A research team at the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) has developed a two-layer metalens with anti-reflection built into the lens itself. A metalens controls light with an array of tiny pillars (nanopillars) on its surface. Because the pillar dimensions vary from place to place, a conventional uniform anti-reflection coating has difficulty suppressing reflection across the entire surface. The team addressed this by stacking titanium dioxide () nanopillars in two layers and optimizing the shape of each one individually. Focusing efficiency measured at a near-infrared wavelength of 1,310 nm reached about 90%, roughly 4 percentage points higher than the about 86% of a single-layer lens made through the same fabrication process. The reduction in reflectance was confirmed in simulations, and the measured gain in focusing efficiency supports it.
Why a uniform coating can't adequately prevent reflection in metalenses
Metalenses, which are increasingly replacing plastic lenses in applications such as smartphone depth sensors, are ultra-thin optical elements that arrange tiny nanostructures on a surface to precisely control the phase of light. But a surface that looks uniform to the naked eye is actually made up of many nanostructures (meta-atoms) of different dimensions. Each pillar imparts a different phase delay to light, so the effective refractive index and reflectance also vary by location.
This structural characteristic has been an obstacle to applying conventional anti-reflection technology. Uniform thin-film coatings used on ordinary glass lenses are designed to match optical impedance according to the refractive index and cancel reflection at the interface. In a metalens, where the effective refractive index changes from place to place, adding a uniform film cannot provide the optimal anti-reflection condition for every nanostructure.
In a paper published in Nature Nanotechnology on November 20, 2026, the team led by Professor Federico Capasso reported a new design approach to this problem. The researchers patterned the two layers independently and modeled each combination of upper and lower nanopillars as a single thin-film stack. Each pair handles both phase control of the light and impedance matching to suppress reflection. Davide Cassara, co-first author who led the experiments, explained in Harvard SEAS's announcement that "in a two-layer design, adjusting the upper pillar to match the shape of the lower one makes local reflection control possible."
In the optimized structure, the lower layer is 1,300 nm thick and the upper layer 290 nm, with nanopillars arranged at 600 nm intervals. The pillars have square cross-sections to reduce differences in characteristics depending on polarization direction. Design-stage simulations confirmed that both the two-layer structure and a single-layer reference structure with a thickness of 1,300 nm could control the phase of transmitted light from 0 to . The minimum transmittance of the nanopillar set used in the design was calculated to improve from 86.7% in the single-layer structure to 97.6% in the two-layer structure.
Simulated reflectance falls from 5.1% to 1.2%, and measured focusing efficiency also improves
To verify the design, the team designed a metalens for a wavelength of 1,310 nm with a diameter of 250 , a focal length of 1 mm, and a numerical aperture (NA) of 0.18. It focuses light inside the glass substrate, and the single-layer comparison lens used the same aperture diameter and focal length.
Full-device finite-difference time-domain (FDTD) simulations predicted that reflectance at the design wavelength would fall from 5.1% for the single-layer lens to 1.2% for the two-layer lens. This is below the 3.3% Fresnel reflectance of a bare glass surface, and corresponds to a reduction of about 76% in the reflection loss that occurred in the single-layer structure. Predicted focusing efficiency was 91.1% for the two-layer lens and 87.4% for the single-layer lens, and the Strehl ratio, which indicates focusing quality, was about 0.99 for both. In other words, losses were predicted to fall while the precision of focusing light almost unchanged.
The wavelength and incident-angle dependence of the anti-reflection effect was also evaluated by simulation. The two-layer lens was predicted to maintain lower reflectance than an air-glass interface over a wavelength span exceeding 350 nm, and at the 1,310 nm design wavelength to keep reflectance below 2% over an incident-angle range of . In this range, there was almost no difference in characteristics between TE and TM polarization.
| Study / device | Materials and structure | Design wavelength / lens geometry | Comparison | Reflectance | Focusing efficiency | Scope and caveats |
|---|---|---|---|---|---|---|
| This two-layer metalens (Harvard SEAS, 2026) | Freestanding two-layer nanopillars. Lower layer 1,300 nm, upper layer 290 nm | 1,310 nm, diameter 250 , focal length 1 mm, NA 0.18 | Single-layer control on the same substrate | 1.2% (full-device simulation). Not directly measured | About 90% (measured), 91.1% (simulated) | Monochromatic focusing in the near-infrared. Batch-to-batch reproducibility and numerical error ranges not stated in the paper |
| This single-layer control (Harvard SEAS, 2026) | Single-layer , 1,300 nm thick | 1,310 nm, diameter 250 , focal length 1 mm, NA 0.18 | Normalized to light transmitted through an unprocessed area of the same fused silica substrate | 5.1% (full-device simulation). Bare glass is 3.3% | About 86% (measured), 87.4% (simulated) | Control device that went through the same processing steps as the two-layer device |
| Earlier anti-reflection metalens study (2023) | Multilayer structure of and . Processed with a single lithography step | 461 nm (visible), low and high NA | Visible-light metalens of the same type without an anti-reflection structure | Not stated. Mainly evaluated improvements in transmittance and focusing efficiency | About 55% → 90% at low NA, about 35% → 65% at high NA (values reported in the earlier study) | Wavelength, materials, structure, and measurement conditions differ, so it cannot be directly compared with this device |
The actual device fabrication also included measures to keep the comparison conditions equal. The two-layer lens and the single-layer control lens were fabricated on fused silica substrates of the same 500 thickness. The single-layer side went through all the other processing steps, omitting only the pattern exposure of the second layer, so that the first-layer processing conditions and post-processing effects were matched. This made it possible to compare the effect of adding the two-layer structure.
