Roadmap on photonic metasurfaces Open Access

Sebastian A. Schulz,^* Rupert F. Oulton,^** and Mitchell Kenney^***

^*[email protected]

^**[email protected]

^***[email protected]

Over the last decade, the topic of metasurfaces has flourished like no other field in optics. This is primarily due to the appeal of compact flat optical components with wide design flexibility inherent to these artificial two-dimensional materials. As can be seen in this roadmap, metasurfaces can be made from a wide range of materials, from metals and other plasmonic materials to dielectrics and even flexible and conformable platforms. In all cases, a metasurface's overall properties follow not only from the those of the constituent materials but also from the shape and arrangement of meta-atoms, engineered on nanoscopic dimensions. The broad range of metasurface applications available stem from the rich physics that emerges from the arrangement and geometry of the meta-atoms; by arranging these into periodic or disordered distributions, that display local or non-local responses, as well as exhibiting conventional or topological characteristics, these “design parameters” allow us to influence light's amplitude, phase, polarization, and nonlinearity.

This document offers a comprehensive overview of the physics and applications of photonic metasurfaces which sets the context for discussion of future research directions. As commissioning editors, we have invited 20 perspectives from the community, where each has given equal importance to reviewing the historic development of the field, presenting highlights of current research and then pointing to future challenges and outlook. Andrea Alu sets the scene with a general introduction (I). The roadmap is then organized into three sections of perspectives: ten perspectives on the applications of photonic metasurfaces; three perspectives on the fabrication and material platforms; and finally, six perspectives on the emerging topics and the theory of metasurfaces. The roadmap explore several key application areas for metasurfaces, including metasurfaces for: the control of light emission (II); nonlinear optics (III); biosensing (IV); passive radiative cooling (V); photocatalysis (VI); orbital angular momentum and wavefront control (VII); holography and structural color (VIII); augmented reality (IX); quantum state generation (X); and finally metasurface based lasers (XI). The section on fabrication and material platforms includes solution derived metasurface fabrication (XII); semiconductor based metasurfaces (XIII); and flexible metasurfaces (XIV). The final six perspectives explore metasurfaces for or with time modulation (XV); disorder (XVI); epsilon-near-zero characteristics (XVII); deep learning (XVIII); complex-valued singularities (XIX); and, last but not least, multipolar characteristics (XX).

In contrast to previous reviews (e.g., Refs. 1–3) and roadmaps,⁴ which focus exclusively on either past or future development of the field, we have aimed instead to provide a comprehensive overview that gives equal weight to past, present and future of the selected topics. This gives the necessary context for the discussion of future research directions and outlook whilst also providing a valuable up to date resource by indexing the rapidly amassed literature in this field.

This roadmap is truly the culmination of a world-wide effort of leading contributors from across the metasurface community. We are grateful for their time, effort, and dedication to this compilation of perspectives. We hope that this Roadmap on Photonic Metasurfaces will provide you—the reader—with an up to date reference of the current status, understanding, and direction of photonic metasurface research; whether you are already an experienced researcher looking for a contemporary summary of the research and direction of the field or whether you are looking for a comprehensive introduction to this fascinating area.

Andrea Alù

[email protected]

The field of photonics relies on gaining control over light flows to advance a wide and rapidly growing range of technologies, including energy, sensing, imaging, and computing. Photonics research has been recently unveiling exciting opportunities both in the context of basic science and of its engineering and technological impact. A major role in this recent progress has been played by ultrathin surfaces engineered at the nanoscale, known as metasurfaces,⁵ which have been enhancing the way we tailor optical wavefronts over an ultrathin platform, opening a paradigm of compactification of photonic devices for extreme light control. In turn, metasurfaces have also been unveiling unique forms of light–matter interactions emerging from their subwavelength light confinement and extreme light control at the nanoscale. The recent surge of interest for metasurfaces in the photonics community has been driven by the introduction of the “generalized Snell's laws of refraction,”^5,6 according to which a transverse gradient of phase discontinuities imparted by a tailored array of nanostructures can offer enhanced control over wave transmission through a planar interface. These concepts have been building on well-established technologies in the realm of microwave components^7,8 and “flat” diffractive optical elements,^9,10 established fields of research for several decades. Over the years, more sophisticated metasurface designs have been demonstrating further mastery over all properties of light, enhancing efficiency, bandwidth, polarization, and wavefront control through metasurfaces,^11–16 which, in turn, have been translated into ultrathin devices for lensing, holograms and lasing, among many other applications.

While the early demonstrations of metasurfaces were mostly based on plasmonic nanostructures, which enable field enhancement and light focusing at deeply subwavelength scales, the associated material loss have driven the field toward alternative materials, including high-index dielectrics and doped semiconductors. Today, the possibilities offered by metasurfaces to efficiently control light with nanoscale resolution over an ultrathin platform have translated into myriads of opportunities, not only limited to high-profile academic research, but also to commercial companies.^17,18 The progress in this area has been truly impressive, leveraging the combination of theoretical advances in our understanding of the interactions of light with nanostructures, and progress in nanofabrication of a wide range of materials, also over large area. In turn, the discoveries driven by photonic metasurface research have also translated back to lower frequencies, where the fields of reflect- and transmit-arrays, as well as frequency-selective surfaces, originally demonstrated wavefront and spectrum control for radio-waves, driving the interest in a new wave of hardware platforms known as reconfigurable intelligent surfaces for wireless communications.¹⁹ 

The first demonstrations of optical metasurfaces, e.g., Fig. 1(a), were limited in terms of efficiency and of functionalities, due to a variety of challenges. For the most part, they relied on metallic resonant elements, associated with significant absorption, and on polarization conversion mechanisms, which tend to be inefficient and limited by symmetries, to control the amplitude and phase of the scattered fields. The overall efficiency and the type of transformations that could be implemented on the incoming wavefront were significantly limited in these first demonstrations.¹⁴ By contrast, over the years new generations of photonic metasurfaces have flourished into a plethora of exciting functionalities for holography,²⁰ multi-functional and multi-wavelength operation,²¹ lensing and imaging,²² and several other impressive demonstrations of complex wavefront shaping using ultrathin optical devices, becoming competitive with bulky optical technologies in terms of several performance metrics. Figure 1 shows a few demonstrations of the state of the art of this technology, in which tailored nanostructured apertures are able to imprint the desired wavefront to the electromagnetic fields at the wavelength of interest. Figure 1(b), for instance, shows high-efficiency holograms created by tailored nanostructured surfaces, and Fig. 1(c) an ultrathin metalens that can focus light at multiple frequencies.

Several design principles have been explored to tailor the optical wavefront, spanning from coupled resonances locally tuned to control the amplitude and phase²³ to polarization conversion⁵ and geometric phase concepts,²⁴ each coming with their own advantages and tradeoffs. For instance, coupled resonances tend to introduce unwanted frequency dispersion, modulating the scattering amplitude as we tailor the local phase of each metasurface elements. Polarization conversion can powerfully tailor the phase of the scattered fields, but its efficiency is typically limited. Geometric phase concepts can partially address these challenges, but they are conventionally limited to circularly polarized responses.

While all these design principles offer a powerful playground for wavefront control, metasurfaces often face fundamental constraints in terms of efficiency and bandwidth of wavefront manipulation, which stem from their ultrathin footprint and inherent symmetry constraints. Over the years, several of these bounds have been unveiled,^11,14,25,26 indicating that extreme compactification of optical devices comes at a price, which needs to be taken into account when compared to other more bulky solutions. For instance, it is hard to imagine that a fully metasurface-based approach may be able to compete with a sophisticated EOS camera lens, but metasurfaces may be able to replace specific components. Hybrid approaches leveraging metasurfaces appear to be certainly competitive, and metasurfaces may be excellent at specific tasks, or for targeted wavelengths of operation.

While the conventional approach to metasurface design has been the one of locally patterning the aperture with nanoscale resonators that control point by point the response, a few more recent approaches have been opening new design degrees of freedom. For instance, metagratings^27,28 have been demonstrating that extreme control over the optical wavefront can be achieved with high efficiency, for instance yielding near-grazing beam steering. Metagratings are based on grating resonances that stem from an overarching long-range periodicity of the structure, while the individual metasurface elements are designed to control the way the incident wavefront locally couples to the available diffraction orders. These principles have been further empowered by nonlocal metasurfaces, in which the coupling among distant elements across the metasurface aperture is not fought against, but rather leveraged through more sophisticated design principles that exploit lattice resonances locally perturbed by symmetry-breaking defects.²⁹ Nonlocal metasurfaces provide a more sophisticated control over the spectral response, and open new opportunities for wavefront manipulation, including wavefront selectivity,³⁰ and multi-functionality.³¹ 

The recent progress in metasurface design principles, fabrication and also tunability of their response has led to a surge of exciting applications. Figure 2 summarizes a few relevant highlights of metasurfaces that realize complex structured wavefronts, relevant for e.g., optical communications and holography,³² for eye tracking applications,³³ for quantum technologies,³⁶ for catalysis and chemical reactions,³⁴ and for augmented reality (AR).³⁵ These breadth of applications holds the promise of a vibrant future for metasurfaces, opening exciting prospects for the near future.

The opportunities for photonic metasurfaces have been growing by the day, fostered by continuous progress on multiple fronts, which drives the future of this research area. From the modeling perspective, enhancing the ultimate performance of metasurfaces requires sophisticated modeling tools that can capture and optimize their design. Given that most metasurfaces are non-periodic, rational design tools, fast simulation and optimization techniques have been emerging as exciting prospects, for instance in the context of adjoint methods, topology optimization and machine learning. Physics-driven design and optimization approaches, such as the use of singularities in the complex frequency plane, topological concepts, bound states in the continuum (BIC) and exceptional points are emerging directions that unveil new optical phenomena and add new designer tools for photonic metasurfaces.^37–40 

Emerging applications of nonlocal metasurfaces include the possibility of filtering and processing images in momentum space, realizing Fourier and even nonlinear operations over an ultrathin, efficient platform. The idea is to leverage the engineered nonlocality in metasurfaces to perform mathematical operations on the incoming images,^41–43 and even realize optical analog computers that can solve complex mathematical problems^44–46 [Figs. 3(a) and 3(b)]. Tailored nonlocalities in metasurfaces can also add spatial and temporal coherence, ideally suited to pattern and control thermal emission and photoluminescence. These ideas are paving the way to the realization of ultrathin surfaces that emit tailored wavefronts with desired amplitude, phase, and polarization without the need for an external coherent source driving them.^47,48 Figure 3(c) shows a thermal metasurface, based on an underlying periodic lattice that can control the degree of temporal and spatial coherence endowed to thermal emission, and whose local perturbations control the polarization and wavefront shape of thermal emission. Future efforts may be able to demonstrate ultrathin patterned surfaces that embed their own optical sources and pattern them with extreme flexibility.

On the material front, the use of two-dimensional (2D) materials integrated with metasurfaces, such as graphene⁵¹ and transition metal dichalcogenides⁴⁹ [Fig. 3(d)], and even direct patterning of bulk 2D materials, holds the promise for exciting opportunities both in the context of basic science and applications. These materials offer interesting forms of light–matter interactions in the form of plasmon, exciton, phonon polaritons, in which light and matter are so intertwined to form quasi-particles. Combined with metasurface concepts, these phenomena can further boost light control, and impart exotic photonic features to polaritonic responses. Polaritonic materials can be also ideally suited to boost optical nonlinearities, opening exciting opportunities to extend metasurface operations and wavefront control to nonlinear optical processes, such as wave mixing, frequency generation, limiting, up- and downconversion.^52,53 These materials can also be exciting prospects to efficiently integrate and pattern optical gain in metasurfaces, paving the way to a plethora of interesting non-Hermitian wave phenomena.^38,39,54

Nonlinearities can also offer powerful tools to reconfigure and modulate in time the metasurface response, which becomes crucial to make an impact in many technologies. Beyond the importance of manipulating in real-time the spatial degrees of freedom of the incoming wavefront, creating enhanced forms of spatial light modulators,⁵⁵ but also their temporal and frequency content. Suitable temporal modulation schemes can break reciprocity and efficiently mix frequencies,⁵⁶ as well as induce nontrivial parametric phenomena, including Doppler shifts, and active beam steering,⁵⁰ largely expanding the reach and opportunities offered by optical metasurfaces, as schematically shown in Fig. 3(e). For various applications, real-time dynamic programmability may become necessary moving forward the field of metasurfaces, for instance in the context of image processing, dynamic scene creation, and holograms. The integration of 2D and polaritonic materials may enable faster and more efficient forms of modulation.

Finally, photonic metasurfaces may truly flourish once its realization becomes fully compatible with large-area and inexpensive fabrication techniques, pushing metasurfaces from proof of concept implementations to foundry-level production. Roll-to-roll, nanoimprinting, and self-assembly techniques appear promising in this context, and efforts to develop design principles compatible with the constraints of these techniques in terms of materials and disorder tolerance are becoming necessary. Overall, photonic metasurfaces have a bright future ahead, continuously evolving along the years with new concepts, new material playgrounds and a highly interdisciplinary broad research community. As the field continues to mature, the exotic wave phenomena at the basis of photonic metasurfaces are empowering photonic technologies and exciting several industries.

Our work on these topics has been supported by the Simons Foundation and the Air Force Office of Scientific Research.

Ayesheh Bashiri, Zlata Fedorova, Radoslaw Kolkowski, A. Femius Koenderink, and Isabelle Staude^*

^*[email protected]

Optical metasurfaces are two-dimensional (2D) arrangements of subwavelength scale building blocks known as meta-atoms with engineered scattering properties. In contrast to the passive metasurfaces discussed in most other sections of this Roadmap, which manipulate the light propagating from a distant light source, light-emitting metasurfaces incorporate nanoscale light sources in their architecture, allowing the coupling of the emission from the sources to the far-field.⁵⁷ Light-emitting metasurfaces offer various functionalities including photoluminescence enhancement,⁵⁸ tailored emission directionality,^48,59,60 improved quantum efficiency,⁶¹ color conversion,⁶² and controlled degree of coherence,⁶³ thus offering important opportunities for compact and efficient light sources such as lasers (see also Sec. XI), LEDs,⁶⁴ and single-photon sources⁶⁵ for quantum technologies (see also Sec. X). Other applications range from displays and optical communication to biomedical applications, and energy harvesting. Note that, in this section, the term “light emission” specifically refers to the fluorescent and photoluminescent sources, while thermal emission is addressed in Secs. I and V.

Light-emitting metasurfaces can be categorized into two main platforms: plasmonic and high-refractive-index all-dielectric. Plasmonic metasurfaces are composed of metallic meta-atoms, where the large confinement of electromagnetic fields in their surroundings can result in a strong reduction of the radiative lifetime of the emitters coupled to the metasurface.⁶⁶ However, the applicability of these metasurfaces is limited due to their intrinsic Ohmic losses. All-dielectric metasurfaces, on the other hand, can exhibit low absorption losses and support both electric and magnetic multipolar Mie-type modes without demanding complex geometries.⁶⁷ Multipolar superposition then allows for tailoring directional properties, e.g., through the Kerker effect, which facilitates applications in beam steering, lensing, and beam shaping.

Various types of emitters have been considered as active components of light-emitting metasurfaces, such as quantum dots (QDs), dyes, and semiconductors. Recently, many studies concentrated on the coupling of optical metasurfaces with 2D materials. In the family of 2D materials, layered systems of transition metal dichalcogenides (TMDs)⁶⁸ have attracted special attention due to their unique optoelectronic properties, including strong photoluminescence, excitonic response at room temperature, and circular dichroism caused by their valley-selective optical transitions.⁶⁹ 

The key mechanism for controlling light emission by metasurfaces is the engineering of the local density of photonic states (LDOS) through either the localized resonances provided by individual meta-atoms or the collective resonances originating from their overall arrangement or both. In periodic metasurfaces, collective resonances can be regarded as photonic Bloch bands (as in photonic crystals) transformed by the resonant meta-atoms into surface lattice resonances (SLRs).⁷⁰ They also include dark modes and quasi bound states in the continuum (quasi-BICs), in which radiative loss is suppressed by destructive interference.⁷¹ Collective resonances can provide high LDOS at the band edges, together with control of spatial coherence and directivity of emission.^59,72 Engineering the individual meta-atoms allows to further tailor the light emission properties, employing both local and nonlocal effects. For example, asymmetric meta-atoms may allow a controlled fraction of the light to be radiated away by turning the perfectly nonradiating BICs into quasi-BICs.⁷³ Moreover, spatial variation of the meta-atom design can allow for the emission of light fields with shaped wavefronts through metasurface defined spatially varying phase or amplitude profiles.^48,60,74

Although most of the light-emitting metasurfaces demonstrated so far facilitate tailoring of spontaneous emission in the weak coupling regime, their potential for strong coupling and lasing has also been demonstrated.^75,76 The latter is discussed in detail in Sec. XI. Semiclassical strong coupling (ensemble of many emitters) is achieved when the coupling rate of emitters to the photonic resonances exceeds the photonic loss rate, and the rate of spontaneous decay is gauged by the emission spectral bandwidth. This can be achieved by aligning the active material's emission bandwidth with metasurface resonances featuring sharp line widths while aiming to maximize the oscillator strength of the dipole transition. Low-threshold lasers, on the other hand, require high-quality (high-Q) optical cavities and precise spatial and spectral overlap of their modes with the gain material. This can be achieved by engineering the parameters of the metasurface integrated with the gain medium.

The relevant performance factors for light-emitting metasurfaces depend on the specific target application. For instance, to impact high-power LED lighting one must contend with 0.5–1 A/mm² current density (blue emitting LEDs), or equivalently for phosphor converting layer of order 1 W/mm² emitted power, and 1 W/mm² thermal load due to the Stokes shift. This places huge challenges on photostability and absorption coefficients. Another application is light sources for projection or image-projecting metasurfaces. In contrast to general lighting, such applications require a small etendue. Metasurface-based displays must emit a set of selected wavelengths, which can be achieved, e.g., by energy transfer between different types of emitters. For applications like, e.g., AR glasses, it is not overall emission power output that matters, but the combination of emissive behavior with transparency for most of the ambient light. Some factors are relevant for multiple applications, e.g., brightness, photostability, power efficiency, fabrication reliability, scalability, and cost. In addition, a frequently desired property is post-fabrication tunability. Modulating the emission and/or achieving multiple and adjustable functionalities in one design are the key features of light-emitting metasurfaces that can bring them closer to real-world applications.⁷⁷ 

Figure 4 summarizes many of the aspects of light-emitting metasurfaces discussed in this section.

The emission from nanoscale light sources incorporated into metasurfaces can be tailored through a modification of excitation rate (for optical pumping), of radiative decay rate, and of directionality of the emission.⁷⁸ Metasurfaces can strongly confine the excitation field in the emissive layer, resulting in an efficient incoupling and further enhancement of the emission. Radiative decay rate enhancement is associated with the Purcell effect and achieved through modifications of the LDOS by resonant metasurfaces. For classical cavities, it is well known that the Purcell factor scales approximately with the ratio of the quality factor Q and the mode volume V. Simple Mie-resonant all-dielectric metasurfaces exhibit Q-factors on the order of 10–100 and moderate Purcell factors.⁷⁹ However, they contribute effectively to the radiative decay rate enhancement due to their negligible absorption losses. This is especially beneficial for emitters with intrinsically high quantum yields. In contrast, plasmonic metasurfaces can offer much smaller mode volumes, resulting in larger Purcell factors,⁸⁰ at the cost of enhancing both the radiative and nonradiative decay rate, thus reducing the quantum yield for highly efficient emitters. Hence, they can only be usefully employed for boosting the quantum yield of rather inefficient emitters⁷⁸ or should be operated in the regime of high-Q SLRs, where the hybridization between plasmon and diffraction resonances is tuned to trade in plasmonic confinement for the quality factor. Indeed, collective metasurface resonances such as SLRs or quasi-BICs generally support significantly high Q-factors (in particular for low-loss dielectric implementations) at the cost of relatively high mode volumes. The high Q-factors make them suitable for lasing, entangled-photon generation, and luminescence enhancement.⁷¹ For example, a 40-fold enhancement by quasi-BIC for color centers in silicon metasurfaces has been reported⁸¹ [see Figs. 5(a)–5(d)]. For germanium QDs, over three orders of magnitude luminescence enhancement was obtained in dielectric Fano-resonant metasurfaces.⁵⁸ Efficient lasing and strong coupling enabled by collective metasurface resonances have been demonstrated, e.g., in a plasmon-exciton-polariton laser, exhibiting a reduced threshold in the strong coupling regime.⁸² 

Other crucial enabling features of metasurfaces are their areal nature and the diversity of far-field spatial character that their modes can be engineered to support, in particular for spatially varying architectures. As such, the radiative decay rate enhancement can also serve to tailor the emission directionality by channeling the emitted light into carefully tailored spatial modes. This is usually achieved by adjusting the meta-atom geometry or their arrangement,⁸³ which can span from a basic periodic lattice to a complex, spatially inhomogeneous distribution, as well as the substrate/superstrate material. Tailoring directionality is key for realizing light-emitting devices with efficient outcoupling,⁸⁴ emission into a reduced solid angle for augmented/virtual reality (AR/VR) applications³⁵ and generating arbitrary emission patterns,^48,60,85 e.g., for projectors or light-field displays. The combination of the high Q-factors and topological properties of quasi-BICs can also be harnessed for beam shaping, e.g., for generation of vortex laser beams.⁸⁶ 

Many envisioned applications of light-emitting metasurfaces require dynamic, controllable, and reversible modulation of light emission. Such post-fabrication tunability can be achieved using active materials, which change their optical properties under electrical, optical, or thermal stimuli. Recently, sub-picosecond modulation of photoluminescence has been achieved in semiconductor metasurfaces through optically induced free-carrier effects,⁸⁸ which constitutes the current state of the art in modulation speed. On the other hand, large modulation depths can be provided by liquid crystals (LCs) or changing the metasurface geometry, e.g., by stretchable deformation or using micro-electro-mechanical systems (MEMS). The use of both approaches for actively tuning spontaneous emission has already been demonstrated,^89,90 making them promising future platforms for dynamic light-emitting metasurface devices.

