Design Method and System of a Hot Electron Detector Based on a Lightweight Gradient Boosting Machine

Abstract

The present invention discloses a design method and system of a hot electron detector based on a lightweight gradient boosting machine, relating to the technical field of device design. The key points of its technical solution are as follows: Based on the initial sample data of a layered nanophotonic device, a lightweight gradient boosting machine model is trained and constructed, and each sample data includes geometric parameters and the corresponding reflection spectrum; the target reflection spectrum is input into the pre-constructed lightweight gradient boosting machine model to obtain the predicted geometric parameters. The present invention uses a method of using a lightweight gradient boosting machine to fit the mapping relationship between the structural parameters and optical response of a hot electron detector and then performing inverse design as needed; this method can learn the functional relationship between the structural parameters and optical response of a Tamm state plasmonic hot electron detector, and optimize the working performance of the device through prediction and design as needed; the lightweight gradient boosting machine can perform preprocessing on the data through its own feature importance algorithm, and select the data with effective contribution.

Description

Technical Field

[0001] The present invention relates to the technical field of device design, and more specifically, it relates to a design method and system for a hot electron detector based on a lightweight gradient boosting machine. Background Art

[0002] As an important branch of machine learning, deep learning has changed the development directions of many fields and industries. The design problem of neural networks usually involves establishing a neural network with a functional relationship between inputs and outputs. However, a simple neural network cannot converge when facing samples with a one-to-many mapping relationship. Nano-photonic devices are a type of device with broad application prospects, and their performance and characteristics are affected by their structural parameters. Among them, hot electron detectors provide research ideas for optical communication due to their low development cost, high sensitivity, and easy integration.

[0003] In traditional design methods, the arrangement and combination of data are usually verified through finite element analysis. However, this method has problems such as the occupation of computing resources and a large amount of time consumption, and it is impossible to traverse all possible parameter combinations. In recent years, remarkable progress has been made in using deep learning algorithms to assist in the design of nano-photonic devices, greatly improving the design efficiency. However, the current deep learning design is a multi-to-multi prediction method, and the obtained objective function is a fitting function with the overall as the target, which makes it difficult to meet the accuracy of data in some regions. In addition, the current deep learning generally directly applies data without preprocessing the original data and cannot extract effective features, which may lead to a decrease in overall accuracy.

[0004] Therefore, how to research and design an optical device design method for a Tamm state plasmonic hot electron detector that can overcome the above defects is an urgent problem for us to solve at present. Summary of the Invention

[0005] To solve the deficiencies in the prior art, the purpose of the present invention is to provide a design method and system for a hot electron detector based on a lightweight gradient boosting machine, which uses the lightweight gradient boosting machine to fit the mapping relationship between the structural parameters and optical response of the hot electron detector and then reverse design as needed; this method can learn the functional relationship between the structural parameters and optical response of the Tamm state plasmonic hot electron detector, and optimize the working performance of the device through prediction and design as needed; the lightweight gradient boosting machine can preprocess data through its own feature importance algorithm and select data with effective contributions.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] In the first aspect, a design method of a hot electron detector based on a lightweight gradient boosting machine is provided, including the following steps:

[0008] Train and construct a lightweight gradient boosting machine model based on the initial sample data of the layered nanophotonic device. Each sample data includes geometric parameters and the corresponding reflection spectrum.

[0009] Input the target reflection spectrum into the pre-constructed lightweight gradient boosting machine model to obtain the predicted geometric parameters.

[0010] Furthermore, the lightweight gradient boosting machine model includes:

[0011] A forward prediction network, trained and constructed with the geometric parameters in the initial sample data as the input and the reflection spectrum in the initial sample data as the output.

[0012] An inverse design network, trained and constructed with the reflection spectrum in the effective sample data as the input and the geometric parameters in the effective sample data as the output.

[0013] Among them, the effective sample data is specifically:

[0014] Calculate the contribution value of the eigenvalue data in the initial sample data through the feature importance method, and screen out the effective sample data from the initial sample data through the contribution value.

[0015] Furthermore, the process of obtaining the effective sample data is specifically:

[0016] Calculate the contribution value of the reflection rate points corresponding to the reflection spectrum and geometric parameters through the feature importance method of the lightweight gradient boosting machine.

