The study asks how a manufacturable polymer metalens can bring the red, green and blue channels of a compact near-eye display to one focus. A differentiable inverse-design framework selects cylindrical pillars from a fabrication-constrained height-and-diameter library and maximizes the weakest RGB focusing efficiency. Two-photon lithography produces the three-dimensional master; a conformal silicon dioxide layer enables soft-mold nanoimprint replication. Simulated efficiencies are 33%, 34% and 34% at 450, 532 and 635 nm; measured efficiencies are 27.1%, 28.2% and 29.1%, with diffraction-limite

NATURE COMMUNICATIONS · 2026

A forest of polymer nanopillars brings red, green and blue light to the same focus. The secret is a new way to choose their three-dimensional shapes.

A research experience based on Jeon et al.

01 / THE OPTICAL IDEA

A lens without the usual curve.

A conventional lens bends light through a curved surface. A metalens coordinates light with an array of tiny structures.

Each pillar changes the phase of the light passing through it. Arrange those local delays correctly and the emerging waves meet at a focus. But different colors respond differently to the same geometry. This paper gives the design another degree of freedom: pillar height, as well as diameter.

YOUR FIRST EXPERIMENT

Bring the colors together.

Switch the geometry. Then move the observation plane to find the sharpest spot.

One geometry must serve three wavelengths. Add height as a second design freedom.

Illustrative optical model · in-focus widths from Fig. 4. Diameter-only focal separations and the color target are schematic; off-plane spots use a Gaussian teaching model.

Focusing efficiency · 1.147 µm FWHM

Focusing efficiency · 1.376 µm FWHM

Focusing efficiency · 1.646 µm FWHM

THE COMPUTATIONAL STEP

Choose a printable shape. Help the weakest color.

The design searches a library of printable cylinders, then evaluates how the whole lens focuses light. The final lens uses 51 distinct diameter–height combinations, listed in Supplementary Table 1.

A library of light responses

Rigorous coupled-wave analysis calculates each candidate’s phase and amplitude at the three target wavelengths.

One real pillar per position

Eight phase-compatible candidates are shortlisted for each ring. Hard Gumbel–Softmax selects one shape, while a gradient estimator lets the design keep improving.

Optimize the weakest channel

Wave propagation predicts the lens’s focusing efficiency. A minimax objective improves the lowest RGB efficiency, balancing simulated results to 33%, 34% and 34%.

02 / FROM SHAPES TO MATTER

Write a lens with light.

Two-photon polymerization turns a focused laser into a tiny three-dimensional pen.

Only near the intense laser focus does the IP-DIP2 polymer resin solidify. Stacking these small written volumes builds pillars of different heights; laser power changes their diameter.

The researchers used a Quantum X shape printer with 780 nm femtosecond pulses. This is how the height freedom in the optical design becomes a physical lens.

Square lattice pitch

Designed pillar heights

The nanopillar workshop

900 nm tall · about 60 nm wider than at 12.5 mW.

Schematic array; the height control sets the central reference pillar. Reported calibration: approximately +20 nm in diameter per +0.5 mW at 1000 µm/s. High-power bridging is shown in Supplementary Note 2; merged pillars are excluded from the library.

PRINT ONCE. REPLICATE.

A master worth keeping.

A 20 nm conformal SiO₂ coating protects the printed master. A soft mold then transfers the pillar pattern into UV-curable resin by nanoimprinting.

The paper reports no observed morphological degradation in the fifth and tenth replication cycles. Try casting with and without the protective step.

From master to replica

Protection helps the master release from the soft mold.

Schematic demonstration of adhesion and release, not a quantitative failure model. The paper examines replicas made from the same soft mold after five and ten cycles.

03 / THE DISPLAY CONNECTION

Small optics. Full-color ambition.

A thin lens can bring the optics closer to the display — and potentially closer to the eye.

The authors demonstrated a full-color near-eye display on an OLED optical bench. Its red, green and blue channels share a focal plane, opening a route toward more compact display optics.

The design targets three wavelengths, rather than the entire visible spectrum. The supplementary results show focal shifts away from those bands; the OLED’s roughly 40 nm emission widths also contribute residual blur.

FROM THE SUPPLEMENTARY MATERIAL

Watch the authors’ near-eye experiment.

A moving pattern puts the color alignment to the test. The five columns show the target, the RGB achromatic lens, and the blue, green and red focal planes of a chromatic lens.

Jeon et al. (2026), Supplementary Movie 1 · © The Author(s) · CC BY-NC-ND 4.0. Format converted to WebM; no content edits.

LOOK AT THE EVIDENCE

The lens, as measured.

The original figures connect fabrication, microscopy and the optical test. Open a figure to explore the details.

Printed, molded, replicated.

The process diagram and micrographs reveal the polymer master, soft mold and imprinted lens.

Three sharp focal spots.

Measured Strehl ratios of 0.90, 0.89 and 0.93 accompany the RGB spot profiles at the shared focus.

Figures 3 and 4 reproduced without content changes from Jeon et al. (2026), © The Author(s), CC BY-NC-ND 4.0.

THE RESEARCH PAPER

Differentiable library-based inverse design of achromatic metalens for full-color near-eye displays

Youngsun Jeon, Cherry Park, Shiqi Hu, Junhwa Seong, Seokwoo Kim, Jaekyung Kim, Sangwon Lee, Jihwan An & Junsuk Rho

Nature Communications 17, article 6310 (2026)

This experience draws on the paper’s inverse-design method, fabrication Methods, Figs. 3–4, and supplementary analysis of wavelength-dependent focus and replication.

The optical bench is a teaching model. Its split-focus comparison is schematic, and its off-plane point-spread functions are illustrative Gaussians anchored to the measured in-focus FWHM values. Efficiencies, focal dimensions and Strehl ratios identified as measured are experimental results; the 33%, 34% and 34% optimization values are simulated.

The nanopillar workshop shows how power and height affect shape using the paper’s reported relative diameter calibration. The replication scene illustrates the reported role of a low-damage, conformal SiO₂ coating deposited at 60 °C; its animation is not a measured failure rate.

The generated graphical abstract is an AI-generated conceptual illustration, not to scale. Original paper figures are reproduced without content changes under CC BY-NC-ND 4.0.

Jeon et al. (2026) · © The Author(s) · CC BY-NC-ND 4.0 · reproduced without content changes