Metasurfaces are subwavelength arrays of engineered resonators that impose amplitude, phase and polarisation control on a wave in a layer thinner than the wavelength, and the 2026 literature shows the field splitting into two active branches: resonance-engineered optical components (sensing, absorption, nonlinear conversion, analogue image processing) and programmable radio-frequency surfaces proposed for 6G MIMO.
Summary
A metasurface is a planar array of scatterers whose spacing and size are below the operating wavelength, so the array does not diffract in the usual sense but instead acts as an engineered boundary condition that rewrites the amplitude, phase and polarisation of transmitted or reflected waves. “Flat optics” is the application framing: replacing curved, bulk elements with patterned films. The physical machinery is resonance. Dielectric elements use Mie-type resonances, metallic ones use plasmonic modes, and both can be pushed into interference-driven regimes such as bound states in the continuum, lattice resonances and anapoles. A recent perspective argues these platform-specific vocabularies (photonic crystals, plasmonics, dielectric metasurfaces) describe the same underlying object, the eigenmodes of an open system, seen through different scattering features.
The parameters that decide a metasurface are quality factor, loss, field overlap and fabrication tolerance, and they trade against each other. High Q concentrates the field and sharpens the spectral response but is fragile to dissipation and geometric error: an all-metallic design that balances a Fabry-Perot BIC dark mode against a Rayleigh-anomaly lattice resonance reached a measured Q of 2180 (2800 in theory) with nearly 99% absorption, and held that behaviour across pillar heights of 70 to 120 nm and radii of 210 to 280 nm. The opposite strategy also works: deliberately low-Q all-dielectric surfaces with engineered evanescent modes can beat the sensing performance normally expected of low-Q structures by maximising near-field overlap with the analyte. Quasi-BIC resonances are usually created by breaking geometric symmetry; one 2026 result instead breaks it with material anisotropy, inserting a 20 nm BeS layer with permittivity anisotropy of about 0.27 into a symmetric TiO2 nanobar pair to obtain Q of roughly 181 while leaving the geometry untouched, and uses the resulting transfer function for optical edge detection.
Three other threads matter. Tunability: phase-change chalcogenides offer refractive index changes above 0.7 on switching, and a two-photon printing route now writes them as freeform 3D structures rather than patterned thin films; hydrostatic pressure up to 5 GPa shifts a silicon nitride Fabry-Perot resonance by up to 30 nm and changes index by up to 4%, enabling a polarisation-converting surface intended for extreme environments. Active and nonlinear function: a thin-film lithium niobate metasurface performs wavelength-selective second-harmonic wavefront shaping, converting a Gaussian pump near 1100 nm into a first-order Hermite-Gaussian mode at 550 nm, and Er-implanted Si3N4 nanocylinder arrays enhance room-temperature telecom-band photoluminescence by a factor of about 18 with a tenfold lifetime reduction via the Purcell effect, on a CMOS-compatible platform. Design tooling: differentiable multiple-scattering solvers now return exact gradients with respect to geometry and position for both finite clusters and infinite metasurfaces, motivated explicitly by sub-10 nm fabrication expanding the design space, and simplified multipolar models reduce design complexity by exploiting the origin dependence of multipole moments.
Separately, the same idea has been transplanted to radio frequencies as programmable surfaces for wireless. Here the objects are waveguide-fed metasurface antennas for extremely large MIMO, stacked intelligent metasurfaces that cascade several programmable layers, and flexible surfaces that morph their 3D geometry to add spatial degrees of freedom. That branch is currently dominated by modelling accuracy and channel estimation, not hardware.
Viability (4/5)
On the optical side the evidence is experimental and quantitative rather than promissory. A measured Q of 2180 with close to 99% absorption, robust across a wide geometric window, is a fabricated device. Erbium-doped Si3N4 metasurfaces show a measured photoluminescence enhancement of about 18 and a nearly tenfold lifetime reduction in agreement with simulation. Nonlinear wavefront shaping in thin-film lithium niobate is a demonstrated mode conversion, not a design study. Pressure tuning to 5 GPa with a 30 nm resonance shift is measured. The modelling stack has also matured: physically consistent dipolar and multipolar frameworks with passivity constraints, symmetry formalisms for multi-order gratings, and differentiable solvers that give exact geometric gradients.
The caveats are specific. Q values in the hundreds to low thousands are laboratory measurements on small areas; none of the sources reports wafer-scale uniformity, yield, environmental stability or switching endurance for the tunable variants. The RF branch scores lower on its own: the waveguide-fed antenna, stacked and flexible metasurface papers present frameworks, full-wave simulations and estimation algorithms, and explicitly flag that accurate modelling of mutual coupling and inter-layer coupling remains challenging. A 4 reflects the optical hardware; taken alone the wireless work would be a 2.
