Calculation Method Applied to Intelligent Optical Computing Reverse Design Model, Intelligent Optical Computing Reverse Design Method and System

Abstract

The present disclosure relates to the field of optical computing technologies, and in particular, to a calculation method, an intelligent optical computing reverse design method, and a system applied to an intelligent optical computing reverse design model. The method includes: controlling an input division module to divide an input electromagnetic field into a set of input electromagnetic field segments based on the Huygens-Fresnel law; controlling a wavefront prediction module to input each input electromagnetic field segment in the set of input electromagnetic field segments and the structural parameters corresponding to the intelligent optical computing device to be designed into an electromagnetic neural local solver corresponding to the input electromagnetic field segment to obtain a set of output electromagnetic field segments; controlling an output splicing module to splice all the output electromagnetic field segments in the set of output electromagnetic field segments based on the Huygens-Fresnel law to obtain an output electromagnetic field under a global response, so as to adjust the structural parameters according to the output electromagnetic field. The present disclosure adopting the above solution can achieve the reverse design of an intelligent optical computing device.

Description

Technical Field

[0001] The present disclosure relates to the field of optical computing technologies, and particularly to a computing method, an intelligent optical computing reverse design method, and a system applied to an intelligent optical computing reverse design model. Background Art

[0002] With the rapid development of the fields of artificial intelligence and scientific computing, the complexity and scale of computing requirements are also constantly increasing. However, existing electronic computing technologies are limited by Moore's Law, and their performance is gradually approaching saturation, making it difficult to effectively meet the increasingly stringent requirements for computing power and power consumption of large-scale complex algorithms. Light has natural advantages such as high throughput and low latency during propagation, and optical computing technology using photons instead of electrons as the computing carrier is regarded as the key to breaking the existing computing bottleneck. Summary of the Invention

[0003] The present disclosure aims to at least partly solve one of the technical problems in the related art.

[0004] To this end, the first object of the present disclosure is to propose an intelligent optical computing reverse design model to perform reverse design for an intelligent optical computing device and achieve a balance between design efficiency and accuracy.

[0005] The second object of the present disclosure is to propose an intelligent optical computing reverse design method.

[0006] The third object of the present disclosure is to propose an intelligent optical computing reverse design architecture.

[0007] To achieve the above object, an embodiment of the first aspect of the present disclosure proposes an intelligent optical computing reverse design model, including:

[0008] An input partitioning module, configured to partition the input electromagnetic field into a set of input electromagnetic field segments based on the Huygens - Fresnel law according to the total wavefront width of the input electromagnetic field and the wavefront width corresponding to each electromagnetic neural local solver in the electromagnetic neural local solver set;

[0009] A wavefront prediction module, configured to input each input electromagnetic field segment in the set of input electromagnetic field segments and the structural parameters corresponding to the intelligent optical computing device to be designed into the electromagnetic neural local solver corresponding to the input electromagnetic field segment, so that the electromagnetic neural local solver predicts an output electromagnetic field segment according to the input electromagnetic field segment and the structural parameters, and obtains a set of output electromagnetic field segments;

[0010] An output splicing module, configured to splice all the output electromagnetic field segments in the set of output electromagnetic field segments based on the Huygens - Fresnel law to obtain an output electromagnetic field under the global response, so as to adjust the structural parameters according to the output electromagnetic field.

[0011] Optionally, when the input partitioning module is used to partition the input electromagnetic field into a set of input electromagnetic field segments according to the total wavefront width of the input electromagnetic field and the wavefront width corresponding to each electromagnetic neural local solver in the electromagnetic neural local solver set, it is specifically used for:

[0012] Partition the total wavefront width of the input electromagnetic field based on the wavefront width corresponding to each electromagnetic neural local solver in the electromagnetic neural local solver set to obtain a partitioning result;

[0013] If the partitioning result indicates that there is an input electromagnetic field segment in the input electromagnetic field with a width less than the wavefront width, zero padding is added to the end side and / or the end of the input electromagnetic field until the width of each input electromagnetic field segment is the wavefront width, obtaining a set of input electromagnetic field segments, where the input electromagnetic field segment is regarded as a secondary wave generated by a central source point on the input electromagnetic field segment.

[0014] Optionally, when the output splicing module is used to splice all the output electromagnetic field segments in the output electromagnetic field segment set, it is specifically used for:

[0015] Coherently superpose all the output electromagnetic field segments in the output electromagnetic field segment set, where the output electromagnetic field segment is regarded as an envelope surface generated by a secondary wave source on the output electromagnetic field segment.

[0016] Optionally, when the electromagnetic neural local solver is used to predict an output electromagnetic field segment according to the input electromagnetic field segment and the structural parameters, it is specifically used for:

[0017] Perform a complex linear connection on the input electromagnetic field segment to represent the non-physical information propagation process of the input electromagnetic field segment in the to-be-designed intelligent optical computing device, obtaining the input electromagnetic field segment after non-physical propagation;

[0018] Perform a real non-linear connection on the structural parameters, and align the information of the connected structural parameters with the input electromagnetic field segment after non-physical propagation to represent the physical information propagation process of the input electromagnetic field segment after propagation in the to-be-designed intelligent optical computing device, obtaining the input electromagnetic field segment after physical propagation;

[0019] Perform a complex linear connection on the input electromagnetic field segment after physical propagation to represent the non-physical information propagation process of the input electromagnetic field segment after physical propagation in the to-be-designed intelligent optical computing device, obtaining the output electromagnetic field segment.

[0020] Optionally, when the electromagnetic neural local solver is used to perform a complex linear connection, it is specifically used for:

[0021] Obtain the first weight matrix and the first intercept term used for complex linear connection, and perform complex linear connection based on the first weight matrix and the first intercept term, where the first weight matrix and the first intercept term are obtained by training an initial electromagnetic neural local solver.

[0022] Optionally, when the electromagnetic neural local solver is used for real - number non - linear connection, it is specifically used for:

[0023] Obtain the second weight matrix, the second intercept term, and the non - linear operation function used for real - number non - linear connection, and perform complex linear connection based on the second weight matrix, the second intercept term, and the non - linear operation function, where the second weight matrix, the second intercept term, and the non - linear operation function are obtained by training an initial electromagnetic neural local solver.

[0024] Optionally, before inputting each input electromagnetic field segment in the input electromagnetic field segment set and the structure parameters corresponding to the intelligent optical computing device to be designed into the electromagnetic neural local solver corresponding to the input electromagnetic field segment, the wavefront prediction module is further used for:

[0025] Obtain a training sample set, where each training sample in the training sample set includes an input electromagnetic field training sample and a structure parameter training sample;

[0026] Perform gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set until an electromagnetic neural local solver that meets the training requirements is obtained;

[0027] Obtain a test sample set, where each test sample in the test sample set includes an input electromagnetic field test sample and a structure parameter test sample;

[0028] Test the electromagnetic neural local solver based on the test sample set. If the test result does not meet the test requirements, perform gradient backpropagation training on the electromagnetic neural local solver again until an electromagnetic neural local solver with a test result meeting the test requirements is obtained.

