Can transparent, low-refractive-index polymers route visible colors efficiently at the pixel scale? Pan and colleagues inverse-design a four-nanopillar RGGB color router and fabricate it in IP-L resin using two-photon polymerization. Simulations give approximately 87% overall visible transmission and experiments approximately 70%, with measured peak routing efficiencies of about 40%, 51% and 48% for blue, green and red. A projector-and-microscope imaging experiment with a monochrome CMOS camera demonstrates color reconstruction and shorter exposures. This is an optical proof of concept rather

A camera usually throws light away to see in color. Four tiny, transparent, 3D-printed pillars offer another way: send each color where it belongs.

What if color came from redirecting light, rather than absorbing it?

A conventional Bayer filter lets one color through each pixel and absorbs much of the rest. Pan and colleagues instead design a microscopic landscape in low-index polymer. Light scatters from the pillars, interacts between them, and concentrates into a familiar red–green–green–blue pattern. The material is clear; its geometry does the work.

Find the color plane.

Move the observation plane upward. A mixed field resolves into four color regions. Then illuminate the whole structure with just one band, or compare it with an absorptive filter.

Top view of the planeFour 1 µm-square regions. The two green regions together form the green channel.

A conceptual teaching model inspired by Supplementary Video 1. Paths and spot shapes illustrate the mechanism; they are not an electromagnetic simulation.

The whole array participates.

Scattering, coupling and near-field interference make the four-pillar response different from simply adding four isolated pillars.

One clear polymer.

All posts use the same transparent IP-L polymer, with refractive index about 1.52. Pillar names refer to the focal region above them, not to dyes.

The plane matters.

At the designed focal plane, the light falls into blue, two green and red collection regions. Each is 1 × 1 µm within a 2 × 2 µm unit cell.

Color appears as focus moves.

The microscope records a patterned array under white illumination. As the focus travels away from the printed structure, distinct colored spots emerge. This depth-dependent separation is the inspiration for the movable plane in the virtual lab.

Which pixel gets a photon?

Sweep the wavelength through the visible spectrum. These curves replot the paper’s supplied numeric data. Compare the fabricated device with its simulated design.

Transmission at the designed color plane

Figure 2b–c source data, replotted. Readouts interpolate between supplied wavelengths. The source reports G1 + G2 together; this symmetric display divides it equally between the two green regions. Percentages refer to incident light over the whole unit cell.

Measured average transmission Reported experimentally across the visible spectrum.

Simulated average transmission Reported for the inverse-designed structure.

Transmission is not the same as correct routing.

The white curve tracks light transmitted through the cell. The colored curves track light arriving in specific collection regions. Measured peak routing is about 40% for blue, 51% for the combined green channel and 48% for red, at different wavelengths.

Why optimize the weakest color?

The design searches six geometric variables: three heights and three radii, with the second green post repeated by symmetry. Particle swarm optimization maximizes the weakest target RGB response, helping all three channels perform together.

Figure 2 · Pan et al., Nature Communications (2026). © The Authors 2026. Reproduced unchanged under CC BY-NC-ND 4.0. Open the source figure ↗

See the brightness arrive.

More collected light changes how long a camera needs to expose. Increase the exposure and compare the paper’s measured white-light brightness curves.

Illustrative mosaics driven by measured brightness. Each response is normalized to its own plateau.

The paper reports comparable image brightness at 4 ms with the router and 10 ms with the filter. In the supplied Figure 4 source data, the router reaches its normalized plateau at 4 ms; the filter reaches its plateau at 12.5 ms.

Figure 4 brightness data are interpolated above. The demonstration uses a projector, optical components and a monochrome CMOS camera to record the routed pattern, then reconstructs color from the channel intensities.

Write the shape of color.

Two-photon polymerization turns the optimized design into physical nanopillars. A tightly focused laser writes the structure inside a photosensitive resin.

A low-index material. A deliberately shaped field.

The experiment prints a 1 × 1 mm array containing 500 × 500 unit cells. Three pillar geometries, arranged into repeating four-post cells, are enough to produce the RGGB collection pattern.

Search the geometry

Simulate candidate heights and radii. Improve the smallest target-channel efficiency so one strong color does not hide a weak one.

Write in clear resin

A Nanoscribe Photonic GT writes IP-L layer by layer, using concentric laser paths. Develop the resin to leave the printed posts.

Locate the color plane

Illuminate and scan the microscope focus. Measure the separated colors at the designed 5 µm observation distance.

Reconstruct RGGB

Read the four intensity regions on a monochrome camera. Combine the red, green and blue signals into a color image.

The actual printed geometry B: height 1.2 µm, diameter 0.56 µm. G1 and G2: height 1.9 µm, diameter 0.48 µm. R: height 2.0 µm, diameter 0.68 µm.

Two-photon printing in the Methods IP-L photoresist · 63× objective · numerical aperture 1.4. Dip-in laser lithography · 5000 µm/s scan speed · 27.5 mW. Fabrication paths detailed in Supplementary Note 8.

A better photon economy.

Redirecting photons can make color imaging brighter at shorter exposure times, especially when light is scarce. This work shows that a low-index, printable polymer can support that strategy, expanding the materials available for pixel-level color routing.

The imaging experiment is an optical proof of concept. Its reconstructed bird image is stitched from 17 × 18 scanned blocks, each containing 40 × 40 router cells. Direct sensor integration will need control of alignment, spacing and illumination angle; stable routing is reported up to 6°. The authors also propose using printed structures as masters for replication.

3D printed designer color routers with low refractive index for low-light imaging

Cheng-Feng Pan, Hao Wang, Bochang Wu, Max J. H. Tan, Shutao Zhang, Hongtao Wang, Wang Zhang, Zhenchao Liu, Lei Chen, Tomohiro Mori, Md Abdur Rahman, Xiaoyan Zhou, Qifeng Ruan, Cheng-Wei Qiu & Joel K. W. Yang. Nature Communications (2026). DOI: 10.1038/s41467-026-76072-8.

Original Figure 2 and Supplementary Video 1 are credited at their point of use. Numeric curves are replotted from the supplied source data. The graphical abstract is an AI-generated conceptual illustration; the interactive optical lab is a conceptual teaching model.

AI-generated conceptual illustration.