Metamaterials: From Fundamental Physics To Intelligent Design

Received: 30 April 2022 | Revised: 4 June 2022 | Accepted: 19 June 2022
DOI: 10.1002/idm2.12049
REVIEW
Metamaterials: From fundamental physics to intelligent

design
Ji Chen1,2 | Shanshan Hu1 | Shining Zhu1 | Tao Li1

1
National Laboratory of Solid State Microstructures, Key Laboratory of Intelligent Optical Sensing and Manipulations, Jiangsu Key Laboratory of
Artificial Functional Materials, College of Engineering and Applied Sciences, School of Physics, Nanjing University, Nanjing, China
2
National Mobile Communications Research Laboratory, School of Information Science and Engineering, Frontiers Science Center for Mobile
Information Communication and Security, Southeast University, Nanjing, China
Correspondence
Tao Li, Ji Chen, and Shining Zhu, National Abstract
Laboratory of Solid State Microstructures, Metamaterials are artificial structures with the ability to efficiently control
Key Laboratory of Intelligent Optical
Sensing and Manipulations, Jiangsu Key
light‐field, attracting intensive attention in the past few decades. People have
Laboratory of Artificial Functional studied the working principles, design strategies, and fabrication methods of
Materials, College of Engineering and metamaterials, making this field cross and combine with many disciplines,
Applied Sciences, School of Physics,
Nanjing University, 210093 Nanjing, including physics, material science, electronics, and chemistry. In recent
China. years, with the rapid development of high‐efficiency and multifunctional
Email: taoli@nju.edu.cn, jichen@seu.edu.cn,
metasurfaces, which are a two‐dimensional version of metamaterials, great
and zhusn@nju.edu.cn
efforts have been made to push this material to practical applications. In
Funding information particular, the introduction of artificial intelligent (AI) algorithms enables
National Key R&D Program of China, metamaterials‐based photonic devices that exhibit excellent performances and
Grant/Award Number: 2016YFA0202103;
Fundamental Research Funds for the
intelligent functionalities. In this review, we first introduce the basic concepts,
Central Universities, working principles, design methods, and applications of metamaterials, and
Grant/Award Numbers: 2242022R10128, then focus on the rapidly developing metamaterials research combined with
2242022k30006; National Natural Science
Foundation of China, AI algorithms. Finally, we conclude this review with personal perspectives on
Grant/Award Numbers: 12104223, the current problems and future directions of metamaterials research and
91850204
developments.
KEYWORDS
artificial intelligence, artificial neural network, inverse design, metamaterial, metasurface
1 | INTRODUCTION polarizers, and possess a very limited range of permittivity,

leading to their bulky size, and curved shapes to ensure
The ability to efficiently manipulate electromagnetic (EM) enough accumulation of propagating phase. However, it is
waves is always highly desired for its great significance in still far from satisfying people's desire for arbitrary
many aspects, like, imaging, information communications, manipulation of EM waves. Metamaterials are artificial
target detection, energy utilizations, and optical encryp- materials composed of an array of nanostructures, which
tions. Conventional optical components are composed function as the atoms and molecules in traditional
of natural materials, such as lenses, waveplates, and materials and are termed as meta‐atoms. Different from
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided
the original work is properly cited.
© 2022 The Authors. Interdisciplinary Materials published by Wuhan University of Technology and John Wiley & Sons Australia, Ltd.
Interdisciplinary Materials. 2023;2:5–29. onlinelibrary.wiley.com/r/interdisciplinarymaterials | 5

2767441x, 2023, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/idm2.12049 by Cern Library, Wiley Online Library on [28/03/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 | CHEN ET AL.
natural materials, the properties of metamaterials are not propagating effect inside the metamaterials by introducing
determined by the intrinsic properties of chemical constit- abrupt changes of optical properties which are achieved by
uents but rather determined by the specific structures of interactions between light and scatterers.[11–14] Through
meta‐atoms.[1,2] Through efficient interactions between arranging scatterers with various geometric parameters in
incident EM waves and the meta‐atoms, metamaterials certain arrays, spatially varying optical responses can be
possess powerful light‐field manipulation ability and exhibit formed at will. Owing to the advantages like ultrathin
fascinating physical properties some of which are even not thickness, relatively low fabrication difficulty, and low‐loss
achievable in natural existing materials.[3–8] optical field modulation, metasurfaces have been exten-
If we describe all materials in parameter space with sively studied in recent years and shown promising
axes of permittivity (ε) and permeability (μ) (Figure 1A), prospects in realistic applications.[15–28] Numerous func-
which are two basic parameters to describe how EM tionalities have been demonstrated, such as metalens,
fields interact with materials,[9] most dielectric materials beam engineering, waveplates, polarizers, and holograms,
will locate in the quadrant where ε and μ are all positive. the performances of which strongly rely on the design,
Quadrant with ε < 0 and μ > 0 covers materials of metals and optimization of metastructures.
and ferroelectric materials, while some ferrite materials With the continuous development of metamaterials
are included in the opposite quadrant with ε > 0 and and metasurfaces, researchers have gradually shifted their
μ < 0. However, natural existing materials with simulta- attention from explorations of new physical principles to
neously negative permittivity and permeability have not versatile functionalities and devices based on this newly
been found so far. Through careful structural design, in emerging optical design. Thus various optimization algo-
principle metamaterials can exhibit arbitrary values of rithms are used to seek the optimal structure parameters
ε and μ, which leads to the significant capability of for realizing specific functions.[29–31] However, due to the
metamaterials to control EM waves. Therefore despite more and more complex desired functions and growing
Veselago theoretically investigated materials with simul- demands on the devices' performances, simple and regular
taneously negative permittivity and permeability as early meta‐atoms will no longer meet the requirements, instead
as 1968,[10] this kind of material was eventually achieved of sophisticated structures being used to construct the
by using metamaterials after several decades. However, metadevices. This will undoubtedly lead to a very large
despite metamaterials possessing the strong ability to space of parameters that can hardly be optimized by
manipulate light fields, the nonnegligible losses and the traditional algorithms. In recent years, artificial intelligence
fabrication challenges of complicated three‐dimensional (AI) algorithms especially artificial neural networks
(3D) structures hinder its development toward practical (ANNs) have been successfully applied in complex
applications. metastructures designing, endowing the metadevices with
Metasurfaces are generally created by using optically richer functions, and excellent performances.[32–34] This
thin films and assembled by arrays of subwavelength successful combination points to a promising direction of
scatterers, which can be viewed as a two‐dimensional (2D) metadevices development, hence in this review, we will
version of metamaterials. The working principles of pay much attention and fully discussion to this direction.
metasurfaces are totally different from 3D metamaterials This review is organized as follows, after introducing
in that metasurfaces break out the dependence of the basic concepts and development trend of metamaterials
F I G U R E 1 Working principles of three‐dimensional (3D) metamaterials. (A) Material parameter space characterized by permittivity
and permeability. (B) Schematic of split‐ring resonators. (C) Schematic of periodic infinite wires. Three figures are all reproduced with
permission.[7] Copyright 2011, Royal Society of Chemistry.
CHEN ET AL. | 7
in Section 1, successively we present the working to design metamaterials to surmount obstacles of nature,
principles and applications of 3D metamaterials and such as magnetism at optical frequencies. By mimicking
metasurfaces in Sections 2 and 3, respectively. Subse- a usual liquid crystal (LC) circuit, consisting of a plate
quently, traditional structural design methods includ- capacitor with capacitance C and a magnetic coil with
ing numerical simulations and classical optimization inductance L, people have successfully realized artificial
algorithms are introduced in Section 4. Following in magnetic “atoms,” termed split‐ring resonators (SRRs;
Section 5 is the focus of this review, AI design methods Figure 1B). A magnetic field penetrating the resonator
which will be discussed in detail to show how will induce a strong circulating current and thus result in
metamaterials are endowed with the properties of an effective magnetic moment.[35–38] The effective
intelligence. Finally, in Section 6 this review is capacitance C and inductance L of SRRs are related to
concluded with personal perspectives on the current the structure parameters marked in the figure, which are
problems and future directions of this rapidly growing shown as C = ε0 εC ωt / d and L = μ0 l 2/ t , respectively.
research field. These lead to the eigenfrequency
1 1 ε0 d
ω0 = = . (2)
2 | WORKING PRINCIPL ES OF 3 D LC l εC ω
M E T A M A T E R I A L S AN D T H E I R
APPLICATIONS John Pendry derived the effective relative magnetic
permeability of an SRR array which is given by[39]
2.1 | Working principles
Fω2
μ r,eff (ω) = 1 − , (3)
Permittivity (ε) and permeability (μ) are two basic ω2 − ω02 + iΓω
parameters of materials to describe the EM responses
to external EM waves, which usually can be described by where F is the filling ratio of the SRR and Γ is the
the Drude–Lorentz model shown as[9] damping term. If the magnetic response of the SRR is



sufficiently strong, μ r,eff with a negative value can be




2

ωp,e achieved.


