Multi-wavelength high-color-difference metasurface color holographic device and design method thereof

Abstract

The invention discloses a multi-wavelength high-color-difference metasurface color holographic device and a design method thereof.The method comprises the steps that firstly, a nano-column super unit is simulated and then serves as training data of a four-port neural network model, a well-trained forward prediction model is obtained, a prediction response database is constructed, and the prediction response database is used as a prediction result; and according to the phase demand of the target image, a corresponding unit structure is matched in the prediction response database, the multi-wavelength high-chromatic-aberration metasurface color holographic device is designed, and full-mode electromagnetic simulation is carried out. According to the scheme, when holographic imaging is achieved, the RGB energy efficiency is improved, and positive illumination high-efficiency metasurface color holographic imaging can be effectively achieved. The method belongs to a non-filtering type multi-wavelength high-color-difference metasurface color holographic device design method, information of three channels of red, green and blue is coded into sub-pixels at the same time, each sub-pixel responds to RGB light, and compared with a traditional mode that RGB sub-pixels respond respectively, the space utilization rate and energy efficiency of three primary colors are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of metasurfaces, in particular to a multi-wavelength high-color-difference metasurface color holographic device and a design method thereof. Background Art

[0002] Metasurface color holography is one of the common applications of metasurface dispersion control technology, and often realizes independent encoding and reconstruction of RGB three-channel information by precisely controlling the dispersion characteristics of multi-wavelength light. There are three common implementation means for metasurface color holographic imaging technology, including K-space frequency multiplexing, polarization multiplexing, and spatial multiplexing.

[0003] The method of K-space frequency multiplexing uses a unit structure with broadband response. Based on the principle of geometric phase, multiple holograms are reconstructed on all wavelength channels, and the spatial position of holographic imaging is moved by off-axis illumination to display the target image in a preset target area. This method effectively eliminates imaging crosstalk and has high-quality holograms. In theory, only the corresponding monochromatic target image located in the preset target space area in each color component participates in the final color holographic imaging, wasting the light field energy corresponding to other non-target images, and its energy efficiency is low. In addition, the construction of the off-axis holographic imaging optical path is relatively complex, and the optical path debugging is difficult. Positive illumination color holography includes polarization multiplexing and spatial multiplexing schemes.

[0004] The method of polarization multiplexing is based on a unit structure with broadband response. The phase distribution of the target hologram is encoded into independent polarization channels, and the channels are separated according to polarization filtering. However, this method filters the outgoing light, reducing the energy efficiency of the device.

[0005] The method of spatial multiplexing uses a superpixel structure / spatially sub-regional metasurface structure to achieve high-color-difference holographic display. The superpixel structure integrates the sub-unit structures that regulate each wavelength in a super unit, and each sub-unit structure only responds to a single wavelength; the spatially sub-regional metasurface structure divides regions on the metasurface, and each region on the metasurface regulates a single wavelength respectively. The spatial multiplexing method is the most intuitive method to realize high-color-difference metasurface holography.

[0006] However, the coupling between adjacent units in the spatial multiplexing structure will generate electromagnetic crosstalk, resulting in low imaging quality, and the wavelength-selective regulation method often causes huge waste of incident light energy, and the energy efficiency of color holographic imaging is relatively low.

[0007] In summary, there are still relatively serious problems with the energy efficiency corresponding to traditional metasurface color holographic imaging. The off-axis holographic imaging optical path is complex to build, and the on-axis illumination holographic imaging optical path is easy to build. However, currently, in on-axis illumination metasurface color holography, the incident light filtering method is often used to weaken the crosstalk between different color holographic images, which causes great energy loss and limits the energy efficiency of holographic imaging. How to achieve high chromatic aberration metasurface color holography using a non-filtering method is the key to realizing a high-energy efficiency color holographic device. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a multi-wavelength high chromatic aberration metasurface color holographic device and its design method. This method designs a holographic device that conforms to the target phase distribution by obtaining the predicted response data of the supercell at different wavelengths through a multi-port deep neural network model trained with simulation response data.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] The design method of the multi-wavelength high chromatic aberration metasurface holographic device provided by the present invention includes the following steps:

[0011] Step 1: Construct a simulation supercell, set the structural parameters of the simulation supercell; obtain the phase response data of the simulation supercell at different wavelengths, and construct a simulation response database;

[0012] Step 2: Build a multi-port deep neural network model. The multi-port deep neural network model is trained with the data in the simulation response database as training data to obtain a trained multi-port deep neural network model;

[0013] Step 3: Input the generated random structural parameters into the trained multi-port deep neural network model to obtain the response results of the nano-pillar supercells with different parameters at different wavelengths, and establish a predicted response database;

[0014] Step 4: Select the structural parameters of the supercells that need to be designed for the metasurface structure according to the predicted response database, and form a metasurface holographic device according to the structural parameters of the supercells.

[0015] Further, the simulation supercell includes a substrate and a plurality of nano-pillars with multi-wavelength responses arranged on the substrate. The nano-pillars can generate phase response data at different wavelengths by setting geometric parameters.

[0016] Further, the structural parameters of the simulation nano-pillar supercell include length L, width W, and rotation angle θ. Among them, length L = 160 - 280 nm, width W = 40 - 160 nm, rotation angle θ = 0 - 180°, and the height of each nano-pillar is 1 μm.

[0017] Further, there are 4 nanocolumns.

[0018] Further, the multi-port deep neural network model includes a number of input layers, a number of hidden layers, and an output layer;

[0019] Each port of the input layer inputs the geometric parameters of the corresponding nanocolumn respectively;

[0020] A number of hidden layers are independently arranged at each input layer port; the last hidden layer of each input layer port is spliced to form a connection hidden layer; after the connection hidden layer, it is connected to the output layer through a number of hidden layers;

[0021] The output layer is used to output the complex amplitude response result after the nanocolumn is regulated, and the complex amplitude response result includes the real part and the imaginary part of the response at different wavelengths.

[0022] Further, the multi-port deep neural network model is a four-port deep neural network model; the four-port deep neural network model includes:

[0023] Ports of four independent input layers, which respectively receive the structural parameters L, W, θ of four nanocolumns; a hidden layer structure is sequentially connected after the input layer; the hidden layers of the four ports are spliced and processed through a number of hidden layer nodes, and the output layer returns the real part and the imaginary part of the complex amplitude response at different wavelengths; the activation function uses the Tanh function, and the loss function is the mean absolute error MAE.

[0024] Further, the acquisition of the target phase distribution is carried out in the following manner:

[0025] Use the G-S algorithm to iteratively optimize the phase distribution of the RGB channels, and pre-compensate the wavelength-related imaging size difference; adjust the target image size to be inversely proportional to the wavelength, and iteratively correct the amplitude and phase distribution through Fourier transform until it converges to the preset target.

[0026] Further, the different wavelengths include red, green, and blue light.

[0027] Further, for the training of the multi-port deep neural network model in step 2, the Adam optimizer is used to update the weight parameters of the model; the mean absolute error MAE is used to represent the loss function between the network output and the true value, and the specific form is:

[0028]

[0029] Among them, n is the number of training data, y i is the actual response result, and h(x i ) is the corresponding response prediction result.

[0030] The multi-wavelength high chromatic aberration metasurface color holographic device provided by the present invention is obtained through the above-mentioned multi-wavelength high chromatic aberration metasurface color holographic device design method, and includes a metasurface structure composed of a plurality of periodically arranged supercells; the supercell includes a substrate and a plurality of nanocolumns with multi-wavelength responses arranged on the substrate, and the nanocolumns can generate phase modulation at different wavelengths by setting structural parameters.

