Medium-wave infrared focal plane optical filter realized by metasurface and reverse design method of medium-wave infrared focal plane optical filter

Abstract

The invention discloses an optical filter for realizing a medium-wave infrared focal splitting plane on a metasurface and a reverse design method. According to the light field regulation and control characteristics of the metasurface, spectrum routing and focusing functions of a specified medium-wave infrared wave band can be completed only by using the single-layer all-silicon medium metasurface, so that the filtering effect similar to that of a Bayer optical filter is achieved, the spectrum energy utilization rate is remarkably improved compared with that of the Bayer optical filter, an end-to-end reverse method is used in the design thought, and the design cost is reduced. Structural parameters are optimized in a staged mode through a genetic algorithm and a binary search algorithm (DBS), the advantages of all the algorithms are brought into full play, the design threshold is lowered, the optimization freedom degree is improved, and therefore the optimization performance can be improved. Therefore, key device support can be provided for low-illumination mid-infrared night vision color imaging application, and a new design thought is expanded for the technical field of integration and miniaturization.

Description

Technical Field

[0001] The present invention belongs to the field of optical imaging and spectral signal processing, and in particular relates to a mid-infrared spectral router similar to a Bayer filter in an all-silicon dielectric metasurface and its structural inverse design using a multi-objective genetic algorithm. Background Art

[0002] Medium-wave infrared band imaging is widely used in military reconnaissance, security surveillance, unmanned driving, and aerospace remote sensing because of its ability to work day and night and passive imaging. The medium-wave infrared filter is one of the key components on the detector in this band. Especially in low-light night vision environments, the filter design directly affects the spectrum and color imaging quality of the detector. Common design methods include pre-filter wheel devices or back-end image enhancement and pseudo-color algorithms. However, traditional hardware designs such as filter wheels are limited by low energy utilization and difficult mechanical system design, and the back-end algorithm also has the disadvantages of long training time and overfitting. The above methods are difficult to meet the application requirements with the increasing requirements of integrated miniaturization and low latency, which seriously limits the development of the medium-wave night vision imaging field.

[0003] Metasurface is a new type of two-dimensional planar metamaterial composed of a single-layer sub-wavelength metal or dielectric nanoantenna array. Metasurfaces with wavelength-level thickness can arbitrarily control the phase, polarization, intensity and other parameters of light. Nowadays, the ultra-high resolution, multi-channel and multi-dimensional characteristics of metasurface light field control have realized a variety of functions, including aberration-free microlenses, high-quality holographic technology, and beam deflection and generation. Metasurface filters are favorable candidates for improving spectral energy utilization under low illumination and realizing spectral routing.

[0004] Traditional metasurface design is achieved by determining the structural parameters and arrangement distribution of nanoantennas. However, such forward design requires a lot of electromagnetic theory and engineering experience, and the design requires step-by-step scanning of parameters to determine the optimal structure. This greatly increases the design threshold and cost, and it is also impossible to achieve more complex functions.

[0005] In summary, it is necessary to propose a metasurface filter that realizes multi-channels in the mid-infrared band and a targeted reverse design method of the metasurface to provide key device support for spectral routing and low-illumination night vision color imaging applications. Summary of the invention

[0006] To solve the above problems, the present invention proposes for the first time a mid-wave infrared (3μm~5μm) metasurface filter, and provides a targeted inverse design algorithm, which utilizes the light field control characteristics of the metasurface to realize the routing of four different bands of spectra to the pointed focal plane. Unlike the unidirectional spectral scattering mechanism, this metasurface achieves significant improvements in energy utilization and segmented spectrum collection efficiency, thereby showing higher imaging intensity than common imaging sensors in the fields of low-light night vision imaging and mid-wave infrared color imaging.