Moreover, an important area of application for light-emitting metasurfaces is the creation and manipulation of circularly polarized light—a property that is often desired in classical and quantum optical information processing, communication, sensing, and displays. Several strategies have emerged in this field. In one scenario, circular polarization of the emission is enforced by the metasurface design, while the emitter can be arbitrary. This can be achieved, e.g., by using chiral meta-atoms, photonic spin-valley locking⁹¹ or optical Rashba effect, where lack of inversion symmetry causes splitting of optical spin-polarized states in momentum space. Alternatively, the utilization of chiral quasi-BICs was introduced as a compelling method for narrowband directional chiral light emission⁸⁷ [see Figs. 5(e)–5(h)]. In the second scenario, chiral metasurfaces are combined with emitters of circularly polarized light. Prominent examples include the usage of chiral metasurfaces for circular dichroism enhancement or selective coupling with valley-polarized excitons in 2D TMDs. The latter is often motivated by the demands of valleytronics. As to the valleytronic applications of chiral metasurfaces, they, however, face notable limitations as they often promote only one specific valley and would require nanoantenna geometry changes for switching. This leads us to the third scenario, where achiral metasurfaces enable control over the circularly polarized emission. Specifically, achiral metasurfaces were proven to allow for control of the lifetime, the degree of valley polarization, and the spectral shape of chiral 2D TMD emission.^69,92 An example of chiral 2D TMDs coupled to the achiral metasurfaces is shown in Figs. 5(i)–5(k). However, a comprehensive understanding of the coupling mechanisms in such complex hybrid systems, which would be essential for testing their applicability for valleytronics, is still missing. Notably, coherent coupling between excitons and meta-atom resonances offers new avenues for the tunability of chiral properties, including via the magneto-optical effect.⁹³ 

Various numerical methods, including rigorous coupled-wave analysis (RCWA), finite element method (FEM), and finite-difference time-domain (FDTD), are used to simulate light-emitting metasurfaces.^94,95 RCWA is well-suited for periodic metasurfaces with low refractive index contrast excited by plane waves, and directly computes reflected and transmitted diffraction orders, but requires extra steps for near-field reconstruction. FEM and FDTD can simulate both periodic structures with plane wave excitation and finite structures with open boundary conditions. Importantly, for light-emitting metasurfaces, point dipole excitation mimicking a nanoscale emitter is possible for these techniques.⁹⁶ Emission calculations for infinite periodic systems are not as straightforward, since periodic boundary conditions replicate a single emitter in the unit cell to an infinite coherent periodic set. For such systems a Floquet transformation is required, also known in the RF antenna community as the so-called “array scanning method.”⁹⁷ Alternatively, reciprocity-based methods that relate fluorescence to absorption can be used, which are particularly powerful to simulate extended ensembles of incoherent emitters.⁹⁸ 

Among the nanofabrication techniques, electron beam lithography (EBL) and focused ion beam milling (FIB) provide state-of-the-art nanoscale precision while preserving high flexibility, and were generally empowering research on metasurfaces over the last decade.⁹⁹ The key challenge for the fabrication of light-emitting metasurfaces is the integration of nanoscale emitters into the metasurface architecture. Typically, this is achieved either by fabricating the metasurfaces directly from light-emitting materials or by hybridizing passive metasurfaces with active materials post-fabrication. A typical materials choice for the first approach are direct bandgap semiconductors, including epitaxially grown structures incorporating quantum wells or quantum dots.¹⁰⁰ The latter approach can be achieved, e.g., by spin-coating the metasurfaces with nanoemitter or dye-containing polymers^100,101 or by depositing or chemically binding emitters such as QDs on the surface. 2D or layered active semiconductors are obtained from bulk crystals via exfoliation or grown directly by chemical vapor deposition (CVD). Among the integration techniques, one distinguishes dry and wet transfer as well as various strategies of direct integration.⁶⁸ 

While the field of light-emitting metasurfaces has witnessed significant progress to date, several challenges remain to be overcome to further develop the metasurface platforms for future practical applications. Here, we provide some suggestions for progress in this regard.

As light-emitting metasurfaces consist of two primary components, namely the metasurface itself and the nanoscale emitter, robust material platforms for both, as well as effective integration techniques are required. Importantly, high radiation efficiencies can only be obtained by using dielectric materials with minimal losses. With much of the recent research on passive metasurfaces resorting to a convenient implementation in silicon, metasurfaces made from wide-bandgap materials such as titanium dioxide (TiO₂), gallium phosphide (GaP), zirconium dioxide (ZrO₂), or lithium niobate (LiNbO₃) can be game-changers for the visible spectral range, which is of particular interest for applications of light-emitting metasurfaces.

Regarding the emitters, high quantum efficiency along with strong optical and thermal stability is desired. Importantly, for both metasurface and emitters, fitting the industrial needs and compatibility with electrical devices must be considered. Here, electrical driving schemes are of particular interest, considering that most research works on light-emitting metasurfaces still rely on optical pumping.

As such, many hybrid material platforms and integration schemes remain to be explored. For example, fluorescent ions such as trivalent lanthanides could be doped into the constituent materials of the metasurface during the fabrication process or implanted afterwards using ion implantation.

Apart from materials and integration, important challenges remain in terms of nanofabrication. First, Q-factors in low-loss quasi-BIC metasurfaces are currently limited by fabrication imperfections such as roughness.¹⁰² Thus, it is highly desirable to further improve the structure quality to unlock a range of effects dependent on high Q-factors. Second, another grand challenge in terms of fabrication, especially regarding application perspectives, is scalability. While this holds for most metasurface applications, the challenge is even greater for light-emitting metasurfaces due to the need to integrate the active material. For example, although deep-UV photolithography has been proposed as a viable solution for complementary metal–oxide–semiconductor (CMOS)-compatible wafer-scale metasurface production,¹⁰³ light-emitting materials are usually not CMOS-compatible, requiring new approaches to be developed. For example, a combination of nanoimprint lithography and selective area sublimation has been proposed for large-scale production of GaN light-emitting metasurfaces.¹⁰⁴ Other alternatives may include, e.g., laser-induced transfer, suitable for creating large-scale periodic arrays of light-emitting nanoparticles.¹⁰⁵ Third, the precise and deterministic placement of the active material only at desired locations within the metasurface geometry still presents a challenge. This can be, e.g., the gap between two meta-atoms where the fields are significantly enhanced, asymmetric locations within the unit cell for beam-steering effects, or positions featuring specific local field characteristics such as chiral density. Suitable placement strategies may be adopted from the field of nanoantennas, with approaches relying on two-step electron-beam lithography showing particular potential for metasurfaces. The large areas that the metasurface spans also suggest that nanoimprint methods may be used to print emissive materials in a registry with meta-atoms if sufficiently precise spatial registration can be achieved.

A critical issue concerning a wide range of emitters, spanning from fluorescent dyes to QDs and perovskites is photobleaching, where the excess energy delivered during the excitation process results in the reduction of the emitted light intensity from the emitters over time. Photobleaching is attributed to both pumping conditions and intrinsic properties of the emitters. Therefore, an adequate emitter choice is crucial for minimizing sample photodamage. In this regard, interesting opportunities may be offered by highly stable emitters like nitrogen vacancies and nanodiamonds.¹⁰⁶ 

The light–matter interaction in the optical frequency range mainly refers to the interaction of the electric field with electric dipolar (ED) electronic transitions. However, certain solid-state emitters, such as trivalent lanthanides and transition-metal-doped crystals, exhibit robust magnetic dipolar (MD)-dominated electronic transitions as a consequence of a selection rule forbidden ED transition.¹⁰⁷ Due to a strong optical magnetic response, all-dielectric metasurfaces were suggested as exquisite candidates for tailoring the fluorescence properties of MD transitions, opening an exciting pathway toward exploring magnetic light–matter interactions at the nanoscale^57,72 [see Figs. 5(l)–5(o)]. A particular challenge associated with commonly studied MD transitions is their slower radiative decay rate and consequently lower emission intensity compared to their ED counterparts. However, in a recent study, a significantly bright and fast optical-frequency MD radiation is identified in 2D hybrid organic-inorganic perovskites (HOIPs) such as butylammonium lead iodide (BA₂PbI₄).¹⁰⁸ It is shown that the MD radiative rate in such systems is up to 3 orders of magnitude faster than previously established MD transitions. Additionally, similar to trivalent lanthanides, 2D HOIPs exhibit spectrally separated ED and MD radiations. MD radiation properties, e.g., emission directionality of such emitters, can be engineered by integrating them with suitable nanophotonic platforms. Particularly, metasurfaces supporting high-Q resonances are elegant tools for enabling directional color and/or polarization routing of the emission from MD and ED transition channels. This opens new perspectives on realizing spectrometer-free, low-loss, and CMOS-compatible nanophotonic devices for imaging, sensing, and probing.

When dealing with 2D materials as active components, their ultrathin nature presents both advantages and drawbacks. Notably, it entails limited overlap with the excitation field, resulting in a low signal in absolute units. To address this, employing high-Q metasurfaces is needed to boost local pump field enhancements, while also being key for superior Purcell enhancements. A significant limitation of all valleytronic devices is the cryogenic temperature requirement due to the presence of phonon-assisted intervalley scattering that prohibits valley polarization under normal conditions. A promising avenue toward realizing non-zero valley polarization at room temperature involves the utilization of TMD-containing van-der-Waals heterostructures. These artificial materials can support a flurry of interesting effects and applications, one of them being that valley polarization can be augmented through mechanisms such as charge or spin transfer and other proximity effects. Adding to the general fabrication challenge, for realistic device applications methods for scalable and high-quality fabrication of 2D materials and heterostructures have to be developed. Here, the CVD-based phase engineering techniques can pave the way for novel heterostructures and junctions with unique electronic phases and customized properties.

Regarding the design of light-emitting metasurfaces, a grand challenge is posed by the high computational demand of suitable approaches like those based on the reciprocity principle. These demands are typically higher than for other metasurface applications: the emission process needs multi-wavelength (optically pumped) and multi-physics analysis (including, e.g., electrical and thermal effects), coherent or incoherent summing over many source positions, and when translated by reciprocity into a scattering problem it involves a continuum of wave vectors. Finally, emissive devices such as LEDs often rely on metasurfaces in complex vertically stratified stacks. Generally in metasurface research, there is a growing preference for optimization algorithms over human-driven approaches due to their capability of efficient systematic investigation of parameters, as detailed in Secs. XVIII and XX. New, highly performing algorithms may thus prove especially useful for active architectures and help to come up with spatially variant architectures of unprecedented complexity or help to tackle specific design issues associated with the integration of emitters. For example, quasi-BICs can be engineered to suppress absorption losses, enabling, e.g., plasmonic-dielectric metasurfaces with theoretically arbitrary high Q-factors,¹⁰⁹ which can potentially lead to high-performance metasurface designs incorporating light-emitting and thus inevitably also absorbing materials.

To be competitive as commercial light sources, metasurfaces must reach specific performance values. A challenge for high power applications, such as LEDs is the required combination of high emitter density, high emitter efficiency, and resistance to photobleaching and thermal load. Inherent to the process of emission is that the Stokes shift energy is dissipated as heat, often meaning 0.5–1 eV of dissipated energy for every emitted photon. For example, a metasurface-based LED should withstand of order 1 A/mm² input current while maintaining power-to-photon conversion and power handling capacity sufficient for providing 1 W/mm² emission intensity. These goals require proper material engineering to obtain sufficiently bright and photostable emitters—the associated dissipated power, or in the case of a color converting layer, the Stokes-energy loss in such systems is also of order 1 W/mm². For metasurface lasers, it is important to realize that a very high benchmark is set by III–V surface-emitting lasers that provide continuous-wave output power on the level of 100 W and brightness >1 GW/cm²/sr at power-to-photon conversion >0.5 W/A, while maintaining decent beam quality (M²). This typically requires large-area single-mode operation, which is particularly challenging for metasurfaces. However, the above figures have recently been approached by photonic crystal surface-emitting lasers.¹¹⁰ 

In addition to simple light sources, many of the emerging applications of light-emitting metasurfaces, such as LiFi, LIDAR, AR/VR, require a reliable way of tuning the emission. In that regard, LCs and MEMS are the most mature technologies, which are also CMOS-compatible and therefore suitable for large-scale and low-cost production.^89,90 Both approaches can offer a large modulation depth while providing a modulation speed that is sufficient for most commercial uses. However, direct integration of MEMS and LCs for specific applications of light-emitting metasurfaces remains to be developed.

In addition to the real-world applications, the research on light-emitting metasurfaces continues to explore unconventional directions such as nontrivial topology,¹¹¹ non-Hermitian physics/parity-time (PT) symmetry,¹¹² and temporal photonics.⁵⁰ Topological robustness enables various systems (including, e.g., photonic topological insulators, BICs) to exhibit fragile properties regardless of imperfections, which has been studied especially in the context of large-area single-mode lasing.¹¹³ Photonic systems with gain and loss can support non-Hermitian features known as exceptional points (EPs), which can be used to control the directionality and polarization of emitted light.⁵⁴ In temporal photonics, sufficiently fast time modulation can give rise to exotic physical phenomena related to light emission and lasing, such as new mechanisms of light amplification.¹¹⁴ Currently, the concepts of topological, non-Hermitian, and temporal photonics remain mainly at the level of fundamental research. However, in the future, some of these concepts may revolutionize the field of light-emitting metasurfaces, enabling unprecedented performance and novel functionalities.

Light-emitting metasurfaces represent a cutting-edge area of research in nanophotonics, offering dynamic control over the properties of emitted light from the incorporated nanoscale light sources. Furthermore, thanks to recent developments in the field of nanofabrication and material processing, light-emitting metasurfaces hold great promise for a wide range of practical applications from advanced displays and communication systems to imaging devices, sensors, and beyond. Although challenges such as energy efficiency and integration with existing technologies need to be addressed to reach their full potential, the ongoing research and development in this area lead us to expect a bright future for advanced light-emitting metasurface devices with novel functionalities as well as their integration into real-world applications. Finally, metasurfaces also offer a unique platform for future scientific applications investigating fundamentally new directions in the area of light emission.

I.S., A.B., and Z.F. acknowledge financial support from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through the International Research Training Group (IRTG) 2675 “Meta-ACTIVE,” Project No. 437527638, and through the Emmy Noether Program, Project No. STA 1426/2-1. The work of A.F.K. is part of the Dutch Research Council (NWO) and was performed at the research institute AMOLF. R.K. acknowledges the support of the Research Council of Finland (Grant Nos. 347449 and 353758).

Xiaofei Xiao, John Yang, and Rupert. F. Oulton^*

^*[email protected]

The field of nonlinear optics, emerging in the 1960s alongside lasers, explores light–matter interactions under extreme light intensities.¹¹⁵ Effects, such as harmonic generation, three- or four-wave mixing and parametric amplification, have expanded the operational bandwidth of coherent light sources from the extreme ultraviolet to terahertz and produced methods to measure and control light on ultrafast time scales. Applications extend to quantum optics, where parametric downconversion and spontaneous four-wave mixing produce correlated and entangled photon pairs for quantum enhanced sensing, imaging and computing.^116–121 

Nonlinear processes such as harmonic generation and four-wave mixing can be remarkably efficient when beams being mixed propagate in phase over many millimeters of interaction in a nonlinear crystal. Phase-matching brings material and technical restrictions as it comes at the cost of reduced tuning range and material flexibility. For example, phase-matching may only be achieved at specific wavelengths and temperatures and the large interaction lengths require low absorption. For some photonic applications, new nonlinear materials that can be integrated with the most relevant photonic platforms are important, such as polymers^122–124 and 2D materials.^125–128 Nonlinear optical metasurfaces also hold promise for addressing these challenges [Figs. 6(a) and 6(b)]. First, optical nanostructures can be designed at the nanoscale to enhance local fields and meet symmetry requirements to produce strong nonlinear effects, as we discuss in detail below. Second, nonlinear metasurfaces do not use phase-matching but instead rely on their high intrinsic nonlinearity to produce a useful nonlinear effect. This alleviates tuning, bandwidth and material absorption restrictions. Third, nanostructured metasurfaces can be integrated with a wide variety of nonlinear materials to further strengthen their intrinsic nonlinearity. Finally, their engineered nature allows for nonlinear wavefront control. These design features have led to a burgeoning field of nonlinear metasurfaces.

In this perspective, we present the state-of-the-art in nonlinear photonic metasurfaces, by reviewing the rich variety of mechanisms to engineer a nonlinear response at the nanoscale. Before considering future directions for this field, we present a short study of how the metasurface interaction area affects nonlinear mixing efficiency as a function of incident beam powers and areas. To conclude, we identify the most useful metrics for assessing nonlinear metasurface performance and future direction this field might take.

Resonances play a crucial role in nonlinear optical metasurfaces, since the linear optical response often dictates the nonlinear one.¹²⁹ The relationship between linear and nonlinear susceptibilities is well known,¹¹⁵ but are not often utilized in phase-matched processes due to the constraints of absorption on interaction length. However, thin metasurfaces can utilize resonance to produce strong nonlinearities. Through structural resonances, metasurface can also enhance the electric and magnetic fields of incident light. These are usually categorized into local and collective resonances.¹³⁰ Local resonances, such as surface plasmons,¹³¹ Mie modes of metallic and dielectric nanoparticles,^132,133 and Fano resonances,^134,135 utilize nanostructures that operate independently across the metasurface, while collective resonances, such as guided-mode resonances,^136,137 surface lattice resonance,^70,138,139 and bound states in the continuum^140,141 (BIC), access mutual coherences between nanostructures.

Localized surface plasmon resonances (LSPR) are collective oscillation of free electrons in metal nanoparticles (typically gold or silver) excited by incident light, where resonance is controlled via size, shape, and particle composition. When integrated into metasurfaces, these resonances enhance second harmonic generation (SHG)¹⁴³ [as shown in Fig. 6(c)] as well as a range of mixing effects^53,158–160 including third harmonic generation (THG), four-wave mixing, and sum frequency generation, due to their ability to provide strong local light field enhancement. Wavefront control has also been shown [Fig. 6(d)] using the globally varied local nonlinear harmonic response of gold nanoantennas of C3 and C4 rotational symmetries, respectively.¹⁴⁴ Although metal provide excellent confinement, the high nonlinearity of the metal is difficult to use, since metals do not easily admit electric fields—this means interaction volumes are relatively small. Simple plasmonic resonances also provide relatively weak local field enhancement and structures with sharp features and gap that do enhance fields have low damage threshold.^132,161 For example, typical quantum efficiencies of SHG are $∼ 10 − 9$⁠.

Metasurfaces made of high refractive index dielectrics, such as silicon, germanium, and gallium phosphide, have emerged recently as a competitive alternative to plasmonic metasurfaces, due to low material loss, larger interaction volumes and the capability to engineer their magnetic response. Their efficiencies compared to plasmonic metasurfaces show improvement by several order of magnitude, mainly due to the higher damage threshold (see Sec. III C).^151,162 These metasurfaces use Mie resonances to control the nonlinear interaction. Although they exhibit some ability to enhance optical fields, the larger interaction volume provides the largest benefit.

Individual Mie resonance modes,^132,135,145 such as electric, magnetic dipoles, and higher-order Mie resonances, have been explored individually to enhance nonlinear conversion efficiency beyond plasmonic metasurfaces.^{132,133,163,164} For example, Fig. 6(e) shows that THG conversion efficiency in silicon nanodisks is enhanced by magnetic dipole resonances by two orders of magnitude compared to bulk silicon.¹⁴⁵ 

To increase the performance further, resonant coupling is often used. Interference of multiple Mie resonances may result in far-field destructive interference, consequently enhancing the nonlinear effect. For example, the anapole state, associated with the destructive interference of dipole and toroidal moments, suppresses scattering to produce near-field enhancements.¹⁶⁵ Both fundamental and a higher-order anapole states in individual all-dielectric nanodisks have been implemented.^146,166,167 Figure 6(f) shows a fundamental anapole state in a single Ge nanodisk for THG.¹⁴⁶ Here, a minimum in extinction cross section corresponds to a maximum electric energy within the material, producing THG quantum efficiencies as large as $10 − 6$⁠. Coupled resonators also provide stronger nonlinear responses through their mutual coupling.^168–171 For example, Fig. 6(g) shows a nanotrimer of three silicon nanocylinders that improves THG via the interference of electric and magnetic dipoles.¹⁴⁷ 

The nonlinearity of nanostructures can be further improved by combining the confinement capabilities of metallic resonances and the interaction volume and radiative control of dielectric resonances.^172,173 For example, the plasmonic resonance of a Au nanoring around a Si nanodisk supporting an anapole state¹⁴⁸ [Fig. 6(h)] shows THG quantum efficiency as high as $10 − 4$⁠.

Fano resonance arises from the interference between a broad continuum of states and a narrow discrete resonance.¹³⁴ In metasurfaces, this can be realized through the coupling of plasmonic elements with a continuum of modes, leading to asymmetric line shapes and sharp spectral features. The intense and localized electromagnetic fields associated with Fano resonances can greatly boost nonlinear frequency mixing processes.^149, Figure 6(i) shows enhanced high-harmonic generation from an all-dielectric metasurface,¹⁴⁹ where non-perturbative high-harmonic generation is enhanced in a Fano-resonant Si metasurface. Figure 6(j) shows magnetic-Fano-resonance enhanced THG signal via interplay between collective and individual optically induced magnetic responses in quadrumers made of identical dielectric nanoparticles.¹⁵⁰ Plasmonic nanostructures show similar nonlinear enhancements.^169,174 We note nonlinear generation enhanced by Fano resonances are usually anisotropic with respect to the excitation polarization and have narrow operation bands due to their resonant nature.

