A metamaterial light absorber and its structure prediction method and prediction device
By using metal-die-non-metal three-layer structure and four-quadrant arrangement pattern structure in metamaterial optical absorbers, combining feature extraction of crystalline and amorphous absorption spectrum, the low freedom of microstructure design in the prior art is solved. The problem of degree has achieved high degree of freedom structural design and diverse spectral absorption, which improves absorption rate and modulationability.
Patent Information
- Application Number
- CN202311729559.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-14
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2043-12-14
AI Technical Summary
The low degree of freedom of existing metamaterial optical absorbers in microstructure design limits their modulationability and diversity of absorption spectrum, resulting in limited development of multifunctional absorption and wide fields.
Using a metal-die-non-metal three-layer structure, the dielectric layer is embedded with phase change material layer and the metal layer is embedded with silica layer. A high degree of freedom is constructed through four-quadrant arrangement and image encoding, and based on crystalline and A neural network prediction model for feature extraction of amorphous absorption spectra to predict structural parameters.
The structural diversity and reconstructibility of the metamaterial light absorber are realized, the width and light absorption rate of the spectral absorption band are improved, and the microstructure of complex metamaterials can be accurately predicted, and the design time is shortened.
Smart Images

Figure CN118348620B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of metamaterial light absorbers, and in particular to a metamaterial light absorber and a structure prediction method and a prediction device thereof. Background Art
[0002] Metamaterial light absorbers belong to the field of optical components. They are materials with specially designed structures. Their unique electromagnetic properties break through the physical limits of traditional media and can effectively absorb electromagnetic radiation in frequency bands including visible light, infrared and microwaves. This gives them broad application potential and prospects in the fields of optics, electromagnetic wave control and energy collection.
[0003] At present, the research on metamaterial light absorbers aims to make full use of the microstructure advantages of metamaterials to achieve efficient electromagnetic wave absorption or reflection, improve energy efficiency, reduce electromagnetic interference, improve sensor performance, and make it widely used in solar energy collection, optical stealth and image processing. Its structural parameters significantly affect its absorption characteristics.
[0004] Therefore, designing the microstructure of metamaterials to construct light absorbers has become the most cutting-edge research hotspot in this field.
[0005] Traditional metamaterial light absorbers are designed and constructed based on optimization algorithms, and the optimal structural parameters are found through complex simulations. This method usually requires researchers to have rich experience and takes a lot of time to explore the optimal structural parameters. At present, metamaterial device design methods based on deep learning have become a research trend, which greatly improves the efficiency of design. However, researchers generally use fixed, single structures to obtain data sets in the process of structural design, such as only changing the length and / or width of the microstructure and / or the shape of the graph, and then use the data set obtained by this method to train the neural network, so that the trained neural network can only predict structural parameters containing the same features. This low-degree-of-freedom microstructure design method limits the modulatability of metamaterial light absorbers, reduces the diversity of the absorption spectrum of metamaterial light absorbers, and limits the development of metamaterial light absorbers in multiple functions and wide fields. Summary of the invention
[0006] Based on this, an object of the present invention is to provide a metamaterial light absorber, which has high modulability, structural diversity, a wide spectral absorption band and high light absorption rate.
[0007] The technical solution adopted by the present invention is as follows.
[0008] In a first aspect, the present invention proposes a metamaterial light absorber, comprising a substrate layer, a dielectric layer and an absorption anti-reflection layer arranged in sequence, the dielectric layer comprising a phase change material layer and a first silicon dioxide layer, the phase change material layer is embedded in the first silicon dioxide layer and forms a surface contact with the substrate layer; the absorption anti-reflection layer comprises a metal layer and a second silicon dioxide layer, the metal layer is embedded in the second silicon dioxide layer and forms a surface contact with the first silicon dioxide layer; the phase change material layer has a graphic structure, the graphic structure is evenly divided into four module graphics, the four module graphics are arranged in a four-quadrant manner and the image of each module graphic is mirror-symmetric with the image of the adjacent module graphics; the metal layer and the phase change material layer have the same graphic structure and correspond one to one.
[0009] Furthermore, the module graph is divided into n 2 A coding unit is defined, and a state value of "empty" or "real" is assigned to the coding unit. When the state value of the coding unit is "empty", it indicates that there is no phase change material or metal at the coding unit; when the state value of the coding unit is "real", it indicates that there is a phase change material or metal at the coding unit, wherein n is a coding parameter and n is an integer.
[0010] Furthermore, the encoding parameter n≥2.