[0029] Optionally, when the wavefront prediction module is used to perform gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set, it is specifically used for:

[0030] In the process of performing gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set, any training sample in the training sample set is input into the initial electromagnetic neural local solver to obtain a predicted value corresponding to the any training sample, and a first error value between the predicted value and the true value corresponding to the any training sample is determined. The network parameters of the initial electromagnetic neural local solver are updated based on the first error value to minimize the loss function.

[0031] Optionally, before the wavefront prediction module is used to determine the first error value between the predicted value and the true value corresponding to any training sample, it is further used to:

[0032] Input the any training sample into a numerical calculation solver to obtain the true value corresponding to the any training sample.

[0033] To achieve the above object, an embodiment of the second aspect of the present disclosure proposes an intelligent optical computing reverse design method, including:

[0034] Obtain a sample data set, where the sample data set includes input electromagnetic field samples;

[0035] Based on the sample data set and the intelligent optical computing reverse design model shown in any one of the foregoing first aspects, train the structural parameters of the intelligent optical computing device to be designed to obtain target structural parameters that meet the training requirements;

[0036] Construct an intelligent optical computing device corresponding to the target structural parameters.

[0037] Optionally, the training of the structural parameters of the intelligent optical computing device to be designed to obtain target structural parameters that meet the training requirements includes:

[0038] Input the structural parameters of the intelligent optical computing device to be designed and any input electromagnetic field sample in the sample data set into the intelligent optical computing reverse design model to obtain a predicted output electromagnetic field;

[0039] Input the structural parameters and the any input electromagnetic field sample into a numerical calculation solver to obtain the true output electromagnetic field corresponding to the any input electromagnetic field sample;

[0040] Determine a second error value between the predicted output electromagnetic field and the true output electromagnetic field. If the second error value is greater than the error threshold, update the structural parameters until the second error value is not greater than the error threshold to obtain target structural parameters that meet the training requirements.

[0041] To achieve the above object, an embodiment of the third aspect of the present disclosure proposes an intelligent optical computing reverse design architecture, including:

[0042] A sample acquisition unit for acquiring a sample data set, where the sample data set includes input electromagnetic field samples;

[0043] A parameter training unit for training the structural parameters of the intelligent optical computing device to be designed based on the sample data set and the intelligent optical computing inverse design model shown in any one of the foregoing first aspects, so as to obtain target structural parameters that meet the training requirements;

[0044] A device construction unit for constructing an intelligent optical computing device corresponding to the target structural parameters.

[0045] In summary, the intelligent optical computing inverse design model, method and architecture provided by the present disclosure can directly, quickly and accurately predict the full-wave electromagnetic field distribution based on an arbitrarily varying input field and a sub-wavelength structure of any size by using an electromagnetic neural local solver as a local electromagnetic field solver and connecting the locally predicted output electromagnetic fields by using the Huygens-Fresnel global method. Therefore, the inverse design of the structural parameters of the intelligent optical computing device to be designed can be realized based on the predicted output electromagnetic field.

[0046] Some of the additional aspects and advantages of the present disclosure will be given in the following description, some will become obvious from the following description, or will be understood through the practice of the present disclosure. Description of the Drawings

[0047] The above and / or additional aspects and advantages of the present disclosure will become obvious and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0048] Figure 1 is a schematic structural diagram of an intelligent optical computing inverse design model provided by an embodiment of the present disclosure;

[0049] Figure 2 is a design concept diagram of an intelligent optical computing inverse design model provided by an embodiment of the present disclosure;

[0050] Figure 3 is a design concept diagram of a local solver for an electromagnetic field fast simulation operator provided by an embodiment of the present disclosure;

[0051] Figure 4 is a schematic flow diagram of an intelligent optical computing inverse design method provided by an embodiment of the present disclosure;

[0052] Figure 5 is a schematic structural diagram of an intelligent optical computing inverse design architecture provided by an embodiment of the present disclosure. Detailed Embodiments

[0053] Embodiments of the present disclosure will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals designate like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present disclosure and should not be construed as limiting the present disclosure.

[0054] Optical computing hardware is expected to drive the development of artificial intelligence in the future due to its advantages over electronic implementations in terms of low latency, minimum power consumption, high parallelism, and high speed, and as a form of analog computing eliminating the need for digital-to-analog conversion. Currently, optical neural networks (Optical neural network, analog-domain optical intelligent computing) are mainly implemented by means such as interference based on Mach–Zehnder interferometers (MZIs), diffraction using training modulation units, and wavelength division multiplexing implemented by microrings. However, due to their large footprint, the integration density and scale of these optical neural networks are significantly limited.

[0055] Artificially designed subwavelength structures, known as meta-structures, may significantly improve the integration level due to their subwavelength superatomic size. The interaction between light and subwavelength structures enables the modulation of the phase and amplitude of the light wavefront, which has a wide range of applications in fields such as focusing, imaging, vortex generation, and spectroscopy. Compared with traditional spatial light diffraction plates, subwavelength structures have a more compact size, possess the ability of high integration, and can achieve on-chip all-optical diffraction neural networks through the cascading of multiple levels of subwavelength structures. The diffractive optical computing super-system uses multi-layer subwavelength structures to flexibly modulate the wavefront and is expected to become the next-generation computing hardware platform with its compact volume, high computing density, and low power consumption.

[0056] Early subwavelength structure designs mainly relied on intuitive manual parameter adjustment or prior knowledge using analytical models, which led to a large number of trial-and-error processes, making the design process inefficient and arduous. In recent years, inverse design has provided a new paradigm for device design by adjusting structural parameters to optimize design objectives to achieve the desired optical response, greatly improving design efficiency and device performance. Optimization algorithms mainly include heuristic algorithms, gradient optimization, etc. These methods typically use iterative numerical simulation algorithms such as the finite element method (FEM), Finite Difference Time Domain (numerical calculation method), and rigorous coupled-wave analysis (RCWA) to obtain the forward evaluation of the device. These methods solve electromagnetic field constraints and Maxwell's equations by replacing differentiation with differences, providing accurate and reliable results, but they have a high computational cost in terms of memory usage and computational time. The inefficiency of numerical methods may lead to a time-consuming and costly inverse design process because the forward evaluation is continuously repeated in step-by-step iterations. To alleviate the computational burden, neural networks are used as substitutes for numerical simulation methods to accelerate inverse design. In recent years, various deep learning models have been combined with inverse design to achieve better performance and higher efficiency, including discriminative architectures such as fully connected neural networks (fuzzy clustering neural network, FCNN), Convolutional Neural Networks (CNN), and Recurrent Neural Network (RNN), as well as generative models such as Variational Autoencoder (VAE) and Generative Adversarial Networks (GAN). These models have complex architectures and a large number of connections and weights, showing strong learning capabilities, but at the same time consuming huge computational resources. Essentially, these methods are not specifically designed to handle the interaction between electromagnetic waves and subwavelength structures, resulting in model redundancy and waste of computational resources. In addition, there are other specific limitations in previous designs using deep learning. These works only focus on indirect properties such as spectra or transmittance, rather than the light field itself. Moreover, these methods lack flexibility because they ignore variable input fields and limit the device size to a fixed value. In addition, in large-scale design, the design space has an extremely high degree of freedom (Degrees of freedom, DOF), which requires an extremely large number of simulations and optimizations. With the increase in device size and the complexity of functions, all these problems pose a considerable challenge to inverse design.