ε (ω ) = 1 − ,

r In addition to SRR, Pendry et al. also proposed basic

2
ω2 − ω0,e + iγe ω




(1) metamaterial structures to implement artificial electric


2
ωp,m


μ r (ω ) = 1 − , responses that are dilute metals with extremely low
2
ω2 − ω0,m + iγm ω plasma frequency.[40] The structure is composed of
periodic infinite wires arranged in a simple cubic lattice
Here ωp, ω0, and γ are the plasma frequency, the (Figure 1C). Detailed derivations show that the effective
resonant frequency, and the damping factor with sub- permittivity of the system is given by
scripts ‘e’ and ‘m’ represent electric and magnetic
2
response, respectively. In naturally existing materials ωp,eff
εr,eff = 1 − , (4)
these parameters are generally fixed which are related to ω (ω + iγeff )
the intrinsic properties of the materials. Taking noble
metals as an example, ω0,e is taken as zero under the free 2
where ωp,eff , and γeff are, respectively, the effective
electron approximation, which means the permittivity of plasma frequency and damping factor which are related
metals is always negative below the plasma frequency. 2 2πc 2
with the structure parameters, shown as ωp,eff = d 2 ln(d /r )
On the contrary, the permittivity of dielectric materials is ε0 d 2ωp2
almost positive. In contrast to the apparent electrical and γeff = πr 2σ
, respectively.
responses to incident light, the magnetic response in
natural substances is significantly weaker at optical
frequency, leading to the magnetic permeability being 2.2 | Negative‐index metamaterials
nearly unity, that is, μ = 1 holds for all naturally existing
materials. By using the combination of periodic SRR and wires as the
Metamaterials are artificial structures that allow us to basic structure, the composed metamaterials will exhibit
tailor the EM material properties to a previously any desired permittivity and permeability after well
unprecedented degree. That is permittivity and perme- determining the structural parameters. A representative
ability can be controlled using properly designed example is the realization of negative‐index metamaterials
structures. Thanks to this complete freedom, it is possible with ε < 0 and μ < 0, which was proposed by Veselago
8 | CHEN ET AL.
many decades ago.[10] Figure 2A shows the schematic a manner consistent with Snell's law in their work.[41,48]
of negative refraction phenomena. Smith's group first Since then, magnetic metamaterials and negative‐index
experimentally demonstrated negative‐index metamater- metamaterials have aroused strong research interest, a
ials at the microwave frequency, by designing the series of related works were reported within several years.
structure consisting of a periodic array of copper SRRs The working wavelengths of the realized metamaterials
and wires (Figure 2B). Negative refraction was observed in covered not only the microwave but also the visible
F I G U R E 2 Applications of metamaterials. (A) Schematic of negative refraction phenomenon. (B) Negative‐index metamaterials (left)
and the experimental results (right) working at microwave frequencies. Reproduced with permission.[41] Copyright 2001, American
Association for the Advancement of Science. (C) Optical superlensing experiment. Reproduced with permission.[42] Copyright 2005,
American Association for the Advancement of Science. (D) Schematic of the experimental setup for hyperlens demonstration. Reproduced
with permission.[43] Copyright 2007, American Association for the Advancement of Science. (E) Light propagating through the cloaking
material. Reproduced with permission.[44] Copyright 2006, American Association for the Advancement of Science. (F) Experimental setup
for cloaking at microwave frequency. Reproduced with permission.[45] Copyright 2006, American Association for the Advancement of
Science. (G) Illustration of a metasurface skin cloak. Reproduced with permission.[21] Copyright 2015, American Association for the
Advancement of Science. (H) The working principle of an illusion device that transforms the stereoscopic image of the object (a man) into
that of the illusion (a woman). Reproduced with permission.[46] Copyright 2009, American Physical Society. (I) Analogue of light deflection
in a gravitational field (left) and microstructured optical waveguide (right). Reproduced with permission.[47] Copyright 2013, Springer
Nature.
CHEN ET AL. | 9
regime.[37,49–51] However, due to the very shot working tailoring the material property.[57–59] The invariance of
wavelength at visible, the structure size of SRR is also Maxwell's equations implies that the coordinate transfor-
drastically reduced, which not only brings great challenges mation can instead be applied to the permittivity and
to fabrication but also ultimately affects the magnetic permeability, giving rise to the prescription of the spatial
response in SRR. profile of the material that enables us to achieve desired
To circumvent this difficulty, researchers used fishnet light flow. Following rigorous mathematical derivation to
structure[52,53] as an alternative design to realize design metamaterials for transformation optics, research-
negative‐index materials other than a combination of ers have been able to realize many exotic optical
metal wires and SRRs, which are composed of two layers phenomena, including invisible cloak, illusion optics,
of metal meshes with a dielectric spacer layer in the and optical black holes.
middle, as shown in the figure. The top and bottom metal Transformation optics render an object invisible by
films form a loop and the induced current is interrupted guiding the flow of light around the hidden object
by the hole which also forms a capacitor between the two without disturbance to the internal object (Figure 2E).
films giving a resonant permeability response. Thus the In 2006, transformation optics for cloaking was indepen-
paired stripes oriented parallel to the electric field dently presented by Pendry et al.[44] and Leonhardt.[60]
provide negative permittivity while the pairs of stripes Soon after a 2D cloak prototype composed of SRRs was
parallel to the magnetic field offer negative permeability. designed in an experiment to validate cloaking phe-
This fishnet structure is much easier to fabricate than the nomena in the microwave spectrum (Figure 2F).[45]
combination of metal wires and SRRs, providing a new Besides the complete cloak, the carpet cloak aiming to
direction for the design of negative refractive index give all cloaked objects the appearance of a flat
materials. conducting sheet was extensively studied since first
One of the most striking properties of negative‐index proposed by Li and Pendry.[61] Carpet cloak relieves the
material is that a slab negative‐index material can be a loss issue by avoiding the usage of resonating elements
“perfect lens” in which the evanescent waves, instead of hence is easier to be realized at optical frequencies
decaying, are in fact enhance through the slab.[54] This (Figure 2G).[62–65] More than in EM waves, transforma-
offers the possibility to make a perfect image without any tion optics have been used as a universal tool to engineer
deterioration, which was first proposed by Pendry.[55] The sound, heat flow, elastic waves, water waves, and so
evanescent waves of a light passing through an object on.[66–71] Illusion optics could transform an object into an
always carry the subwavelength detail of the object. illusion of another object (Figure 2H) by using the
However, the evanescent waves decay exponentially in complementary medium and the restoring medium
mediums with a positive refractive index so that they designed from transformation optics.[46,58] To achieve a
cannot be collected at the image plane by a conventional total illusion effect for all angles, the far‐field EM
lens, resulting in a diffraction‐limited image. But the responses of the original object plus the illusion device
evanescent waves propagating in a negative‐index mate- must be equal to the illusion object for any incident
rial will have opposite behavior that is the evanescent waves. However, it is far from being implemented
waves can be strongly enhanced across the media. So if the experimentally, as it requires the negative‐index materi-
middle of the object plane and the image plane is filled als used in the experiment must not have large
with a proper combination of positive‐ and negative‐index absorption as well as defects, which is very difficult to
slabs, the object wave can be completely recovered to form achieve in existing fabrication techniques.[72] Another
a perfect image on the image plane, as shown in the fascinating application of transformation optics is mi-
figure. After Pendry proposed the concept, researchers micking universal phenomena on optical chips. Sheng
used various kinds of metamaterials structure to realize et al. experimentally demonstrated an optical analog of
superlens and experimentally demonstrated the perfect the effects of gravity on the motion of light rays,
imaging at microwave and optical frequencies including light deflection and photon capture.[47] The
(Figure 2C,D).[42,43,56] “gravitational field” effect in their experiment is achieved
by a microstructured waveguide spin‐coated in the
presence of a microsphere (Figure 2I), the refractive
2.3 | Transformation optics index distribution of which is similar to the form of
gravitational field distribution around celestial bodies.
In addition to negative refractive index materials, Through this method, they also experimentally emulated
another important research field of metamaterials is the formation of Einstein's ring, which is a space‐time
transformation optics, the goal of which is to precisely curvature phenomenon predicted by Einstein's General
control the flow of light in desirable manners by spatially Relativity.[73]
10 | CHEN ET AL.
Although 3D metamaterials have been successfully metasurfaces and propagation phase metasurfaces.
used to solve many scientific issues and demonstrated Next, we will introduce these different types of
exotic optical phenomena, it is still a big challenge to metasurfaces in detail.
implement practical devices, due to the complex structures,
large propagation loss, and high fabrication requirements.
3.1 | Resonant metasurfaces
3 | METASURFACE A ND The early proposed metasurfaces mainly used plasmonic

METALENS metal resonant units to control the light field. For example,
Yu et al. designed a metasurface composed of V‐shaped
In the past decade, ultrathin, ultralight, and high‐ antennas (Figure 3B), the double resonance effect which
efficiency metasurfaces have been widely studied, which guarantees the light control can cover the full phase
are the 2D version of metamaterials and also possess modulation range in mid‐infrared.[11] Similarly, metasur-
powerful control abilities to EM waves. Different from faces composed of H‐shaped antennas or SRRs have also
the working principles of 3D metamaterials, the manip- been proposed to achieve arbitrary phase modulation,
ulation of EM waves by metasurface is through the owning to the high order resonances in the unit
interactions between the metaunit structure and the structures.[80] In addition, Ni et al. applied the Babinet
incident waves, resulting in abrupt light properties principle to design a metasurface with hole structures as
changes of the emergent light on the 2D plane. the resonant units (Figure 3C), which can also control
Compared with 3D metamaterials, metasurfaces are not phase at will by setting appropriate parameters.[74]
only easier to be fabricated, but also avoid the propaga- Although plasmonic resonant structures possess advan-
tion loss of light in bulk materials, so great attention has tages of small size and strong EM field locality,[81,82] they
been received since they were first proposed by Capasso's suffer from serious loss problems, especially at optical
group in 2011.[11] Capasso's group used Fermat's princi- frequency, which limits the performances of the
ple to derive that the reflective and refractive behavior of metasurface‐based devices. To improve the efficiency,
light on metasurfaces can be expressed by generalized low‐loss dielectric Huygens' metasurfaces have been
Snell's law (Figure 3A): proposed (Figure 3D),[75] the high refractive index




dielectric units which support strong EM resonance modes,

 sin(θr ) − sin(θi ) = 2πn i dx ,

λ 0 dϕ known as Mie resonances.[83] Researchers have found that





the dielectric units can generate electrical and magnetic

 sin(θt ) n t − sin(θi ) n i = 2π dx .
(5)