[0007] The design method of the medium-wave infrared metasurface filter based on inverse design includes the following steps:

[0008] Step 1: Under the plane incident light in the mid-wave infrared band and the same polarization condition, the geometric parameters of the elliptical superatom used to constitute the metasurface are scanned and calculated using the time-domain finite difference method simulation, and the optical response under the corresponding dimension is obtained, so as to obtain the relationship between the superatom under different geometric parameters and the amplitude and phase regulation of the light field. The geometric parameters of the superatom include the long axis length a, the short axis length b, the columnar height h, and the distribution period p of the superatom. The ellipse does not rotate, that is, it is a standard ellipse.

[0009] Step 2: In the design stage, the metasurface is divided into four regions of equal area, which are mapped to the target distribution areas of the four bands on the target plane. Then the entire metasurface is "pixelated", and the size of each pixel unit is p. Each pixel unit (square micro-nano column) is assigned a super atom or air of the size selected in step 1, represented by "1" or "0" respectively. In this way, the metasurface plane is mapped to a square matrix containing only "0" and "1".

[0010] Step 3: Genetic algorithm is used to randomly generate k initial populations containing binary coding patterns of the metasurface according to the row and column sizes of the matrix selected by step 2. These coding numbers are defined as material or free space, and the fitness function of the routing properties of the metasurface of each binary coding pattern is defined, that is, the fitness F.

[0011]

[0012] Among them, λ m1 and λ m2 Used to indicate the minimum and maximum wavelengths, the subscript m represents the corresponding four bands, namely, the minimum and maximum wavelengths of a are 3μm and 3.5μm, b are 3.5μm and 4μm, c are 4μm and 4.5μm, and d are 4.5μm and 5μm. a , T b , T c , T dare the ratios of the light intensity of each band area on the imaging plane to the light intensity incident on the metasurface area, and A, B, C, and D are the adaptive weights of each band integral. The purpose of this function is to maximize the light energy of the characteristic bands in different band areas, so as to achieve the purpose of transmitting the light of different bands to the designated routes and improve the overall utilization of the spectrum.

[0013] Step 4: Use Lumerical FDTD solutions and Python co-simulation to set the initial weight factor and calculate the fitness of k hypersurface individuals in the initial population binary coding mode.

[0014] Step 5: Set the number of iterations to g and the crossover probability to p c , mutation probability p a According to the parent generation obtained in step 4, k probability values ​​of being selected are generated. According to the principle of roulette, the higher the fitness function value, the greater the probability of being selected.

[0015] Step 6: Select two parents in a cycle and perform crossover mutation according to a certain probability to obtain a child, and use new individuals to replace a certain number of individuals with poor fitness in the parent generation to finally obtain a new generation. Simulate and calculate the fitness.

[0016] Step 7: Repeat step 6 until the iteration ends or the fitness function converges. The values ​​of the four weight factors are adaptively changed according to certain rules to ensure that the objective function is continuously optimized during the iteration process. Output the hypersurface binary coding matrix with the maximum fitness after optimization.

[0017] Step 8: In order to further optimize the fitness function to improve the degree of freedom, the improved binary search algorithm DBS is used for further iterative optimization. The iteration number g1 is set, an initial cell is selected as the optimization starting point, and the state thereof is changed, including the flipping of "0" and "1", and the three rotation angles of the elliptical superatom around the center of the ellipse (45°, 90°, 135°). The fitness function in each state is calculated by joint simulation, and the state with the maximum fitness is retained. Each cell is scanned step by step until the fitness function converges, and the weight factor remains unchanged. Finally, the hypersurface coding pattern with the maximum fitness is output.

[0018] The method also includes step 9: according to the optimal geometric structure parameters obtained in step 8, a metasurface for a medium-wave infrared filter is manufactured, and the metasurface is applied in the field of medium-wave infrared detection to improve low-light imaging performance.

[0019] The beneficial effects of the present invention are:

[0020] (1) The present invention proposes a new metasurface filter routing design, which can achieve spectral routing and filtering effect only through a single-layer metasurface structure, instead of losing too much spectral energy information like common filters; it can be seen that the present invention greatly improves the spectral energy utilization rate of mid-infrared imaging in terms of hardware, and greatly reduces the bloat of the filtering system. It also provides a new design idea for the application of medium-wave night vision color imaging.

