CN109117575B - Method and device for determining structural parameters of surface plasmon waveguide system - Google Patents

Abstract

The method and the device for determining the structure parameters of the surface plasmon waveguide system provided by the embodiment of the invention are used for obtaining the structure type of the surface plasmon waveguide system. And based on the structure type, obtaining target electromagnetic response data of the surface plasmon waveguide system aiming at the structure type, wherein the target electromagnetic response data is used for indicating the performance of a target device of the surface plasmon waveguide system, and inputting the target electromagnetic response data into a pre-trained reverse calculation model corresponding to the structure type to obtain target structure parameters corresponding to the structure type. The inverse calculation model corresponding to this structure type is: and the neural network model is obtained by training in advance by using a plurality of preset sample structure parameters corresponding to the structure type and a plurality of sample electromagnetic response data respectively corresponding to each sample structure parameter one by one. Compared with a target structure parameter determination mode which needs multiple times of manual participation, the determination efficiency of the target structure parameters is improved.

Description

Method and device for determining structural parameters of surface plasmon waveguide system

technical field

The invention relates to the technical field of photonic devices, in particular to a method and device for determining the structural parameters of a surface plasmon waveguide system.

Background technique

SPPs (Surface Plasmon Polaritons, surface plasmon polaritons) waveguide systems are nanoscale photonic devices that can excite and transmit SPPs, such as nanoscale optical switches, modulators, logic gates, sensors and other photonic devices. The SPPs waveguide system can utilize the characteristics of SPPs to form electromagnetic oscillations through the interaction of free electrons and photons in the metal surface area, break the diffraction limit of light, and perform subwavelength or even nanometer-sized light manipulation, thereby achieving low loss and long distances. At the same time, the size of nanometer scale is suitable for on-chip integration to realize high transmission rate optical data transmission.

In practical applications, different SPPs waveguide systems can have different structural types. For example, Fig. 1(a) and Fig. 1(b) show two different structures of SPPs waveguide systems. The structure shown in FIG. 1( a ) includes a part constituting the structure: the resonant cavity 0 . The structure shown in Fig. 1(b) does not include the part: resonant cavity 0 . For the SPPs waveguide system of any structure, the device performance of the SPPs waveguide system can be measured according to the electromagnetic response data of the SPPs waveguide system of the structure, such as transmission spectrum, field distribution, etc., and the SPPs waveguide system of the structure can be used to excite and transmit SPPs. ability. For the SPPs waveguide system of this structure, structural parameters, such as the size of each part constituting the structure, are important factors to determine the electromagnetic response data of the SPPs waveguide system of this structure. Therefore, before fabricating the SPPs waveguide system of a certain structure, it is necessary to consider how to obtain the target structure parameters of the structure that can achieve the target electromagnetic response data, so as to fabricate the SPPs waveguide system of the structure with the target device performance according to the target structure parameters.

In order to obtain the target structure parameters of the above-mentioned structure that can achieve the target electromagnetic response data, the simulation results are usually obtained by using the set structure parameters for the structure, and it is judged whether the simulation results conform to the target electromagnetic response data. If not, according to the obtained simulation results, manually calculate the structural parameters that need to be adjusted, and then use the adjusted structural parameters to simulate to obtain new simulation results, and return to the step of judging whether the simulation results conform to the target electromagnetic response data. This cycle is repeated until structural parameters that match the target electromagnetic response data are obtained. However, the structural parameters need manual calculation and adjustment during the simulation process, and multiple simulations and multiple manual adjustments make the process of determining the target structural parameters complicated and inefficient.

SUMMARY OF THE INVENTION

The purpose of the embodiments of the present invention is to provide a method and device for determining the structural parameters of a surface plasmon waveguide system, so as to improve the efficiency of determining the structural parameters without the need for repeated manual participation. The specific technical solutions are as follows:

In a first aspect, an embodiment of the present invention provides a method for determining structural parameters of a surface plasmon waveguide system, the method comprising:

Obtain the structure type of the surface plasmon waveguide system;

Based on the structure type, obtain target electromagnetic response data for the surface plasmon waveguide system of this structure type; wherein, the target electromagnetic response data is used to indicate the target device performance of the surface plasmon waveguide system;

Input the target electromagnetic response data into the pre-trained reverse calculation model corresponding to the structure type, and obtain the target structure parameters corresponding to the structure type;

Wherein, the reverse calculation model corresponding to the structure type is: pre-determined multiple sample structure parameters corresponding to the structure type and multiple sample electromagnetic response data corresponding to each sample structure parameter in advance. The trained neural network model.

In a second aspect, an embodiment of the present invention provides an apparatus for determining structural parameters of a surface plasmon waveguide system based on a neural network, the apparatus comprising:

The acquisition module is used to obtain the structure type of the surface plasmon waveguide system; based on the structure type, the target electromagnetic response data of the surface plasmon waveguide system for the structure type is obtained; wherein, the target electromagnetic response data is used to indicate Target device performance of surface plasmon waveguide systems;

The target structure parameter determination module is used for inputting the target electromagnetic response data into the pre-trained reverse calculation model corresponding to the structure type to obtain the target structure parameter corresponding to the structure type;

Wherein, the reverse calculation model corresponding to the structure type is: pre-determined multiple sample structure parameters corresponding to the structure type and multiple sample electromagnetic response data corresponding to each sample structure parameter in advance. The trained neural network model.

In a third aspect, an embodiment of the present invention provides an electronic device, the device comprising:

A processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the bus; the memory is used to store computer programs; the processor is used to execute the programs stored in the memory to achieve The steps of the method for determining the structural parameters of the surface plasmon waveguide system provided by the first aspect above.

In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, where a computer program is stored in the storage medium, and when the computer program is executed by a processor, implements the surface plasmon waveguide system provided in the first aspect above. Steps of a method for determining structural parameters.