Collective high-Q resonances in all-dielectric metasurfaces, such as bound states in the continuum (BICs) can offer large improvements in nonlinear metasurface response. BICs are eigenstates of a system that exist within the continuum but do not radiate energy into the surrounding space. They were initially introduced in quantum mechanics and later extended to photonics.^{71,140,141,175,176} Mathematical BICs may have vanishing resonant linewidths. In practice, due to finite sample size, material absorption, small amount of radiation leakage, and structural imperfections, BICs manifest with reasonably large Q factors, known as quasi-BICs. The long lifetimes of quasi-BICs are highly desirable for concentrating light in time and thus the nonlinear optical process.^177,178 Figure 6(k) shows SHG in a dielectric (gallium phosphide) metasurface enhanced by two asymmetric cylinders with quasi-BIC resonances.¹⁵² 

Quasi-BICs can also be engineered as individual local resonators by exploiting the interference of several Mie modes, similar to the anapole state discussed earlier.^146,166,167 For instance, AlGaAs cylinders placed on an engineered three-layer substrate (SiO₂/ITO/SiO₂) show suppressed radiative losses, as shown in Fig. 6(l), boosting SHG efficiency.¹⁵¹ 

Surface lattice resonances (SLR) also achieve high Q-factors. When a grating Rayleigh anomaly condition crosses a LSPR, a sharp spectral feature occurs, due to the hybridization of the surface and localized resonances.⁷⁰ The SLR feature is controlled via nanostructure shape and lattice parameters. The effect of SLRs at the fundamental¹³⁹ and the generated nonlinear modes¹³⁸ has been studied for boosting the nonlinear optical response.^179, Figure 6(m) shows SLRs used to enhance SHG from a plasmonic nanoparticle array made of split-ring resonators¹³⁸ where the SHG intensity improved more than a thirty-fold. Nonlinear metasurfaces based on both localized and collective plasmonic resonances have also enabled the generation of controllable terahertz waves,^180,181 using mechanisms like ponderomotive acceleration of photo-ejected elections¹⁸² and optical rectification.¹⁸³ 

At the boundary between metals and dielectrics, the nonlinearity of materials with epsilon-near-zero (ENZ) points have also been studied^184–188 (see Sec. XVII). Where ENZ occurs, the surface normal electric field of incident light diverges when the imaginary part of the permittivity approaches zero. This produces field enhancements inversely proportional to ENZ film thickness¹⁸⁹ that boost nonlinear effects.^184–188 Their unique nonlinear optical response and the required deep sub-wavelength thickness of the ENZ film also makes them appealing for high-harmonic generation.¹⁵³ Indium-doped cadmium oxide thin films (75 nm) show harmonic generation up to the ninth order,¹⁵³ as shown in Fig. 6(n). More on these materials is presented in Sec. XVII of this roadmap. Figure 6(o) shows a metasurface combining plasmonic antennas with an ENZ material.¹⁵⁴ Here, the narrow spectral range and ENZ position of the underlying ENZ material could be engineered to significantly boost the nonlinear optical response.

In attempts to maximize the nonlinearity of metasurfaces, researchers have integrated nonlinear materials with metasurfaces to exploit both material and structural resonances. For example, a 100-path spontaneous parametric downconversion photon-pair source was created by integrating a metalens array (10 × 10 array of GaN nanopillars) with a β-barium borate (BBO) nonlinear crystal.¹⁹⁰ Another example uses low-loss nonlinear polymer within nanoplasmonic systems.^{124,191–193} Figure 6(p) shows efficient four-wave mixing over micrometer-scale interaction lengths at telecommunications wavelengths on silicon can be achieved by combining an integrated plasmonic gap waveguide with a nonlinear organic polymer. Similar ideas have been used to focus light into a nanoscale cavity to enable frequency upconversion from Mid infrared to visible,^155,194 via molecular vibrations [Fig. 6(q)]. Highly resonant BIC metasurfaces may also be integrated with highly nonlinear materials. Figure 6(r) shows a radial BIC was used to boost SHG from monolayers of transition metal dichalcogenide.¹⁵⁶ This brings two advantages: the two-dimensional materials are straightforward to integrate within metasurfaces and their atomic thickness minimizes the deformation of the resonance.

Finally, we briefly consider the tensorial nature of non-linear interactions and their corresponding selection rules. Local field enhancement and resonances are not alone sufficient for strong nonlinear effects to be achieved. The non-linear polarization created by pump light in a metasurface must also be able to radiate to the far field to complete the nonlinear process. This requires that the nonlinear tensor components of the metasurface be matched to the desired input and output beams. Moreover, the interaction of the input and output beams with the metasurface should be optimized by both impedance matching, and spatial mode overlap at the various interacting frequencies.^151,195,196 Ideally, the suppression of all radiation channels apart from those for power injection and extraction has been proposed to increased coupling to and from the metasurface,¹⁵⁷ as schematically shown in Fig. 6(s). Indeed, this has been experimental investigated both in plasmonic¹⁵⁸ [as shown in Fig. 6(t)] and all-dielectric^145,197,198 metasurfaces.

Conventional nonlinear materials produce large conversion efficiencies by mediating nonlinear processes over a large interaction volume.¹¹⁵ Thus, large area metasurfaces should produce the highest nonlinear generation rates. To elaborate on this, consider the following analysis of nonlinear frequency mixing at a metasurface.

These SHG efficiencies are both influenced by mode area, A₀, and pulse duration, τ₀. Clearly, the best conversion rates occur when the incident average pump power is high and the mode area and pulse duration are small. Inevitably, SHG is limited by the damage threshold intensity of a metasurface, I_D. In our experiments with plasmonic metasurfaces using near diffraction limited beams from either mode-locked oscillators (Ω = 80 MHz, $τ 0 = 150$ fs) or continuous wave lasers, we have found damage occurs at about 5 mW average power. For a diffraction limited area of $0.25 μ m 2 , I D ∼ 10 11$ W cm^–2 for pulsed operation and $I D , c w ∼ 10 6$ W cm^–2 for CW. Damage here is due to sample heating, and so the lower repetition rates of amplified laser systems can allow higher intensities. Dielectric metasurfaces have slightly higher damage threshold intensities than plasmonic one, but are not immune to damage at intensities $I D > 10 12$ W cm^–2 where ionization effects occurs. We note that at these intensities, non-perturbative nonlinear process start to emerge, and the preceding analysis requires reevaluation.

The quantum efficiency and average SHG power are plotted in Figs. 7(b) and 7(c) as a function of normalized power and mode area. Along the power axis, normalization against both I_D and diffraction limited mode area, A_D, leaves the axis dimension-less. Likewise, we have also normalized the beam area against A_D. [We assumed a pump wavelength, $λ = 1 μ m$ with a diffraction limited mode area: $A D = π ( 0.41 λ ) 2$⁠.]

The power independent efficiency, η_p, might seem like the most appropriate metric, but it predicts optimal conversion efficiency at low beam area. Indeed, this is where the nonlinear effect is strongest, since the beam intensity is maximized. However, the scaling of SHG with pump power squared suggests that large mode areas and high pump powers should produce optimal SHG. The quantum efficiency proves to be a more useful metric; although it varies with pump power, it can be evaluated in a way that is independent of both beam area and pulse duration, $η q = α n P ¯ 0 / [ 2 2 A 0 Ω τ 0 ] = α n I 0 / [ 2 2 ]$⁠. As shown in Fig. 7(b), contours of η_q are linear and pass through the origin, so the gradient gives the intensity. If the damage threshold of the metasurface is known, its limiting performance can be completely specified at any power and beam area.

Figure 7(c) shows a plot of average SHG power as a function of average pump power and beam area. The plot confirms that optimal SHG occurs when the pump power is maximized and the mode area is minimized to operate at the damage threshold intensity. This power-area scaling presents challenges for applications of metasurfaces. The write field of electron beam lithography is typically $100 × 100 μ m 2$⁠, representing an upper limit on metasurface area. A beam area of $10 4 μ m 2$⁠, at damage threshold intensity requires an average beam power of $∼ 100$ W!

Metasurfaces cannot compete with phase-matched nonlinear processes in terms of efficiency, and the power-area analysis above suggests there are a few routes forward. Metasurfaces offer strong resonant nonlinearities, wave-front control and the possibility to integrate new materials. In this final section, we review applications in wave-front control, optical switching, and accessing non-perturbative nonlinear physics at reduced power.

The capability to locally engineer optical nonlinearity in metasurfaces provides the means to engineer local nonlinear phase. Here, sub-wavelength nanostructures are tailored to control the phase and amplitude and direction of the nonlinear polarization.^2,199,200 Recently, nonlinear wavefront control has been demonstrated using plasmonic^144,201,202 and all-dielectric^203,204 nanostructures with specific structural symmetries from the extreme ultraviolet to terahertz.^{53,180,181,205} Beam shaping based on binary nonlinear phase and holography based on continuous nonlinear phase have been demonstrated by nonlinear optical wavefront engineering techniques.^144,206,207 Currently, there are a few studies in this area and with the attention being dedicated to controlling local nonlinear phase, conversion efficiencies are relatively low. Additionally, wavefront engineering techniques for various optical beams,^53,208 such as multiple optical vortex beams (Sec. VII), have been realized using linear optical metasurfaces. Transferring these techniques developed for linear optical metasurfaces into the nonlinear regime will open a new avenue for the realization of compact and ultrafast nonlinear optical devices.

Although nonlinear metasurfaces open new avenues in flat nonlinear optics, the majority cannot be actively tuned post-fabrication despite this being a key feature in some linear metasurfaces (see Secs. II, XIV, and XV). Dynamic control over nonlinear metasurfaces under external stimuli is highly desirable in future. The mechanisms of the tunability include but are not limited to thermo-optic effects, free-carrier effects, and phase transitions.²⁰⁹ The electrical modulation of a nonlinear response based on electric-field-induced SHG or optical rectification has been investigated in both plasmonic or dielectric metasurfaces.^210–214 While this research direction is in its infancy, there are clear opportunities that play to the strengths of metasurfaces: thin films can be electrically accessed and modulated and resonant nonlinearities are more amenable to tuning with stronger modulations in material parameters.

The intrinsic nonlinearity of materials is measured by the nonlinear susceptibility,¹¹⁵  $χ ( n )$⁠. For example, SHG is governed by the value of $χ ( 2 )$ and naturally transparent crystals have $χ ( 2 ) ∼$ 1–100 pm V^–1. When multiplied by incident electric fields of light, the dimensionless product indicates nonlinear strength. For an n-order process, the dimensionless parameter $χ ( n ) | E | n − 1$ is a metric of nonlinearity. For example, SHG with 1 W of CW light over a $10 4 μ m 2$ beam area, has $χ ( 2 ) | E | ∼ 10 − 8$– $10 − 6$⁠. Meanwhile, strong nonlinear effects occur for $χ ( n ) | E | n − 1 ∼ 1$⁠, where the perturbation description of the nonlinear polarization breaks down. Achieving this condition requires either more intense light or a more nonlinear material. For example non-perturbative effects such as high harmonic generation occur at intensities $I > 10 13 W cm − 2$ in gases, and slightly lower in solids. While amplified pulsed lasers can access non-perturbative intensities, lowering the intensity at which non-perturbative effects begin is appealing for reducing cost, complexity, size, weight and power. The increased nonlinearity of metasurfaces are a promising route.

Nanostructures have been used to control non-perturbative high harmonic generation effects both in isolated structures^215–219 and metasurfaces.^{149,220–224} High-harmonic generation has also been shown in ENZ based metasurfaces.¹⁶⁵ Such applications play to the strengths of metasurfaces and would enable an expansion of research teams able to study non-perturbative nonlinear physics across different disciplines by mitigating the need for expensive and bespoke laser systems. Metasurface nonlinearity can be extremely large: to our knowledge, $χ ( 2 ) = 1.2$ μm V^–1 is the highest nonlinear response reported for a metallic metasurface coupled to inter-sub-band quantum well transitions in the mid-infrared,^52,225 yielding $χ ( 2 ) | E | ∼ 1$ under CW illumination. Such giant nonlinear responses have been seen in metasurfaces combining both electromagnetic and material resonances.

William J. Peveler and Alasdair W. Clark^*

^*[email protected]

The advancement of technologies for the rapid and sensitive detection of biological macromolecules (e.g., nucleic acids, proteins, glycans) and whole organisms (e.g., viruses, bacteria, and other pathogens), are crucial to furthering our understanding of biological systems, to the development of new diagnostic tests and healthcare tools, to improved drugs and therapeutics, and for ensuring the safety of our food and water supplies. Therefore, as new developments are made in materials science, one of the first application areas of these new materials is often sensors. This is also true of nanophotonic metasurfaces. Indeed, metasurface biosensors have emerged as one of the most notable success stories in the field of nanoscale metamaterial research, transitioning quickly from an academic curiosity to practical, commercial sensing devices that are having impact in medical diagnostics, environmental monitoring, and pharmaceutical research. The term “nanophotonic metasurface” relates to devices comprised of two-dimensional arrangements of nanoscale building blocks (meta-atoms), whose collective optical properties are derived from their rational design and engineered array properties, rather than purely from their chemical composition.¹⁹⁹ These surfaces can be constructed using metallic or dielectric meta-atoms which have optical resonance properties defined by their size, shape, composition, spacing, and crucially for biological detection, the refractive index of their local surroundings.²²⁶ Therefore, their resonances properties can be altered by individual biomolecules interacting with their surface. The nanoscale nature of the meta-atoms mean they are often of comparable size to these biomolecules, meaning that even single-molecular localization on a meta-atom surface can produce resonance shifts easily measurable in the far-field by simple white light illumination; shifts that can also provide information on interaction kinetics between molecules.²²⁷ Resonant meta-atoms are also able to confine light to extremely small volumes, producing many orders of magnitude enhancement to the electric field around certain facets of the nanostructure (e.g., at sharp corners and inside small gaps). Since many vibrational spectroscopy techniques have efficiencies that scale with electric-field strength, the signals recorded from molecules within these enhanced regions is greatly improved. These two aspects of resonant meta-atoms make them excellent candidates for applications where label-free, hyper-sensitive detection is required, and for applications where the interaction strength between pairs of molecules is being assessed. It has been possible to produce structures with bespoke optical properties for decades, but recent strides in fabrication technology have greatly improved the efficiency, speed, and cost-effectiveness of producing metasurface sensors. Coupled with their ease of excitation (e.g., illumination with incandescent light bulbs, LEDS, laser diodes, etc.) and the low complexity of their optical measurement, these advancements pave the way for portable, low-power, point-of-care, and field-deployable sensing solutions. In this brief roadmap we will cover a small number of research developments in the field of metasurface biosensors (some of which appear in Fig. 8), and provide our outlook on what the future of this fast-moving field may hold.

Metasurface biosensors can be broadly categorized in two groups: plasmonic (where meta-atoms are metallic) and dielectric (where meta-atoms are, typically, silicon). Plasmonic metasurfaces exploit localized surface plasmon resonances (LSPR), a nanoscale variant of the long-established biosensor technology of surface plasmon resonance (SPR).¹ Instead of generating surface plasmons that propagate along the air-metal interface of bulk, thin-film metals (as in SPR), LSPR is generated by the oscillatory interaction of light with the free electrons in the metal nanostructures. These structures are smaller than the wavelength of the exciting light, producing resonantly driven, highly localized electric fields. In contrast to SPR devices, LSPR metasurfaces require no special coupling schemes, meaning they are trivial to excite, and produce fields with very short decay lengths (only probing their immediate surroundings²²⁸—i.e., bound molecules, not molecules in the bulk), meaning they are not thrown off by environmental fluctuations (e.g., temperature). While plasmonic nano-resonators are decades old, new research into dielectric resonators is revealing key advantages. Metallic metasurfaces suffer from high losses, strong dispersion, and poor magnetic response, whereas dielectric resonance, the result of Mie resonances which generate oscillating displacement currents in response to the driving field polarization, exhibit low loss, sharper bandwidths, little heating, and can support strong magnetic dipoles.²²⁹ Despite their different resonance mechanisms, the outcomes are similar: nanostructures, resonantly driven by light, with properties tunable by altering their geometry, and which are either sensitive to their surroundings (i.e., the presence of a biomolecule), or that produce a highly localized electric field that can be used to enhance existing sensing techniques. In terms of sensor production, both wide-area, relatively inexpensive techniques, like colloidal lithography, and smaller-area, relatively slow, more expensive techniques like electron-beam lithography are commonly used in the research space. However, advances in rapid nanolithography techniques, such as nano-imprint lithography and extreme UV photolithography, mean that, even for e-beam-developed geometries, there is a clear path to practical, high-throughput production of metasurface biosensors.

The resonances that unlock applications in biosensing produce highly confined field enhancements at the meta-atom surface. Therefore, only molecules resident within this near-field (typically in direct contact with the meta-atoms) are going to influence, or be influenced by, the resonance, and be detected. So, to function as a sensor, mechanisms are required to ensure that analytes are located within the sensing volume long enough to measure. As a result, in almost all the following biosensor examples, bio/chemical modification of the metasurface is required, usually with a receptor molecule that is complementary to the target analyte (e.g., nucleic acids, proteins, antibodies, etc.). Once located there are several techniques that allow molecular presence to be confirmed, some of which can also provide information about molecular structure and binding strength of those analytes. The examples below include technologies developed to detect biomarkers of specific diseases. In those cases, the detection limits achieved are within the range of clinical relevance.

The most widely employed sensing modality involves monitoring the resonance frequency of the metasurface as it interacts with its surroundings—operating on the principle that this frequency is linked to the refractive index immediately adjacent to the nanostructure. Therefore, when biomolecules bind to or approach the nanostructure's surface, they induce an easily observable resonance shift that provides a quantitative measure of molecular presence, binding events, and interaction kinetics.²³⁰ 

As a widely used technique, there are thousands of examples of different geometries, materials, and operating frequencies that have been explored. What follows is a small selection of recent examples. Gold is the most popular material for these sensors because of its chemical stability, simplicity of modification with bio-receptors, and the ease by which visible resonances can be excited using white light.²³¹ For over two decades, the most popular surface geometry has been circular nanodisks, a geometry that can be produced using many different fabrication techniques. Since controlled array spacing is not necessarily required for many biosensing tasks, a number of wide-area fabrication examples (e.g., colloidal fabrication) exist. Examples include hole-mask colloidal lithography and photolithography to produce gold nanodisks for multiplexed measurement of diagnostic-relevant protein-protein binding affinity at pM concentration.²³² Similar fabrication techniques have been used to detect single-base DNA mutations,²³³ and others have shown that by coating the nanodisks with molecularly imprinted polymers (MIPS) instead of traditional “complementary” receptor molecules, label-free quantification of many polyphenol-protein interactions can be made.²³⁴ Other examples of wide-area fabrication for scalable sensor production include nanosphere lithography to produce Au nanopyramids to measure real-time protein (α–Synuclein)-small molecule interactions,²³⁵ and the use of nanoimprinting to create indented gold structures (“nanocaves”) for detecting tumor markers in human serum (carcinoembryonic antigen at 5 ng/ml).²³⁶ Electron-beam lithography enables greater metasurface control, providing more options for geometry and layout variation (e.g., multi-layered biosensors with internal self-referencing),²³⁷ and receptor localization (e.g., receptor “patches” adjacent to the meta-atoms for biomolecule-mediated metasurface reconfiguration).^238,239 As such, periodic arrays of e-beam produced gold structures have been used to detect prostate specific antigen (PSA), in serum, at 500 pg/ml.²⁴⁰ In a variation of the periodic nanodisk structure, Shen et al. demonstrated gold mushroom arrays with high-intensity coupled resonance modes (between a gold film and elevated gold disks) for detecting alpha-fetoprotein (a liver cancer and fetal development marker) at 15 ng/ml.²⁴¹ Similarly, high-intensity fields can be produced in small gaps between discrete structures, as demonstrated with IR-resonant nanorods for detecting human antibody IgG down to 30 pM,²⁴² and to monitor peptide-induced neurotransmitters from synaptic vesicle mimics.²⁴³ 

Recently, as the benefits of dielectric resonators has become apparent, similar functionality has been demonstrated with dielectric metasurfaces. Silicon nanodisks have been used to monitor biotin-streptavidin interaction at $10 − 10$ M,²⁴⁴ to detect breast cancer biomarkers at 0.7 ng/ml,²⁴⁵ and to identify PSA at 1.6 ng/ml (below diagnostic requirements).²⁴⁶ Low-cost fabrication has also been explored in this space, with PSA detection shown using silicon disks produced via colloidal lithography,²⁴⁷ and, in efforts to produce cost-effective meta-sensors with CMOS camera compatible readouts, Triggs et al. demonstrated chirped silicon-nitride gratings for characterizing antibody binding.²⁴⁸ More complex shapes are also emerging, with crescent structures supporting quasi-BIC (bound states in the continuum)¹⁴⁰ having achieved 0.16 nM protein detection,²⁴⁹ and where, notably, Altug's laboratory have manipulated the angles and spacings of germanium and silicon nano-ellipse structures, and combined them with hyperspectral imaging, to fingerprint various biomolecular events at high sensitivities.^250,251

Chirality is a fundamental property of many biomolecules, including amino acids and nucleotides, where molecules exist in two forms that are mirror images of each other, known as enantiomers. This characteristic is crucial in biochemistry and pharmacology because enantiomers of the same substance can exhibit vastly different biological activities, with one form potentially beneficial and the other harmful. Chiroptical spectroscopic techniques can identify and characterize chiral biomacromolecules, but their sensitivities can be greater than six orders of magnitude lower than refractometric plasmonic sensors,²⁵² meaning they are largely unsuitable for detection of enantiomers at diagnostically relevant concentrations (which can be pg/ml). The enhanced fields exhibited by resonant metasurfaces can significantly enhance the chiroptical response, enabling more sensitive detection of chirality. By using geometrically chiral plasmonic structures, so-called superchiral fields can be generated,²⁵³ optimizing the selective absorption of light by chiral molecules and enhancing CD signals. This provides a more efficient means for the detection, analysis, and separation of enantiomers, offering improved sensors for pharmaceutical development and diagnostics. The quest to generate superchiral fields for these purposes has led to a number of plasmonic metasurfaces which exhibit complex meta-atom geometries. Notable examples include the use of gold shuriken indentations, fabricated by injection molding, to detect pg quantities of helical biopolymers,²⁵² perform multiplexed sensing of proteins and virions,²⁵⁴ and record the chiro-optical response of type II dehydroquinase.²⁵⁵ Another example of a chiral metasurface sensor demonstrated the use of a 3D core-shell nanohelix structure, constructed via ion beam deposition of a Pt core, surrounded by the conductive polymer P-oPD, to achieve detection of DNA binding protein 43 (relevant for neurodegenerative disease) at a concentration of 10 fM.²⁵⁶ 

