Photonic crystal performance prediction method based on photonic crystal surface emitting laser

Abstract

The invention relates to a photonic crystal performance prediction method based on a photonic crystal surface emitting laser, and the method comprises the steps: carrying out the discretization of photonic crystal unit cells, and obtaining a discretization grid; discrete Fourier transform is carried out on the discretized grid to obtain Fourier coefficients of all levels, a position coding array is defined, and a splicing array of the photonic crystal layer is obtained according to the position coding array and real parts and imaginary parts of the Fourier coefficients of all levels; the thickness and dielectric constant of each layer of the epitaxial structure and parameters of the photonic crystal structure composed of the discretized grids are used as input parameters, and performance index parameters of the photonic crystal structure are simulated and calculated through a semi-analytical method coupling wave theory; performing optimization training processing on the neural network model by taking the thickness of each layer of the epitaxial structure and the splicing array as input and taking the performance index parameter as output; and inputting the to-be-predicted photonic crystal into the trained neural network model to obtain a performance test result of the to-be-predicted photonic crystal. According to the invention, efficient prediction of the photonic crystal performance is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optoelectronic technology, in particular to a method for predicting the performance of a photonic crystal based on a photonic crystal surface emitting laser. BACKGROUND

[0002] As an optical material with a periodic dielectric structure, photonic crystals have a wide application prospect in the field of optoelectronic technology due to their unique optical properties such as photonic bandgap and high refractive index contrast. Photonic crystal surface emitting lasers are a new type of laser that uses photonic crystal structure to achieve efficient light emission, and have important application value in the fields of optical communication, optical interconnection, optical display, etc.

[0003] However, the performance of photonic crystals is significantly affected by their structural parameters such as the thickness of each layer of the epitaxial structure, the dielectric constant, and the geometry of the photonic crystal unit cell. Traditional photonic crystal surface emitting lasers can only be designed with simple geometric patterns such as a single circular hole or a single triangular hole, and there is no effective design method for complex pattern combinations such as manifolds or multiple simple pattern combinations. Traditional methods for predicting the performance of photonic crystals mainly rely on complex physical models and numerical simulations such as the Finite Element Method (FEM), the Finite-Difference Time-Domain (FDTD) method, and trial-and-error methods or parameter scanning methods. If the Finite-Difference Time-Domain method is used for optical design, even a single simulation can take several hours, and in the actual global optimization process, the calculation speed is too slow, and it may take several days or even months to optimize the required photonic crystal structure design. These traditional methods have the problems of low efficiency, high computational cost, and long time consumption, which cannot meet the demand for rapid design and optimization. SUMMARY

[0004] The present application provides a method for predicting the performance of a photonic crystal based on a photonic crystal surface emitting laser, aiming to at least solve one of the technical problems existing in the prior art.

[0005] The technical solution of the present application is a method for predicting the performance of a photonic crystal based on a photonic crystal surface emitting laser, which includes: Discretizing the photonic crystal unit cell to obtain a discretized grid; Performing a discrete Fourier transform on the discretized grid to obtain Fourier coefficients at each level, removing duplicates and decomposing the Fourier coefficients at each level into real and imaginary parts, defining a position encoding array, and obtaining a photonic crystal layer stitching array based on the position encoding array, the real and imaginary parts of the Fourier coefficients at each level; The input data set of the photonic crystal structure parameters is generated by the thickness of each layer of the epitaxial structure of the photonic crystal and the discretization grid of the photonic crystal unit cell, and the performance index parameters of the photonic crystal structure are simulated and calculated by the semi-analytical coupled wave theory method with the input data set as the input parameters; The parameter optimization training process of the neural network model is performed with the thickness of each layer of the epitaxial structure of the photonic crystal and the splicing array of the photonic crystal layer as the input and the performance index parameters of the photonic crystal structure as the output, and a trained neural network model is obtained. The performance test result of the photonic crystal to be predicted is obtained by inputting the photonic crystal to be predicted into the trained neural network model.

[0006] According to some embodiments of the present application, the discretization processing of the photonic crystal unit cell includes: For the photonic crystal unit cell region, a plurality of two-dimensional Gaussian functions are randomly generated n The expression of each two-dimensional Gaussian function is:

[0007] wherein, g i (x,y) is a two-dimensional Gaussian function, A i is a two-dimensional Gaussian function amplitude control term, x is the horizontal coordinate in the photonic crystal unit cell, y is the vertical coordinate in the photonic crystal unit cell, σ x , σ y is the variance of the two-dimensional Gaussian function in the x direction, y direction, x 0,i , y 0,i is a two-dimensional Gaussian function peak position control term; The n two-dimensional Gaussian functions are added to form a composite function, and the expression of the composite function is:

[0008] wherein, G i (x,y) is the summation of n two-dimensional Gaussian functions, g i (x,y) is a two-dimensional Gaussian function; A threshold value T0 , taking contour lines G(x,y)= T 0 is a hole boundary; pixelating the hole profile of the hole boundary into a 32x32 two-dimensional permittivity array as a discretization grid.

[0009] According to some embodiments of the present application, further comprising: calculating the value of the two-dimensional permittivity of each point in the discretization grid; sorting the value of the two-dimensional permittivity of each point in the discretization grid according to a randomly generated filling factor between (0, 1) as a reference, and taking the value of the two-dimensional permittivity closest to the filling factor as a reference threshold; judging the size of the value of the two-dimensional permittivity of each point in the discretization grid and the reference threshold; setting the grid points with the value of the two-dimensional permittivity greater than the reference threshold to the permittivity of the first medium, and setting the grid points with the value of the two-dimensional permittivity less than the reference threshold to the permittivity of the second medium; forming a discrete photonic crystal structure according to the permittivity of each point in the discretization grid.

[0010] According to some embodiments of the present application, the discrete Fourier transform is performed on the discretization grid to obtain Fourier coefficients at each level, and the Fourier coefficients at each level are de-duplicated and decomposed into real and imaginary parts, comprising: extracting Fourier coefficients by performing two-dimensional Fourier transform on the discretization grid to obtain complex Fourier coefficients; decomposing the complex Fourier coefficients to obtain real and imaginary parts, and constructing network input features according to the real and imaginary parts; performing complex conjugate redundancy removal on the complex Fourier coefficients, and retaining half of the complex Fourier coefficients to avoid redundancy.

[0011] According to some embodiments of the present application, the definition of the position encoding array comprises: introducing a learnable position encoding to the Fourier coefficients at each level; adding the same embedding item to the real and imaginary parts of each Fourier coefficient to obtain a position encoding array, and the expression of the position encoding array is:

[0012] wherein Input is the position encoding array, ξ is the Fourier coefficient, PE is the position encoding, (m,n) denotes the index of the level corresponding to the position encoding, R denotes the real part, Irepresents an imaginary part.

