Process constraint based automated design method and apparatus for anti-diffraction pupil filter

Abstract

The present application relates to the field of optical technology, and provide a kind of process constraint-based anti-diffraction pupil filter automation design method and device.The process constraint-based anti-diffraction pupil filter automation design method described above, including: constructing hybrid spectrum parameter space based on the microstructure of pupil edge, hybrid spectrum parameter is mapped to the edge radial offset sequence of spatial domain;Based on the closed vector graphics of smooth edge profile, distance field is mapped to two-dimensional pupil transmittance distribution function;Based on two-dimensional pupil transmittance distribution function and Fourier optics scalar diffraction theory, construct full-link optical simulation model;Based on the gradient of composite loss function about hybrid spectrum parameter calculated by automatic differentiation framework, and the parameter is iteratively updated using numerical optimization algorithm.The process constraint-based anti-diffraction pupil filter automation design method described above successfully solves the problem that the theoretical model is disconnected from the entity processing in the general computing optical design.

Description

Technical Field

[0001] This invention relates to the field of optical technology, and in particular to an automated design method and apparatus for anti-diffraction pupil filters based on process constraints. Background Technology

[0002] In high-precision optical imaging and laser transmission systems, the pupil (or aperture stop) is a core component controlling the numerical aperture and energy transmission of the light beam. The geometric microstructure of its edge has a decisive influence on the system's point spread function (PSF) and the final imaging quality. Traditional variable apertures are usually composed of multiple overlapping blades, forming a complex polygonal aperture that is approximately circular. According to the theory of physical optics diffraction, light waves undergo strong diffraction when passing through the abrupt truncation point at the edge of the aperture. For polygonal apertures with straight edges, the diffraction energy is highly concentrated in the frequency domain, forming strong anisotropic energy divergence (i.e., higher-order diffraction sidelobes) around the light spot.

[0003] This phenomenon of stray diffraction light is particularly detrimental in dark-field detection, strong point light source observation, or high dynamic range (HDR) imaging scenarios. For example, in the detection of defects on semiconductor wafer surfaces, strong diffraction spikes can mask weak scattering signals, leading to false positives or false negatives; in high-power laser systems, concentrated diffraction hotspots can even damage subsequent optical components. Therefore, how to suppress diffraction noise in specific directions from a physical optics perspective is a key technical challenge for improving the signal-to-noise ratio and detection limit of high-end optical systems.

[0004] To suppress diffraction side lobes, existing technical solutions mainly include using apodization filters or pursuing a high-precision ideal circular aperture. Apodization technology softens edge diffraction by introducing a coating with gradually varying transmittance, but its precision coating process is extremely complex and costly, and it is prone to thermal damage under high-energy laser irradiation. While mechanical circular apertures offer relatively uniform diffraction distribution, their mechanical structure is complex and assembly errors are difficult to avoid in precision equipment requiring continuous aperture adjustment, making it difficult to meet micron-level adjustment requirements.

[0005] In recent years, modulating diffraction energy distribution by designing "irregular pupil filters" with specific microscopic undulations at their edges has gradually become a research hotspot in computational optics. However, existing design methods face significant technical bottlenecks when applied to engineering manufacturing: the edge curves generated by conventional algorithms are often extremely fragmented, containing numerous high-frequency burrs or submicron-level features. While these features perform well in mathematical simulations, they completely exceed the precision limits of current precision lithography, laser micro / nano processing, or electrical discharge machining, resulting in theoretical design drawings that cannot be converted into qualified high-precision optical components, and a severe disconnect between the actual performance and simulation results. Therefore, providing a new design method for pupil filters has become an urgent problem to be solved in the industry. Summary of the Invention

[0006] This invention provides an automated design method and apparatus for anti-diffraction pupil filters based on process constraints, which solves the defect in the conventional algorithm that generates edge curves containing a large number of high-frequency glitch or submicron-level features, causing a disconnect between the design and the actual product.

