle locked basis function, the basis function corresponding to each rotation angle, and the spatial distribution of the established space-frequency joint modulation unit and phase structure are converted into frequency domain feature vectors by the 4-f system.

[0088] S2022: The frequency domain components of the micromirror pattern are selected through a dynamic energy gate. When the phase structure rotates to a predetermined angle, the main lobe of the basis function it generates scans the frequency domain components of the micromirror pattern like a probe. The projection energy threshold is used as the matching criterion: the basis function will only be activated when the frequency domain energy covered by the main lobe of the basis function exceeds the set dynamic threshold.

[0089] S2023: The retained basis function components carry dual information, including both the rotation parameters of the phase structure and the local frequency domain characteristics of the micromirror pattern, and spontaneously reconstruct a low-resolution optical field with a specific phase difference during inverse transformation.

[0090] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the rotation angle of the phase structure first locks the main lobe direction characteristics of the basis function. The basis function corresponding to each rotation angle is converted into a frequency domain feature vector by the established space-frequency joint modulation unit and the spatial distribution of the phase structure through a 4-f system. Secondly, the frequency domain components of the micromirror pattern achieve feature selection through a dynamic energy gate. When the phase structure rotates to a predetermined angle, the main lobe of the generated basis function will scan the frequency domain components of the micromirror pattern like a probe. The projection energy threshold is used as the matching criterion: the basis function will only be activated when the frequency domain energy covered by the main lobe of the basis function exceeds the set dynamic threshold. Finally, the retained basis function components carry dual information, including both the rotation parameters of the phase structure and the local frequency domain characteristics of the micromirror pattern, and spontaneously reconstruct a low-resolution light field with a specific phase difference during the inverse transformation. The above scheme achieves feature decoupling and reconstruction under joint spatial-frequency modulation through the synergistic effect of rotation angle-constrained main lobe orientation detection of the basis function, dynamic energy threshold screening, and a dual information retention mechanism. The linkage matching between the phase structure rotation angle and the micromirror pattern frequency components ensures that the basis function main lobe captures only frequency components consistent with the current spatial modulation state, avoiding irrelevant noise interference and improving the accuracy of feature extraction. The projection energy threshold, as a matching criterion, automatically filters low-energy noise components, retaining only effective frequency components and optimizing the signal-to-noise ratio in subsequent reconstruction. Simultaneously, this threshold mechanism, synchronized with the rotation angle, forms a closed-loop control, ensuring that the frequency feature selection strictly corresponds to the spatial modulation state. The activated basis function components simultaneously encode the rotation parameters of the phase structure and the local frequency features of the micromirror pattern, enabling the optical field reconstruction in the inverse transform stage to not only recover the original amplitude information but also maintain a precise phase difference distribution, thereby achieving sub-pixel-level resolution enhancement.

[0091] In summary, this embodiment, through the tight coupling of space-frequency joint modulation, dynamic energy screening, and dual information preservation, enables the system to efficiently extract key features in the frequency domain and accurately reconstruct them in the spatial domain, ultimately achieving high-resolution optical field synthesis under diffraction effects.

[0092] Example 7: Based on Example 6, the process of decomposing the modulated optical field to the frequency domain provided in this embodiment of the invention includes the following steps:

[0093] S20211: After the spatial phase delay distribution corresponding to each rotation angle is transformed by the 4-f system front-end, it forms a cluster of characteristic basis functions with specific orientations in the frequency domain. The main lobe direction of the basis functions is controlled by the rotation angle, so that the frequency domain decomposition has a deterministic mapping relationship between angle and frequency.

[0094] S20212: The binary modulation information of the micromirror is encoded into the frequency domain energy distribution. When the composite light field output by the space-frequency joint modulation unit passes through the front lens of the 4-f system, its spatial distribution is converted into a frequency domain characteristic spectrum, which contains two key components: the basis function framework generated by the phase structure and the energy filling mode of the micromirror pattern modulation, which together constitute the input conditions for subsequent projection matching.