The ability to confine light to incredibly small volumes make metasurfaces ideal for increasing the efficiency of vibrational spectroscopy techniques. Many techniques can benefit from these properties, with Infrared Absorption Spectroscopy, Surface-Enhanced Infrared Absorption (SEIRA), and Terahertz spectroscopy, having benefited from metasurface enhancement (using IR resonating microstructures).^257,258 However, the most common use of metasurface-enhanced vibrational spectroscopy is seen in Surface Enhanced Raman Spectroscopy (SERS).^259,260 When light passes through a molecule with certain attributes, it may undergo Raman scattering; inelastic scattering that produces a fingerprint of molecular vibrations from that molecule, elucidating chemical composition and structure. However, Raman scattering is inefficient; large concentrations are required to obtain signals. SERS amplifies scattering from molecules adsorbed on nanostructured surfaces, enabling detection of individual molecules and events. Since SERS enhancement scales with electric-field, meta-atoms are engineered to exhibit sharp edges or small gaps to concentrate the electric-field. Sensing requires the excitation laser line crossover with the (typically broad) meta-atom plasmon, making surface tuning simple, even with less-controlled techniques (e.g., colloid-mediated lithography). Indeed, the vast majority of SERS work is performed using cheap, effective, metal colloid. However, commercial and research examples of metasurface SERS systems do exist when signal reproducibility or more sophisticated functionality are required. One such example used coupled gold triangular trimer structures designed to exhibit multiple resonance modes spanning the visible and mid-IR.²⁶¹ With tip separation of a few nanometers, the trimers promoted large field in these volumes. The authors demonstrated that these multi-resonant structures were multi-functional, and could be used as SERS substrates for detecting DNA hybridization, and as surface enhanced fluorescence and SEIRA substrates.²⁶¹ DNA hybridization has also been detected via SERS using metasurfaces consisting of polarization-sensitive silver split-ring resonators^262,263 and by using the hybridization events to drive individual nanoparticle localization into the center of gold bowtie structures.²³⁹ Combining SERS with chiral nanostructures has led to chiral discrimination of amino-acids via SERS.²⁶³ Recent work on dielectric metasurfaces has shown that, much like for refractometric sensing, these surfaces may hold advantages over plasmonics for SERS biosensing. While plasmonic meta-atoms can require high laser intensities, leading to heating and molecular damage, dielectric meta-atoms exhibit little absorption, minimizing heat, while retaining nanoscale light confinement. However, although there is activity in dielectric SERS,²⁶⁴ and some examples of dielectric SERS metasurface substrates do exist (e.g., silicon dimers to detect β-carotenal molecules)²⁶⁵ the enhancement factors that can be achieved in Si systems are significantly lower than those seen in plasmonic systems (10³ for Si vs $10 6 – 10 1 4$ in metal systems),²⁶⁰ meaning that their impact in biosensing may be limited.

Surfaces that exhibit structured “negative” features, such as engineered holes and voids, also make for effective biosensors.²⁶⁶ Extraordinary Optical Transmission (EOT) through plasmonic nanoapertures is a phenomenon where light passes through holes in metal films at efficiencies higher than classical theories predict. A combined effect of propagating and localized surface plasmon resonances, EOT acts to concentrate and funnel light through the nanoapertures, significantly enhancing the transmitted light intensity, in doing so generating enhanced fields within the aperture itself. Thus, if molecules can be located inside these holes, refractometric and enhanced spectroscopy sensors can be created. Gold nanohole arrays have been shown to be highly sensitive to antibody binding events,²⁶⁷ capable of responding to PSA in a 96-well plate compatible metasurface format,²⁶⁸ used for detecting numerous exosomes for cancer diagnostics (measurements that can also monitor treatment efficacy),^269,270 for detecting pancreatic cancer,²⁷¹ and for identifying virus-like particles and assessing viricidal drug candidates.²⁷² Hole structures also bring the possibility to create nanofluidic flow-through systems, where the apertures serve dual roles as both optical modulators and liquid delivery channels. Examples of these systems include circular gold apertures for label free DNA detection,²⁷³ single base DNA discrimination using gold nanopores and electro-plasmonic trapping,²⁷⁴ and bowtie shaped gold pores to detect DNA²⁷⁵ and beta-amylase protein translocation.²⁷⁶ 

In all the above examples, recognition of the target and retention in the sensing volume has been achieved using a specific binding partner or interaction. For example, complementary base pairing of DNA, or a specifically chosen antibody or affinity protein for the target. Such approaches are well established and many methods of engineering metasurfaces with suitable chemistries attached have been achieved. However, there are also drawbacks to such an approach that are now becoming more widely recognized.

The first consideration is one of multiplexing and the number of targets that need to be detected. In the paradigm considered above, assuming perfect receptor or sensing element specificity, “n” functional sensing elements are required for n different targets. As the understanding and appreciation of multi-biomarker solutions for disease detection or biological systems measurement grows, particularly with advances in “omic” technologies, n can approach a very large number, rendering it impractical or impossible to build such a large number of sensors.²⁸⁰ Second, again led by the increased amount of -omic data available, there is an understanding that not all biomarkers or target molecules are actually known at the outset. These “unknown-unknowns” might enable greater insight or more accurate diagnosis but simply have not yet been linked to the condition in hand. Finally, there is a question of just how specific a sensor can be, particularly when targeting small molecules or macromolecules with few features to bind.

Thus, the idea of cross-reactive arrays is now gaining traction. Such arrays contain multiple sensing elements, sensitive to multiple or many different targets, and, rather than generating a single target response, generate a pattern or fingerprint when exposed to a sample. The sensing fingerprint is then tied back to the sample contents by statistical analysis and pattern recognition. This cross-reactive approach breaks the n for n paradigm and enables a holistic approach to sample measurement where all biomarkers in the sample have a chance to be measured, not just those known about at the outset.²⁸¹ The sensor targeting chemistry can be tuned to be more specific or more cross-reactive as the user desires, and a “training and testing” model is used to gather representative data that informs a statistical model or machine learning for further “unknown” sample analysis.²⁸² 

Incorporating this approach with metasurfaces allows optically addressable arrays to be created where changes in absorbance, reflectance, optical helicity, or other outputs are transduced to create a fingerprint, with spatial or wavelength based multiplexing possible for each individual part of the sensor array.²²⁷ 

There are many examples of the method using plasmonic nanoparticles in solution, deposited inside well-plates, or drop-cast onto surfaces as discrete sensing areas that display differential reactivity directly with target compounds or biomarkers to alter their plasmonic color.²⁸³ For example, Hormozi-Nezhad and co-workers have made extensive studies in this area, differentiating pesticides,²⁸⁴ measuring urinary biomarkers for neurology,²⁸⁵ and also amyloid proteins among many other bio-targets.²⁸⁶ Biogenic amines as markers of food spoilage can also be detected in this manner,²⁸⁷ as well as the different bacteria themselves.²⁷⁸ In each case, the growth, etching, and aggregation of metallic nanomaterials that leads to their color change, is influenced by the different reactive chemistry included in each part of the array (typically a well in a 96 well plate), and the chemical interactions with the sample under test. It is clear that a reactive plasmonics approach such as this could be further integrated into arrays of plasmonic materials as part of a metasurface, however, this would likely have to be a single-use sensor, so may have limited utility in practical applications.

A second approach is to functionalize the metasurface elements with different cross-reactive chemistries, less specific than an antibody but still imparting a degree of selective binding to nearby biomolecules. We have already exemplified this approach with cross-reactive metasurface arrays applied to the sensing of small molecules in whiskey and water.^277,288 In each case, a sensor was produced that consisted of an array of plasmonic metasurfaces (many separate metasurfaces on the same substrate) that were each modified with a self-assembled monolayer (SAM) of a different small molecule, imparting varying local charge, acidity, nucleophilicity and hydrophobicity to each metasurface. The metasurfaces were then exposed as an ensemble to different samples, reacting with molecules in solution to alter the local dielectric environment of each metasurface differently, creating a varying plasmonic fingerprint based on color shift across the array. These fingerprints were analyzed with principal component analysis (PCA) and linear discriminant analysis (LDA) to create tools capable of either discriminating whiskeys based on their chemical content, or treated and untreated water samples based on their mineral and organic carbon content. It is clear how such an approach can now be extended to direct biosensing, and is an area of ongoing research.

In a similar approach, but with a different transduction method, Stevens et al. demonstrated a gold nanopillar metamaterial, where each region (sensing element) again featured a different SAM.²⁷⁹ The metamaterial created strong SERS hotspots between the elements, and the interaction of the SAM with the targets (small biomolecules, extracellular vesicles, cell lysates) ensured different molecules could enter the hotspots, driving differential SERS spectra across the array. Two companion cell lines (one cancerous, one not) were easily identified based on their differential SERS spectra across select array elements, coupled with PCA and LDA analysis.

The coupling of this approach with the exquisite sensitivity of metasurfaces, their greatly improved multiplexing capacities, and advanced fabrication methods should make metasurface sensing arrays a hugely powerful tool in bioanalaysis of the future, capable of operating in both liquid and gas sensing modes to deliver pattern recognition of a wide variety of targets. Key to this success will be the choice of interfacial cross-reactive chemistries to deliver the best interactions with the analytes under study, as well as exploitation of novel transduction modes possible with advanced metasurfaces, to increase the data density that can be collected. As more and more metasurface-based tools become portable, thanks to their compatibility with simple optical platforms and even smartphone read outs, there is huge potential here for delivering ultra-sensitive biomolecule detection with specific or cross-reactive sensing, at the point of need.

George Perrakis^*, Anna C. Tasolamprou, and Maria Kafesaki

^*[email protected]

From a thermodynamic point of view, exploiting a $∼ 3$ K heat sink (ultra-cold universe) of infinite heat capacity, earth objects can reach temperatures below the ambient air temperature (⁠ $∼ 300$ K). This is the principle behind passive radiative cooling (PRC). Specifically, PRC technology aims to exploit the coldness of the Universe for cooling purposes without any electricity input required.²⁸⁹ It relies on Earth's atmosphere transparency between 8 and 13 μm. This transparency window coincides with the peak of the black-body thermal radiation spectrum of earth objects at typical temperatures (⁠ $∼ 300$ K). Thus, any earth object facing the sky can release heat to the outer space, in the form of thermal electromagnetic (EM) radiation, lowering its temperature.

Such a passive strategy (i.e., that cools without any electricity input) could significantly impact global energy consumption (15% decrease).²⁹⁰ The use of PRC as a passive cooling mechanism in human infrastructure elements like buildings, for instance, may lead to lower indoor temperature and alleviate the need for excessive air conditioning, with potential for energy savings and improved efficiency.²⁹⁰ Many other devices and technologies, such as solar cells, thermal textiles, and thermophotovoltaics, can highly benefit from PRC.^291,292

Despite the practical benefits and high potential impact, PRC has three main limitations.^293–298 First, peak cooling demand occurs during daytime, when incident sunlight heats the objects, Fig. 9(a). Therefore, one will need a structure that reflects the entire solar spectrum, Figs. 9(b) and 9(d).²⁹⁶ Second, PRC technology may be counterproductive in year-round assessments, due to provision of undesirable cooling during cold days. Therefore, adaptive PRC may be required, i.e., based on dynamic modulation techniques to tune thermal emission depending on the ambient air temperature.²⁹⁷ Third, current PRC technology is highly unstable, strongly affected by climate or varying weather conditions (e.g., humidity or clouds, resulting to reduction of atmospheric transparency) and by surrounding constructions [e.g., high-rise buildings, providing additional heat or obstructing the heat release, Fig. 9(c)], causing drop of the radiative cooling power. These limitations are a direct consequence of the symmetry of absorption-emission met in most thermal emitters, owing to the fact that PRC configurations are made of reciprocal materials. Specifically, reciprocal materials, characterized by symmetric permittivity and permeability tensors, satisfy Kirchhoff's law of thermal radiation, which states that their spectral, directional emissivity (ε) equals their spectral, directional absorptivity (α). To this end, the two main existing strategies to stabilize PRC (i.e., avoid fluctuations in the radiative cooling power) are (i) high spectral selectivity [maximizing the thermal radiation in the atmospheric transmission window and keeping it minimum everywhere else, Figs. 9(b) and 9(d)]²⁹⁹ to mitigate the sensitivity to humidity, and (ii) enhanced directionality, also to a limited solid angle (beaming), to avoid additional heat gains from the surrounds, Fig. 9(c).³⁰⁰ 

Conventional approaches to meet these stringent requirements, however, come with large space demands, often requiring complex shelter and shading systems, and therefore limit applicability, often reducing also the emission power.²⁹³ For real objects in local thermal equilibrium, Kirchhoff's law of thermal radiation still holds at every combinatorial specific set of wavelength (λ), angle (θ), and polarization (p).³⁰⁰ The field of photonics though, which explores the use of light molding techniques in the visible and infrared, assisted by the recent advances in nano-fabrication, is capable of engineering ε at selected λ, θ, and p, i.e., using the concepts of photonic crystals (PCs), metasurfaces, and metamaterials. Metasurfaces and metamaterials, for instance, involve structuring the material at a subwavelength scale. Relying on local and non-local resonant phenomena (as a result of proper structuring, i.e., of suitably designed building-blocks, known as meta-atoms), metamaterials and metasurfaces provide unique optical properties and enable novel ways to tailor light. The careful engineering of shape, size, and arrangement of meta-atoms can imprint the desired response, such as enhanced directionality,³⁰⁰ or tune the spectral bandwidth of operation,³⁰¹ also offering extra advantages, such as ultra-low thickness and weight,³⁰⁵ and flexibility. Therefore, metamaterials/metasurfaces have been employed as coatings for PRC applications in recent years [see Fig. 9(b)].^305–307 Additionally, to date, considerable attention has been paid to utilizing metasurfaces and metamaterials also to relax the constraints coming from space-reversal symmetry or by reciprocity (connected to time-reversal symmetry) via (i) breaking the time-reversal symmetry with non-reciprocal materials and gratings, e.g., magneto-optic materials, time-modulated systems, or Weyl semimetals,³⁰⁸ or (ii) breaking the space-reversal symmetry of transmission with reciprocal but anisotropic or bi-anisotropic systems, such as metallic gratings with subwavelength slits, gradient metasurfaces, hyperbolic metamaterials, and chiral structures.^309,310 Consequently, it would be of paramount importance to fabricate non-reciprocal thermal emitters (coolers) or asymmetric configurations that would emit or allow heat in certain directions without absorbing or allowing incoming heat.^293,294 Currently and in recent years, there is significant research on examining different photonic strategies in terms of applicability, feasibility, and potential of enhancing and stabilizing PRC. In this perspective, we provide an overview of the recent and ongoing research on metasurfaces, metamaterials, and PCs as means to control the fundamental aspects of thermal radiation for cooling purposes. Such aspects are related to (i) spectral selectivity, (ii) tunability, (iii) directionality, and (iv) asymmetric propagation. Finally, we discuss the challenges and potential future directions.

Spectral selectivity. As mentioned already, spectral selectivity is a quite important feature of a radiative cooling system, as it can allow emission/absorption only in the beneficial for cooling $∼ 8 – 13$ μm spectral window (atmospheric transparency window) and reject the non-beneficial parts of the EM spectrum, Figs. 9(b) and 9(d). In 2015, Hossain et al.³⁰¹ fabricated an anisotropic, conical-shaped waveguide-metamaterial emitter composed of 14 alternating layers of aluminum (Al) and germanium (Ge) (size $∼ 2$ μm, i.e., thickness and bottom diameter) to realize polarization-insensitive, highly selective PRC, Fig. 10(a). They demonstrated ultra-high emissivity (>0.85) and broadband operation at $∼ 8 – 13$ μm, owing to the slow-light modes at different wavelengths (with peaks' emissivity >0.9) and the tapered shape. They measured a remarkably high cooling power, of 116.6 W/m², at ambient air temperature and a temperature of $12.2 °$C below ambient.

Apparently, fabrication feasibility, scalability, and cost-effectiveness are also vital for promoting PRC as a viable energy technology. With this in mind, Zhai et al.,³⁰² two years later, demonstrated a highly transparent hybrid metamaterial cooler composed of resonant polar dielectric (silica) microspheres (⁠ $∼ 4$ μm diameter) randomly distributed in a polymeric matrix (50-μm-thick), Fig. 10(b). By accessing the high-order Fröhlich resonances of the polar silica spheres, the metamaterial provided larger than 0.93 and broadband infrared emissivity across the atmospheric window. When backed with a silver coating to reflect solar radiation, the metamaterial showed a noon-time radiative cooling power of 93 W/m² under direct sunshine. The high significance of this work is related to the high-throughput and economical roll-to-roll manufacturing of the metamaterial combined with the high PRC performance.

Although the impact of PRC on emitters' surface temperature was early highlighted, its impact on cooling inner (enclosed) spaces (that trap heat by the greenhouse effect) remained unclear until 2020. In 2020, Heo et al.²⁹⁹ presented a Janus thermal emitter for cooling enclosed spaces, consisting of an Ag-coated quartz grating and a polydimethylsiloxane (PDMS) layer on quartz substrate's front and rear side, Fig. 10(c). The Janus emitter acted as selective emitter on the top side, to dissipate heat to outer space, and as broadband emitter on the bottom side, to draw heat from the inner space, Fig. 10(c). The selective emitter showed a sub-ambient cooling of $∼ 6 °$C, obtained by exploiting spoof surface plasmon polaritons (sSPPs) to achieve near-ideal selectivity. Notably, the temperature of an object inside the enclosure was $∼ 4 °$C lower than that achieved when using a conventional radiative cooler with reflective bottom surface. These results demonstrate that metasurfaces and metamaterials can provide high-efficiency PRC (⁠ $4 − 12 °$C) even paired with high-throughput, economical, and scalable manufacturing, promoting PRC as a viable energy technology.

Tunability. As mentioned also earlier, the ability to tune a cooler's thermal emission depending on the environmental conditions allows for an adaptive PRC system whose cooling power switches off when not needed, e.g., in cold nights or winter times, avoiding overcooling that increases the heating cost. In 2018, Ono et al.³¹¹ proposed an one-dimensional (1D)-PC consisting of two components, a spectrally reflective filter and a switchable radiative cooler. The on-top spectrally selective filter was a 11-layer stack of alternating Ge/MgF₂ layers, which served as a reflector of solar radiation and a passband filter with high transmissivity at $8 – 13$ μm. The radiative cooler at the bottom consisted of a VO₂/MgF₂/W tri-layer, where VO₂ (material with phase transition hysteresis) was the switching component. The angle and polarization averaged emissivity (⁠ $0 ° – 90 °$⁠) in the wavelength range of $8 – 13$ μm was 0.64 in the metallic state of VO₂. On the contrary, the averaged emissivity dropped to 0.05 when VO₂ was in insulating phase, leading to a 0.59 modulation performance. At “on” state, the radiative cooler reached an equilibrium temperature around 10 °C below ambient. At low ambient temperatures (⁠ $≤ 25 °$C, when cooling is not needed) the system switched off, accompanied by a sudden reduction of the cooling power, maintaining cooler's temperature near the critical temperature (⁠ $∼ 25 °$C). The same year (2018), Wu et al.³⁰³ proposed a phase-change metasurface consisting of periodic VO₂/SiO₂/VO₂ cavities supporting a thermally switchable Fabry–Pérot (FP)-like fundamental resonance mode at $8 – 13$ μm, Fig. 10(e). The thermal emissivity there switched from $∼ 0.2$ to $∼ 0.65$ (⁠ $∼ 0.45$⁠) with a critical (VO₂ phase-change) temperature $∼ 68 °$C. The cooling power was calculated 118 and 528 W/m² at device temperatures 67 and $69 °$C, respectively, compared to 1.3 and 187 W/m² for a simple 300-nm-thick VO₂ film. In 2020, Zhang et al.³¹² proposed a trapezoidal hyperbolic metamaterial (HMM) emitter composed of Ge and VO₂ for self-adaptive PRC and a MgF₂/Ge multilayer filter placed on top for reflecting solar radiation and allowing selective transparency (at $8 – 13$ μm). When VO₂ was in metallic phase, close-to-unity, broadband (>4 μm), and angle-insensitive emission was achieved due to the slow-light waveguide mode in the trapezoidal HMM, while with the presence of the filter on the top, system's emissivity peaked (>0.95) only at three wavelengths (8.4, 9.7, and 11.8 μm), creating spectral-selective emission (at $8 – 13$ μm). The thermal emissivity modulation performance was $∼ 0.67$ (from $∼ 0.20$ to 0.87), leading to a cooling power >100 W/m² (up to 127.8 W/m²) when cooler's temperature was $69 °$C, and 27.9 W/m² at $68 °$C. Considering scalability, flexibility, and low cost, Tang et al.,²⁹⁷ in 2021, fabricated a phase-change metamaterial consisting of a 2D array of thin W_xV_1−xO₂ blocks embedded in a BaF₂ dielectric layer on an Ag film, in various design configurations, for production in a roll-to-roll fashion, Fig. 10(d). When W_xV_1−xO₂ was in the insulating phase, the metamaterial was mainly transparent to the IR radiation at $8 – 13$ μm and highly emissive when it switched to the metallic state. The thermal emissivity was further amplified by a designed photonic resonance involving adjacent W_xV_1−xO₂ blocks and by the $1 4$-wavelength cavity formed in BaF₂ due to the bottom ultra-thin Ag layer, Fig. 10(d). This flexible, temperature-adaptive metamaterial switched its thermal emissivity from 0.2 to 0.9 (0.7) at $8 – 13$ μm when the surface temperature raised above $∼ 22 °$C. The metamaterial was $2 °$C warmer than two reference (non-adaptive) roof coatings when the ambient temperature was below $∼ 22 °$C.