[0011] Furthermore, the phase change material layer is germanium antimony selenium tellurium alloy Ge2Sb2Se4Te (GSST).
[0012] Compared with the prior art, the metamaterial light absorber described in the present invention achieves high-degree-of-freedom graphic design by setting four module graphics arranged in a four-quadrant manner in the phase change material layer and the metal layer, and the image of each module graphic is mirror-symmetric with the image of the adjacent module graphic, and constructing the graphic structure of the light absorber unit structure in a way of image encoding the module graphics, so as to achieve high-degree-of-freedom graphic design, so that the metamaterial microstructure design has structural diversity and reconfigurability, and based on this, a large amount of structure-absorption spectrum simulation data can be provided for the subsequent training structure prediction model; at the same time, the mirror-symmetric mapping of the module graphics makes the designed metamaterial light absorber insensitive to light polarization; in addition, the light absorber uses the phase change material Ge2Sb2Se4Te (GSST) with an extremely high quality factor (FOM) as the dielectric layer, and by changing the phase change material GSST from a crystalline state to an amorphous state, the light absorber can achieve a change in the absorption spectrum without changing the structure, thereby achieving a reconfigurable effect.
[0013] In a second aspect, the present invention provides a method for predicting the structure of a metamaterial light absorber, comprising the following steps:
[0014] S10: converting the input crystalline absorption spectrum data into a first vector, and sequentially shrinking and expanding the first vector to output a first three-dimensional matrix;
[0015] S20: converting the input amorphous absorption spectrum data into a second vector, and sequentially shrinking and expanding the second vector to output a second three-dimensional matrix;
[0016] S30: splicing the first three-dimensional matrix and the second three-dimensional matrix, and outputting a spliced matrix;
[0017] S40: extracting features from the concatenated matrix and outputting a feature matrix;
[0018] S50: Expand the feature matrix and output a feature vector;
[0019] S60: performing feature conversion on the feature vector and outputting predicted structural parameters;
[0020] The weights and bias parameters used in executing steps S10, S20, S40, and S60 are corrected and optimized by performing simulation calculations on the light absorber described in any of the above items to obtain the structure-absorption spectrum data set of the phase change material in the crystalline and amorphous states.
[0021] Furthermore, the step S10 includes the following sub-steps:
[0022] S11: converting the input crystalline absorption spectrum data into a first vector;
[0023] S12: After shrinking and expanding the first vector in sequence, a first three-dimensional matrix is output; wherein the shrinking is performed according to the following formula:
[0024]
[0025] Where, e represents the crystalline absorption spectrum, e T represents the transposed matrix of the crystalline absorption spectrum, V1 and b1 represent the weight and bias parameters used in the first dimension transformation, respectively. k is a constant representing the weight The number of
[0026] Furthermore, the step S20 includes the following sub-steps:
[0027] S21: converting the input amorphous absorption spectrum data into a second vector;
[0028] S22: After shrinking and expanding the second vector in sequence, a second three-dimensional matrix is output; wherein the shrinking is performed according to the following formula:
[0029]
[0030] Where, e′ represents the amorphous absorption spectrum, e′ Trepresents the transposed matrix of the amorphous absorption spectrum, V2 and b2 represent the weight and bias parameters used in the second dimension transformation, respectively. k′ is a constant representing the weight The number of
[0031] Compared with the prior art, the structure prediction method described in the present invention adopts a neural network prediction model that simultaneously extracts features from the input crystalline and amorphous absorption spectra to predict structural parameters. The predicted parameter results are accurate and have good prediction effects. It can realize the prediction and construction of the microstructure of complex metamaterial light absorbers with multiple absorption characteristics.
[0032] In a third aspect, the present invention provides a structure prediction device for a metamaterial light absorber, comprising a first data processor, a second data processor, a matrix splicer, a convolutional neural network, a vector tiler, and a fully connected neural network;
[0033] The first data processor is used to convert the input crystalline absorption spectrum data into a first vector, and output a first three-dimensional matrix after shrinking and expanding the first vector in sequence;
[0034] The second data processor is used to convert the input amorphous absorption spectrum data into a second vector, and output a second three-dimensional matrix after shrinking and expanding the second vector in sequence;
[0035] The matrix splicer is used to splice the first three-dimensional matrix and the second three-dimensional matrix and output a spliced matrix;
[0036] The convolutional neural network is used to extract features from the concatenated matrix and output a feature matrix;
[0037] The vector flattener is used to expand the feature matrix and output a feature vector;
[0038] The fully connected neural network is used to perform feature conversion on the feature vector and output predicted structural parameters;
[0039] The weights and bias parameters of the first data processor, the second data processor, the convolutional neural network and the fully connected neural network are simulated and calculated using the light absorber described in any one of the above items to obtain the structure-absorption spectrum data set correction and optimization of the phase change material in the crystalline and amorphous states.