[0007] This invention provides an automated design method for anti-diffraction pupil filters based on process constraints, comprising: constructing a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components based on the microstructure of the pupil edge; constructing a differentiable inverse frequency domain transform operator to map the mixed spectral parameters into a spatial domain edge radial offset sequence; performing convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism to obtain a smooth edge contour of the pupil; constructing a closed vector graphic based on the smooth edge contour and establishing a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field into a two-dimensional pupil transmittance distribution function; constructing a full-link optical simulation model based on the two-dimensional pupil transmittance distribution function and Fourier scalar diffraction theory; constructing a composite loss function, calculating the gradient of the composite loss function with respect to the mixed spectral parameters based on an automatic differentiation framework, and iteratively updating the parameters using a numerical optimization algorithm until the overall system performance index converges.

[0008] According to the present invention, an automated design method for an anti-diffraction pupil filter based on process constraints is provided. The design method further includes: when constructing a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components, allowing the numerical optimization algorithm to perform controllable energy injection on specific spatial frequency bands.

[0009] According to the present invention, an automated design method for anti-diffraction pupil filters based on process constraints is provided. In the microstructure construction based on the pupil edge, a mixed spectral parameter space containing low-frequency base components and high-frequency detail components is constructed. The low-frequency base components are used to control the macroscopic geometric topology and fractal roughness of the pupil edge, and the high-frequency detail components are used to modulate the microscopic fluctuations at a specific spatial frequency.

[0010] According to the present invention, an automated design method for an anti-diffraction pupil filter based on process constraints is provided. The step of performing convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism includes: integrating a continuously differentiable smoothing filter regularization module into the generation link of the edge radial offset sequence; the smoothing filter regularization module is used to construct a corresponding smoothing convolution operator based on the minimum feature size and physical transfer function of the target processing technology; and performing convolution processing on the edge radial offset sequence using the smoothing convolution operator.

[0011] According to the present invention, an automated design method for an anti-diffraction pupil filter based on process constraints is provided. The step of constructing a closed vector graphic based on the smooth edge contour and establishing a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, and mapping the distance field to a two-dimensional pupil transmittance distribution function, further includes: nonlinearly mapping the distance field to the two-dimensional pupil transmittance distribution function with a value in the range of 0-1 based on a soft threshold activation function with a continuous gradient.

[0012] According to the present invention, an automated design method for anti-diffraction pupil filters based on process constraints is provided. The construction of the full-link optical simulation model includes: simulating the projection geometric distortion effect of the pupil under different off-axis field of view angles based on the affine transformation matrix during the simulation process, and calculating the point spread function of the central field of view and multiple typical edge fields of view in parallel.

[0013] According to the present invention, an automated design method for an anti-diffraction pupil filter based on process constraints is provided. The design method further includes: fusing evaluation indicators of different fields of view based on a weighted summation method.

[0014] This invention also provides an automated design device for an anti-diffraction pupil filter based on process constraints, comprising: a first construction module, used to construct a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components based on the microstructure of the pupil edge, construct a differentiable inverse frequency domain transform operator, and map the mixed spectral parameters to a spatial domain edge radial offset sequence; a processing module, used to perform convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism to obtain a smooth edge contour of the pupil; an establishment module, used to construct a closed vector graphic based on the smooth edge contour, and establish a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field to a two-dimensional pupil transmittance distribution function; a second construction module, used to construct a full-link optical simulation model based on the two-dimensional pupil transmittance distribution function and Fourier scalar diffraction theory; and a calculation module, used to construct a composite loss function, calculate the gradient of the composite loss function with respect to the mixed spectral parameters based on an automatic differentiation framework, and iteratively update the parameters using a numerical optimization algorithm until the overall system performance index converges.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the automated design method for anti-diffraction pupil filters based on process constraints as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the automated design method for anti-diffraction pupil filters based on process constraints as described above.

[0017] The present invention provides an automated design method for anti-diffraction pupil filters based on process constraints. By constructing a hybrid spectral parameter space containing low-frequency floor components and high-frequency detail components, and by performing convolution processing on the radial offset sequence of the edges based on a smoothing regularization mechanism, it successfully solves the starburst interference problem of traditional polygonal pupils and the common problem of the disconnect between theoretical models and physical fabrication in computational optics design. Without relying on prior human knowledge, it automatically designs high-performance anti-diffraction pupils with excellent optical imaging quality and reliable engineering feasibility. This provides a new and industrially feasible technical path for stray light suppression in high-end photographic lenses and precision optical instruments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the automated design method for anti-diffraction pupil filters based on process constraints provided by the present invention.

