A metalens replaces a curved glass lens with a flat, wafer-fabricated array of subwavelength nanostructures that impose a designed phase profile on light; it is now delivering measured wins in narrowband laser, sensing and instrument optics while still failing system-level tests as a full-colour imaging eyepiece.
Summary
A conventional lens bends light by accumulating optical path difference through curved bulk glass. A metalens does the same job in a film a fraction of a wavelength thick, by patterning a dielectric surface with a dense array of subwavelength scatterers (meta-atoms, typically nanopillars or nanorods) whose local geometry sets the local phase delay. Arrange the phase to be hyperbolic and you get a focus. Because the whole device is a lithographic pattern on a flat wafer, it can in principle be made with semiconductor tooling, integrated monolithically with other photonics, and made from materials that no one can grind into a lens shape.
The physics that decides everything is dispersion. Standard designs couple the phase to meta-atom size, which ties phase directly to wavelength and produces severe chromatic aberration. That is tolerable, even useful, at a single laser line, and it is the reason the strongest results in the current literature are narrowband: a 7.2 mm diamond metalens holding two foci under 25 W pulsed irradiation with only 25.5 µm focal drift over an hour, a silicon photonic beam steerer that puts the collimating lens on-chip and steers continuously over 62° silicon photonic beam steerer based on metalens focal plane, and a quasi-optical coupler for a 1.63 THz superconducting mixer. Chromaticity can even be the product: a Moiré metalens pair turns phase-wrapping dispersion into a one-to-one mapping between rotation angle and focused wavelength, giving a 405-980 nm spectrometer.
The other parameters that decide a design are efficiency (which degrades with local deflection angle across the aperture, so high-NA and off-axis performance is not free), polarisation sensitivity, aperture size versus fabrication cost, and whether the device is fixed or reconfigurable. Reconfigurability is being pursued through phase-change chalcogenides such as Sb2S3 and GST, which switch refractive index non-volatilely between amorphous and crystalline states, giving varifocal or bifocal operation, and through all-optical pumping with sub-picosecond switching.
Viability (4/5)
The evidence base is dominated by built and measured hardware, not simulation. A monolithic diamond metalens of 7.2 mm aperture produced two foci 200 µm apart at 4 mm focal length and survived continuous-wave irradiation at 8.25 kW for 30 s without structural degradation. A fully meta-optical stereomicroscope reached 435 nm lateral and 1026 nm depth resolution. A metalens focal plane array steered a beam continuously over 62° with 19 dB average sidelobe suppression. Predictive design is also credible: a physics-informed network predicted a metasurface nanolaser threshold of 565 µJ/cm² and 620 nm emission to within 1% of measurement.
The counter-evidence is sharp and specific. A single-layer silicon nitride metalens tuned to the three RGB OLED emission peaks looked excellent on the usual PSF and MTF metrics under coherent illumination, but in a realistic VR testbed with incoherent OLED illumination and a dynamic-pupil eye model it produced pronounced ghosting and background haze, traced to the intrinsic multifocal nature of polychromatic diffractive focusing. That result also indicts the field’s standard benchmarking practice. Viability is therefore high where the source is narrowband and low where it is broadband and incoherent, and the near-achromatic aperiodic architecture that might fix this is so far theoretical only.
TLDR: Fabricated devices with measured performance across visible, infrared and terahertz; the failure mode is application-specific, not fundamental.
Drivers (4/5)
On demand, the pull is explicit in the sources: VR/AR visors need compact, lightweight optics; solid-state LiDAR and free-space optical comms want chip-scale beam steering without external collimation lenses; high-power laser manufacturing wants multi-beam throughput without bulky, alignment-sensitive, damage-prone optical trains; terahertz heterodyne receivers want compact, scalable replacements for machined elliptical silicon lenses.
TLDR: Demand is compactness and monolithic integration in AR, LiDAR and laser processing; supply is nanoimprint wafer manufacturing plus rapidly improving inverse-design tooling.