[0031] The beneficial effects of the present invention are as follows:

[0032] The multi-wavelength high chromatic aberration metasurface color holographic device and its design method provided by the present invention belong to a non-filtering multi-wavelength high chromatic aberration metasurface holographic device design method. By simultaneously encoding the red (R), green (G), and blue (B) channel information into sub-pixels, each sub-pixel responds to RGB light. Compared with the traditional RGB sub-pixel response mode, this method greatly improves the spatial utilization rate and energy efficiency of the three primary colors. This method is based on a deep learning scheme. First, the four-nanocolumn unit structure is simulated, and then it is used as the training data of a four-port DNN (Deep Neural Networks) model to obtain a well-trained forward prediction model and construct a prediction response database. Finally, according to the phase requirements of the target image, the corresponding unit structure is matched in the prediction response database, and a multi-wavelength high chromatic aberration metasurface color holographic device is designed and full-mode electromagnetic simulation is carried out. The effects of this invention are: when implementing holographic imaging, the RGB energy efficiencies of this scheme reach 53.60%, 66.23%, and 67.55% respectively, and this scheme can effectively achieve high-efficiency positive-illumination metasurface color holographic imaging.

[0033] The multi-wavelength response metasurface holographic device provided by the present invention uses data-driven screening and inverse design. First, the phase response data of nanocolumns to R / G / B under different parameters are collected through simulation or experiment, and a "structural parameter-phase mapping" relationship library is established to accelerate neural network training and screening.

[0034] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to make the objectives, technical solutions, and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration.

[0036] Figure 1 It is a flowchart of a high chromatic aberration metasurface design based on a deep learning model.

[0037] Figure 2 It is a schematic diagram of the supercell structure.

[0038] Figure 3 They are the complex amplitude response distribution diagrams corresponding to the red, green, and blue channels of the simulation response database respectively.

[0039] Figure 4 It is a schematic diagram of the four-port DNN model.

[0040] Figure 5 It is the training process of the forward prediction model.

[0041] Figure 6 It is a schematic diagram of the RGB amplitude error and phase error distribution.

[0042] Figure 7 It is a schematic diagram of the G-S algorithm flow and its RGB channel phase distribution.

[0043] Figure 8 It is a multi-wavelength high chromatic aberration metasurface color holographic device.

[0044] Figure 9 It is a simulation result diagram.

[0045] Figure 10 It is a schematic diagram of the theoretical effect. Specific implementation manners

[0046] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the embodiments cited are not intended to limit the present invention.

[0047] As Figure 1 shown, Figure 1 It is the design process of the high chromatic aberration metasurface based on the deep learning model. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device provided in this embodiment. The multi-port deep neural network model provided in this embodiment is set as a four-port deep neural network model according to the actual situation, that is, the four-port DNN (Deep Neural Networks) model is used to encode the information of the red (R), green (G), and blue (B) channels into each four-nanopillar subunit structure, improving the space utilization rate of the metasurface device to alleviate the problem of low energy efficiency of the metasurface color holography using the spectral filtering method. The main steps of the method provided in this embodiment are as follows:

[0048] Step 1: Construct a simulation supercell, use the CST Microwave Studio simulation software to simulate the four-nanopillar supercell, and construct a simulation response database;

[0049] Step 2: Build a four-port DNN model and use the simulation response database as the training data for the four-port DNN model;

[0050] Step 3: Use the trained model to predict the response results of random supercells under three channels, and establish a predicted response database;

[0051] Step 4: For the target image, use the G-S (Gerchberg-Saxton) algorithm to obtain the phase modulation requirements of the RGB channels corresponding to the target image, and screen the corresponding structures in the predicted response database according to the target phase distribution to design a multi-wavelength high chromatic aberration metasurface color holographic device;

[0052] Step 5: Use CST software to simulate the metasurface device to obtain the transmitted light field distribution under plane wave illumination, and then use diffraction theory programming to calculate its final holographic imaging.

[0053] As Figure 2 shown, Figure 2 the front view of the unit structure in (a); the top view of the unit structure in (b). The multi-wavelength high chromatic aberration metasurface color holographic device in this embodiment adopts a supercell composed of four different nanocolumns, and there are 12 change parameters in total, with a high design freedom. Among them, the material used for the nanocolumns in the four-nanocolumn supercell is titanium dioxide (TiO2), and the underlying substrate uses silicon dioxide (SiO2). In the supercell, the height of each nanocolumn is 1μm, and it has different lengths (L), widths (W), and rotation angles (θ) along the x direction. The substrate period P is 0.6μm. The structural parameters of the supercell can be described by the parameter array {S} = {L1, W1,

[0054] θ1, L2, W2, θ2, L3, W3, θ3, L4, W4, θ4}, where the numbers of the letters in the parameter array represent the nanocolumns from 1 to 4.