[0020] Figure 2 These are comparison images of simulated effects using a traditional flat rectangular aperture under multiple different fields of view.

[0021] Figure 3 The above is a comparison of the simulation effects of the automated design method for anti-diffraction pupil filters based on process constraints provided in the embodiments of the present invention under different fields of view after automated optimization.

[0022] Figure 4 It is a simulated image obtained using traditional algorithms.

[0023] Figure 5 The image is a simulated image obtained using the automated design method for anti-diffraction pupil filters based on process constraints provided in this embodiment of the invention.

[0024] Figure 6 This is a schematic diagram of the automated design device for anti-diffraction pupil filters based on process constraints provided by the present invention.

[0025] Figure 7This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined with Figures 1-7 This invention describes an automated design method, design apparatus, electronic device, and storage medium for anti-diffraction pupil filters based on process constraints.

[0028] like Figure 1 As shown, in an embodiment of the present invention, the automated design method for anti-diffraction pupil filters based on process constraints specifically includes the following steps: Step 101: Based on the microstructure of the pupil edge, construct a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components, construct a differentiable inverse frequency domain transformation operator, and map the mixed spectral parameters into a spatial domain edge radial offset sequence.

[0029] Specifically, in this embodiment, a frequency domain analysis method is used to perform high-dimensional characterization of the radial vector at the edge of the pupil filter, constructing a hybrid spectral parameter space containing low-frequency basis components and high-frequency detail components. The low-frequency basis components are used to control the macroscopic geometric topology and fractal roughness of the pupil edge, while the high-frequency detail components are used to precisely modulate the microscopic fluctuations at specific spatial frequencies. By constructing a differentiable inverse frequency domain transform operator, the hybrid spectral parameters are mapped to a spatial domain edge radial offset sequence, thereby achieving decoupled control of the pupil microstructure at different scales.

[0030] Step 102: Based on the smoothing regularization mechanism, the edge radial offset sequence is convolved to obtain the smooth edge contour of the pupil.

[0031] To address the problem of excessively fragmented pupil edge structures, even containing deep submicron burrs, making them unmanufacturable in traditional computational optics design, this embodiment integrates a continuously differentiable smoothing filter regularization module into the generation chain of the edge radial offset sequence. During optimization, this smoothing filter regularization module performs real-time convolution processing on the pupil edge radial offset sequence, forcibly filtering out high-frequency abrupt noise exceeding the limits of physical processing precision, ensuring that the output geometric contour possesses smoothness and continuity that meets industrial-grade manufacturing standards.

[0032] Step 103: Construct a closed vector graphic based on the smooth edge contour, and establish a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field to a two-dimensional pupil transmittance distribution function.

[0033] Specifically, in this embodiment, a closed vector graphic is constructed based on a smooth edge contour, and a distance field mapping relationship is established from the image plane sampling grid to the boundary of the closed vector graphic. By introducing a soft-threshold activation function with continuous gradients, the distance field is nonlinearly mapped to a two-dimensional pupil transmittance distribution function (i.e., a soft pupil mask) with values ​​ranging from 0 to 1. This step overcomes the mathematical non-differentiability barrier of traditional binary masks, opens up the gradient backpropagation channel from the underlying geometric parameters to the top-level optical response function, and lays the mathematical foundation for gradient-based automated optimization.

[0034] Step 104: Based on the two-dimensional pupil transmittance distribution function and Fourier optical scalar diffraction theory, construct a full-link optical simulation model.