[0095] The symmetric optical architecture of the S20213:4-f system ensures a geometric correspondence between the spatial phase delay and the frequency-domain basis function orientation, while the amplitude modulation information of the micromirror is losslessly converted to the frequency-domain energy envelope.

[0096] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, firstly, the spatial phase delay distribution corresponding to each rotation angle is transformed by the front-end of the 4-f system to form a cluster of characteristic basis functions with specific orientations in the frequency domain. The main lobe direction of the basis functions is controlled by the rotation angle, so that the frequency domain decomposition has a deterministic mapping relationship between angle and frequency. Secondly, the binary modulation information of the micromirror is encoded into the frequency domain energy distribution. When the composite light field output by the space-frequency joint modulation unit passes through the front lens of the 4-f system, its spatial distribution is converted into a frequency domain characteristic spectrum, which contains two key components: the basis function framework generated by the phase structure and the energy filling mode of the micromirror pattern modulation, which together constitute the input conditions for subsequent projection matching. Finally, the symmetrical optical architecture of the 4-f system ensures that the spatial phase delay and the orientation of the frequency domain basis functions satisfy a geometric correspondence, while the amplitude modulation information of the micromirror is losslessly converted into the frequency domain energy envelope. The process of decomposing the modulated optical field to the frequency domain in the above scheme achieves a precise conversion of spatial modulation information to frequency domain features through the synergistic effect of the generation of basis function clusters constrained by rotation angle, frequency domain encoding of binary modulation information, and the fidelity transfer mechanism of the optical architecture. The strict correspondence between the phase structure rotation angle and the main lobe direction of the basis functions ensures a predictable geometric correlation between the spatial modulation parameters and the frequency domain decomposition results, providing a structured basis function framework for projection matching. The binary amplitude modulation of the micromirror and the spatial delay of the phase structure are represented in the frequency domain as energy-filling modes and basis function frameworks, respectively. The dual-channel encoding mechanism preserves the complete dimension of the original modulation information, enabling the frequency domain feature spectrum to simultaneously contain amplitude and phase features. The symmetric architecture of the 4-f system maintains a strict geometric correspondence between the spatial phase distribution and the frequency domain basis function orientation, while ensuring energy conservation of amplitude modulation information during the space-to-frequency conversion process, providing a distortion-free frequency domain input for dynamic thresholding.

[0097] In summary, this embodiment achieves accurate decomposition of the modulated light field into frequency domain features by organically combining angle-frequency mapping, dual-channel information encoding, and optical fidelity conversion. Its output has both a structured basis function framework and accurate energy distribution characteristics, laying the foundation for frequency domain analysis for frequency domain screening and sub-pixel-level reconstruction.

[0098] Example 8: As Figure 4 As shown, based on Example 1, the process of forming a target circuit pattern by accumulating exposure dose provided in this embodiment of the invention includes the following steps:

[0099] S301: The reconstructed high-resolution light field forms a multi-frame low-dose exposure pattern with sub-pixel displacement characteristics on the image plane. The intensity distribution of each frame pattern comes from the frequency domain coupling result of the corresponding phase structure and the micro-mirror pattern. The sub-pixel level light field shift between frames is controlled by the phase structure rotation timing, and the cumulative dose spatially covers all high-frequency details of the target pattern.

[0100] S302: The nonlinear response characteristics of photoresist to the projected light field convert the continuous dose distribution into a binary pattern. When the exposure dose of a single frame is lower than the photoresist threshold, the dose of multiple frames superimposed reaches the critical value and triggers a photochemical reaction. This, in conjunction with the reconstructed sub-pixel light field spatial distribution, enables the resolution of the final developed pattern to break through the diffraction limit of the single-frame optical system.

[0101] S303: The redundant exposure introduced by the phase structure rotation compensates for the dose quantization error of the micromirror discretization refresh, while the light field reconstructed by frequency domain coupling suppresses the image distortion caused by the proximity effect by controlling the dose spatial gradient of each frame.

