Design method and device of metasurface converging lens

Abstract

The application discloses a design method and device of a metasurface converging lens, and the method comprises the following steps: obtaining a sample set of a unit band, wherein each unit band in the sample set comprises a plurality of units arranged in a line; calculating the light field distribution of the metasurface converging lens corresponding to each unit band in the sample set; taking the sample set and the light field distribution as training samples to train a deep generation model for analyzing the relationship between the unit band and the light field distribution, and then designing the metasurface converging lens. The method uses symmetry to reduce the dimension of the metasurface converging lens, simplifies the metasurface converging lens into lower-dimensional data, greatly reduces the consumption of computing resources when training the model corresponding to the low-dimensional data and the light field distribution, avoids searching in the parameter space with exponential dimension explosion, and the model corresponding to the metasurface converging lens and the light field distribution has low cost, high speed and high precision, can be used in the design of a super-large parameter space, and better meets the design requirements.

Description

Technical Field

[0001] This invention relates to the fields of optoelectronic material design and deep learning, and particularly to a design method and apparatus for a metasurface converging lens. Background Technology

[0002] Traditional converging lenses rely on factors such as their thickness, shape, and the refractive index of the material to control the focal point, making it difficult to achieve the desired focal length thickness and miniaturization. With the advent of metasurface technology, miniaturization of converging lenses has made new progress. Converging lenses constructed from metasurfaces achieve miniaturization by adjusting the optical path through periodically or quasi-periodically arranged subwavelength scatterers on their surface.

[0003] In the process of developing this invention, the inventors discovered the following problems with existing design methods for metasurface converging lenses:

[0004] Metasurface converging lenses are composed of tens of millions of micro-nano primitives. Existing design methods for metasurface converging lenses include researchers optimizing parameters based on their expertise and training models of the relationship between metasurface converging lenses and corresponding light field intensities using neural networks. Among these methods, manual optimization is inefficient, while training models using neural networks requires a large amount of computational resources, resulting in high design costs, low efficiency, and the potential for introducing discretization errors that lead to inaccurate simulation results.

[0005] Therefore, current design methods cannot meet the requirements of metasurface converging lenses, thus restricting their development. Summary of the Invention

[0006] To address at least one of the aforementioned problems in the prior art, the present invention aims to provide a design method and apparatus for metasurface converging lenses that consumes less computational resources, generates models efficiently, and achieves high accuracy.

[0007] To achieve the above-mentioned objective, one embodiment of the present invention provides a design method for a metasurface converging lens, wherein the metasurface converging lens includes a circular substrate and a plurality of primitives uniformly distributed on the substrate, and the design method includes the following steps:

[0008] Obtain basic parameters, wherein the basic parameters include the parameters of the substrate;

[0009] Obtain a sample set of primitive bands, wherein each primitive band in the sample set includes multiple primitives arranged linearly, the primitive band passes through the center of the metasurface converging lens, the geometric parameters of the primitives in each primitive band are different, and the number of primitive bands in the sample set meets the requirements of gradient backpropagation optimization.

[0010] Based on the aforementioned basic parameters, the light field distribution of the metasurface converging lens corresponding to each primitive band in the sample set is calculated, and each light field distribution is used as a label for the corresponding primitive band.

[0011] Using the sample set and the label as training samples, a deep generation model is trained to obtain a deep generation model for analyzing the relationship between primitive bands and light field distribution.

[0012] Based on the aforementioned depth generation model, the corresponding target primitive band is determined according to the target light field distribution;

[0013] A metasurface converging lens is designed based on the target element band and the basic parameters.

[0014] As a further improvement of the present invention, the step of obtaining the sample set of primitive bands includes:

[0015] Obtain the initial parameters of the primitive band, wherein the initial parameters include the initial geometric parameters of each primitive;

[0016] Each time, the initial geometric parameters of at least some of the primitives in the primitive band are randomly increased or decreased within the error range to obtain a new primitive band. This step is repeated multiple times to obtain the sample set.

[0017] As a further improvement of the present invention, the step of obtaining the initial parameters of the primitive band includes:

[0018] Obtain an initial phase distribution, wherein the initial phase distribution includes the phase value corresponding to each primitive position;

[0019] Acquire phase distribution relationship data, wherein the phase distribution relationship data includes the geometric parameters of the primitives and the phase corresponding to the geometric parameters;

[0020] Based on the initial phase distribution and the phase distribution relationship data, the geometric parameters of the primitive corresponding to each primitive position are determined, and the initial parameters of the primitive band are generated.

[0021] As a further improvement to the present invention, the basic parameters also include wavelength and focal length.

[0022] The method for calculating the initial phase distribution is as follows:

[0023] Based on the parabolic phase formula, the phase value corresponding to each primitive position is calculated according to the wavelength and the focal length.

[0024] As a further improvement of the present invention, the element band is a plurality of elements on the diameter of the metasurface converging lens;

[0025] The step of randomly increasing or decreasing the initial geometric parameters of at least a portion of the primitives within the error range includes:

[0026] While keeping the geometric parameters of the primitives in the designated central region of the primitive band unchanged, the initial geometric parameters of at least some of the primitives in the radius region on one side outside the designated central region are randomly increased or decreased within an error range, wherein the error range is determined according to the process conditions.

[0027] As a further improvement of the present invention, the initial geometric parameters of each element are random values ​​within the process dimension range.