[0055] The simulation response data in this embodiment is the phase response data of nanocolumns under different parameters for different wavelengths collected in advance through simulation, and a "structural parameter-phase mapping" relationship library is established to accelerate the training of the neural network.

[0056] The predicted response database in this embodiment is the response data and corresponding structural parameter data of the pre-trained four-port DNN model predicting random unit structures under red, green, and blue light illuminations, which increases the candidate unit structures.

[0057] In this embodiment, the commercial software CST Microwave Studio is used to model the supercell to construct a simulation response dataset. First, the simulation units and frequency range are set to ensure that the frequency unit is THz. At the same time, the geometric dimensions of all models are unified based on nm. The refractive index parameter of TiO2 and the dielectric constant of SiO2 are added. The background is set to vacuum, covering the unit structure of the simulation. In terms of boundary condition settings, the x and y directions are configured as unit cell boundaries to simulate the periodic structure; while the z direction is set as an open boundary to simulate the open space. These settings together ensure the accuracy and effectiveness of the simulation. After completing the basic configuration, a substrate with a size of 600 nm × 600 nm is constructed using SiO2. On top of this substrate, four TiO2 nanocolumns are designed, and the geometric parameters of these nanocolumns are carefully regulated: the length (L) is in the range of 160 nm to 280 nm, the width (W) is in the range of 40 nm to 160 nm, and the rotation angle (θ) is in the range of 0° to 180°. To ensure the accuracy of the simulation, a local mesh refinement strategy is implemented in the unit structure area, and the accuracy and precision of the simulation are further improved by finely regulating the number and size of the meshes in the simulation area. In terms of the simulation method, the frequency-domain simulation technology is selected, and left-handed circularly polarized light of three colors RGB with frequencies of 474.08 THz, 563.91 THz, and 634.25 THz (corresponding wavelengths 632.8 nm, 532 nm, 473 nm) is used as the light source to irradiate the unit structure vertically along the substrate direction. To effectively monitor and record the simulation results, an electric field monitor is set 300 nm above the pixel unit to collect the amplitude and phase information of the outgoing right-handed circularly polarized light, which serves as the key response results for subsequent analysis.

[0058] As Figure 3 shown, Figure 3 (a), (b), and (c) in Figure 3Shows the schematic diagram of the complex amplitude distribution of 29,239 randomly generated structures under the normal incidence illumination of RGB three lights. In order to effectively reduce the difficulty in the training process, this embodiment preprocesses the original data set. Considering that the response characteristics of structures with small amplitudes are often complex and difficult to accurately predict, which will bring unnecessary challenges to the training of the neural network. Therefore, this embodiment sets the amplitude threshold to 0.03 and eliminates all structures with amplitudes less than this threshold. This approach aims to optimize the data set so that the neural network can focus more on learning the response data rules of structures with clear response characteristics, thereby obtaining higher prediction accuracy. In addition, it can be observed that the amplitude responses corresponding to all structures show a non-uniform distribution, and the number of structures in the regions with low and high efficiency is also different. However, the phase distribution of the structures shows a relatively uniform distribution in the interval of 0 to 2π.

[0059] Since the multi-wavelength high chromatic aberration metasurface color holographic device designed in this embodiment is a pure phase hologram and focuses on the phase modulation of the structure, the electromagnetic response corresponding to this data set can meet the encoding requirements of the multi-wavelength high chromatic aberration metasurface color holographic device and can be used for the training of the deep learning model in this work.

[0060] Construction and training of the forward prediction model in this embodiment: After collecting the electromagnetic response data of the four-nanopillar supercell, a deep learning model is constructed for the forward design of the multi-wavelength high chromatic aberration metasurface color holographic device. This embodiment adopts a DNN architecture based on four-port input to solve the problem of rapid prediction of the electromagnetic response of the four-nanopillar supercell and realize the function of predicting the response according to the structural parameters. Compared with the traditional single-input DNN model, the multi-input DNN model has multiple independent input layers, which can extract structural features from the input structural parameters respectively and perform calculations and processing in the subsequent hidden layers of the network. This multi-input architecture shows higher applicability when dealing with supercells with multiple subunit structures. In this way, the network can learn the complex relationships between the structure and the electromagnetic response and between different wavelength channels. This in-depth understanding and utilization of the structural features enable the multi-input DNN to more accurately predict and optimize the performance of the supercell during the forward design process.