λ 0 dϕ resonances of comparable intensity and overlapping
spectral. Kivshar's group proposed that by properly
designing the unit parameters, metasurfaces with high
Where λ 0 is the working wavelength, nt and n i are transmittance can be realized.[22,75] However, the working
the refractive index of the medium on both sides of bandwidth of resonant metasurface is narrow, as the
the interface, θi , θ r, and θ t are the angles of incidence, modulation is very sensitive to the structural parameters,
reflection, and transmission, respectively, dϕ / dx is the which also places high requirements for fabrication
gradient phase introduced by metasurface. According accuracy.
to generalized Snell's law, metasurfaces designed to
satisfy specific phase gradients can refract light in any
direction. On the other hand, if we look from the 3.2 | Geometric phase metasurface
perspective of Huygens–Fresnel's principle, the me-
tasurface units can be regarded as the secondary The geometric phase metasurfaces consist of anisotropic
wavelet sources, the radiation fields from which scattering units with uniform geometric parameters
interfere in space to form any desired light‐field (Figure 3E), and the phase control is realized by
distribution. changing the orientations of units. Hasman and other
There are two types of metasurfaces, working with researchers found that when circularly polarized light is
different principles. First is the resonant metasur- incident on a dipole antenna, the modulated light will
faces, in which the abrupt phase change of EM waves partially convert into the opposite circular polarization
is caused by resonances in unit structures, including with a phase change determined only by the orientation
plasmon resonance and dielectric resonance. How- of the dipole antenna.[84–86] Soon after Zhang's group
ever, the other is the nonresonant type, including systematically described the method of obtaining gradi-
geometric (or Pancharatnam–Berry [PB]) phase ent phase by rotated nanorods.[76] The modulated
CHEN ET AL. | 11
F I G U R E 3 Various types of metasurfaces. (A) Schematic of the generalized Snell's law. Reproduced with permission.[11] Copyright
2011, American Association for the Advancement of Science. (B) Resonant metasurface is composed of V‐shaped antennas. Reproduced
with permission.[11] Copyright 2011, American Association for the Advancement of Science. (C) Babinet‐inverted plasmonic metalens.
Reproduced with permission.[74] Copyright 2013, Springer Nature. (D) High‐efficiency dielectric Huygens' metasurface. Reproduced with
permission.[75] Copyright 2015, WILEY‐VCH. (E) The geometric phase metasurface. Reproduced with permission.[76] Copyright 2012,
American Chemical Society. (F) Propagation phase metasurface. Reproduced with permission.[77] Copyright 2015, Springer Nature.
(G) Broadband achromatic metalens with a combination of resonant phase and geometric phase. Reproduced with permission.[78] Copyright
2017, Springer Nature. (H) Multiplexed phase modulation under different polarizations is realized by a combination of geometric phase and
propagation phase. Reproduced with permission,[79] Copyright 2017, American Physical Society.
transmission field under the normal incidence of handed circular polarization. The first term on the right
circularly polarized light is found as side of the equal sign in Equation (6) represents light
with the same polarization as the incident light, while
Et = tˆ (θ )|σ  = o |σ  + o | − σ , (6)
t + te t − te ∓i2σθ the second term is the orthogonal circularly polarized
e
2 2 light with an extra phase of ∓ 2σθ . This extra phase is the
geometric phase, also known as the PB phase. According
where to and te are the complex transmission coefficients to this equation, if the nanorod is rotated from 0 to π rad,
along the long and short axes of the nanorod, respec- the phase control of orthogonal circularly polarized light
tively, and σ = 1/−1 corresponds to left‐handed/right‐ will cover the full 2π range. Thanks to the simple unit
12 | CHEN ET AL.
structure and wavelength‐independent phase modulation ultrathin, ultralight, and flat architectures favor minimiza-
mechanism, geometric phase metasurfaces have been tion and compact devices. Second, their function multi-
widely used in recent years.[87–91] However, due to its plexing and expansion abilities provide flexible wavefront
design for circularly polarized light, additional polarizing shaping, polarization control, and spectrum tailoring.
and analyzing optical elements are required for high‐ Metalens can focus normally incident light into a point if
quality light control, which inevitably reduces the light a hyperbolic phase distribution is designed, expressed as
energy efficiency.
2π
φ (r ) = (f − r 2 + f 2 ), (7)
λ
3.3 | Propagation phase metasurface
where λ is the working wavelength, r is the radial
The units in this type of metasurface are dielectric nanoposts position from the center, and f is the focal length. Early
with a high aspect ratio, which can be regarded as truncated realized metalenses were composed of plasmonic nanos-
dielectric waveguides (Figure 3F). By changing the cross‐ tructures,[104–106] the large intrinsic losses which lead to
section size, nanoposts will possess different effective low working efficiencies of metalenses. In 2016, Capas-
propagation constants, resulting in required phase accumu- so's group proposed a TiO2‐based metalens with an
lation via propagation over the nanoposts.[92,93] Although efficiency of 86%,[107] the good imaging performance of
high aspect ratio nanoposts place high fabrication require- which attracted much attention. Since then great efforts
ments, such polarization‐insensitive and high‐efficient have been made by researchers to further improve the
metasurfaces are of great use in practical applications. performances of metalenses, including efficiency, work-
Faraon's group realized a series of metasurfaces‐based ing bandwidth, and field of view (FOV).
devices by using high refractive index silicon nanoposts as Due to the diffractive optical properties of metalenses,
the units.[77,94–98] In 2018, Dong's group achieved a metasur- they have obvious chromatic aberration without special
face of 80% transmittance in the visible regime, which is design, which significantly limit their applications,
composed of silicon nitride nanoposts.[99] especially in color imaging and display. To address this
Recently, strategies that combine two or more types of issue, achromatic metalenses working at several discrete
metasurfaces have been proposed to enrich the control wavelengths were achieved by integrating multiple lenses
functions and improve control performances. The V‐shaped into the same platform.[108–114] Research further realized
metal antennas proposed by Capasso's group in 2011 continuous broadband achromatic metalenses by optimi-
contain the modulations of the resonance phase and the zation of metastructures.[115,116] Through this method,
geometric phase.[11] In the designed eight diffraction units, Capasso's group demonstrated a reflective achromatic
the first four are resonant units that realize phase metalens that have an invariant focal length from 490 to
modulation from 0 to π through different resonance modes. 550 nm.[115] By exploiting the geometric phase combined
Then after rotating 90°, the latter four units can be obtained with phase compensation from well‐designed integrated‐
to cover the remaining π to 2π phase modulation range. resonant elements, Tsai and Li's joint group successively
Continuous broadband achromatic metalenses are another realized broadband achromatic metalenses in the near‐
example that benefits from the combination of resonant infrared (Figure 4A) and the visible.[78,122,123] Capasso's
phase and geometric phase (Figure 3G), which will be group combined geometric phase and resonance phase
discussed later. Moreover, researchers have also studied within the nanofins to simultaneously control the phase
the combination of geometric phase and propagation and group delay dispersion, which also successfully
phase (Figure 3H), based on which they realized not resulted in broadband achromatic metalens in the visible
only multiplexed phase modulation under different (Figure 4B).[117]
polarizations,[79] but also simultaneously independent The FOV is one of the key indicators of imaging
control of amplitude, phase, and polarization.[100–102] systems, which is usually limited by off‐axis aberration
(especially coma aberration). In traditional large‐FOV
imaging systems, compound lenses with bulky thickness
3.4 | Metalens are used to eliminate coma aberrations. While for the
ultimate goal of compact imaging, people are working for a
On the basis of different types of metasurfaces, rich singlet or few‐layered metalenses to achieve the same
functionalities have been implemented, among which function. Thus, in recent years, great efforts have been
metalens is the most concerned one due to the extremely made to enlarge the FOV based on metalenses. However, a
wide applications in optical systems. Metalenses offer two traditional single metalens still has a limited FOV, due to
major advantages in imaging technology.[103] First, the the fact that the focusing phase profile can hardly be
CHEN ET AL. | 13
F I G U R E 4 Metalenses. (A) Reflective achromatic metalens with 450 nm bandwidth in infrared. Reproduced with permission.[78] Copyright
2017, Springer Nature. (B) Transmissive achromatic metalens with 200 nm bandwidth in visible. Reproduced with permission.[117] Copyright
2018, Springer Nature. (C) Metalens doublet with a viewing angle of ±30°. Reproduced with permission.[118] Copyright 2016, Springer Nature. (D)
Monolayer quadratic metalens with a viewing angle of ±70°. Reproduced with permission.[119] Copyright 2020, American Chemical Society. (E)
Dielectric‐gapped landscape quadratic metalens with a viewing angle near 180°. Reproduced with permission.[120] Copyright 2021, Wiley‐VCH.
(F) Metalens array‐based wide‐angle camera, which can realize high‐quality imaging with a viewing angle as large as 120°.[121]
satisfied for all incident angles.[124] Recently, several types high‐quality imaging with a viewing angle as large as
of metalens designed to enlarge FOV have been proposed, 120°, by a very compact meta device (Figure 4F).[121]
including optimized multiparametric geometries,[125,126]
monolayer quadratic metalens (Figure 4D),[119,127,128] air‐
gapped landscape quadratic metalens,[129–131] dielectric‐ 4 | META STRUCTURES DESIGN
gapped landscape quadratic metalens (Figure 4E),[120] and AND OPTIMIZATION
metalens doublets (Figure 4C).[118,132] However, these
methods either sacrifice imaging performances or imaging 4.1 | Design through numerical
efficiencies. Li's group proposed a metalens array‐based simulations
wide‐angle solution, through which the angular focusing
problem was transformed into the multiple distributions in Although there are clear physical models in classic
the planar dimension. And they successfully realized metastructures, it is often difficult to analytically derive
14 | CHEN ET AL.
F I G U R E 5 Metastructure simulation and optimization. (A) Schematic of a rectangular meta‐atom capped by SU‐8 polymer and the
simulated phase maps plotted for transverse electric (TE) and transverse magnetic (TM) polarizations. Reproduced with permission.[94] Copyright
2018, Springer Nature. (B) Photonic beam splitter designed by the direct‐binary search algorithm. Reproduced with permission.[134] Copyright
2015, Springer Nature. (C) Topology‐optimized freeform metasurfaces for beam deflector with ultrahigh efficiency. Reproduced with
permission.[135] Copyright 2017, American Chemical Society. (D) Multilayered 2D wide‐angle metalens designed by a topology‐optimized
algorithm. Reproduced with permission.[125] Copyright 2018, American Physical Society.
CHEN ET AL. | 15
the structure parameters, due to the complexity of the optimizers like the Genetic Algorithm (GA) will be a good
structure or the deviations from reality and ideal models. choice, for it is good at finding the optimal design from
So in most metamaterial works, people retrieve the large solution spaces for noncontiguous structure.[140] But if
structure parameters from numerical simulations by the structure is contiguous, the Ant Colony Optimization
using some commercial software, including Lumerical algorithm will be better than GA.[141] Then for problems
Finite‐Difference Time‐Domain (FDTD) solutions, COM- with continuous input space, there will be another
SOL Multiphysics, CST Microwave Studio, ANSOFT, and question: Is it a finite element problem? If yes, the topology
so forth. With this software, people can construct the optimization method will be a good candidate. Topology
models of metastructures and simulate the optical optimization refers to the idea of optimizing a 2D or 3D
responses under different parameters based on numerical system comprising an array of elements (Figure 5C,D),
algorithms, such as FDTD and Finite Element each of which contains parameters requiring
Method.[30,133] In early 3D metamaterials works, trans- adjustment.[125,135,142–144] Otherwise, a number of
mission and reflection of the composite metamaterials gradient‐based algorithms can be used for optimization,
are first calculated by the simulation software, and then including Newton's method[145] and the multiobjective
used to derive the permittivity and permeability. In Gradient Descent algorithm.[146] There are still other cases
metasurface design, generally, only a single unit model of problems and their corresponding suitable algorithms,
with periodic boundary conditions is simulated. Then the detailed discussion seen in Campbell et al.[30]
phase or transmission responses corresponding to differ- We can see that these optimization algorithms are
ent structure parameters are obtained from the simulated rule‐based approaches containing iterative searching
data library to guide the design of the entire metasurface steps in a case‐by‐case manner. It is not friendly for
device (Figure 5A). inexperienced researchers to select the correct algorithms
for different application scenes, which requires research-
ers to be familiar with the characteristics of each
4.2 | Inverse design by classical algorithm. In addition, these algorithms usually need a
optimization techniques lot of computation sources including power and time and
hence are insufficient for complex design while the
For some classical functions, under the guidance of dimension of parametric space and number of
physical models, the metadevice can generally be designed constraints increases.
by parametric sweeps in simulation software. However, if
the device function is too complex, the construction units
for metamaterials with the common physical model cannot 5 | DESIGN METASTRUCTURES
meet the requirement, and the parametric sweeps will be WITH AI ALGORITHMS
insufficient. Complex, compound, or even random shapes
of structures will probably work, the huge parameter space 5.1 | Basic concepts
makes the parametric sweeps almost impossible to
complete the structure design. In this section, we will In recent years, as an important part of AI algorithms,
introduce the inverse design methods, that is, find the data‐driven machine learning algorithms have developed
parameters of a metasurface to exhibit the desired function rapidly in the fields of industrial production, scientific
with the assistance of optimization algorithms. The inverse research, medical diagnosis, economic analysis, and
design problem can be solved as an optimization problem so on.[147] Machine learning algorithms establish links
that minimizes or maximize an objective function of between input parameters and output results by looking
the unit cell parameters subject to some constraints for the laws in a large amount of data. This process is
(Figure 5B).[29,31,134,136–139] The objective function is the known as the learning or training process, which can
mathematical expression of what functionality the constitute a mathematical model to connect the input
metadevice is to achieve. and output data. Once the model is trained, it can be
What kind of optimization algorithm is chosen for a used to make predictions on unknown data, which looks
particular metadevice design is of importance, since there is like intelligent behaviors. When training or using the
no single optimization algorithm that is perfectly suited for model, no physical or mathematical meaning of
all problems. To choose the best optimization strategy, one the specific problem needs to be understood, thus
needs to determine which category his design belongs to machine learning method is universal and can be used
according to some questions. For example, is the input to solve various problems, including classification,
space discrete? If yes, that means there is a finite number of regression, clustering, and structured prediction.[148–151]
potential solutions to search from. In this condition, global It should be noted that big data are of great importance
16 | CHEN ET AL.
for the performance of machine learning methods. final result is output by neurons in the output
Generally, higher quality and a larger amount of data layer.[157,158] The training process of the neural network
result in more accurate predictions. Usually, it takes a lot is using a large amount of data to continuously iterate
of time and effort to collect a large amount of data and the weights of connections between neurons with the
train the model, but using the model is convenient and help of methods, such as the error back‐propagation
time‐saving. So the machine learning method is more (BP) algorithm until the output results from the neural
suitable for situations where the model needs to be used network can be consistent with the ground truth. After
repeatedly after it is established. training, the ANN can accurately map the input to the
In the past few decades, machine learning has desired output.
developed many classical algorithms, including support
vector machine, decision tree, Bayesian classifier,