[0021] (2) The present invention provides a phased hybrid reverse design algorithm, which optimizes the parameters of the metasurface structure in phases by combining the genetic algorithm and the binary brute force search algorithm DBS. Combining the characteristics of the two greatly improves the degree of freedom of reverse design. Compared with the traditional forward design that requires complex algorithms and steps, this design method is simpler and has higher design accuracy.

[0022] (3) Compared with common metasurface design methods, this method realizes end-to-end design, which only requires determining the target requirements, without the user having to spend time processing and analyzing during the design process. All processes come from FDTD software and automatic optimization in python, which greatly reduces the cost of manual design and improves ease of use. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings, in which:

[0024] Figure 1 A schematic diagram of the routing of four bands of medium-wave infrared provided by the present invention;

[0025] Figure 2 A schematic diagram of a super-atom simulation model constructed by the electromagnetic simulation method of the present invention;

[0026] Figure 3 Designing an initial structure for the metasurface of the binary coding mode provided by the present invention;

[0027] Figure 4 Schematic diagram of the design process in Example 1 of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0029] This embodiment uses a hybrid reverse design based on genetic algorithm and DBS algorithm to route the optical filter using all-silicon super-surface. The all-silicon substrate material and the super-atom material are both silicon.

[0030] (1) In this embodiment, the incident light source is set to the 3μm-5μm band, the phase is 0, the normalized amplitude is 1, and the polarization state is x-polarized plane light. Lumerical FDTD solutions is used to establish an elliptical silicon superatom model, and the boundary condition is set to periodic PML. Since the size of a single pixel of the mid-infrared detector is 15×15μm 2 , the period of the superatom is set to 1.5μm. The simulation scans the amplitude and phase of the transmission scattering field corresponding to different sizes of superatoms (major axis a, minor axis b, column height h, period p). The superatoms with full-band coverage of 2π phase and high transmittance are selected.

[0031] (2) Divide the metasurface into 20×20 cells, each of which has two states, “0” and “1”, representing the remaining air after etching and the metaatom selected in step (1), respectively. Use a genetic algorithm, set the maximum number of iterations, and randomly generate k initial populations containing binary coding patterns of the metasurface. In the simulation software, each random structure is uniformly set to have a silicon substrate; four monitors are placed to receive transmission field data, corresponding to the four pixel surfaces of the detector; the obtained simulation data are defined as T a 、T b 、T c 、T d Four adaptive weight factors are set, which are functions of the four transmittances of the iteration number i and the previous generation i-1 of the current iteration number, namely:

[0032]

[0033] Furthermore, the fitness function F is defined as:

[0034]

[0035] Thus, the corresponding fitness value is obtained according to the simulation calculation of random k initial structures.

[0036] (3) Set the crossover probability p c , mutation probability p a According to the parent generation obtained in step (2), k probability values ​​of being selected are generated. According to the principle of roulette, the higher the fitness function value, the greater the probability of being selected, so the probability of the th parent generation being selected is p i , is defined as:

[0037]

[0038] (4) According to the probability value of step (3), two parents are selected cyclically, and a child is obtained by crossover and mutation according to the crossover and mutation probability, and a certain number of individuals with poor fitness in the parent generation are replaced with new individuals. Finally, a new first generation is obtained.

[0039] (5) Set the maximum iteration number g, repeat step (3), sort the fitness of each generation of population during the process, and make an iterative convergence curve according to the maximum fitness of each generation to monitor the optimization trend in real time. Stop the repetition when the curve converges or reaches the maximum iteration number g. Obtain the hypersurface structure parameters with the maximum fitness at this time.

[0040] (6) Perform improved DBS processing based on the structural parameters obtained in step (5), set the iteration number g1, select the first initial cell in the upper left corner as the optimization starting point, change its state, including the flipping of "0" and "1", and the three rotation angles of the elliptical superatom around the center of the ellipse (45°, 90°, 135°), and record the states corresponding to these angles as "2", "3", and "4". Use Lumerical FDTD solutions and python co-simulation to calculate the fitness function in each state and retain the state with the maximum fitness. The weight factors are all set to 1. Then scan and calculate each cell point by point from left to right and from top to bottom, and finally generate a new generation.