[0028] As a further improvement of the present invention, the deep generative model is a generative adversarial network model, and the conditional generative adversarial network model includes a generator network and a discriminator network;

[0029] The step of training the deep generative model includes:

[0030] Using the sample set as the true values ​​and the light field distribution as the constraints, the deep generative model is trained through adversarial exercises between the generator network and the discriminator network.

[0031] As a further improvement of the present invention, the basic parameters include the arrangement of the basic elements; the step of designing a metasurface converging lens based on the target element band and the basic parameters includes:

[0032] Based on the arrangement, determine the position of the center of each element on the metasurface converging lens;

[0033] Based on the distance from the center of the primitive to the center of the circle, an equivalent primitive corresponding to the center of the primitive is determined on the target primitive band, wherein the distance from the center of the equivalent primitive to the center of the circle is the same as the distance from the center of the primitive to the center of the circle.

[0034] Based on the primitive or the equivalent primitive, determine the geometric parameters of the primitive corresponding to the center of each primitive;

[0035] By combining the aforementioned basic parameters and the geometric parameters of each element, the design parameters for the metasurface converging lens are generated.

[0036] As a further improvement of the present invention, the arrangement method includes a Cartesian network arrangement;

[0037] The step of determining the position of the center of each element on the surface of the metasurface converging lens according to the arrangement includes:

[0038] According to the arrangement, a uniform Cartesian network is set on the surface of the metasurface converging lens, wherein each grid point of the Cartesian network is located at the center of the primitive.

[0039] or,

[0040] The arrangement method includes a circular array arrangement;

[0041] The step of determining the position of the center of each element on the surface of the metasurface converging lens according to the arrangement includes:

[0042] According to the arrangement, multiple annular arrays are set on the surface of the metasurface converging lens, and the position of the center of each element is determined according to the number of elements on each annular ring in the annular array arrangement.

[0043] As a further improvement of the present invention, the step of determining the equivalent primitive corresponding to the lattice point on the target primitive band includes:

[0044] Based on the distance from the center of the target primitive to the center of the circle, and the geometric parameters corresponding to the primitive, determine the distance parameter relationship between the distance and the geometric parameters;

[0045] The geometric parameters of the equivalent primitive are determined based on the distance from the grid point to the center of the circle and the relationship between the distance parameters.

[0046] To achieve one of the above-mentioned objectives, an embodiment of the present invention provides a design apparatus for a metasurface converging lens, the metasurface converging lens comprising a circular substrate and a plurality of primitives uniformly distributed on the substrate, the design apparatus comprising:

[0047] A parameter acquisition module is used to acquire basic parameters, wherein the basic parameters include the parameters of the substrate;

[0048] The sample acquisition module is used to acquire a sample set of primitive bands, wherein each primitive band in the sample set includes multiple primitives arranged linearly, the primitive band passes through the center of the metasurface converging lens, the geometric parameters of the primitives in each primitive band are different, and the number of primitive bands in the sample set meets the requirements of gradient backpropagation optimization.

[0049] The light field distribution calculation module is used to calculate the light field distribution of the metasurface converging lens corresponding to each primitive band in the sample set by combining the basic parameters, and to use each light field distribution as a label for the corresponding primitive band.

[0050] The model training module is used to train the deep generation model by using the sample set and the label as training samples, and to obtain a deep generation model for analyzing the relationship between primitive bands and light field distribution.

[0051] The size determination module is used to determine the corresponding target primitive band based on the target light field distribution according to the depth generation model.

[0052] The design parameter generation module is used to design a metasurface converging lens based on the target primitive band and the basic parameters.

[0053] To achieve one of the above-mentioned objectives, one embodiment of the present invention provides an electronic device, comprising:

[0054] Storage module, used to store computer programs;

[0055] The processing module, when executing the computer program, can implement the steps in the above-described design method for metasurface converging lenses.

[0056] To achieve one of the above-mentioned objectives, one embodiment of the present invention provides a readable storage medium storing a computer program that, when executed by a processing module, can implement the steps in the above-described design method for a metasurface converging lens.

[0057] Compared with existing technologies, the present invention has the following advantages: Before model training, the design method and apparatus for the metasurface converging lens first reduces the dimensionality of the metasurface converging lens using symmetry, simplifying it into lower-dimensional data, such as data of a one-dimensional array of primitive geometric dimensions. When training the model corresponding to the light field distribution of this lower-dimensional data, the consumption of computing resources is greatly reduced compared to the traditional training of the generative model corresponding to the light field distribution of two-dimensional or three-dimensional data of the metasurface converging lens. It avoids searching in a parameter space with exponentially exponential dimensionality and can quickly train the expected generative model. This generative model can be used in the design of ultra-large parameter spaces. In addition, the introduction of low-dimensional data also allows the model to fully learn more possible errors. Therefore, the model for generating the correspondence between the metasurface converging lens and the light field distribution is low-cost, fast, and accurate, better meeting the design requirements. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the structure of a metasurface converging lens according to an embodiment of the present invention;

[0059] Figure 2 This is a schematic diagram of the structure of a metasurface converging lens according to another embodiment of the present invention;

[0060] Figure 3 This is a flowchart of a design method for a metasurface converging lens according to an embodiment of the present invention;

[0061] Figure 4 This is a schematic diagram illustrating the determination of the initial geometric parameters of a primitive based on the parabolic phase formula in one embodiment of the present invention.