[0061] As Figure 4 shown, Figure 4 is the schematic diagram of the four-port DNN model. In the figure, Densel-Dense4 represents the fully connected layer, which is used to integrate and transform the features extracted from the previous layers; Concatenate represents the concatenation layer, which is used to merge multiple tensors.

[0062] Each port of the four-port DNN model constructed in this embodiment independently processes the geometric parameters of a nanocolumn structure, i.e., {S1} = {L1, W1, θ1}; {S2} = {L2, W2, θ2}; {S3} = {L3, W3, θ3}; {S4} = {L4, W4, θ4} (all are normalized); four hidden layers are connected after each input layer, and the number of neural nodes is 200, 150, 100, 100 in sequence.

[0063] The weights of the hidden layer neurons between different ports are not shared, ensuring that the structural parameters of each subunit structure can be independently and fully processed. This helps the neural network model to deeply understand the light modulation mechanism of each subunit. After the structural parameters are processed by four hidden layers, the last hidden layer is concatenated respectively to form a hidden layer containing 400 neurons. This is because considering that the incident light will be affected by the interaction between channels after being modulated by the nanosquare columns, this design helps the model to capture the interaction between different channels, so as to more accurately predict the optical response of the metasurface.

[0064] Finally, three hidden layers with the number of neural nodes 150, 100, 100 in sequence are connected for processing, and the response result is obtained at the output layer. The output result of this output layer is the real part and the imaginary part of the complex amplitude response result of the RGB three lights after being modulated by the four-nanocolumn supercell, i.e., {R} = {R_real, R_imag, G_real, G_imag, B_real, B_imag}; considering that the value range of the output complex amplitude result is between -1 and 1, the Tanh function is used as the activation function from the input layer to the output layer.

[0065] During the training process of the model, the Adam optimizer is used to update the weight parameters of the model. The initial learning rate is set to 10 -5 . The mean absolute error (MAE) is used to represent the loss function between the network output and the true value, and the specific form is:

[0066]

[0067] where n is the number of training data, h(x i ) is the corresponding response prediction result, and y i is the actual response result.

[0068] During the training process, 95% of the simulation response database is randomly divided into the training set, and 5% is divided into the test set. The computer configuration used for training is an i9-11900K CPU, an NVIDIA GeForce GTX3070 Ti GPU, and 128GB ddr4 RAM.

[0069] The training process is asFigure 5 As shown Figure 5 This is the training process of the forward prediction model. The forward prediction model is trained for 5000 rounds, and both the training loss and the validation loss show good convergence characteristics, indicating that the fitting ability of the model on the training data reaches an excellent level. The entire training process takes about 5.35 hours.

[0070] To verify the accuracy of the trained model, 500 structures are randomly selected as the validation set to evaluate the performance of the trained model. Table 1 shows the actual responses and predicted responses of some structures. It can be seen that most of the prediction results are relatively consistent with the actual results, and only a very few results show large deviations.

[0071] As Figure 6 shown Figure 6 the (a) RGB amplitude error distribution; (b) RGB phase error distribution in Figure 6 Details the complex amplitude (amplitude and phase) error distribution of the four-port DNN model for 500 structures under red, green, and blue light illumination. The error distribution plot shows that the data points exhibit an obvious concentration trend, with most error values being small and closely clustered in the low-error region. This indicates a low overall data error level and high data accuracy and consistency. Through calculation and statistical analysis, the average amplitude error of the model for red light is 0.058, for green light is 0.046, and for blue light is 0.057; the corresponding RGB phase errors are 25.33°, 9.54°, and 13.28° respectively. These results fully verify the high performance of the model in predicting the modulation results of incident light on four-nanometer pillar unit structures. Compared with traditional numerical calculation methods, the model can significantly reduce the calculation time and achieve fast and relatively accurate electromagnetic response prediction.