[0035] Specifically, the generated two-dimensional pupil transmittance distribution function is used as the complex amplitude entrance pupil function of the optical system. Combined with Fourier scalar diffraction theory, a full-link optical simulation model incorporating wavefront propagation, phase modulation, and intensity detection is constructed. During the simulation, an affine transformation matrix is ​​introduced to accurately simulate the projection geometric distortion effects of the pupil at different off-axis field-of-view angles (including aperture scaling, translation, and shearing deformation), and the point spread function (PSF) of the central field of view and several typical edge fields is calculated in parallel. Furthermore, the nonlinear photoelectric response characteristics and saturation cutoff effect of high-performance image sensors (such as CCD / CMOS) are simulated to generate a simulated imaging image reflecting the actual detection effect.

[0036] Step 105: Construct a composite loss function, calculate the gradient of the composite loss function with respect to the mixed spectral parameters based on the automatic differentiation framework, and iteratively update the parameters using a numerical optimization algorithm until the overall system performance index converges.

[0037] Specifically, a composite loss function is constructed, comprising spatial domain morphological constraints, frequency domain energy distribution constraints, and image gradient consistency constraints. The spatial domain morphological constraints aim to optimize the radial symmetry and energy concentration of the point spread function; the frequency domain energy distribution constraints aim to suppress the energy of higher-order diffraction sidelobes in specific directions and eliminate diffraction artifacts. The gradient of the composite loss function with respect to the mixed spectral parameters is calculated using an automatic differentiation framework, and the parameters are iteratively updated using a numerical optimization algorithm until the overall system performance converges. The final output is a pupil design scheme that combines excellent diffraction suppression performance with good manufacturability. Optionally, in embodiments of this invention, the numerical optimization algorithm can be an adaptive moment estimation algorithm or an adaptive gradient algorithm.

[0038] The automated design method for anti-diffraction pupil filters based on process constraints provided in this invention constructs a hybrid spectral parameter space containing low-frequency floor components and high-frequency detail components. By performing convolution processing on the radial offset sequence of the edges based on a smoothing regularization mechanism, it successfully solves the starburst interference problem of traditional polygonal pupils and the common problem of the disconnect between theoretical models and physical fabrication in computational optics design. Without relying on prior human knowledge, it automatically designs high-performance anti-diffraction pupils with excellent optical imaging quality and reliable engineering feasibility. This provides a new and industrially feasible technical path for stray light suppression in high-end photographic lenses and precision optical instruments.

[0039] Furthermore, in embodiments of the present invention, the automated design method for anti-diffraction pupil filters based on process constraints further includes: when constructing a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components, allowing numerical optimization algorithms to controllably inject energy into specific spatial frequency bands to generate edge microstructures capable of destructive interference, thereby canceling out specific order diffraction bright spots caused by pupil aperture truncation from the essence of physical optics.

[0040] Specifically, traditional optimization algorithms (such as standard gradient descent) aim to uniformly decay energy (i.e., reduce loss) across all frequency bands, pursuing a monotonically decreasing overall loss function. In this embodiment, allowing the numerical optimization algorithm to inject energy into specific spatial frequency bands is a non-monotonic or multi-scale strategy. To achieve more efficient global convergence, it strategically and temporarily increases updates at a certain scale to accelerate the convergence of important features.

[0041] Furthermore, in an embodiment of the present invention, the step of performing convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism includes: integrating a continuously differentiable smoothing filter regularization module into the generation link of the edge radial offset sequence. This smoothing filter regularization module is used to construct a corresponding smoothing convolution operator based on the minimum feature size and physical transfer function of the target processing technology, and then using the smoothing convolution operator to perform convolution processing on the edge radial offset sequence. In this embodiment, the target processing technology includes: laser micro / nano processing, electron beam lithography, or precision etching, etc.

[0042] In an embodiment of the present invention, when constructing a full-link optical simulation model, the end-to-end physical optics diffraction simulation link with multiple parallel fields of view integrates the evaluation indicators of different fields of view through weighted summation, driving the numerical optimization algorithm to find the globally optimal pupil shape that can balance the imaging quality of the central field of view and the edge field of view, thus avoiding the occurrence of a local optimal solution where the central field of view is improved while the edge field of view is deteriorated.