[0040] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is a schematic diagram of a unit structure of an embodiment of a metamaterial light absorber of the present invention;
[0042] Figure 2 A structural prediction model diagram of the metamaterial light absorber of the present invention;
[0043] Figure 3 It is a flow chart of structure prediction of the metamaterial light absorber of the present invention;
[0044] Figure 4 A first actual expected absorption spectrum structure prediction and a predicted structure simulation spectrum absorption diagram of an embodiment of the present invention;
[0045] Figure 5 It is a second actual expected absorption spectrum structure prediction and predicted structure simulation spectrum absorption diagram of one embodiment of the present invention;
[0046] Figure 6 A first ideal expected absorption spectrum structure prediction and a predicted structure simulation spectrum absorption diagram of an embodiment of the present invention;
[0047] Figure 7 This is a second ideal expected absorption spectrum structure prediction and predicted structure simulation spectrum absorption diagram of an embodiment of the present invention. DETAILED DESCRIPTION
[0048] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention.
[0049] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that, unless otherwise stated, "multiple" refers to two or more; "and / or" refers to and includes any or all possible combinations of one or more associated listed items; "first", "second", "third", etc. are only used to distinguish, rather than to describe a specific order or sequence, and cannot be understood as indicating or implying relative importance.
[0050] Existing metamaterial device design methods based on deep learning generally use fixed, single structures in the process of microstructure design, such as only changing the length and / or width and / or shape of the microstructure, and then obtaining a data set of the absorption spectrum corresponding to the microstructure based on this, and using the data set obtained based on this method to train the neural network. Because the existing methods cannot consider the corresponding absorption spectrum that does not exist for a fixed, single specific structure, the trained neural network can only predict the structural parameters containing the same characteristics. This limitation hinders the accurate prediction of the corresponding structural parameters by a specific absorption spectrum. This low-degree-of-freedom, single microstructure design method limits the modulatability of metamaterial light absorbers, reduces the diversity of the absorption spectrum of metamaterial light absorbers, and limits the development of metamaterial light absorbers in multiple functions and wide fields.
[0051] To solve the above problems, the present invention proposes a metamaterial light absorber, which has a three-layer structure of metal-medium-non-metal, wherein the dielectric layer is a silicon dioxide layer with a phase change material layer embedded in it, and the non-metal layer is a silicon dioxide layer with a metal layer embedded in it. By setting four module graphics arranged in a four-quadrant manner in the same manner in the phase change material layer and the metal layer, and the image of each module graphic is mirror-symmetric with the image of the adjacent module graphic, and constructing a graphic structure of the light absorber unit structure in a way of image encoding the module graphics, a high degree of freedom of graphic design is achieved, so that the metamaterial microstructure design has structural diversity and reconfigurability, and based on this, a large amount of structure-absorption spectrum simulation data can be provided for the subsequent training structure prediction model.
[0052] Based on this, a metamaterial light absorber structure prediction method is proposed. The neural network prediction model is used to extract features from the input crystalline and amorphous absorption spectra to predict the structural parameters. The predicted parameter results are accurate and have good prediction effects. It can realize the prediction and construction of the microstructure of complex metamaterial light absorbers with multiple absorption characteristics.
[0053] The unit structure of the metamaterial light absorber and the structure prediction method thereof described in the present invention will be described in detail below.
[0054] (I) Construction of light absorber unit structure
[0055] 1. Physical structure
[0056] See also Figure 1 (a) and Figure 1 (b) The unit structure of the metamaterial light absorber described in the present invention is a three-layer structure of metal-dielectric-non-metal, and the three-layer structure is, from bottom to top, a base layer 10, a dielectric layer 20, and an absorption anti-reflection layer 30. The length and width of the base layer 10 and the dielectric layer 20 are both equal to a, and the length and width of the absorption anti-reflection layer 30 are both equal to b, and a>b.
[0057] The base layer 10 is made of titanium metal, and has a height H1.