[0057] For example, when light propagates in the fundamental slab mode in the Transverse Electric (TE) wave mode, the on-chip propagation process is modeled by two parts: the slab mode propagation model and the optical phase modulation model. Among them, the slab mode propagation model has been discussed in detail and well-verified in many previous studies. Its calculation results are almost exactly the same as those of the numerical calculation method solver. The slab mode propagation it adopts is derived based on the Fourier optics method and can be expressed as:

[0058]

[0059] Among them, represents the electric field after optical phase modulation, and are the discrete Fourier transform and the inverse discrete Fourier transform respectively, while is the filtering matrix:

[0060] In the actual calculation of the computer, the electric field is a string of discrete numerical values uniformly distributed in space z and can be expressed as:

[0061]

[0062] Performing the discrete Fourier transform on these data can obtain the magnetic field spatial frequency components of the magnetic field along the z direction:

[0063]

[0064] The value of the filtering matrix can be obtained by simulating through numerical calculation methods, specifically through the method of looking up a table (Look-up Table):

[0065]

[0066] The final result of calculating the optically phase-modulated electric field is:

[0067]

[0068] Previously, the physical modeling error of the on-chip optical network was mainly attributed to the optical phase modulation model. Currently, optical phase modulation can be modeled through an analytical model or numerical simulation. The analytical model is derived based on the simplification of the physical process, such as the effective refractive index method and the Fourier optics method. Although these methods can quickly calculate the output field distribution, they cannot guarantee the accuracy of the modeling. While numerical simulation can ensure the accurate modeling of the designed device, for large-scale, multi-level cascaded, and high-degree-of-freedom computational supersystems, the device optimization time and computational cost will be extremely high.

[0069] In addition, different from the design of metasurfaces or some single-layer subwavelength structures, the cascading of multi-layer structures leads to the accumulation of physical modeling errors, which has a significant impact on the effectiveness of simulation computing systems. The accuracy of existing advanced analytical models cannot meet the requirements of large-scale computing supersystems. Therefore, when devices designed using analytical methods are deployed in actual physical experiments, their performance drops significantly, which requires additional complex calibration to improve performance. The lack of reliable and efficient physical modeling tools hinders the development of large-scale optical computing in performing complex tasks. Currently, there is no tool or model that can reliably and efficiently address such problems.

[0070] That is to say, the on-chip computing supersystem composed of multi-layer subwavelength structures is expected to become the next-generation computing hardware, with the ability of light-speed processing and low-power consumption characteristics. However, the current design paradigm hinders its development. So far, neither numerical methods nor analytical analysis methods can achieve a balance between design efficiency and accuracy.

[0071] The present disclosure will be described in detail below with reference to specific embodiments.

[0072] Figure 1 The following is a schematic structural diagram of an intelligent optical computing inverse design model provided by an embodiment of the present disclosure. As Figure 1 shown, the intelligent optical computing inverse design model includes:

[0073] An input partitioning module, configured to partition the input electromagnetic field into a set of input electromagnetic field segments based on the Huygens-Fresnel law, according to the total wavefront width of the input electromagnetic field and the wavefront width corresponding to each electromagnetic neural local solver in the electromagnetic neural local solver set.

[0074] A wavefront prediction module, configured to input each input electromagnetic field segment in the set of input electromagnetic field segments and the structural parameters corresponding to the intelligent optical computing device to be designed into the electromagnetic neural local solver corresponding to the input electromagnetic field segment, so that the electromagnetic neural local solver predicts an output electromagnetic field segment according to the input electromagnetic field segment and the structural parameters, and obtains a set of output electromagnetic field segments.

[0075] An output splicing module, configured to splice all the output electromagnetic field segments in the set of output electromagnetic field segments based on the Huygens-Fresnel law to obtain an output electromagnetic field under the global response, so as to adjust the structural parameters according to the output electromagnetic field.

[0076] According to some embodiments, the electromagnetic neural local solver refers to the local solver in electromagnetic neural networks (Electromagnetic neural networks, fast electromagnetic field simulation operator), that is, the local solver of the fast electromagnetic field simulation operator. The local solver of the fast electromagnetic field simulation operator is a neural network with a physically inspired architecture that can predict the output electromagnetic field based on the given input electromagnetic field and structural parameters. The local solver of the fast electromagnetic field simulation operator is a key building block for the intelligent optical computing inverse design model to effectively predict the input electromagnetic field. It first considers the variable input field in the inverse design network. Through physically inspired modifications, it can successfully perform full-wave predictions accurately and efficiently, accurately predict the electromagnetic field distribution according to arbitrarily varying input fields, and exhibit good interpretability and generalization ability.

[0077] In some embodiments, the structural parameters refer to the structural parameters of the sub-wavelength structure used by the intelligent optical computing device to be inversely designed when modulating the input light. The intelligent optical computing device can be a device corresponding to the computing super-system. The computing super-system includes a linear system and a nonlinear system.

[0078] In some embodiments, the wavefront width corresponding to the local solver of the fast electromagnetic field simulation operator can be adjusted according to the actual application scenario. For example, the wavefront width corresponding to the local solver of the fast electromagnetic field simulation operator can be 0.6 micrometers.

[0079] It should be noted that when the total wavefront width of the input electromagnetic field is significantly larger than the wavefront width corresponding to the local solver of the fast electromagnetic field simulation operator. For example, in a subsequent task to be carried out, the total wavefront width of the designed input electromagnetic field is 100 micrometers, and the wavefront width that the local solver of the fast electromagnetic field simulation operator can handle is 0.6 micrometers. In this case, the total wavefront width of the input electromagnetic field far exceeds the wavefront width that a single local solver of the fast electromagnetic field simulation operator can handle, with a difference of three orders of magnitude. If a single neural network is directly applied to a 100-micrometer-long device, many problems will occur. First, the complexity of the problem increases sharply. As the application area grows linearly, the cost of training the network, including generating the dataset, increases multiplicatively. In other words, the computational cost required to train a single forward prediction network for a 100-micrometer-wide device is at least several orders of magnitude larger than that of training a network for a 0.6-micrometer-wide device. In addition, even if a great deal of effort is made to train this single neural network, the network lacks flexibility and generality. That is, once the device size changes, a new network will have to be trained from scratch.

