Metasurface Router Boosts Camera Light Efficiency
A neural-network-designed nanoscale device transmits far more light to image sensors than conventional colour filters.
Updated on Nov. 11, 2026 in Materials Science
Researchers have developed a nanoscale colour router that steers red, green and blue light to the correct detector pixels instead of absorbing most of it, as conventional filters do. The experimental device transmitted 88.8% of incident light, compared with 24.7% for a conventional colour-filter array in the researchers' reconstruction.
Why it matters
Absorptive filters impose a basic limit on how efficiently cameras use available light, a constraint that matters most in low-light conditions. The approach shows that neural networks can make otherwise impractical nanoscale optical designs feasible.
The experimental metasurface transmitted 88.8% of incident light versus 24.7% for a reconstructed conventional colour-filter array, about 3.6 times more. Residual colour crosstalk and a modest loss of spatial resolution remain unresolved.
The details
The device is a patterned layer of silicon-nitride nanopillars, called a metasurface, that steers red, green and blue light toward the matching detector pixels rather than filtering out the rest. The researchers trained a neural network on 140,000 simulated structures and combined it with a genetic algorithm that progressively refined the most promising designs, which evaluated optical fields about 230,000 times faster than the original simulations.
The Big Picture
The metasurface departs from the Bayer filter, which discards most of the spectrum at each pixel to achieve colour separation. By redirecting rather than absorbing light, the approach addresses a limit built into camera sensors for decades.
The researchers say the approach could eventually improve smartphone cameras, low-light and night-vision imaging, security cameras, medical endoscopes and microscopes. Commercialization timelines were not provided in the source material.
The takeaway
Gains in camera performance may come from redesigning how light is directed at the sensor, not just from improving the sensor itself. Neural-network-assisted design could make complex nanoscale optics practical to develop.
Further reading
Explore more research in Materials Science.
Source note: This article includes information reported by Physics World.