[0013] According to some embodiments of the present application, the performance index parameters of the photonic crystal structure are simulated and calculated by the semi-analytical coupled-wave theory, including: The thicknesses and dielectric constants of the layers of the epitaxial structure of the photonic crystal, and the discrete photonic crystal structure formed according to the dielectric constants of the points in the discretization grid, together constitute the photonic crystal structure parameter generation input data set, and the input data set is taken as an input parameter, The surface emission efficiency, the quality factor, the radiation constant and the threshold gain of the photonic crystal structure are obtained by simulation and calculation through the semi-analytical coupled-wave theory and the finite element method; The logarithmic value of the quality factor is used as a training target to narrow the numerical range and stabilize the simulation and calculation process; A preset number of simulation data sets are constructed, and the simulation data sets are divided into a training data set and a test training set.

[0014] According to some embodiments of the present application, the neural network model includes 4 fully connected layers, the number of neurons of the 4 fully connected layers is 1024, 512, 256 and 64 in sequence, each fully connected layer is matched with a nonlinear activation function, the thicknesses of the layers of the epitaxial structure of the photonic crystal and the splicing array of the photonic crystal layer are taken as inputs, the performance index parameters of the photonic crystal structure are taken as outputs, the parameters of the neural network model are optimized and trained, and a trained neural network model is obtained, including: The real part of the Fourier coefficient is defined as R, the imaginary part is defined as I, the position encoding array is defined as PE1 and PE2, and PE1=PE2; The splicing array of the photonic crystal layer is obtained according to the position encoding array, the real part and the imaginary part of the Fourier coefficient, and the expression of the splicing array of the photonic crystal layer is [R+PE1, I+PE2]; The thicknesses of the layers of the epitaxial structure of the photonic crystal and the splicing array [R+PE1, I+PE2] of the photonic crystal layer are taken as inputs, the surface emission efficiency, the quality factor, the radiation constant and the threshold gain of the photonic crystal structure are taken as outputs, the thicknesses of the layers of the epitaxial structure of the photonic crystal and the splicing array [R+PE1, I+PE2] of the photonic crystal layer sequentially pass through 4 fully connected layers and 4 nonlinear activation functions for optimization and training, and a trained neural network model is obtained.

[0015] According to some embodiments of the present application, the neural network model further includes an output layer, the output layer includes 1 neuron, and the method further includes: The Sigmoid activation function is used for the surface emission efficiency to normalize the output range to (0, 1); The log value of the quality factor is processed by an identity mapping to maintain consistency of the output data.

[0016] The technical scheme of the present application also relates to a computer device comprising a memory and a processor, wherein the processor implements the above method when executing a computer program stored in the memory.

[0017] The technical scheme of the present application also relates to a computer readable storage medium having computer program instructions stored thereon, wherein the computer program instructions are executed by a processor to implement the above method.

[0018] The prediction method based on the photonic crystal performance of the photonic crystal surface emitting laser provided by the embodiment of the present application at least has one of the following advantages or beneficial effects: the discretization processing of the photonic crystal unit cell can divide the continuous physical space into discrete grid points, realizing the conversion of the complex continuous problem into a form that can be processed by numerical calculation. The discrete Fourier transform is performed on the discretized grid to obtain Fourier coefficients at each level, the purpose being to convert the information in the spatial domain into information in the frequency domain. The Fourier transform can extract the periodic characteristics of the photonic crystal unit cell, and the Fourier coefficients contain the frequency information of the discretized grid, which can reflect the contribution of the periodic structure of the photonic crystal to the wave vector space. The Fourier coefficients are processed to avoid repeated calculation, and then the Fourier coefficients are decomposed into real and imaginary parts, respectively representing the amplitude and phase information of the physical quantity. Then, a position coding array is defined to mark the position information of each grid point in space. According to the position coding array, the real and imaginary parts of the Fourier coefficients, a splicing array is generated. By defining the position coding array, the recognition ability of the neural network model to the spatial distribution characteristics can be strengthened, and the performance prediction accuracy can be improved.

[0019] The thickness, dielectric constant of each layer of the epitaxial structure of the photonic crystal, and the discretized grid of the single photonic crystal unit cell are combined into an input data set of the photonic crystal structure parameters, the semi-analytical method and the coupling wave theory are used to simulate and calculate the photonic crystal structure to obtain the performance index parameters, the epitaxial structure parameters of the photonic crystal and the splicing array are used as inputs, and the performance index parameters of the photonic crystal structure are used as outputs, and the neural network model is trained. By adjusting the parameters of the neural network model through a large number of data input and output samples, a trained neural network model is obtained, which can accurately predict the performance index of the photonic crystal structure. The to-be-predicted photonic crystal is input into the trained neural network model, and the generalization ability of the neural network is used to quickly predict the performance of the unknown structure. The trained neural network model outputs the performance index prediction result of the photonic crystal, thereby realizing efficient prediction of the performance of the photonic crystal and improving the accuracy of the performance prediction result of the photonic crystal.

[0020] Additionally, additional aspects and advantages of the present application will be set forth in part in the description that follows, and in part will become apparent to those having ordinary skill in the art upon examination of the following or can be learned from practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a general flowchart of a method for predicting photonic crystal performance based on a photonic crystal surface emitting laser provided by an embodiment of the present application; Figure 2 is a first detailed flowchart of a method for predicting photonic crystal performance based on a photonic crystal surface emitting laser provided by an embodiment of the present application; Figure 3 is a detailed flowchart of step S200 in the method for predicting photonic crystal performance based on a photonic crystal surface emitting laser provided by an embodiment of the present application; Figure 4 is a detailed flowchart of step S300 in the method for predicting photonic crystal performance based on a photonic crystal surface emitting laser provided by an embodiment of the present application; Figure 5 is a detailed flowchart of step S400 in the method for predicting photonic crystal performance based on a photonic crystal surface emitting laser provided by an embodiment of the present application; Figure 6 is a structural schematic diagram of a neural network model provided by an embodiment of the present application; Figure 7 is a structural schematic diagram of a discretized grid provided by an embodiment of the present application. DETAILED DESCRIPTION

[0022] The concept, specific structure and generated technical effects of the present application will be described clearly and completely below in combination with embodiments and drawings, so as to fully understand the purpose, scheme and effects of the present application.

[0023] It should be noted that, unless otherwise specified, when a certain feature is referred to as being "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. The singular forms "a", "an" and "the" used in this text are also intended to include the plural forms, unless the context clearly indicates otherwise. In addition, unless otherwise defined, all technical and scientific terms used in this text have the same meaning as understood by those skilled in the art. The terms used in the specification herein are only used to describe specific embodiments, and are not intended to limit the present application. The term "and / or" used in this text includes any combination of one or more related listed items.