[0062] Figure 5 This is a schematic diagram of phase distribution relationship data according to an embodiment of the present invention;

[0063] Figure 6 This is a schematic diagram of a deep generative model according to an embodiment of the present invention;

[0064] Figure 7 This is a schematic diagram of the initial geometric parameters of a primitive based on random value generation, according to another embodiment of the present invention;

[0065] Figure 8 This is a schematic diagram illustrating how the geometric parameters of each primitive are determined using a Cartesian network and interpolation method according to an embodiment of the present invention.

[0066] Figure 9 This is a schematic diagram showing the relationship between the distance from the center of the primitive to the center of the circle and the geometric parameters in an embodiment of the present invention;

[0067] Figure 10 This is a schematic diagram of a design device for a metasurface converging lens according to an embodiment of the present invention;

[0068] Among them, 100 is the metasurface converging lens; 10 is the substrate; 20 is the primitive; and 21 is the simulation region. Detailed Implementation

[0069] The present invention will now be described in detail with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of the present invention.

[0070] One embodiment of the present invention provides a design method and apparatus for metasurface converging lenses that consumes less computational resources, generates models efficiently, and achieves high accuracy.

[0071] Metasurface converging lens

[0072] The metasurface converging lens 100 in this embodiment has the following structure: Figure 1 and 2 As shown, the system includes a circular substrate 10 and a plurality of primitives 20 uniformly distributed on the substrate 10. These primitives 20 have a structure composed of subwavelength scatterers arranged periodically or quasi-periodically. Each primitive 20 is individually designed according to the desired optical field requirements of the metasurface converging lens 100. The design includes material selection, structural geometry parameters, etc., thereby enabling the metasurface converging lens 100 to modulate the amplitude and phase of light, achieving the desired optical response. Compared to traditional lenses, the metasurface converging lens 100 represents an advancement in miniaturization and integration, allowing the entire optical system to be more compact in size.

[0073] The specific design aspects of the metasurface converging lens 100 include material design and the individual design of the geometric parameters of each element. These geometric parameters include the shape, height, period, and corresponding dimensional and rotational angle parameters of the element. The material parameters and the geometric parameters of each element can be obtained through the design method and apparatus for the metasurface converging lens described below, and then the required process parameters can be generated based on these geometric parameters.

[0074] The shape of the primitive can be linear, cylindrical, cuboid, elliptical cylinder, hollow elliptical body, hollow cuboid, etc. Different shapes can correspond to some identical and some different dimensional parameters and rotation angle parameters. Examples of identical parameters include period and height, while examples of different parameters include the radius of a cylinder, the radius and width of a cuboid, and the major and minor axes of an elliptical cylinder. Different primitives can have different rotation angle parameters. Figure 1 and 2 For example, the shape of primitive 20 in the figure is cylindrical. If the material parameters, the height parameter of primitive 20, and the period of primitive 20 are specified in advance, the radius of primitive 20 can be used as a trainable parameter. Only the radius of primitive 20 needs to be determined to generate the corresponding metasurface converging lens 100. In addition, Figure 1 and 2 Only one side of the metasurface converging lens 100 is shown in the image. The other side of the metasurface converging lens 100 can be an optical surface, or it can also be provided with a primitive 20.

[0075] The basic units 20 are evenly distributed on the surface of the substrate 10. The meaning of "evenly distributed" is explained in the following text. Figure 1 and 2 As shown, taking the circular cross-section of the basic element 20 as an example, the centers of all basic elements 20 are equidistantly distributed. Moreover, these basic elements 20 are centrally symmetrically distributed. Each basic element 20 rotates 180° relative to the center and has another basic element 20 that is consistent with it, so that there is no directional requirement for the incident direction of the incident light.

[0076] In addition, there are various ways in which the basic element 20 can be arranged on the substrate 10. These arrangements can all be designed using the design method of the metasurface converging lens 100 described below. Two implementation methods are shown in this article.

[0077] Figure 1An arrangement of primitives 20 with reference to a Cartesian grid is shown, where the lateral spacing between the centers of these primitives 20 is equal to their longitudinal spacing. Correspondingly, simulation regions 21 of the primitives 20 can be divided on the metasurface converging lens 100, each simulation region 21 containing one primitive 20. The simulation region 21 is square, and its side length is equal to the distance between the centers of two adjacent primitives 20. Thus, the periodic parameters of the primitives 20 described above can be determined by specifying the side length of the simulation region 21 or the distance between the centers of two adjacent primitives 20.

[0078] Figure 2 The diagram illustrates a circular array with multiple primitives 20 arranged at equal distances from the center. The number of primitives 20 is equal on the same radius. Here, the number of primitives 20 arranged at a specified distance from the center and the spacing between two adjacent primitives 20 on the same radius can be used as its period.

[0079] In the above text, the periodic parameter can be used as a hyperparameter. Other parameters that can be used as hyperparameters include the thickness of the substrate 10, the length of the simulation region 21 in the thickness direction, the type of incident light, the focal length, and the wavelength.

[0080] Design methods for metasurface converging lenses

[0081] The following section describes the design method for the metasurface converging lens described above. This method allows for the design of metasurface converging lenses that meet the requirements of the target light field distribution. The design process includes three aspects: generating and selecting training samples, training the model, and determining the design parameters based on the model.