TLDR: Optical metasurfaces are demonstrated in hardware with measured figures of merit; the RF programmable branch is still models and simulation.
Drivers (3/5)
Supply side is clear. The design space is being opened by fabrication capability, with inverse-design tooling built specifically because sub-10 nm features expand the parameter space by orders of magnitude. New material platforms are arriving in parallel: phase-change chalcogenides with index contrast above 0.7 now printable in freeform 3D, thin-film lithium niobate for nonlinear function, anisotropic van der Waals crystals such as ReS2 for polaritonic engineering, and CMOS-compatible Si3N4 for active emitters. Modelling that is cheap enough to sit inside an optimiser is itself a driver.
Demand side is stated but not evidenced. The wireless papers invoke 6G and beyond, extremely large MIMO and scalable antenna implementation as the pull; the optical papers point at substance detection, reconfigurable and programmable photonic circuits and integrated on-chip light sources. None of the sources contains market size, procurement, volume or customer evidence, so the demand argument rests on roadmap language rather than data. That ceiling is why this is a 3 and not higher.
TLDR: Supply-side push from fabrication and inverse design is well documented; demand is asserted through 6G and sensing rather than measured.
Novelty (3/5)
The comparisons in the sources are mostly against the metasurface literature itself, not against bulk optics. The all-metallic design is better than conventional plasmonic metasurfaces in that it escapes the usual dissipation-versus-Q trade-off, reaching a measured Q of 2180 with 99% absorption where plasmonic surfaces normally cannot have both. The low-Q sensing work claims sensitivity beyond what is typically reported for low-Q structures, but quantifies the improvement relative to literature norms rather than a single named benchmark. The anisotropy-driven quasi-BIC is better than geometric symmetry breaking in kind rather than degree: Q of about 181 is unremarkable, but the resonance becomes tunable by material choice while the geometry is fixed. Nonlinear wavefront control gains wavelength selectivity that geometric-phase approaches lack, which makes previously static nonlinear responses spectrally addressable.
Two entries are more than incremental. Direct 3D printing of high-index phase-change chalcogenides lifts a real constraint, since existing implementations rely on 2D or quasi-3D patterning of coated films and therefore cannot prototype freeform geometry. On the wireless side, stacked surfaces are argued to enable richer wave manipulation than single-layer metasurfaces, and morphing surfaces to add spatial degrees of freedom unavailable to rigid arrays. Both remain unvalidated in hardware in these sources. Net: real but mostly intra-field advances, hence 3.
TLDR: The paradigm is mature; the genuine 2026 novelty is in control mechanisms and modelling, with baselines usually qualitative.
Diffusion (unscored)
Every source here is a 2026 preprint reporting physics, device demonstration or modelling. There is no information on production volumes, yield, cost per die, supplier ecosystems, product integration, qualification or standardisation. The nearest things to adoption signals are indirect: Si3N4 metasurfaces are described as a route to integrating light sources into CMOS-compatible photonic devices, the pressure-tunable platform is aimed at deep-ocean and planetary environments, and the RF papers position programmable surfaces within 6G system architectures. None of these is evidence that anyone is buying, building at volume or standardising.
What the sources do let one infer about barriers is narrow but useful. Fabrication tolerance is treated as a design objective, which implies it is a live obstacle: the all-metallic absorber’s selling point is that near-perfect absorption survives a wide geometric window. For the wireless branch, the stated barriers are modelling fidelity under strong mutual coupling and the difficulty of acquiring channel state information when the surface has many programmable elements or deforms continuously. Those are necessary but nowhere near sufficient conditions for a diffusion score.
TLDR: The sources contain no adoption, manufacturing or commercial evidence, so no diffusion judgement is defensible.
Impact (3/5)
The breadth is the argument for impact. Within a single month of literature, metasurfaces are shown performing near-perfect narrowband absorption, refractive-index and absorption-based substance detection, Purcell-enhanced telecom-band emission on a CMOS-compatible substrate, frequency conversion with simultaneous spatial mode shaping, analogue spatial differentiation and edge detection in the optical domain, and strong light-matter coupling architectures for polariton chemistry and transport. If the wireless branch works, the claimed prize is scalable extremely large MIMO apertures and channel diagonalisation performed in the analogue domain.
The constraint on the score is that value is asserted at the component level only. No source reports what a metasurface replaces, at what cost, with what power or size saving, or against what incumbent performance. A Q of 2180 and a Purcell factor of 18 are good numbers in a physics paper; neither tells you the value of the resulting product. There is also no evidence here on the highest-profile flat-optics claim, the substitution of bulk imaging lenses, so that part of the impact case cannot be assessed from this material at all. A 3 records credible, broad, unquantified value.