[0024] It is to be understood that, although the terms first, second, third, etc. can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first element could also be termed a second element, and, similarly, a second element could also be termed a first element, without departing from the scope of the present application. The use of any and all examples, or exemplary language (e.g., "such as", "for instance", etc.) provided herein is intended merely to better illuminate embodiments of the application and does not pose a limitation on the scope of the application unless otherwise claimed.

[0025] Traditional photonic crystal surface emitting lasers tend to be designed with simple geometric patterns such as a single circular hole, a single triangular hole, etc., and there is no effective design method for complex pattern combinations such as manifolds, combinations of multiple simple patterns, etc. Traditional photonic crystal performance prediction methods mainly rely on complex physical models and numerical simulations, such as the Finite Element Method (FEM), the Finite-Difference Time-Domain (FDTD), and trial-and-error methods or parameter scanning methods. If the Finite-Difference Time-Domain method is used for optical design, even a single point simulation takes several hours, and in the actual global optimization process, the calculation speed is too slow, and it may take several days or even months to optimize the required photonic crystal structure design. These traditional methods have the problems of low efficiency, high calculation cost, and long time consumption, and it is difficult to meet the needs of rapid design and optimization.

[0026] Based on this, the embodiment of the present application provides a photonic crystal performance prediction method based on a photonic crystal surface emitting laser. By discretizing the photonic crystal unit cell, extracting its discrete Fourier coefficients, and combining the position encoding technology, the photonic crystal structure parameters are converted into the input features of the neural network. The performance index parameters of the photonic crystal are simulated and calculated by the semi-analytical method coupled wave theory, and are used as the training target of the neural network, and finally the rapid prediction of the photonic crystal performance is realized. This method not only can significantly improve the prediction efficiency of the photonic crystal, but also can improve the accuracy of the prediction results of the photonic crystal performance to a certain extent, and provides a new technical means for the design and optimization of the photonic crystal.

[0027] Reference Figure 1 As shown in the figure, Figure 1 is a general flowchart of a photonic crystal performance prediction method based on a photonic crystal surface emitting laser provided by the embodiment of the present application. The photonic crystal performance prediction method based on a photonic crystal surface emitting laser includes but is not limited to steps S100 to S500, specifically, S100: Discretize the photonic crystal unit cell to obtain a discretized grid; S200: Discrete Fourier transform is performed on the discretized grid to obtain Fourier coefficients at each level, the Fourier coefficients at each level are de-duplicated and decomposed into real and imaginary parts, a position encoding array is defined, and a splicing array of the photonic crystal layer is obtained according to the position encoding array, the real and imaginary parts of the Fourier coefficients at each level; S300: An input data set is generated by taking the thickness and dielectric constant of each layer of the epitaxial structure of the photonic crystal and the discretized grid of the photonic crystal unit cell together to form a photonic crystal structure parameter, and the input data set is taken as an input parameter to simulate and calculate the performance index parameter of the photonic crystal structure by a semi-analytical method coupled wave theory; S400: The parameters of the neural network model are optimized and trained by taking the thickness of each layer of the epitaxial structure of the photonic crystal and the splicing array of the photonic crystal layer as inputs and the performance index parameter of the photonic crystal structure as outputs, to obtain a trained neural network model; S500: The photonic crystal to be predicted is input into the trained neural network model to obtain the performance test result of the photonic crystal to be predicted.

[0028] In some embodiments of the present application, the photonic crystal performance prediction method based on the photonic crystal surface emitting laser includes: discretizing the photonic crystal unit cell, which can divide the continuous physical space into discrete grid points, and realize the conversion of the complex continuous problem into the form that can be processed by numerical calculation. The discretized grid obtained after discretization can be regarded as a digital representation of the photonic crystal unit cell, and each grid point corresponds to a physical quantity (such as dielectric constant, etc.). Discrete Fourier transform is performed on the discretized grid to obtain Fourier coefficients at each level, which aims to convert the information in the spatial domain into the information in the frequency domain. Fourier transform can extract the periodic characteristics of the photonic crystal unit cell, and the Fourier coefficients contain the frequency information of the discretized grid, which can reflect the contribution of the periodic structure of the photonic crystal to the wave vector space. The Fourier coefficients are de-duplicated to avoid repeated calculation, and then decomposed into real and imaginary parts to represent the amplitude and phase information of the physical quantity. Then, a position encoding array is defined to mark the position information of each grid point in space. According to the position encoding array, the real and imaginary parts of the Fourier coefficients, a splicing array is generated. This array can be regarded as a digital feature representation of the photonic crystal unit cell, which is used for subsequent calculation and modeling. By defining the position encoding array, the recognition ability of the neural network model to the spatial distribution characteristics can be enhanced, and the performance prediction accuracy can be improved.

[0029] The thickness of each layer of the epitaxial structure of the photonic crystal, the dielectric constant, and the discretized grid of the single photonic crystal cell, and other parameters are combined into an input data set of photonic crystal structure parameters, which collectively determines the physical characteristics of the photonic crystal. The semi-analytical method and the coupling wave theory are used to simulate and calculate the photonic crystal structure to obtain performance index parameters. This step is an accurate calculation based on a physical model. The epitaxial structure parameters and the splicing array of the photonic crystal are used as inputs, and the performance index parameters of the photonic crystal structure are used as outputs to train the neural network model. Through a large number of data input and output samples, the parameters of the neural network model are adjusted to obtain a trained neural network model that can accurately predict the performance index of the photonic crystal structure. The new or to-be-predicted photonic crystal is input into the trained neural network model, and the generalization ability of the neural network is used to quickly predict the performance of the unknown structure. The trained neural network model outputs the performance index prediction result of the photonic crystal, thereby realizing efficient prediction of the performance of the photonic crystal and improving the accuracy of the performance prediction result of the photonic crystal.

[0030] It can be understood that the photonic crystal structure includes an epitaxial structure and a photonic crystal layer, and the method for predicting the performance of the photonic crystal based on the photonic crystal surface-emitting laser provided in the embodiments of the present application first generates an input data set of the photonic crystal structure. For the epitaxial structure, the thickness and dielectric constant of each layer are directly used as inputs. For the photonic crystal layer, a “sampling-threshold method” is used. For the region in the photonic crystal cell, a function z=f(x,y) is defined, where f is a binary function (which can be a complex function, a polynomial function, an exponential function, a logarithmic function, a Gaussian function, a Lorentz function, or a combination of linear or multiplication or convolution thereof). The photonic crystal cell is discretized, and the discretized grid is calculated. Discrete Fourier transform is performed on the discretized grid to obtain Fourier coefficients at each level. After removing the duplicate Fourier coefficients, the real and imaginary parts are separated and flattened into one-dimensional arrays R and I. In order to ensure the correspondence between the real part R and the imaginary part I, position encoding arrays PE1 and PE2 (PE1=PE2) are defined, and the final photonic crystal layer is input in the form of a splicing array [R+PE1, I+PE2]. It should be noted that PE1 and PE2 here can be function position encoding or parameter position encoding for training.