[0082] The following is combined with Figures 3-8 This invention describes a method for designing a metasurface converging lens according to an embodiment of the present invention. Although this application provides method operation steps as shown in the following embodiments or flowcharts, the execution order of steps that do not logically have a necessary causal relationship, based on conventional or non-inventive effort, is not limited to the execution order provided in the embodiments of this application. For example, the acquisition order of steps S10 and S20 below does not distinguish between time and sequence.

[0083] Specifically, refer to Figure 3 As shown, the design method of a metasurface converging lens in this embodiment includes the following steps S10 to S70, wherein steps S10 to S30 are the generation and selection of training samples, steps S40 to S50 are the training of the model, and steps S60 to S70 are the determination of design parameters based on the model.

[0084] Training sample generation and selection

[0085] Step S10: Obtain basic parameters, wherein the basic parameters include the parameters of the substrate.

[0086] The parameters of the substrate can be the material and thickness of the substrate. Basic parameters can also include the periodic parameters mentioned above, the length of the simulation region in the thickness direction, the type of incident light, the focal length and wavelength, the shape of the primitive, and even the height of the primitive, etc. For more descriptions of parameters, please refer to the structural description section of the metasurface converging lens above. These basic parameters can be preset before model training in the following text.

[0087] Here, if a parameter is defined as a basic parameter, meaning it doesn't participate in the model training described below, and the final designed metasurface converging lens is produced with that set value, then if it's not used as a basic parameter but as a trainable parameter, the model training process also includes training these parameters, such as the selected material, the shape of the primitive, and its height. These can be used as either basic parameters or trainable parameters as described below. For simplicity, the following explanation uses a cylinder with all primitives of a specified height as an example, with the radius of the primitive as the trainable parameter.

[0088] Step S20: Obtain the sample set of the primitive band.

[0089] Each primitive band in the sample set includes multiple primitives arranged linearly. The primitive band passes through the center of the metasurface converging lens. The geometric parameters of the primitives in each primitive band are different. The number of primitive bands in the sample set meets the requirements of gradient backpropagation optimization.

[0090] A primitive band is a row of primitives passing through the center of a circle in a metasurface converging lens. The way the primitives pass through the center can be either the radius or the diameter. By arranging the primitives in a primitive band, the entire metasurface converging lens can be simplified into a band-shaped structure, thereby converting two-dimensional data or higher-dimensional data (the number of types of trainable parameters determines the number of dimensions) into lower-dimensional data. In this embodiment, it is one-dimensional data, and the primitive band is a set of one-dimensional arrays {R}, that is, an array of radii.

[0091] Preferred, with Figure 4 and 6 For example, the basic unit bands in the following text are explained using diameter as an example, and the diagram also includes the corresponding... Figure 1 The simulated areas are the surface converging lens arrangement patterns of the Chinese Super League, and these simulated areas are arranged horizontally in sequence. Figure 4 and Figure 6 In the diagram, the x-direction is the arrangement direction of the primitives, and the vertical y-direction is the side length of the simulation region.

[0092] Furthermore, the step of obtaining the sample set of the primitive band includes steps S21 and S22.

[0093] Step S21: Obtain the initial parameters of the primitive band, wherein the initial parameters include the initial geometric parameters of each primitive.

[0094] In step S21, obtaining the initial parameters of the primitive band includes several implementation methods, two of which are described below:

[0095] Implementation Method 1

[0096] Step S211: Obtain the initial phase distribution, wherein the initial phase distribution includes the phase value corresponding to each primitive position.

[0097] Step S212: Obtain phase distribution relationship data, wherein the phase distribution relationship data includes the geometric parameters of the primitive and the phase corresponding to the geometric parameters.

[0098] Step S213: Based on the initial phase distribution and the phase distribution relationship data, determine the geometric parameters of the primitive corresponding to each primitive position, and generate the initial parameters of the primitive band.

[0099] Step S211 also includes multiple implementation methods, to Figure 4 For example, the basic parameters also include wavelength and focal length, and the calculation method for the initial phase distribution is as follows:

[0100] Based on the parabolic phase formula, the phase value corresponding to each primitive position is calculated according to the wavelength and the focal length. The calculation formula for the parabolic phase formula is as follows:

[0101]

[0102] Where x is the coordinate position, λ is the wavelength, and f is the focal length. The phase formula determines the phase value corresponding to the position of x.

[0103] In addition, since the initial phase distribution is only used as a reference, the final generated primitive band does not necessarily conform to a certain formula. Therefore, step S211 can also be generated according to other phase formulas, so that the given primitive distribution conforms to the approximate phase distribution law.

[0104] In step S212, when only the radius of the primitive is considered as a trainable parameter, the following parameters can be referenced: Figure 5 As shown, the horizontal coordinate R is the radius of the primitive, and the vertical coordinate... The curve represents the phase corresponding to the primitive, and it is also the phase distribution data. In addition, the phase distribution data can also be presented in tabular form. Here, the same phase can correspond to multiple radii. Therefore, for simplicity, we can select only this interval of the phase distribution data based on the process size range of the radius of the primitive generated during processing.

[0105] In step S213, based on the results of steps S211 and S212, the geometric parameters corresponding to the initial phase distribution are determined, i.e. Figure 5 The radius R of the primitive in the equation.