In addition to modulation of thermal emissivity, the simultaneous modulation also of solar heating (i.e., absorbing/reflecting near- and short-wave-IR radiation and remaining transparent or semi-transparent in the visible spectrum), e.g., in smart and/or thermochromic windows, may further improve the regulation of heat flow to maintain a system's temperature near a desired set point, increasing energy-savings. In 2020, Wang et al.³¹³ proposed a compound metasurface of large and small cross-shape resonators for realizing self-adaptive PRC and solar heating in one system. The core of the self-adaptive response is the utilization of a PVP (pNIPAM) spacer below crosses, which can expand (contract) with heat and contract (expand) with cold, increasing or decreasing crosses' spacing. The absorptivity/emissivity modulation performance in both solar and $8 – 13$ μm spectral regimes was $∼ 0.8$ (from $∼ 0.1$ to $∼ 0.9$⁠) within a narrow bandwidth. The results showed that the compound metasurface can remain cool at $35 °$C and warm at $25 °$C. In 2021, Wang et al.³⁰⁴ fabricated a scalable smart window based on W-doped VO₂ nanoparticles dispersed in a PMMA solution on top of a PMMA spacer for simultaneous thermal emissivity and solar reflectivity modulation, Fig. 10(f). Interestingly, the critical temperature was tuned through W-doping around $27.5 °$C (i.e., much lower than that of pure VO₂, $68 °$C). The stacking formed a FP resonator with weak resonance (at $8 – 13$ μm) at low temperatures and strong FP resonation effect at higher temperature (due to the insulator-to-metal transition of VO₂), enhancing emissivity. The emissivity at low temperatures was 0.21 and at high temperatures 0.61 (0.4 modulation performance), combined with near-IR solar modulation of 9.3%, due to the change of the transmittance of VO₂ with temperature, Fig. 10(f). The proposed smart window yielded higher energy-saving performance than a commercial low-ε window, up to 324.6 MJ/m², revealing the importance of integrating PRC modulation in smart and/or thermochromic windows.

Stabilizing PRC. In 2018, Wong et al.²⁹⁴ proposed an asymmetric EM transmission structure (ATS) placed on top of a thermal emitter/cooler to reduce its dependence on humidity and clouds that lower its radiative cooling power, Fig. 10(g). The proposed ATS, consisting of a tapered metallic grating, exhibited high-contrast asymmetric transmission at $8 – 13$ μm (⁠ $∼ 0.8$ forward and $∼ 0.4$ backward, ratio $∼ 2$⁠), Fig. 10(g). Therefore, the ATS, on one hand, permits outgoing transmission of thermal radiation emitted by the radiative cooler, and, on the other hand, it reflects incoming radiation from the sky that lies within the same wavelengths [see also Fig. 10(h)]. Theoretical predictions under humid, semi-transparent sky revealed 57% (50 W/m²) recovery of cooling power, translating to a radiative cooler's temperature $8 °$C lower than the standalone radiative cooler (without ATS). In this case, however, the metallic grating is made of silver, Fig. 10(g), which is a reciprocal material, meaning the directional absorption must equal the directional emission. Therefore, the amount of recovered cooling power should be reconsidered.³¹⁴ Two years later, Wei et al.²⁹⁵ proposed an ATS placed on top of a thermal emitter/cooler to mitigate the combined effects of weather (humidity and clouds) and terrain [e.g., obstruction effect of high-rise buildings, Fig. 9(c)] on its radiative cooling power, Fig. 10(h). Using a glass imprinted with properly designed micro-structures [Fig. 10(h)], a high forward-backward transmissivity contrast ratio at $8 – 13$ μm was achieved (⁠ $∼ 0.7$ forward and $∼ 0.1$ backward, ratio $∼ 7$⁠). Theoretical predictions revealed that the radiative cooler integrated with the ATS has an insensitive (no matter the conditions and terrain) equilibrium temperature $∼ 10 °$C below ambient, compared to $≤ 2 °$C of the standalone thermal emitter/cooler. As in the previous case, though, the equilibrium temperature should be reconsidered.³¹⁴ 

In 2023, Cho et al.³⁰⁰ reported directional, polarization-insensitive thermal emission, which facilitates PRC from the sidewalls of buildings [Fig. 10(i)], see also purple arrows in Fig. 9(c). They fabricated hexagonally arrayed holes (12-μm pitch, 5.5-μm depth, and 10-μm diameter) realized by sub micrometer-thick SiO₂/AlO_x double shells (100/100 nm), resulting to a hollow cavity film, Fig. 10(i). The cavity film exhibited average emissivity values of 0.51–0.62 at angles (θ) $60 ° – 75 °$ and $0.29 – 0.32$ at $5 ° – 20 °$⁠, yielding a parabolic antenna-shaped distribution. The angular selectivity peaked at four different wavelengths at $8 – 13$ μm (8, 9.1, 10.9, and 12 μm), identified as Berreman modes (only in p-polarization at epsilon-near-zero (ENZ) wavelengths of SiO₂) and photon-tunneling modes (at maximum negative permittivity wavelengths of AlO_x), creating broadband (at $∼ 8 – 13$ μm), angle-selective emission bands, resulting to superior directional PRC compared to an isotropic black body-like thermal emitter (up to $4 °$C). We note that Wang et al.³¹⁵ proposed the same year that directional thermal emission can also be realized in a broader spectrum beyond the previously considered ENZ and Berreman mode region. A two-phase metamaterial emitter composed of only two materials (material 1 containing subwavelength particles of material 2 on top of a perfect electrical conductor) was proposed showing numerically strong (ε > 0.8) directional (80  $°$ ± 5°) broadband thermal emission, where the bandwidth of directional thermal emission can be controlled through the methods of gradient ENZ and effective medium theory. These results suggest that by utilizing metasurfaces and metamaterials, PRC can be less affected by the environmental conditions and the surroundings, which promise all-weather and all-terrain applications.

Regarding non-reciprocal emitters, where Kirchhoff's law of thermal radiation does not hold, i.e., emissivity is not equal to absorptivity, their impact on PRC is not extensively examined in the literature. Two recent studies in 2023, by Liu et al.³¹⁶ and Shayegan et al.,³¹⁷ demonstrated direct observation of Kirchhoff's law violation under a moderate external magnetic field (<1.5 T). They reported inequality between emissivity and absorptivity over a broad MIR band, matching also $8 – 13$ μm.³¹⁷ In Liu et al. case, they used a photonic design that supports a guided-mode resonance coupled to a magneto-optic material³¹⁶ while in the Shayegan et al. case, they employed magnetized gradient epsilon-near-zero thin films.³¹⁷ 

Using concepts coming from metasurfaces, metamaterials, and PCs research, recent studies have demonstrated tunable (>0.4) and high-efficiency PRC (⁠ $∼ 4 − 12 °$C below ambient) during daytime under direct sunlight, even combined with high-throughput, economical, and scalable manufacturing, Fig. 10. Here, we emphasize on flexible, roll-to-roll processing technologies that earned sufficient attention as a future contender on the PRC industry manufacturing roadmap [see Figs. 10(b) and 10(d)].^297,302 In this case, radiative coolers' reliability should be further examined in relation to integrated polymers' degradation (thermally- or light-induced due to ultraviolet (UV) radiation) or to metal reflectors' oxidation by oxygen and moisture. Employing chemical additives, coatings, or hybrid organic-inorganic ultra-thin PRC multilayers³⁰⁶ may improve polymers' outdoor performance, although thin polymer films with extended outdoor lifetimes are already available.³⁰² Moreover, a way to avoid the metal layers or more intricate 1D multilayers (commonly used to reflect sunlight) and still strongly reflect sunlight (including UV radiation) is by integrating layers with light-scattering pores on the nano- and micro-scale.³¹⁸ Porous structures' durability, though, affected by dust or moisture, should be further examined. Alternatively, to avoid porosity and efficiently reflect solar light, paints based on a binder (e.g., polymeric) with various fillers and high-index pigments to scatter light (such as TiO₂, SiO₂, Al₂O₃, and BaSO₄ nano- and micro-particles) could be utilized,³¹⁸ i.e., “complementary” to porous structures. Finally, minimizing non-radiative heat gains from conduction and convention [wind, Fig. 9(b)] should also be addressed for efficient PRC below ambient air temperature (i.e., besides avoiding the radiative heat gains).²⁹⁶ This could be achieved by tuning metamaterial's thermal conductivity, which can be achieved by changing its chemistry or micro-structure.^319,320

In addition to the above, great effort has been conducted to stabilize PRC subject to unideal conditions such as clouds, humid subtropical climate, and high-rise buildings [Fig. 9(c)].^{294,295,300,315} Despite the demonstrated superior directional PRC (up to $4 o$ C compared to an isotropic black body-like thermal emitter), enhanced selectivity and directionality come with reduced sky access and emission power [Figs. 9(b)–9(d)],²⁹⁶ limiting heat dissipation to outer space and, therefore, cooling power. Moreover, integrating ATSs composed of reciprocal materials to break the symmetry of transmission as means to stabilize PRC [Figs. 10(g) and 10(h)]^294,295 is still under consideration.³¹⁴ Theoretical studies proved that asymmetric transmission is achievable using reciprocal materials but without increasing the overall radiative cooling power.³¹⁴ Therefore, it would be interesting to experimentally examine the applicability and potential of this photonic strategy on stabilizing PRC compared to conventional approaches for enhancing directionality and selectivity, given also the plethora of photonic designs for asymmetric transmission proposed in the literature (e.g., gratings, arrays, and chiral metamaterials) for various applications, even at $8 – 13$ μm.³²¹ We note that, according to recent studies,^293–295 asymmetric transmission and PRC cannot be enabled in the same component or material (e.g., one grating or metamaterial), which may increase PRC system's complexity, Fig. 10(h).

A promising route to stabilize PRC, maintaining high cooling power enabled by a single component, is to avoid the reciprocity constraints with non-reciprocal materials.^316,317 Maximal violation of detailed balance in thermal radiation (absorbed equals emitted) currently faces several challenges, as it requires a temporal or magnetically induced modulation.³²² The design, simulation, and optimization of such approaches, though, are inherently complex while the involved phenomena are typically resonant.³²² Consequently, novel solutions to achieve high-contrast asymmetric propagation with broadband operation in MIR are required.³¹⁶ Moreover, the impact of non-reciprocal thermal emission on PRC, to our knowledge, has not been experimentally demonstrated yet, also due to limited fabrication feasibility, scalability, and increased cost in standard lithographic techniques. Toward this direction, emerging low-cost, robust, and scalable approaches could be employed, such as two-photon polymerization (2PP), 3D/4D-printing, and laser-induced periodic surface structures (LIPSS) with ultrashort pulsed lasers, for nano-structuring a wide range of materials (metals, semiconductors, dielectrics, and polymers). Decoupling α and ε can lead to novel functions, ranging from reducing losses from solar radiation re-emission in solar energy harvesting systems (occurring due to reciprocity) to radiative camouflage and optimum PRC.^292,317

G.P., A.C.T., and M.K. acknowledge support by the Hellenic Foundation for Research and Innovation (HFRI) under “Sub-action 2 for Funding Projects in Leading-Edge Sectors—RRFQ: Basic Research Financing (Horizontal support for all Sciences),” Project ID 15117 (MultiCool).

Anastasiia Zaleska, Wayne Dickson, David Richards, and Anatoly V. Zayats

[email protected]

Chemical reactions facilitated by light are a cornerstone of clean energy conversion, environmental remediation and material synthesis. While photochemical transformations can take place when reactants are directly illuminated with light, the use of various material systems, often nanostructured, as photocatalysts, can enable a broader class of chemical reactions and can offer enhanced efficiency and selectivity in light-driven processes. Nanomaterials, particularly nanoparticles, play a crucial role in photocatalytic applications by providing large surface areas for molecular adsorption and unique electronic properties, enabling efficient light absorption and facilitating controlled photochemical transformations. The size, shape, and composition of nanomaterials and their surface chemistry can be engineered to optimize photocatalytic activity and selectivity, thereby enhancing the overall efficiency of light-driven reactions. Both semiconductor and metallic nanostructures can be used for photocatalysis, exploiting strong light absorption in the designed spectral range, resulting in electron–hole pair excitation through interband transitions in semiconductors and metals, or the generation of energetic hot electrons using surface plasmon resonances in metallic nanostructures. An additional side-effect of such electronic excitations is the rise of the local temperature of the material which influences the reactivity through thermal effects, as in conventional thermal catalysis. Overall, the integration of nanostructures with photochemistry and photocatalysis holds significant promise for advancing sustainable technological solutions.

In this respect, metamaterials and metasurfaces provide a highly flexible platform for engineering enhanced light–matter interactions which control electric field localization and its polarization as well as local temperature, while providing large surface areas important for chemistry.^323,324 The planar geometry of metasurfaces also facilitates their integration into photo-electrochemical systems, while the use of metaparticles—nanoparticles engineered using metamaterial design principles³²⁵—provides the combination of increased surface area and engineered optical response with the advantage of scaleable fabrication afforded by nanoparticle synthesis. Furthermore, the opportunity to engineer these nanoparticles as a coating on powders is preferred for some photocatalytic applications.

Many realizations of metallic (plasmonic) and semiconductor metasurfaces and metamaterials have been recently developed and investigated for enhancing and controlling photocatalytic and photo-electrocatalytic transformations. While plasmon resonances have been widely employed in metallic catalysts, which provide enhanced light absorption and hot-carrier excitation not limited by a bandgap, semiconductor, and dielectric metastructures are also actively considered for enhancing light absorption at sub-bandgap energies and engineering strong absorption in catalytically inert dielectrics decorated with metal or semiconductor catalysts. These often exploit archetypical metasurface designs based on nonradiative high-index nanophotonics, such as multipolar Mie resonances and bound states in the continuum.

Photocatalysts based on semiconductor materials, such as titanium dioxide (TiO₂) and gallium phosphide (GaP) are traditionally attracting significant attention because of the possibility to utilize solar energy directly.^326,327 They generally benefit from high chemical stability and reactivity, are nontoxic and inexpensive. Holes in the valence band and electrons in the conduction band, excited upon interband light excitation, can then interact with molecules on the surface promoting chemical transformations. However, optical absorption in semiconductor materials is limited by their bandgap. TiO₂ has a relatively large bandgap of about 3.2 eV, which makes this catalyst predominantly active under UV light, in the wavelength range shorter than 400 nm, compared with a bandgap of around 2.2 eV for GaP. This means TiO₂ photocatalysts can harvest only a few percent of the solar spectrum energy (about 3%–5%). Additionally, the photocatalytic activity of semiconductor materials is limited by the rapid recombination of photogenerated electron–hole pairs and insufficient charge carrier transfer. Extensive research efforts have been dedicated toward the improvement of their photocatalytic performance through material modification and incorporation or doping with other functional materials for visible light utilization.

Metasurfaces patterned on TiO₂ and GaP have been designed to increase the absorption of light and broaden their working spectral range.^328–331 For TiO₂ metasurfaces, doping can be first used to introduce defect states near the bandgap edges, and the associated absorption is then increased at selected wavelengths of visible light by engineering nonradiative modes via the interplay between electric and magnetic dipoles of the individual nanostructures forming the metasurface (meta-atoms), lattice modes, or the effects of bound states in the continuum. While the defect related excitation of electron–hole pairs may have its drawbacks, primarily in lower carrier mobility, the enhanced absorption at the resonances in the visible spectral range can be sufficient to demonstrate photocatalytic transformations at wavelengths where reactivity would be negligible without the metasurface. The impact of patterning also increases the surface area available for photocatalytic activity, which may be the dominant effect, as was demonstrated for NO degradation, when a 20-nm-thick TiO₂ film was conformally coated onto a photonic crystal, with some contribution also from the waveguided mode excitation enabled by the photonic crystal.³³² This approach can be applied to various types of semiconductor material systems by varying the active materials, leading to substantial improvements in air/water contamination, water splitting or artificial photosynthetic processes (Fig. 11).

The engineered enhanced light absorption and planar geometries of metasurfaces enable their use as a photo-electrode material,³³³ demonstrating, in the case of a GaP-nanodisk-based metasurface, an overall photocurrent enhancement by more than fivefold compared to a planar GaP film under hydrogen evolution reaction conditions [Fig. 11(c)].³³¹ Nanoimprint lithography techniques and electrochemical approaches enable the rapid development of large-area nanopatterned surfaces of various materials necessary for industrial applications as a new generation of highly efficient solar meta-electrodes.

Dielectric (catalytically inactive) metasurfaces decorated with metal catalytic nanoparticles can be employed as an antenna-reactor photocatalyst where the virtually lossless metasurface funnels light to efficiently drive a chemical reaction.³³⁴ By combining a Si₃N₄ metasurface exhibiting strong quasi-bound states in the continuum (quasi-BIC) resonances and large electric field enhancements, with Ni nanoparticles as reactors, this hybrid metasurface-based catalytic system was demonstrated to efficiently drive H₂ dissociation under the resonant illumination [Fig. 11(d)]. The experimental and theoretical analysis suggest that both photothermal heating and electronic transitions at the quasi-BIC-Ni surface facilitate catalytic chemistry (the thermal effect also cannot be ruled out in the above-discussed semiconductor metasurfaces when defect-related transitions are considered). This approach, based on engineered dielectric metasurfaces combined with otherwise weak absorbing metal catalysts, provides a universal platform which can be exploited for various combinations of dielectrics and catalytic metals.

In contrast to dielectric and semiconductor materials, plasmonic nanostructures provide the unique possibility to engineer light absorption throughout visible and near-infrared spectral ranges with spectral selectivity as well as strong electromagnetic field enhancement. Surface plasmons excited in metal nanostructures by light absorption decay either radiatively through re-emission of photons or non-radiatively via, e.g., Landau damping on a femtosecond timescale, resulting in hot carriers with energies comparable to the energy of the exciting photons.³³⁷ These highly energetic hot carriers generated from the non-radiative decay of surface plasmons can either directly interact with molecular adsorbates on the metal surface or be transferred into adjacent semiconductor or metal co-catalysts efficiently inducing chemical transformations.

Plasmonic hot-electrons can be harvested in metal-semiconductor Schottky barriers, by combining the semiconductor catalyst with plasmonic nanostructures to achieve efficient electron–hole pair separation to prevent their direct recombination, thus increasing the lifetime of the hot-carriers and prolonging their opportunity to interact with adsorbed molecules. One drawback of metal–semiconductor heterostructure catalysts is the low charge mobility from the Schottky contacts. Many designs use a combination of plasmonic metasurfaces on or in a catalytic semiconductor layer simply to increase the local temperature and enhance thermal catalysis.

Alternatively, by incorporating strong plasmonic effects from metallic (Au, Cu, Al, etc.) nanostructures with highly catalytic active materials, core-shell or decorated nanostructures can be employed for various types of reactions. Plasmonic metasurfaces decorated with transition metals (Pd, Pt) can enable efficient coupling of light energy into a catalyst through strong electromagnetic field enhancement, hot-carrier generation and photothermal effects. Such metaparticles (or hetero-nanoparticles) have demonstrated a significant increase in photocatalytic activity over monometallic particles by efficient transfer of hot carriers to the catalytic surface for various reactions.^338–341 

As an alternative to conventional plasmonic metals, titanium nitride (TiN) has started attracting significant attention for photocatalysis. TiN metasurfaces can be designed to exhibit broadband optical absorption in the visible range (i.e., an average of more than 92% in the 400–750 nm spectral range).³³⁶ When coated with a polymeric photocatalyst, plasmon-enhanced hydrogen production from water under visible-light illumination was achieved, with the hydrogen evolution rate increased by 300% compared to a smooth TiN film [Fig. 11(f)]. The increased efficiency is attributed to a combination of enhanced light absorption, carrier separation, hot carrier transfer and thermal effects induced by the plasmonic metasurface. Refractory plasmonic TiN metasurfaces were also used for driving heterogeneous photothermal catalytic reactions. Self-assembled TiN cylindrical nanocavities act as plasmonic “nanofurnaces” capable of reaching temperatures above 600 °C under moderately concentrated solar irradiation (∼20 Suns). Upon decoration with catalytic Rh nanoparticles, the obtained photothermal metasurface achieved a high rate of CO₂ production under solar-spectrum-simulated light intensities³⁴² as well as in the reverse water gas shift reaction.³⁴³ 

Metasurfaces based on arrays of strongly coupled bimetallic core–shell nanoparticles were also exploited to enhance photo-electrocatalytic activity for hydrogen evolution reactions.³³⁵ Large-area Cu–Pt nanoparticle lattices fabricated by combining top-down lithography and solution-based chemistry support two different types of plasmon modes, localized surface plasmons from individual particles and surface lattice resonances (SLRs) from the 2D lattice, that increased the catalytic activity under white-light illumination up to 60% [Fig. 11(e)].

In addition to optical excitation by light illumination, hot electrons can also be generated electrically by excitation with tunneling electrons in tunnel junctions.³⁴⁴ Recent realization of an electrically driven plasmonic nanorod metamaterial provides the opportunity to use an electron tunneling effect for the simultaneous excitation of hot electrons and surface plasmons, providing the means to realize a new kind of hot-electron-activated nanoreactor. By constructing a high-density array of plasmonic tunnel junctions at the top surface of a plasmonic metamaterial composed of vertically oriented gold nanorods, the efficient electrical excitation of the plasmonic modes of the metamaterial by inelastic electron tunneling can be realized. During the tunneling process, the majority of electrons (∼99.9%) tunnel elastically in the electrically driven nanorod metamaterials, appearing as energetic hot electrons in the nanorod tips. The highly efficient and confined hot-electron generation makes the tunnel junctions highly reactive and opens up opportunities for the precise activation of chemical reactions in the junctions, which can be further detected with high sensitivity by observing the light emission from the tunnel junction or by observing changes in the tunneling current due to the extreme sensitivity of this highly confined tunneling process to any changes in the junction.