[0031] When generating the data set, the thickness and dielectric constant of each layer of the epitaxial structure of the photonic crystal, the discretized grid of the photonic crystal cell, and other parameters are combined to generate an input data set of the photonic crystal structure parameters. The semi-analytical method and the coupling wave theory are used to simulate and calculate the performance index parameters of the photonic crystal structure, such as the quality factor, the surface radiation efficiency, the radiation constant, and the threshold gain.

[0032] When training the neural network model, the thickness of each layer of the epitaxial structure of the photonic crystal, and the splicing array [R+PE1, I+PE2] of the photonic crystal layer are taken as inputs, and the performance index parameter of the photonic crystal structure is taken as output. The parameters of the neural network model are optimized and minimized to evaluate the loss function of the difference between the output result and the simulation result. Finally, the trained neural network model is used as a fast solver to give the performance test result of the photonic crystal to be predicted.

[0033] The trained neural network model can quickly predict the performance of unknown structures through the generalization ability of the neural network. The trained neural network model outputs the performance index prediction result of the photonic crystal, thereby realizing efficient prediction of the performance of the photonic crystal and improving the accuracy of the performance prediction result of the photonic crystal.

[0034] In some embodiments of the present application, the step S100 in the method for predicting the performance of the photonic crystal of the photonic crystal surface emitting laser includes but is not limited to the following steps: Discretizing the photonic crystal unit cell to obtain a discretized grid includes: S101: For the photonic crystal unit cell region, randomly generating n two-dimensional Gaussian functions, and the expression of each two-dimensional Gaussian function is:

[0035] wherein, g i (x,y) is a two-dimensional Gaussian function, A i is a two-dimensional Gaussian function amplitude control term, x is the horizontal coordinate in the photonic crystal unit cell, y is the vertical coordinate in the photonic crystal unit cell, σ x , σ y is the variance of the two-dimensional Gaussian function in the x direction, y direction, x 0,i , y 0,i is a two-dimensional Gaussian function peak position control term; S102: Adding the n two-dimensional Gaussian functions to form a composite function, and the expression of the composite function is:

[0036] wherein, G i (x,y) is a two-dimensional Gaussian function, nSum of two-dimensional Gaussian functions, g i (x,y) two-dimensional Gaussian function; S103: Set threshold T 0 , take contour G(x,y)= T 0 hole boundary; S104: Pixelate the hole profile of the hole boundary into a 32x32 two-dimensional dielectric constant array as a discretization grid.

[0037] Randomly generate n two-dimensional Gaussian functions in the photonic crystal unit cell region, the parameters (such as amplitude A i , center coordinates (x 0 ,y 0 ) , standard deviation σ x and σ y ) of each two-dimensional Gaussian function can be randomly selected to simulate holes of different positions and sizes. Add these n two-dimensional Gaussian functions to form a composite function, which can represent the superposition effect of multiple holes. Set a threshold T 0 , used to distinguish hole regions and non-hole regions. Extract the hole boundary by contour G(x,y) = T 0 , which represents the set of all points whose function values are equal to T 0 at this threshold T 0 , and the area surrounded by these points can be regarded as the boundary of the hole.

[0038] Pixelate the extracted hole boundary profile to generate a 32x32 two-dimensional dielectric constant array, which can be regarded as a discretization grid of the photonic crystal unit cell.

[0039] In the embodiments of the present application, the discretization grid is generated by the two-dimensional Gaussian cross-section fitting method, which can generate a hole profile with smooth boundary, facilitating photonic crystal structure modeling and subsequent processing. By randomly generating Gaussian functions, a variety of complex hole structures can be simulated, increasing the diversity of photonic crystal design. The shape of the Gaussian function can be adjusted by parameters, thereby generating holes of different sizes and shapes. Pixelating the hole profile into a discretization grid facilitates subsequent numerical calculation and simulation, and this method has high computational efficiency.

[0040] Referring to Figure 2 as shown, Figure 2 is the first detailed flowchart of the method for predicting the performance of photonic crystals based on photonic crystal surface emitting lasers provided by the embodiments of the present application. After the discretization of the photonic crystal unit cell is performed in step S100 to obtain a discretized grid, the method for predicting the performance of photonic crystals based on photonic crystal surface emitting lasers further includes but is not limited to steps S110 to S150, specifically, S110: calculating the value of the two-dimensional dielectric constant of each point in the discretized grid; S120: sorting the value of the two-dimensional dielectric constant of each point in the discretized grid with a randomly generated filling factor between (0, 1) as a reference, and taking the value of the two-dimensional dielectric constant closest to the filling factor as the reference threshold; S130: judging the size of the value of the two-dimensional dielectric constant of each point in the discretized grid and the reference threshold; S140: setting the grid points with the value of the two-dimensional dielectric constant greater than the reference threshold as the dielectric constant of the first medium, and setting the grid points with the value of the two-dimensional dielectric constant less than the reference threshold as the dielectric constant of the second medium; S150: forming a discrete photonic crystal structure according to the dielectric constant of each point in the discretized grid.

[0041] In some embodiments of the present application, the discretized grid is to divide the photonic crystal unit cell area into a series of discrete points, each point representing a small physical area. At each grid point, the corresponding dielectric constant is calculated according to whether it is in the hole region or the non-hole region. A randomly generated filling factor between (0, 1) is taken as a reference. The filling factor is a random number between 0 and 1, representing the proportion of the hole region in the photonic crystal unit cell. The values of the two-dimensional dielectric constant of each point in the discretized grid are sorted from the minimum value to the maximum value. According to the filling factor, the value closest to the filling factor in the sorted dielectric constant values is found. This value of the two-dimensional dielectric constant is taken as a reference threshold for distinguishing the hole region and the non-hole region. For each point in the discretized grid, its two-dimensional dielectric constant value is compared with the reference threshold. The grid points with the two-dimensional dielectric constant value greater than the reference threshold are set as the dielectric constant of the first medium (usually high dielectric constant). The grid points with the dielectric constant value less than the reference threshold are set as the dielectric constant of the second medium (usually low dielectric constant). Finally, a discrete photonic crystal structure is formed according to the dielectric constant of each point in the discretized grid. This structure can be used for subsequent simulation calculation and performance analysis.