The embodiments of the present invention provide a method and device for determining the structural parameters of a surface plasmon waveguide system, by obtaining the structure type of the surface plasmon waveguide system. Based on the structure type, the target electromagnetic response data for the surface plasmon waveguide system of this structure type is obtained, wherein the target electromagnetic response data is used to indicate the target device performance of the surface plasmon waveguide system, and the target electromagnetic response data is The response data is input into the pre-trained reverse calculation model corresponding to the structure type, and the target structure parameter corresponding to the structure type is obtained. Wherein, the reverse calculation model corresponding to the structure type is: pre-determined multiple sample structure parameters corresponding to the structure type and multiple sample electromagnetic response data corresponding to each sample structure parameter in advance. The trained neural network model. Compared with the traditional method of determining the target structure parameters that requires multiple manual participation, the mapping relationship between the structure parameters and the electromagnetic response data is established by using the reverse calculation model obtained by pre-training. When the target structure parameter corresponding to the type is selected, the target electromagnetic response data corresponding to the structure type is input into the pre-trained reverse calculation model, and the target structure parameter corresponding to the structure type that can achieve the target electromagnetic response data can be obtained. The determination of the structural parameters does not require manual adjustment of structural parameters and multiple simulations, thereby improving the efficiency of determining the target structural parameters.

Description of drawings

In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that are required in the description of the embodiments or the prior art.

Fig. 1(a) and Fig. 1(b) are respectively the structural example diagrams of the surface plasmon waveguide system;

FIG. 2 is a schematic flowchart of a method for determining structural parameters of a surface plasmon waveguide system according to an embodiment of the present invention;

3 is a schematic diagram of a structure of a reverse calculation model according to an embodiment of the present invention;

4 is a schematic flowchart of a training process of a reverse computing model according to an embodiment of the present invention;

FIG. 5( a ) is a schematic structural diagram of a metal-dielectric-metal waveguide coupling cavity structure according to an embodiment of the present invention;

FIG. 5(b) is a schematic diagram of the transmission spectrum of the electromagnetic response data of the sample according to an embodiment of the present invention;

FIG. 6(a) is a schematic diagram of the actual value and the predicted value of the structural parameter according to an embodiment of the present invention;

FIG. 6(b) is a schematic diagram of the actual transmission spectrum and the predicted transmission spectrum according to an embodiment of the present invention;

7 is a schematic flowchart of a training process of a reverse computing model according to another embodiment of the present invention;

8 is a schematic structural diagram of a forward prediction model according to an embodiment of the present invention;

9 is a schematic structural diagram of a reverse calculation model and a forward prediction model connected in series according to an embodiment of the present invention;

10(a) is a schematic diagram of the actual transmission spectrum and the predicted transmission spectrum of the training data according to an embodiment of the present invention;

10(b) is a schematic diagram of the actual transmission spectrum and the predicted transmission spectrum of the test data according to an embodiment of the present invention;

11 is a schematic flowchart of a method for determining parameters of a surface plasmon waveguide system according to another embodiment of the present invention;

12 is a schematic diagram of an optimized transmission spectrum according to an embodiment of the present invention;

13 is a schematic structural diagram of an apparatus for determining parameters of a surface plasmon waveguide system according to an embodiment of the present invention;

FIG. 14 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed ways

In order to make those skilled in the art better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described implementation Examples are only some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The following first introduces a method for determining the structural parameters of the surface plasmon waveguide system according to an embodiment of the present invention.

The method for determining the structural parameters of the surface plasmon waveguide system provided by the embodiments of the present invention can be applied to electronic devices with data processing capabilities, and the electronic devices may include desktop computers, portable computers, Internet TVs, smart mobile terminals, wearable It is not limited here, and any electronic device that can implement the embodiments of the present invention belongs to the protection scope of the embodiments of the present invention.

As shown in FIG. 2 , the flow of a method for determining the structural parameters of a surface plasmon waveguide system according to an embodiment of the present invention may include:

S201, obtaining the structure type of the surface plasmon waveguide system.

Different surface plasmon waveguide systems have different structure types, and accordingly, for surface plasmon waveguide systems with different structure types, the structural parameters that need to be determined are also different. For example, for the structure type shown in Fig. 1(a), compared with the structure type shown in Fig. 1(b), the size of the resonant cavity 0 not included in the structure type shown in Fig. 1(b) also needs to be determined. Therefore, when determining the structural parameters of the surface plasmon waveguide system, it is necessary to obtain the structural type of the device first, so as to determine the target structural parameters suitable for the structural type subsequently.

S202 , based on the structure type, obtain target electromagnetic response data for the structure type; wherein, the target electromagnetic response data is used to indicate the performance of the target device of the surface plasmon waveguide system.

S203, input the target electromagnetic response data into the pre-trained reverse calculation model corresponding to the structure type, to obtain target structure parameters corresponding to the structure type.

The electromagnetic response data can specifically include transmission spectrum, field distribution, etc., which are used to measure the device performance of the surface plasmon waveguide system and determine the ability of the surface plasmon waveguide system to excite and transmit SPPs. The target electromagnetic response data may be specifically determined according to the requirements of the device performance of the surface plasmon waveguide system before fabricating the surface plasmon waveguide system of the structure type.

For the surface plasmon waveguide system of any structure, the structural parameters of the device may specifically include the dimensions of each part constituting the structure, such as the length and width of the resonant cavity 0 in the structure shown in FIG. 1(a).

Wherein, the reverse calculation model corresponding to the structure type is: pre-determined multiple sample structure parameters corresponding to the structure type and multiple sample electromagnetic response data corresponding to each sample structure parameter in advance. The trained neural network model.

Specifically, as shown in FIG. 3 , the structure of the reverse calculation model according to an embodiment of the present invention may include: an input layer, a hidden layer, and an output layer. The acquired data in the transmission spectrum corresponding to the structure type is taken as the target electromagnetic response data t ₁ , t ₂ ,..., t _m , which is input to the reverse calculation model through the input layer, and is calculated by the hidden layer and output by the output layer. Target structure parameters p ₁ , p ₂ , . . . , p _k , where k is the total number of target structure parameters corresponding to the target electromagnetic response data. m is the total number of target electromagnetic response data, the horizontal axis in the transmission spectrum represents the wavelength of the incident light entering the surface plasmon waveguide system, and the vertical axis represents the transmission spectrum of the surface plasmon waveguide system corresponding to different incident light wavelengths over rate. Exemplarily, for the surface plasmon waveguide system with the structure shown in FIG. 1( a ), the target structure parameters p ₁ , p ₂ , . . . , p _k may specifically be the length, width and waveguide coupling distance between resonators.