[0080] Therefore, inspired by the principle of optical diffraction propagation, a global method with physical information is proposed to seamlessly connect multiple local solvers of electromagnetic field fast simulation operators. The global method with physical information is a strategy to extend the forward model to arbitrarily large sizes. It is a physically inspired strategy, inspired by the Huygens-Fresnel law, which can successfully piece together the local responses predicted by the local solvers of electromagnetic field fast simulation operators physically to form a global response, so as to obtain an arbitrarily large electromagnetic field distribution. This physically inspired strategy enables the model to be flexibly applied to devices with non-fixed sizes, enabling the model to handle inverse design problems of any scale without increasing the complexity of the problem or reducing the design performance. More importantly, this method aligns the process of wavefront stitching with the real physical process, significantly reducing the stitching error compared with the previous non-physically inspired methods. In addition, in the previous methods, the network structure lacked scalability, restricting the ability of the algorithm to adjust parameters within a fixed-size region. However, with the introduction of the global method with physical information, the network demonstrates strong scalability. Taking the inverse design of an optical neural network as an example, the intelligent optical computing inverse design model can design an optical intelligent computing with an arbitrarily wide and deep simulation domain.

[0081] It should be noted that the intelligent optical computing inverse design model provided by the embodiments of the present disclosure, by using the local solver of the electromagnetic field fast simulation operator as the local electromagnetic field solver and connecting the predicted local output electromagnetic fields by using the global method with physical information, can directly, quickly, and accurately predict the full-wave electromagnetic field distribution based on arbitrarily varying input fields and sub-wavelength structures of arbitrary sizes, so as to realize the inverse design of the structural parameters of the intelligent optical computing device to be designed based on the predicted output electromagnetic field. More importantly, the previous deep learning frameworks predicted indirect physical quantities, such as transmittance or spectrum, to meet the requirements of specific tasks. In contrast, the intelligent optical computing inverse design model can directly predict the electromagnetic field components. If transmittance is required, it can be calculated from the electromagnetic field components. The scalability and flexibility of the intelligent optical computing inverse design model meet unprecedented inverse design requirements and can handle any scale, high degrees of freedom, and complex functions, which exceed the capabilities of other deep learning-based methods.

[0082] Optionally, Figure 2 is a design concept diagram of an intelligent optical computing inverse design model provided by the embodiments of the present disclosure. As Figure 2 shown, the device is optimized through iterative inference and adjustment to generate the desired electromagnetic response. The large scale, high design degrees of freedom, and complex functions of the device, as well as the challenging forward modeling, are the main obstacles faced by inverse design. The proposed model aims to solve these problems.

[0083] It should be noted that in the global method with physical information, according to the Huygens - Fresnel law, during the propagation of the light field, each point on the wavefront can be regarded as a central source, generating secondary waves; subsequently, these secondary waves are coherently superposed to form a new wavefront. That is to say, as Figure 2 shown,

[0084] The input partitioning module can partition the total wavefront width of the input electromagnetic field based on the wavefront width corresponding to each electromagnetic neural local solver in the set of electromagnetic neural local solvers, obtaining a partitioning result; if the partitioning result indicates that there is an input electromagnetic field segment in the input electromagnetic field with a width less than the wavefront width, zero padding is added at the end side and / or the terminal side of the input electromagnetic field until the width of each input electromagnetic field segment is the wavefront width, obtaining a set of input electromagnetic field segments, where the input electromagnetic field segment is regarded as a secondary wave generated by the central source point on the input electromagnetic field segment.

[0085] The output splicing module can perform coherent superposition on all the output electromagnetic field segments in the set of output electromagnetic field segments, where the output electromagnetic field segment is regarded as an envelope surface generated by the secondary wave source on the output electromagnetic field segment.

[0086] For example, first, a wavefront with a width of 100 microns can be divided into multiple segments with a width of 0.6 microns, and each segment can be regarded as an independent point on the wavefront, generating secondary waves. Then, each 0.6 - micron wavefront segment and its corresponding structural parameter with a width of 3.0 microns are input into the electromagnetic field fast - simulation operator local solver, which will generate output electromagnetic field segments with a width of 3.0 microns, and these output electromagnetic field segments represent the envelope surface generated by the secondary wave source. Finally, the predicted output electromagnetic field segments are coherently superposed to obtain the final wavefront modulation result of the 100 - micron - wide diffractive sub - wavelength structure on the input field.

[0087] According to some embodiments, Figure 3 is a design concept diagram of an electromagnetic field fast - simulation operator local solver provided by an embodiment of the present disclosure. As shown in Figure 3 , it shows the interpretable correspondence between the electromagnetic field fast - simulation operator local solver architecture and the light propagation process. The physical process of wavefront modulation can be divided into three stages, where the first stage and the third stage are non - physical information propagation stages, and the second stage is the physical information propagation stage. When the electromagnetic neural local solver is used to predict the output electromagnetic field segment according to the input electromagnetic field segment and the structural parameter, it is specifically used for:

[0088] Performing a complex linear connection on the input electromagnetic field segment to represent the non - physical information propagation process of the input electromagnetic field segment in the intelligent optical computing device to be designed, obtaining the input electromagnetic field segment after non - physical propagation;

[0089] Perform a real - valued non - linear connection on the structural parameters, and align the information of the connected structural parameters with the input electromagnetic field segment after non - physical propagation to represent the physical information propagation process of the input electromagnetic field segment in the to - be - designed intelligent optical computing device, obtaining the input electromagnetic field segment after physical propagation;

[0090] Perform a complex - valued linear connection on the input electromagnetic field segment after physical propagation to represent the non - physical information propagation process of the input electromagnetic field segment after physical propagation in the to - be - designed intelligent optical computing device, obtaining the output electromagnetic field segment.

[0091] According to some embodiments, the ways of information alignment include, but are not limited to, element - wise multiplication operations, concatenation, outer product, attention mechanisms, and other mathematical operation methods. Among them, the element - wise multiplication operation corresponds to the physical process of optical phase modulation, and can be replaced by other mathematical operations according to the specific physical process required to be designed in the to - be - designed intelligent optical computing device; concatenation is to connect two vectors along a certain dimension, retaining the original electromagnetic field information; the outer product refers to generating an outer - product matrix of two vectors, which is used to capture the second - order interaction between features; the attention mechanism uses the Transformer model to align different modal information. In this embodiment, the attention mechanism can be used to align the electromagnetic field information and the sub - wavelength physical structure information.

[0092] For example, x , y can be used to represent the information alignment operation on two complex - valued signals. In this embodiment, x and y are respectively the connected structural parameters and the input electromagnetic field segment after non - physical propagation.

[0093] In some embodiments, the non - physical information propagation process, for example, refers to the propagation process of the input signal in the slab mode outside the sub - wavelength structure in the to - be - designed intelligent optical computing device (although the distance is short, it does exist).

[0094] In some embodiments, the physical information propagation process, for example, refers to the optical phase modulation process of the input signal passing through the sub - wavelength structure in the to - be - designed intelligent optical computing device.

[0095] According to some embodiments, in electromagnetic field signal processing (such as communication, radar, electromagnetic compatibility analysis, etc.), electromagnetic field data is usually represented in complex form (including amplitude and phase information). Complex - valued linear connection is a key operation that maps complex features to a unified semantic space, and its core is to perform information processing through linear transformation in the complex domain.