[0042] In one embodiment of the present application, judging the size of the value of the two-dimensional dielectric constant of each point in the discretized grid and the reference threshold includes: In the hole area (i.e. G(x,y) < T 0 The dielectric constant can be set to a lower value (for materials with a low dielectric constant, such as air) In the non-hole area (i.e. G(x,y) ≥ T 0 region), the dielectric constant can be set to a higher value (high dielectric constant materials such as GaAs).

[0043] It should be noted that this method performs oversampling and then takes the average to reduce the sudden change of adjacent grids.

[0044] Reference Figure 3 As shown, Figure 3 This is a detailed flow chart of step S200 in the method for predicting photonic crystal performance based on a photonic crystal surface emitting laser provided by an embodiment of the present invention. In step S200, a discrete Fourier transform is performed on the discretized grid to obtain Fourier coefficients at each level. The Fourier coefficients at each level are deduplicated and decomposed into real and imaginary parts, including but not limited to steps S210 to S230. Specifically, S210: performing a two-dimensional Fourier transform on the discretized grid to extract Fourier coefficients to obtain complex Fourier coefficients; S220: Decompose the complex Fourier coefficients to obtain real and imaginary parts, and construct network input features based on the real and imaginary parts; S230: Perform complex conjugate redundancy removal on the complex Fourier coefficients, and retain half of the complex Fourier coefficients to avoid redundancy.

[0045] In some embodiments of the present invention, a method for predicting the performance of a photonic crystal based on a photonic crystal surface emitting laser includes: performing a Fourier transform on a discretized grid to convert information in the spatial domain into information in the frequency domain to obtain complex Fourier coefficients ξm,n, The complex Fourier coefficients are decomposed to obtain the real part ξm,nR and the imaginary part ξm,nI. The real part ξm,nR and the imaginary part ξm,nI represent the amplitude and phase information of the Fourier coefficients, respectively. These coefficients reflect the contribution of the periodic structure of the photonic crystal to the wave vector space. The real and imaginary parts constitute the network input features. The decomposition of the real and imaginary parts provides richer feature information, which facilitates the training and prediction of machine learning models. These features can be used in subsequent neural network model training. Due to the properties of the Fourier transform, the complex Fourier coefficients have complex conjugate symmetry, so that the Fourier coefficients satisfy ξ−m,−n=ξm,n∗. To avoid redundancy, only half of the complex Fourier coefficients can be retained. Complex conjugate redundancy removal is performed on the complex Fourier coefficients, reducing the amount of data and improving computational efficiency.

[0046] In some embodiments of the present invention, the position coding array defined in step S200 in the method for predicting the performance of photonic crystals based on photonic crystal surface emitting lasers includes but is not limited to steps S240 to S250. Specifically, S240: Introducing learnable positional encoding for Fourier coefficients at all levels; S250: Add the same embedding term to the real and imaginary parts of each Fourier coefficient to obtain a positional encoding array. The expression of the positional encoding array is:

[0047] Among them, Input is the position encoding array, ξ are the Fourier coefficients, PE is the position code, (m,n) Indicates the level subscript corresponding to the position code, R represents the real part, I represents the imaginary part. For example, Represents the real part of the (1, 2)-order Fourier coefficient.

[0048] Introducing learnable positional encoding when processing Fourier coefficients is a technique that enhances the capabilities of neural network models. It helps them better understand the spatial relationships in data structures, thereby improving their expressive power. Positional encoding is used for sequential data processing. For Fourier coefficients, positional encoding provides an additional dimension to the real and imaginary parts of each Fourier coefficient. This dimension contains positional information, allowing the neural network model to learn the differences in the contributions of Fourier coefficients at different positions to the final output.

[0049] It's important to note that positional encodings can be initialized to random values ​​and then optimized during neural network model training. Positional encodings can be fixed or learnable parameters. If learnable, they are updated via backpropagation. Positional encodings are learnable, meaning they can automatically adapt to data to better suit a specific task.

[0050] Reference Figure 4 As shown, Figure 4 This is a detailed flow chart of step S300 in the method for predicting the performance of a photonic crystal based on a photonic crystal surface emitting laser provided by an embodiment of the present invention. In step S300, the performance index parameters of the photonic crystal structure are calculated by semi-analytical coupled wave theory simulation, including but not limited to steps S310 to S340. Specifically, S310: generate an input data set from the thickness and dielectric constant of each layer of the epitaxial structure of the photonic crystal, and the discrete photonic crystal structure formed according to the dielectric constant of each point in the discretization grid, and use the input data set as an input parameter; S320: simulate and calculate the surface emission efficiency, quality factor, radiation constant and threshold gain of the photonic crystal structure by the semi-analytical method coupled wave theory and the finite element method; S330: use the logarithmic value of the quality factor as a training target to reduce the numerical range and stabilize the simulation calculation process; S340: construct a predetermined number of simulation data sets, and divide the simulation data sets into a training data set and a test training set.

[0051] In some embodiments of the present application, the performance index parameters of the photonic crystal structure are simulated and calculated by the semi-analytical method coupled wave theory, including: generating an input data set from the thickness and dielectric constant of each layer of the epitaxial structure of the photonic crystal, and the discrete photonic crystal structure formed according to the dielectric constant of each point in the discretization grid, and using the input data set as an input parameter.

[0052] The semi-analytical method coupled wave theory is a method for analyzing the electromagnetic wave propagation characteristics of periodic structures (such as photonic crystals), which combines analytical solutions and numerical calculations to improve computational efficiency and accuracy. The finite element method is a numerical method for solving problems involving partial differential equations, which can be used to accurately simulate the distribution and propagation of electromagnetic fields in the simulation of photonic crystal structures. Through the above simulation method, the surface emission efficiency, quality factor, radiation constant and threshold gain of the photonic crystal structure are calculated. Then, the logarithmic value of the quality factor is used as a training target, which is to reduce the numerical range and make the simulation calculation process more stable, and also helps the training and convergence of the neural network model.

[0053] A simulation data set containing 200,000 samples is constructed, which will be used to train and test the neural network model. The simulation data set is divided into a training data set and a test training set, of which 95% is used as the training data set and 5% is used as the test training set. The training data set is used to train the neural network model, and the test training set is used to evaluate the performance of the neural network model.

