• Photonics Insights
  • Vol. 1, Issue 1, R01 (2022)
Qian Ma1、2、†, Che Liu1、2, Qiang Xiao1、2, Ze Gu1、2, Xinxin Gao1、2, Lianlin Li3, and Tie Jun Cui1、2、*
Author Affiliations
  • 1State Key Laboratory of Millimeter Waves, Southeast University, Nanjing, China
  • 2Institute of Electromagnetic Space, Southeast University, Nanjing, China
  • 3State Key Laboratory of Advanced Optical Communication Systems and Networks, Department of Electronics, Peking University, Beijing, China
  • show less
    DOI: 10.3788/PI.2022.R01 Cite this Article Set citation alerts
    Qian Ma, Che Liu, Qiang Xiao, Ze Gu, Xinxin Gao, Lianlin Li, Tie Jun Cui. Information metasurfaces and intelligent metasurfaces[J]. Photonics Insights, 2022, 1(1): R01 Copy Citation Text show less
    Development of metamaterials, artificial intelligence, and their integration to result in intelligent metamaterials.
    Fig. 1. Development of metamaterials, artificial intelligence, and their integration to result in intelligent metamaterials.
    Theories of information metasurfaces. (a), (b) Characterization of metasurface by digital coding and its scattering features. (c)–(e) Convolution operation of the digital coding metasurface, from the coding-pattern domain to the scattering-pattern domain. (f), (g) Information entropy of the digital coding metasurface, offering the information measurement from the coding pattern to the scattering pattern. (a), (b) Adapted from Ref. [66], Copyright 2014, with permission from Springer Nature, licensed under CC-BY-NC-SA 3.0. (c)–(e) Adapted from Ref. [97], Copyright 2016, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (f), (g) Adapted from Ref. [96], Copyright 2016, with permission from Springer Nature, licensed under CC-BY-NC-ND 4.0.
    Fig. 2. Theories of information metasurfaces. (a), (b) Characterization of metasurface by digital coding and its scattering features. (c)–(e) Convolution operation of the digital coding metasurface, from the coding-pattern domain to the scattering-pattern domain. (f), (g) Information entropy of the digital coding metasurface, offering the information measurement from the coding pattern to the scattering pattern. (a), (b) Adapted from Ref. [66], Copyright 2014, with permission from Springer Nature, licensed under CC-BY-NC-SA 3.0. (c)–(e) Adapted from Ref. [97], Copyright 2016, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (f), (g) Adapted from Ref. [96], Copyright 2016, with permission from Springer Nature, licensed under CC-BY-NC-ND 4.0.
    Principle of the reprogrammable plasmonic topological insulator and the experimental demonstration. (a) Schematic of the reprogrammable topological insulator, where each unit can be programed by FPGA to establish distinct topological routes. (b) Detailed structure of a 2-bit unit cell, in which six PIN diodes are integrated on the six branches. (c) Four typical states when different on–off states are applied, encoded as units 0, 1, 2, and 3. (d) Band diagrams of a crystal with the designed 2-bit unit cell. (e) First Brillouin zones of units 1 and 2. (f) Topological phase transition and valley–chirality properties of units 1 and 2. (g)–(i) Measured near-field distributions of three typical topological routes. Adapted from Ref. [75], Copyright 2021, under a Creative Commons Attribution 4.0 International License.
    Fig. 3. Principle of the reprogrammable plasmonic topological insulator and the experimental demonstration. (a) Schematic of the reprogrammable topological insulator, where each unit can be programed by FPGA to establish distinct topological routes. (b) Detailed structure of a 2-bit unit cell, in which six PIN diodes are integrated on the six branches. (c) Four typical states when different on–off states are applied, encoded as units 0, 1, 2, and 3. (d) Band diagrams of a crystal with the designed 2-bit unit cell. (e) First Brillouin zones of units 1 and 2. (f) Topological phase transition and valley–chirality properties of units 1 and 2. (g)–(i) Measured near-field distributions of three typical topological routes. Adapted from Ref. [75], Copyright 2021, under a Creative Commons Attribution 4.0 International License.