The method for determining the structural parameters of the surface plasmon waveguide system provided by the embodiment of the present invention is compared with the traditional method for determining the target structural parameters that requires multiple manual participation. Therefore, when determining the target structure parameters, the target electromagnetic response data is input into the pre-trained reverse calculation model, and the target structure parameters that can achieve the target electromagnetic response data can be obtained. , the determination of the target structure parameters does not require manual adjustment of the structure parameters and multiple simulations, thereby improving the efficiency of the determination of the target structure parameters.

As shown in FIG. 4, the flow of the training process of the reverse computing model according to an embodiment of the present invention may include:

S401: Input the sample electromagnetic response data corresponding to each sample structure parameter one-to-one into the current neural network model respectively, and obtain the predicted structure parameter corresponding to each sample electromagnetic response data. When the current neural network model is used for the first time, it is the default initial neural network model.

Referring to the structure of the surface plasmon waveguide system shown in Fig. 1(a) and Fig. 1(b), in general, a surface plasmon waveguide system of a certain structure type will contain a number of different parts, which need to be targeted for Each site determines the target structural parameters. Correspondingly, it is necessary to set the sample structure parameters for each part and obtain the electromagnetic response data corresponding to each sample structure parameter one-to-one.

Therefore, for a surface plasmon waveguide system with multiple structural parameters, optionally, multiple sample structural parameters corresponding to the structure can be set by using the following steps:

Obtain multiple preset initial structure parameters, preset parameter adjustment accuracy and preset parameter variation range.

For each preset initial structure parameter, the precision is adjusted according to the preset parameter, and within the preset parameter variation range, at least one parameter change value corresponding to the preset initial structure parameter is determined.

Based on the preset initial structure parameter and the corresponding at least one parameter change value, at least one sample structure parameter corresponding to the preset initial structure parameter is determined.

The electromagnetic response data of multiple samples corresponding to each sample structure parameter one-to-one can be obtained by using the following steps:

Using the determined at least one sample structure parameter corresponding to each preset initial structure parameter, a plurality of sample electromagnetic response data corresponding to each sample structure parameter one-to-one is obtained by sampling.

For example, referring to FIG. 5( a ), set the sample structure parameters corresponding to the MDMWCC structure (metal-dielectric-metal waveguide coupled with cavities, metal-dielectric-metal waveguide coupled cavity structure) and the parameters of each sample structure one by one. The step of corresponding electromagnetic response data of multiple samples includes: acquiring preset initial structure parameters corresponding to the structure: in the MDMWCC structure, the length I ₁ =480nm, the width W ₁ =100nm of the resonant cavity 1, the resonant cavity 1 and the resonant cavity are The waveguide coupling distance g ₁ =255nm between 2; the length I ₂ =520nm, the width W ₂ =100nm of the resonator 2, the waveguide coupling distance g ₂ =330nm between the resonator 2 and the resonator 3; Length I ₃ =520 nm, width W ₃ =100 nm. The preset parameter adjustment precision can be 1nm, and the preset parameter variation range can be a closed interval [preset initial structure parameter -20nm, preset initial structure parameter +20nm].

For ease of understanding, the description is made with respect to the length I ₁ =480 nm of the resonant cavity 1 in the preset initial structural parameters. According to the preset parameter adjustment accuracy of 1 nm, within the closed interval of the preset parameter variation range, determine the parameter variation values -20 nm, -19 nm, . . . , 19 nm, 20 nm corresponding to the length I ₁ =480 nm of the resonant cavity 1 . Based on the length I ₁ =480 nm of the resonant cavity 1 and the corresponding parameter change values above, the sample structure parameters 460 nm, 461 nm, ..., 499 nm, 500 nm corresponding to the length I ₁ =480 nm of the resonant cavity 1 are determined. Similarly, the width W ₁ of the resonant cavity 1 , the waveguide coupling distance g ₁ between the resonant cavity 1 and the resonant cavity 2 , the length I ₂ of the resonant cavity 2 , the width W ₂ , and the distance between the resonant cavity 2 and the resonant cavity 3 can be determined. The waveguide coupling distance g ₂ , the length I ₃ and the width W ₃ of the resonant cavity 3 correspond to at least one sample structure parameter respectively. Of course, the determined change value of at least one parameter corresponding to the preset initial structure parameter may specifically be based on the quantity requirement of the sample data to adjust the precision according to the preset parameter for each preset initial structure parameter. Within the variation range, it is determined that at least one parameter variation value corresponding to the preset initial structure parameter is the random number generated by the random number generation rule.

Using the at least one sample structure parameter determined above, a simulation is performed by Lumerical FDTD Solutions, a software used for simulation design of photonics products, to obtain the electromagnetic response corresponding to the sample structure parameter, such as the transmission shown in Figure 5(b). spectrum. Then, a set of sample electromagnetic response data corresponding to the electromagnetic response is obtained based on MC sampling (Monte Carlo, sampling for generating random numbers of a specified distribution). The combination of the sample structure parameters and the corresponding sample electromagnetic response data is called a set sample. The number of sample data can be determined according to specific needs, such as 15,000 groups.

In addition, the preset initial structural parameters, the preset parameter adjustment accuracy, and the preset parameter variation range may specifically be received after input by the designer of the surface plasmon waveguide system, or may be obtained from pre-stored preset initial structural parameters. , obtained from the storage module of the preset parameter adjustment accuracy and the preset parameter variation range.

The sample structure parameters and corresponding sample electromagnetic response data obtained from the above steps, and the input preset initial neural network model can specifically set the number of layers and the number of neurons in each layer according to historical experience, or can be based on training A neural network model with a specified number of layers and neurons determined by the historical experience of the neural network model.