[0058] The dielectric layer 20 includes a phase change material layer 21 and a first silicon dioxide layer 22, and the height thereof is H2. The phase change material layer 21 is a germanium antimony selenium tellurium alloy Ge2Sb2Se4Te (GSST), which is embedded in the first silicon dioxide layer 22 and forms a surface contact with the base layer 10, and the height thereof is H4. On the one hand, the quality factor FOM of the selected GSST is large, which is two orders of magnitude higher than the classical phase change alloy germanium antimony tellurium Ge2Sb2Te5 (GST-225), and it has unprecedented broadband optical transparency in almost the entire infrared spectrum; on the other hand, the use of silicon dioxide-coated phase change material GSST as a dielectric layer has a plasma resonance effect.
[0059] The absorption anti-reflection layer 30 includes a metal layer 31 and a second silicon dioxide layer 32, and its height is H3. The metal layer 31 is titanium metal, which is embedded in the second silicon dioxide layer 32 and forms a surface contact with the first silicon dioxide layer 22, and its height is H5. The use of silicon dioxide to coat titanium metal can reduce the reflection effect of the metal layer and increase the absorption rate.
[0060] 2. Graphic structure
[0061] See also Figure 1 (c) The phase-change material layer 21 in the unit structure is divided into four module patterns, namely the first module pattern, the second module pattern, the third module pattern, and the fourth module pattern, and each module pattern is arranged in a four-quadrant manner; and the first module pattern is symmetrical with the second module pattern about the Y axis, the second module pattern is symmetrical with the third module pattern about the X axis, the third module pattern is symmetrical with the fourth module pattern about the Y axis, and the fourth module pattern is symmetrical with the first module pattern about the X axis. Dividing the phase-change material layer 21 into four module patterns and arranging them in a four-quadrant manner to be symmetrical about the X axis and the Y axis can reduce the number of structural parameters that need to be changed and speed up the data acquisition; and this symmetric mode makes the manufactured metamaterial light absorber insensitive to the polarization of light.
[0062] Divide the first module graph into n 2 coding units, n is a coding parameter, n≥2, and n is an integer. The coding unit is assigned a state value of "empty" or "real". When the state value of the coding unit is "empty", it means that there is no phase change material at the coding unit; when the state value of the coding unit is "real", it means that there is a phase change material at the coding unit. 2 Assign values to the coding units to get A graphics state.
[0063] The pattern structure of the metal layer 31 in the unit structure is arranged the same as that of the phase change material layer 21 , and the pattern structures correspond one to one.
[0064] (II) Constructing an absorption spectrum simulation data set based on a unit structure
[0065] The present invention uses FDTD Solutions software to perform simulation design and obtain data sets.
[0066] 1. Unit structure parameter setting
[0067] The coding parameter of the phase change material layer 21 is set to 3, then a single module pattern in the phase change material layer 21 has 9 coding units, the entire phase change material layer 21 has 36 coding units, and the length and width of each coding unit are fixedly set to 5nm. At the same time, the graphic structure of the metal layer 31 is set to be the same as that of the phase change material layer 21, that is, the transformation is performed simultaneously and the transformation pattern is the same.
[0068] The height of the base layer 10 is set to H1=500nm; the height of the first silicon dioxide layer 22 is set to H2=150nm~200nm; the height of the phase change material layer 21 is set to H4=50nm~100nm; the height of the second silicon dioxide layer 32 is set to H3=150nm~200nm; the height of the metal layer 31 is set to H5=50nm~100nm.
[0069] The length and width of the base layer 10 and the dielectric layer 20 are set to a=410 nm, the length and width of the absorption anti-reflection layer 30 are set to b=300 nm, and the other areas are left as air.
[0070] 2. Other parameter settings
[0071] 1) Set up the light source
[0072] The light source is set to a plane wave, the frequency thereof is set to 300nm-2700nm, and the incident direction thereof is set to vertically enter the absorption anti-reflection layer 30 of the unit structure from above downward.
[0073] 2) Set up reflectance and transmittance monitors
[0074] A reflectivity monitor is arranged above the light absorber unit structure and within a distance from the light source; and a transmittance monitor is arranged within a distance from the light source passing through the rear end of the light absorber unit structure, so as to obtain the reflectivity distribution and transmittance distribution of the light source frequency in the range of 300nm-2700nm, calculate the corresponding absorption spectrum according to the absorbance formula, and map it into 201 point outputs.
[0075] 3) Set boundary conditions
[0076] The boundary conditions of the simulation interval are as follows: the X direction is Periodic, the Y direction is Periodic, and the Z direction is PML.