ANNs, and so forth. Among them, deep learning based 5.2 | Neural network‐assisted
on multilayer ANNs is one of the most popular machine metadesigns
learning methods, which has been successfully applied
in autonomous driving, target recognition, machine We can use ANN to establish relationships between the
translation, and speech recognition.[152–156] In a biolog- metamaterials structure parameters and the optical
ical neural network, the basic unit is a neuron, which is responses (Figure 6). The inverse process of the ANN is
connected to other neurons. If the electrical potential of to derive the corresponding parameters from the target
a neuron exceeds a threshold, it will be activated and function, which can be applied for metastructure design.
send chemicals to other connected neurons whose The forward process is also very useful, as it provides a
electrical potential thereby also be changed. ANN is a simple mathematical way to obtain physical phenomena
mathematical algorithm model that imitates the behav- from structure parameters, which otherwise needed to be
ioral characteristics of biological neural networks and calculated by time‐consuming FDTD or other simulation
performs distributed parallel information processing. In methods.
an ANN model, each neuron receives input signals from A representative example is using neural networks to
other connected neurons. These input signals are design structures of core–shell metamaterial unit, which is
transmitted through weighted connections to receive composed of multilayers with different thicknesses and
the total input value which will be then compared with dielectric permittivity (Figure 7B). Peurifoy et al. first
the neuron's threshold and processed through an applied ANN to study the scattering properties of core–shell
activation function to produce the neuron's output. An nanoparticles.[165] The inputs of the network are the
ANN consists of an input layer, an output layer, and thickness of the layers ranging from 30 to 70 nm and the
more than one hidden layer. The neurons in the input outputs are the scattering spectrums in the wavelength
layer receive external data, which then be processed range between 400 and 800 nm. After training, the ANN
through hidden layers and the output layer, and the can generate the scattering spectrum for a given set of
F I G U R E 6 Neural network‐assisted metadesigns. The ANN mathematically connects the metastructures and the physical phenomena.
The inverse process can be applied for intelligent metadesigns from target functions. The forward process is useful for the quick prediction
of physical responses from certain metastructures. ANN, artificial neural network.
CHEN ET AL. | 17
F I G U R E 7 Intelligent meta‐applications. (A) Full‐color generation with HfO2 metasurface designed by deep learning. Reproduced with
permission.[159] Copyright 2019, Royal Society of Chemistry. (B) An ANN for studying the scattering properties of multilayer core–shell
nanoparticles. Reproduced with permission.[160] Copyright 2019, American Chemical Society. (C) Phase recovery and holographic image
reconstruction using a deep neural network. Reproduced with permission.[161] Copyright 2018, Springer Nature. (D) High‐density planar
optical information storage is enabled by deep learning. Reproduced with permission.[162] Copyright 2019, Springer Nature.
(E) Programmable intelligent metasurface used for imaging and recognizing. Reproduced with permission.[163] Copyright 2019, Springer
Nature. (F) Deep‐learning‐enabled self‐adaptive microwave cloak without human intervention. Reproduced with permission.[164]
Copyright 2020, Springer Nature.
parameters very precisely within a fraction of a second. In supplement ANN with other methods were also proposed
addition, the ANN can also be used as an inverse design in the past 3 years to further improve the model
tool, which determines the structure parameters from a performances.[160,167,168]
target spectra. Hu et al. completed the same task with an Apart from the research on nanoparticle scattering
improved ANN structure called a modified denoising properties, more works focus on using ANNs to realize the
autoencoder neural network.[166] A series of studies that overall functions of metadevices, including metalenses,
18 | CHEN ET AL.
metaholograms, structural colors, and many other func- and self‐interference‐related artifacts (Figure 7C).[161] On the
tions. Sometimes if the parameter space is too large, or the basis of this achievement, 2 years later their group used a
metastructures are difficult to parameterize, simple fully GAN to perform cross‐modality image transformation from
connected ANN will not work well. Therefore, in practical a digitally BP hologram into a speckle‐ and artifact‐free
applications, some advanced ANNs models are used for bright‐field microscopy image.[173] This deep‐learning‐
metadevices design, such as convolutional neural net- enabled holographic microscopy enables the reconstruction
works (CNNs), recurrent neural networks, generative of volumetric samples from a single‐hologram measurement
adversarial networks (GAN), and some improved net- without any mechanical scanning. Gu's group used a
works based on these basic ones that are made according multilayer ANN to statistically learn the relationship
to specific problems. We would like to introduce between a given 3D vectorial field in the image space and
several examples of metadevices benefiting from neural the 2D vector field distribution in the hologram plane. And
network approaches. they successfully demonstrate 3D vectorial holography for
the time‐efficient and accurate reconstruction of 3D vectorial
holographic images.[174] Cui's group proposed an efficient
5.3 | Intelligent meta‐applications noniterative algorithm for designing metasurface holograms
by a deep‐learning structure composed of unsupervised
Starting from metalens, one of the most successful variational autoencoder (VAE) and conditional generative
applications of metasurfaces. An et al. proposed a adversarial networks (cGANs). The usage of unsupervised
broadband achromatic metalens in the near‐infrared design VAE structure makes their system easily trained from
based on deep neural networks (DNN), which connect scratch and the noniterative structure of the generator in
phases of metaunits and their geometric parameters. Two cGANs enables the realization of holographic imaging with
achromatic metalenses composed of elliptical nanopillars high quality and high efficiency.[175] Ma et al. proposed and
and nanofins were designed by the DNN and simulated, experimentally demonstrated the application of statistical
demonstrating the validity of their method.[169] Fan and Su machine learning in the design and optimization of complex
combined different numbers of nanofins as the metalens multifunctional metasurfaces.[176] They successfully realized
unit and used DNN to train their data library. The well‐ metasurface focusing lenses and holograms, which showed
trained model then predicted new data with the higher up to eight controllable functionalities for different combi-
transmission, which could contribute to a broadband nations of working frequencies and polarization states, far
achromatic metalens in the visible with focusing efficiency beyond the design capability of traditional approaches.
as high as 45%.[170] Gu et al. designed a bifocal metalens Besides, researchers have also used deep‐learning methods
that focuses the orthogonal polarization components of to achieve metasurface holograms with complex functions
incident light to two independent spots. In their work, a or high performances, such as multicolor holograms[177] and
DNN is trained to refine the unit cell simulation result of high‐efficiency holograms.[178]
parameter sweep and benefits the efficient design of the Structural colors based on absorbing or scattering of
metalens.[171] Zarei and Khavasi employed an inverse light in metasurfaces may be widely used for digital
design method combining FDTD parameter sweep with displays, cryptography, optical data storage, security, and
Fourier‐optics simulations, which performs like error BP in many other applications.[179,180] The deep‐learning
neural networks. Through this method, a number of method will be a powerful tool to minimize the
thermally controlled varifocal metalens doublets and computation cost and maximize the design efficiency
triplets were designed based on the thermo‐optical effect for structural color design. Yu's group trained a
of silicon.[172] bidirectional DNN, the forward modeling process which
Holography is also an important application of can accurately predict the color generated by random
metasurfaces, which can fully demonstrate the flexible silicon nanostructures, while the inverse design process
phase control ability of metasurfaces. With the help of which can accurately output the device geometries for at
deep‐learning methods, researchers have successfully im- least one million different colors.[181] Hemmatyar et al.
proved the imaging quality, imaging speed, and imaging used a deep‐learning algorithm based on a dimensional-
efficiency of metasurface holography. Ozcan's group ity reduction technique to design and optimize dielectric
demonstrated that a CNN can perform phase recovery metasurfaces consisting of a square‐lattice array of hafnia
and holographic image reconstruction after training. nanopillars and generated vivid and pure structural color
Compared with existing holographic phase recovery with a wide color gamut (Figure 7A).[159] Rho's group
approaches, the deep‐learning‐based technique they employed a reinforcement learning method called
employed is faster to compute and reconstruct Q‐learning for optimization of color generation from
high‐quality images of objects by eliminating twin‐image dielectric nanostructures. The model is capable of finding
CHEN ET AL. | 19
the best possible solution in millions of choices without of programmable meta‐atoms can be quickly switched, it
ever experiencing it, thus helping the authors to find enables the information coding and processing on the
geometrical properties from a possible 34.5 million physical level of metamaterials in real‐time.[198] Intelligent
solutions that can generate much purer red, green, and algorithms are extremely useful for realizing self‐adaptive
blue colors compared with other reported results.[182] programmable metamaterials, for the reason that they can
Macias' group proposed a metamodel‐based optimization be used as a quick feedback mechanism, redefining
scheme that combines ANNs and a particle swarm programming sequences of the programmable metamater-
optimization to tailor the chromatic response of periodic ials to adjust to environmental changes. Li et al. combined
metastructures, including one‐dimensional gratings[183] machine learning techniques with a 2‐bit digital metasur-
and arrays of nanoparticles with arbitrary cross‐ face to realize a microwave reprogrammable imager. After
section,[184] and experimentally reproduced arbitrarily a period of training, the proposed microwave imager can
prescribed colors from the employed model. In addition, directly produce high‐quality imaging and high‐accuracy
there are multiple other works using deep learning to object recognition without the need for costly computa-
efficiently design structural colors in plasmonic and tional image reconstruction.[199] They also used an
metamaterial structures.[185–187] integrated network composed of three ANNs to simulta-
Other practical metasurfaces applications designed by neously realize a smart metasurface imager and recognizer.
intelligent algorithms include light sails,[188,189] optical The network they trained can transform measured micro-
information storage (Figure 7D),[162] perfect absorbers,[190] wave data into images of the whole human body, classify
biosensors,[191–194] and energy conversions.[195] In these specifically designated spots within the image and recog-
metasurface designs, intelligent algorithms perform not nize human hand signs instantly at a Wi‐Fi frequency of
only faster, but also more powerful than traditional 2.4 GHz (Figure 7E).[163]
methods. On the one hand, intelligent algorithms allow a In the field of wireless communications, intelligent
much larger parameter space for optimization, which reflecting surfaces (IRSs) recently have received much
increases the probability of finding the optimal solution. interest in the academia and industry for they are
On the other hand, intelligent algorithms establish special promising in next‐generation (i.e., 6G) communications to
correlations between metasurfaces structures and physical achieve faster and more reliable data transmissions.[200,201]
phenomena, which are difficult to be represented by IRS comprises an array of reflective units, each of which
physical models or mathematical expressions. Most of the can independently incur some change to the incident
introduced metadevices above are for specific application signal, such as the phase, amplitude, frequency, or
scenarios, as the structures are fixed after design. Besides, polarization, without consuming transmit power. In
there are some works that applied dynamically reconfigur- essence, an IRS intelligently configures the wireless
able metamaterials. With the assistance of intelligent environment to help the transmission between the sender
algorithms, the metastructures can make adaptive adjust- and receiver, when direct communications have bad
ments according to external environment changes or qualities. Liaskos et al. used deep learning for configuring
functions switch, which look more like an intelligent IRS to aid wireless communications. Specifically, wireless
device. These metamaterials are often termed digital propagation is regarded as a neural network in their work,
metamaterials or programmable metamaterials, which we where IRS units are neurons and their cross‐interactions
will discuss below. are links. The trained network could learn the propagation
basics of IRS and configures them to the optimal
setting.[202] Taha et al. trained an ANN using only IRS
5.4 | Reconfigurable metamaterials units connected to the baseband of the IRS controller,
with self‐adaptive properties which was then used for channel estimation in IRS
communications and guiding the IRS to learn the optimal
Recently the concept of digital metamaterials has been interaction with the incident signals.[203]
proposed,[196,197] the constitutive 1‐bit meta‐atom is digita- In recent years, research has tried to endow
lized as “0” or “1” corresponding to two opposite EM invisibility cloaks with intelligence, independence, and
responses. By incorporating an electric diode into the unit self‐adaptability. Chen's group applied deep‐learning
cell of metamaterials, the scattering state can be controlled algorithms into an invisibility cloak and built an
by applying different biased voltages across the diode. intelligent cloak that exhibited a millisecond response
When the digital meta‐atom is programmable, a metama- time to an everchanging incident wave and the sur-
terial can be used to realize different functions when rounding environment without any human intervention
programmed with different coding sequences via the field (Figure 7F).[164] The reconfigurable metasurface they
programmable gate array (FPGA). Moreover, as the states used is composed of varactor diodes, the reflection
20 | CHEN ET AL.
spectrum of which can be dynamically altered by feeding lens size increases. Since the lateral parameters of
different bias voltages across it. Incoming waves sensed nanostructures cannot change so large in very limited
by detectors in real‐time are processed by the ANN to unit cells, to access a large compensation phase these
calculate voltages, which then will be applied across the nanoposts need to have a very large height with a large
diodes to adjust the scattering spectrum and thus aspect ratio that undoubtedly adds great challenges to
produce a scattered wave similar to that generated by nanofabrication. In fact, there have been several papers
the bare surrounding without a hidden object. concerning on the issue, and the relation between the
maximum radius of the ideal achromatic lens was
derived approximately as[206]
6 | DISCUS SIONS A ND
PERSPECTIVES H Δn NA
R  ωc . (8)
Δω 1 − 1 − NA2
After more than 10 years of development, metamaterials
research has undergone a series of changes from physical From Equation (8) we may find the Rmax only has a
phenomenon demonstration to devices development, from positive relationship with the height (H) of nanoposts,
exploring principles to improving devices performances, while is inverse to the bandwidth (Δω) and NA. This
from single function to multifunctional integration, and relation is also valid for another kind of flat lens—
from conventional design to intelligent design. Although multilevel diffractive lens (MDL), which can be regarded
the working principles of metamaterials have been well as a paralleled strategy for an achromatic flat lens.[207]
studied and based on which many practical applications Different from metalens, the MDL stemmed from the
have been realized, there are still some key scientific diffractive optics with optimized complex ring heights,
issues and technical bottlenecks waiting to be solved and where sufficient phase compensation against the mate-
overcome especially in metasurfaces applications. In this rial and diffractive dispersion can be properly achieved.
section, we will comment on some directions, challenges, There have been a number of achromatic MDL reports
and perspectives of metamaterial research. showing relatively larger lens diameters compared with
metalens thanks to its higher thickness, though it may
have some flaws in angular dispersion and efficiency.
6.1 | Full control of metamaterial Undoubtedly, there is still improvement space by further
angular dispersions increasing the thickness with structural optimizations.
A key scientific problem that currently restricts metalens