[0106] Implementation Method 2

[0107] The initial geometric parameters of each primitive are random values ​​within the process dimension range. Figure 6 As shown, a set of vertical radii is randomly generated. It is sufficient that this radius is within the process dimension range, that is, the value that can be satisfied during processing. In extreme cases, all the radii of the primitives can be equal.

[0108] The difference between Implementation 1 and Implementation 2 is that in Implementation 1, the arrangement of primitives roughly meets the phase requirements, so the model training speed is fast and the number of samples required can be relatively small. Since Implementation 2 is randomly generated, the fitting speed may be slower and the number of samples required may be larger.

[0109] Step S22: Each time, the initial geometric parameters of at least some of the primitives in the primitive band are randomly increased or decreased within the error range to obtain a new primitive band, and this step is repeated multiple times to obtain the sample set.

[0110] The element band can be multiple elements on the diameter of the metasurface converging lens. In step S22, only half of the elements are randomly added or removed. The parameters of the other half of the elements, which are symmetrical about the center, are kept consistent with those of the elements at their symmetrical positions to maintain the central symmetry of the metasurface converging lens. In addition, the error range in step S22 can be determined according to the process conditions.

[0111] The following explanation uses the example of the initial geometric parameters of the primitive band originating from the parabolic phase formula of the local periodic approximation in step S211: the process size range is 70nm to 220nm, with an accuracy of 10nm. For a primitive structure with an initial radius of 100nm, its radius will be randomly selected between [70, 80, ..., 140, 150]. This not only sets up the parameter space for the subsequent optimization process but also simulates the process errors generated during mass production in reality, improving the robustness of the finally trained model.

[0112] Furthermore, based on the parabolic phase formula calculated using the local periodic approximation in step S211, if the required converging lens area is too large, step S22 can maintain the initial parameters of the designated central region of the primitive band to satisfy the local periodic approximation unchanged, and only optimize the primitive radius parameters of the edge regions outside the designated central region. In other words, step S22 only randomly increases or decreases the primitives in the edge regions far from the center (i.e., the center of the circle) within the error range. Combined with symmetry, this means that only at least some of the primitives in the radius region on one side outside the designated central region have their initial geometric parameters randomly increased or decreased within the error range. This leverages the characteristic that the size of primitives closer to the center generally remains gradually changing, and the local periodic approximation still maintains good accuracy, thereby avoiding the problem of excessive computational resources consumed in subsequent optimization processes due to an excessive number of primitive structures.

[0113] For example, for a primitive band with 500 primitives, specifying 300 primitives centered on the central region ensures that the radius parameters of these 300 primitives conform to the initial parameters of the parabolic phase formula. Only the radius parameters of the 200 primitives at the edge of the primitive band need adjustment. Considering symmetry, the number of primitives requiring adjustment can be further reduced to 100. This effectively reduces the computational resources consumed when designing large-size converging lenses.

[0114] Since each primitive can be randomly generated multiple times, many non-repeating primitive bands can be obtained. For example, if the number of primitives to be adjusted is 100, even if each primitive has only 10 possible variations, the corresponding number of variations is 10 to the power of 100. Therefore, in this embodiment, the order of magnitude of these primitive bands in the sample set can be expanded to tens of thousands, such as one hundred thousand primitive bands. The number of primitive bands expanded in this way has been experimentally verified and meets the requirements for model training.

[0115] Step S30: Based on the basic parameters, calculate the light field distribution of the metasurface converging lens corresponding to each primitive band in the sample set, and use each light field distribution as the label of the corresponding primitive band.

[0116] The light field distribution of the metasurface converging lens corresponding to the primitive zone can be calculated using programs such as meep and FDTD. The light field distribution is a two-dimensional light field distribution at the plane position corresponding to the focal point, and these light field distributions can be used as labels for the primitive zone.

[0117] Model training

[0118] The above steps have completed the preparation of basic parameters and sample data. Next, we will first determine the deep learning model to be used for training. In this embodiment, the deep learning model is a deep generative model, which can be a generative adversarial network (GAN) model, more preferably a conditional generative adversarial network (CGAN) model. Other deep learning models can also be long short-term memory network models, recurrent neural network models, attention models, Bayesian optimization models, etc.

[0119] The following explanation uses a conditional generative adversarial network (GAN) model as an example. Before model training, step S40, preparation for model training, can be performed. Step S40 specifically includes the following steps S41 to S46:

[0120] Step S41: Training parameter processing.

[0121] Normalizing the sample set of training parameters and the light field distribution used as labels, so that the range and distribution of the data are kept within a reasonable range, such as [0,1] or [-1,1], helps to stabilize the training of the conditional generative adversarial network model.

[0122] Step S42: Define the forward propagation function.

[0123] The conditional generative adversarial network model includes a generator network and a discriminator network. During forward propagation, the generator network contains five transposed convolutional layers, each followed by a batch normalization and ReLU activation layer, and the last layer is activated by the Tanh function. The discriminator network contains five convolutional layers, each followed by a batch normalization and Leaky ReLU activation layer, and the last layer is activated by the Sigmoid function.

[0124] Step S43: Model initialization. The parameter weights in the generator and discriminator networks are randomly orthogonally initialized. The randomly generated weight matrices are orthogonalized to ensure that the initial weights are independent and orthogonal. This approach helps avoid redundancy and overfitting issues during subsequent training.

[0125] Step S44: Define the loss function.