[0077] 3. Get simulation data
[0078] Within the setting range of the height parameter H2 of the first silicon dioxide layer 22, the height parameter H4 of the phase change material layer 21, the height parameter H3 of the second silicon dioxide layer 32 and the height parameter H5 of the metal layer 31, different height parameter combinations are selected according to the arrangement and combination rules. and at the selected height parameter combination Under this condition, each coding unit is randomly assigned a value to obtain graphic structures of various forms; after simulation, the initial data set of structure-absorption spectrum is obtained.
[0079] In specific implementation, within the setting range of each height parameter, a set of data is taken every 50 nm, and the combination of height parameters has 2 4 =16; Under each height parameter combination, the states of 36 coding units are randomly changed, and 2 9 =512 combinations; then under the unit structure parameters of 4 height parameters and 9 coding unit state values, the entire unit structure can obtain 2 13 =8192 initial sample data label values.
[0080] Set the phase change material GSST to a crystalline state, simulate the label value using the time-domain finite element difference method, and obtain a crystalline data set of structure-absorption spectrum in represents the structural parameters of the unit structure, The crystalline absorption spectrum corresponding to the unit structure obtained by the time-domain finite element difference method is shown, and the absorption spectrum mapping output 201 data.
[0081] Set the phase change material GSST to an amorphous state, simulate the label value using the time-domain finite element difference method, and obtain an amorphous data set of structure-absorption spectra The amorphous absorption spectrum corresponding to the unit structure obtained by the time-domain finite element difference method is shown, and the absorption spectrum mapping output 201 data.
[0082] The crystalline data set and the non-crystalline data set were spliced into a total data set.
[0083] (III) Constructing a unit structure prediction model
[0084] See also Figure 2The prediction model of the unit structure of the metamaterial light absorber constructed by the present invention is based on the deep learning development framework of Pytorch and the Python development language, and includes a first data processor, a second data processor, a matrix splicer, a convolutional neural network, a vector tiler and a fully connected neural network.
[0085] The first data processor is used to execute step S10: converting the input crystalline absorption spectrum data into a first vector, and successively shrinking and expanding the first vector to output a first three-dimensional matrix.
[0086] Specifically, the first data processor includes a first vector converter and a first dimension transformation network.
[0087] The first vector converter is used to execute step S11: converting the input crystalline absorption spectrum data into a first vector.
[0088] Based on the above parameter settings, in the specific implementation, the 201 data output by the crystalline absorption spectrum mapping are converted into a first vector, the dimension of the first vector is [1×201], and the structural parameters corresponding to the 201 data are 13, which are height parameter values and 3 2 =9 coding unit state values.
[0089] The first dimensional transformation network is used to execute step S12: shrinking and expanding the first vector in sequence, and outputting a first three-dimensional matrix.
[0090] In a specific implementation, the first vector [1×201] is first shrunk into a vector of dimension [1×64], and the shrinkage of the first dimensional transformation network is performed according to the following formula:
[0091]
[0092] Where, e represents the crystalline absorption spectrum, e T represents the transposed matrix of the crystalline absorption spectrum, V1 and b1 represent the weight and bias parameters of the first dimension transformation network respectively, k is a constant representing the weight According to the quantity. V1 and b1 process the first vector with a dimension of [1×201] to obtain vector data with a dimension of [1×64]; then, the vector data with a dimension of [1×64] is expanded into a three-dimensional matrix with a dimension of [8×8×64].
[0093] The second data processor is used to execute step S20: convert the input amorphous absorption spectrum data into a second vector, and successively shrink and expand the second vector to output a second three-dimensional matrix.
[0094] Specifically, the second data processor includes a second vector converter and a second dimension transformation network.
[0095] The second vector converter is used to execute step S21: converting the input amorphous absorption spectrum data into a second vector.
[0096] Based on the above parameter settings, in the specific implementation, the 201 data output by the amorphous absorption spectrum mapping are converted into a second vector, the dimension of the second vector is [1×201], and the structural parameters corresponding to the 201 data are 13, which are height parameter values and 3 2 =9 coding unit state values.
[0097] The second dimensional transformation network is used to execute step S22: shrinking and expanding the second vector in sequence, and outputting a second three-dimensional matrix.
[0098] In a specific implementation, the second vector [1×201] is first shrunk into a vector of dimension [1×64], and the shrinkage of the second dimensional transformation network is performed according to the following formula:
[0099]
[0100] Where, e′ represents the amorphous absorption spectrum, e′ T represents the transposed matrix of the amorphous absorption spectrum, V2 and b2 represent the weight and bias parameters of the second dimension transformation network, respectively. k′ is a constant representing the weight According to the quantity. V2 and b2 process the second vector with a dimension of [1×201] to obtain vector data with a dimension of [1×64]; then, the vector data with a dimension of [1×64] is expanded into a three-dimensional matrix with a dimension of [8×8×64].