and related metadevices is the uncontrollability of their 6.3 | Active control of metamaterials
phase distributions to incident EM waves at different in the optical regime
angles. It is a very important issue in metalens aberration
correction and wide working angle metadevices. Although Controlling the optical properties of metamaterials
some angular dispersion has been manipulated by actively in real‐time could take their functionalities to
coupling effect[204] or tilting units,[205] the control abilities the next level and expand the boundaries of fundamental
and performances are still far from arbitrary angular optical science.[208] In the millimeter‐wave and micro-
control. In particular, the problem of low modulation wave regime, varactor diodes have been successfully
efficiency to large‐angle incident light has not been well utilized to constitute reconfigurable metasurfaces, which
solved in principle, which will be the bottleneck in can accomplish various functions dynamically via
realizing high‐efficiency wide‐angle metadevices. FPGA.[196] However, this kind of diode is no longer
applicable in the visible and near‐infrared, researchers
have proposed a series of other means to achieve
6.2 | Constraints among metasurface dynamic control function either by tuning the metama-
frequency dispersion, phase distribution, terial's dielectric permittivity or changing its topology.
and working efficiency One way to tune a metamaterial's properties in real‐time
is reconfiguring the physical shape or arrangement of its
In achromatic metadesigns, the large diffractive disper- nanoantennas through micro‐electro‐mechanical sys-
sion needs to be compensated by the constituting tems, with which a varifocal metalens is realized with a
nanostructures, which mainly depends on the propaga- switching speed on the order of kHz to MHz.[209] Liu's
tion phase in the transmission scheme. The required group realized dynamic generation of programmable
phase compensation would be huge as the achromatic images by integrating metamaterials with LCs,[210] which
CHEN ET AL. | 21
can also be used for real‐time scanning and 3D imaging. through FDTD simulations or experimental characteriza-
By using metamaterials fabricated on top of phase‐ tions. Another challenge is that the large solution space
change materials, large amplitude and phase changes will make it difficult for even a very complicated neural
will be achieved under a thermal or electrical network to learn the intrinsic relationship between
impulse.[211] Through this method, dynamical metasur- metastructures and the physical responses.[215] New
face beam‐steering and reconfigurable diffraction‐limited approaches that extend beyond standard deep‐learning
metalenses have been realized.[212,213] Besides electrical methods are required for large‐scale metamaterial design.
control is a more widely used technique to tune a One possible way to address this problem is to use
material's optical properties. Transparent conducting unsupervised or semisupervised learning methods. They
oxides materials enable strong modulation of their require only a small number of labels and usually serve as
optical properties under an applied bias, with which tools for data clustering and dimensionality reduction,
Park et al. achieved wide‐angle beam steering with which could remarkably reduce the time consumption for
independent control of phase and amplitude.[214] The preparing training data.[216,217] Using the insights from
above‐mentioned methods successfully realized active physics as a model prior to deep learning is also an
control of metamaterials in specific functions. Active effective way when the amount of training data is limited.
metamaterials with high switch speed (GHz to THz), Different from the nonphysical applications, metamaterial
high reliability, and the ability that each pixel is designing is governed by certain physical laws or models,
controllable, still remain to be studied, which are of which if applied to deep learning, the training data and
great significance in optical communications, light network parameters are not wasted on “learning the
detection and ranging (LIDAR) and 3D imaging devices, physics.” This will undoubtedly not only simplify the
such as augmented and virtual reality. deep‐learning model but also skip the collection of a large
amount of data.[218–220]
6.4 | Fabrication of large‐scale