[0126] The joint operation and improvement of the generator network G and the discriminator network D in the competition are accomplished through the loss function. The loss function can calculate the loss value, which in turn guides the learning and adjustment of the model. Through the gradient backpropagation method, the gradient calculated by the loss function is passed to each parameter of the model, thereby updating the parameters and obtaining the target parameters.

[0127] The loss function in this embodiment is calculated using the binary cross-entropy criterion, and can be expressed as:

[0128] min G max D l(G,D)=E x→pdata {logD(x,y)}+E z→pz(z) {log(1-D(G(z,y)))},

[0129] Where E is the prediction result, pdata is the distribution of the training data, pz(z) is the distribution of the latent vector, and log(D(x,y))+log(1-D(G(z,y))) is the loss L of the discriminator. D log(D(G(z,y))) is the generator loss L. G The goal of the training process is to maximize L. D and minimized L G .

[0130] Step S45: Optimizer selection.

[0131] The optimizer can be the Adam optimizer. The input to the Adam optimizer is the model parameters and the learning rate. In this embodiment, the optimization parameters β1 are set to 0.5 and β2 are set to 0.999. In this embodiment, the optimizers for both the generator network G and the discriminator network D are Adam.

[0132] Step S46: Setting parameters for dynamic adjustment of the learning rate.

[0133] Reducing the model's learning rate after each training period can prevent overfitting after multiple training sessions. In this embodiment, a stepped learning rate scheduler can be used to adjust the learning rate, which is reduced after a certain number of training sessions using a stepped learning rate scheduler.

[0134] Step S50: Use the sample set and the label as training samples to train the deep generation model and obtain a deep generation model for analyzing the relationship between primitive bands and light field distribution.

[0135] During model training, the sample set is used as the true value and the light field distribution is used as the constraint condition. The deep generative model is trained through adversarial exercises between the generator network and the discriminator network.

[0136] Determine design parameters based on the model.

[0137] Step S60: Based on the depth generation model, determine the corresponding target primitive band according to the target light field distribution.

[0138] In this step, the target light field distribution is the light field distribution corresponding to the metasurface converging lens required in the current design. Since the depth generation model has established the correspondence between the light field distribution and the primitive band, the target light field distribution can be input into the depth generation model to determine the corresponding target primitive band. Taking the radius of the primitive, which is the parameter to be designed, as an example, the radius distribution on a primitive band corresponding to the target light field distribution can be obtained, and other parameters can refer to the above basic parameters.

[0139] Step S70: Design a metasurface converging lens based on the target primitive band and the basic parameters.

[0140] Corresponding to Figure 1 and 2 The basic parameters include the arrangement pattern of the primitives. For different arrangement patterns, the corresponding metasurface converging lenses are determined. Step S70 includes multiple implementation manners, and two implementation manners are shown below. Among them, implementation manner a corresponds to the metasurface converging lens with the primitive arrangement form referring to the Cartesian grid above Figure 1 and implementation manner b corresponds to the metasurface converging lens with the primitive arrangement form of the circular array above Figure 2 in

[0141] Implementation manner a

[0142] Refer to Figure 7 shown. [[ID=​​​​​​​​​​​​​​​​​​​​ Figure 7 The distance from the center of the basic unit to the center of the circle, with the ordinate R and Figure 5 The meaning of R in these steps is the same, referring to the radius of the primitive. Step S72 also includes:

[0147] Step S721: Determine the distance parameter relationship between the distance and the geometric parameters based on the distance from the center of the target primitive to the center of the circle and the geometric parameters corresponding to the primitive.

[0148] The target primitive carries the distance r from the primitive center to the center of the circle, and the corresponding geometric parameter R of the primitive, which corresponds to... Figure 8 The data consists of multiple discrete points, and a curve can be fitted based on these discrete points, such as... Figure 8 The distance parameter relationship curve in the curve.

[0149] Step S722: Determine the geometric parameters of the equivalent primitive based on the distance from the grid point to the center of the circle and the relationship between the distance parameters.

[0150] Let's go back to this point. Figure 7 Taking one of the primitives as an example, this primitive falls between two primitives on the target primitive band based on its distance to the center of the circle. The distance to the center of the circle is r. Based on the distance r and Figure 8 The corresponding geometric parameter R can be determined, thus corresponding to... Figure 7 The radius of this primitive is R.

[0151] Step S73: Determine the geometric parameters of the primitive at each grid point location based on the primitive or the equivalent primitive.

[0152] Step S74: Combine the basic parameters and the geometric parameters of each primitive to generate the design parameters for the metasurface converging lens.

[0153] By determining the geometric parameters and basic parameters of all the basic elements in step S73, all parameters of the entire metasurface converging lens can be obtained.

[0154] Implementation method b

[0155] Implementation b corresponds to the metasurface converging lens with a ring array of basic elements. Since each basic element on the metasurface converging lens can be identified on the basic element band with a corresponding basic element of the same size, steps S71 and S72 of implementation a are not required. The design of the metasurface converging lens can be completed directly from step S73.

[0156] Compared with the prior art, this embodiment has the following beneficial effects:

[0157] (1) The design method and device of the metasurface converging lens first reduces the dimension of the metasurface converging lens to one-dimensional data before model training. When training the model of the correspondence between the one-dimensional data and the light field distribution, the consumption of computing resources is greatly reduced compared with the traditional training of the model of the correspondence between the two-dimensional data or even three-dimensional data of the metasurface converging lens and the light field distribution. The expected model can be generated quickly. At the same time, the introduction of one-dimensional data also enables the model to fully learn more possible errors. Therefore, the model of the correspondence between the metasurface converging lens and the light field distribution is low in cost, fast and accurate, and better meets the design requirements.