For example, the preset initial neural network model may be a 6-layer neural network model, and the number of neurons in each layer is 200-100-200-500-100-8, wherein the number represents the number of neurons in each layer, The number of neurons in the first layer and the last layer respectively corresponds to the number of input sample electromagnetic response data and the number of corresponding sample structure parameters. The structure of the preset initial neural network model can be the same as the reverse design model shown in FIG. 3 of the present invention. It can be understood that, in the training process of the reverse calculation model, the neural network model used for the first time is a preset initial neural network model.

The sample electromagnetic response data in the above embodiment of FIG. 5(b) are respectively input into the current neural network model, and the predicted structural parameters corresponding to each sample electromagnetic response data are obtained. Of course, the obtained predicted structural parameters include: predicted length of resonator 1, predicted width, predicted waveguide coupling distance between resonator 1 and resonator 2, predicted length of resonator 2, predicted width, predicted resonator 2 and resonator The waveguide coupling distance between the cavities 3, the predicted length of the resonant cavity 3, and the predicted width.

S402 , according to the obtained multiple predicted structural parameters and the first preset cost function, determine whether the current neural network model has converged, if so, go to S403 , if not, go to S404 to S405 . The first preset cost function is a function set based on a sample structure parameter.

Specifically, judging whether the current neural network model is converging may be, aiming at minimizing the cost function, and calculating the minimum value of the first preset cost function. When the minimum value is obtained, it means that the current neural network model has converged, and when the minimum value has not been obtained. , it means that the current neural network model does not converge. Specifically, the first preset cost function may be Formula 1:

J ² =u/v,

Among them, J is the error between the sample structure parameter corresponding to the sample electromagnetic response data and the predicted structure parameter corresponding to the sample electromagnetic response data, u is the residual sum of squares of the sample structure parameter and the predicted structure parameter, v is the sample structure parameter and the total sample structure Sum of squares of parameter means.

S403, determining the current neural network model as a reverse calculation model.

After the inverse calculation model is obtained, the same method as the above-mentioned embodiment of FIG. 5 can be used to set multiple test structure parameters and obtain the corresponding test electromagnetic response data as a test set to verify the training effect of the inverse calculation model. For example, 500 sets of test data can be set, and the test electromagnetic response data in the 500 sets of test data can be input into the reverse calculation model to obtain predicted structural parameters corresponding to the test electromagnetic response data. The corresponding test structure parameters and the predicted structure parameters corresponding to the test electromagnetic response data are input into the first preset cost function, and the score of the cost function corresponding to the test data is obtained as 0.9029. The sample structure parameters in the embodiment of FIG. 5 and the predicted structure parameters corresponding to the sample electromagnetic response data are input into the first preset cost function, and the score of the cost function corresponding to the sample data is obtained as 0.9169. The difference in the score of the cost function corresponding to the test data and the sample data is within the expected range, so it can be determined that the training effect of the inverse calculation model reaches the expected level, and the expected inverse calculation model is established.

S404 , using a preset gradient function, and adopting a stochastic gradient descent method to adjust the model parameters of the current neural network model to obtain a new neural network model.

则所调整的模型参数为权重w。其中，b为偏置，x为目标电磁响应数据，n为神经网络的神经元个数。For example, the current neural network model can be

Then the adjusted model parameter is the weight w. Among them, b is the bias, x is the target electromagnetic response data, and n is the number of neurons in the neural network.

The preset gradient function can be specifically formula 2:

Among them, v _t is the gradient of the current model parameter, _gt is the initial gradient of the current model parameter, η is the learning rate, θ _t+1 is the new model parameter to be acquired, and θ _t is the current model parameter.

S405, update the current neural network model to the obtained new neural network model, and return to executing S401.

The current neural network model is updated to the obtained new neural network model, so as to iterate on the basis of the latest neural network model whose model parameters are adjusted each time until a converged current neural network model is obtained.

In general, when the current neural network model converges, the model parameters of the neural network model have reached the target value that the structural parameters can be accurately obtained. At this time, the accuracy of the predicted structural parameters obtained by the current neural network model conforms to the training objective of the neural network model. At this time, the real structure parameters represented by the sample structure parameters are close to the predicted structure parameters output by the current neural network model, and the real structure parameters represented by the test structure parameters are close to the predicted structure parameters corresponding to the test electromagnetic response data output by the current neural network model. , as shown in Figure 6(a). At the same time, the transmission spectrum corresponding to the sample structure parameters is close to the transmission spectrum corresponding to the predicted structure parameters corresponding to the sample electromagnetic response data. The predicted transmission spectrum is close.

However, when the neural network model converges, the accuracy of the prediction result of the model may not reach the target level. To this end, as shown in FIG. 7 , the flow of the training process of the reverse computing model according to another embodiment of the present invention may include:

S701: Input the sample electromagnetic response data corresponding to each sample structure parameter one-to-one into the current neural network model, respectively, to obtain the predicted structure parameter corresponding to each sample electromagnetic response data. When the current neural network model is used for the first time, it is the default initial neural network model.

S702, according to the obtained multiple predicted structural parameters and the first preset cost function, determine whether the current neural network model has converged, if so, execute S703 to S704, and if not, execute S707 to S708. The first preset cost function is a function set based on a sample structure parameter.

S701 to S702 are the same steps as S401 to S402 in the embodiment of FIG. 4 , and details are not repeated here. For details, please refer to the description of the embodiment of FIG. 4 .

S703, save the obtained multiple predicted structure parameters.

S704: Determine whether the degree of fit of the obtained plurality of predicted structural parameters and the corresponding plurality of sample structural parameters is less than a preset degree of fit threshold. If yes, go to step S705, if not, go to step S706.

The sample structure parameters represent the real structure parameters of the surface plasmon waveguide system, and the fitting degree of the obtained multiple predicted structure parameters and the corresponding multiple sample structure parameters is used to indicate the closeness between the predicted structure parameters and the real structure parameters degree. The higher the degree of fit, the closer the predicted structural parameters are to the real structural parameters, and the higher the accuracy of the prediction results of the current neural network model.