[0017] If the contribution value of the reflection rate point is greater than 1, the reflection spectrum and geometric parameters corresponding to the corresponding reflection rate point are valid data, and all the valid data constitutes the effective sample data.

[0018] Furthermore, the process of obtaining the initial sample data is specifically:

[0019] Use the finite-difference time-domain method to construct the nanostructure of the layered Tamm state plasma to obtain geometric parameters.

[0020] And, obtain the corresponding reflection spectrum by characterizing the optical response of the nanostructure.

[0021] Furthermore, during the training process of the lightweight gradient boosting machine model, the learning rate is gradually reduced, and the depth of the tree is increased or the number of leaf nodes is adjusted.

[0022] Furthermore, the lightweight gradient boosting machine model uses a many-to-one cyclic prediction method to achieve cyclic prediction of each geometric parameter.

[0023] Among them, each geometric parameter corresponds to an objective function.

[0024] Furthermore, the predicted geometric parameters include the number of distributed Bragg reflector layers and the thicknesses of the layers corresponding to gold, titanium dioxide, and silicon dioxide.

[0025] In a second aspect, a hot electron detector design system based on a lightweight gradient boosting machine is provided, including:

[0026] A model training module for training and constructing a lightweight gradient boosting machine model based on the initial sample data of the layered nanophotonic device, where each sample data includes geometric parameters and the corresponding reflection spectrum;

[0027] A parameter prediction module for inputting the target reflection spectrum into the pre-constructed lightweight gradient boosting machine model to obtain the predicted geometric parameters.

[0028] In a third aspect, a computer terminal is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for designing a hot electron detector based on a lightweight gradient boosting machine as described in any item of the first aspect is implemented.

[0029] In a fourth aspect, a computer-readable medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for designing a hot electron detector based on a lightweight gradient boosting machine as described in any item of the first aspect can be implemented.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. The method for designing a hot electron detector based on a lightweight gradient boosting machine provided by the present invention uses the lightweight gradient boosting machine to fit the mapping relationship between the structural parameters and the optical response of the hot electron detector and then performs reverse design as needed; this method can learn the functional relationship between the structural parameters and the optical response of the Tamm state plasmon hot electron detector, and optimize the working performance of the device through prediction and design as needed; the lightweight gradient boosting machine can perform preprocessing on the data through its own feature importance algorithm and select the data with effective contribution.

[0032] 2. The present invention realizes a one-to-many prediction method, establishes an objective function for each predicted eigenvalue, specifically solves the problem of computing resource consumption in traditional designs, and this method can also explore more possible combinations of structural parameters, improving the efficiency of accurately designing a hot electron detector as needed. This method provides ideas for the research of Tamm state plasmon hot electron detectors. Description of the Drawings

[0033] The accompanying drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0034] Figure 1 It is an exemplary schematic diagram of the Tamm state plasma hot electron detector in Embodiment 1 of the present invention;

[0035] Figure 2 It is a side view and set parameters of the sensor for detecting polarized light in Embodiment 1 of the present invention;

[0036] Figure 3 It is the design architecture and flowchart of the lightweight gradient boosting machine in Embodiment 1 of the present invention;

[0037] Figure 4 It is the data feature effective processing flowchart of the lightweight gradient boosting machine in Embodiment 1 of the present invention;

[0038] Figure 5 It is a schematic diagram of the importance score of data features of the reflection point in Embodiment 1 of the present invention;

[0039] Figure 6 It is the flowchart of using multi-to-one loop prediction in Embodiment 1 of the present invention;

[0040] Figure 7 It is an effect display diagram of constructing the parameter mapping of the nanophotonic device by the forward prediction network in Embodiment 1 of the present invention;

[0041] Figure 8 It is an effect display diagram of constructing the parameter mapping of the nanophotonic device by the inverse design network in Embodiment 1 of the present invention;

[0042] Figure 9 It is an effect display diagram of the problem of designing a hot electron detector as needed in the inverse design in Embodiment 1 of the present invention;

[0043] Figure 10 It is a detailed flowchart of lightweight design-on-demand and verification in Embodiment 1 of the present invention;

[0044] Figure 11 It is the detailed information of designing a hot electron detector as needed in Embodiment 1 of the present invention;

[0045] Figure 12 It is the system block diagram in Embodiment 2 of the present invention. Detailed implementation manners

[0046] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings. The illustrative embodiments and descriptions thereof of the present invention are only used to explain the present invention and do not limit the present invention.