[0054] The high-precision simulation calculation of the photonic crystal structure is performed by coupling the wave theory and the finite element method through the semi-analytical method, surface emission efficiency, quality factor, radiation constant and threshold gain of the photonic crystal structure can be quickly predicted, and the performance of the photonic crystal is quickly predicted; the logarithmic value of the quality factor is used as a training target to reduce the numerical range and stabilize the simulation calculation process, and the design parameters of the photonic crystal are optimized by using the prediction results, so that the efficiency of the photonic crystal design and performance prediction can be significantly improved, and the cost of experiments and simulations can be reduced.

[0055] In some embodiments of the present application, the method for predicting the performance of the photonic crystal of the photonic crystal surface emitting laser further comprises constructing a neural network model, the constructed neural network model is a full connection feedforward neural network, the neural network model comprises four full connection layers, the number of neurons of the four full connection layers is 1024, 512, 256 and 64 in sequence, and each full connection layer is matched with a nonlinear activation function. Figure 5 Figure 5 is a detailed flowchart of step S400 in the method for predicting the performance of the photonic crystal of the photonic crystal surface emitting laser provided by the embodiments of the present application, in S400, the thickness of each layer of the epitaxial structure of the photonic crystal and the splicing array of the photonic crystal layer are taken as inputs, the performance index parameters of the photonic crystal structure are taken as outputs, the parameters of the neural network model are optimized and trained, and a trained neural network model is obtained, which includes but is not limited to steps S410 to S430, and specifically, S410: defining the real part of each level of Fourier coefficient as R and the imaginary part as I, defining the position encoding array as PE1 and PE2, and PE1=PE2; S420: obtaining the splicing array of the photonic crystal layer according to the position encoding array, the real part and the imaginary part of each level of Fourier coefficient, and the expression of the splicing array of the photonic crystal layer is [R+PE1, I+PE2]; S430: taking the thickness of each layer of the epitaxial structure of the photonic crystal and the splicing array [R+PE1, I+PE2] of the photonic crystal layer as inputs, and taking the surface emission efficiency, the quality factor, the radiation constant and the threshold gain of the photonic crystal structure as outputs, the thickness of each layer of the epitaxial structure of the photonic crystal and the splicing array [R+PE1, I+PE2] of the photonic crystal layer are sequentially subjected to optimization training through four full connection layers and four nonlinear activation functions, and a trained neural network model is obtained.

[0056] ​The real part of each level of Fourier coefficient is defined as R, and the imaginary part is defined as I, and the position encoding array is defined as PE1 and PE2, and PE1=PE2; the position encoding is used to provide additional position information for the neural network model, helping the neural network model to understand the spatial structure of the data. The splicing array of the photonic crystal layer is obtained according to the position encoding array, the real part and the imaginary part of each level of Fourier coefficient, and the expression is [R+PE1, I+PE2]. This splicing array combines the Fourier coefficient and the position encoding to form the input feature of the neural network model.

[0057] The thickness of each layer of the epitaxial structure of the photonic crystal and the splicing array [R+PE1, I+PE2] of the photonic crystal layer are taken as inputs. The input data successively passes through 4 fully connected layers and 4 nonlinear activation functions. The fully connected layer is used to extract features, and the nonlinear activation function such as PReLU is used to introduce nonlinearity, so that the neural network model can learn complex function mapping; the surface emission efficiency, the quality factor, the radiation constant and the threshold gain of the photonic crystal structure are taken as outputs, and these performance indicators are the prediction targets of the neural network model.

[0058] The neural network model parameters are adjusted by an optimization algorithm (such as gradient descent) to minimize the difference between the predicted output and the true value, and this process involves the definition of a loss function, such as mean square error (MSE). During the training process, the validation set is used to evaluate the performance of the neural network model to prevent overfitting. The neural network model is optimized by adjusting hyperparameters (such as learning rate, regularization coefficient, etc.).

[0059] In the embodiments of the present application, the input features are enhanced by Fourier coefficients and position encodings, and the prediction ability of the neural network model is improved. The nonlinear activation function is used to make the neural network model learn complex nonlinear relationships. The constructed neural network model can automatically predict the performance indicators of the photonic crystal, improving the design efficiency. This method can significantly improve the efficiency of photonic crystal design and performance prediction, and reduce the cost of experiments and simulations.

[0060] In some embodiments of the present application, the neural network model further includes an output layer, and the output layer includes 1 neuron, and the method further includes: S440: using a Sigmoid activation function for the surface emission efficiency to normalize the output range to (0, 1); S450: adopting identity mapping processing for the logarithmic value of the quality factor to keep the consistency of the output data.

[0061] In some embodiments of the present application, the surface emission efficiency is normalized using a Sigmoid activation function, the output range of the Sigmoid function is (0, 1), which makes it very suitable for normalizing the output that should theoretically vary between 0 and 1; the logarithmic value of the quality factor adopts an identity mapping, that is, directly outputting the linear combination result of neurons without any nonlinear transformation, which can maintain the consistency of data, without any form of scaling or transformation of data, and directly output the logarithmic value of the quality factor. The prediction of the neural network model will directly reflect the relationship between the input features and the output, which can quickly predict the performance of unknown structures, realize efficient prediction of the performance of photonic crystals and improve the accuracy of the performance prediction results of photonic crystals.

[0062] In some embodiments of the present application, the optimization training method of the neural network model further comprises: using an Adam optimizer to optimize and train the neural network model, and setting the learning rate of the Adam optimizer to be between 0.0005 and 0.0015, the batch size to be 64, 128, 256, 512, and improving the generalization ability of the neural network model. The Adam optimizer is an adaptive gradient descent optimizer that combines the advantages of AdaGrad and RMSProp, which can automatically adjust the learning rate and usually has good convergence. SmoothL1Loss is a loss function between L1 and L2 loss, which is not sensitive to outliers, and the parameter beta defines the transition point from L1 loss to L2 loss, which is usually set between 0.05 and 0.15. The maximum number of iterations of the neural network model trained on the training data is set to 100, which improves the performance of the neural network model and avoids overfitting. The early stopping mechanism is a technique to prevent overfitting, which will terminate the training when the loss on the validation dataset does not improve after a certain number of iterations.

[0063] The following describes the prediction method of the performance of the photonic crystal photonic crystal surface emitting laser provided by the embodiments of the present application with a complete embodiment.