    Space–time-coding digital metasurface. (a) Conceptual illustration. (b), (c) Examples of space–time coding matrices for harmonic beam steering. (d)–(g) Distributions of the equivalent magnitudes and phases based on amplitude and phase modulations. (h), (i) Corresponding far-field scattering patterns at harmonic frequencies for AM and PM. Adapted from Ref. [82], Copyright 2018, under a Creative Commons Attribution 4.0 International License.
    Fig. 4. Space–time-coding digital metasurface. (a) Conceptual illustration. (b), (c) Examples of space–time coding matrices for harmonic beam steering. (d)–(g) Distributions of the equivalent magnitudes and phases based on amplitude and phase modulations. (h), (i) Corresponding far-field scattering patterns at harmonic frequencies for AM and PM. Adapted from Ref. [82], Copyright 2018, under a Creative Commons Attribution 4.0 International License.
    Space- and frequency-multiplexing wireless communication system based on the space–time-coding metasurface. (a) Conceptual illustration. (b) Experimental scenario. Adapted from Ref. [81], Copyright 2021, under a Creative Commons Attribution 4.0 International License.
    Fig. 5. Space- and frequency-multiplexing wireless communication system based on the space–time-coding metasurface. (a) Conceptual illustration. (b) Experimental scenario. Adapted from Ref. [81], Copyright 2021, under a Creative Commons Attribution 4.0 International License.
    Intelligent designs of meta-atoms. (a) Flowchart of the BPSO algorithm together with the CST Microwave Studio. The BPSO algorithm controls the update of meta-atoms, and CST provides the reflection phases of the current meta-atoms. (b) Models and reflection phases and amplitudes of the paired meta-atoms with 90° phase difference. (c) Flowchart of DDQN method used to optimize the meta-atoms. The DDQN model predicted the optimal current update actions of meta-atom structure and material parameters by learning from the interactions with numerical simulations. (a), (b) Adapted from Ref. [188], Copyright 2017, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License. (c) Adapted from Ref. [194], Copyright 2019, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License.
    Fig. 6. Intelligent designs of meta-atoms. (a) Flowchart of the BPSO algorithm together with the CST Microwave Studio. The BPSO algorithm controls the update of meta-atoms, and CST provides the reflection phases of the current meta-atoms. (b) Models and reflection phases and amplitudes of the paired meta-atoms with 90° phase difference. (c) Flowchart of DDQN method used to optimize the meta-atoms. The DDQN model predicted the optimal current update actions of meta-atom structure and material parameters by learning from the interactions with numerical simulations. (a), (b) Adapted from Ref. [188], Copyright 2017, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License. (c) Adapted from Ref. [194], Copyright 2019, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License.
    Illustration of design flowcharts for CNN. (a) Flowchart of the BPSO algorithm together with the prediction CNN to design the anisotropic coding meta-atom. The nearly real-time reflection-phase prediction of CNN accelerates immensely the whole procedure and makes it possible for simultaneous optimizations of TE and TM responses. (b) Contrastive flowchart of the design process of the REACTIVE method and the conventional metasurface design method. As a non-iterative method, REACTIVE could generate the probable digital meta-atom structures in seconds when given the design target of the reflection coefficients. (a) Adapted from Ref. [180], Copyright 2018, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) Reprinted from Ref. [182], Copyright 2019, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim.
    Fig. 7. Illustration of design flowcharts for CNN. (a) Flowchart of the BPSO algorithm together with the prediction CNN to design the anisotropic coding meta-atom. The nearly real-time reflection-phase prediction of CNN accelerates immensely the whole procedure and makes it possible for simultaneous optimizations of TE and TM responses. (b) Contrastive flowchart of the design process of the REACTIVE method and the conventional metasurface design method. As a non-iterative method, REACTIVE could generate the probable digital meta-atom structures in seconds when given the design target of the reflection coefficients. (a) Adapted from Ref. [180], Copyright 2018, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) Reprinted from Ref. [182], Copyright 2019, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim.