[0043] To demonstrate the effectiveness of the process-constrained automated design method for anti-diffraction pupil filters provided in this invention in suppressing diffraction, and its engineering application potential in the manufacturing of high-end precision optical components, the following section uses this method to optimize the design of a rectangular aperture optical system. Because rectangular apertures have mutually perpendicular straight edges, they correspond to extremely strong orthogonal diffraction components in the Fourier frequency domain, typically producing the most severe cross-shaped high-order diffraction side lobes (i.e., diffraction artifacts). This experiment aims to eliminate this persistent diffraction structure through an automated algorithm, while ensuring that the final pupil profile strictly meets the process standards for precision photolithography mask fabrication or laser micro / nano processing.

[0044] The experiment was run on a workstation equipped with high-performance GPU computing nodes, and a fully differentiable optical forward propagation model was built based on a deep learning framework (such as PyTorch).

[0045] Parameter settings: The simulated wavelength was set to 555nm (typical precision detection band) as the benchmark evaluation wavelength. To simulate the imaging differences of a large numerical aperture precision optical system at different field of view angles, the system was configured with a multi-field parallel sampling mode, including: (a) central field of view (0.0 field of view); (b) horizontal 0.7 field of view; (c) vertical 0.7 field of view; and (d) diagonal full field of view (1.0 field of view). This configuration covers the complete effective light-transmitting area from the optical axis center to the edge of the image field. The sampling grids of both the pupil plane and the image plane were set to high-resolution matrices to ensure resolution of sub-micron diffraction fringe details.

[0046] Initial state: The initial pupil is set to a standard rectangular aperture. The edge of this aperture consists of four perfectly straight geometric lines, without any form of frequency domain modulation or edge modification.

[0047] Optimization objective definition: Set a multidimensional composite loss function to minimize the spatial energy distribution of the point spread function in the non-circular region, especially to suppress higher-order diffraction maxima extending along the horizontal and vertical axes, and to forcibly constrain the rate of curvature change of the pupil edge to ensure physical manufacturability.

[0048] The experiment initiated an automated iterative optimization process based on gradient descent, executing a total of 1500 iterations. The optimization process went through the following three evolutionary stages: Phase 1: Rapid Response and Reconstruction of Spectral Parameters (Step 1-500) In the initial optimization phase, the numerical optimization algorithm quickly identified the main frequency domain components causing the crosshair effect. The system maps the geometric features of the pupil edge to a high-dimensional spectral parameter space, and breaks the periodicity of the original rectangular edge by significantly adjusting the phase and amplitude parameters of the "foundation spectrum" and "detail spectrum". At this point, the originally straight edge begins to show slight undulations, and the diffraction energy begins to diffuse from the concentrated bright line to the surrounding space.

[0049] Phase Two: Continuous Constraints on Embedded Process Filtering (Step 500-1000) As optimization progresses, the pursuit of ultimate optical performance often leads to the generation of extremely complex edge textures through purely mathematical optimization. This is where the smoothing filter regularization module plays a crucial role. This module simulates the physical integral effect of machining tools, performing real-time smoothing on the edge contours generated in each iteration. This mechanism forcibly filters out any tiny burrs and sub-pixel-level abrupt changes in the edge radial offset sequence that are smaller than the machining precision, forcing the optimizer to find the optically optimal solution while maintaining edge smoothness. This ensures that the geometric evolution of the edges is always confined within the physically realizable manifold space.

[0050] Phase 3: Fine-tuning of multi-field consistency (Step 1000-1500) In the final stage of optimization, the numerical optimization algorithm focuses on balancing the performance differences between different fields of view. By taking into account the pupil projection distortion in off-axis fields of view, the numerical optimization algorithm fine-tunes the edge fluctuation frequency, so that the pupil can still maintain good diffraction suppression capability even in edge fields of view where perspective distortion occurs.

[0051] When the aforementioned automated iterative optimization process based on gradient descent is applied to the automated design method of anti-diffraction pupil filters based on process constraints, in step 101, a mixed spectrum is generated using a power-law basis and a Gaussian mixture model, and a texture is generated through inverse Fourier transform. In step 102, a differentiable Gaussian convolution is applied to the generated edges to forcibly filter out unmanufacturable high-frequency glitches. In step 103, when constructing closed vector graphics based on smooth edge contours, the offsets of the four sides are combined into closed geometric polygon vertices. In step 104, parallel processing modules for different fields of view perform wavefront propagation calculations and apply a sensor model to generate a simulated imaging map reflecting the actual detection effect. In step 105, the symmetry of the light spot is optimized, the diffraction sidelobe energy is suppressed, and the image gradient consistency is maintained. The gradient is calculated, gradient clipping is applied, all learnable parameters are updated using an optimizer, and the final design scheme is output when the loss function converges.