[0102] The working principle and beneficial effects of the above technical solution are as follows: First, the reconstructed high-resolution light field forms a multi-frame low-dose exposure pattern with sub-pixel displacement characteristics on the image plane. The intensity distribution of each frame pattern comes from the frequency domain coupling result of the corresponding phase structure and the micromirror pattern. The sub-pixel level light field shift between frames is controlled by the phase structure rotation timing, and the accumulated dose spatially covers all high-frequency details of the target pattern. Second, the nonlinear response characteristics of the photoresist to the projected light field convert the continuous dose distribution into a binary pattern. When the exposure dose of a single frame is lower than the photoresist threshold, when the dose of multiple frames reaches the critical value, a photochemical reaction is triggered. This reaction works synergistically with the reconstructed sub-pixel light field spatial distribution, so that the resolution of the final developed pattern breaks through the diffraction limit of the single-frame optical system. Finally, the redundant exposure introduced by the phase structure rotation compensates for the dose quantization error of the micromirror discretization refresh, while the light field reconstructed by frequency domain coupling suppresses the pattern distortion caused by the proximity effect by controlling the dose spatial gradient of each frame. The above scheme, through the cumulative exposure dose to form the target circuit pattern, exhibits the synergistic effect of high-resolution light field modulation, nonlinear photochemical conversion, and dose error compensation mechanism, achieving physical transfer of sub-pixel precision patterns. Specifically, this manifests as: complete reconstruction of high-frequency details; the intensity distribution generated by the sub-pixel displacement light field based on phase structure rotation timing control and the frequency domain coupling of the micromirror; and the formation of a continuous dose distribution through the spatial superposition of multiple frames of low-dose exposure, ensuring the complete preservation of high-frequency components of the target pattern during cumulative exposure. Super-diffraction-limited imaging utilizes the nonlinear response of the photoresist threshold characteristics to the sub-pixel-level light field, transforming the sub-threshold dose of discrete frames into the binarized edges of the developed pattern. Its effective resolution is determined by the dose gradient of multiple superimposed frames, breaking through the spatial bandwidth limitations of a single-frame optical system. Enhanced pattern fidelity is achieved through the exposure redundancy introduced by phase structure rotation, eliminating dose step errors caused by micromirror refresh, while the spatial gradient control of the frequency domain reconstructed light field suppresses dose diffusion effects during development, jointly ensuring the edge positioning accuracy and linewidth consistency of the pattern.

[0103] In summary, this embodiment transforms the calculated subpixel light field information into a physical pattern that meets the requirements of nanoscale size control through a triple mechanism of spatiotemporal dose modulation, threshold response conversion, and system error compensation. Its core lies in utilizing the synergistic effect of optical modulation and photochemical properties to achieve lossless transfer from digital calculation to physical structure.

[0104] Example 9: As Figure 5 As shown, based on Embodiments 1-8, the sparse compression reconstruction LDI subpixel image processing system provided in this embodiment of the invention includes:

[0105] The image compression module is responsible for partitioning at least two phase structures on at least one rotatable transparent glass slide, with the rotation frequency of the transparent glass slide synchronized with the pattern refresh frequency of the digital micromirror; and pre-calculating the low-resolution image to be exposed corresponding to each phase structure using a computer image compression algorithm.

[0106] The pattern matching module is responsible for loading the calculated low-resolution image to be exposed onto the digital micromirror in a time sequence. The display time of each pattern is matched with the rotation position of the corresponding phase structure. The pattern of the digital micromirror and the phase-encoded mask are coupled in the frequency domain through a 4-f optical system, and the high-resolution exposure pattern at the sub-pixel level is reconstructed by using the diffraction effect.

[0107] The patterning module is responsible for projecting the high-resolution light field of the reconstructed high-resolution exposure pattern onto the photoresist surface, forming the target circuit pattern by accumulating the exposure dose; after development, it obtains a micro-nano structure with sub-pixel precision, completing the entire process of conversion from computation to physical pattern.