[0096] In some embodiments, when an electromagnetic neural local solver is used for complex linear connection, a first weight matrix and a first intercept term used for complex linear connection can be obtained, and complex linear connection is performed based on the first weight matrix and the first intercept term.

[0097] For example, if z 1 is the output complex signal, is the first weight matrix, x 1 is the input signal for complex linear connection, 1 is the first intercept term, then the complex linear connection can generally be expressed as:

[0098]

[0099] According to some embodiments, when performing real-number nonlinear connection on structural parameters, complex interactions between parameters and high-dimensional complex information can be captured through nonlinear transformation modeling.

[0100] In some embodiments, when an electromagnetic neural local solver is used for real-number nonlinear connection, a second weight matrix, a second intercept term, and a nonlinear operation function used for real-number nonlinear connection can be obtained, and complex linear connection is performed based on the second weight matrix, the second intercept term, and the nonlinear operation function.

[0101] For example, if 2 is the output nonlinear signal, 2 is the input signal for real-number nonlinear connection, is the second weight matrix, 2 is the second intercept term, is the nonlinear operation function, then the real-number nonlinear connection can generally be expressed as:

[0102]

[0103] It should be noted that, different from the previous practice of ignoring the input waveform, the electromagnetic field fast simulation operator local solver takes into account the changes in the input electromagnetic field. Based on a dual-input architecture design, this electromagnetic field fast simulation operator local solver can predict the output electromagnetic field results according to both the input electromagnetic field and the structural parameters. In addition, it is worth noting that generally, neural networks are regarded as black boxes lacking interpretability. However, in this electromagnetic field fast simulation operator local solver, the input electromagnetic field and optical phase modulation are encoded in complex form instead of separately processing the real and imaginary parts. Different parts of the network are connected in different ways. The input electromagnetic field and the output electromagnetic field are linearly connected by complex weights, while the input structural parameters are nonlinearly connected by real weights. This physically inspired neural network is carefully designed to be physically interpretable and has a convincing explanation for the physical meaning contained in this network architecture, and can demonstrate strong physical interpretability.

[0104] According to some embodiments, this electromagnetic field fast simulation operator local solver can be obtained by pre-training an initial electromagnetic field fast simulation operator local solver. That is to say, the wavefront prediction module can also be used for:

[0105] Obtain a training sample set, where each training sample in the training sample set includes an input electromagnetic field training sample and a structural parameter training sample;

[0106] Perform gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set until an electromagnetic neural local solver that meets the training requirements is obtained;

[0107] Obtain a test sample set, where each test sample in the test sample includes an input electromagnetic field test sample and a structural parameter test sample;

[0108] Test the electromagnetic neural local solver based on the test sample set. If the test result does not meet the test requirements, re-perform gradient backpropagation training on the electromagnetic neural local solver until an electromagnetic neural local solver with a test result that meets the test requirements is obtained.

[0109] In some embodiments, as Figure 3 shown in, when the wavefront prediction module is used to perform gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set, it is specifically used for:

[0110] During the process of performing gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set, input any training sample in the training sample set into the initial electromagnetic neural local solver to obtain a predicted value corresponding to any training sample, and determine a first error value between the predicted value and the true value corresponding to any training sample. Update the network parameters of the initial electromagnetic neural local solver based on the first error value to minimize the loss function.

[0111] Among them, by inputting any training sample into a numerical calculation solver, the true value corresponding to any training sample can be obtained. The numerical calculation solver can be, for example, a finite-difference time-domain method solver.

[0112] It should be noted that when performing gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set, the training parameters include but are not limited to the parameters used in the above-mentioned complex linear connection and real-number non-linear connection. That is to say, , , , 1, 2 can be obtained by training the initial electromagnetic neural local solver.

[0113] Taking a scenario as an example, in the case where a single electromagnetic field fast simulation operator local solver can handle the wavefront modulation problem within a finite region with a width of 900 nanometers, after the input-output pairs are randomly parameterized, they can be simulated and generated using numerical calculation methods. A total of 90,000 input-output pairs are used to form the training sample set and test sample set of the network, among which 80,000 are used as the training sample set and 10,000 are used as the test sample set. The shown training learning curve indicates that the test error converges after more than 70 training cycles.

[0114] It is worth noting that although the simulation of the above network dataset is quite large and requires a large amount of computing resources, however, compared with those methods that need to repeatedly access numerical calculation method solvers with high computational costs and time-consuming, this simulation work can be regarded as a one-time investment. In other words, the electromagnetic field fast simulation operator local solver can be used as a fast-response alternative to inefficient numerical simulations.

[0115] According to some embodiments, the intelligent optical computing device that needs to be reverse-designed includes a single-layer metal wire composed of meta-atoms, and the corresponding structural parameters of these single-layer metal wires are the structural parameters corresponding to the intelligent optical computing device. Among them, the thickness of the silicon film is 220 nanometers, the width of the meta-atom is 100 nanometers, and the period of the diffractive metal wire is 200 nanometers. The amplitude and phase modulation of the optical field of the intelligent optical computing device can be controlled by changing the gap length.

[0116] In some embodiments, thanks to the physically inspired modification, the electromagnetic field fast simulation operator local solver also exhibits excellent generalization ability. Under the conditions of the same network depth, comparable number of parameters, and the same training sample set, the electromagnetic field fast simulation operator local solver has stronger learning ability than non-physically inspired networks. As Figure 3As shown, it shows the wavefront profiles in the y-z plane of the output electromagnetic fields calculated by the local solver of the electromagnetic field fast simulation operator, the numerical calculation method solver, and the traditional non-physical heuristic method on the same test sample, which is generated in a different way from any training sample in the training sample set. Among them, the local solver of the electromagnetic field fast simulation operator successfully predicted results consistent with the simulation results of the numerical calculation method, while the non-physical heuristic method failed to achieve the same accuracy. The physically informed local solver of the electromagnetic field fast simulation operator performs better in both the prediction accuracy of amplitude and phase because the local solver of the electromagnetic field fast simulation operator essentially satisfies the linear response constraints of the physical system.

[0117] In summary, the intelligent optical computing inverse design model provided by the embodiments of the present disclosure integrates deep learning technology with basic physical principles, incorporates physical heuristic corrections into deep learning to obtain strong physical interpretability and generalization ability, achieves better performance, can exhibit excellent interpretability and generalization ability beyond traditional networks, can realize the reliable, efficient and flexible inverse design of intelligent optical computing devices with large scale, high degrees of freedom and complex functions, and achieves a balance between the accuracy and efficiency of the design. In addition, considering that the electromagnetic field itself is a composite field with amplitude and phase, the electromagnetic field information and optical phase modulation information are encoded as complex tensors; furthermore, considering that light propagation can be regarded as a linear process of Fourier filtering, a network structure with complex-valued linear connections is introduced into the local solver of the electromagnetic field fast simulation operator. These physically inspired modifications can help the local solver of the electromagnetic field fast simulation operator extract the underlying physical information hidden in the data, enabling the local solver of the electromagnetic field fast simulation operator to essentially satisfy some physical constraints of the system, which endows the local solver of the electromagnetic field fast simulation operator with generalization ability beyond traditional network architectures and ensures high-precision prediction ability.