[0041] (7) Repeat step (6), and make an iterative convergence graph based on the number of iterations and the maximum fitness of each generation. Repeat step (6) until the curve converges.

[0042] The final output is the hypersurface coding mode with the highest fitness, where 0 to 4 all have corresponding hyperatom states.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A design method for a medium-wave infrared metasurface filter based on inverse design, characterized in that: The following steps are involved: Step 1: Under the same polarization condition of plane incident light, use the finite difference time domain method to scan and calculate the geometric parameters of the elliptical super-atom used to constitute the metasurface, and find the super-atom A (a, b, h, p) that can achieve the best light field transmittance and cover the 2π phase. The geometric parameters of the super-atom parameters include the ellipse major axis a, the ellipse minor axis b, the height h, and the distribution period p. Step 2: "Pixelate" the entire metasurface. Each pixel unit has a size of p. Each pixel unit (square micro-nano column) is assigned a metaatom or air (the metaatom is etched), represented by "1" or "0" respectively. Map the metasurface plane into a square matrix containing only "0" and "1". Step 3: Based on the genetic algorithm, define the routing attribute fitness of each hypersurface with binary coding pattern, and randomly generate k initial populations containing hypersurface binary coding patterns according to the row and column sizes in step 2. Step 4: Set the initial weight factor and calculate the fitness of k hypersurface individuals in the initial population binary coding mode. Step 5: Based on the parent generation obtained in step 4, use the roulette principle to generate k probability values ​​of being selected. Step 6: Select two parents in a cycle and perform crossover mutation according to a certain probability to obtain a child generation, and use new individuals to replace a certain number of individuals with poor fitness in the parent generation to finally obtain a new generation. Simulate and calculate the fitness. Step 7: Repeat step 6 until the iteration ends or the fitness function converges. The values ​​of the four weight factors are adaptively changed according to certain rules to ensure that the objective function is continuously optimized during the iteration process. Output the hypersurface binary coding matrix with the maximum fitness after optimization. Step 8: Use the improved binary search algorithm DBS for further iterative optimization. Select an initial cell as the optimization starting point and change the state in it, including the flipping of "0" and "1" and the three rotation angles of the elliptical superatom around the center of the ellipse (45°, 90°, 135°). Simulate and calculate the fitness function in each state, where the weight factor remains unchanged. Keep the state with the maximum fitness, scan each cell step by step, and generate a new generation. Until the fitness function converges, the hypersurface coding pattern with the maximum fitness is finally output.

2. The method for designing a medium-wave infrared metasurface filter based on inverse design as claimed in claim 1, characterized in that: The method further includes step 9, wherein a metasurface for a medium-wave infrared filter is manufactured according to the optimal geometric structure parameters obtained in step 8, and the metasurface is applied in the field of medium-wave infrared detection to improve low-light imaging performance.

3. The method for designing a medium-wave infrared metasurface filter based on inverse design as claimed in claim 1, characterized in that: In step 3, the fitness function is as follows: Among them, λ m1 and λ m2 Used to indicate the minimum and maximum wavelengths, the subscript m represents the corresponding four bands, namely, the minimum and maximum wavelengths of a are 3μm and 3.5μm, b are 3.5μm and 4μm, c are 4μm and 4.5μm, and d are 4.5μm and 5μm. a , T b , T c , T d are the ratios of the light intensity of each band area on the imaging plane to the light intensity incident on the metasurface area, and A, B, C, and D are the adaptive weights of the integral of each band.

4. The method for designing a medium-wave infrared metasurface filter based on reverse design as claimed in claim 1, characterized in that: In step 5, the probability of an individual being selected according to the roulette principle is defined as:

5. The method for designing a medium-wave infrared metasurface filter based on inverse design as claimed in claims 1 and 3, characterized in that: In step 6, the adaptive weight factor is a function of the iteration number i and the four transmittances of the generation i-1 before the current iteration number, that is:

Images