Photonic metasurfaces have emerged as a promising platform for driving photocatalytic reactions. To develop a highly efficient photocatalyst with a photonic metasurface, several key design concepts need to be considered (Fig. 12). First of all, scalability of the fabrication method. For commercial exploitation, high durability, low unit cost, and efficient use of materials are undoubtedly the most appealing criteria for economic viability and sustainable chemical/energy production. Most of the aforementioned metasurfaces were fabricated using EBL or lithographic techniques which are relatively expensive fabrication processes and do not offer a large catalytic surface area. Nanoimprint lithography and self-assembled techniques can be used instead to reduce the fabrication cost.

To justify the use of metasurfaces, it will be advantageous to move from expensive noble metals and employ sustainable earth-abundant materials. For example, plasmonic metasurfaces made of transition metal nitrides and phosphides, CuS, MXenes, copper, and aluminum nanostructures need to be further investigated as a platform for engineering light absorption for photocatalysis and photo-electrocatalysis. The quality of semiconductor and plasmonic materials is of utmost importance. For plasmonic photocatalysts, different applications in photocatalytic metadevices may favor either rough or monocrystalline and ultrasmooth interfaces depending on the requirements of hot electron transfer to either molecules or semiconductor co-catalysts.³⁴⁵ 

In this regard, geometries and constituent materials can be optimized by using numerical simulations to tailor the desired engineering optical response for reactions that require specific energies. Furthermore, advanced ab initio numerical methods could help deepen understanding of reaction mechanisms under specific light illumination conditions (intensity, light polarization) and optimize optical properties to create a database of photonic metasurfaces suitable for photocatalysis. Electron and thermal effects of the metasurface illumination under both solar and pulsed laser illumination, as well as hot-electron transfer processes to reactants, need to be optimized. Machine learning and deep neural networks (DNNs) have recently been demonstrated to be powerful tools for designing efficient light-harvesting metasurfaces due to their extraordinary capability to find solutions from a large data set or parameter space.³⁴⁶ 

Product selectivity is a crucial aspect of any photocatalytic reaction that remains an ongoing concern and challenge. By engineering the synergetic effects of electron transfer and temperature, as well as the choice of the materials for adsorption, there is an opportunity to explore the prescribed modification of the energy reaction landscape to steer reactions toward the desired products. Many tests of the catalytic properties of plasmonic metastructures are performed under pulsed laser illumination which favors plasmonically derived hot-electron generation, and the peculiarities of CW illumination need to be taken into account. Selectivity in the synthesis of chiral molecules is the utmost challenge for pharmaceutical industries, and chiral metasurfaces may play an important role in steering the chiral reactions under circularly polarized light illumination.

Significant attention has been paid recently to the control of the chemical reactions through optical cavity modes, addressing specific vibrational modes of the molecules to influence selectivity.³⁴⁷ These studies have been performed until now in optical cavities which select the spectrum of the electromagnetic modes available for strong-coupling to vibronic modes of the reactants. Metasurfaces and metaparticles may play an important role in these photochemical transformations as they can be engineered to provide high quality-factor vibrational modes for coupling to molecular vibrational states.³⁴⁸ 

Metasurfaces and metamaterials providing precisely designed light–matter interactions offer an exquisite platform for understanding the mechanisms of light-induced chemical transformations, hence revealing routes to optimization. Through targeting either electron-induced and/or thermal effects via the design of the composition and structure of metamaterials, understanding of the role and interplay between electronic and thermal contributions in the reaction can be achieved, and their relative weight can be adjusted to control reaction pathways. Beyond fundamental research, metasurfaces can be readily applied in synthesis of high-value chemicals where small quantities of products, cost and femtosecond laser illumination are not prohibitive issues. Similarly, they provide an excellent platform for multiplexed functionalities, such as sensing and photochemical transformations, useful for environmental applications and pollution remediation.

In order to advance metamaterial-based catalysis concepts into mainstream photocatalysis, the problem of scalable production must be addressed together with a reduction of the associated costs, which can be achieved by moving away from top-down approaches toward self-assembly, electrochemistry, laser printing, or nanoimprint lithography for roll-to-roll fabrication. The planar nature of metasurfaces and metamaterials is advantageous for photo-electrochemical reactors and generally for photochemistry, where light-management issues are important considerations. Beyond these, the transfer of the light-harvesting properties designed for metasurfaces into a metaparticle platform holds much promise for the application of these designs in a scaleable manner compatible with chemical industry standards.

Harnessing metasurfaces and metamaterials as photocatalysts has emerged in the catalysis community because of their ability to harvest sustainable solar energy to promote chemical reactions. It is a comparatively new area of photonics and metamaterial research that is growing rapidly and showing promising results. Due to the ongoing climate and energy crisis, much research effort is currently focused on finding catalysts to combat environmental problems such as CO₂ reduction, NOx decomposition, and ammonia production to name but a few, and to provide renewable and clean energy with sustainable input through water splitting, etc. Metasurfaces, metamaterials and metaparticles have an important role to play in achieving net-zero pollution.

This work was supported by the UK EPSRC under Project No. EP/W017075/1.

Haoran Ren, Yuri Kivshar, and Stefan Maier^*

^*[email protected]

Structuring light in multiple degrees of freedom, from spatial to temporal, holds great promise for advancing modern photonics from both fundamental and applied aspects. In particular, structured-light fields have proven useful for numerous photonic applications, including super-resolution microscopy,³⁴⁹ optical trapping,³⁵⁰ multimode imaging,^351,352 and holography,^32,208,353 optical communications,^354,355 nonlinear light conversion,³⁵⁶ chiral sensing,^357,358 and quantum information processing.^359–361 Conventional generation of structured light typically relies on multiple cascaded phase and wave plates in bulky form, imposing major challenges for their practical applications. Over the past decade, flat optics, also known as ultrathin metasurfaces,⁵ have become a new topic in the photonics community. Recently, such metasurfaces have been largely developed for the generation, manipulation, and detection of structured light. Here we summarize the key physical concepts demonstrated for generation of various kinds of structured light with metasurfaces. Apart from the generation, we highlight some exciting applications of metasurface manipulation of structured light, including but not limited to metasurface lasers with tailored spatial modes, metafibres capable of transforming arbitrary vector beams, vectorial metasurfaces that can create optical skyrmions, high-bandwidth twisted light holography, vortex metalens for edge enhancement, and metasurface-structured quantum emission. We also discuss some of the major challenges that lie ahead for the large deployment of structured-light metasurfaces. Finally, we provide some insights into recent new advances in metaphotonics that could potentially mitigate these challenges and push the field forward by creating more research impact.

In meta-optics, both plasmonic antennas and dielectric pillars have been used for the metasurface generation of different structured light fields in real space. For instance, a helical phase profile that carries the orbital angular momentum (OAM) has been imprinted by a phase-only metasurface, with the use of a set of meta-atoms that offer a complete phase modulation (from 0 to $2 π$⁠). Ultrathin plasmonic metasurfaces based on near-field mode hybridization have been designed to implement the OAM phase [Fig. 13(A-a)].⁵ High-index dielectric nanopillars with strong mode confinement, operating as truncated waveguides, have also been used to realize different phase vortices [Fig. 13(A-b)].⁸⁵ Huygens' metasurfaces of high transmission efficiency offer an alternative platform to realize phase vortices through spectrally overlapping electric and magnetic Mie resonances [Fig. 13(A-c)].^362,363 Apart from the use of phase-sensitive elements for creating phase vortices, plasmonic and dielectric nanostructures can be designed to possess strong polarization birefringence, exhibiting different phase accumulations for the polarization along the long and short axes [Fig. 13(B)].^16,20,364 Geometric metasurfaces based on the Pancharatnam–Berry phase principle have been used to create polarization-encoded phase vortices through the in-plane rotation of an identical asymmetric nanostructure. Meta-atoms with both varying orientation and size have been further developed to control arbitrary orthogonal polarizations in an independent manner.³⁶⁵ As such, the output phase response for orthogonal polarization incidence can be decoupled from each other, leading to the generalized spin-to-OAM conversion of light that creates different OAM modes at orthogonal polarization outputs.³⁶⁶ Such metasurfaces are designed by exploiting the complete and independent phase and polarization control by single meta-atoms based on the Jones calculus, named as J-plate metasurfaces [Fig. 13(C)]. Notably, J-plate metasurfaces pave the way of using a single metasurface for creating arbitrary structured light on the hybrid-order Poincaré sphere.³⁶⁷ 

Unlike traditional refractive or diffractive optics constrained from a limited access to multi-dimensional wavefront shaping, meta-optics has offered an unprecedented opportunity for structuring light fields in multiple degrees of freedom, from transverse to longitudinal (propagation direction) planes, as well as from spatial to temporal domains. Propagation-invariant or non-diffracting light beams, such as caustic beams whose transverse intensity distribution remains invariant over a significant propagation distance, exhibit high robustness and a self-healing behavior.³⁶⁸ This has inspired metasurface generation of multiple phase vortices³⁶⁹ as well as on-demand polarization transformations³⁷⁰ along the optical path, which may inspire new directions in singular optics toward 3D space. In addition, temporal shaping of light, especially in the ultrafast domain, has benefited wide-ranging photonic applications.^50,371 Spatio-temporal metasurfaces have thus been created by combining a frequency comb source with a passive metasurface, capable of steering optical beams with a wide steering angle in just a few picoseconds.⁵⁰ In addition, metasurface-enabled pulse shapers have been demonstrated by embedding a dielectric metasurface in the focal plane of a Fourier-transform pulse shaping setup.³⁷¹ Such a Fourier-transform pulse shaper has recently allowed the generation of spatiotemporal optical vortices carrying previously unexplored transverse OAM modes.^372,373

Over the last decade, structured-light metasurfaces have found many applications in the optics-related fields. Here we present only a few highlights, including but not limited to intracavity metasurface lasers, metafibers harnessing structured-light fields, generation of optical skyrmions in free space, metasurface holography using twisted light modes, vortex metalenses for edge detection, and the manipulation of quantum emission. Incorporating structured-light metasurfaces in a solid-state or fiber laser cavity has enabled coherent light emission with tailored spatial mode profile, such as high-purity vortices³⁷⁴ and vortex arrays,³⁷⁵ in a compact form factor [Fig. 13(D)]. This paves the way for the development of the next generation of miniaturized laser sources with tailored spatiotemporal mode control. The current use of structured light in optical fiber science and technology is limited by mode mixing, and hence generation of structured light is most usually handled outside the fiber via bulky optics in free space. A new metafiber platform has been introduced for implementing ultrathin metasurfaces directly on the end face of optical fibers.^376,377 Polymeric metasurfaces, with unleashed height degree of freedom, were 3D laser nanoprinted and interfaced with polarization-maintaining single-mode fibers. Multiple metasurfaces were interfaced on the fiber end-faces, capable of transforming the fiber output into arbitrary selected structured-light fields on the hybrid-order Poincaré sphere [Fig. 13(E)].³⁷⁷ 

Allowing subwavelength-scale-digitization of optical wavefronts to achieve complete control of light at interfaces, metasurfaces are particularly suited for the realization of planar holograms that promise new applications in high-capacity information technologies. Recent advances in metasurfaces and Fourier holography have offered an unprecedented opportunity to control optical wavefront in the spatial-frequency domain. Metasurface holograms have shown the independent control of both the 3D polarization³⁷⁸ and the OAM states^32,208,353 of individual spatial-frequency components, represented by different image pixels reconstructed from the metasurface holograms in the Fourier plane. It opens the door to the use of a single metasurface for creating optical skyrmions with topologically nontrivial 3D vectorial textures³⁷⁹ [Fig. 13(F)], as well as for reconstructing distinctive images from different OAM states [Fig. 13(G)].^32,208 Meanwhile, structured light has fueled many photonic applications in imaging, microscopy, and optical communications. For instance, the donut-shaped OAM beam in the focal region of a microscopic objective lens can deactivate fluorophores, minimizing the effective area of illumination and leading to super-resolution imaging. In addition, a spiral phase metasurface incorporated in a Fourier filtering optical setup can perform a 2D spatial differentiation operation and achieve isotropic edge detection.³⁸⁰ A recent polarization-multiplexed metalens integrated with a polarization camera has allowed single-shot edge detection of indoor and outdoor scenes under ambient sunlight illumination [Fig. 13(H)].³⁸¹ Practical OAM communication systems suffer from turbulence-induced phase distortions to the propagating beams, decreasing the orthogonality of OAM modes through introduced modal crosstalk. An ultrathin OAM mode-sorting metasurface was recently used for characterizing the OAM orthogonality breakdown under different turbulence conditions.³⁸² It allows the measurement of the whole OAM spectrum at the same time, suggesting that metasurfaces with a small form factor can be integrated with practical communication systems for compact, fast, and efficient detection of twisted light in turbulence environments.

Structured photons in the quantum regime promise enhanced security by harnessing structured light as high-dimensional quantum states. An ultrathin polarization-beam-splitting metalens was demonstrated for arbitrary structuring quantum emission at room temperature [Fig. 13(I)].³⁵⁹ The quantum metalens can collect the quantum emission from ultra-bright defects in hexagonal boron nitride (hBN), as well as enable simultaneous manipulation of multiple degrees of freedom of a quantum light source, including directionality, polarization, and the OAM of light. Dielectric metasurfaces can also be used for the manipulation and even direct generation of entangled photon pairs.³⁸³ A geometric metasurface was first used in a spontaneous parametric downconversion setup for the nonlocal spin and OAM correlations on entangled biphoton states.³⁶¹ In this context, quantum state engineering mainly relies on the nonlinear optical effects, including spontaneous parametric downconversion and four-wave mixing, where one or two pump photons spontaneously decay into a photon pair. These nonlinear effects typically require the use of bulky nonlinear crystals by matching the momentum of the participating photons. Nonlinear metasurfaces have subwavelength thickness and allow the relaxation of this constraint,³⁴ opening new possibilities of quantum state engineering. Recently, metasurface generation and manipulation of entangled photons via spontaneous parametric downconversion have been demonstrated by high-quality-factor nonlinear metasurfaces made of lithium niobate¹³⁷ and gallium arsenide³⁸⁴ materials. These results suggest that metasurfaces can offer a versatile platform for creating complex photon quantum states toward high-dimensional photonic quantum processing.

Structured light, with many millions of spatial modes in a tiny cross section area of light, has shown great potential in transforming the landscapes of a wide range of photonic applications, from classical to quantum regimes. Meanwhile, metasurfaces have provided scientists and engineers with versatile tools to generate, manipulate, and sculpt optical fields. Although numerous advancements have been made in both the structured light and metasurface fields, there remain certain challenges in the science and technical aspects of structured-light metasurfaces. Even though metasurfaces benefit from the use of subwavelength meta-atoms to digitize optical wavefronts, the accuracy and efficiency of their amplitude, phase, and polarization modulations still need to be improved. For instance, for a phase-type metasurface, increasing the number of gray levels is important to reach near-unity diffraction efficiency, but this is indeed challenging for metasurfaces that rely on size-varying meta-atoms to realize the phase modulation. Recent developments in 3D laser-nanoprinted metasurfaces,^{208,376,377,385} multilayer,³⁸⁶ and cascaded metasurfaces,³⁸⁷ as well as in hybrid metasurfaces that combine both refractive and diffractive optical elements³⁸⁸ may alleviate this major challenge by unlocking 3D design degrees of freedom.

Current structured-light metasurfaces still require external source excitation, making them difficult to be integrated on-chip. Integration of metasurfaces with on-chip light sources, waveguides, and detectors will provide an exciting opportunity for future metasurface systems-on-a-chip. Recently, generation of free-space structured light with programmable integrated photonics was demonstrated.³⁸⁹ More complications of using structured-light fields include strong modal coupling and intermodal crosstalk, due to free-space turbulence, large beam divergence, and misalignment in the detection process. Despite a great success has been made in the metasurface generation and manipulation of structured-light fields, metasurface-based selective detection of complex spatial modes, for instance, arbitrary selected states on the hybrid-order Poincaré sphere, is still elusive today. Nonlinear metasurfaces have been developed to observe high-order nonlinear optical spin–orbit interactions,³⁹⁰ as well as non-reciprocal asymmetric generation of visible images.³⁹¹ Owing to the use of subwavelength meta-atoms, nonlinear metasurfaces hold tremendous promise for enhancing the toolkit for structured light–matter interactions beyond the limits of linear optics. While major challenges still exist on both science and technical aspects of structured-light metasurfaces, we believe these challenges are rather likely the driving forces to inspire more innovations and push the field forward.

The authors acknowledge financial support from the Australian Research Council DECRA Fellowship DE220101085 (H.R.) and Australian Research Council Discovery Project DP220102152 (H.R., S.A.M.). S.A.M. additionally acknowledges the Lee-Lucas Chair in Physics.

Xianzhong Chen^* and Muhammad Afnan Ansari

^*[email protected]

Optical holography has been used for several decades for reconstructing 3D images by shaping the wavefront of a light beam. Unlike multiview parallax and photography techniques, a hologram consists of complete information of an object. The recording in the holographic technique is not an image itself, rather it contains random patterns of varying phase or intensity.³⁹² The holograms can be reconstructed either by the interference of a reference light beam with a scattered light beam from a real target or by calculating the phase profile of the light beam at the interface of a hologram. The phase profile can be encoded into the spatial light modulators or surface structures using lithography. The process involving the computer-generated phase information is known as computer-generated holography (CGH).³⁹³ In the traditional holography and diffraction optics, the light field is regarded as a scalar field with two parameters, i.e., intensity and phase of the light beam.³⁹⁴ With the ongoing advancements in nanofabrication techniques, the last decade has witnessed the rapid development of ultrathin and flat optical devices with unique degrees of freedom. One degree of freedom is the independent control of polarization distribution in the engineered light path, which remains the cornerstone for multifunctional holography, e.g., vectorial holography.³⁹⁵ The traditional bulky optics were also used to a certain extent to introduce the concept of polarization holography for two-channel orthogonal polarization control using birefringent materials.³⁹⁶ These polarization-dependent computer-generated holograms are sensitive to the polarization state of the incident light beam and capable of separating readout light through its polarization to generate distinct holographic images. This led to various novel applications such as multi-optical switching and image processing.^397,398 However, the conventional holographic recording medium and diffraction optics suffered from a limited degree of freedom and were unable to achieve simultaneous control of polarization state and spatial light field distributions. The form and magnitude of such type of birefringence were limited to naturally occurring crystals. On the other hand, materials with structures smaller than working wavelengths provided an alternative method to enhance the controllability of birefringence.^399,400 The drawback of materials with subwavelength structures was their extreme anisotropic requirement with very high aspect ratios.

To tackle the challenges in conventional optics, metasurfaces have provided more degrees of freedom and better performance, particularly manifesting polarization-dependent physical channel multiplexing.^401–405 Initial demonstration of polarization multiplexed holograms based on plasmonic metasurfaces suffered from poor quality and low efficiency.^5,406 Wen et al. proposed and experimentally demonstrated highly efficient helicity multiplexed metahologram with good image quality in the near-infrared and visible range.⁴⁰⁷ In this work, the issues of poor image quality and low efficiency were resolved by leveraging the high efficiency and broadband property of reflective metasurfaces.²⁰ Two off-axis holographic images were reconstructed under the illumination of a circularly polarized (CP) light beam. The images were symmetrically distributed and interchangeable by controlling the helicity of the CP light beam as shown in Fig. 14(a).⁴⁰⁷ The right panel of Fig. 14(a) shows the change of holographic images (bee and flower) when the polarization of the input light beam is changed from left circular polarization (LCP) to right circular polarization (RCP). The helicity multiplexed property for CP light was attained by interleaving two metasurfaces comprised of plasmonic nanorods. Polarization-sensitive metaholograms for input light beams with elliptical, linear, and arbitrary orthogonal polarizations have also been reported in the literature.³⁶⁵ Helicity multiplexed metaholograms have attracted various exciting applications in optical field manipulation such as stimuli-responsive (e.g., pressure, voltage, temperature, and gas) and tunable metaholograms.^408–410 For example, a volatile gas sensor based on helicity multiplexed metahologram integrated with liquid crystals was presented to sense isopropyl alcohol gas by generating a visual alarm.⁴⁰⁸ In this work, the response of gas was recorded with different geometries of the designer liquid crystal (nematic and isotropic) by integration with an asymmetric metasurface. The proposed non-interleaved asymmetric metasurface^411–413 employed both geometric and propagation phase for the phase modulation. The overall phase retardation was controlled by controlling the cell order of liquid crystals through the external stimulus, i.e., the presence of gas. The tunable phase retardation adjustment of output light beam with desired polarization state using asymmetric metahologram and liquid crystals have opened new avenues to detect chemical and biomedical/biological substances in real-time for public health and environmental monitoring.^408,409 Recent studies on optical polarization multiplexing have further demonstrated the ability of metasurfaces to realize independent amplitude and phase modulations with non-orthogonal polarizations, which further expanded the ability of metaholograms for the full vectorial holography.^407,414,415

Furthermore, holography has been widely used to achieve beam shaping,⁴¹⁶ surface plasmon holographic displays,⁴¹⁷ data storage,⁴¹⁸ optical trapping,⁴¹⁹ diffractive laser tweezers,⁴²⁰ and digital holographic microscopy.⁴²¹ Although two-dimensional (2D) holograms were experimentally reported using metamaterials and metasurfaces, none of them have achieved 3D image reconstruction using CGH in the visible domain. To fulfill the essence of holography in 3D space, Huang et al. presented a relatively simpler approach to realizing 3D holography using ultrathin metasurface consisting of subwavelength metallic resonators with varying orientation as shown in Fig. 14(b).³⁹² The phase information of the 3D object was calculated using commercially available CGH algorithm.⁴²² In the CGH algorithm, the 3D target object [Fig. 14(c)] was considered as a collection of many point sources. The complex amplitude at the hologram interface was numerically calculated through the superposition of the wavefronts of all point sources. Upon the illumination of the CP light beam, the metasurface reconstructed a real 3D holographic image within the Fresnel range of hologram as shown in Figs. 14(b) and 14(d). As explained earlier, the incorporation of spatially varying polarization along with the phase distribution is essential to fully unlock the potential of metasurfaces for increased information storage. Yue et al. proposed such a technique to encrypt a 2D grayscale image (of James Clerk Maxwell) in a structured laser beam with inhomogeneous polarization distribution using metasurface based on Malus' law as shown in Fig. 14(e).⁴²³ The result of the intensity profile of the encrypted grayscale image was revealed with an analyzer (linear polarizer) as shown in Fig. 14(f). To further explain the technique, a selected portion from the eyebrow area (10 × 10 pixels²) is enlarged in Fig. 14(g) to show the intensity and its corresponding pixel-level polarization distribution. It is noted that the resultant beam exhibited a subwavelength resolution. Similarly, metasurfaces can also be used to encode colorful images in inhomogeneous polarization profiles based on Malus's law. Zang et al. experimentally demonstrated simultaneous encryption of intensity and color information into a wavelength-sensitive polarization distribution.⁴²⁴ This technique was used to embed high-resolution color images with precisely controlled contrast and brightness onto a dielectric metasurface as shown in Figs. 14(h)–14(j). The target high-resolution color image of a rose was selected which encompassed green and red pixels with the variation of spatial brightness as shown in the top panel of Fig. 14(i). The corresponding wavelength-dependent polarization distributions were different for both colors and mapped by controlling the orientation and the feature size of the dielectric nanopillars as shown in the bottom panel of Fig. 14(i). Vivid red petals and green leaves were revealed with the right transmission axis of the analyzer under the illumination of 660 and 550 nm wavelengths. The same method was also used to produce polarization-encrypted colorful images using additive color mixing of the primary colors as shown in Fig. 14(j). In this work, a supercell contains nanopillars with different feature sizes, which respond to different wavelengths.⁴²⁴ By integrating such high-resolution polarization encryption techniques with holography, many diverse applications can be realized in information security, anticounterfeiting, and high-density data storage. For example, an application of optical information security was experimentally demonstrated by integrating an arbitrary polarization distribution for image concealment and a hybrid CGH as shown in Fig. 14(k).⁴²⁵ The functionality was achieved by simultaneously concealing a high-resolution image in the polarization distribution and encoding holographic images through the superposition of two-phase profiles for orthogonal CP light beams. As a result, a metasurface was used to reconstruct holographic images at the observation plane either under the illumination of linearly polarized light beam without the analyzer or under the illumination of RCP light beam as shown in Fig. 14(m). The desired concealed image was revealed with a predesigned angle combination of the analyzer and input linear polarizer as shown in Fig. 14(l). In contrast, the other transmission axis combination of the analyzer and input linear polarizer could not generate meaningful information. The work also investigated the simultaneous incorporation of polarization-sensitive and insensitive holograms with high-resolution image encryption in the polarization distribution, which further enhanced the information capacity and hence can be utilized for high-level anti-counterfeiting.