[0118] The embodiments of the present disclosure also propose an intelligent optical computing inverse design method. This method can be applied to the intelligent optical computing inverse design architecture.

[0119] Exemplarily, as Figure 4 shown, the intelligent optical computing inverse design method includes the following steps:

[0120] S101, obtaining a sample data set;

[0121] According to some embodiments, the sample data set refers to a set composed of at least one input electromagnetic field sample.

[0122] In some embodiments, the sample data set can be obtained by preprocessing the original sample data in the original sample data set into wavefront data.

[0123] S102. Based on the sample data set and the intelligent optical computing inverse design model, train the structural parameters of the intelligent optical computing device to be designed to obtain the target structural parameters that meet the training requirements.

[0124] According to some embodiments, the intelligent optical computing inverse design model is the model provided in the foregoing embodiments, and the explanations of the embodiments of the intelligent optical computing inverse design model also apply to the intelligent optical computing inverse design model of this embodiment, which will not be elaborated here.

[0125] In some embodiments, during the process of training the structural parameters of the intelligent optical computing device to be designed, the structural parameters of the intelligent optical computing device to be designed and any input electromagnetic field sample in the sample data set can be input into the intelligent optical computing inverse design model to obtain a predicted output electromagnetic field; the structural parameters and any input electromagnetic field sample are input into the numerical calculation solver to obtain the true output electromagnetic field corresponding to any input electromagnetic field sample; determine the second error value between the predicted output electromagnetic field and the true output electromagnetic field. If the second error value is greater than the error threshold, update the structural parameters until the second error value is not greater than the error threshold to obtain the target structural parameters that meet the training requirements.

[0126] S103. Construct an intelligent optical computing device corresponding to the target structural parameters.

[0127] It should be noted that currently, due to the simple architecture, analog-domain optical intelligent computing can only handle relatively basic tasks. Only by expanding the scale and increasing the complexity of the optical network can more intelligent optical computing tasks be executed. Typical methods include increasing the number of neurons, the number of layers, and introducing non-linear activation. However, since optical intelligent computing is essentially the implementation of analog computing, the analog errors in optical computing pose the biggest obstacle to expanding the scale of the optical network and realizing more complex functions. Nevertheless, by introducing the intelligent optical computing inverse design method provided in the embodiments of the present disclosure, even in large-scale and non-linear computing architectures, analog errors can be effectively solved.

[0128] Taking a scenario as an example, a three-layer analog-domain optical intelligent computing task for handwritten digit classification provided by the embodiments of the present disclosure. Among them, the on-chip diffraction super-system is reverse-designed to perform the handwritten digit recognition task, including preprocessing, optical phase modulation, and output conversion; that is, before entering the optical network, the data undergoes some preprocessing steps, including downsampling a (28, 28) grayscale image to (4, 4), and then flattening it into a (16,) vector; the on-chip optical neural network for optical phase modulation consists of three layers of diffractive sub-wavelength structures, each layer with a width of 300 microns and having 1000 independently controllable optical network neurons, and the distances between the input surface and each layer and between the layers and the output layer are fixed at 60 microns. The handwritten digit signal is encoded as the amplitude of the input optical field and is input into the planar waveguide through 16 reverse-tapered waveguides. The optical field propagates through the waveguide and then is modulated by the three layers of diffractive sub-wavelength structures in sequence. Finally, the optical field reaches the output surface, where there are 10 output reverse-tapered waveguides. Ten intensity values are recorded through the output ports, and the classification result is determined by the highest intensity value among all 10 output ports.

[0129] Among them, the three-layer analog-domain optical intelligent computing model is a combination of the above optical phase modulation model and the planar mode propagation model. The MNIST dataset can be used as the original sample dataset to train the three-layer analog-domain optical intelligent computing, and the MNIST dataset is divided into a training set containing 50,000 samples and a test set containing 10,000 samples. During the training process, the intelligent optical computing reverse design model can be used as the forward propagation model to process the input signal. The structural parameters of the optical network are input into the intelligent optical computing reverse design model, and through the propagation calculation of the actual optical response of the computing network, the calculated response is compared with the target optical response, and the mean-square error (MSE) is used as the loss function of the optical network. In addition, the loss function can also be regarded as the design objective function of the reverse design task. The optimization process is to adjust and optimize the structural parameters of the sub-wavelength structure through backpropagation gradients. During this process, the three-layer analog-domain optical intelligent computing can be deployed on TensorFlow, and the Adam optimizer is selected as the optimizer for this task.

[0130] In some embodiments, the intelligent optical computing reverse design model is replaced by a numerical calculation method solver and a Fourier optics method model respectively to compare their prediction capabilities. Among them, the Fourier optics method model is considered to be the best-performing analytical model so far, and the result calculated by the Fourier optics method can be regarded as the benchmark. The numerical calculation method is a very reliable full-wave numerical method for electromagnetic field simulation, and its simulation result can be regarded as the true value.

[0131] According to some embodiments, a six - layer analog - domain optical intelligent computing for voice command recognition tasks provided by the embodiments of the present disclosure. As shown, it presents a more complex on - chip optical diffraction computing architecture, which extends the analog - domain optical intelligent computing to six layers and introduces three non - linear layers into the network. The architecture of this six - layer analog - domain optical intelligent computing includes pre - processing, optical phase modulation, non - linear activation, and output conversion. It shows the performance comparison results of linear analog - domain optical intelligent computing and non - linear analog - domain optical intelligent computing in voice command classification. The results indicate that after introducing non - linearity, the learning ability of the network can be enhanced, thereby improving the classification accuracy.

[0132] According to some embodiments, for the application of optical computing in audio classification tasks, such as vowel classification, these tasks involve a small number of categories and low feature dimensions, so as to be compatible with a completely linear optical computing framework. Therefore, the embodiments of the present disclosure can select a language instruction data set as the original sample data set. This data set contains a total of 34 categories of voice data. Among them, ten categories can be selected for the classification tasks of the six - layer analog - domain optical intelligent computing. The language instruction data set has a total of 20,000 samples, of which 16,000 samples are used as the training set and 4,000 samples are used as the test set. Each sample is usually a voice segment with a duration of about 1 second and a sampling rate of 16,000 Hz. These voice samples are recorded and organized from different speakers using various recording devices. Therefore, accents, background noise, and device - specific artifacts pose great challenges to the classification tasks. For such complex tasks, a simple shallow - layer linear optical network is obviously insufficient. Therefore, a deep non - linear optical network is needed to solve this problem. At the same time, the intelligent optical computing reverse design method proposed by the present disclosure is used to improve the accuracy of the physical modeling of this six - layer analog - domain optical intelligent computing, and enhance the efficiency and fidelity of the design.