[0108] The working principle and beneficial effects of the above technical solution are as follows: The image compression module of this embodiment sets at least two phase structures in sections on at least one rotatable transparent glass slide, and the rotation frequency of the transparent glass slide is synchronized with the pattern refresh frequency of the digital micromirror; the low-resolution image to be exposed corresponding to each phase structure is pre-calculated by a computer image compression algorithm; the pattern matching module loads the calculated low-resolution image to be exposed onto the digital micromirror in a time sequence, and the display time of each pattern is matched with the rotation position of the corresponding phase structure; the pattern of the digital micromirror and the phase encoding mask are coupled in the frequency domain by a 4-f optical system, and a high-resolution exposure pattern at the sub-pixel level is reconstructed by using the diffraction effect; the pattern forming module projects the high-resolution light field of the reconstructed high-resolution exposure pattern onto the photoresist surface, and forms the target circuit pattern by accumulating the exposure dose; after development, a micro-nano structure with sub-pixel precision is obtained, completing the entire process conversion from calculation to physical pattern. The phase structure of the above scheme establishes a strict mapping relationship between physical encoding and computational image with the spatiotemporal synchronization mechanism of the digital micromirror. Phase-locked control of the rotation and refresh frequencies ensures precise matching of the corresponding compressed image for each phase window. Computer pre-computation algorithms convert the static mask in traditional photolithography into a dynamically programmable spatiotemporal encoded sequence. The 4-f system couples the digital micromirror array and phase mask in the frequency domain, overcoming the inherent resolution limitations of optical systems through diffraction effects. Frequency-domain modulation reconstructs discrete low-resolution image sequences into continuous sub-pixel exposure patterns, achieving synergistic optimization of spatial sampling rate and optical transfer function. The cumulative exposure mechanism transforms multiple frames of low-dose exposure in time into high-resolution patterns in space, and the threshold response characteristics of the photoresist convert the sub-pixel light field distribution into the physical edge morphology after development. Through computational optics, pattern transfer exceeding hardware resolution is achieved, forming a closed-loop conversion chain from digital bitstream to physical devices.

[0109] In summary, this embodiment constitutes a complete technology chain for computational optical lithography. By using temporal control, the spatial resolution constraint in traditional lithography is transformed into a computable temporal-spatial joint optimization problem, achieving super-resolution manufacturing under hardware limitations.

[0110] In this embodiment, if the phase structure used is a binary structure of 0 and π that is easy to fabricate, it will result in a flat-top light intensity non-uniformity of <1.5% in the exposed lines, such as... Figure 6 As shown, since photosensitive emulsion has a light sensitivity threshold, its exposure accuracy is determined by the contrast between light and no light. Therefore, in practical applications, this non-uniformity will not affect the accuracy of the circuit board and can be fully applied to LDI to achieve sub-pixel resolution enhancement.

[0111] Example 10: Based on Examples 1-8, the application of the sparse compression reconstruction method for LDI subpixel images provided in this embodiment of the invention includes:

[0112] During the phase encoding design phase, based on the wiring characteristics of the target circuit board, four sets of orthogonal phase structures are configured on the rotating glass slide; the rotation period of the phase structure is synchronized with the refresh rate of the digital micromirror at 120Hz; the original Gerber file is decomposed into eight sets of low-resolution mask patterns using a compressed sensing algorithm.

[0113] In the dynamic light field reconstruction stage, the digital micromirror cycles through the compressed mask pattern at a refresh rate of μs; the 4-f optical system couples the rotating phase structure with the digital micromirror pattern in the frequency domain; and reconstructs the sub-pixel light field with an equivalent step of 0.5 μm through the diffraction effect.

[0114] In the circuit pattern transfer stage, the reconstructed light field forms cumulative exposure on the dry film photoresist surface. An equivalent 5μm linewidth is achieved through 16 subpixel shift exposures. After development, a precision circuit with a minimum linewidth / line spacing of 10μm / 10μm is obtained.