2D high-resolution image encryption based on spatially varying polarization distributions has been realized with metasurfaces.^423–425 However, in addition to the phase, intensity, and color, the encryption of spatially varying polarization distribution in a 3D space remains a challenge. To generate the desired wavefront and dynamically manipulate the polarization distribution in a 3D space, an independent pixel-level control of phase and polarization is simultaneously required along the light's propagation. The biggest challenge to achieve such functionality is the limited degree of freedom in the current metasurface design. Although few studies have shown unprecedented abilities to break the conventional limits to some extent,^{415,427–429} they are still unable to address the complete and dynamic control of light fields in a 3D space. Therefore, a solution is required to address this problem at the fundamental level where new control variables can be incorporated into the metasurface design procedure. The required control variables for an arbitrary 3D polarization structure include wavelength, wavefront positioning (x, y, z), and dynamic polarization rotation in a 3D space. Much work has been done to address one control parameter at a time,^{407,412,424,426,430} however, simultaneous control of all parameters is scarce and challenging. The combinational control of more than one parameter is also expected to yield unique optical field manipulation. Recently, Zang et al. proposed a metalens with polarization functionality through the superposition of both RCP and LCP components of the light beam at each focal spot under the illumination of linearly polarized incident light beam.⁴³¹ The authors started from a simple lens model by incorporating an additional term for the polarization rotation $ϕ$ in the phase profiles of LCP and RCP components. Since a linearly polarized incident light beam can be divided into LCP and RCP light components, therefore, a portion of the transmitted wave is converted to LCP light to focus with an additional phase shift $− ϕ$⁠, and the other portion is converted to RCP light with an additional shift of $ϕ$⁠. The idea was then extended to realize multiple focal points using the superposition method with independent polarization rotations as shown in Fig. 15(a). Wang et al. presented an off-axis multi-foci metalens model to incorporate polarization rotations in four different focal points at different locations on the observation plane as shown in Fig. 15(b).⁴³² Unlike the previous work where only linear polarization rotations are created, Wang et al. introduced RCP, LCP, and linear polarization rotations for the application of polarimetry to detect the polarization of arbitrary incident light beams. The ellipticity, handedness, and major axis of the polarization state of the incident light beam were determined through the measured intensities of the four focal points.⁴³² Both designs have provided a unique insight into a pixel-level simultaneous control of wavefront along with the independent polarization rotations. More specifically, both unique ideas of polarization rotation at different focal lengths (f1 and f2) along the z-direction⁴³¹ and the off-axis (xi, yi) multi-foci metalens⁴³² have led to the generation of 3D polarization structures. An optical focal curve can be regarded as a continuous structure if the number of focal points along the curve becomes infinite. Wang et al. started from a muti-foci metalens with discretized focal points to a continuous optical curve along an arbitrary trajectory by increasing the number of focal points.⁴³³ In the next step, the 2D polarization curve model was transformed to 3D polarization knot design by introducing the longitudinal parameter (z) as an additional control variable in the phase terms using the superposition method. As a result, 3D light fields along with spatially varying inhomogeneous polarization distributions and intensity patterns were generated in the form of 3D optical knots as shown in Fig. 15(c).⁴³³ The predesigned polarization distribution was confirmed through the modulated intensity patterns with an analyzer (linear polarizer). In addition to the predefined inhomogeneous polarization distribution, the polarization rotation at each point on the 3D trajectory was dynamically modulated by changing the polarization state of the incident light beam. The complete evolution of the 3D structure was also observed and measured using a CCD camera as shown in Fig. 15(d). Intaravanne et al. introduced a new control variable (wavelength λ) in the design procedure along with polarization rotation $ϕ$ and 3D focal position (u, v, f) to demonstrate wavelength selective 3D polarization structures as shown in Fig. 15(e).^434,435 At a given observation plane, different 3D polarization knots were generated by changing the wavelength of the incident light beam as shown in the left and right panels of Fig. 15(e). The right panel and left panels present the intensity distribution of corresponding 3D polarization structures with and without the analyzer, respectively.

The ability to generate arbitrary wavelength-dependent 3D polarization structures with ultrathin metasurfaces can be of interest to many pragmatic applications such as integration optics for compact spectrometers, and longitudinally variable structures for 2D and 3D optical image steganography. For example, motivated by the intrinsic dispersion property and ability of metalens to generate multiple focal points at will, the wavelength of the incident light beam can be detected accurately by mapping the wavelength information to the intensity distribution. A compact spectrometer was realized by tailoring the dispersion based on the model of multi-foci metalens as shown in Fig. 15(f).⁴³⁶ The metasurface-based spectrometer can focus and split the monochromatic [Fig. 15(g)] and polychromatic light beams [Fig. 15(h)] of different wavelengths to desired positions on an observation plane with 1 nm resolution. This unique dispersion control and high nanometer spectral resolution were achieved over a broadband range in the visible domain. The application of the proposed multi-foci metalens-based spectrometer proved that this technology has clear potential for development in on-chip integrated photonics where wavelength detection, information processing, and spectral analysis are required in a compact platform. Like other emerging technologies, further improvements in the current design are required and new challenges need the attention of the research community. For example, the demonstrated metasurfaces have a low conversion efficiency, which is associated with the symmetric spin–orbit interactions and can be resolved by employing the nanoresonators with asymmetric spin–orbit interactions.^411–413 Furthermore, the polarization manipulation can be extended to incorporate arbitrary polarization states, including linear, elliptical, and circular polarization states. The integration of metasurfaces with materials such as 2D transition metal dichalcogenides,⁴³⁷ conducting oxides,⁴³⁸ liquid crystals,^408,409,439 and active materials²⁰⁹ can add the extra degree of freedom and perform complicated optical tasks that are impossible or extremely challenging only with passive metasurface design.

Yuhui Gan, Arseny Alexeev, Thomas F. Krauss, and Andrea Di Falco^*

^*[email protected]

Augmented reality (AR) superimposes computer-generated content onto a transparent medium such that the user can perceive the real world augmented with the artificial content; this contrasts with virtual reality (VR), whereby a computer generates all the content. In AR applications, the need for a transparent medium precludes the use of conventional displays. This requirement for transparency poses challenges, but it also promotes widespread acceptance of the technology. It benefits specialized applications such as assisted surgical operations or equipment maintenance, as well as end-user applications such as smartphones.

The main elements of an AR system are the image projector, which creates the digital image, and the optical combiner, which blends digital and real images before delivering the combined picture into the user's eyes.⁴⁴⁰ Depending on the system architecture, the image projector consists of a light source (laser or LED), a spatial light modulator that modulates the light to create an image, e.g., based on liquid crystals or digital micro-mirrors, and image-forming optics such as lenses and mirrors. The optical combiner has been the subject of intense research over the last few years and is the focus of this paper. Combiners started as bulky and irregularly shaped glass elements,⁴⁴⁰ as shown in Fig. 16(a), they then evolved into sophisticated photonic devices, usually involving nanostructures [Figs. 16(b) and 16(d)].

The most common configuration used for optical combiners to date is an optical waveguide that delivers light toward the observer via a Holographic Optical Element (HOE)⁴⁴¹ or a Surface Relief Grating (SRG)⁴⁴² [Fig. 16(b)]. Both HOEs and SRGs are manufactured using nanoscale lithographic techniques. Another type of optical combiner uses metalenses [Fig. 16(c)],⁴⁴³ which are actively being explored for AR applications due to their low weight and their ability to imprint any desired phase distribution onto the incoming wave; therefore, metalenses can replace freeform optical elements due to their advantageous form factor and functionality. We note that several reviews describing AR systems using both waveguide combiners and metalenses are already available,⁴⁴⁴ which this perspective article aims to complement. Here, we first introduce the key parameters and figures of merit of an AR near-eye transparent display and then briefly review the state-of-the-art of various meta-optical technologies that are being used in consumer-grade AR glasses. Finally, we discuss the performance tradeoffs of each approach, including the use of transparent holographic displays based on metasurfaces. The holographic approach eliminates the need for an image projector, thus achieving a minimalistic form factor while delivering superior optical performance, as shown in Fig. 16(d).

AR display systems must meet a variety of performance metrics, which can sometimes conflict one another. For example, the field of view (FoV) describes the range of angles at which optical rays are delivered to the human eye to form an image on the retina. The FoV is critical, as it defines the perceived size of the virtual object generated by the AR glasses. A typical value for heads-on AR displays is around 60° (diagonally)⁴⁴⁴ while, for reference, the human eye's FoV is approximately 120°;⁴⁴⁰ thus, it is desirable to extend the FoV of AR glasses. However, such an extension poses real challenges for the display brightness and resolution, as the properties of the image are determined by the luminous source. If the FoV increases, the luminous energy must be distributed over a larger range of solid angles, so the luminous energy per solid angle will suffer. Similarly, for a larger FoV, the resolution decreases, as there is only a fixed number of pixels available. A large FoV also has an impact on the eyebox, which describes the geometrical region of eye positions that contribute to image formation without vignetting. AR glasses must have a large eyebox to ensure tolerance to eye movements and to accommodate different users. The size of the eyebox is measured in mm in the x and y directions in the plane coinciding with the plane of the eye hole, where a minimum of 6 mm is required to accommodate inter-pupil distance variations of 95% of the human population. High-performance displays achieve >10 mm in both directions. The product of the FoV and the eyebox is comparable to the etendue of an optical system, as it represents the product of an area and an angle; this product is fixed by the image projector. For a given projector, an increase in the FoV directly results in a decrease in the eyebox size. As discussed further below, this conservation of etendue can be overcome by the method of “pupil replication”—which replicates the image source over the surface of the optical combiner, which is why the waveguide approach is so attractive. Finally, images must be created with good color and brightness uniformity, irrespective of FoV and eyebox, so this requirement cannot be traded off.

Freeform Optics—The simplest way to deliver light from a projector to the eye is to reflect the projector signal toward the eye by a suitably shaped, partially reflective mirror, as shown in Fig. 16(a). This approach has the advantage of requiring minimal image pre-processing and results in a relatively simple system architecture while delivering good FoV (90°) and eyebox (>10 mm) as well as producing high-quality images. The main disadvantage of this approach is the need for a bulky, inconvenient headset that imposes an unnatural look on the user due to the shape and weight of the freeform mirror and its tilt. Additionally, this design is fundamentally restricted by the conservation of etendue, as discussed before.

Waveguides—Currently, the most popular implementation of the AR optical combiner function consists of an imaging waveguide with diffractive elements. In a typical arrangement, three linear diffraction gratings are used, (i) to deliver the image rays created by the projector into the first diffracted order below the angle of total internal reflection of the glass-air interface, thus in-coupling image rays into the waveguide; (ii) split the image rays into higher diffraction orders, thereby replicating the image pupil; and (iii) diffract the image rays at multiple points into the user's eye, thus, outcoupling the image rays from the waveguide. The key advantage of this approach is that the waveguide produces copies of the image pupil, thereby surpassing the fundamental trade-off between the FoV and the eyebox imposed by the conservation of etendue, achieving FoV > 60° and eyebox > 10 mm without the need to use bulky free-form optics. The gratings are typically Bragg gratings recorded in holographic media or Surface Relief Gratings—SRGs. The technology of holographic Bragg gratings is very mature but this solution only offers low refractive index contrast. This intrinsically leads to a high selectivity of the diffraction process in terms of angle and wavelength, therefore limiting the FoV of the entire system.⁴⁴⁴ Despite these shortcomings, this approach is well suited for applications that do not require large FoV (for example, displaying notifications or giving instructions for navigation, for automotive applications). SRGs are, in turn, fabricated by nano-structuring waveguide surfaces and possess higher refractive index contrast. Therefore, SRGs support larger FoVs. Such gratings can now be fabricated with high accuracy and high manufacturing yield because of advances in nanofabrication, including the use of nanoimprint lithography [see Fig. 17(a)].⁴⁴⁵ This classical three-element grating scheme was pioneered by Nokia⁴⁴⁰ and then adopted by other companies, including Microsoft, Magic Leap, Vuzix, and others. A more advanced version of the waveguide combiner scheme (e.g., that introduced by WaveOptics uses two-dimensional SRGs to combine the expander and the out-coupler elements in one hybrid unit). In this configuration, a larger area of the combiner is available for the outcoupling, resulting in an increased eyebox of >10 mm in both directions. Realizing these advantages is, however, a challenging task. The grating must not only have minimal dependence of the diffraction efficiency on the wavelength and incident angle, but also balance the contributions from the image rays that are incident from different directions, as specific optical rays can reach a given extraction point via multiple routes and not just through a single geometrical path as in the classical configuration, see Fig. 17(b). In this way, the chromatic and angular aberrations can be minimized to produce a uniform virtual image. These requirements can only be met with the introduction of further degrees of freedom, which has led to the development of 3D metagratings—true nano-photonics objects sculpted into complex 3D shapes.⁴⁴⁶ Each element of such a 3D metagrating acts as a sub-wavelength antenna. The full response of the grating is then defined by the convolution of the interference pattern of the entire grating with the radiation profile of each grating element.⁴⁴⁷ While traditional SRGs are patterned at scales greater than the wavelength, these advanced metagratings require a lithography resolution below 50 nm and high-precision alignment.

Metalenses—Metalenses utilize subwavelength structures to manipulate light at the nanoscale. With their ultra-thin size, sub-wavelength modulation scale, and ability to imprint any desired phase front onto the incoming signal, metalenses can be used to make flat versions of traditional lenses⁴⁴⁸ reducing the size, complexity, and cost of the optical system.^449,450 Remarkably, metalenses can also be engineered to respond differently to light with specific polarization, wavelength, or angle of incidence.^451,452 This flexibility makes them very attractive for AR applications.^453–455 In the example of Fig. 16(c), metalenses are used in the Maxwellian configuration, whereby the image is focused on the pupil and the virtual image is always in focus, independent of the accommodation of the eye lens.⁴⁵⁶ This approach offers advantages in terms of FoV. However, the eyebox for the Maxwellian configuration is very small.⁴⁴⁴ While metalenses offer superior control over the wavefront of the transmitted light, they are characterized by intrinsic tradeoffs.⁴⁵⁷ The main trade-off is the fact that a typical metalens design strongly depends on the wavelength. The choice of thin film thickness, meta-atom size and unit cell, as well as the use of Fresnel zones to accommodate multiple 2π phase shifts, are all wavelength-dependent. Moreover, the unit cell size depends on the numerical aperture (NA) according to the Nyquist criterion as a <λ/(2 NA), which means that the higher NA required for AR systems to achieve reasonably large FoV also demands a smaller unit cell size. This quickly becomes very challenging in terms of lithographic requirements, especially for operation in the visible. All these requirements trade-off against each other; for example, the wavelength dependence can be addressed by controlling the dispersion in each meta-atom by substructuring the unit cell and using two meta-atoms instead of one.⁴⁵⁸ This is an elegant solution which, however, restricts the unit cell size for lithographic reasons, so only a low NA or longer wavelength operation can be achieved. Another example is the control of aberrations, where in particular, designed phase profiles have been used to demonstrate a very large FoV,⁴⁵⁹ but only for a single wavelength.

When it comes to AR applications, metalenses face the same challenge as standard freeform optical elements in terms of balancing a sufficiently large FoV (>60°) with a wide enough eyebox (>10 mm), while possessing a small form-factor for the convenience of the user. However, the main limitation is that due to the design and manufacturing challenges we mention above, modern metalenses are yet to become suitable for image-forming optical systems. As a result, many metalenses are designed for security camera operations that do not require multiple colors and that work best in the near-infrared.

Holographic Metasurfaces—In the scheme shown in Fig. 16(d), the microdisplay projector is replaced by a simple laser or an LED. This choice shifts the physical location of the digital information from the image projector to the optical combiner, which is much simpler, but restricts the system to fixed images stored in the hologram.^35,460 While some degree of image multiplexing is possible, the number of independent images that can be stored in the combiner is then rather limited. A solution would be the deployment of dynamically tuned metasurfaces to create AR glasses capable of displaying dynamic images. Such a dynamic holographic display could be achieved using a variety of material platforms and actuation mechanisms, as discussed in Sec. IX C.

Waveguides and Metalenses—Metalenses and metagratings used in AR optical combiners exhibit noticeable similarities in terms of controlling a given wavefront and producing a desired optical response. Both approaches utilize engineered patterns of sub-wavelength antennas to create a desired functionality. Likewise, both in metalenses and metagratings, the complexity of the meta-atom design is now surpassing initial simplistic designs by shaping or substructuring the meta-atoms. By exploiting all the degrees of freedom of metaoptics design, it is now possible to improve the brightness and color uniformity of AR systems.

While the design complexity of the meta-atoms and the manufacturing of metalenses and metagratings have already merged, these two approaches employ fundamentally different architectures. Given the advanced technology readiness level reached by the waveguide-style combiner, it is anticipated that waveguide combiners will be the choice for the next few generations of commercially available AR systems. The main advantage that motivates this choice is that the waveguide architecture addresses the issue of the FoV—eyebox trade-off, through the pupil replica mechanism. Importantly, from an engineering point of view, in metagratings, it is also possible to manage the extraction efficiency and thus maximize the extracted energy along the propagation path of the image rays. The main challenge for waveguide-based combiners remains to achieve good color and brightness uniformity. These parameters are still limited by chromatic and angular aberrations. Regardless of the coupling mechanism, it is extremely challenging to in-couple and out-couple all the image rays with the same efficiency, independently of their wavelength and incident angle. To this end, the shape of the 3D meta-atoms will require further optimization. Importantly, this development will need to include consideration of the manufacturing at the scale of the metagratings, e.g., using nano-imprinting technologies and using materials with a sufficiently high index of refraction. Current efforts in this direction tend to fabricate gratings with a refractive index greater than 2.5 by imprinting patterns into polymers and using them as an etching mask.⁴⁶¹ This approach can employ a wide range of conventional optical materials but possesses the challenge of accurately translating complex metagrating shapes into dielectric materials through a single-step etch process [see, e.g., Fig. 17(b)]. Alternative approaches include imprinting into a polymeric matrix with added nanoparticles, which have been used to create 3D nano-structures with a refractive index greater than 2^462,463 avoiding the etching step.

While waveguide combiners will likely remain the standard approach, they will improve on their limitations by incorporating some of the benefits of metalenses. Next-generation combiners will likely exploit the polarization-dependent response of meta-atoms, implement pupil replica schemes, be achromatic, and offer a high level of sophistication of response. In the process, they will also tackle other key issues like the unwanted scattering of external light, known as the “rainbow effect,” and the outcoupling of the digital image away from the user's eyes, referred to as “glowing eye.” This will be achieved by designing metagratings that have an asymmetric optical response depending on the direction of the incident light.