S705, adjust the physical structure of the current neural network model to obtain a new neural network model, and perform step S708.

When the fitting degree of multiple predicted structural parameters and the corresponding multiple sample structural parameters is less than the preset fitting degree threshold, it indicates that the accuracy of the prediction results of the current neural network model has not reached the target level. Therefore, the physical properties of the current neural network model can be adjusted At the same time, the new neural network model with the adjusted physical structure is used as the current neural network model, and the training is continued to obtain a reverse calculation model whose prediction accuracy is improved to reach the target level. Wherein, adjusting the physical structure of the current neural network model may specifically include: increasing or decreasing the number of layers of the current neural network model, and increasing or decreasing the number of neurons in each layer of the current neural network model. The specific increase or decrease can be determined according to historical experience.

S706, determining the current neural network model as a reverse calculation model.

When the fitting degree of multiple predicted structural parameters and the corresponding multiple sample structural parameters is not less than the preset fitting degree threshold, it indicates that the accuracy of the prediction results of the current neural network model has reached the target level, and there is no inaccurate prediction result of the above model. problem, therefore, the current neural network model can be determined as a reverse computational model

S707, using a preset gradient function, and adopting a stochastic gradient descent method to adjust the model parameters of the current neural network model to obtain a new neural network model.

S708, the current neural network model is updated to the obtained new neural network model, and the process returns to S701.

S707 to S708 are the same steps as S404 to S405 in the embodiment of FIG. 4 , which are not repeated here, and refer to the description of the embodiment of FIG. 4 for details.

In addition, due to the characteristic of the correspondence between the electromagnetic response data of the surface plasmon waveguide system and the structural parameters, different structural parameters may correspond to the same electromagnetic response data. In the process of training the reverse calculation model, multiple structural parameters correspond to the same electromagnetic response data, which may lead to the use of the cost function to obtain multiple model parameters corresponding to the same error between the sample data and the predicted data, which cannot be calculated. The minimum value, which in turn causes the problem that the current neural network model cannot converge.

To this end, optionally, before S404 in the embodiment of FIG. 4 , the method for determining parameters of the surface plasmon waveguide system in the embodiment of the present invention may further include:

The resulting multiple predicted structure parameters are saved.

The stored multiple predicted structural parameters are respectively input into the pre-trained forward prediction model to obtain multiple first predicted electromagnetic response data corresponding to each predicted structural parameter one-to-one. The forward prediction model is a neural network model trained by using multiple preset sample structure parameters and multiple sample electromagnetic response data corresponding to each sample structure parameter one-to-one.

Among them, the forward prediction model is a model with opposite input and output to the reverse calculation model, and is used to obtain electromagnetic response data based on the input structural parameters. Specifically, as shown in FIG. 8 , the structure of the forward prediction model according to an embodiment of the present invention may include: an input layer, a hidden layer, and an output layer. For example, the forward prediction model may specifically be a 6-layer neural network 8-50-200-500-100-200, wherein the input layer is 8 neurons for inputting 8 structural parameters. The hidden layer is 4 layers, and the number of neurons in each layer is 50, 200, 500, 100 respectively. The output layer is 200 neurons, which are used to output 200 discrete electromagnetic response data of the transmission spectrum. The structural parameters p ₁ , p ₂ ,..., p _k are input into the forward prediction model through the input layer, and are calculated by the hidden layer, and the predicted electromagnetic response data t ₁ , t ₂ ,..., t are output from the output layer. _m , where the predicted electromagnetic response data is the data in the predicted transmission spectrum.

In this embodiment, as shown in FIG. 9 , after the forward prediction model is connected in series with the reverse calculation model, the input structural parameters are a plurality of stored predicted structural parameters obtained from the reverse calculation model. When the forward prediction model is used alone, the input structural parameters can be the real structural parameters.

Of course, the process of obtaining a forward prediction model by training a plurality of preset sample structure parameters and a plurality of sample electromagnetic response data corresponding to each sample structure parameter one-to-one in advance is the same as that of the embodiment shown in FIG. 4 of the present invention. The process of training to obtain a reverse design model is similar. The difference is that during training, the input sample data, the sample data based on which to judge whether to converge, and the cost function used are different.

Correspondingly, after the positive prediction model is obtained by training, the same method as the embodiment of FIG. 5 of the present invention can be used to set multiple (for example, 500 groups) test structure parameters and corresponding test electromagnetic response data, which are used to verify the positive prediction model. To predict the training effect of the model. Specifically, the verification method is the same as the above-mentioned verification method for verifying the training effect of the reverse calculation model, and the difference lies in the verification of the neural network model, the input data used, and the output data obtained. For example, after feeding the test data into the forward prediction model, the resulting cost function has a score of 0.9269, and after feeding the training data into the forward prediction model, the resulting cost function has a score of 0.9356. The scores of the two cost functions are within the expected score range. Therefore, it is determined that the training effect of the forward prediction model reaches the expected level, and the expected forward prediction model is established.

At the same time, the prediction accuracy of the forward prediction model also reaches the target level. When the test electromagnetic response data is used as the real data, the corresponding predicted electromagnetic response data obtained by the forward prediction model is close to the real data. For example, referring to Fig. 10(a), in the training data, any set of structural parameters I ₁ =486 nm, I ₂ =555 nm, I ₃ =585 nm, w ₁ =93 nm, w ₂ =91 nm, w ₃ =115 nm, g ₁ = The actual transmission spectrum corresponding to 294 nm, g ₂ =320 nm is close to the predicted transmission spectrum. When the training electromagnetic response data is used as the real data, the corresponding real transmission spectrum obtained by the forward prediction model is close to the predicted transmission spectrum. For example, referring to Fig. 10(b), in the test data, any set of structural parameters I ₁ =466 nm, I ₂ =524 nm, I ₃ =589 nm, w ₁ =115 nm, w ₂ =93 nm, w ₃ =90 nm, g ₁ = The actual transmission spectrum corresponding to 280 nm, g ₂ =335 nm is close to the predicted transmission spectrum.