[0115] The working principle and beneficial effects of the above technical solution are as follows: Breakthrough process: Compared to the 25μm limit of traditional LDI, it achieves 10μm-level line processing; exposure efficiency is increased by 3 times (multiple pattern exposures are completed in a single rotation). Cost control: High-precision exposure is achieved using low-resolution DMD devices, reducing laser power requirements (50mW can achieve the equivalent effect of 100mW). Production line adaptability: Compatible with existing PCB dry process production lines, it can directly replace traditional LDI optical modules.

[0116] This embodiment can be applied in IC substrate manufacturing: to realize the processing of microbump arrays with a 40μm pitch; to complete the manufacturing of silicon interposer boards with 10 arbitrary interconnects; and to pattern the RDL layer of wafer-level packaging (WLP).

[0117] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of equivalents of this invention, this invention is also intended to include these modifications and variations.

Claims

1. A method for sparse compression reconstruction of LDI subpixel images, characterized in that, Includes the following steps: At least two phase structures are partitioned on at least one rotatable transparent glass slide, and the rotation frequency of the transparent glass slide is synchronized with the pattern refresh frequency of the digital micromirror; a low-resolution image to be exposed corresponding to each phase structure is pre-calculated using a computer image compression algorithm. The calculated low-resolution image to be exposed is loaded onto the digital micromirror in a time sequence, and the display time of each pattern is matched with the rotation position of the corresponding phase structure. The pattern of the digital micromirror and the phase-encoded mask are coupled in the frequency domain through a 4-f optical system, and the high-resolution exposure pattern at the sub-pixel level is reconstructed by using the diffraction effect. The high-resolution light field of the reconstructed high-resolution exposure pattern is projected onto the photoresist surface, and the target circuit pattern is formed by accumulating the exposure dose; After development, subpixel precision micro-nano structures are obtained, completing the entire process of conversion from computation to physical graphics; The process of pre-calculating the low-resolution image to be exposed corresponding to each phase structure includes the following steps: The optical transfer characteristics of each phase structure on the rotating transparent glass slide are quantified, and a model of its correspondence with spatial frequency is established. Each phase structure is regarded as an individual frequency domain filter, and its rotation position forms a fixed mapping with a specific frequency band component. Based on the frequency domain characteristics of the phase structure, a set of overcomplete basis function libraries is generated. Each basis function corresponds to the modulation mode of the phase structure on the incident light at a specific rotation angle. The combination of basis functions that best characterizes the frequency domain features of the target pattern is selected by sparse optimization algorithm. The target high-resolution image is decomposed into multiple sub-band components in the frequency domain. Matching and tracking are performed based on the obtained basis function library. The frequency domain component that best matches the modulation characteristics of the current phase structure is extracted first. The remaining residual image is transferred to the matching process of the next phase structure. The process is repeated until all phase structures are matched or the residual is lower than the set threshold. The frequency domain components obtained from the decomposition are converted into low-resolution control commands. Each component corresponds to a display frame of the digital micromirror. The frame display timing is strictly synchronized with the rotation phase of the phase structure. The frame content is converted into an achievable micromirror array state through deconvolution processing.

2. The processing method for sparse compression reconstruction of LDI subpixel images as described in claim 1, characterized in that, The process of selecting the combination of basis functions that best characterizes the frequency domain features of the target image using a sparse optimization algorithm includes the following steps: Using the established correspondence model, the optical transmission characteristics at each rotation angle are converted into N-dimensional feature vectors, with each dimension of the vector corresponding to the energy coupling coefficient of a specific spatial frequency; the feature vectors of all phase structures at all rotation positions constitute the original basis function space, whose dimension is consistent with the frequency domain decomposition depth of the target image. For the frequency domain sub-band components of the target image, calculate the projection energy ratio between each basis function and the current sub-band component; set a dynamic energy threshold function, whose independent variables include: the main frequency band weight of the correspondence model corresponding to the current rotation angle of the phase structure, the cumulative frequency domain energy of the previous phase structure that has been matched, and only retain basis functions whose projection energy ratio exceeds the threshold to enter the candidate set; Set dual constraints on candidate basis functions, including positive and negative constraints: use the selected basis functions as the initial population and perform frequency domain topology adjustment: the output basis function combination satisfies the condition that the joint frequency domain coverage of the combination reaches more than a preset percentage of the total energy of the target image.