TLDR: Demonstrated functions span sensing, absorption, on-chip emission, analogue computing and wireless, but no source quantifies system-level value.
Timing Soon (2-5yr)
The technology is not one clock. Passive resonant optical surfaces, absorbers, sensors and analogue image-processing elements already exist as measured devices with performance figures that would be usable if manufacturing held up. The gating question for those is process, not physics, and the sources are silent on process. Tunable platforms are one step behind: phase-change chalcogenide printing is a new fabrication capability with no cycling or device data reported, and 5 GPa pressure tuning is by construction a niche actuation route.
The programmable RF branch is the latest. Its 2026 output is modelling frameworks, cascade optimisation and channel estimation algorithms validated numerically, tied to 6G timelines that these sources do not date. Treat that as a separate, later clock. The 2 to 5 year band applies to component-level optical function; anything requiring a programmable, reconfigurable or morphing surface in a deployed system is unsupported by the evidence here.
TLDR: Passive resonant components have measured performance now; tunable, nonlinear and programmable variants are earlier and the wireless branch has no hardware evidence yet.
Overrated or underrated? Fairly rated
The physics is real, reproducible and advancing on several fronts at once, and the modelling and inverse-design tooling has caught up enough that designs are now optimised rather than guessed. What has changed in 2026 is less about flat lenses and more about resonance engineering as a general capability: balancing radiative against dissipative loss to get high Q and near-total absorption together, or deliberately abandoning high Q when near-field overlap matters more. An investor should read the field as a component technology maturing in specific narrow functions, not as a single platform about to displace bulk optics.
The honest caveat is that the sources supplied cannot settle the commercial question. There is no manufacturing, yield, cost or adoption data of any kind, and the wireless application that carries the largest claimed value, programmable surfaces in 6G MIMO, currently exists only as models and estimation algorithms with acknowledged coupling problems. Anyone quoting the 6G metasurface story as near-term is running ahead of this evidence. Fairly rated on the physics, unassessable on the market.
Prediction
By June 2028, a peer-reviewed hardware measurement of a stacked intelligent metasurface will report over-the-air MIMO channel diagonalisation on a physical multi-layer prototype; absent that, the stacked-metasurface line remains a modelling exercise.
Evidence base
- An all-metallic plasmonic metasurface combining a Fabry-Perot BIC dark mode with a Rayleigh-anomaly lattice resonance measured Q of 2180 (2800 theoretical) with nearly 99% absorption, retained across pillar heights of 70 to 120 nm and radii of 210 to 280 nm (29 May 2026).
- Er-implanted Si3N4 nanocylinder metasurfaces gave a room-temperature telecom-band photoluminescence enhancement of about 18 at a 390 nm radius, with a nearly tenfold lifetime reduction attributed to the Purcell effect, and a fourfold emission increase as implantation range moved from 20 to 80 nm (5 May 2026).
- A quasi-BIC created by inserting a 20 nm anisotropic BeS layer into a symmetric TiO2 nanobar pair reached Q of about 181 for an index anisotropy of about 0.11, produced a transfer function with a notch at normal incidence and 180 degree phase reversal, and demonstrated edge detection on a USAF 1951 chart (2 June 2026).
- Hydrostatic pressure up to 5 GPa shifted a silicon nitride Fabry-Perot resonance by up to 30 nm and reduced relative refractive index by up to 4%, enabling a reported first extreme-pressure-tunable polarisation-converting metasurface (15 May 2026).
- A thin-film lithium niobate metasurface with two regions of distinct resonance-engineered phase response converted a Gaussian pump near 1100 nm into a first-order Hermite-Gaussian second-harmonic mode at 550 nm, adding wavelength selectivity that geometric-phase nonlinear metasurfaces lack (2 June 2026).
- Differentiable T-matrix multiple-scattering tooling returning exact gradients for finite clusters and infinite metasurfaces was motivated explicitly by sub-10 nm fabrication expanding the design space by orders of magnitude (28 May 2026).
Open questions
- Do the measured high-Q resonances (Q around 2180 with 99% absorption) survive wafer-scale fabrication variation, angular spread and temperature, or only over the small geometric windows reported on test samples?
- How many switching cycles, and at what optical loss penalty, can 3D-printed phase-change chalcogenide structures sustain, given the >0.7 index change is the whole basis of the tunability claim?
- Are the coupled-dipole and multi-port S-parameter models for waveguide-fed and stacked metasurfaces accurate against measured hardware, not just full-wave simulation, once real mutual and inter-layer coupling is present?
- Does the pilot overhead of tensor-based channel estimation for stacked or morphing surfaces stay small enough to leave a net spectral-efficiency gain in a real system?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.