Engineers in Singapore built polymer nanopillar arrays that steer red, green and blue light to the right pixels instead of absorbing it, producing truer colour images at one-third the exposure time of a standard filter.

Every digital camera, from a smartphone to a scientific instrument, has the same problem in the dark. Most colour image sensors rely on a mosaic of tiny filters, called a Bayer filter, laid over the pixels. Each filter lets through one colour, red, green or blue, and absorbs the rest. The result is that roughly two-thirds of the light reaching the sensor is simply thrown away.

In bright daylight, that waste hardly matters. In dim conditions, it means more image noise, longer exposures and blurred motion. Researchers at the Singapore University of Technology and Design (SUTD) now report a different approach in Nature Communications: a 3D-printed structure that sorts incoming light by colour and sends each colour to the right place, rather than discarding most of it.

“A Bayer filter separates colours by letting one colour through and absorbing the others,” said Professor Joel Yang of SUTD, who led the research. “A colour router instead steers those colours towards the corresponding sensing regions, allowing three times more of the incoming light to be used.”

Sorting light instead of blocking it

The idea of a colour router is not new, but earlier versions have relied on materials with a high refractive index, such as titanium dioxide or gallium nitride. Those materials bend light strongly, which makes it easier to redirect different wavelengths, but they require complex and costly fabrication.

The SUTD team, which included researchers from Singapore, China and Japan, set out to do the job with low-index polymers and glasses. These materials are cheaper and easier to manufacture at scale, but they interact with light more weakly, which makes precise control harder.

“Building on our experience with 3D-printed polymer optics, we wanted to explore whether carefully optimised geometry could overcome that limitation,” said Pan Cheng-Feng, an SUTD PhD candidate and the study’s lead author.

Four pillars working as one

Each repeating unit of the router consists of four cylindrical nanopillars, printed using a technique called two-photon polymerisation. In this method, a tightly focused laser hardens a liquid resin only at the point where it is most intense, allowing the printing of three-dimensional features as small as 150 nanometres, far thinner than a human hair.

Each unit corresponds to the familiar Bayer arrangement of one red, two green and one blue pixel. But the pillars do not simply each handle one colour.

“The four pillars act as a coordinated optical system,” Pan said. “A pillar positioned beneath the red sensing region is not simply an independent ‘red pillar’. In our design, the desired routing emerges from the collective response of the complete arrangement.”

To find the right combination of pillar heights and radii, the team used particle swarm optimisation, a computational method inspired by the collective movements of flocking birds, in which many candidate solutions explore a design space and share information about the best results. The researchers guided the search using whichever colour channel performed worst, ensuring that the design did not favour one colour at the expense of the others.

From 90 per cent on paper to 70 in practice

In simulations, the optimised design achieved routing efficiencies of almost 90 per cent across the visible spectrum. Commercial Bayer filters typically achieve around 20 to 35 per cent.

The team then built physical devices containing 250,000 unit cells across an area of one square millimetre. These reached about 70 per cent efficiency. The researchers attribute the gap between simulation and experiment to small deviations introduced during fabrication.

To test real-world imaging, the researchers paired the router with a monochrome sensor, which records only brightness. Because the router delivers each colour to a known location, colour information can be reconstructed from the pattern of brightness. Images captured this way showed better colour fidelity at one-third of the exposure time needed by a standard filter-based sensor.

Shorter exposures matter in low light. Every extra fraction of a second a shutter stays open increases the chance that a moving subject, or the photographer’s hand, will blur the image. Cameras compensate by boosting the sensor’s sensitivity, which amplifies noise along with the signal. Capturing more of the available light at the outset sidesteps both trade-offs.

Built with manufacturing in mind

The researchers designed the structure in what they call a “2.5D” format, with a flat base and no overhangs or enclosed cavities. That simplicity matters because it should make the design easier to replicate using moulding or other mass-production methods rather than printing each device individually.

Yang was clear that the 3D-printed devices are a proof of concept rather than the final product. “Nanoscale 3D printing is just the beginning as it allowed us to check that our concepts and designs work,” he said. “The path to manufacturing will require updated designs optimised for foundry compatible materials and processes.”

The next steps include integrating the router directly onto a working image sensor, reducing leakage of light between colour channels, improving performance when light arrives at different angles, and demonstrating that the structures can be replicated reliably.

Angle sensitivity is an important hurdle. In a real camera, light reaches the edges of the sensor at steeper angles than at the centre, and a router optimised for light arriving straight on may perform differently across the frame. Channel leakage, in which some light meant for one colour spills into a neighbouring pixel, can blur colour accuracy.

“Together, these steps would move the technology from a laboratory demonstration to a practical imaging component,” Yang said.

Beyond the smartphone

The applications are not limited to consumer cameras. The researchers suggest the approach could lead to compact multispectral sensors, which capture more than three bands of colour, and could support fluorescence imaging of living cells, where light levels are often low and prolonged exposure to bright illumination can damage the sample.

In many kinds of scientific imaging, every photon counts. A sensor that wastes less light could allow researchers to use gentler illumination, capture faster events or see fainter signals, all without changing the camera’s basic architecture.

---

Notes

1. Pan, C.-F., et al. “3D printed designer colour routers with low refractive index for low-light imaging.” Nature Communications, 2026. DOI: 10.1038/s41467-026-76072-8.
2. Singapore University of Technology and Design. “Routing the way to sharper and brighter low light photography with tiny 3D printed colour routers.” EurekAlert!, 22 October 2026. https://www.eurekalert.org/news-releases/1148839.

Photo by Lachlan Ross on Pexels.com