3. The processing method for sparse compression reconstruction of LDI subpixel images as described in claim 2, characterized in that, For the positive constraint, the main lobe direction of the basis function must have an angle less than the maximum modulation direction of the current phase structure that is less than the set tolerance; for the negative constraint, the sidelobe energy distribution of the basis function must not have a spectral overlap of more than 5% with the matched frequency domain components.

4. The processing method for sparse compression reconstruction of LDI subpixel images as described in claim 2, characterized in that, The process of setting the dynamic energy threshold function includes the following steps: Based on the established correspondence model, the distribution of the main frequency band weights under the current phase structure rotation angle is extracted, reflecting the modulation efficiency advantage of the target frequency band energy at a specific rotation angle; the main frequency band weights serve as the basic adjustment factor of the threshold function, and in the initial screening stage, the efficient basis function that best matches the current rotation angle is preferentially retained. The cumulative frequency domain energy distribution of the preceding phase structure is introduced as a dynamic correction term of the threshold function, and the energy ratio of the covered frequency domain sub-bands is updated in real time, so that the threshold function automatically improves the screening sensitivity of the remaining unmatched high-frequency or low-energy sub-bands as the optimization process progresses. The ratio of the projected energy of the basis function to the current sub-band component is compared with a dynamic threshold. The main frequency band weight dominates the initial value of the threshold, and the basis function that is strongly correlated with the current rotation angle is given priority. The cumulative energy distribution dynamically lowers the threshold corresponding to the covered frequency band, forcing the algorithm to migrate to the unmatched frequency band. Only when the projected energy ratio of the basis function satisfies both constraints is it included in the candidate set.

5. The processing method for sparse compression reconstruction of LDI subpixel images as described in claim 1, characterized in that, The process of reconstructing a sub-pixel-level high-resolution exposure pattern using diffraction effects includes the following steps: The synchronous frequency timing of phase structure rotation and micromirror refresh is obtained. Each phase structure activates the corresponding pre-calculated low-resolution image to be exposed at a specific rotation angle. The spatial distribution characteristics of the phase structure and the micromirror pattern are conjugate on the object plane of the 4-f system, forming a space-frequency joint modulation unit. By utilizing the Fourier transform characteristics of the 4-f system, the modulated optical field is decomposed into the frequency domain. Each phase structure rotation position corresponds to a set of characteristic basis functions, and its main lobe direction is uniquely determined by the current rotation angle. The frequency domain components of the micromirror pattern are projected and matched with the phase structure basis functions, and only basis function components whose projected energy exceeds the dynamic threshold are retained. The filtered frequency domain components are inversely transformed to the image plane by a 4-f system. The basis function groups activated at different rotation angles form multiple low-resolution light fields with sub-pixel phase differences on the image plane. During the exposure time, they are incoherently superimposed through diffraction effects. The superposition result is equivalent to frequency domain interpolation reconstruction of the target high-resolution image.

6. The processing method for sparse compression reconstruction of LDI subpixel images as described in claim 5, characterized in that, The discretized phase delay generated by the partitioned layout of the phase structure in the space-frequency joint modulation unit performs directional diffraction of the incident light field; the binary amplitude modulation of the micro-mirror pattern and the phase delay field are coupled by complex multiplication in the frequency domain.

7. The processing method for sparse compression reconstruction of LDI subpixel images as described in claim 1, characterized in that, The process of projecting and matching the frequency domain components of the micromirror pattern with the phase structure basis functions includes the following steps: The modulated optical field is decomposed into the frequency domain, and the main lobe direction characteristics of the phase structure rotation angle are locked. The basis functions corresponding to each rotation angle are converted into frequency domain feature vectors by the spatial distribution of the established space-frequency joint modulation unit and phase structure through a 4-f system. The frequency domain components of the micromirror pattern are selected through a dynamic energy gate. When the phase structure rotates to a predetermined angle, the main lobe of its generated basis function scans the frequency domain components of the micromirror pattern like a probe. The projection energy threshold is used as the matching criterion: the basis function is activated only when the frequency domain energy covered by the main lobe of the basis function exceeds the set dynamic threshold. The retained basis function components carry dual information: they contain both the rotation parameters of the phase structure and the local frequency domain characteristics of the micromirror pattern. During the inverse transform, they spontaneously reconstruct a low-resolution optical field with a specific phase difference.