[0158] (2) The design method of metasurface converging lens can be combined with a depth generation model, such as the conditional generative adversarial network model (CGAN). The size of the metasurface converging lens primitives designed by this method does not need to change uniformly. The size design space of each primitive can have a greater degree of freedom, thereby giving the metasurface converging lens primitives a greater degree of expression freedom, and thus meeting more design requirements for light field distribution.

[0159] Design device for metasurface converging lenses

[0160] In one embodiment, a design device for a metasurface converging lens is provided, such as Figure 9 As shown. The design device for this metasurface converging lens includes the following modules, and the specific functions of each module are as follows:

[0161] A parameter acquisition module is used to acquire basic parameters, wherein the basic parameters include the parameters of the substrate;

[0162] The sample acquisition module is used to acquire a sample set of primitive bands, wherein each primitive band in the sample set includes multiple primitives arranged linearly, the primitive band passes through the center of the metasurface converging lens, the geometric parameters of the primitives in each primitive band are different, and the number of primitive bands in the sample set meets the requirements of gradient backpropagation optimization.

[0163] The light field distribution calculation module is used to calculate the light field distribution of the metasurface converging lens corresponding to each primitive band in the sample set by combining the basic parameters, and to use each light field distribution as a label for the corresponding primitive band.

[0164] The model training module is used to train the deep generation model by using the sample set and the label as training samples, and to obtain a deep generation model for analyzing the relationship between primitive bands and light field distribution.

[0165] The size determination module is used to determine the corresponding target primitive band based on the target light field distribution according to the depth generation model.

[0166] The design parameter generation module is used to design a metasurface converging lens based on the target primitive band and the basic parameters.

[0167] It should be noted that for details not disclosed in the design apparatus for the metasurface converging lens in this embodiment of the invention, please refer to the details disclosed in the design method for the metasurface converging lens in this embodiment of the invention.

[0168] Those skilled in the art will understand that the schematic diagram is merely an example of a design device for a metasurface converging lens and does not constitute a limitation on the terminal device of the design device for a metasurface converging lens. It may include more or fewer components than shown, or combine certain components, or different components. For example, the design device for a metasurface converging lens may also include input / output devices, network access devices, buses, etc.

[0169] The design apparatus for metasurface converging lenses may further include computing devices such as computers, laptops, handheld computers, and cloud servers, as well as, but not limited to, processing modules, storage modules, and computer programs stored in the storage modules and executable on the processing modules, such as the metasurface converging lens design method program described above. When the processing module executes the computer program, it implements the steps in the various metasurface converging lens design method embodiments described above, for example... Figure 3 The steps are shown.

[0170] In addition, the present invention also proposes an electronic device, which includes a storage module and a processing module. When the processing module executes the computer program, it can implement the steps in the above-mentioned design method of metasurface converging lens, that is, implement the steps in any one of the technical solutions in the above-mentioned design method of metasurface converging lens.

[0171] The electronic device can be part of a design device integrated into a metasurface converging lens, a local terminal device, or part of a cloud server.

[0172] The processing module can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processing module is the control center of the metasurface converging lens design apparatus, connecting all parts of the entire metasurface converging lens design apparatus via various interfaces and lines.

[0173] The storage module can be used to store the computer programs and / or modules. The processing module realizes various functions of the metasurface converging lens design device by running or executing the computer programs and / or modules stored in the storage module and calling the data stored in the storage module. The storage module may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc. In addition, the storage module may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0174] For example, the computer program can be divided into one or more modules / units, which are stored in a storage module and executed by a processing module to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in a metasurface converging lens design device.

[0175] Furthermore, one embodiment of the present invention provides a readable storage medium storing a computer program that, when executed by a processing module, can implement the steps in the above-described metasurface converging lens design method, that is, implement the steps in any of the technical solutions in the above-described metasurface converging lens design method.

[0176] If the modules integrated in the metasurface converging lens design method are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processing module, it can implement the steps of the various method embodiments described above.

[0177] The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording media, U disks, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added to or subtracted according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0178] It should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0179] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A design method for a metasurface converging lens, the metasurface converging lens comprising a circular substrate and a plurality of primitives uniformly distributed on the substrate, characterized in that, The design method includes the following steps: Obtain basic parameters, wherein the basic parameters include the parameters of the substrate; Obtain a sample set of primitive bands, wherein each primitive band in the sample set includes multiple primitives arranged linearly, the primitive band passes through the center of the metasurface converging lens, the geometric parameters of the primitives in each primitive band are different, and the number of primitive bands in the sample set meets the requirements of gradient backpropagation optimization. Based on the aforementioned basic parameters, the light field distribution of the metasurface converging lens corresponding to each primitive band in the sample set is calculated, and each light field distribution is used as a label for the corresponding primitive band. Using the sample set and the label as training samples, a deep generation model is trained to obtain a deep generation model for analyzing the relationship between primitive bands and light field distribution. Based on the aforementioned depth generation model, the corresponding target primitive band is determined according to the target light field distribution; A metasurface converging lens is designed based on the target element band and the basic parameters.