    Schematic diagrams of inverse ANNs. (a) Schematic diagram of an inverse ANN retrieving the relationship between spectrum response and meta-atom structure. After being well trained, the inverse ANN could directly output meta-atom structures with corresponding input spectrum responses. (b) Flowchart of GAN for inverse design of 2D meta-atoms with arbitrary patterns. The pre-trained PNN acted as a simulator that could form a VAE when concatenated to the generator. (c) Schematic diagram of an inverse-design GAN with latent space. A well-designed sampling strategy was designed to sample data from this latent space as parts of the inverse generator’s input to guarantee the diversity of design results. (a) Adapted from Ref. [66], Copyright 2019, with permission from American Chemical Society. (b) Reprinted from Ref. [197], Copyright 2018, with permission from American Chemical Society. (c) Reprinted from Ref. [199], Copyright 2019, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim.
    Fig. 8. Schematic diagrams of inverse ANNs. (a) Schematic diagram of an inverse ANN retrieving the relationship between spectrum response and meta-atom structure. After being well trained, the inverse ANN could directly output meta-atom structures with corresponding input spectrum responses. (b) Flowchart of GAN for inverse design of 2D meta-atoms with arbitrary patterns. The pre-trained PNN acted as a simulator that could form a VAE when concatenated to the generator. (c) Schematic diagram of an inverse-design GAN with latent space. A well-designed sampling strategy was designed to sample data from this latent space as parts of the inverse generator’s input to guarantee the diversity of design results. (a) Adapted from Ref. [66], Copyright 2019, with permission from American Chemical Society. (b) Reprinted from Ref. [197], Copyright 2018, with permission from American Chemical Society. (c) Reprinted from Ref. [199], Copyright 2019, with permission from Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim.
    Intelligent design of the metasurface pattern. (a) Experimental setup of metasurface cloak controlled in real time by a pretrained ANN, which can learn the mapping from the needed reflection spectra together with the features of the incident wave to the corresponding bias voltages of the meta-atoms. (b) Schematic diagram of the physics-assisted unsupervised GAN for real-time holography. The generator together with the EM propagation process makes up the VAE structure. A discriminator was used to improve the imaging quality of the generator. (a) Adapted from Ref. [48], Copyright 2020, with permission from Springer Nature. (b) Reprinted from Ref. [88], Copyright 2021, with permission from Optica Publishing Group, under a Creative Commons Attribution 4.0 International License.
    Fig. 9. Intelligent design of the metasurface pattern. (a) Experimental setup of metasurface cloak controlled in real time by a pretrained ANN, which can learn the mapping from the needed reflection spectra together with the features of the incident wave to the corresponding bias voltages of the meta-atoms. (b) Schematic diagram of the physics-assisted unsupervised GAN for real-time holography. The generator together with the EM propagation process makes up the VAE structure. A discriminator was used to improve the imaging quality of the generator. (a) Adapted from Ref. [48], Copyright 2020, with permission from Springer Nature. (b) Reprinted from Ref. [88], Copyright 2021, with permission from Optica Publishing Group, under a Creative Commons Attribution 4.0 International License.
    Reprogrammable metasurface imager integrated with machine-learning algorithm. (a) Schematic of the machine-learning algorithm. (b) Meta-atom structure and the metasurface. (c) Real-time imaging through a wall. (d), (e) Experimental measurements of different body gestures and the related imaging results. (f) Classification rate of different algorithms. Adapted from Ref. [89], Copyright 2019, under a Creative Commons Attribution 4.0 International License.
    Fig. 10. Reprogrammable metasurface imager integrated with machine-learning algorithm. (a) Schematic of the machine-learning algorithm. (b) Meta-atom structure and the metasurface. (c) Real-time imaging through a wall. (d), (e) Experimental measurements of different body gestures and the related imaging results. (f) Classification rate of different algorithms. Adapted from Ref. [89], Copyright 2019, under a Creative Commons Attribution 4.0 International License.
    Intelligent microwave imager and recognizer. (a) Application scenario of the intelligent metasurface. (b) System architecture of the intelligent metasurface. (c) Meta-atom responses at different digital states. (d) Programmable manipulations of EM focusing for different functions including hand gestures and vital inspection. Adapted from Ref. [87], Copyright 2019, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License.