While the intrinsic limitations of metalenses make it harder to see them prevailing as combiners in AR glasses, we see potential in their use as AR technology integrated with LiDAR and Time-of-Flight sensors, and as optics for facile recognition applications.⁴⁶⁴ These applications have much less stringent achromatic requirements, as they solely need to operate at a predetermined wavelength and exclusively within the near-infrared (NIR) spectrum. This has the additional benefit of enabling the use of silicon technology, such as nanolithography, which is readily compatible with scaled-up manufacturing processes with high precision in extra small feature size.^465,466

Freeform combiners with metasurfaces—Another possible direction in the development of metasurfaces for AR glasses is their use to enhance freeform optical elements.⁴⁶⁷ This type of hybrid freefrom-meta optics uses both the controlled phase response typical of metamaterials and the geometry of the freeform elements to fix both the image quality and the aberrations. Patterning curved substrates is challenging, but recent results have shown that this is a practically viable solution.⁴⁶⁷ Alternative approaches include the use of conformable metasurfaces, which have the advantage of decoupling the fabrication constraints from the shape of the substrate to which they are applied.^468,469 This manufacturing approach could also be used to create curved waveguide combiners to mimic curved lenses used in most modern spectacles by first manufacturing diffraction gratings on a transparent flexible film and then transferring this film onto a curved waveguide combiner. This type of combiner would be less bulky than conventional freeform combiners and have a more natural look, however, still be constrained by the conservation of the etendue. It is, therefore, more likely that freeform optics enhanced by metasurfaces will find better use in the image projector, working on a broadband range with a wide field of view.

Holographic metasurfaces—Finally, we would like to imagine ultimate AR glasses based on holographic metamaterials. Current solutions only offer static images, at best multiplexed in the same device, using e.g., polarization, wavelength, the direction of the incoming light, and their shape.^470,471 Future solutions would deploy active tuning mechanisms. The most promising solutions include the incorporation of liquid crystal cells coupled to metasurfaces filters,^472,473 metasurfaces based on phase change materials^474,475 and variants of transparent conductive oxide modulators.^476,477 Ideally, the holographic metasurfaces should be tuned electrically, at the level of the single meta-atom, over a sufficiently wide modulation range, and reversibly over a meaningful number of cycles, cycle number still being an issue with phase change materials. Despite the fast and meaningful progress,⁴⁷⁸ at the time of writing this article, it is not clear which, if any, of these approaches will yield the magnitude in change of the optical response required by AR applications, with low dispersion and low loss. If this important issue could be solved, however, such glasses would be lightweight with a form factor barely different from modern spectacles. Their frame would be thin yet contain all the electronics and the light source. The degree of freedom available in the design would also enable the implementation of pupil replica strategies, thus surpassing the etendue constraints. By scanning a laser over various areas of a holographic metasurface and dynamically altering the optical response of specific meta-atoms, it becomes possible to generate wavefronts in any desired shape and direction. This approach could realize the goal of consumer-grade AR glasses that seamlessly integrate digital and real-world visuals.

To conclude, there have been major developments in AR systems over the last few years. Traditional methods have evolved to incorporate more sophisticated strategies to manage the delivery of light to the eye, incorporating nanopatterned optical interfaces. In their evolution, these systems have crossed paths and aligned with the progress made in the various areas of meta-optics, which is indeed one of the most versatile nanophotonic platforms available to optical engineers. Here, we have discussed both the differences and the similarities between the different platforms and practical solutions. While some considerations are unavoidably speculative, and AR systems of the future may look very different from current implementations, meta-optics has already carved a place for itself in this future.

Sylvain D. Gennaro, Tomás Santiago-Cruz, Igal Brener, and Maria V. Chekhova^*

^*[email protected]

In the early 20th century, the understanding of light underwent a significant transformation through the discovery of the photoelectric effect and the development of quantum theory.⁴⁷⁹ Pioneering physicists of the time postulated the existence of quanta of light called photons—a term coined by the American physicist Gilbert N. Lewis in 1926 from the Greek word “phōs,” meaning “light.”⁴⁸⁰ However, it was not until the 1970s that physicists learned to generate single photons and even more fascinating states—pairs of such photons that behave as a single quantum object, and exhibit entanglement. Subsequent advancements enriched the toolbox of quantum optics with other quantum states of light, such as squeezed states, Schrödinger cat states, cluster states, and many others, whose description in terms of classical electromagnetic theory leads to paradoxes. These quantum states of light are now pivotal instruments for quantum photonic technologies, including quantum information processing, communication, metrology, sensing, and imaging.⁴⁸¹ 

Over the years, scientists have developed a plethora of single-photon sources⁴⁸² using atoms,^483,484 trapped ions,⁴⁸⁵ single molecules in solids,⁴⁸⁶ epitaxially grown⁴⁸⁷ and colloidal⁴⁸⁸ quantum dots, carbon nanotubes,⁴⁸⁹ color centers in diamond,⁴⁹⁰ and defects in two-dimensional materials.⁴⁹¹ The most convenient sources of entangled photons are created using nonlinear optical effects: spontaneous parametric downconversion (SPDC) in second-order nonlinear crystals^492–497 and spontaneous four-wave mixing (SFWM) in fibers.^498,499 SPDC and SFWM can also produce single photons if one photon of a pair is used to herald the presence of the other photon. However, the most significant application of these nonlinear effects is to generate and engineer entanglement between the two photons, whether in polarization or continuous variables, such as time/frequency and position/momentum. With stronger pumping, both SPDC and SFWM produce higher fluxes of photon pairs, forming different kinds of squeezed vacuum states. Being brighter than single- or few-photon states, squeezed vacuum and other squeezed states are especially useful for quantum measurement and sensing.

In recent years, optical metasurfaces have been developed to manipulate and control the amplitude, phase, and polarization of light.^{1,13,500–502} The latter manipulation combined with enhanced efficiencies due to engineered resonances has been used to promote the classical process of harmonic generation at the nanoscale.^{53,130,135,503–519} More recently, metasurfaces have evolved into platforms for the generation and manipulation of quantum light. Thanks to their flexibility in engineering high-quality optical resonances, metasurfaces can control and amplify various forms of spontaneous emission—from the emission of single photons by solid-state emitters^{58,100,395,489,520–525} to the generation of photon pairs through nonlinear optical effects.^{137,384,526–529} In addition to their compact nature, metasurfaces are multifunctional, capable of combining the generation of quantum light with tasks like sorting,⁵³⁰ filtering,⁵³¹ polarization transformation,^532,533 sensing,⁵³⁴ and engineering photon entanglement.^535,536 Metasurfaces thus present a promising avenue to generate and control complex quantum states of light at the nanoscale. In this short perspective, we direct our focus toward the unique abilities of optical metasurfaces at enhancing the emission of single photons or photon pairs through the engineering of optical resonances.

One of the fundamental building blocks of optical quantum technologies consists of single-photon sources. A prominent challenge in this field is thus the search for single-photon sources that emit indistinguishable photons with well-defined polarization, high purity, and high emission rates. Many traditional quantum emitters (QEs) typically face issues such as low quantum yields and emission rates, a broad emission spectrum, and omnidirectional emission.⁵²⁵ 

Recently, metasurfaces have facilitated the enhancement of spontaneous emission rates,^{58,100,395,489,520–525} beam steering,^{48,88,537,538} and polarization control of emitters^370,539 using optical resonances. According to the Fermi golden rule, the emission rate of a QE increases with the local electromagnetic density of states in its environment. As demonstrated by Purcell in 1946, the spontaneous emission rate in an optical cavity is proportional to the ratio between the quality (Q) factor of the cavity and the optical mode volume.⁵⁴⁰ Consequently, optical resonances with high quality factors and/or small optical mode volumes are sought after for this purpose. From a quantum optical perspective, effects of spontaneous emission—such as single-photon emission by atoms or solid-state emitters, and the generation of photon pairs through SPDC and SFWM—can be elucidated as being stimulated by the vacuum field. Despite the vacuum field having an average value of zero, it exhibits non-zero fluctuations around this mean. The spectral density of these fluctuations is uniform across all states in the k-space and is equal to one photon per state (per mode).⁴⁷⁹ A resonance augments the density of states for the resonant mode. Given the uniform population of states due to quantum vacuum fluctuations, the resonance intensifies the spectral brightness of these fluctuations at its wavelength and for the corresponding orientation of the optical field. This explains the enhancement of all spontaneous emission effects for resonant modes of optical metasurfaces.

Similar enhancement of emission can be obtained using plasmonic nanoantennas and metasurfaces. Plasmonic metasurfaces harness the power of surface plasmons—collective electron oscillations at interfaces between metals and dielectrics—to amplify single-photon emission. By tailoring light–matter interactions within nanoscale mode volumes, these structures improve the efficiency of QEs. For instance, plasmonic metasurfaces were shown to enhance the spontaneous emission rate of embedded InAs quantum dots, as illustrated in Fig. 18(a).⁵²³ The quantum dot entered a superlinear photoluminescence regime, which was attributed to the excitation of surface plasmon polaritons (SPPs) and the subsequent room-temperature generation of hot electrons. Plasmonic metasurfaces can also enhance the spontaneous emission of isolated defects in 2D materials, such as hexagonal boron nitride (hBN), that are known to emit single photons at room temperature. For instance, Fig. 18(b) depicts a metasurface composed of gold nanoparticle arrays that support localized surface plasmon resonances. The strong confinement of the electric field near the nanoparticle surface intensifies the spontaneous emission, aligning with the Purcell enhancement in the weak coupling regime.⁵²¹ 

Plasmonic metasurfaces, while offering remarkable opportunities for enhancing single-photon generation through their ability to concentrate light in small volumes, face limitations in achieving high-Q resonances. These limitations arise due to the inherent Ohmic loss of plasmonic materials. Researchers are thus actively exploring alternative high refractive index dielectric materials to achieve higher Q-factors. All-dielectric metasurfaces offer a promising route to enhance single-photon generation at visible and near infrared wavelengths through the excitation of Mie-type resonances and high-Q bound states in the continuum (BIC). For instance, as depicted in Fig. 18(c), an early experimental demonstration shows an all-dielectric metasurface, made of arrays of high-index nanoresonators supporting magnetic dipole modes, modifying the single-photon emission of oxygen-doped single-walled carbon nanotubes (SWCNTs), by a moderate factor of $0.8 − 4.0$⁠. Moreover, single-dopant polarization experiments show an anomalous photoluminescence polarization rotation due to coupling individual SWCNTs to silicon nanoresonators, showcasing the exciting potential of dielectric metasurfaces to control the polarization of QEs.⁴⁸⁸ 

Using optical modes with higher Q-factors, single-photon generation can be further enhanced. For instance, consider a metasurface composed of Mie-resonant GaAs nanocylinders embedding five layers of InAs quantum dots [Fig. 18(d) ¹⁰⁰]. Although the primary objective of this study was to modify the photoluminescence polarization and directionality through the outcoupling of quantum dots emission to the magnetic quadrupolar modes, the authors also observed a noteworthy 19-fold enhancement in the emission rate when compared to an unstructured surface. This improvement in the spontaneous emission rate aligns with expectations, given the diminished Q factors associated with the resonances involved.

All-dielectric metasurfaces can capitalize on high-Q optical resonances such as symmetry-protected quasi-BICs.^506–508 BICs are discrete-energy modes that do not radiate and possess energy levels that coincide with a continuous spectrum of radiating modes.¹⁴¹ Leveraging their high-Q resonances can lead to increased spontaneous emission. For instance, as depicted in Fig. 18(e), a 1000-fold enhancement of photoluminescence was observed for four layers of self-assembled Ge quantum dots (QDs) embedded within a metasurface composed of a periodic lattice of asymmetric air holes in a silicon-on-insulator (SOI) slab. This metasurface exhibited quasi-BICs with a Q-factor of approximately 1000.⁵⁸ 

Similarly, a metasurface employing broken-symmetry GaAs resonators, arranged in a square lattice, showcased a several-hundred-fold enhancement in the photoluminescence of monolithically embedded, self-assembled, near-infrared InAs/InGaAs QDs [Fig. 18(f)]. Notably, the symmetry protected quasi-BICs reshaped the emission of the QDs, making it unidirectional along the normal to the metasurface.⁵²² 

Optical metasurfaces can go beyond the mere enhancement of spontaneous emission and tailor its polarization and orbital angular momentum properties. Examples are metasurfaces designed in the configuration of a bullseye target through concentric dielectric rings [Fig. 18(g)]⁵³⁹ or anisotropic nanobricks [Fig. 18(h)],⁵⁴¹ allowing the emission of single photons with controllable polarization. Consider a nitrogen-vacancy (NV) center at the center of concentric dielectric nanoridges positioned above a silver film, as shown in Fig. 18(g). This arrangement emits single photons with well-defined circular polarization (chirality >0.8) and remarkable directionality (collection efficiency reaching up to 92%), a result of the tailored phase of the field scattered from the QE and the corresponding excited surface plasmon polaritons.⁵³⁹ Similarly, Fig. 18(h) illustrates how a metasurface composed of dielectric anisotropic nanobricks, enveloping a room-temperature single-photon emitter—germanium vacancy center (GeV-ND) positioned on a silica (SiO₂) spacer-covered silver (Ag) substrate—excites radially diverging SPP waves. In this manner, metasurfaces can confer topological charges to emitted single photons, thus enabling the emission of linearly polarized vortex beams. The authors additionally demonstrated the realization of multiplexing single-photon emission channels with orthogonal linear polarizations carrying distinct topological charges and demonstrate their entanglement.⁵⁴¹ A further application of this concept involves the construction of a dielectric bifocal metalens atop a SiO₂ and gold film, housing a quantum dot within the SiO₂ layer [Fig. 18(i)]. The metalens is designed with its two foci aligned with the positions of the quantum dot and its mirrored image. Consequently, each photon can be emitted along one of the two distinct directions, depending on its polarization.⁵⁴² 

To conclude, we have surveyed some of the recent work in using metasurfaces to improve the performance of quantum emitters embedded within them, achieved through optical resonance engineering. In addition, metasurfaces can also tailor the emission direction and polarization. Shifting our focus, we will now turn to showcasing how nonlinear metasurfaces can facilitate the generation of photon pairs via SPDC.

Quantum state engineering largely relies on SPDC^{492,493,496,497} and SFWM.^498,499 These processes lead to the spontaneous emission of entangled photon pairs, known as biphotons, while adhering to principles of energy and momentum conservation. SPDC and SFWM serve as the foundation for generating a diverse range of quantum states, spanning from heralded single photons and higher-order Fock states to squeezed states and multiphoton non-Gaussian states like Schrödinger cat states or NOON states.

In contrast to bulk crystals, SPDC at the nanoscale relaxes the requirement for longitudinal momentum conservation,⁵⁴³ resulting in a broad range of frequencies and angles for emitted photon pairs. For instance, the broad frequency-angular spectrum of SPDC in ultrathin nonlinear films [Fig. 19(a)]⁵⁴³ leads to a heightened degree of frequency and angular entanglement, along with short correlation times and subwavelength correlation distances.¹¹⁶ Moreover, the relaxation of momentum conservation offers the potential for tunable polarization entanglement,⁵⁴⁴ thereby expanding the possibilities for quantum state engineering. However, in optical antennas and metasurfaces, resonances serve to select the wavelengths, directions, and polarizations of enhanced photon emission. Through meticulous selection and design of optical modes and resonances, the generation of entangled photons with improved efficiency can be achieved with control over their emission direction and creation of intricate quantum states.^{116,137,384,527–529}

Observing SPDC at the nanoscale presents challenges. SPDC is inherently a weak parametric process, and experiments with nanophotonic samples require prolonged data acquisition to detect a signal. To find suitable materials and structures, one can use classical processes such as second harmonic generation (SHG) and sum frequency generation (SFG).⁵²⁷ These processes share the same phase matching condition and effective nonlinearity as SPDC. Alternatively, difference frequency generation (DFG) can be employed within stimulated emission tomography (SET) to investigate the efficiency of SPDC and the entanglement of biphotons.^{116,543,545,546}

For instance, SFG was used to predict the enhancement of SPDC from a nonlinear plasmonic metagrating made of periodic silver nanostrips atop a thin lithium niobate (LiNbO₃) film and a silver substrate. Through optical excitation, localized surface and gap plasmon resonances were formed around the edges of the nanostrips and within the nanogaps. This amplified the optical field in the nonlinear LiNbO₃ layer.⁵⁴⁷ However, because plasmonic-based systems have high Ohmic losses, they also emit incoherent light, which would reduce the signal-to-noise ratio of the generated biphotons.⁵⁴⁸ 

Compared to their plasmonic counterparts, metasurfaces crafted from materials with high second-order susceptibility, such as GaAs,^{509,510,518,549} AlGaAs,^503,513,550 and other materials,^512,551,552 are attractive alternatives. The initial endeavor to observe SPDC created within nanostructures was centered around a single crystalline AlGaAs nanocylinder, characterized by Mie-like resonances at both the fundamental and down-converted wavelengths [Fig. 19(b)].⁵⁵³ The histogram of the photon arrival time difference revealed a minor peak above the background of accidental coincidences. However, the peak-to-background contrast (referred to as CAR, coincidences-to-accidentals ratio) was insufficient to confirm the emission of SPDC pairs. Given that thermal light can also lead to a peak at zero arrival time difference,⁵⁵⁴ a signature of two-photon emission is the peak exceeding the background (CAR > 1). Another hallmark of biphoton emission is the reciprocal relationship between the CAR and the pump power. The first demonstration of nanoscale SPDC using this approach emerged in the context of a nonlinear film.¹¹⁶ Very recently, SPDC was achieved in a single LiNbO₃ nanoresonator of height 500 nm and diameter $1 μ$m, which is, perhaps, the most compact 3D source of biphotons to date.⁵⁵⁵ 

One approach to achieve higher biphoton rates involves using resonances in optical metasurfaces, which enhance the vacuum field and hence the spontaneous emission of photon pairs via SPDC. This was first realized in 2021 using arrays of truncated LiNbO₃ nanopyramids [Fig. 19(c)]. These arrays exhibited Mie-like electric dipole resonances at telecom wavelengths.⁵²⁶ Given the high reflectivity of the metasurfaces at resonance wavelengths and their low reflectivity elsewhere, an amplified emission of frequency-degenerate biphotons was observed in the backward direction. Notably, the rate of pair generation surpassed that of an unpatterned LiNbO₃ film of equivalent thickness and crystalline orientation by two orders of magnitude. Remarkably, there was no discernible indication of biphotons emitted in the forward direction.⁵⁵⁶ The resonances characterizing the metasurface were adjustable by altering the size and period of the pyramidal nanoresonators, thereby enabling the tuning of the biphoton emission wavelength. However, we note that that only frequency degenerate SPDC exhibited a noticeable rate enhancement.

The enhancement achieved in SPDC due to Mie resonances is generally moderate. For more substantial enhancements in biphoton generation rates for specific directions, optical resonances with higher Q-factors, such as photonic quasi-BIC^384,527,528 and guided-mode resonances,¹³⁷ come into play. Symmetry-protected BICs can prevent, via design, the outcoupling of radiation in the direction normal to the metasurface.⁵⁴⁶ However, a symmetry-breaking perturbation of a resonator or the metasurface lattice can convert a symmetry-protected BIC into a quasi-BIC. Quasi-BIC resonances,⁵²⁵ with their elevated Q-factors, can lead to significant enhancements of biphoton generation rates compared to those obtained from traditional Mie metasurfaces. For example, this enhancement was expected in a meta-grating configuration comprising (110)-cut Al_0.18Ga_0.82As nanofins atop a dielectric substrate [Fig. 19(d)],⁵²⁸ and within an array of AlGaAs nanocylinders with pairs of holes [Fig. 19(e)].⁵²⁷ Experimentally, tunable biphoton generation mediated by quasi-BIC has been observed in a metasurface consisting of GaAs nanoresonators arranged in a square lattice [Fig. 19(f)]. The shape of each nanoresonator was designed to disrupt the rotational C2 and C4 symmetries of the lattice.³⁸⁴ This metasurface facilitated not only degenerate, but also non-degenerate SPDC, with the photons of a single pair emitted at different wavelengths. The wavelength separation between the two photons reached almost 200 nm without significant efficiency loss. By coherently exciting this metasurface at various wavelengths, photons at multiple distinct wavelengths can be entangled via pairwise coupling, offering the potential for generating more intricate quantum states. These methods of quantum state engineering, made feasible by the relaxation of momentum conservation and the exploitation of high-Q resonances, thus hold tremendous appeal.

Metasurfaces can also impart distinct directionalities to SPDC. Despite the relaxation in phase matching constraints, even within ultrathin layers, the probability of both photons being emitted backward remains exceedingly low. However, metasurface resonances engineered to exhibit enhanced reflectivity enable such backward pair emission.⁵²⁶ Conversely, resonances configured for enhanced transmissivity result in the forward emission of biphotons.³⁸⁴ By strategically designing a metasurface to possess high reflectivity at the resonance of a quasi-BIC and high transmissivity elsewhere, bi-directional emission of photon pairs was observed⁵²⁹ [Fig. 19(g)]. In this work, the photon at the resonant wavelength was predominantly emitted backwards, while its entangled partner was emitted forwards. The integration of quantum- and nanophotonics thus presents a promising avenue for generating complex photon states encompassing more than two entangled photons. Additionally, as exemplified in Fig. 19(h), guided-mode resonances have facilitated the creation of spatially entangled photon pairs with a degenerate narrowband emission spectrum spanning approximately 3 nm around 1570 nm.¹³⁷ This achievement was realized using a metasurface comprised of a silica metagrating atop a LiNbO₃ film. Notably, the biphoton generation rate was amplified by over 450 times in comparison to an unpatterned film.

Finally, we note other avenues for generating photon pairs at the nanoscale. These alternatives include the incorporation of a metalens—constructed from a 10 × 10 array of GaN nanopillars—onto the surface of a 0.5 mm β-barium borate (BBO) crystal,¹⁹⁰ which enables the generation of pairs in multiple modes as shown in Fig. 19(i). Additional methods encompass the utilization of LiNbO₃ microcubes⁵⁵⁷ [Fig. 19(j)] and GaAs nanowires,⁵⁵⁸ to amplify the biphoton rate while retaining a certain degree of engineering flexibility offered by nanophotonics. In a different vein from SPDC, some groups are investigating SFWM in doubly resonant gold plasmonic nano-antennas and metasurfaces.^158,511,546