[0133] In some embodiments, for the selected ten categories, the input sample data with the true label of "eight" is forward - inferred respectively using a numerical calculation method solver, a Fourier - optics method model, and an intelligent optical computing reverse design model. The intelligent optical computing reverse design method provided by the present disclosure has achieved sufficient accuracy in physical modeling. The classification accuracy verified by the numerical calculation method solver can reach 80%, even when the depth and complexity of the network increase. The classification accuracy of the Fourier - optics method model is only 16%, slightly higher than random guessing.

[0134] It should be noted that when the intelligent optical computing reverse design method is applied to achieve reliable and efficient reverse design, it can provide a more accurate and rapid electromagnetic field prediction tool. On the one hand, due to the simplification and abstraction of the actual physical process, the analytical model can quickly calculate approximate results. However, its accuracy is not high, and after passing through multiple-layer structures, the errors will accumulate, resulting in a decrease in the fidelity of the final device performance. In addition, for some analytical models, such as the Fourier optics method model, a table lookup process is used for solution, which makes it difficult to implement gradient backpropagation. On the other hand, the numerical method solves the time-dependent partial differential equations through the finite difference method, providing accurate results at the cost of significant spatial and temporal computational resources. By using the data-driven deep learning method to learn the diffraction process, the time-consuming and resource-intensive numerical simulation process can be eliminated, thus achieving direct and accurate prediction of the electromagnetic field. In addition, the end-to-end learning architecture does not require specialized physical knowledge or analytical capabilities.

[0135] Through the intelligent optical computing reverse design model, the computing supersystem for performing handwritten digit recognition and voice command recognition can be reliably and efficiently reverse-designed. Verified by numerical calculation methods, it is proved to have high efficiency and high fidelity, and surpasses the comprehensive performance of existing numerical, analytical, or other deep learning-based methods. It can reduce the modeling error by two orders of magnitude and ensure consistent high precision in both near-field and far-field scenarios. Additionally, although the speed of the intelligent optical computing reverse design model is slightly slower than that of the Fourier optics method model, it is 17,000 times faster than numerical simulation. Moreover, its output is differentiable with respect to the input, allowing for error backpropagation. As an accurate and rapid forward solver, the intelligent optical computing reverse design model can ensure high fidelity and efficiency in design, and is suitable for scenarios that require a large number of evaluations.

[0136] It should be noted that although only the design of the supersystem composed of one-dimensional subwavelength structures is shown in the above embodiments, if the one-dimensional tensor in the intelligent optical computing reverse design model is extended to two dimensions, the efficient and accurate reverse design of two-dimensional subwavelength structures can be achieved. In addition, by increasing the number of channels in the tensor, each channel representing different frequency components of the electromagnetic field, the reverse design of devices based on the frequency-domain response can also be realized.

[0137] In summary, the method provided by the embodiments of the present disclosure innovates a design paradigm by adopting an intelligent optical computing inverse design model, which can ensure the balance between high efficiency and high fidelity, has good scalability, and shows stronger flexibility compared with the previous inverse design methods based on deep learning. More importantly, this flexible paradigm can be applied to the unprecedented challenging designs of large-scale, high-degree-of-freedom, and functionally complex devices, such as on-chip optical diffraction networks, thereby further promoting the development of computing supersystems and paving the way for the development of large-scale, highly integrated, and powerful optical intelligent computing.

[0138] To implement the above embodiments, the present disclosure also proposes an intelligent optical computing inverse design architecture.

[0139] As Figure 5 shown, the intelligent optical computing inverse design architecture includes:

[0140] A sample acquisition unit, configured to acquire a sample data set, where the sample data set includes input electromagnetic field samples;

[0141] A parameter training unit, configured to train the structural parameters of the intelligent optical computing device to be designed based on the sample data set and the intelligent optical computing inverse design model shown in the foregoing embodiments, so as to obtain target structural parameters that meet the training requirements;

[0142] A device construction unit, configured to construct an intelligent optical computing device corresponding to the target structural parameters.

[0143] Optionally, when the parameter training unit is configured to train the structural parameters of the intelligent optical computing device to be designed to obtain target structural parameters that meet the training requirements, it is specifically configured to:

[0144] Input the structural parameters of the intelligent optical computing device to be designed and any input electromagnetic field sample in the sample data set into the intelligent optical computing inverse design model to obtain a predicted output electromagnetic field;

[0145] Input the structural parameters and any input electromagnetic field sample into a numerical calculation solver to obtain the true output electromagnetic field corresponding to any input electromagnetic field sample;

[0146] Determine a second error value between the predicted output electromagnetic field and the true output electromagnetic field. If the second error value is greater than the error threshold, update the structural parameters until the second error value is not greater than the error threshold to obtain target structural parameters that meet the training requirements.

[0147] It should be noted that the foregoing explanations of the embodiments of the intelligent optical computing inverse design method are also applicable to the intelligent optical computing inverse design architecture of this embodiment, and will not be elaborated here.

[0148] In summary, the architecture provided in this embodiment innovates a design paradigm by adopting an intelligent optical computing reverse design model, which can ensure the balance between high efficiency and high fidelity, has good scalability, and shows stronger flexibility compared with previous reverse design methods based on deep learning. More importantly, this flexible paradigm can be applied to the unprecedented challenging design of large-scale, high-degree-of-freedom and functionally complex devices, such as on-chip optical diffraction networks, thus further promoting the development of computing supersystems and paving the way for the development of large-scale, highly integrated and powerful optical intelligent computing.

[0149] The collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information involved in this disclosure all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0150] It should be noted that personal information from users should be collected for legal and reasonable purposes and should not be shared or sold outside of these legal uses. In addition, such collection / sharing should be carried out after obtaining the informed consent of the user, including but not limited to notifying the user to read the user agreement / user notice and signing an agreement / authorization including authorizing relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others with access to the personal information data comply with their privacy policies and procedures.

[0151] This disclosure anticipates providing embodiments for users to selectively block the use or access of personal information data. That is, this disclosure anticipates providing hardware and / or software to prevent or block access to such personal information data. Once personal information data is no longer needed, the risk can be minimized by restricting data collection and deleting the data. In addition, when applicable, personal identifiers are removed from such personal information to protect the privacy of the user.

[0152] In the technical solution of this disclosure, the acquisition, transmission, storage, use, processing, etc. of data all comply with the relevant provisions of national laws and regulations.

[0153] It should be noted that in the embodiments of this disclosure, some existing industry solutions such as certain software, components, models, etc. may be mentioned. They should be considered exemplary, and their purpose is only to illustrate the feasibility in the implementation of the technical solution of this application, but it does not mean that the applicant has already or necessarily used this solution.

[0154] In the descriptions of the foregoing embodiments, the descriptions with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples.

[0155] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0156] Any process or method description in a flowchart or described in other ways herein may be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logic function or process, and the scope of the preferred embodiments of the present disclosure includes additional implementations, where the functions may be performed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present disclosure pertain.