Since one structural parameter of the surface plasmon waveguide system corresponds to one electromagnetic response data, after inputting the predicted structural parameters into the forward prediction model, multiple predicted structures corresponding to the same electromagnetic response data in the reverse calculation model can be calculated The parameter is mapped to a first predicted electromagnetic response data in the forward prediction model, so as to avoid the situation where the cost function has multiple solutions and the minimum value cannot be determined. Specifically, as shown in FIG. 9 , a reverse calculation model and a forward prediction model are connected in series according to an embodiment of the present invention.

Whether the current neural network model is converged is determined according to the obtained plurality of first predicted electromagnetic response data and a second preset cost function, where the second preset cost function is a function set based on the sample electromagnetic response data.

Since the above steps use the forward prediction model to realize the mapping of the unique solution to the output of the current neural network, therefore, based on the first predicted electromagnetic response data obtained above, it can be judged whether the current neural network model used for training to obtain the reverse calculation model is not convergence. Specifically, it can be determined whether the current neural network model has converged according to the obtained plurality of first predicted electromagnetic response data and the second preset cost function. The second preset cost function may be a cost function similar to formula 1, except that the sample data is adjusted to the sample electromagnetic response data, and the predicted data is adjusted to the first predicted electromagnetic response data.

If it is, the current neural network model is determined as the inverse computation model.

If not, perform the steps of using the preset gradient function to adjust the model parameters of the current neural network model to obtain a new neural network model.

In practical applications, there may also be situations in which new electromagnetic response data needs to be determined due to device performance upgrades in the surface plasmon waveguide system and changes in device performance requirements. In this regard, the traditional technology determines new electromagnetic response data through complex formula operations according to the upgraded device performance or changed device performance requirements, and then determines new target structural parameters, which is a complicated process.

In order to simplify the process of determining new target structure parameters on the basis of historical electromagnetic response data, optionally, as shown in FIG. 11 , the process of a method for determining parameters of a surface plasmon waveguide system according to another embodiment of the present invention , the method can include:

S1101, obtaining the structure type of the surface plasmon waveguide system.

S1102 , based on the structure type, obtain target electromagnetic response data for the structure type; wherein, the target electromagnetic response data is used to indicate the performance of the target device of the surface plasmon waveguide system.

S1103: Input the target electromagnetic response data into the pre-trained reverse calculation model corresponding to the structure type, to obtain target structure parameters corresponding to the structure type.

S1101 to S1103 are the same steps as S201 to S203 in the embodiment shown in FIG. 2 of the present invention, and are not repeated here. For details, refer to the description of the embodiment shown in FIG. 1 of the present invention.

S1104: Acquire second predicted electromagnetic response data corresponding to the target structure parameter.

The acquisition method of the second predicted electromagnetic response data corresponding to the target structural parameters may specifically include: inputting the target structural parameters into a pre-trained forward prediction model, and obtaining two descriptions of the second predicted electromagnetic response data corresponding to the target structural parameters . The forward prediction model has the same structure as the forward prediction model in the embodiment of FIG. 8 . The details are not repeated here. For details, refer to the description of the embodiment of FIG. 8 of the present invention. Alternatively, the determined target structure parameters can be used to perform simulation simulation through Lumerical FDTD Solutions, a software used for simulation design of photonics products, to obtain electromagnetic responses corresponding to the target structure parameters. Then, a set of second predicted electromagnetic response data corresponding to the electromagnetic response is acquired based on MC sampling (Monte Carlo, sampling for generating random numbers of a specified distribution).

S1105 , based on the second predicted electromagnetic response data, obtain a first transmission spectrum corresponding to the target structure parameter.

S1106: Adjust the first transmission spectrum to obtain an optimized transmission spectrum, and use the electromagnetic response data in the optimized transmission spectrum as the optimized electromagnetic response data.

S1107: Input the optimized electromagnetic response data into the pre-trained reverse calculation model to obtain optimized structural parameters, and determine the optimized structural parameters as target structural parameters.

For example, after obtaining the target structural parameters, it is necessary to upgrade the performance of the nanodevice: optimize the transmittance at the wavelength of 900nm. Referring to FIG. 12 , the first transmission spectrum corresponding to the target structure parameter can be obtained based on the second predicted electromagnetic response data obtained from the target structure parameter. The optimized transmission spectrum was obtained by shifting the transmittance at the wavelength of 900 nm of the first transmission spectrum from 0.05 to 0.75. By inputting the electromagnetic in the translated transmission spectrum into the pre-trained inverse calculation model, the optimized structural parameters can be obtained, for example, l ₁ =486nm, l ₂ =550nm, l ₃ =608nm, w ₁ = 89.7 nm, w ₂ =96 nm, w ₃ =94 nm, g ₁ =290 nm, g ₂ =341 nm. After the optimized structural parameters are determined as the target structural parameters, the surface plasmon waveguide system fabricated based on the new target structural parameters is a device with improved performance.

In practical applications, a surface plasmon waveguide system can be fabricated based on the target structural parameters of the surface plasmon waveguide system obtained in any of the above embodiments, but the device structure error caused by factors such as temperature, equipment accuracy, and material in device fabrication unavoidable. When using the surface plasmon waveguide system, it is necessary to refer to the structural error to determine the actual parameters and performance of the surface plasmon waveguide system. Therefore, after the surface plasmon waveguide system is fabricated, it is necessary to determine the structural errors of the device.

To this end, optionally, after S203 in the embodiment shown in FIG. 2 of the present invention, the method for determining the structural parameters of the surface plasmon waveguide system of the present invention may further include:

Based on the target structural parameters, a surface plasmon waveguide system is fabricated.

Collect the measured electromagnetic response data of the fabricated surface plasmon waveguide system.

Input the measured electromagnetic response data into the pre-trained reverse calculation model to obtain the measured structural parameters.

Since the pre-trained inverse calculation model establishes the mapping relationship between the electromagnetic response data and the structural parameters, the fabricated surface plasmon waveguide system can be obtained by inputting the measured electromagnetic response data into the pre-trained inverse calculation model. the measured structural parameters.