In optical measurements near a wavelength of 1,310 nm, focusing efficiency was calculated using light transmitted through an unprocessed area of the same substrate as the reference. The result was about 90% for the two-layer lens and about 86% for the single-layer control lens, confirming an improvement of about 4 percentage points in actual measurements as well. The two-layer lens maintained higher focusing efficiency than the single-layer lens over a wavelength span of at least 200 nm, and both devices showed near-diffraction-limited focusing performance across the measured range.
However, reflectance itself was not measured directly. Light reflected from a metalens scatters over a wide range of angles, so a detector with a finite aperture cannot collect enough of the reflected light, causing systematic measurement errors. In addition, the main text of the paper does not report error ranges for the key focusing efficiency values or quantitative data showing reproducibility across different fabrication batches.
What drove the efficiency gain, and how it differs from earlier work
The improvement of about 4 percentage points is not the result of a major improvement in the lens's imaging performance itself. The light intensity distribution at the focus at the design wavelength was nearly identical for the two-layer and single-layer lenses, and the Strehl ratio obtained from measurements was also about 0.99 for both. The ability to accurately shape the light's wavefront was roughly equivalent in the two.
An analysis that broke down the loss factors showed that focusing efficiency improved mainly because reflection decreased and more light passed through the lens. The measured efficiency gain of about 4 percentage points agrees well quantitatively with the reduction in reflectance predicted by simulation (from 5.1% to 1.2%).
This is not the first metalens with anti-reflection functionality. In 2023, an anti-reflection metalens using tantalum oxide and silicon dioxide (/) was reported. That method deposits multiple dielectric films and then processes them in a single nanolithography step, and at a wavelength of 461 nm it improved focusing efficiency from about 55% to 90% at low NA and from about 35% to 65% at high NA. However, because the materials, structures, operating wavelengths, and measurement conditions differ, the efficiency figures alone cannot be used to rank it against the 1,310 nm metalens in this study.
In addition, Capasso's group demonstrated a freestanding two-layer metasurface with a total thickness of 1,200 nm in research published in Nature Communications on April 1, 2025. In that technique, the upper and lower nanostructures are formed independently using two-step lithography and selective development, achieving 80% diffraction efficiency in a reflective polarization-control element using visible light at a wavelength of 560 nm. The present study builds on this two-layer fabrication technology and develops it into a design approach that incorporates anti-reflection functionality.
Technical constraints on the path to practical use, and the industrial partnership with Corning
What was demonstrated is a metalens for monochromatic near-infrared focusing, with a freestanding structure formed on a glass substrate. Harvard SEAS cites AR/VR devices, facial recognition, spectrometers, biological imaging, and optical trapping of cold atoms in quantum computers as possible future applications, but finished products for these uses have not been developed.
Design constraints also remain. The design approach is based on an effective medium approximation that holds in a regime where the optical response of periodically arranged nanostructures is dominated by the fundamental Bloch mode. The team compared the transfer matrix method with rigorous coupled-wave analysis (RCWA) and confirmed that the nanostructures used in the design operate in this single-mode regime.
On the fabrication side, the current method for making freestanding two-layer structures has the constraint that the upper pillars cannot be made thicker than the lower ones. When the same material is used for both layers, this structural condition also imposes the restriction on the effective refractive index. The team points out that a hybrid stacked structure combining different materials could relax this constraint.
Achromatic lenses that correct chromatic aberration and the addition of new optical functions that are difficult in single-layer structures are also being considered as applications of two-layer and multilayer structures. Extension to other materials such as silicon is also considered possible, but each material would require its own design and two-layer fabrication process.
The research was conducted as part of an industry-supported project by Corning. It was funded by the U.S. Air Force Office of Scientific Research (AFOSR) and also used Harvard's nanofabrication facility, which is supported by the U.S. National Science Foundation (NSF). Corning principal researcher Paulo Dainese also participated as a co-author, and Harvard's Office of Technology Development (OTD) says the collaboration with Corning has produced technology protected as intellectual property and that it is exploring commercialization opportunities.
At this point, no shipments of mass-produced products or independent reproduction tests by third parties have been reported. For practical use, in addition to demonstrating performance in structures combining dissimilar materials, data on device-to-device variation across multiple fabrication batches and direct evaluation of reflection and scattering losses will be important.