metamaterial devices 6.6 | Optical ANNs enabled
by metamaterials
Since the structure size of each metamaterial unit is on
the scale of tens of nanometers, it usually requires Besides employing intelligent algorithms for nanophotonics
trillions of units to build a macroscopic metadevice, and design as we have discussed above, another meaningful
each unit also has multiple structural parameters. combination of intelligent algorithms and nanophotonics is
Therefore, if all the parameters of the units do not have constructing photonic neural networks that could emulate
repeatability or regularity, the parameters that need to be the structural and functional features of biological neural
input in the fabrication process will reach the order of networks. As the photon is an important information carrier
Tbit. Although there are some data compression meth- exhibiting a broad bandwidth and low transmission
ods, to achieve universal large‐area fabrication, data scattering compared with electrons, the development of
import and export will be a huge challenge. At the same ANNs based on nanophotonics devices will achieve orders‐
time, it is also a huge challenge to ensure the accuracy of‐magnitude improvements in both computational speed
and uniformity of each unit on such a scale, as well as the and energy consumption over existing solutions based on
stability of the fabrication equipment. Recent studies electronics.[221] In recent years, a variety of ways to
have shown that the higher performance of metasurfaces implement optical neural networks have been proposed,
will lead to larger aspect ratio nanopillar structural units such as ANNs based on layers of diffractive materials in free
(such as 100:1), which is also a huge challenge for space (Figure 8A)[222,226] and ANNs based on photonic
current micro–nano fabrication technology. circuits on a chip.[227,228] As a kind of flexible diffractive
material, metasurfaces or 3D metamaterials are very suitable
for implementing the first kind of optical ANNs. Luo et al.
6.5 | Intelligent optimization realized a polarization‐multiplexed metasurface‐based
of large‐scale metamaterials optical neural network that can perform various complex
object recognition tasks.[229] The intelligent device they
Similar to the problem in the fabrication of large‐scale proposed is composed of one layer of metasurface integrated
metamaterial, when the metamaterial size increases, into a complementary metal‐oxide semiconductor (CMOS)
the number of structural parameters will explode.One imaging sensor, which can mimic an ANN with a single
challenge is that a training set with a large amount of data hidden layer. Gu's group fabricated multilayer diffractive
is needed to be produced first, which is time‐consuming metastructures by 3D nanoprinting which were also
22 | CHEN ET AL.
F I G U R E 8 Metamaterials‐based optical neural network. (A) Multilayer diffractive deep neural network for classifying and imaging.
Reproduced with permission.[222] Copyright 2018, American Association for the Advancement of Science. (B) Machine learning decryptor
integrated with a CMOS sensor. Reproduced with permission.[223] Copyright 2021, Springer Nature. (C) Schematic illustration of optical
logic operations by a diffractive neural network. Reproduced with permission.[224] Copyright 2020, Springer Nature. (D) A programmable
diffractive deep neural network platform able to handle various intelligent tasks. Reproduced with permission.[225] Copyright 2022, Springer
Nature. CMOS, complementary metal‐oxide semiconductor.
integrated on a CMOS chip (Figure 8B). Their experimental realization of optical nonlinear activation functions such
results demonstrated the application of their device as as laser‐cooled atom with electromagnetically induced
power‐efficient optical decryptors and secure functional transparency[230] or using nonlinear materials, including
displays.[223] Chen's group used the compound Huygens' crystals, polymers, or semiconductors.[231] However,
metasurface to construct a diffractive neural network and how can these methods be combined with multilayer
realized all seven basic types of optical logic operations in a metasurfaces, especially dynamically tunable ones in the
compact system (Figure 8C).[224] The above‐mentioned optical regime, still requires a lot of researches.
metasurfaces‐based optical neural networks are all passive
ones that are once these devices are fabricated only a fixed
intelligent task can be performed. Cui's group applied their 6.7 | Metasurfaces integrated with
digital‐coding metasurfaces technique to constructing a other nanostructures
programmable diffractive DNN which they termed a
programmable AI machine (Figure 8D).[225] This intelligent Owning to the excellent flexibility in light‐field manipula-
machine can handle various deep‐learning tasks for wave tion, metasurfaces have been combined with other
sensing, including image classification, mobile communica- materials or nanostructures to realize ultracompact devices
tion coding, and real‐time multibeam focusing fully showing with high performances. For example, researchers fabri-
what the next‐generation of intelligent materials will look cated hybrid metasurfaces by combining plasmonic nanos-
like. Although a series of inspiring results have been tructures with transition metal dichalcogenides (TMDCs),
achieved on metastructures‐based optical neural networks, such as WS2, which can simultaneously enhance and
there is still room for improvement. One key issue is the manipulate the nonlinear processes in TMDC, facilitating
realization of the optical nonlinear activation function, nonlinear valleytronic nanodevices.[232–234] Li et al. inte-
which can endow optical neural networks with certain grated a metalens array with a nonlinear crystal, and
computational power for better performances in learning demonstrated a 100‐path spontaneous parametric down‐
and predicting. Some previously reported optical neural conversion photon‐pair source in a 10 × 10 array, indicating
networks lack the nonlinear activation part, which leads to a new platform for integrated quantum devices.[235] Xie
their poor performances and limited functionalities. In et al. realized monolithic integration of dielectric metasur-
recent years, research has proposed several methods for the faces with vertical‐cavity surface‐emitting lasers that enable
CHEN ET AL. | 23
arbitrary control of the laser beam profiles with high [7] Liu Y, Zhang X. Metamaterials: a new frontier of science
efficiency.[236] This function can be implemented in various and technology. Chem Soc Rev. 2011;40(5):2494‐2507.
wide‐field applications, such as optical fiber communica- [8] Zheludev NI, Kivshar YS. From metamaterials to metade-
vices. Nat Mater. 2012;11(11):917‐924.
tions, laser printing, face recognition, and ultracompact
[9] Jackson JD. Classical electrodynamics. Wiley; 1999.
LIDAR.
[10] Veselago VG. The electrodynamics of substance with
In conclusion, the current metamaterial research has simultaneously negative values of ε and μ. Sov Phys Usp.
progressed from the initial study of the EM response of 1968;10(4):509‐514.
nanostructures and the simple control of the optical field [11] Yu N, Genevet P, Kats MA, et al. Light propagation with
to the development of photonics devices with specific phase discontinuities: generalized laws of reflection and
functionalities and performance enhancement. As a large refraction. Science. 2011;334(6054):333‐337.
number of metamaterials‐based works have been dem- [12] Yu N, Capasso F. Flat optics with designer metasurfaces.
Nat Mater. 2014;13(2):139‐150.
onstrated, the bottleneck problems encountered in
[13] Genevet P, Capasso F, Aieta F, Khorasaninejad M, Devlin R.
practical applications have gradually become prominent
Recent advances in planar optics: from plasmonic to
and attracted researchers' attentions. However, with the dielectric metasurfaces. Optica. 2017;4(1):139‐152.
continuous progress of new design methods, new [14] Kruk S, Kivshar Y. Functional meta‐optics and nanophotonics
material systems, and fabrication technologies, we governed by Mie resonances. ACS Photonics. 2017;4(11):
believe that the key problems discussed above can be 2638‐2649.
further solved. Especially with the utilization of AI [15] Aieta F, Genevet P, Kats MA, et al. Aberration‐free ultrathin
photonics technologies, many fundamental problems in flat lenses and axicons at telecom wavelengths based on
plasmonic metasurfaces. Nano Lett. 2012;12(9):4932‐4936.
hardware can be alleviated by sophisticated algorithms,
[16] Liu Z, Li Z, Liu Z, et al. High‐performance broadband circularly
which also bring excellent performances to the new polarized beam deflector by mirror effect of multinanorod
generation of metamaterials‐based photonic devices. metasurfaces. Adv Funct Mater. 2015;25(34):5428‐5434.
[17] Bao Y, Yu Y, Xu H, et al. Full‐colour nanoprint‐hologram
ACKNOWLEDGMENTS synchronous metasurface with arbitrary hue‐saturation‐
This study is supported by the National Key R&D brightness control. Light Sci Appl. 2019;8(1):1‐10.
Program of China (2016YFA0202103), the National [18] Liang H, Martins A, Borges BHV, et al. High performance
metalenses: numerical aperture, aberrations, chromaticity,
Natural Science Foundation of China (Nos. 12104223
and trade‐offs. Optica. 2019;6(12):1461‐1470.
and 91850204), and the Fundamental Research Funds for
[19] Lee D, Gwak J, Badloe T, Palomba S, Rho J. Metasurfaces‐
the Central Universities (Nos. 2242022R10128 and based imaging and applications: from miniaturized optical
2242022k30006). Tao Li thanks the support from components to functional imaging platforms. Nanoscale
Dengfeng Project B of Nanjing University. Adv. 2020;2(2):605‐625.
[20] Chen WT, Zhu AY, Capasso F. Flat optics with dispersion‐
CONFLI CT OF I NTER EST engineered metasurfaces. Nat Rev Mater. 2020;5(8):604‐620.
The authors declare no conflict of interest. [21] Ni X, Wong ZJ, Mrejen M, Wang Y, Zhang X. An ultrathin
invisibility skin cloak for visible light. Science. 2015;349(6254):
1310‐1314.
ORCID
[22] Wang L, Kruk S, Tang H, et al. Grayscale transparent
Ji Chen http://orcid.org/0000-0002-9376-6077 metasurface holograms. Optica. 2016;3(12):1504‐1505.
Shanshan Hu http://orcid.org/0000-0002-0174-1156 [23] Totero Gongora JS, Miroshnichenko AE, Kivshar YS,
Shining Zhu http://orcid.org/0000-0002-3472-6497 Fratalocchi A. Anapole nanolasers for mode‐locking and
Tao Li http://orcid.org/0000-0003-0049-471X ultrafast pulse generation. Nat Commun. 2017;8(1):1‐9.
[24] Khorasaninejad M, Capasso F. Metalenses: versatile multi-
REFERENCES functional photonic components. Science. 2017;358(6367):
[1] Engheta N, Ziolkowski RW. Metamaterials: Physics and eaam8100.
Engineering Explorations. John Wiley & Sons; 2006. [25] Lalanne P, Chavel P. Metalenses at visible wavelengths:
[2] Cai W, Shalaev V, Paul DK. Optical metamaterials: funda- past, present, perspectives. Laser Photon Rev. 2017;11(3):
mentals and applications. Phys Today. 2010;63(9):57‐58. 1600295.
[3] Smith DR, Pendry JB, Wiltshire MCK. Metamaterials and [26] Tittl A, Leitis A, Liu M, et al. Imaging‐based molecular
negative refractive index. Science. 2004;305(5685):788‐792. barcoding with pixelated dielectric metasurfaces. Science.
[4] Ramakrishna SA. Physics of negative refractive index 2018;360(6393):1105‐1109.
materials. Rep Prog Phys. 2005;68(2):449‐521. [27] Krasnok A, Tymchenko M, Alù A. Nonlinear metasurfaces: a
[5] Fang N, Xi D, Xu J, et al. Ultrasonic metamaterials with paradigm shift in nonlinear optics. Mater Today. 2018;21(1):8‐21.
negative modulus. Nat Mater. 2006;5(6):452‐456. [28] Yesilkoy F, Arvelo ER, Jahani Y, et al. Ultrasensitive
[6] Fok L, Ambati M, Zhang X. Acoustic metamaterials. MRS hyperspectral imaging and biodetection enabled by
Bull. 2008;33(10):931‐934. dielectric metasurfaces. Nat Photonics. 2019;13(6):390‐396.
24 | CHEN ET AL.
[29] Molesky S, Lin Z, Piggott AY, Jin W, Vucković J, [48] Smith DR, Padilla WJ, Vier DC, Nemat‐Nasser SC,
Rodriguez AW. Inverse design in nanophotonics. Nat Schultz S. Composite medium with simultaneously negative
Photonics. 2018;12(11):659‐670. permeability and permittivity. Phys Rev Lett. 2000;84(18):
[30] Campbell SD, Sell D, Jenkins RP, Whiting EB, Fan JA, 4184‐4187.
Werner DH. Review of numerical optimization techniques for [49] Yen TJ, Padilla WJ, Fang N, et al. Terahertz magnetic
meta‐device design. Opt Mater Express. 2019;9(4):1842‐1863. response from artificial materials. Science. 2004;303(5663):
[31] Li W, Meng F, Chen Y, Li Yf, Huang X. Topology 1494‐1496.
optimization of photonic and phononic crystals and meta- [50] Linden S, Enkrich C, Wegener M, Zhou J, Koschny T,
materials: a review. Adv Theory Simul. 2019;2(7):1900017. Soukoulis CM. Magnetic response of metamaterials at 100
[32] Ma W, Liu Z, Kudyshev ZA, Boltasseva A, Cai W, Liu Y. terahertz. Science. 2004;306(5700):1351‐1353.
Deep learning for the design of photonic structures. Nat [51] Soukoulis CM, Linden S, Wegener M. Negative refractive
Photonics. 2021;15(2):77‐90. index at optical wavelengths. Science. 2007;315(5808):47‐49.
[33] Ma L, Li J, Liu Z, et al. Intelligent algorithms: new avenues for [52] Zhang S, Fan W, Panoiu NC, Malloy KJ, Osgood RM,
designing nanophotonic devices. Chin Opt Lett. 2021;19(1): Brueck SRJ. Experimental demonstration of near‐infrared
11301. negative‐index metamaterials. Phys Rev Lett. 2005;95(13):
[34] Krasikov S, Tranter A, Bogdanov A, Kivshar Y. Intelligent 137404.
metaphotonics empowered by machine learning. Opto‐ [53] Dolling G, Enkrich C, Wegener M, Soukoulis CM, Linden S.
Electron Adv. 2022;5(3):210124‐210147. doi:10.29026/oea. Simultaneous negative phase and group velocity of light in a
2022.210147 metamaterial. Science. 2006;312(5775):892‐894.
[35] Katsarakis N, Koschny T, Kafesaki M, Economou EN, [54] Zhang X, Liu Z. Superlenses to overcome the diffraction
Soukoulis CM. Electric coupling to the magnetic resonance limit. Nat Mater. 2008;7(6):435‐441.
of split ring resonators. Appl Phys Lett. 2004;84(15):2943‐2945. [55] Pendry JB. Negative refraction makes a perfect lens. Phys
[36] Baena JD, Bonache J, Martin F, et al. Equivalent‐circuit Rev Lett. 2000;85(18):3966‐3969.
models for split‐ring resonators and complementary split‐ [56] Grbic A, Eleftheriades GV. Overcoming the diffraction limit
ring resonators coupled to planar transmission lines. IEEE with a planar left‐handed transmission‐line lens. Phys Rev
Trans Microwave Theory Tech. 2005;53(4):1451‐1461. Lett. 2004;92(11):117403.
[37] Zhou J, Koschny T, Kafesaki M, Economou EN, Pendry JB, [57] Leonhardt U, Philbin TG. Transformation optics and the
Soukoulis CM. Saturation of the magnetic response of split‐ geometry of light. In: Progress in Optics. Vol 53. Elsevier;
ring resonators at optical frequencies. Phys Rev Lett. 2009:69‐152.
2005;95(22):223902. [58] Chen H, Chan CT, Sheng P. Transformation optics and
[38] Aydin K, Bulu I, Guven K, Kafesaki M, Soukoulis CM, metamaterials. Nat Mater. 2010;9(5):387‐396.
Ozbay E. Investigation of magnetic resonances for different [59] Pendry JB, Aubry A, Smith DR, Maier SA. Transformation
split‐ring resonator parameters and designs. New J Phys. optics and subwavelength control of light. Science. 2012;
2005;7(1):168. 337(6094):549‐552.
[39] Pendry JB, Holden AJ, Robbins DJ, Stewart WJ. Magnetism [60] Leonhardt U. Optical conformal mapping. Science. 2006;
from conductors and enhanced nonlinear phenomena. IEEE 312(5781):1777‐1780.
Trans Microwave Theory Tech. 1999;47(11):2075‐2084. [61] Li J, Pendry JB. Hiding under the carpet: a new strategy for