[0072] Table 1 Comparison of actual and predicted phases of some units in the test set

[0073]

[0074] As Figure 7 shown Figure 7(a) Schematic diagram of the G-S algorithm process; (b) Phase distribution of the RGB channels. In this embodiment, the device and its simulation results are carried out as follows. After the construction and training of the four-port DNN model are completed, the G-S algorithm is used to obtain the phase distribution of the hologram at the target wavelength as the design goal of the multi-wavelength high chromatic aberration metasurface color holographic device. The G-S algorithm is a classic phase modulation requirement calculation method based on iterative optimization. By alternately constraining the amplitude distribution and phase information of the target light field, it gradually approaches the phase solution that meets the preset conditions and is widely used in fields such as optical holography, metasurface phase distribution design, and wavefront modulation device design. In the design of metasurface functional devices, this algorithm can inversely optimize the phase modulation required for metasurface units at different spatial positions through the intensity distribution of the target light field. However, when dealing with multi-wavelength color holographic imaging targets, different wavelengths will produce holographic images of different sizes in the reconstruction plane due to diffraction effects. Therefore, a wavelength-dependent pre-compensation mechanism needs to be introduced: Theoretically, for two-dimensional holographic projection, the longer the wavelength, the larger the size of the holographic image measured at the same spatial position, and the size of its holographic image is proportional to the wavelength.

[0075] Based on this physical relationship, it is necessary to perform wavelength-dependent preprocessing on the target images of the RGB three channels to eliminate the holographic program size mismatch caused by wavelength differences, and then apply the G-S algorithm to iteratively optimize the phase distributions corresponding to each RGB channel respectively.

[0076] As Figure 7 shown, (a) in

[0077] First, preset the amplitude distribution |f| and the random / estimated phase to synthesize the complex amplitude

[0078] Secondly, obtain the image-plane complex amplitude g = |g|e iΨ ;

[0079] Then, retain the image-plane phase information and replace it with the preset target amplitude |g′|;

[0080] Finally, perform the inverse Fourier transform on the corrected image-plane complex amplitude g′ = |g′|e iΨ to update the object-plane distribution and replace the amplitude distribution. Through the closed-loop iteration of the forward transform (from the spatial domain to the frequency domain) and the inverse transform (from the frequency domain to the spatial domain), the target phase reconstruction is finally achieved.

[0081] As Figure 8 shown, Figure 8For the metasurface holographic device, after obtaining the target phase modulation requirements for the corresponding RGB channels in this embodiment, it is necessary to construct a prediction response database to facilitate screening out the structures that match the target phase modulation requirements. One million supercells are randomly generated, where the random structure length ranges from 160 nm to 280 nm, the width ranges from 40 nm to 160 nm, and the rotation angle ranges from 0° to 180°. The above-trained forward prediction model is used to predict the phase and amplitude results under the illumination of RGB three kinds of light for the above structural parameters, and a prediction response database is constructed. Directly retrieve the required structural design metasurface holographic sample in the database to obtain Figure 8 the multi-wavelength high chromatic aberration metasurface color holographic device in

[0082] The size of the high chromatic aberration metasurface holographic device in this embodiment is 60 μm * 60 μm, that is, 100 pixels * 100 pixels. The CST software is used to perform full-wave simulation calculations on this device. The following is a detailed introduction to the CST full-wave simulation environment settings:

[0083] First of all, the basic configuration of the software needs to be carried out. Open the software and set the required units (the length unit is μm, and the frequency is THz), add the SiO2 substrate, and add the TiO2 material parameters. At the same time, set the background to normal.

[0084] Subsequently, import the full-wave structure, set the height of the structure layer to 1 μm, select the material as TiO2, and set the boundary conditions to open in the x, y, and z directions.

[0085] Next, configure the incident light in the form of a plane wave, and clearly specify it as left-handed circularly polarized light illumination, while ensuring that the mesh division meets the simulation requirements.

[0086] After that, set up an electric field monitor to monitor the simulation process. After confirming that all settings are accurate, start the time-domain simulation.