[0101] The matrix stitching device is used to execute step S30: stitching the first three-dimensional matrix and the second three-dimensional matrix, and outputting a stitching matrix.
[0102] In a specific implementation, the first three-dimensional matrix and the second three-dimensional matrix with dimensions of [8×8×64] are spliced into a spliced matrix with dimensions of [8×8×128].
[0103] The convolutional neural network is used to execute step S40: extracting features from the concatenated matrix and outputting a feature matrix.
[0104] In the specific implementation, the convolutional neural network extracts features from the concatenated matrix with a dimension of [8×8×128] and outputs a feature matrix with a dimension of [2×2×256].
[0105] The vector flattener is used to execute step S50: unfold the feature matrix and output a feature vector.
[0106] In specific implementation, the vector flattener expands the feature matrix with a dimension of [2×2×256] and flattens it into a vector with a dimension of [1×1024].
[0107] The fully connected neural network is used to execute step S60: performing feature conversion on the feature vector and outputting predicted structural parameters.
[0108] In the specific implementation, the fully connected neural network performs feature transformation on the feature vector with a dimension of [1×1024] and outputs the predicted structural parameters with a dimension of [1×13].
[0109] (IV) Training prediction model
[0110] The total data set is divided into the first data set in a ratio of 8:2 and the second dataset The first dataset The second dataset is used for training the unit structure prediction model. Used for testing and validation of unit structure prediction models.
[0111] The first dataset The structure-absorption spectrum data in is input into the unit structure prediction model, and the unit structure prediction model is trained. The specific training process is as follows:
[0112] S10 inputs the first data set The crystalline absorption spectrum data in a certain set of data in the first vector are converted into the first vector, and the first vector is contracted and expanded in sequence to output a first three-dimensional matrix;
[0113] S20 converts the input amorphous absorption spectrum data in the same group of data used in S10 into a second vector, and successively shrinks and expands the second vector to output a second three-dimensional matrix;
[0114] S30: splicing the first three-dimensional matrix and the second three-dimensional matrix, and outputting a spliced matrix;
[0115] S40 performs feature extraction on the concatenated matrix and outputs a feature matrix;
[0116] S50 expands the feature matrix and outputs a feature vector;
[0117] S60 performs feature conversion on the feature vector and outputs predicted structural parameters;
[0118] S70 calculates the loss value of the predicted structure parameters and the label structure parameters, and updates the weight and bias parameters in the model through the back propagation algorithm according to the loss value. The loss value is calculated according to the following formula:
[0119]
[0120] in, represents the predicted structural parameters, Represents the tag structure parameters.
[0121] Specifically, the weights and bias parameters that need to be updated in the model include the first dimension transformation network, the second dimension transformation network, the convolutional neural network, and the fully connected neural network.
[0122] Repeat steps S10 to S70 until the loss value becomes stable or reaches the iteration number or is less than the iteration threshold.
[0123] In this embodiment, the hyperparameter of the prediction model training is set to learning rate = 1×e -5 , the optimizer is Adam optimizer. This setting allows the prediction model to fully explore and avoid falling into local optimality. The training effect is also optimal.
[0124] At this point, the trained optimal structure prediction model is obtained.
[0125] (V) Simulation test analysis
[0126] (1) Prediction unit structure
[0127] See also Figure 4 , Figure 5 , obtaining two different actual expected absorption spectra, namely a first actual expected absorption spectrum and a second actual expected absorption spectrum. The first actual expected absorption spectrum and the second actual expected absorption spectrum are respectively input into the trained unit structure prediction model to obtain corresponding predicted structure parameters.
[0128] See also Figure 6 , Figure 7 , set two different ideal expected absorption spectra, namely the first ideal expected absorption spectrum and the second ideal expected absorption spectrum, wherein the first ideal expected absorption spectrum is an input absorption spectrum of y equal to 0.9 when used as an amorphous state and a sine function when used as an input absorption spectrum of a crystalline state; the second ideal expected absorption spectrum is an input absorption spectrum of y equal to 0.95 when used as an amorphous state and a crystalline state. The first ideal expected absorption spectrum and the second ideal expected absorption spectrum are respectively input into the trained unit structure prediction model to obtain the corresponding predicted structure parameters.