[40] Pendry JB, Holden AJ, Stewart WJ, Youngs I. Extremely low cloaking. Phys Rev Lett. 2008;101(20):203901.
frequency plasmons in metallic mesostructures. Phys Rev [62] Valentine J, Li J, Zentgraf T, Bartal G, Zhang X. An optical
Lett. 1996;76(25):4773‐4776. cloak made of dielectrics. Nat Mater. 2009;8(7):568‐571.
[41] Shelby RA, Smith DR, Schultz S. Experimental verification of [63] Gabrielli LH, Cardenas J, Poitras CB, Lipson M. Silicon
a negative index of refraction. Science. 2001;292(5514):77‐79. nanostructure cloak operating at optical frequencies. Nat
[42] Fang N, Lee H, Sun C, Zhang X. Sub‐diffraction‐limited Photonics. 2009;3(8):461‐463.
optical imaging with a silver superlens. Science. 2005;308 [64] Ergin T, Stenger N, Brenner P, Pendry JB, Wegener M.
(5721):534‐537. Three‐dimensional invisibility cloak at optical wavelengths.
[43] Liu Z, Lee H, Xiong Y, Sun C, Zhang X. Far‐field optical Science. 2010;328(5976):337‐339.
hyperlens magnifying sub‐diffraction‐limited objects. Science. [65] Gharghi M, Gladden C, Zentgraf T, et al. A carpet cloak for
2007;315(5819):1686. visible light. Nano Lett. 2011;11(7):2825‐2828.
[44] Pendry JB, Schurig D, Smith DR. Controlling electromag- [66] Zhang S, Genov DA, Sun C, Zhang X. Cloaking of matter
netic fields. Science. 2006;312(5781):1780‐1782. waves. Phys Rev Lett. 2008;100(12):123002.
[45] Schurig D, Mock JJ, Justice BJ, et al. Metamaterial [67] Farhat M, Guenneau S, Enoch S. Ultrabroadband elastic
electromagnetic cloak at microwave frequencies. Science. cloaking in thin plates. Phys Rev Lett. 2009;103(2):24301.
2006;314(5801):977‐980. [68] Zhang S, Xia C, Fang N. Broadband acoustic cloak for
[46] Lai Y, Ng J, Chen H, et al. Illusion optics: the optical ultrasound waves. Phys Rev Lett. 2011;106(2):24301.
transformation of an object into another object. Phys Rev [69] Zigoneanu L, Popa BI, Cummer SA. Three‐dimensional
Lett. 2009;102(25):253902. broadband omnidirectional acoustic ground cloak. Nat
[47] Sheng C, Liu H, Wang Y, Zhu SN, Genov DA. Trapping light Mater. 2014;13(4):352‐355.
by mimicking gravitational lensing. Nat Photonics. [70] Xu H, Shi X, Gao F, Sun H, Zhang B. Ultrathin three‐
2013;7(11):902‐906. dimensional thermal cloak. Phys Rev Lett. 2014;112(5):54301.
CHEN ET AL. | 25
[71] Park J, Youn JR, Song YS. Hydrodynamic metamaterial [91] Fang B, Wang Z, Gao S, Zhu S, Li T. Manipulating guided
cloak for drag‐free flow. Phys Rev Lett. 2019;123(7):74502. wave radiation with integrated geometric metasurface.
[72] Lai Y, Ng J, Chen HY, Zhang ZQ, Chan CT. Illusion optics. Nanophotonics. 2022;11(9):1923‐1930.
Front Phys China. 2010;5(3):308‐318. [92] Hsiao H, Chu CH, Tsai DP. Fundamentals and applications
[73] Sheng C, Bekenstein R, Liu H, Zhu S, Segev M. Wavefront of metasurfaces. Small Methods. 2017;1(4):1600064.
shaping through emulated curved space in waveguide [93] Kamali SM, Arbabi E, Arbabi A, Faraon A. A review of
settings. Nat Commun. 2016;7(1):1‐8. dielectric optical metasurfaces for wavefront control.
[74] Ni X, Ishii S, Kildishev AV, Shalaev VM. Ultra‐thin, planar, Nanophotonics. 2018;7(6):1041‐1068.
Babinet‐inverted plasmonic metalenses. Light Sci Appl. [94] Faraji‐Dana M, Arbabi E, Arbabi A, Kamali SM, Kwon H,
2013;2(4):e72. Faraon A. Compact folded metasurface spectrometer. Nat
[75] Decker M, Staude I, Falkner M, et al. High‐efficiency dielectric Commun. 2018;9(1):1‐8.
Huygens' surfaces. Adv Opt Mater. 2015;3(6):813‐820. [95] Arbabi E, Kamali SM, Arbabi A, Faraon A. Vectorial
[76] Huang L, Chen X, Mühlenbernd H, et al. Dispersionless holograms with a dielectric metasurface: ultimate polariza-
phase discontinuities for controlling light propagation. tion pattern generation. ACS Photonics. 2019;6(11):
Nano Lett. 2012;12(11):5750‐5755. 2712‐2718.
[77] Arbabi A, Horie Y, Ball AJ, Bagheri M, Faraon A. [96] Faraji‐Dana M, Arbabi E, Kwon H, et al. Hyperspectral
Subwavelength‐thick lenses with high numerical apertures imager with folded metasurface optics. ACS Photonics.
and large efficiency based on high‐contrast transmitarrays. 2019;6(8):2161‐2167.
Nat Commun. 2015;6(1):1‐6. [97] Kamali SM, Arbabi E, Faraon A. Metasurface‐based
[78] Wang S, Wu PC, Su VC, et al. Broadband achromatic optical compact light engine for AR headsets. In: Optical Design
metasurface devices. Nat Commun. 2017;8:1‐9. Challenge. Vol 11040. SPIE; 2019:1104002.
[79] Mueller JPB, Rubin NA, Devlin RC, Groever B, Capasso F. [98] Kamali SM, Arbabi E, Kwon H, Faraon A. Metasurface‐
Metasurface polarization optics: independent phase control generated complex 3‐dimensional optical fields for interfer-
of arbitrary orthogonal states of polarization. Phys Rev Lett. ence lithography. Proc Natl Acad Sci. 2019;116(43):
2017;118(11):113901. 21379‐21384.
[80] Sun S, He Q, Xiao S, Xu Q, Li X, Zhou L. Gradient‐index [99] Fan ZB, Shao ZK, Xie MY, et al. Silicon nitride metalenses
meta‐surfaces as a bridge linking propagating waves and for close‐to‐one numerical aperture and wide‐angle visible
surface waves. Nat Mater. 2012;11(5):426‐431. imaging. Phys Rev Appl. 2018;10(1):14005.
[81] Chen J, Li T, Wang S, Zhu S. Multiplexed holograms by [100] Arbabi A, Horie Y, Bagheri M, Faraon A. Dielectric
surface plasmon propagation and polarized scattering. Nano metasurfaces for complete control of phase and polarization
Lett. 2017;17(8):5051‐5055. with subwavelength spatial resolution and high transmis-
[82] Chen J, Chen X, Li T, Zhu S. On‐chip detection of orbital sion. Nat Nanotechnol. 2015;10(11):937‐943.
angular momentum beam by plasmonic nanogratings. Laser [101] Wang H, Zhang Z, Zhao K, Liu W, Wang P, Lu Y.
Photon Rev. 2018;12(8):1700331. Independent phase manipulation of co‐ and cross‐
[83] Shamkhi HK, Baryshnikova KV, Sayanskiy A, et al. polarizations with all‐dielectric metasurface. Chin Opt Lett.
Transverse scattering and generalized Kerker effects in all‐ 2021;19(5):53601.
dielectric Mie‐resonant metaoptics. Phys Rev Lett. 2019; [102] Fan Q, Liu M, Zhang C, et al. Independent amplitude
122(19):193905. control of arbitrary orthogonal states of polarization via
[84] Shitrit N, Bretner I, Gorodetski Y, Kleiner V, Hasman E. dielectric metasurfaces. Phys Rev Lett. 2020;125(26):267402.
Optical spin Hall effects in plasmonic chains. Nano Lett. [103] Li T. New opportunities for metalenses in imaging applica-
2011;11(5):2038‐2042. tions. Sci China: Phys Mech Astron. 2020;63:284231.
[85] Kang M, Chen J, Wang XL, Wang HT. Twisted vector field [104] Chen X, Huang L, Mühlenbernd H, et al. Dual‐polarity
from an inhomogeneous and anisotropic metamaterial. plasmonic metalens for visible light. Nat Commun.
J Opt Soc Am B. 2012;29(4):572‐576. 2012;3(1):1‐6.
[86] Kang M, Feng T, Wang HT, Li J. Wave front engineering [105] Monticone F, Estakhri NM, Alu A. Full control of nanoscale
from an array of thin aperture antennas. Opt Express. optical transmission with a composite metascreen. Phys Rev
2012;20(14):15882‐15890. Lett. 2013;110(20):203903.
[87] Huang L, Chen X, Mühlenbernd H, et al. Three‐ [106] Arbabi A, Faraon A. Fundamental limits of ultrathin
dimensional optical holography using a plasmonic metasur- metasurfaces. Sci Rep. 2017;7(1):1‐9.
face. Nat Commun. 2013;4(1):1‐8. [107] Khorasaninejad M, Chen WT, Devlin RC, Oh J, Zhu AY,
[88] Zheng G, Mühlenbernd H, Kenney M, Li G, Zentgraf T, Capasso F. Metalenses at visible wavelengths: diffraction‐
Zhang S. Metasurface holograms reaching 80% efficiency. limited focusing and subwavelength resolution imaging.
Nat Nanotechnol. 2015;10(4):308‐312. Science. 2016;352(6290):1190‐1194.
[89] Li J, Yuan Y, Wu Q, Burokur SN, Zhang K. Dual‐band [108] Avayu O, Almeida E, Prior Y, Ellenbogen T. Composite
independent phase control based on high efficiency functional metasurfaces for multispectral achromatic optics.
metasurface. Chin Opt Lett. 2021;19(10):100501. Nat Commun. 2017;8(1):1‐7.
[90] Chen C, Song W, Chen JW, et al. Spectral tomographic [109] Lin D, Holsteen AL, Maguid E, et al. Photonic multitasking
imaging with aplanatic metalens. Light Sci Appl. 2019;8(1): interleaved Si nanoantenna phased array. Nano Lett.
1‐8. 2016;16(12):7671‐7676.
26 | CHEN ET AL.
[110] Li L, Yuan Q, Chen R, et al. Chromatic dispersion [129] Shalaginov MY, An S, Yang F, et al. Single‐element
manipulation based on metasurface devices in the mid‐ diffraction‐limited fisheye metalens. Nano Lett. 2020;20(10):
infrared region. Chin Opt Lett. 2020;18(8):82401. 7429‐7437.
[111] Arbabi E, Arbabi A, Kamali SM, Horie Y, Faraon A. [130] Fan CY, Lin CP, Su GDJ. Ultrawide‐angle and high‐efficiency
Multiwavelength polarization‐insensitive lenses based on metalens in hexagonal arrangement. Sci Rep. 2020;10(1):1‐9.
dielectric metasurfaces with meta‐molecules. Optica. [131] Engelberg J, Zhou C, Mazurski N, Bar‐David J,
2016;3(6):628‐633. Kristensen A, Levy U. Near‐IR wide‐field‐of‐view Huygens
[112] Khorasaninejad M, Aieta F, Kanhaiya P, et al. Achromatic metalens for outdoor imaging applications. Nanophotonics.
metasurface lens at telecommunication wavelengths. Nano 2020;9(2):361‐370.
Lett. 2015;15(8):5358‐5362. [132] Groever B, Chen WT, Capasso F. Meta‐lens doublet in the
[113] Aieta F, Kats MA, Genevet P, Capasso F. Multiwavelength visible region. Nano Lett. 2017;17(8):4902‐4907.
achromatic metasurfaces by dispersive phase compensation. [133] Achouri K, Caloz C. Design, concepts, and applications of
Science. 2015;347(6228):1342‐1345. electromagnetic metasurfaces. Nanophotonics. 2018;7(6):
[114] Arbabi E, Arbabi A, Kamali SM, Horie Y, Faraon A. High 1095‐1116.
efficiency double‐wavelength dielectric metasurface lenses [134] Shen B, Wang P, Polson R, Menon R. An integrated‐
with dichroic birefringent meta‐atoms. Opt Express. nanophotonics polarization beamsplitter with 2.4 × 2.4 μm2
2016;24(16):18468‐18477. footprint. Nat Photonics. 2015;9(6):378‐382.
[115] Khorasaninejad M, Shi Z, Zhu AY, et al. Achromatic [135] Sell D, Yang J, Doshay S, Yang R, Fan JA. Large‐angle,
metalens over 60 nm bandwidth in the visible and metalens multifunctional metagratings based on freeform multimode
with reverse chromatic dispersion. Nano Lett. 2017;17(3): geometries. Nano Lett. 2017;17(6):3752‐3757.
1819‐1824. [136] Bendsoe MP, Sigmund O. Topology Optimization: Theory,
[116] Arbabi E, Arbabi A, Kamali SM, Horie Y, Faraon A. Methods, and Applications. Springer Science & Business
Controlling the sign of chromatic dispersion in diffractive Media; 2003.
optics with dielectric metasurfaces. Optica. 2017;4(6):625‐632. [137] Jensen JS, Sigmund O. Topology optimization for nano‐
[117] Chen WT, Zhu AY, Sanjeev V, et al. A broadband photonics. Laser Photon Rev. 2011;5(2):308‐321.
achromatic metalens for focusing and imaging in the [138] Li J, Cao M, Liang W, Zhang Y, Xie Z, Yuan X. Inverse
visible. Nat Nanotechnol. 2018;13(3):220‐226. design of 1D color splitter for high‐efficiency color imaging.
[118] Arbabi A, Arbabi E, Kamali SM, Horie Y, Han S, Faraon A. Chin Opt Lett. 2022;20(7):73601.
Miniature optical planar camera based on a wide‐angle [139] Su L, Piggott AY, Sapra NV, Petykiewicz J, Vuckovic J.
metasurface doublet corrected for monochromatic aberra- Inverse design and demonstration of a compact on‐chip
tions. Nat Commun. 2016;7(1):1‐9. narrowband three‐channel wavelength demultiplexer. ACS
[119] Martins A, Li K, Li J, et al. On metalenses with arbitrarily Photonics. 2018;5(2):301‐305.
wide field of view. ACS Photonics. 2020;7(8):2073‐2079. [140] Holland JH. Genetic algorithms. Sci Am. 1992;267(1):66‐73.
[120] Zhang F, Pu M, Li X, et al. Extreme‐angle silicon infrared [141] Dorigo M, Birattari M, Stutzle T. Ant colony optimization.
optics enabled by streamlined surfaces. Adv Mater. IEEE Comput Intell Mag. 2006;1(4):28‐39.
2021;33(11):2008157. [142] Lin Z, Liu V, Pestourie R, Johnson SG. Topology optimiza-
[121] Chen J, Ye X, Gao S, et al. Planar wide‐angle‐imaging camera tion of freeform large‐area metasurfaces. Opt Express.
enabled by metalens array. Optica. 2022;9(4):431‐437. 2019;27(11):15765‐15775.
[122] Wang S, Wu PC, Su VC, et al. A broadband achromatic [143] Phan T, Sell D, Wang EW, et al. High‐efficiency, large‐area,
metalens in the visible. Nat Nanotechnol. 2018;13(3):227‐232. topology‐optimized metasurfaces. Light Sci Appl. 2019;8(1):1‐9.
[123] Lin RJ, Su VC, Wang S, et al. Achromatic metalens array for [144] Wang EW, Sell D, Phan T, Fan JA. Robust design of topology‐
full‐colour light‐field imaging. Nat Nanotechnol. 2019;14(3): optimized metasurfaces. Opt Mater Express. 2019;9(2):469‐482.
227‐231. [145] Fliege J, Drummond LMG, Svaiter BF. Newton's method for
[124] Luo X, Zhang F, Pu M, Guo Y, Li X, Ma X. Recent advances multiobjective optimization. SIAM J Optim. 2009;20(2):
of wide‐angle metalenses: principle, design, and applica- 602‐626.
tions. Nanophotonics. 2022;11(1):1‐20. [146] Désidéri JA. Multiple‐gradient descent algorithm (MGDA)
[125] Lin Z, Groever B, Capasso F, Rodriguez AW, Lončar M. for multiobjective optimization. Comptes Rendus Math.
Topology‐optimized multilayered metaoptics. Phys Rev 2012;350(5‐6):313‐318.
Appl. 2018;9(4):44030. [147] Jordan MI, Mitchell TM. Machine learning: trends, perspec-
[126] Hao C, Gao S, Ruan Q, et al. Single‐layer aberration‐ tives, and prospects. Science. 2015;349(6245):255‐260.
compensated flat lens for robust wide‐angle imaging. Laser [148] Sra S, Nowozin S, Wright SJ. Optimization for Machine
Photon Rev. 2020;14(6):2000017. Learning. MIT Press; 2012.
[127] Pu M, Li X, Guo Y, Ma X, Luo X. Nanoapertures with [149] Alzubi J, Nayyar A, Kumar A. Machine learning from
ordered rotations: symmetry transformation and wide‐angle theory to algorithms: an overview. J Phys Conf Ser. 2018;
flat lensing. Opt Express. 2017;25(25):31471‐31477. 1142:012012.
[128] Lassalle E, Mass TWW, Eschimese D, et al. Imaging [150] Carleo G, Cirac I, Cranmer K, et al. Machine learning and
properties of large field‐of‐view quadratic metalenses and the physical sciences. Rev Mod Phys. 2019;91(4):45002.
their applications to fingerprint detection. ACS Photonics. [151] Mahesh B. Machine learning algorithms—a review. Int J Sci
2021;8(5):1457‐1468. Res (IJSR)[Internet]. 2020;9:381‐386.
CHEN ET AL. | 27
[152] Deng L, Hinton G, Kingsbury B. New types of deep neural [170] Fan CY, Su GDJ. Time‐effective simulation methodology for
network learning for speech recognition and related broadband achromatic metalens using deep neural net-
applications: an overview. In: 2013 IEEE International works. Nanomaterials. 2021;11(8):1966.
Conference on Acoustics, Speech and Signal Processing. [171] Gu Y, Hao R, Li EP. Independent bifocal metalens design
IEEE; 2013:8599‐8603. based on deep learning algebra. IEEE Photonics Technol
[153] Eitel A, Springenberg JT, Spinello L, Riedmiller M, Lett. 2021;33(8):403‐406.
Burgard W. Multimodal deep learning for robust RGB‐D [172] Zarei S, Khavasi A. Inverse design of on‐chip thermally
object recognition. In: 2015 IEEE/RSJ International Confer- tunable varifocal metalens based on silicon metalines. IEEE
ence on Intelligent Robots and Systems (IROS). IEEE; 2015: Access. 2021;9:73453‐73466.
681‐687. [173] Wu Y, Luo Y, Chaudhari G, et al. Bright‐field holography:
[154] Voulodimos A, Doulamis N, Doulamis A, Protopapadakis E. cross‐modality deep learning enables snapshot 3D imaging
Deep learning for computer vision: a brief review. Comput with bright‐field contrast using a single hologram. Light Sci
Intell Neurosci. 2018;2018:7068349. Appl. 2019;8(1):1‐7.
[155] Zhao ZQ, Zheng P, Xu St, Wu X. Object detection with deep [174] Ren H, Shao W, Li Y, Salim F, Gu M. Three‐dimensional
learning: a review. IEEE Trans Neural Networks Learn Syst. vectorial holography based on machine learning inverse
2019;30(11):3212‐3232. design. Sci Adv. 2020;6(16):eaaz4261.