[0087] After the simulation is completed, extract the optical field distribution data after the incident light is transmitted through the electric field monitor. Use MATLAB to extract the optical field results of the right-handed circularly polarized light, and analyze the diffraction propagation characteristics of the optical field based on the angular spectrum propagation theory.

[0088] As Figure 9 shown, Figure 9 is the simulation result diagram, where (a) is the illumination effect of red light; (b) is the illumination effect of green light; (c) is the illumination effect of blue light; (d) is the illumination effect of red, green, and blue lights simultaneously; they are the results of the high chromatic aberration metasurface color holographic device under the normal incidence illumination of RGB three kinds of light sources, and the final combined effect.

[0089] From Figure 9As can be seen, when irradiated with light of a specific color, each part of a different color appears at a different position, and the result clearly shows that the four-nanometer square columns designed in this embodiment can modulate three kinds of light simultaneously, and Figure 9 It also shows the effect of color overlay: the area where red and green overlap appears yellow, the area where green and blue overlap shows cyan, the area where blue and red overlap shows purple, and when red, green, and blue lights overlap simultaneously, it shows white. This result fully demonstrates that this metasurface holographic device can not only achieve red, green, and blue monochromatic holographic imaging displays, but also can mix and achieve holographic imaging of any color. In addition, it is observed that in the case of red light illumination, the influence of crosstalk images is relatively small, the influence of crosstalk images is the largest under blue light illumination, and green light ranks second. This is because when matching the corresponding unit structure according to the target phase in the prediction response database, the errors between the three channels are not balanced, resulting in interference images in the simulation results.

[0090] To measure the energy utilization rate of the multi-wavelength high chromatic aberration metasurface color holographic device, the energy utilization rate is defined here as:

[0091]

[0092] where E RCP is the energy of the outgoing right-handed circularly polarized light (corresponding to the electric field distribution at 0.3 μm on the sample surface), and E LCP is the energy of the incident left-handed circularly polarized light. The calculated energy utilization rate of holographic imaging is 53.60% in the red light channel, 66.23% in the green light channel, and 67.55% in the blue light channel. Different from the traditional spectral filtering type high chromatic aberration metasurface color holography, the method proposed in this embodiment does not rely on spectral filtering to eliminate crosstalk between RGB channels, and each sub-pixel unit can regulate RGB three kinds of light, ensuring that the incident light energy can be fully utilized. However, due to the crosstalk between the three channels, the energy of the crosstalk holographic image will be calculated into the final energy efficiency when calculating the energy efficiency, resulting in the simulation energy efficiency being slightly higher than the actual situation.

[0093] Compared with the above 100 pixel * 100 pixel design, a metasurface holographic device with a larger size will improve the quality of holographic imaging, thereby reducing crosstalk between RGB channels. To verify this mechanism, this embodiment designs a metasurface holographic device with a size of 198 μm * 198 μm (330 pixels * 330 pixels) and analyzes its theoretical results. First, the phase modulation requirement is calculated for the holographic target image using the G-S algorithm, and then the corresponding structure is matched according to the target response in the prediction response database, and the diffracted holographic image formed by the matched electromagnetic response is calculated according to the angular spectrum theory.

[0094] The diffraction result is as Figure 10 shown.Figure 10 From left to right, it shows the imaging effects corresponding to the RGB three channels respectively and the effect of simultaneous illumination. The observation results show that there is still crosstalk in the three channels, but the crosstalk decreases as the size increases. Therefore, increasing the size of the metasurface helps to improve the imaging quality, which provides a theoretical basis for the subsequent realization of high-quality multi-wavelength high chromatic aberration metasurface color holographic devices.

[0095] Figure 10 It is a schematic diagram of the theoretical effect, where (a), (b), and (c) respectively represent the illumination effects of red, green, and blue lights; (d) represents the simultaneous illumination effect of red, green, and blue lights.

[0096] The multi-wavelength high chromatic aberration metasurface color holographic device designed by using deep learning provided in this embodiment realizes the high energy efficiency regulation of the RGB three channels through a four-nanopillar supercell and a four-port DNN model. Compared with the traditional method, this technology does not require filtering the incident light, significantly improving the energy efficiency of the holographic image (53.60% for red light, 66.23% for green light, 67.55% for blue light), and at the same time supports multi-color mixed display. The imaging performance of the designed multi-wavelength high chromatic aberration metasurface color holographic sample is verified by CST software simulation. This method provides a new solution for designing high energy efficiency and high-fidelity color holographic displays, and has important application prospects in the fields of VR / AR (virtual reality / augmented reality), optical encryption, etc.