    Fig. 11. Intelligent microwave imager and recognizer. (a) Application scenario of the intelligent metasurface. (b) System architecture of the intelligent metasurface. (c) Meta-atom responses at different digital states. (d) Programmable manipulations of EM focusing for different functions including hand gestures and vital inspection. Adapted from Ref. [87], Copyright 2019, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License.
    Smart metasurfaces with self-adaptive capabilities. (a), (b) Conceptual illustration of self-adaptively smart metasurfaces. (c) Smart sensing metasurface. (a), (b) Adapted from Ref. [85], Copyright 2019, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License. (c) Reprinted from Ref. [86], Copyright 2020, under a Creative Commons Attribution 4.0 International License.
    Fig. 12. Smart metasurfaces with self-adaptive capabilities. (a), (b) Conceptual illustration of self-adaptively smart metasurfaces. (c) Smart sensing metasurface. (a), (b) Adapted from Ref. [85], Copyright 2019, with permission from Springer Nature, under a Creative Commons Attribution 4.0 International License. (c) Reprinted from Ref. [86], Copyright 2020, under a Creative Commons Attribution 4.0 International License.
    Diffractive deep neural networks. (a), (b) All-optical diffractive deep neural networks (D2NNs) based on 3D-printed materials. (c), (d) Input and output optical fields, where the detector regions are marked with red dashed lines. (e), (f) Confusion matrix and energy distribution percentage for 10,000 different handwritten digits. (g) Nanophotonic media for artificial neural inference. (h) Schematic of DPU. (i) Experiment of DPU. (a)–(f) Adapted from Ref. [147], Copyright 2018, under a Creative Commons Attribution 4.0 International License. (g) Reprinted from Ref. [88], Copyright 2021, with permission from Optica Publishing Group, under a Creative Commons Attribution 4.0 International License. (h), (i) Adapted from Ref. [154], Copyright 2020, with permission from Springer Nature.
    Fig. 13. Diffractive deep neural networks. (a), (b) All-optical diffractive deep neural networks (D2NNs) based on 3D-printed materials. (c), (d) Input and output optical fields, where the detector regions are marked with red dashed lines. (e), (f) Confusion matrix and energy distribution percentage for 10,000 different handwritten digits. (g) Nanophotonic media for artificial neural inference. (h) Schematic of DPU. (i) Experiment of DPU. (a)–(f) Adapted from Ref. [147], Copyright 2018, under a Creative Commons Attribution 4.0 International License. (g) Reprinted from Ref. [88], Copyright 2021, with permission from Optica Publishing Group, under a Creative Commons Attribution 4.0 International License. (h), (i) Adapted from Ref. [154], Copyright 2020, with permission from Springer Nature.
    Programmable artificial neural network and its image recognition. (a) PAIM structure composed of multi-layer information metasurfaces. (b) Diffractive illustration of PAIM. (c), (d) Image recognition of oil paintings for landscape and portraiture. Adapted from Ref. [155], Copyright 2022, with permission from Springer Nature.
    Fig. 14. Programmable artificial neural network and its image recognition. (a) PAIM structure composed of multi-layer information metasurfaces. (b) Diffractive illustration of PAIM. (c), (d) Image recognition of oil paintings for landscape and portraiture. Adapted from Ref. [155], Copyright 2022, with permission from Springer Nature.
    Encoder and decoder in the CDMA scheme and its communication experiment using PAIM. (a) Schematic of CDMA scheme using PAIM, where the first layer and the last four layers are assigned as encoder and decoders, respectively. (b) Experiment of image transmission using the presented CDMA scheme and PAIM. Adapted from Ref. [155], Copyright 2022, with permission from Springer Nature.
    Fig. 15. Encoder and decoder in the CDMA scheme and its communication experiment using PAIM. (a) Schematic of CDMA scheme using PAIM, where the first layer and the last four layers are assigned as encoder and decoders, respectively. (b) Experiment of image transmission using the presented CDMA scheme and PAIM. Adapted from Ref. [155], Copyright 2022, with permission from Springer Nature.
    Qian Ma, Che Liu, Qiang Xiao, Ze Gu, Xinxin Gao, Lianlin Li, Tie Jun Cui. Information metasurfaces and intelligent metasurfaces[J]. Photonics Insights, 2022, 1(1): R01
    Download Citation