2. The design method according to claim 1, characterized in that, The step of obtaining the sample set of primitive bands includes: Obtain the initial parameters of the primitive band, wherein the initial parameters include the initial geometric parameters of each primitive; Each time, the initial geometric parameters of at least some of the primitives in the primitive band are randomly increased or decreased within the error range to obtain a new primitive band. This step is repeated multiple times to obtain the sample set.

3. The design method according to claim 2, characterized in that, The steps for obtaining the initial parameters of the primitive band include: Obtain an initial phase distribution, wherein the initial phase distribution includes the phase value corresponding to each primitive position; Acquire phase distribution relationship data, wherein the phase distribution relationship data includes the geometric parameters of the primitives and the phase corresponding to the geometric parameters; Based on the initial phase distribution and the phase distribution relationship data, the geometric parameters of the primitive corresponding to each primitive position are determined, and the initial parameters of the primitive band are generated.

4. The design method according to claim 3, characterized in that, The basic parameters also include wavelength and focal length. The method for calculating the initial phase distribution is as follows: Based on the parabolic phase formula, the phase value corresponding to each primitive position is calculated according to the wavelength and the focal length.

5. The design method according to claim 4, characterized in that, The element band consists of multiple elements along the diameter of the metasurface converging lens; The step of randomly increasing or decreasing the initial geometric parameters of at least a portion of the primitives within the error range includes: While keeping the geometric parameters of the primitives in the designated central region of the primitive band unchanged, the initial geometric parameters of at least some of the primitives in the radius region on one side outside the designated central region are randomly increased or decreased within an error range, wherein the error range is determined according to the process conditions.

6. The design method according to claim 2, characterized in that, The initial geometric parameters of each primitive are random values ​​within the process dimension range.

7. The design method according to claim 1, characterized in that, The deep generative model is a conditional generative adversarial network model, which includes a generator network and a discriminator network. The step of training the deep generative model includes: Using the sample set as the true values ​​and the light field distribution as the constraints, the deep generative model is trained through adversarial training between the generator network and the discriminator network.

8. The design method according to claim 1, characterized in that, The basic parameters include the arrangement of the primitives; the step of designing a metasurface converging lens based on the target primitive band and the basic parameters includes: Based on the arrangement, determine the position of the center of each element on the metasurface converging lens; Based on the distance from the center of the primitive to the center of the circle, an equivalent primitive corresponding to the center of the primitive is determined on the target primitive band, wherein the distance from the center of the equivalent primitive to the center of the circle is the same as the distance from the center of the primitive to the center of the circle. Based on the primitive or the equivalent primitive, determine the geometric parameters of the primitive corresponding to the center of each primitive; By combining the aforementioned basic parameters and the geometric parameters of each element, the design parameters for the metasurface converging lens are generated.

9. The design method according to claim 8, characterized in that, The arrangement method includes Cartesian network arrangement; The step of determining the position of the center of each element on the surface of the metasurface converging lens according to the arrangement includes: According to the arrangement, a uniform Cartesian network is set on the surface of the metasurface converging lens, wherein each grid point of the Cartesian network is located at the center of the primitive. or, The arrangement method includes a circular array arrangement; The step of determining the position of the center of each element on the surface of the metasurface converging lens according to the arrangement includes: According to the arrangement, multiple annular arrays are set on the surface of the metasurface converging lens, and the position of the center of each element is determined according to the number of elements on each annular ring in the annular array arrangement.

10. The design method according to claim 8, characterized in that, The step of determining the equivalent primitive corresponding to the center of the primitive on the target primitive band includes: Based on the distance from the center of the target primitive to the center of the circle, and the geometric parameters corresponding to the primitive, determine the distance parameter relationship between the distance and the geometric parameters; The geometric parameters of the equivalent primitive are determined based on the distance from the center of the primitive to the center of the circle and the relationship between the distance parameters.

11. A design apparatus for a metasurface converging lens, the metasurface converging lens comprising a circular substrate and a plurality of primitives uniformly distributed on the substrate, characterized in that, The design device includes: A parameter acquisition module is used to acquire basic parameters, wherein the basic parameters include the parameters of the substrate; The sample acquisition module is used to acquire a sample set of primitive bands, wherein each primitive band in the sample set includes multiple primitives arranged linearly, the primitive band passes through the center of the metasurface converging lens, the geometric parameters of the primitives in each primitive band are different, and the number of primitive bands in the sample set meets the requirements of gradient backpropagation optimization. The light field distribution calculation module is used to calculate the light field distribution of the metasurface converging lens corresponding to each primitive band in the sample set by combining the basic parameters, and to use each light field distribution as a label for the corresponding primitive band. The model training module is used to train the deep generation model by using the sample set and the label as training samples, and to obtain a deep generation model for analyzing the relationship between primitive bands and light field distribution. The size determination module is used to determine the corresponding target primitive band based on the target light field distribution according to the depth generation model. The design parameter generation module is used to design a metasurface converging lens based on the target primitive band and the basic parameters.

12. An electronic device, characterized in that, include: Storage module, used to store computer programs; The processing module, when executing the computer program, can implement the steps in the design method of the metasurface converging lens according to any one of claims 1 to 10.

13. A readable storage medium storing a computer program, characterized in that, When executed by the processing module, the computer program can implement the steps in the design method of the metasurface converging lens according to any one of claims 1 to 10.

Images