[0064] The embodiments of the present application construct a neural network model based on discrete Fourier coefficients, and complete the complete data preparation, training process design and parameter tuning, which are described in detail as follows: Data preparation: The construction of the input data set adopts a two-dimensional Gaussian cross-section fitting method (2D-GCF), which generates a hole structure with randomness and smooth boundaries by superimposing three Gaussian functions, and is discretized into a permittivity distribution matrix of 32x32 to obtain a discretized grid. Each discretized grid is then extracted by a two-dimensional Fourier transform to obtain Fourier coefficients ξm,n, and the real and imaginary parts are separated to form the input features (a total of 1024 dimensions) of the neural network model. To enhance the perception of spatial position information of the neural network model, a learnable position encoding is introduced for all Fourier coefficients. According to the position encoding array, the real and imaginary parts of the Fourier coefficients at each level, a splicing array of the photonic crystal layer is obtained; the input data set is generated by the photonic crystal structure parameters composed of the thickness and permittivity of each layer of the epitaxial structure of the photonic crystal and the discretized grid of the photonic crystal unit cell, and the input data set is used as an input parameter. The performance index parameters of the photonic crystal structure are obtained by simulation calculation of the input parameter by the coupled wave theory and the finite element method, including but not limited to two key performance indicators: Surface-Emitting Efficiency (SEE); Quality Factor (Q), and the logarithmic value logQ is used as a training target to reduce the numerical range and improve the training stability.

[0065] Finally, a simulation data set containing 200,000 samples is constructed, which will be used to train and test the neural network model. The simulation data set is divided into a training data set and a test training set, 95% of which is used as the training data set and 5% of which is used as the test training set. The training data set is used to train the neural network model, and the test training set is used to evaluate the performance of the neural network model.

[0066] Neural network model: The constructed neural network model is a fully connected feedforward neural network, and the input is a 1024-dimensional vector after embedding the Fourier coefficients and their position encoding. Referring to Figure 6 , it is shown that Figure 6 is a structural diagram of the neural network model provided by the embodiment of the present application, and the neural network model comprises: First layer: fully connected layer (1024 neurons) + PReLU activation; Second layer: fully connected layer (512 neurons) + PReLU activation; Third layer: fully connected layer (256 neurons) + PReLU activation; Fourth layer: fully connected layer (64 neurons) + PReLU activation; Output layer: 1 neuron, Sigmoid activation function for surface emission efficiency to normalize the output range to (0, 1); identity mapping for the logarithmic value of the quality factor.

[0067] Specifically, for the region within the photonic crystal unit cell, a function z = f(x, y) is defined, where f is a binary function (which can be a complex function, the function can be a polynomial function, an exponential function, a logarithmic function, a Gaussian function, a Lorentz function, etc. and their linear or multiplication or convolution combinations), using the function generate_sample_array as f, which is a linear combination of several two-dimensional quasi-Gaussian functions, eps_sample as the sampling matrix, the grid is discretized into 320*320 points, and a composite function of n quasi-Gaussian functions is generated.

[0068] For the randomly generated filling factor between (0, 1), find the data closest to 100%*filling factor as the threshold eps_thresh. Set the dielectric constant array to GaAs_eps (the dielectric constant of GaAs) or 1.0 (the dielectric constant of air) according to the threshold. Here, it is set to GaAs_eps if it is less than the threshold.

[0069] Then average the super-sampled 320*320 size grid to a 32*32 size two-dimensional dielectric constant array as a discretized grid, as shown in Figure 7 Figure 7 is a structure diagram of the discretized grid provided by the embodiment of the present application. Discrete Fourier transform is performed on the discretized grid, and the conjugate term is removed. Then, for the 32*32 discretized grid, there are 481 items left after removing the conjugate term, and the real part + the imaginary part totals 962 items, which is the input of the neural network model. After 4 layers of non-linear activation and full connection, a performance index of a photonic crystal is output. Finally, the Sigmoid activation function is used for the surface emission efficiency to normalize the output range to (0, 1); the identity mapping is used for the logarithmic value of the quality factor to keep the output consistent with the data. The Sigmoid activation function is used for the surface emission efficiency normalization, and the identity mapping is used for the logarithmic value of the quality factor, i.e. directly outputting the linear combination result of the neurons without any non-linear transformation, which can keep the data consistent and does not perform any form of scaling or transformation, and directly outputs the logarithmic value of the quality factor. The prediction of the neural network model directly reflects the relationship between the input features and the output, can quickly predict the performance of unknown structures, realizes efficient prediction of the performance of photonic crystals, and improves the accuracy of the performance prediction result of photonic crystals.

[0070] ​It should be appreciated that the steps of the methods in accordance with the embodiments of the application can be implemented or performed by a computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable medium. The methods can use standard programming techniques. Each program can be implemented in a high level procedural or object oriented programming language to be executed by a computer system. However, if desired, the programs can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Moreover, the programs can be stored on or downloaded from computer program products, which can be in the form of tangible non-transitory computer readable storage media having stored computer instructions. Examples of computer readable storage media include magnetic disks, magnetic tape, or any other magnetic media, CD-ROM or DVD or any other optical media, punch cards or other punch medium, flash memory, or any other non-transitory memory that stores computer instructions.

[0071] Further, the operations of the processes described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and / or combinations thereof) can be performed under the control of one or more computer systems configured with executable instructions (e.g., computer programs, one or more computer programs or applications, executable instructions, etc.) and can be implemented as code (e.g., executable instructions, one or more computer programs or one or more applications) running on a computer system (e.g., distributed or local) and / or executed by hardware or combinations thereof. The computer programs include program instructions that can be executed by one or more processors.

[0072] Further, the methods can be implemented in any suitable type of computing platform that is operatively connected to any suitable type of computing platform, including but not limited to a personal computer, a mini-computer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the application can be implemented in machine readable code stored on a non-transitory storage medium or device, whether removable or integrated to the computing platform, such as a hard disk, an optical read and / or write storage medium, RAM, ROM, and the like, such that it can be read by a programmable computer to configure and operate the computer to perform the processes described herein when the storage medium or device is read by the computer. Further, the machine readable code, or portions thereof, can be transmitted over a wired or wireless network. The application described herein includes these and other different types of non-transitory computer readable storage media when such media include instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. The application can also include the computer itself when programmed in accordance with the methods and techniques described herein.

[0073] The computer programs can be applied to input data to perform the functions described herein, to transform the input data to generate output data that is stored to a non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In preferred embodiments of the application, the transformed data represents a physical and tangible object, including a particular visual depiction of a physical and tangible object produced on a display.

[0074] The above merely describes preferred embodiments of the present application, and the present application is not limited to the above-described embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application, as long as the same technical effects are achieved by the same means, shall be included in the protection scope of the present application. The technical solutions and / or embodiments within the protection scope of the present application can have various modifications and changes.