[0157] The logic and / or steps represented in the flowchart or otherwise described herein can, for example, be considered a definitional sequence list of executable instructions for implementing logical functions, and can be embodied specifically in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection portion having one or more wirings (electronic device), a portable computer diskette (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or other suitable medium on which a program can be printed, as the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpretation, or otherwise processing as appropriate, and then storing it in a computer memory.

[0158] It should be understood that various parts of the present disclosure can be implemented by hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0159] Those of ordinary skill in the art of this technology can understand that all or part of the steps carried by the method of implementing the above embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0160] In addition, each functional unit in various embodiments of the present disclosure may be integrated into a processing module, may exist separately physically for each unit, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0161] The above-mentioned storage medium may be a read-only memory, a magnetic disk, an optical disc, etc. Although the embodiments of the present disclosure have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present disclosure.

Claims

1. A calculation method applied to an intelligent optical computing reverse design model, characterized in that, The intelligent optical computing reverse design model includes an input division module, a wavefront prediction module, and an output splicing module. The calculation method includes: Controlling the input division module to divide the input electromagnetic field into a set of input electromagnetic field segments based on the Huygens-Fresnel law, according to the total wavefront width of the input electromagnetic field and the wavefront width corresponding to each electromagnetic neural local solver in the electromagnetic neural local solver set; Controlling the wavefront prediction module to input each input electromagnetic field segment in the set of input electromagnetic field segments and the structural parameters corresponding to the intelligent optical computing device to be designed into the electromagnetic neural local solver corresponding to the input electromagnetic field segment, so that the electromagnetic neural local solver predicts the output electromagnetic field segment according to the input electromagnetic field segment and the structural parameters, obtaining a set of output electromagnetic field segments. Among them, a complex linear connection is performed on the input electromagnetic field segment to represent the non-physical information propagation process of the input electromagnetic field segment in the intelligent optical computing device to be designed, obtaining the input electromagnetic field segment after non-physical propagation; a real non-linear connection is performed on the structural parameters, and the connected structural parameters are aligned with the input electromagnetic field segment after non-physical propagation to represent the physical information propagation process of the input electromagnetic field segment after propagation in the intelligent optical computing device to be designed, obtaining the input electromagnetic field segment after physical propagation; a complex linear connection is performed on the input electromagnetic field segment after physical propagation to represent the non-physical information propagation process of the input electromagnetic field segment after physical propagation in the intelligent optical computing device to be designed, obtaining the output electromagnetic field segment; Controlling the output splicing module to splice all the output electromagnetic field segments in the set of output electromagnetic field segments based on the Huygens-Fresnel law, obtaining the output electromagnetic field under the global response, so as to adjust the structural parameters according to the output electromagnetic field.

2. The calculation method according to claim 1, characterized in that The step of dividing the input electromagnetic field into a set of input electromagnetic field segments according to the total wavefront width of the input electromagnetic field and the wavefront width corresponding to each electromagnetic neural local solver in the electromagnetic neural local solver set includes: Dividing the total wavefront width of the input electromagnetic field based on the wavefront width corresponding to each electromagnetic neural local solver in the electromagnetic neural local solver set, obtaining a division result; If the division result indicates that there is an input electromagnetic field segment with a width less than the wavefront width in the input electromagnetic field, zero padding is added at the end side and / or the end of the input electromagnetic field until the width of each input electromagnetic field segment is the wavefront width, obtaining a set of input electromagnetic field segments, where the input electromagnetic field segment is regarded as a secondary wave generated by a central source point on the input electromagnetic field segment.

3. The calculation method according to claim 1, wherein The step of performing complex linear connection includes: Obtaining a first weight matrix and a first intercept term used for complex linear connection, and performing complex linear connection based on the first weight matrix and the first intercept term, where the first weight matrix and the first intercept term are obtained by training an initial electromagnetic neural local solver.

4. The calculation method according to claim 1, characterized in that The step of performing real non-linear connection includes: Obtain the second weight matrix, the second intercept term, and the non-linear operation function used in the real non-linear connection, and perform complex linear connection based on the second weight matrix, the second intercept term, and the non-linear operation function, where the second weight matrix, the second intercept term, and the non-linear operation function are obtained by training an initial electromagnetic neural local solver.

5. The calculation method according to claim 1, characterized in that, Before inputting each input electromagnetic field segment in the set of input electromagnetic field segments and the structural parameters corresponding to the intelligent optical computing device to be designed into the electromagnetic neural local solver corresponding to the input electromagnetic field segment, the method further includes: Obtain a training sample set, where each training sample in the training sample set includes an input electromagnetic field training sample and a structural parameter training sample; Perform gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set until an electromagnetic neural local solver that meets the training requirements is obtained; Obtain a test sample set, where each test sample in the test sample set includes an input electromagnetic field test sample and a structural parameter test sample; Test the electromagnetic neural local solver based on the test sample set. If the test result does not meet the test requirements, re-perform gradient backpropagation training on the electromagnetic neural local solver until an electromagnetic neural local solver with a test result that meets the test requirements is obtained.

6. The calculation method according to claim 5, characterized in that The performing gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set includes: During the process of performing gradient backpropagation training on the initial electromagnetic neural local solver based on the training sample set, input any training sample in the training sample set into the initial electromagnetic neural local solver to obtain a predicted value corresponding to the any training sample, and determine a first error value between the predicted value and the true value corresponding to the any training sample. Update the network parameters of the initial electromagnetic neural local solver based on the first error value to minimize the loss function.

7. An intelligent optical computing reverse design method, characterized in that, Includes: Obtain a sample data set, where the sample data set includes input electromagnetic field samples; Train the structural parameters of the intelligent optical computing device to be designed based on the sample data set and the calculation method applied to the intelligent optical computing reverse design model according to any one of claims 1 to 6 to obtain target structural parameters that meet the training requirements; Construct an intelligent optical computing device corresponding to the target structural parameters.

8. The method according to claim 7, wherein The training the structural parameters of the intelligent optical computing device to be designed to obtain target structural parameters that meet the training requirements includes: Input the structural parameters of the intelligent optical computing device to be designed and any input electromagnetic field sample in the sample data set into the intelligent optical computing reverse design model to obtain a predicted output electromagnetic field; Input the structural parameters and the any input electromagnetic field sample into a numerical calculation solver to obtain the true output electromagnetic field corresponding to the any input electromagnetic field sample; Determine a second error value between the predicted output electromagnetic field and the true output electromagnetic field. If the second error value is greater than the error threshold, update the structural parameters until the second error value is not greater than the error threshold, and obtain the target structural parameters that meet the training requirements.

9. An intelligent optical computing reverse design system, characterized in that, It includes: A sample acquisition unit, configured to acquire a sample data set, where the sample data set includes input electromagnetic field samples; A parameter training unit, configured to train the structural parameters of the intelligent optical computing device to be designed based on the sample data set and the calculation method applied to the intelligent optical computing reverse design model according to any one of claims 1 to 6, and obtain the target structural parameters that meet the training requirements; A device construction unit, configured to construct an intelligent optical computing device corresponding to the target structural parameters.

Images