[0052] Figure 2 and Figure 3 The comparison of key data at the beginning and end of the optimization process clearly reflects the qualitative change in system performance. Figure 2 (a1), (a2), (a3) ​​and (a4) represent the pupil projection mask diagrams under the central field of view and three different edge fields of view, respectively; Figure 2 (b1), (b2), (b3) and (b4) represent the logarithmic heatmaps of the physical point spread function (PSF) under the corresponding field of view; Figure 2 (c1), (c2), (c3), and (c4) represent the final simulated images under the corresponding fields of view. Figure 2 As shown in (a1), (a2), (a3), and (a4), the initial aperture exhibits a standard square structure with straight edges and no modulation features. In the PSF heatmap on logarithmic coordinates, as... Figure 2 As shown in (b1), (b2), (b3), and (b4), the energy distribution exhibits typical Fraunhofer diffraction characteristics. The energy is highly concentrated along two orthogonal axes, and its decay is extremely slow with increasing distance from the center. This "long-tailed" energy distribution is the main physical reason for the dazzling starburst effect. In the final simulated image, as shown... Figure 2 As shown in (c1), (c2), (c3) and (c4), a bright cross-shaped starburst runs through the entire 256x256 pixel viewport, completely obscuring the dark details of the background.

[0053] Figure 3 (a5), (a6), (a7) and (a8) represent the optimized pupil projection mask diagrams under the central field of view and three different edge fields of view, respectively; Figure 3 (b5), (b6), (b7) and (b8) represent the logarithmic heatmaps of the physical point spread function (PSF) under the corresponding optimized pupil, respectively; Figure 3 (c5), (c6), (c7) and (c8) represent the final simulated images under the corresponding fields of view.

[0054] like Figure 3 As shown in (a5), (a6), (a7), and (a8), after 1500 iterations, the pupil edge evolves into a "modulated wave structure" with specific quasi-periodic characteristics. Careful observation of the edge curve reveals that although the overall shape has changed significantly, its local curvature changes are very continuous and gentle. There are no sharp corners, breaks, or high-frequency oscillations, fully conforming to existing CNC precision machining or photolithography mask fabrication process specifications, allowing for direct production without subsequent manual adjustments. Figure 3 As shown in (b5), (b6), (b7), and (b8), the physical point diffusion function thermograms at this point show that the previously concentrated high-intensity diffraction bright lines have been completely dispersed. Figure 3As shown in (c5), (c6), (c7) and (c8), the energy is redistributed to the diffuse circle region around the light spot, exhibiting an isotropic fogging distribution characteristic, which means that the interference enhancement effect in a specific direction is successfully destroyed.

[0055] like Figure 4 and Figure 5 As shown in the simulated image of Step 1500, the previously glaring crosshair has completely disappeared, replaced by a soft, natural glow around the light source. The signal-to-noise ratio of the image is significantly improved, and the background blackness is purer.

[0056] contrast Figure 3 As can be seen from the four sub-figures (a5), (a6), (a7), and (a8), this excellent anti-diffraction performance maintains a high degree of consistency throughout the entire image field. Even in the extreme corner field of view where the pupil experiences projection compression, the edge modulation structure still functions effectively, without any obvious starburst recurrence. The total loss of the system decreased significantly from the initial 43684.86 to 14269.65, a reduction of over 67%. This significant numerical decrease objectively confirms that the automated design method for anti-diffraction pupil filters based on process constraints provided in this embodiment of the invention has extremely high efficiency in diffraction energy control.

[0057] The following describes the automated design device for anti-diffraction pupil filters based on process constraints provided in the embodiments of the present invention. The design device described below can be referred to in correspondence with the design method described above.