[0047] Embodiment 1: A design method for a hot electron detector based on a lightweight gradient boosting machine, comprising the following steps:

[0048] Step 1: Train and construct a lightweight gradient boosting machine model based on the initial sample data of the layered nanophotonic device. Each sample data includes geometric parameters and the corresponding reflection spectrum;

[0049] Step 2: Input the target reflection spectrum into the pre-constructed lightweight gradient boosting machine model to obtain the predicted geometric parameters.

[0050] In Step 1, the lightweight gradient boosting machine model includes a forward prediction network and an inverse design network. The forward prediction network is trained and constructed with the geometric parameters in the initial sample data as the input and the reflection spectrum in the initial sample data as the output; the inverse design network is trained and constructed with the reflection spectrum in the effective sample data as the input and the geometric parameters in the effective sample data as the output.

[0051] Specifically, (1) Use the finite-difference time-domain method to characterize the optical response of the 4- and 5-layered Tamm state plasmon. The scanning parameters are the thickness of the gold layer (90 - 100 nm, 9 times), the thickness of the silica layer (60 - 70 nm, 11 times), the thickness of the titanium dioxide layer (70 - 80 nm, 11 times), and the number of layers (4, 5). A total of 2178 initial sample data are obtained, and the reflection spectrum consists of 253 reflection points. Divide the data set, with the division ratio being 80% for the training set and 20% for the test set. Use the training set data to train the forward prediction network. Obtain the mapping relationship between the geometric parameters and the optical response for this design.

[0052] (2) Use the divided training set data to train the inverse design network. The inverse design network includes an input: optical response; decision tree; parallel optimization; output of the predicted geometric parameters. The optical response input into the inverse design network is traversed through the data. The discretized values are used as indices to accumulate statistics in the histogram. When the data is traversed once, the histogram accumulates the required statistics, and then based on the discrete values of the histogram, the optimal splitting point is searched for by traversing.

[0053] In data preprocessing, calculate the contribution values of the reflection spectrum (2178×253) and the geometric parameters (2178×4) through the feature importance method of the lightweight gradient boosting machine. The contribution value greater than 1.0 is defined as valid data. The valid data set is (2178×201×4). Map the relationship of the data obtained through training, and finally output the predicted value.

[0054] (3) In the on-demand design stage, the predicted geometric parameters are calculated by the finite element difference method to obtain the optical response data, which verifies the prediction performance of the model. By comparing the calculated optical response with the target optical response of the on-demand design, the design performance of the model is determined, providing a basis for the actual fabrication of the sample. Design a hot electron detector with an absorption rate greater than 70% (reflectivity less than 30%) at 660 nm. Input a 201×1 reflection spectrum into the inverse-designed lightweight gradient boosting machine model. The output geometric parameters are 100 nm for gold, 77 nm for titanium dioxide, 60 nm for silicon dioxide, and 5 layers for the distributed Bragg reflector. Calculate the reflection spectrum data of these parameters by finite element difference and compare it with the on-demand design reflection diagram. The absorption peak (reflection) is at 660 nm. This shows that the model can explore as much as possible according to the target optical response to obtain the design geometric parameters, thereby realizing the inverse design of the Tamm state plasmon hot electron detector.

[0055] For example, the device designed by the present invention selects a silicon wafer as the research substrate and metal gold as the main material of the absorber. It is a metal with a high melting point and stability, and it has good plasmonic properties. Based on thermoplasmonics, zero transmittance is guaranteed. The distributed Bragg reflector is composed of silicon dioxide and titanium dioxide. The gold backplane and the distributed Bragg reflector can form a Tamm state plasmon. As Figure 1 shown, the Tamm state plasmon hot electron detector provided by the present invention includes a substrate layer, a gold layer, and an alternating TiO2 / SiO2 layer. The light source is required to be a plane wave, vertically incident on the unit structure from the top, and the wavelength range of the light source is 600 nm - 1000 nm. The distance between the light source and the device is greater than one wavelength. The boundary conditions of the simulation interval are periodic boundary conditions in both the x and y directions, and a perfectly matched layer in the z direction. A reflectivity monitor is set above the light source to obtain the reflectivity distribution between 500 nm and 1000 nm and map it to 201 points for output. The thickness of each layer is between 100 nm and 300 nm, and the other areas are reserved as air. Since the substrate is a silicon wafer, the transmittance is almost 0, and there is no need to set a transmittance monitor. Therefore, the absorption rate value A = 1 - R - T, and T = 0. As Figure 2 shown, its side view shows the parameters and materials.