[0129] (2) Simulation test of predicted structure
[0130] The above predicted structural parameters were input into the FDTD solutions simulation software for verification.
[0131] The simulation results show that the absorption spectrum obtained by the predicted structural parameters is highly consistent with the set expected absorption spectrum as a whole, indicating that the neural network prediction model set by the present invention for simultaneously extracting features from the input crystalline and amorphous absorption spectra can accurately predict the parameters of the unit structure with good prediction effect, and can realize the prediction and construction of the microstructure of complex metamaterial light absorbers with multiple absorption characteristics.
[0132] Taking the predicted structure of the second ideal expected absorption spectrum and the corresponding simulated absorption spectrum as an example, when the crystalline and amorphous states are in the absorption spectrum frequency range of 0.3μm to 2.7μm, their corresponding absorption rates are both greater than 90%, of which the average absorption rate of the crystalline state is 93.6% and the average absorption rate of the amorphous state is 94.2%. Therefore, within this range, the absorption bandwidths corresponding to the amorphous and crystalline states of the metamaterial light absorber proposed in the present invention are both 2.4μm, which completely covers the spectral range of solar energy available to humans. By adjusting the state of the phase change material GSST, different types of absorption spectra corresponding to the same structure can be obtained, realizing the reconfigurability of the metamaterial light absorber.
[0133] At the same time, from Figure 5-Figure 7 It can be seen from the graphic structure of the metamaterial light absorber that the absorption anti-reflection layer of the unit structure proposed in the present invention is not a fixed pattern and different types of unit structures are not obtained by scaling the size of the absorption anti-reflection layer, but by dividing the absorption anti-reflection layer into multiple modules and using mirror-symmetric mapping between modules and combining image encoding within a single module to achieve diversity in prediction structures, so that the prediction network model has high generalization ability after training and can predict structural parameters corresponding to multiple different types of absorption spectra.
[0134] The present invention provides an optimized solution for designing reconfigurable metamaterial devices with multiple absorption characteristics. Compared with the design of traditional metamaterial devices, the unit structure of the metamaterial light absorber proposed in the present invention allows designers to avoid spending too much time looking for the optimal unit structure parameters corresponding to the applied absorption spectrum. The required absorption spectrum can be input into the unit structure prediction model proposed in the present invention to obtain the corresponding structure parameters, which greatly shortens the design time.
[0135] Based on the same inventive concept, the present application also provides an electronic device, which may be a terminal device such as a server, a desktop computing device or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). The device includes one or more processors and a memory, wherein the processor is used to execute a program to implement a structure prediction method for a metamaterial light absorber; and the memory is used to store a computer program executable by the processor.
[0136] Based on the same inventive concept, the present application also provides a computer-readable storage medium, corresponding to the embodiment of the aforementioned metamaterial light absorber structure prediction method, the computer-readable storage medium stores a computer program thereon, and when the program is executed by a processor, the steps of the metamaterial light absorber structure prediction method recorded in any of the above embodiments are implemented.
[0137] The present application may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-usable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
[0138] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, and the present invention is also intended to include these modifications and modifications.
Claims
1. A metamaterial light absorber, comprising a substrate layer, a dielectric layer and an absorption anti-reflection layer arranged in sequence, characterized in that: The dielectric layer includes a phase change material layer and a first silicon dioxide layer, wherein the phase change material layer is embedded in the first silicon dioxide layer and forms a surface contact with the base layer; the absorption anti-reflection layer includes a metal layer and a second silicon dioxide layer, wherein the metal layer is embedded in the second silicon dioxide layer and forms a surface contact with the first silicon dioxide layer; the phase change material layer has a graphic structure, wherein the graphic structure is evenly divided into four module graphics, wherein the four module graphics are arranged in a four-quadrant manner and the image of each module graphic is mirror-symmetrical with the image of the adjacent module graphics; the module graphics are evenly divided into n 2 A coding unit is provided, and a state value of "empty" or "real" is assigned to the coding unit. When the state value of the coding unit is "empty", it indicates that there is no phase change material or metal at the coding unit; when the state value of the coding unit is "real", it indicates that there is a phase change material or metal at the coding unit, wherein n is a coding parameter, n is an integer and n≥2; the graphic structure of the metal layer is the same as that of the phase change material layer and corresponds one to one.
2. The light absorber according to claim 1, characterized in that The phase change material layer is germanium antimony selenium tellurium alloy Ge2Sb2Se4Te (GSST).