[0058] like Figure 6As shown in the embodiments of the present invention, the automated design device for anti-diffraction pupil filters based on process constraints includes: a first construction module 201, a processing module 202, an establishment module 203, a second construction module 204, and a calculation module 205. The first construction module 201 is used to construct a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components based on the microstructure of the pupil edge, and to construct a differentiable inverse frequency domain transform operator to map the mixed spectral parameters into a spatial domain edge radial offset sequence. The processing module 202 is used to perform convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism to obtain a smooth edge contour of the pupil. The establishment module 203 is used to construct a closed vector graphic based on the smooth edge contour and to establish a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field into a two-dimensional pupil transmittance distribution function. The second construction module 204 is used to construct a full-link optical simulation model based on the two-dimensional pupil transmittance distribution function and Fourier scalar diffraction theory. The computation module 205 is used to construct the composite loss function, calculate the gradient of the composite loss function with respect to the mixed spectral parameters based on the automatic differentiation framework, and iteratively update the parameters using a numerical optimization algorithm until the overall system performance index converges.

[0059] like Figure 7 As shown in the figure, this embodiment of the invention also provides a schematic diagram of the physical structure of an electronic device, which may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. Processor 810 can call logic instructions in memory 830 to execute an automated design method for anti-diffraction pupil filters based on process constraints. This design method includes: constructing a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components based on the microstructure of the pupil edge; constructing a differentiable inverse frequency domain transform operator to map the mixed spectral parameters to a spatial domain edge radial offset sequence; performing convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism to obtain a smooth edge contour of the pupil; constructing a closed vector graphic based on the smooth edge contour and establishing a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field to a two-dimensional pupil transmittance distribution function; constructing a full-link optical simulation model based on the two-dimensional pupil transmittance distribution function and Fourier scalar diffraction theory; constructing a composite loss function, calculating the gradient of the composite loss function with respect to the mixed spectral parameters based on an automatic differentiation framework, and iteratively updating the parameters using a numerical optimization algorithm until the overall system performance converges.

[0060] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0061] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the automated design method for anti-diffraction pupil filters based on process constraints provided by the above methods. The method includes: constructing a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components based on the microstructure of the pupil edge; constructing a differentiable inverse frequency domain transform operator to map the mixed spectral parameters to a spatial domain edge radial offset sequence; performing convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism to obtain a smooth edge contour of the pupil; constructing a closed vector graphic based on the smooth edge contour and establishing a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field to a two-dimensional pupil transmittance distribution function; constructing a full-link optical simulation model based on the two-dimensional pupil transmittance distribution function and Fourier optical scalar diffraction theory; constructing a composite loss function, calculating the gradient of the composite loss function with respect to the mixed spectral parameters based on an automatic differentiation framework, and iteratively updating the parameters using a numerical optimization algorithm until the overall system performance index converges.

[0062] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements an automated design method for anti-diffraction pupil filters based on process constraints, as provided by the methods described above. This method includes: constructing a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components based on the microstructure of the pupil edge; constructing a differentiable inverse frequency domain transform operator to map the mixed spectral parameters to a spatial domain edge radial offset sequence; performing convolution processing on the edge radial offset sequence based on a smoothing regularization mechanism to obtain a smooth edge contour of the pupil; constructing a closed vector graphic based on the smooth edge contour and establishing a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field to a two-dimensional pupil transmittance distribution function; constructing a full-link optical simulation model based on the two-dimensional pupil transmittance distribution function and Fourier scalar diffraction theory; constructing a composite loss function, calculating the gradient of the composite loss function with respect to the mixed spectral parameters based on an automatic differentiation framework, and iteratively updating the parameters using a numerical optimization algorithm until the overall system performance index converges.