[0056] After the required simulation conditions are set, the geometric parameter ranges are combined to form an initial data set or arranged and combined according to certain rules to ensure a reasonable coverage range. In the specific implementation of the present invention, there are 11×11×9 = 1089 tag values of initial sample data. The reflection spectra corresponding to their geometric parameters are obtained through numerical simulation by the finite element difference method. The initial sample data is composed of geometric parameters and the corresponding reflection spectra. Then, four layers are selected for value taking, and numerical simulation is also carried out through the finite element difference method in the time domain to obtain a data set composed of 1089 sets of structural parameters and optical responses as the test and verification data set, and the total data set is 2178. In order to enable the model to better learn the mapping relationship between the input data X and Y, it is necessary to normalize the tag values. The linear normalization method is adopted in this specific implementation, that is:

[0057] As Figure 3 shown, during the training process, the lightweight gradient boosting machine model will calculate the loss value according to the difference between the current model prediction value and the true label in each iteration. Specifically, the model uses the mean square error as the loss function to measure the prediction error of the current model. This process calculates the loss for each sample during each training iteration and updates the model parameters through the gradient boosting algorithm.

[0058] In the multi-label regression problem, if the model outputs multiple sets of prediction parameters, each set of predictions will be compared with the corresponding tag values to calculate the loss value. Assuming there are multiple labels, the loss value will be calculated separately for each set of predictions, and the smallest loss value will be taken as the current optimal loss value for updating. This helps the model avoid error accumulation caused by the complexity of the labels. Set the loss value, and determine the smallest loss value by calculating the mean square error between each set of structural parameters and the labels. The role of this step is to ensure that when facing multiple labels, the model can match each label one by one and minimize the overall error to the greatest extent.

[0059] To better obtain the optimal data, the lightweight gradient boosting machine model adopts the Figure 4 feature importance method to measure the contribution of each absorption point feature in all decision trees. If the feature score is greater than 1.0, it means that the feature is contributive, and it can effectively extract data in the forward prediction and provide input data for the inverse design. In actual tests, as Figure 5 shown. Figure 5 (A) shows the contribution values of 253 reflection points, Figure 5 (B) is sorted from high to low. As a result, 52 reflection points with importance lower than 1.0 are screened out.

[0060] As Figure 6As shown in the figure, to solve the problem of large prediction errors in the many-to-many model in traditional designs, the lightweight gradient boosting machine is optimized. After optimization, the model uses a many-to-one cyclic prediction method to achieve cyclic prediction of each parameter and constructs multiple objective functions. This can avoid errors in some regions caused by overall prediction and improve the accuracy of prediction.

[0061] During the training process, to accelerate the convergence of the model and avoid overfitting, the lightweight gradient boosting machine model will dynamically adjust the learning rate and regularization term. At the beginning of training, a relatively large learning rate (e.g., 0.1) can be set to quickly explore the optimal solution. As training progresses, gradually reduce the learning rate and increase the depth of the tree or adjust the number of leaf nodes to further improve the accuracy of the model. Eventually, the model should reach a small loss value and show high prediction accuracy on the test set. Parameter adjustment during the training process: During the training process, some key hyperparameters of the model need to be adjusted: Learning rate: The learning rate controls the step size of model parameter updates in each iteration. It can be initially set to 0.1 and then adjusted gradually. Number of trees: The complexity of the model can be controlled by setting the number of iterations or the number of trees. It can be initially set to 100 and then adjusted according to the change of the loss value. Maximum depth of the tree: The depth of the tree controls the complexity of each tree. It can be initially set to 6 and adjusted according to the performance during the training process. Number of leaf nodes: The number of leaf nodes affects the capacity of the model. Increasing the number of leaf nodes can improve the expressive ability of the model but also increase the risk of overfitting. Regularization parameters: These parameters control the regularization effect of the model and avoid overfitting. By adjusting these hyperparameters and combining the feedback of the loss value during the training process, the model can converge to a better prediction result. During the training process, the model continuously adjusts the structure of the decision tree to minimize the error between the output prediction result and the true label. Eventually, the model will make predictions according to the learned patterns and output the final prediction results on the test set or actual data. These prediction results can be further used for various tasks such as optical response and structural parameter prediction.