8. A processing system for sparse compression reconstruction of LDI subpixel images, characterized in that, Include: The image compression module is responsible for partitioning at least two phase structures on at least one rotatable transparent glass slide, with the rotation frequency of the transparent glass slide synchronized with the pattern refresh frequency of the digital micromirror; and pre-calculating the low-resolution image to be exposed corresponding to each phase structure using a computer image compression algorithm. The process of pre-calculating the low-resolution image to be exposed corresponding to each phase structure includes: The optical transfer characteristics of each phase structure on the rotating transparent glass slide are quantified, and a model of its correspondence with spatial frequency is established. Each phase structure is regarded as an individual frequency domain filter, and its rotation position forms a fixed mapping with a specific frequency band component. Based on the frequency domain characteristics of the phase structure, a set of overcomplete basis function libraries is generated. Each basis function corresponds to the modulation mode of the phase structure on the incident light at a specific rotation angle. The combination of basis functions that best characterizes the frequency domain features of the target pattern is selected by sparse optimization algorithm. The target high-resolution image is decomposed into multiple sub-band components in the frequency domain. Matching and tracking are performed based on the obtained basis function library. The frequency domain component that best matches the modulation characteristics of the current phase structure is extracted first. The remaining residual image is transferred to the matching process of the next phase structure. The process is repeated until all phase structures are matched or the residual is lower than the set threshold. The frequency domain components obtained by decomposition are converted into low-resolution control commands. Each component corresponds to a display frame of the digital micromirror. The frame displays the rotation phase of the phase structure with strict timing synchronization. The frame content is converted into a realizable micromirror array state through deconvolution processing. The pattern matching module is responsible for loading the calculated low-resolution image to be exposed onto the digital micromirror in a time sequence. The display time of each pattern is matched with the rotation position of the corresponding phase structure. The pattern of the digital micromirror and the phase-encoded mask are coupled in the frequency domain through a 4-f optical system, and the high-resolution exposure pattern at the sub-pixel level is reconstructed by using the diffraction effect. The patterning module is responsible for projecting the high-resolution light field of the reconstructed high-resolution exposure pattern onto the photoresist surface and forming the target circuit pattern by accumulating the exposure dose. After development, subpixel-precision micro-nano structures are obtained, completing the entire process of conversion from computation to physical graphics.

9. An application of a sparse compression reconstruction method for LDI subpixel images, used to implement the sparse compression reconstruction method for LDI subpixel images according to any one of claims 1-7, characterized in that, Includes: phase encoding design stage, dynamic light field reconstruction stage, and circuit pattern transfer stage; During the phase encoding design phase, based on the wiring characteristics of the target circuit board, four sets of orthogonal phase structures are configured on the rotating glass slide; the rotation period of the phase structure is synchronized with the refresh rate of the digital micromirror at 120Hz; the original Gerber file is decomposed into eight sets of low-resolution mask patterns using a compressed sensing algorithm. In the dynamic light field reconstruction stage, the digital micromirror cycles through the compressed mask pattern at a refresh rate of μs; the 4-f optical system couples the rotating phase structure with the digital micromirror pattern in the frequency domain; and reconstructs the sub-pixel light field with an equivalent step of 0.5 μm through the diffraction effect. In the circuit pattern transfer stage, the reconstructed light field forms cumulative exposure on the dry film photoresist surface. An equivalent 5μm linewidth is achieved through 16 subpixel shift exposures. After development, a precision circuit with a minimum linewidth / line spacing of 10μm / 10μm is obtained.

Images