[0097] The above-described embodiments are only preferred embodiments given to fully illustrate the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention is subject to the claims.

Claims

1. A design method for a multi-wavelength high chromatic aberration metasurface color holographic device, characterized in that, It includes the following steps: Step 1: Construct a simulation supercell, and set the structural parameters of the simulation supercell; obtain the phase response data of the simulation supercell at different wavelengths, and construct a simulation response database; Step 2: Build a multi-port deep neural network model, and train the multi-port deep neural network model with the data in the simulation response database as training data to obtain a trained multi-port deep neural network model; Step 3: Input the generated random structural parameters into the trained multi-port deep neural network model to obtain the response results of the nano-column supercell with different simulated parameters at different wavelengths, and establish a prediction response database; Step 4: Select the structural parameters of the supercells that need to be designed for the metasurface structure according to the prediction response database, and form a metasurface holographic device according to the structural parameters of the supercells.

2. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 1, wherein The simulation supercell includes a substrate and a plurality of nano-columns with multi-wavelength responses arranged on the substrate. The nano-columns can generate phase response data at different wavelengths by setting geometric parameters.

3. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 2, characterized in that, The structural parameters of the simulation nano-column supercell include length L, width W, and rotation angle θ. Among them, length L = 160 - 280 nm, width W = 40 - 160 nm, rotation angle θ = 0 - 180°, and the height of each nano-column is 1 μm.

4. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 1, wherein There are 4 nano-columns.

5. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 1, characterized in that The multi-port deep neural network model includes several input layers, several hidden layers, and an output layer; Each port of the input layer inputs the geometric parameters of the corresponding nano-column respectively; Several hidden layers are independently set for each input layer port; the last hidden layers of each input layer port are spliced to form a connected hidden layer; after the connected hidden layer, it is connected to the output layer through several hidden layers; The output layer is used to output the complex amplitude response result after the nano-column is regulated. The complex amplitude response result includes the real part and the imaginary part of the response at different wavelengths.

6. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 1, characterized in that The multi-port deep neural network model is a four-port deep neural network model; the four-port deep neural network model includes: Four independent input layer ports, which respectively receive the structural parameters L, W, and θ of the four nano-columns; the input layer is sequentially connected to the hidden layer structure; the hidden layers of the four ports are spliced and processed through several hidden layer nodes, and the output layer returns the real part and the imaginary part of the complex amplitude response at different wavelengths; the activation function uses the Tanh function, and the loss function is the mean absolute error MAE.

7. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 1, wherein, The acquisition of the target phase distribution is carried out in the following manner: Use the G-S algorithm to iteratively optimize the phase distribution of the RGB channels, and pre-compensate the wavelength-related imaging size differences; adjust the target image size to be inversely proportional to the wavelength, and iteratively correct the amplitude and phase distribution through Fourier transform until it converges to the preset target.

8. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 1, characterized in that, The different wavelengths include red, green, and blue light.

9. The design method of the multi-wavelength high chromatic aberration metasurface color holographic device according to claim 1, characterized in that For the training of the multi-port deep neural network model in Step 2, the Adam optimizer is used to update the weight parameters of the model; the mean absolute error MAE is used to represent the loss function between the network output and the true value, and the specific form is: where n is the number of training data, and y i is the actual response result, and h(x i ) is the corresponding response prediction result.

10. Multi-wavelength high chromatic aberration metasurface color holographic device, characterized in that, Obtained by the multi-wavelength high chromatic aberration metasurface color holographic device design method according to any one of the above-mentioned claims 1-9, including a metasurface structure composed of a plurality of periodically arranged supercells; the supercell includes a substrate and a plurality of nanocolumns with multi-wavelength responses arranged on the substrate, and the nanocolumns achieve phase modulation at different wavelengths by setting structural parameters.

Images