[0062] After training the forward prediction network, as Figure 7 shown, it demonstrates the prediction effect of the forward prediction network of the present invention, ensuring the accuracy and high generalization ability of the forward prediction network. Use the forward prediction network to train the inverse design network according to the algorithm block diagram and design the network according to the requirements of this specific implementation example. As Figure 8 shown, for the fitting coefficients of each reflection point of the absorption spectrum, the fitting coefficients are all greater than 0.95. In this specific implementation example, Figure 9 it shows four groups of geometric parameters output by the inverse design network. It can be seen that the error between the predicted parameters and the actual parameters is small (accuracy: 0.99). Figure 10Flowchart of the on-demand designed device. For example: When designing a hot electron detector with an absorption at 660 nm, the corresponding optical response spectrum is input into the lightweight gradient boosting machine. At this time, the model has been trained previously. Obtain the output geometric parameters of the model, and fabricate the device according to the obtained geometric parameters. Before fabricating the device, the optical response can be calculated by the finite element difference method and compared with the on-demand optical response. The geometric parameter values obtained by on-demand design are a gold thickness of 100 nm, a titanium dioxide thickness of 77 nm, a silica thickness of 60 nm, and the number of distributed Bragg reflector layers designed is 5 layers. Verify the accuracy of the model design by using the obtained geometric parameters for finite difference time domain method calculation and comparing the calculated results with the input. The absorption spectrum is calculated as A = 1 - R - T, and T is 0.

[0063] As Figure 11 shown, it includes the input reflection spectrum data, the predicted geometric parameters, and the absorption spectrum calculated by simulation. It is not difficult to see from the results that the lightweight gradient boosting machine is beneficial to the auxiliary design of the Tamm state plasmon hot electron detector and can achieve on-demand design according to the required target optical response, which has certain enlightenment significance. In today's rapid development of nanophotonic devices and the growing demand for high-performance device design, the design method of the present invention has extremely strong supporting power and reference value for the extensive development of future nanophotonic device design.

[0064] As one of the machine learning methods, ensemble learning improves the overall prediction performance by combining the prediction results of multiple models and constructs a more powerful ensemble model by combining multiple weak learners. The advantage of ensemble learning is that it can reduce overfitting, improve the generalization ability of the model, and usually has better performance than a single model. In addition, ensemble learning can also use parallel computing to accelerate the model training process. Compared with neural networks, the lightweight gradient boosting machine can only achieve many-to-one prediction, which exactly conforms to the design of the Tamm state plasmon hot electron detector because its absorption and reflection peaks are single and the geometric parameters also exist independently. In addition, the lightweight gradient boosting machine has the characteristics of fast training and not being easily overfitted, and can effectively solve the problems in traditional design.

[0065] Example 2: A hot electron detector design system based on a lightweight gradient boosting machine, which is used to implement the hot electron detector design method based on a lightweight gradient boosting machine as described in Example 1, as Figure 12 shown, including a model training module and a parameter prediction module.

[0066] Among them, the model training module is used to train and construct a lightweight gradient boosting machine model based on the initial sample data of the layered nanophotonic device. Each sample data includes geometric parameters and the corresponding reflection spectrum. The parameter prediction module is used to input the target reflection spectrum into the pre-constructed lightweight gradient boosting machine model to obtain the predicted geometric parameters.

[0067] The present invention also describes a computer terminal, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the design method of the hot electron detector based on the lightweight gradient boosting machine as described in Embodiment 1.

[0068] The present invention also describes a computer-readable medium, on which a computer program is stored. When the computer program is executed by a processor, it can implement the design method of the hot electron detector based on the lightweight gradient boosting machine as described in Embodiment 1.

[0069] Working principle: The present invention uses a method of fitting the mapping relationship between the structural parameters and optical response of a hot electron detector by a lightweight gradient boosting machine and then performing inverse design as needed. This method can learn the functional relationship between the structural parameters and optical response of a Tamm state plasmonic hot electron detector, and optimize the working performance of the device through prediction and design as needed. The lightweight gradient boosting machine can perform preprocessing on the data through its own feature importance algorithm and select the data with effective contributions.