Based on the measured structural parameters and target structural parameters, the structural errors of the fabricated surface plasmon waveguide system are determined.

Specifically, based on the measured structural parameters and the target structural parameters, the square root error formula or the root mean square error formula can be used to calculate the structural error of the fabricated surface plasmon waveguide system.

Corresponding to the above method embodiments, an embodiment of the present invention further provides an apparatus for determining structural parameters of a surface plasmon waveguide system.

As shown in FIG. 13 , the structure of the apparatus for determining the structural parameters of the surface plasmon waveguide system according to an embodiment of the present invention may include:

An obtaining module 1301 is used to obtain the structure type of the surface plasmon waveguide system; based on the structure type, obtain target electromagnetic response data for the surface plasmon waveguide system of the structure type; wherein, the target electromagnetic response data is used for Indicate the target device performance of the surface plasmon waveguide system;

The target structure parameter determination module 1302 is used to input the target electromagnetic response data into the pre-trained inverse calculation model corresponding to the structure type to obtain the target structure parameter corresponding to the structure type;

Wherein, the reverse calculation model corresponding to the structure type is: pre-determined multiple sample structure parameters corresponding to the structure type and multiple sample electromagnetic response data corresponding to each sample structure parameter in advance. The trained neural network model.

The device for determining the structure parameters of the surface plasmon waveguide system provided by the embodiment of the present invention is compared with the traditional method for determining the target structure parameters that requires multiple manual participation. Therefore, when determining the target structure parameters, the target electromagnetic response data is input into the pre-trained reverse calculation model, and the target structure parameters that can achieve the target electromagnetic response data can be obtained. , the determination of the target structure parameters does not require manual adjustment of the structure parameters and multiple simulations, thereby simplifying the process of determining the target structure parameters.

Corresponding to the above embodiments, an embodiment of the present invention further provides an electronic device, as shown in FIG. 12 , the device may include:

A processor 1201, a communication interface 1202, a memory 1203 and a communication bus 1204, wherein the processor 1201, the communication interface 1202, and the memory communicate with each other through the communication bus 1204 through 1203;

The memory 1203 is used to store computer programs;

The processor 1201 is configured to implement the steps of the method for determining the structural parameters of the surface plasmon waveguide system in any of the foregoing embodiments when executing the computer program stored in the memory 1203.

The electronic device provided by the embodiment of the present invention, compared with the traditional method of determining target structure parameters that requires multiple manual participation, establishes the relationship between the structure parameters and the electromagnetic response data by using the reverse calculation model obtained by pre-training. Therefore, when determining the target structure parameters, the target electromagnetic response data is input into the pre-trained reverse calculation model, and the target structure parameters that can achieve the target electromagnetic response data can be obtained. The determination of the target structure parameters does not need to go through many times Manual adjustment of structural parameters and multiple simulations improve the efficiency of determining target structural parameters.

The above-mentioned memory may include RAM (Random Access Memory, random access memory), and may also include NVM (Non-Volatile Memory, non-volatile memory), for example, at least one disk memory. Optionally, the memory may also be at least one storage device located away from the above-mentioned processor.

The above-mentioned processor may be a general-purpose processor, including a CPU (Central Processing Unit, central processing unit), NP (Network Processor, network processor), etc.; it may also be a DSP (Digital Signal Processor, digital signal processor), an ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array, Field Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

A computer-readable storage medium provided by an embodiment of the present invention is included in an electronic device, and a computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, any one of the above-mentioned embodiments can be implemented. Steps of a method for determining structural parameters of a meta-waveguide system.

A computer-readable storage medium provided by an embodiment of the present invention is included in an electronic device, and a computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, it is different from a traditional target that requires multiple manual participation. Compared with the method of determining the structural parameters, the mapping relationship between the structural parameters and the electromagnetic response data is established by using the reverse calculation model obtained by pre-training. Therefore, when determining the target structural parameters, the target electromagnetic response data is input into the pre-trained The target structure parameters that can achieve the target electromagnetic response data can be obtained by using the reverse calculation model of the model, and the determination of the target structure parameters does not require manual adjustment of the structure parameters and multiple simulations, thereby improving the determination efficiency of the target structure parameters.

In yet another embodiment provided by the present invention, there is also provided a computer program product including instructions, which, when run on a computer, enables the computer to execute the structure of the surface plasmon waveguide system in any of the above embodiments Parameter determination method.

In the above-mentioned embodiments, it may be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented in software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer may be a general purpose computer, special purpose computer, computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions may be downloaded from a website site, computer, server or data center Transmission to another website site, computer, server or data center by means of wired (such as coaxial cable, optical fiber, DSL (Digital Subscriber Line, digital subscriber line) or wireless (such as: infrared, radio, microwave, etc.). A computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The available media can be magnetic media, (eg, floppy disk, hard disk, etc. , magnetic tape), optical media (eg: DVD (Digital Versatile Disc, digital versatile disc)), or semiconductor media (eg: SSD (Solid StateDisk, solid-state hard disk)) and the like.

In this document, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such existence between these entities or operations. The actual relationship or sequence. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device that includes a list of elements includes not only those elements, but also includes not explicitly listed or other elements inherent to such a process, method, article or apparatus. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in a process, method, article or apparatus that includes the element.