3. A method for predicting the structure of a metamaterial light absorber, characterized in that: The following steps are involved: S10: converting the input crystalline absorption spectrum data into a first vector, and sequentially shrinking and expanding the first vector to output a first three-dimensional matrix; S20: converting the input amorphous absorption spectrum data into a second vector, and sequentially shrinking and expanding the second vector to output a second three-dimensional matrix; S30: splicing the first three-dimensional matrix and the second three-dimensional matrix, and outputting a spliced matrix; S40: extracting features from the concatenated matrix and outputting a feature matrix; S50: Expand the feature matrix and output a feature vector; S60: performing feature conversion on the feature vector and outputting predicted structural parameters; The weights and bias parameters used in executing steps S10, S20, S40, and S60 are corrected and optimized by simulating the light absorber according to claim 1 or 2 to obtain the structure-absorption spectrum data set of the phase change material in the crystalline and amorphous states.
4. The prediction method according to claim 3, characterized in that: The step S10 includes the following sub-steps: S11: converting the input crystalline absorption spectrum data into a first vector; S12: After shrinking and expanding the first vector in sequence, a first three-dimensional matrix is output; wherein the shrinking is performed according to the following formula: output1=f(e T W1 [1:k] e+V1e+b1); Where, e represents the crystalline absorption spectrum, e T Represents the transposed matrix of the crystalline absorption spectrum, W1 [1:k] , V1 and b1 represent the weight and bias parameters used in the first dimension transformation, respectively, k is a constant, and represents the weight W1 [1:k] The number of 5. The prediction method according to claim 3 or 4, characterized in that: The step S20 includes the following sub-steps: S21: converting the input amorphous absorption spectrum data into a second vector; S22: After shrinking and expanding the second vector in sequence, output a second three-dimensional matrix; wherein the shrinking is performed according to the following formula: Where, e′ represents the amorphous absorption spectrum, e′ T represents the transposed matrix of the amorphous absorption spectrum, V2 and b2 represent the weight and bias parameters used in the second dimension transformation, respectively. k′ is a constant representing the weight The number of 6. A metamaterial light absorber structure prediction device, characterized in that: comprising a first data processor, a second data processor, a matrix splicer, a convolutional neural network, a vector tiler, and a fully connected neural network; The first data processor is used to convert the input crystalline absorption spectrum data into a first vector, and output a first three-dimensional matrix after shrinking and expanding the first vector in sequence; The second data processor is used to convert the input amorphous absorption spectrum data into a second vector, and output a second three-dimensional matrix after shrinking and expanding the second vector in sequence; The matrix splicer is used to splice the first three-dimensional matrix and the second three-dimensional matrix and output a spliced matrix; The convolutional neural network is used to extract features from the concatenated matrix and output a feature matrix; The vector flattener is used to expand the feature matrix and output a feature vector; The fully connected neural network is used to perform feature conversion on the feature vector and output predicted structural parameters; The weights and bias parameters of the first data processor, the second data processor, the convolutional neural network and the fully connected neural network are corrected and optimized by simulating the light absorber according to claim 1 or 2 to obtain the structure-absorption spectrum data set of the phase change material in the crystalline and amorphous states.
7. The prediction device according to claim 6, characterized in that The first data processor includes a first vector converter and a first dimension transformation network; The first vector converter is used to convert the input crystalline absorption spectrum data into a first vector; The first dimensional transformation network is used to shrink and expand the first vector in sequence to output a first three-dimensional matrix; wherein the shrinkage of the first dimensional transformation network is performed according to the following formula: output1=f(e T W1 [1:k] e+V1e+b1); Where, e represents the crystalline absorption spectrum, e T Represents the transposed matrix of the crystalline absorption spectrum, W1 [1:k] , V1 and b1 represent the weight and bias parameters of the first dimension transformation network respectively, k is a constant, and represents the weight W1 [1:k] The number of 8. The prediction device according to claim 6 or 7, characterized in that: The second data processor includes a second vector converter and a second dimension transformation network; The second vector converter is used to convert the input crystalline absorption spectrum data into a second vector; The second dimensional transformation network is used to shrink and expand the second vector in sequence to output a second three-dimensional matrix; wherein the shrinkage of the second dimensional transformation network is performed according to the following formula: Where, e′ represents the amorphous absorption spectrum, e′ T represents the transposed matrix of the amorphous absorption spectrum, V2 and b2 represent the weight and bias parameters of the second dimension transformation network, respectively. k′ is a constant representing the weight The number of