[0070] In addition, the present invention realizes a one-to-many prediction method and establishes an objective function for each predicted feature value, specifically solving the problem of computational resource consumption in traditional designs. This method can also explore more possible combinations of structural parameters, improving the efficiency of accurately designing a hot electron detector as needed. This method provides ideas for the research of Tamm state plasmonic hot electron detectors.

[0071] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0072] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 and / or means for implementing the functions specified in one or more of the blocks.

[0073] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 and / or means for implementing the functions specified in one or more of the blocks.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 and / or means for implementing the functions specified in one or more of the blocks.

[0075] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Design method of hot electron detector based on lightweight gradient boosting machine, characterized in that, Including the following steps: Training and constructing a lightweight gradient boosting machine model based on the initial sample data of the layered nanophotonic device, where each sample data includes geometric parameters and the corresponding reflection spectrum; Inputting the target reflection spectrum into the pre-constructed lightweight gradient boosting machine model to obtain the predicted geometric parameters; The lightweight gradient boosting machine model includes: a forward prediction network, which is trained and constructed with the geometric parameters in the initial sample data as the input and the reflection spectrum in the initial sample data as the output; a reverse design network, which is trained and constructed with the reflection spectrum in the effective sample data as the input and the geometric parameters in the effective sample data as the output; where the effective sample data is specifically: calculating the contribution value of the eigenvalue data in the initial sample data through the feature importance method, and screening out the effective sample data from the initial sample data through the contribution value.

2. The design method of a hot electron detector based on a lightweight gradient boosting machine according to claim 1, characterized in that The specific process for obtaining the effective sample data is: calculating the contribution value of the reflection rate points corresponding to the reflection spectrum and the geometric parameters through the feature importance method of the lightweight gradient boosting machine; if the contribution value of the reflection rate point is greater than 1, the reflection spectrum and the geometric parameters corresponding to the corresponding reflection rate point are effective data, and all the effective data constitutes the effective sample data.

3. The design method of a hot electron detector based on a lightweight gradient boosting machine according to claim 1, characterized in that The specific process for obtaining the initial sample data is: constructing the nanostructure of the layered Tamm state plasmon by using the finite-difference time-domain method to obtain the geometric parameters; and obtaining the corresponding reflection spectrum by characterizing the optical response of the nanostructure.

4. The design method of a hot electron detector based on a lightweight gradient boosting machine according to claim 1, characterized in that, During the training process of the lightweight gradient boosting machine model, the learning rate is gradually reduced, and the depth of the tree is increased or the number of leaf nodes is adjusted.

5. The design method of a hot electron detector based on a lightweight gradient boosting machine according to claim 1, characterized in that, The lightweight gradient boosting machine model uses a many-to-one cyclic prediction method to achieve cyclic prediction of each geometric parameter; where each geometric parameter corresponds to an objective function.

6. The design method of a hot electron detector based on a lightweight gradient boosting machine according to claim 1, characterized in that, The predicted geometric parameters include the number of layers of the distributed Bragg reflector and the thicknesses of the layers corresponding to gold, titanium dioxide, and silicon dioxide.

7. A hot electron detector design system based on a lightweight gradient boosting machine, characterized in that, Including: A model training module for training and constructing a lightweight gradient boosting machine model based on the initial sample data of the layered nanophotonic device, where each sample data includes geometric parameters and the corresponding reflection spectrum; A parameter prediction module for inputting the target reflection spectrum into the pre-constructed lightweight gradient boosting machine model to obtain the predicted geometric parameters; the lightweight gradient boosting machine model includes: a forward prediction network, which is trained and constructed with the geometric parameters in the initial sample data as the input and the reflection spectrum in the initial sample data as the output; a reverse design network, which is trained and constructed with the reflection spectrum in the effective sample data as the input and the geometric parameters in the effective sample data as the output; where the effective sample data is specifically: calculating the contribution value of the eigenvalue data in the initial sample data through the feature importance method, and screening out the effective sample data from the initial sample data through the contribution value.

8. A computer terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the design method of the hot electron detector based on the lightweight gradient boosting machine according to any one of claims 1-6.

9. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can implement the design method of the hot electron detector based on the lightweight gradient boosting machine according to any one of claims 1-6.

Images