Claims

1. A method for predicting photonic crystal performance based on a photonic crystal surface emitting laser, characterized in that: include: Discretize the photonic crystal unit cell to obtain a discretized grid; Performing a discrete Fourier transform on the discretized grid to obtain Fourier coefficients at each level, deduplicating the Fourier coefficients at each level and decomposing them into real and imaginary parts, defining a position coding array, and obtaining a spliced ​​array of photonic crystal layers based on the position coding array and the real and imaginary parts of the Fourier coefficients at each level; generating an input data set based on the thickness and dielectric constant of each layer of the epitaxial structure of the photonic crystal and the discretized grid of the photonic crystal unit cell, and calculating the performance index parameters of the photonic crystal structure through semi-analytical coupled wave theory simulation using the input data set as input parameters; performing an optimization training process on parameters of a neural network model using the thickness of each layer of the epitaxial structure of the photonic crystal and the spliced ​​array of the photonic crystal layers as input and the performance index parameters of the photonic crystal structure as output to obtain a trained neural network model; The photonic crystal to be predicted is input into the trained neural network model to obtain a performance test result of the photonic crystal to be predicted.

2. The method for predicting photonic crystal performance based on photonic crystal surface emitting laser according to claim 1, characterized in that: The discretization of the photonic crystal unit cell to obtain a discretized grid includes: For the photonic crystal unit cell area, randomly generate n Two-dimensional Gaussian functions, each of which has an expression of: ； in, g i (x,y) is a two-dimensional Gaussian function, A i is the two-dimensional Gaussian function amplitude control term, x is the horizontal coordinate in the photonic crystal unit cell, y is the vertical coordinate in the photonic crystal unit cell, σ x 、 σ y is a two-dimensional Gaussian function in x direction, y The variance in direction, x 0,i 、 y 0,i is the peak position control item of the two-dimensional Gaussian function; The n Two-dimensional Gaussian functions are added together to form a composite function, the expression of the composite function is: ； in, G i (x,y) for n The sum of two-dimensional Gaussian functions, g i (x,y) is a two-dimensional Gaussian function; Setting thresholds T 0 , take the contour line G(x,y)= T 0 is the hole boundary; The hole outline of the hole boundary is pixelated into a 32×32 two-dimensional dielectric constant array as a discretized grid.

3. The method for predicting photonic crystal performance based on photonic crystal surface emitting laser according to claim 2, characterized in that: Also includes: Calculating the value of the two-dimensional dielectric constant of each point in the discretized grid; Using a randomly generated filling factor between (0, 1) as a reference, sorting the values ​​of the two-dimensional dielectric constant of each point in the discretized grid, and using the value of the two-dimensional dielectric constant with the percentile closest to the filling factor as a reference threshold; Determining the value of the two-dimensional dielectric constant of each point in the discretized grid and the size of the reference threshold; The grid points whose two-dimensional dielectric constants are greater than the reference threshold are all set as the dielectric constants of the first medium, and the grid points whose two-dimensional dielectric constants are less than the reference threshold are all set as the dielectric constants of the second medium; A discrete photonic crystal structure is formed according to the dielectric constant of each point in the discretized grid.

4. The method for predicting photonic crystal performance based on photonic crystal surface emitting laser according to claim 1, characterized in that: The discrete Fourier transform is performed on the discretized grid to obtain Fourier coefficients at each level, and the Fourier coefficients at each level are deduplicated and decomposed into real and imaginary parts, including: Performing a two-dimensional Fourier transform on the discretized grid to extract Fourier coefficients to obtain complex Fourier coefficients; Decomposing the complex Fourier coefficients to obtain real and imaginary parts, and forming network input features based on the real and imaginary parts; Complex conjugate redundancy removal is performed on the complex Fourier coefficients, and half of the complex Fourier coefficients are retained to avoid redundancy.

5. The method for predicting photonic crystal performance based on photonic crystal surface emitting laser according to claim 4, characterized in that: The defined position code array includes: Introducing learnable positional encoding for the Fourier coefficients at each level; The same embedding term is added to the real part and the imaginary part of each Fourier coefficient to obtain a position coding array. The expression of the position coding array is: ； Among them, Input is the position encoding array, ξ are the Fourier coefficients, PE is the position code, (m,n) Indicates the level subscript corresponding to the position code, R represents the real part, I Represents the imaginary part.

6. The method for predicting photonic crystal performance based on photonic crystal surface emitting laser according to claim 3, characterized in that: The performance index parameters of the photonic crystal structure calculated by semi-analytical coupled wave theory simulation include: Generating an input data set by using the thickness and dielectric constant of each layer of the epitaxial structure of the photonic crystal and the photonic crystal structure parameters formed by the discrete photonic crystal structure according to the dielectric constant of each point in the discretized grid, and using the input data set as input parameters; The surface emission efficiency, quality factor, radiation constant and threshold gain of the photonic crystal structure are obtained by simulation calculation using semi-analytical coupled wave theory and finite element method. Using the logarithm of the quality factor as a training target to narrow the value range and stabilize the simulation calculation process; A preset number of simulation data sets are constructed, and the simulation data sets are divided into a training data set and a test training set.

7. The method for predicting photonic crystal performance based on photonic crystal surface emitting laser according to claim 6, characterized in that: The neural network model includes four fully connected layers, the number of neurons in the four fully connected layers is 1024, 512, 256, and 64, respectively. Each fully connected layer is matched with a nonlinear activation function. The parameters of the neural network model are optimized and trained using the thickness of each layer of the epitaxial structure of the photonic crystal and the spliced ​​array of the photonic crystal layer as input and the performance index parameters of the photonic crystal structure as output, thereby obtaining a trained neural network model, including: Define the real part of the Fourier coefficients at each level as R and the imaginary part as I, define the position encoding arrays as PE1 and PE2, and PE1=PE2; Obtaining a spliced ​​array of a photonic crystal layer according to the position coding array and the real and imaginary parts of the Fourier coefficients at each level, wherein the expression of the spliced ​​array of the photonic crystal layer is [R+PE1, I+PE2]; Taking the thickness of each layer of the epitaxial structure of the photonic crystal and the spliced ​​array of the photonic crystal layers [R+PE1, I+PE2] as input, and the surface emission efficiency, quality factor, radiation constant and threshold gain of the photonic crystal structure as output, the thickness of each layer of the epitaxial structure of the photonic crystal and the spliced ​​array of the photonic crystal layers are optimized and trained in sequence through 4 fully connected layers and 4 nonlinear activation functions to obtain a trained neural network model.

8. The method for predicting photonic crystal performance based on photonic crystal surface emitting laser according to claim 7, characterized in that: The neural network model further includes an output layer, and the output layer includes one neuron. The method further includes: A Sigmoid activation function is used on the surface emission efficiency to normalize the output range to (0, 1); The logarithmic value of the quality factor is processed by identity mapping so that the output maintains data consistency.

9. A computer device comprising a memory and a processor, characterized in that: The method according to any one of claims 1 to 8 is implemented when the processor executes the computer program stored in the memory.

10. A computer-readable storage medium having program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images