Each embodiment in this specification is described in a related manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the apparatus and electronic device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, refer to the partial descriptions of the method embodiments.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims (9)

1. A method for determining structural parameters of a surface plasmon waveguide system, wherein the method comprises: Obtain the structure type of the surface plasmon waveguide system; Based on the structure type, obtain target electromagnetic response data for the surface plasmon waveguide system of the structure type; wherein, the target electromagnetic response data is used to indicate the target device performance of the surface plasmon waveguide system ; Inputting the target electromagnetic response data into the pre-trained inverse calculation model corresponding to the structure type, to obtain the target structure parameter corresponding to the structure type; wherein, the inverse calculation model corresponding to the structure type is: A neural network model obtained by training in advance using the preset multiple sample structure parameters corresponding to the structure type and the obtained multiple sample electromagnetic response data corresponding to each sample structure parameter one-to-one; acquiring second predicted electromagnetic response data corresponding to the target structure parameter; obtaining a first transmission spectrum corresponding to the target structure parameter based on the second predicted electromagnetic response data; Adjusting the first transmission spectrum to obtain an optimized transmission spectrum, and using the electromagnetic response data in the optimized transmission spectrum as the optimized electromagnetic response data; The optimized electromagnetic response data is input into the pre-trained reverse calculation model to obtain optimized structural parameters, and the optimized structural parameters are determined as target structural parameters. 2. The method of claim 1, wherein the surface plasmon waveguide system has a plurality of structural parameters; The multiple sample structure parameters are set using the following steps: Acquiring multiple preset initial structural parameters corresponding to the multiple structural parameters, the preset parameter adjustment accuracy and the preset parameter variation range; For each preset initial structure parameter, adjust the precision according to the preset parameter, and within the preset parameter variation range, determine at least one parameter change value corresponding to the preset initial structure parameter; determining at least one sample structure parameter corresponding to the preset initial structure parameter based on the preset initial structure parameter and the corresponding at least one parameter change value; The multiple sample electromagnetic response data corresponding to each sample structure parameter one-to-one is obtained by adopting the following steps: Using the determined at least one sample structure parameter corresponding to each preset initial structure parameter, a plurality of sample electromagnetic response data corresponding to each sample structure parameter one-to-one is obtained by sampling. 3 . The method according to claim 1 , wherein the process of performing training using a plurality of preset sample structure parameters and a plurality of sample electromagnetic response data corresponding to each sample structure parameter one-to-one in advance. 4 . ,include: Input the sample electromagnetic response data corresponding to each sample structure parameter into the current neural network model respectively, and obtain the predicted structure parameter corresponding to each sample electromagnetic response data; the current neural network model is a preset initial neural network when it is used for the first time Model; Judging whether the current neural network model converges according to the obtained plurality of predicted structural parameters and a first preset cost function, where the first preset cost function is a function set based on sample structural parameters; If it converges, the current neural network model is determined as a reverse calculation model; If it does not converge, the preset gradient function is used, and the stochastic gradient descent method is used to adjust the model parameters of the current neural network model to obtain a new neural network model; updating the current neural network model to the new neural network model obtained; Returns the step of inputting the sample electromagnetic response data corresponding to each sample structure parameter into the current neural network model respectively. 4. The method according to claim 3, characterized in that, after the current neural network model converges, before the current neural network model is determined as a reverse calculation model, the method further comprises: save the obtained multiple predicted structure parameters; judging whether the degree of fit of the obtained plurality of predicted structural parameters and the corresponding plurality of sample structural parameters is less than a preset fitting degree threshold; If so, adjust the physical structure of the current neural network model to obtain a new neural network model, and return to the step of updating the current neural network model to the obtained new neural network model; If not, the current neural network model is determined to be a reverse calculation model. 5. method according to claim 3, is characterized in that, when current neural network model does not converge, utilize preset gradient function, adjust the model parameter of described current neural network model, before obtaining new neural network model, The method also includes: save the obtained multiple predicted structure parameters; Inputting the stored multiple predicted structural parameters into the pre-trained forward prediction model, respectively, to obtain a plurality of first predicted electromagnetic response data corresponding to each predicted structural parameter one-to-one; the forward prediction model is pre-used The set multiple sample structure parameters and the neural network model obtained by training the multiple sample electromagnetic response data corresponding to each sample structure parameter one-to-one; determining whether the current neural network model converges according to the obtained plurality of first predicted electromagnetic response data and a second preset cost function, where the second preset cost function is a function set based on the sample electromagnetic response data; If so, determining the current neural network model as a reverse calculation model; If not, the step of using the preset gradient function to adjust the model parameters of the current neural network model to obtain a new neural network model is performed. 6. The method according to claim 1, characterized in that, after inputting the target electromagnetic response data into a pre-trained reverse calculation model to obtain target structural parameters, the method further comprises: Based on the target structural parameters, a surface plasmon waveguide system is fabricated; Collect the measured electromagnetic response data of the fabricated surface plasmon waveguide system; Inputting the measured electromagnetic response data into the pre-trained reverse calculation model to obtain measured structural parameters; Based on the measured structural parameters and the target structural parameters, the structural error of the fabricated surface plasmon waveguide system is determined. 7. A device for determining structural parameters of a surface plasmon waveguide system, wherein the device comprises: an obtaining module, configured to obtain the structure type of the surface plasmon waveguide system; based on the structure type, obtain target electromagnetic response data of the surface plasmon waveguide system for the structure type; wherein, the target electromagnetic response The data is used to indicate the target device performance of the surface plasmon waveguide system; The target structure parameter determination module is used to input the target electromagnetic response data into the pre-trained reverse calculation model corresponding to the structure type to obtain the target structure parameter corresponding to the structure type; obtain the target structure parameter corresponding to the target structure based on the second predicted electromagnetic response data; obtain the first transmission spectrum corresponding to the target structure parameter based on the second predicted electromagnetic response data; adjust the first transmission spectrum to obtain an optimized transmission spectrum, and put the optimized transmission spectrum in the The optimized electromagnetic response data is used as the optimized electromagnetic response data; the optimized electromagnetic response data is input into the pre-trained reverse calculation model to obtain the optimized structural parameters, and the optimized structural parameters are determined as the target structural parameters ; Wherein, the reverse calculation model corresponding to the structure type is: a plurality of sample structure parameters corresponding to the structure type and a plurality of samples corresponding to each sample structure parameter one-to-one are used in advance. The neural network model obtained by training the electromagnetic response data. 8. An electronic device, characterized in that it comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface, and the memory complete mutual communication through the bus; the memory is used to store a computer program; the processor , which is used to execute the program stored in the memory to realize the method steps according to any one of claims 1-6. 9 . A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method steps of any one of claims 1-6 are implemented. 10 .

Images