[0063] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0064] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automated design method for anti-diffraction pupil filters based on process constraints, characterized in that, include: Based on the microstructure of the pupil edge, a hybrid spectral parameter space containing low-frequency basis components and high-frequency detail components is constructed. A differentiable inverse frequency domain transformation operator is constructed to map the hybrid spectral parameters into a spatial domain edge radial offset sequence. A continuously differentiable smoothing filter regularization module is integrated into the generation link of the edge radial offset sequence. The smoothing filter regularization module is used to construct a corresponding smoothing convolution operator based on the minimum feature size and physical transfer function of the target processing technology. The smoothing convolution operator is used to perform convolution processing on the edge radial offset sequence to obtain the smooth edge contour of the pupil. Based on the smooth edge contour, a closed vector graphic is constructed, and a distance field mapping relationship is established from the image plane sampling grid to the boundary of the closed vector graphic. The distance field is then mapped to a two-dimensional pupil transmittance distribution function. Based on the two-dimensional pupil transmittance distribution function and Fourier optical scalar diffraction theory, a full-link optical simulation model is constructed. A composite loss function is constructed, and the gradient of the composite loss function with respect to the hybrid spectral parameters is calculated based on the automatic differentiation framework. The parameters are then iteratively updated using a numerical optimization algorithm until the overall system performance index converges.

2. The automated design method for anti-diffraction pupil filters based on process constraints according to claim 1, characterized in that, The design method further includes: When constructing the hybrid spectral parameter space containing low-frequency basis components and high-frequency detail components, the numerical optimization algorithm is allowed to perform controllable energy injection for specific spatial frequency bands.

3. The automated design method for anti-diffraction pupil filters based on process constraints according to claim 1, characterized in that, In the microstructure constructed based on the pupil edge, a mixed spectral parameter space containing low-frequency floor components and high-frequency detail components is constructed. The low-frequency base component is used to control the macroscopic geometric topology and fractal roughness of the pupil edge, while the high-frequency detail component is used to modulate the microscopic undulations at a specific spatial frequency.

4. The automated design method for anti-diffraction pupil filters based on process constraints according to claim 1, characterized in that, The step of constructing a closed vector graphic based on the smooth edge contour, establishing a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, and mapping the distance field to a two-dimensional pupil transmittance distribution function further includes: Based on a soft threshold activation function with a continuous gradient, the distance field is nonlinearly mapped to the two-dimensional pupil transmittance distribution function with values ​​in the range of 0-1.

5. The automated design method for anti-diffraction pupil filters based on process constraints according to claim 1, characterized in that, The construction of the full-link optical simulation model includes: During the simulation, the projection geometric distortion effect of the pupil under different off-axis field of view angles is simulated based on the affine transformation matrix, and the point spread function of the central field of view and multiple typical edge fields of view is calculated in parallel.

6. The automated design method for anti-diffraction pupil filters based on process constraints according to claim 5, characterized in that, The design method further includes: This method integrates evaluation metrics from different fields of view using a weighted summation approach.

7. An automated design device for anti-diffraction pupil filters based on process constraints, characterized in that, include: The first construction module is used to construct a mixed spectral parameter space containing low-frequency basis components and high-frequency detail components based on the microstructure of the pupil edge, and to construct a differentiable inverse frequency domain transformation operator to map the mixed spectral parameters into a spatial domain edge radial offset sequence. The processing module is used to integrate a continuously differentiable smoothing filter regularization module in the generation link of the edge radial offset sequence. The smoothing filter regularization module is used to construct a corresponding smoothing convolution operator based on the minimum feature size and physical transfer function of the target processing technology, and to use the smoothing convolution operator to perform convolution processing on the edge radial offset sequence to obtain the smooth edge contour of the pupil. A module is established to construct a closed vector graphic based on the smooth edge contour, and to establish a distance field mapping relationship from the image plane sampling grid to the boundary of the closed vector graphic, mapping the distance field to a two-dimensional pupil transmittance distribution function; The second construction module is used to construct a full-link optical simulation model based on the two-dimensional pupil transmittance distribution function and Fourier optical scalar diffraction theory. The computation module is used to construct a composite loss function, calculate the gradient of the composite loss function with respect to the hybrid spectral parameters based on the automatic differentiation framework, and iteratively update the parameters using a numerical optimization algorithm until the overall system performance index converges.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the automated design method for anti-diffraction pupil filters based on process constraints as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the automated design method for anti-diffraction pupil filters based on process constraints as described in any one of claims 1 to 6.

Images