[156] Grigorescu S, Trasnea B, Cocias T, Macesanu G. A survey of [175] Liu C, Yu WM, Ma Q, Li L, Cui TJ. Intelligent coding
deep learning techniques for autonomous driving. J Field metasurface holograms by physics‐assisted unsupervised gener-
Rob. 2020;37(3):362‐386. ative adversarial network. Photonics Res. 2021;9(4):B159‐B167.
[157] LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. [176] Ma W, Xu Y, Xiong B, et al. Pushing the limits of
2015;521(7553):436‐444. functionality‐multiplexing capability in metasurface design
[158] Goodfellow I, Bengio Y, Courville A. Deep Learning. MIT based on statistical machine learning. Adv Mater.
Press; 2016. 2022;34:e2110022.
[159] Hemmatyar O, Abdollahramezani S, Kiarashinejad Y, [177] Ma D, Li Z, Liu W, et al. Deep‐learning enabled multicolor
Zandehshahvar M, Adibi A. Full color generation with meta‐holography. Adv Opt Mater. 2022;10:2102628.
fano‐type resonant HfO2 nanopillars designed by a deep‐ [178] Sajedian I, Lee H, Rho J. Double‐deep Q‐learning to
learning approach. Nanoscale. 2019;11(44):21266‐21274. increase the efficiency of metasurface holograms. Sci Rep.
[160] So S, Mun J, Rho J. Simultaneous inverse design of 2019;9(1):1‐8.
materials and structures via deep learning: demonstration [179] Keshavarz Hedayati M, Elbahri M. Review of metasurface
of dipole resonance engineering using core–shell nanopar- plasmonic structural color. Plasmonics. 2017;12(5):1463‐1479.
ticles. ACS Appl Mater Interfaces. 2019;11(27):24264‐24268. [180] Yang B, Cheng H, Chen S, Tian J. Structural colors in
[161] Rivenson Y, Zhang Y, Günaydın H, Teng D, Ozcan A. Phase metasurfaces: principle, design and applications. Mater
recovery and holographic image reconstruction using deep Chem Front. 2019;3(5):750‐761.
learning in neural networks. Light Sci Appl. 2018;7(2):17141. [181] Gao L, Li X, Liu D, Wang L, Yu Z. A bidirectional deep
[162] Wiecha PR, Lecestre A, Mallet N, Larrieu G. Pushing the neural network for accurate silicon color design. Adv Mater.
limits of optical information storage using deep learning. 2019;31(51):1905467.
Nat Nanotechnol. 2019;14(3):237‐244. [182] Sajedian I, Badloe T, Rho J. Optimisation of colour
[163] Li L, Shuang Y, Ma Q, et al. Intelligent metasurface imager generation from dielectric nanostructures using
and recognizer. Light Sci Appl. 2019;8(1):1‐9. reinforcement learning. Opt Express. 2019;27(4):5874‐5883.
[164] Qian C, Zheng B, Shen Y, et al. Deep‐learning‐enabled self‐ [183] González‐Alcalde AK, Salas‐Montiel R, Kalt V, Blaize S,
adaptive microwave cloak without human intervention. Nat Macías D. Engineering colors in all‐dielectric metasurfaces:
Photonics. 2020;14(6):383‐390. metamodeling approach. Opt Lett. 2020;45(1):89‐92.
[165] Peurifoy J, Shen Y, Jing L, et al. Nanophotonic particle [184] Kalt V, González‐Alcalde AK, Es‐Saidi S, Salas‐Montiel R,
simulation and inverse design using artificial neural net- Blaize S, Macías D. Metamodeling of high‐contrast‐index
works. Sci Adv. 2018;4(6):eaar4206. gratings for color reproduction. J Opt Soc Am A. 2019;36(1):
[166] Hu B, Wu B, Tan D, Xu J, Chen Y. Robust inverse‐design of 79‐88.
scattering spectrum in core–shell structure using modified [185] Lee T, Jang J, Jeong H, Rho J. Plasmonic‐ and dielectric‐
denoising autoencoder neural network. Opt Express. based structural coloring: from fundamentals to practical
2019;27(25):36276‐36285. applications. Nano Convergence. 2018;5(1):1‐21.
[167] Wiecha PR, Muskens OL. Deep learning meets nanopho- [186] Baxter J, Calà Lesina A, Guay JM, Weck A, Berini P,
tonics: a generalized accurate predictor for near fields Ramunno L. Plasmonic colours predicted by deep learning.
and far fields of arbitrary 3D nanostructures. Nano Lett. Sci Rep. 2019;9(1):1‐9.
2019;20(1):329‐338. [187] Roberts NB, Keshavarz Hedayati M. A deep learning
[168] Vahidzadeh E, Shankar K. Artificial neural Network‐based approach to the forward prediction and inverse design of
prediction of the optical properties of spherical core–shell plasmonic metasurface structural color. Appl Phys Lett.
plasmonic metastructures. Nanomaterials. 2021;11(3):633. 2021;119(6):61101.
[169] An X, Cao Y, Wei Y, et al. Broadband achromatic metalens [188] Siegel J, Wang AY, Menabde SG, Kats MA, Jang MS,
design based on deep neural networks. Opt Lett. 2021; Brar VW. Self‐stabilizing laser sails based on optical
46(16):3881‐3884. metasurfaces. ACS Photonics. 2019;6(8):2032‐2040.
28 | CHEN ET AL.
[189] Salary MM, Mosallaei H. Photonic metasurfaces as relativ- [207] Banerji S, Meem M, Majumder A, Vasquez FG, Sensale‐
istic light sails for Doppler‐broadened stable beam‐riding Rodriguez B, Menon R. Imaging with flat optics: metalenses
and radiative cooling. Laser Photon Rev. 2020;14(8):1900311. or diffractive lenses? Optica. 2019;6(6):805‐810.
[190] Han C, Zhang B, Wang H, Ding J. Metamaterial perfect [208] Saha S, Shah D, Shalaev VM, Boltasseva A. Tunable
absorber with morphology‐engineered meta‐atoms using metasurfaces: controlling light in space and time. Opt
deep learning. Opt Express. 2021;29(13):19955‐19963. Photonics News. 2021;32(7):34‐41.
[191] Tittl A, John‐Herpin A, Leitis A, Arvelo ER, Altug H. [209] Arbabi E, Arbabi A, Kamali SM, Horie Y, Faraji‐Dana M,
Metasurface‐based molecular biosensing aided by artificial Faraon A. MEMS‐tunable dielectric metasurface lens. Nat
intelligence. Angew Chemie Int Ed. 2019;58(42): Commun. 2018;9(1):1‐9.
14810‐14822. [210] Li J, Yu P, Zhang S, Liu N. Electrically‐controlled digital
[192] Moon G, Choi J‐r, Lee C, Oh Y, Kim KH, Kim D. Machine metasurface device for light projection displays. Nat
learning‐based design of meta‐plasmonic biosensors with Commun. 2020;11(1):1‐7.
negative index metamaterials. Biosens Bioelectron. 2020;164: [211] Ding F, Yang Y, Bozhevolnyi SI. Dynamic metasurfaces
112335. using phase‐change chalcogenides. Adv Opt Mater. 2019;
[193] Wang Y, Li M, Wang Y. Silk: a versatile biomaterial for 7(14):1801709.
advanced optics and photonics. Chin Opt Lett. 2020;18(8): [212] Zhang Y, Fowler C, Liang J, et al. Electrically reconfigurable
80004. non‐volatile metasurface using low‐loss optical phase‐
[194] John‐Herpin A, Kavungal D, von Mücke L, Altug H. change material. Nat Nanotechnol. 2021;16(6):661‐666.
Infrared metasurface augmented by deep learning for [213] Shalaginov MY, An S, Zhang Y, et al. Reconfigurable all‐
monitoring dynamics between all major classes of biomo- dielectric metalens with diffraction‐limited performance.
lecules. Adv Mater. 2021;33(14):2006054. Nat Commun. 2021;12(1):1‐8.
[195] Kudyshev ZA, Kildishev AV, Shalaev VM, Boltasseva A. [214] Park J, Jeong BG, Kim SI, et al. All‐solid‐state spatial light
Machine‐learning‐assisted metasurface design for high‐ modulator with independent phase and amplitude control for
efficiency thermal emitter optimization. Appl Phys Rev. three‐dimensional LiDAR applications. Nat Nanotechnol.
2020;7(2):21407. 2021;16(1):69‐76.
[196] Cui TJ, Qi MQ, Wan X, Zhao J, Cheng Q. Coding [215] Qian C, Chen H. A perspective on the next generation of
metamaterials, digital metamaterials and programmable invisibility cloaks—intelligent cloaks. Appl Phys Lett.
metamaterials. Light Sci Appl. 2014;3(10):e218. 2021;118(18):180501.
[197] Li L, Cui TJ. Information metamaterials—from effective [216] Liu Z, Zhu D, Rodrigues SP, Lee KT, Cai W. Generative
media to real‐time information processing systems. model for the inverse design of metasurfaces. Nano Lett.
Nanophotonics. 2019;8(5):703‐724. 2018;18(10):6570‐6576.
[198] Li L, Jun Cui T, Ji W, et al. Electromagnetic [217] Melati D, Grinberg Y, Kamandar Dezfouli M, et al. Mapping the
reprogrammable coding‐metasurface holograms. Nat global design space of nanophotonic components using machine
Commun. 2017;8(1):1‐7. learning pattern recognition. Nat Commun. 2019;10(1):1‐9.
[199] Li L, Ruan H, Liu C, et al. Machine‐learning [218] Jiang J, Fan JA. Global optimization of dielectric metasur-
reprogrammable metasurface imager. Nat Commun. faces using a physics‐driven neural network. Nano Lett.
2019;10(1):1‐8. 2019;19(8):5366‐5372.
[200] Zhao J. A survey of intelligent reflecting surfaces (IRSs): [219] Barbastathis G, Ozcan A, Situ G. On the use of deep learning
towards 6G wireless communication networks. arXiv Prepr for computational imaging. Optica. 2019;6(8):921‐943.
arXiv190704789. 2019. [220] Zuo C, Qian J, Feng S, et al. Deep learning in optical
[201] Huang C, Zappone A, Alexandropoulos GC, Debbah M, metrology: a review. Light Sci Appl. 2022;11(1):1‐54.
Yuen C. Reconfigurable intelligent surfaces for energy [221] Zhang Q, Yu H, Barbiero M, Wang B, Gu M. Artificial
efficiency in wireless communication. IEEE Trans Wireless neural networks enabled by nanophotonics. Light Sci Appl.
Commun. 2019;18(8):4157‐4170. 2019;8(1):1‐14.
[202] Liaskos C, Tsioliaridou A, Nie S, Pitsillides A, Ioannidis S, [222] Lin X, Rivenson Y, Yardimci NT, et al. All‐optical machine
Akyildiz I. An interpretable neural network for configuring learning using diffractive deep neural networks. Science.
programmable wireless environments. In: 2019 IEEE 20th 2018;361(6406):1004‐1008.
International Workshop on Signal Processing Advances in [223] Goi E, Chen X, Zhang Q, et al. Nanoprinted high‐neuron‐
Wireless Communications (SPAWC). IEEE; 2019:1‐5. density optical linear perceptrons performing near‐infrared
[203] Taha A, Alrabeiah M, Alkhateeb A. Enabling large inference on a CMOS chip. Light Sci Appl. 2021;10(1):1‐11.
intelligent surfaces with compressive sensing and deep [224] Qian C, Lin X, Lin X, et al. Performing optical logic
learning. IEEE Access. 2021;9:44304‐44321. operations by a diffractive neural network. Light Sci Appl.
[204] Zhang X, Li Q, Liu F, et al. Controlling angular dispersions 2020;9(1):1‐7.
in optical metasurfaces. Light Sci Appl. 2020;9(1):1‐12. [225] Liu C, Ma Q, Luo ZJ, et al. A programmable diffractive deep
[205] Verre R, Svedendahl M, Odebo Länk N, et al. Directional neural network based on a digital‐coding metasurface array.
light extinction and emission in a metasurface of tilted Nat Electron. 2022;5(2):113‐122.
plasmonic nanopillars. Nano Lett. 2016;16(1):98‐104. [226] Zhou T, Lin X, Wu J, et al. Large‐scale neuromorphic
[206] Presutti F, Monticone F. Focusing on bandwidth: achro- optoelectronic computing with a reconfigurable diffractive
matic metalens limits. Optica. 2020;7(6):624‐631. processing unit. Nat Photonics. 2021;15(5):367‐373.
CHEN ET AL. | 29
[227] Shen Y, Harris NC, Skirlo S, et al. Deep learning with Shanshan Hu is a Postgraduate at
coherent nanophotonic circuits. Nat Photonics. 2017;11(7): Nanjing University. She received her
441‐446. B.E. degree in Optoelectronic Informa-
[228] Shastri BJ, Tait AN, Ferreira de Lima T, et al. Photonics for
tion Science and Engineering at China
artificial intelligence and neuromorphic computing. Nat
University of Petroleum. She joined the
Photonics. 2021;15(2):102‐114.
[229] Luo X, Hu Y, Li X, et al. Metasurface‐enabled on‐chip National Laboratory of Solid State Micro-
multiplexed diffractive neural networks in the visible. arXiv structures at Nanjing University to pursue her M.S.
Prepr arXiv210707873. 2021. degree under the guidance of Prof. Shining Zhu and
[230] Zuo Y, Li B, Zhao Y, et al. All‐optical neural network with Prof. Tao Li. Her current research interests focus on
nonlinear activation functions. Optica. 2019;6(9):1132‐1137. the metadevices design and metalens imaging
[231] Mengu D, Luo Y, Rivenson Y, Lin X, Veli M, Ozcan A. technology.
Response to Comment on “All‐optical machine learning
using diffractive deep neural networks”. arXiv Prepr
arXiv181004384. 2018. Shining Zhu is a Professor at School of
[232] Spreyer F, Zhao R, Huang L, Zentgraf T. Second harmonic Physics and Principal Investigator (PI) at
imaging of plasmonic Pancharatnam–Berry phase metasur- the National Laboratory of Solid State
faces coupled to monolayers of WS2. Nanophotonics. Microstructures, Nanjing University. He
2020;9(2):351‐360.
is also an Academician of the CAS, and a
[233] Staude I, Pertsch T, Kivshar YS. All‐dielectric resonant
Fellow of OSA, COS, and APS. He got his
meta‐optics lightens up. ACS Photonics. 2019;6(4):802‐814.
[234] Hu G, Hong X, Wang K, et al. Coherent steering of Master's and Ph.D. degrees from the Physics Depart-
nonlinear chiral valley photons with a synthetic Au–WS2 ment of Nanjing University. His main research
metasurface. Nat Photonics. 2019;13(7):467‐472. interests are functionally microstructured materials,
[235] Li L, Liu Z, Ren X, et al. Metalens‐array‐based high‐ including their applications on laser, nonlinear optics,
dimensional and multiphoton quantum source. Science. quantum optics, quantum information, and so forth.
2020;368(6498):1487‐1490.
[236] Xie YY, Ni PN, Wang QH, et al. Metasurface‐integrated
vertical cavity surface‐emitting lasers for programmable Tao Li is a Professor at Nanjing Univer-
directional lasing emissions. Nat Nanotechnol. 2020;15(2): sity. He received his Ph.D. degree at NJU
125‐130. in 2005 and joined NJU as a faculty in
2008. He worked as a visiting scholar at
Nanyang Technology University, Singa-
AUTH OR BI OGRAPH IES
pore, in 2012, and Hong Kong Baptist
University under the support of “K.C. Wong Education
Ji Chen is an Associate Research Fellow
Foundation” scholarship in 2013. He was selected as
and Master Supervisor at Southeast
the “Dengfeng Talent Program B” from Nanjing
University. He received his B.S. degree
University, won the “National Funds for Outstanding
in Material Physics in 2014 and Ph.D.
Young Scientists” and “Young and middle‐aged lead-
degree in Optical Engineering in 2019,
ing scientists” from MOST. His research interest is
both from Nanjing University. From
about the metamaterials, plasmonics, and photonic
2019 to 2021, he has two working experiences in
integration. Till now, he has published >100 papers
China Electronics Technology Group Corporation
with citations of over 5000 and a recent H index of 36
and College of Engineering and Applied Science,
(Web of Science), and presented >50 invited talks at
Nanjing University. He is now working in School of
international conferences and seminars.
Information Science and Engineering, National
Mobile Communications Research Laboratory, South-
east University. His research results have been
published in Nature, Nature Nanotechnology, Nature How to cite this article: Chen J, Hu S, Zhu S, Li
Communications, Optica, Laser & Photonics Reviews, T. Metamaterials: from fundamental physics to
and so forth. His research interests include metaop- intelligent design. Interdiscip Mater. 2023;2:5‐29.
tics, optical wireless communications, advanced doi:10.1002/idm2.12049
imaging technology, and surface plasmon photonics.

Metamaterials: From Fundamental Physics To Intelligent Design

Uploaded by

Copyright:

Available Formats

Metamaterials: From Fundamental Physics To Intelligent Design

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Metamaterials: From Fundamental Physics To Intelligent Design

Uploaded by

Copyright:

Available Formats

Received: 30 April 2022 | Revised: 4 June 2022 | Accepted: 19 June 2022

Metamaterials: From fundamental physics to intelligent

Ji Chen1,2 | Shanshan Hu1 | Shining Zhu1 | Tao Li1

1 | INTRODUCTION polarizers, and possess a very limited range of permittivity,

Interdisciplinary Materials. 2023;2:5–29. onlinelibrary.wiley.com/r/interdisciplinarymaterials | 5

3 | METASURFACE A ND The early proposed metasurfaces mainly used plasmonic

A key scientific problem that currently restricts metalens

6.4 | Fabrication of large‐scale

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.