Novelty (4/5)
The clearest head-to-head is the diamond metalens against a commercial objective plus beam-splitting diffractive optical element in SiC laser scribing: 33.2 µm maximum processing-depth variation versus 319.1 µm, roughly 10x, under sustained 25 W pulsed irradiation. Elsewhere the advantage is architectural rather than incremental. Monolithic on-chip collimation removes the external lens that has blocked chip-scale focal plane arrays, and thermo-optic prisms triple the number of resolvable points by filling in the blind spots of discrete element switching. Going fully meta-optical in a stereomicroscope removes the field-of-view mismatch between meta-optical and conventional elements that limited earlier metalens-assisted systems, yielding 435 nm lateral resolution said to surpass previously reported stereomicroscopes. The Moiré meta-spectrometer trades the conventional footprint/range/resolution triangle for a rotation-angle scan and cuts measurements by nearly 3x with content-adaptive sampling.
Against that, benchmarked directly against refractive and Fresnel eyepieces in a VR system, the single-layer polychromatic metalens was worse in the ways that matter to a viewer. Novelty is real but domain-dependent, and the sources do not show a metalens beating refractive optics on broadband image quality anywhere.
TLDR: Roughly an order of magnitude better than the incumbent in high-power multifocal optics, and enables architectures with no refractive equivalent, but loses to a plain refractive eyepiece in full-colour near-eye imaging.
Diffusion (3/5)
The residual barriers are technical and cultural. Focusing efficiency depends strongly on local deflection angle across the aperture, so metalenses cannot be treated as near-ideal components the way machined elliptical silicon lenses are, and coupling to a given antenna or detector has to be modelled explicitly. Reconfigurable devices add materials risk: crystalline GST is strongly absorbing across the visible to near-infrared, which is why hybrid Si-GST designs are needed, and pushing phase-change tuning into the visible runs into shrinking meta-atom dimensions, tighter phase coverage requirements and intrinsic absorption. Most damaging for adoption timelines, the VR study shows that component-level PSF and MTF under coherent light can be actively misleading about system performance, so integrators should discount published metrics until end-to-end tests exist.
TLDR: A credible mass-manufacturing route exists via nanoimprint, but efficiency, dispersion and the missing system-level validation culture slow entry into anything broadband.
Impact (3/5)
Where it works, the value is structural rather than marginal. Monolithic on-chip collimation removes an entire discrete component from LiDAR and free-space optical link modules and eliminates the blind spots that limited continuous tracking. A single diamond element that replaces an objective plus a DOE, holds focus under 25 W pulsed loading and survives 8.25 kW CW for 30 s changes the cost and reliability envelope of multi-beam laser manufacturing. Light-emitting metalenses that lase at 616 nm with Q above 2050 and retune to 621 nm through phase-change crystallisation collapse source, cavity and wavefront control into one device, which has no conventional analogue.
TLDR: Large in specific niches such as high-power laser processing, chip-scale beam steering and THz receivers; not yet demonstrated as a general replacement for imaging optics.
Timing Soon (2-5yr)
The achromatic imaging path is on a longer clock. The aperiodic identical-nanorod architecture that claims intrinsic near-achromatic focusing at NA 0.4 and 0.8 is presented theoretically, across Si and TiO2, without fabricated devices, and the multifocal artefacts documented in the VR testbed are described as intrinsic to polychromatic diffractive focusing in a single layer. Reconfigurable phase-change devices are at the level of single-device demonstrations with visible-range absorption problems still open.
TLDR: Narrowband sensing and laser-optics deployments are plausible within two to five years on the nanoimprint route; broadband full-colour imaging is further out and may need a different architecture.
Overrated or underrated? Fairly rated
The aggregate rating is about right but the reasoning behind it is usually wrong. The popular case for metalenses, that they will replace bulky refractive optics in cameras and near-eye displays, is the weakest part of the evidence: benchmarked head-to-head against refractive and Fresnel eyepieces in a realistic VR testbed, a single-layer RGB-optimised metalens produced ghosting and haze from intrinsic polychromatic multifocality, despite excellent coherent PSF and MTF. Anyone underwriting metalenses on component-level MTF plots is underwriting the wrong number.
Prediction
By 31 December 2028, metalenses in shipping consumer devices will remain confined to narrowband or non-imaging roles (time-of-flight and structured-light sensing, AR waveguide couplers, secondary sensors), with no mass-market headset using a single-layer polychromatic metalens as its primary full-colour eyepiece.
Evidence base
Open questions
Assessment drafted 2026-08-31 from up to 15 KB sources using the technology-scorecard framework; scores are a draft read pending review.