Helping miniature cameras see more light
3d rendering of eyeo's waveguide structure (image: eyeo)
Dutch deeptech company eyeo is developing a new way for miniature cameras to capture colour without throwing away most of the light that reaches them.
The Eindhoven-based imec spin-off replaces the red, green and blue filters used in conventional image sensors with nanoscale structures that separate light into colours and direct each towards the correct pixel. eyeo says its approach can make sensors three times more light-sensitive while allowing pixels to shrink below 0.5 micrometres.
That combination could prove particularly useful in AI glasses. Cameras built into a lightweight frame must capture usable images in poor light, recognise writing from a distance and compensate for the constant movement of the wearer’s head. Yet they have far less room for large sensors, lenses and batteries than a smartphone.
A camera that collects more of the available light could help AI glasses read a departure board in a dimly lit station, identify an object after sunset or translate a menu in a dark restaurant. Better source images would also give the artificial intelligence interpreting them more reliable information.
Google has said its first Gemini-powered audio glasses are coming later this autumn, with Samsung, Gentle Monster and Warby Parker among its partners. The glasses are intended to take photographs, translate writing and allow wearers to ask Gemini questions about what they see. Google is also working on models that display information directly on their lenses.
eyeo has no announced involvement in these products. Its technology instead addresses a problem shared across the emerging AI-glasses industry: the software may be increasingly capable, but it can only interpret what the camera manages to capture.
For eyeo, the opportunity extends well beyond eyewear. Smartphones, security systems, medical equipment and industrial cameras all stand to benefit from smaller image sensors that perform better in difficult light. First, however, the company must prove that its nanophotonic approach can make the journey from an imec laboratory to mass production.
The image sensor problem inside AI glasses
A digital camera sensor contains millions of light-sensitive sites called pixels. When photons reach a pixel, the sensor converts them into an electrical signal. Collecting more light generally makes it easier to produce a clean image with less electronic noise.
Shrinking an image sensor usually means reducing either the number or size of its pixels. Smaller pixels collect fewer photons, particularly when light is scarce. The resulting image may contain more noise, less dependable colour information and fewer usable details.
That is a particular problem for AI glasses. A human wearer may be able to read the station display while the miniature camera sees an unstable collection of bright letters surrounded by digital noise. If the source image is poor, even a powerful AI model may misread a platform number or departure time.
Phone manufacturers have partly escaped this constraint by using larger sensors, multiple lenses, computational photography and increasingly prominent camera housings. The camera bump has become a familiar feature of modern smartphones.
Glasses have nowhere obvious to put one. Thicker frames add weight, while larger sensors and more intensive image processing can increase power consumption. A device intended to disappear into everyday life cannot feel like a computer strapped to the wearer’s temple.
The problem is not only the amount of light entering the lens. It is what happens to that light when it reaches the camera sensor.
The colour filters that absorb light
Silicon image sensors measure the intensity of light but need additional structures to distinguish colours. Most digital cameras obtain colour information by placing a mosaic of tiny red, green and blue filters over their pixels.
The widely used Bayer pattern contains twice as many green-filtered pixels as red or blue ones, broadly reflecting the human eye’s greater sensitivity to detail in the green part of the spectrum.
Each filter transmits part of the spectrum while absorbing or rejecting much of the rest. Software then uses information from neighbouring pixels to reconstruct the red, green and blue values needed for a full-colour image.
This system has proved extraordinarily successful. Variations of the colour filter array are used in billions of smartphones, vehicles, security cameras and industrial machines.
But it comes with an unavoidable compromise. The filters establish colour partly by discarding light.
Imec, the Belgian semiconductor research centre from which eyeo emerged, says a conventional Bayer filter array can absorb around 70% of the light reaching it. That does not mean every existing camera is literally 70% blind: real performance depends on the filter design, sensor and wavelengths involved. The underlying limitation is nevertheless real. A photon absorbed before reaching the silicon cannot contribute to the electrical signal.
eyeo’s Nanophotonic Color Splitting technology, known as NCOS, is designed to replace absorption with direction.
Instead of using filters to reject unwanted wavelengths, arrays of vertical waveguides separate incoming light into colours and guide it towards different pixels. A conventional filter behaves rather like a roadblock that admits one colour and turns the others away. eyeo is attempting to build a junction that sends each colour down the correct lane.
Nanophotonic colour splitting explained
The structures performing this sorting operate at dimensions comparable to or smaller than the wavelengths of visible light. By controlling their geometry and refractive properties, engineers can alter how light travels through them.
This is nanophotonics: the manipulation of light using structures measured in billionths of a metre.
eyeo says its image-sensor architecture can provide three times the light sensitivity of conventional filtered sensors and enable pixels smaller than 0.5 micrometres. Imec presented the underlying waveguide-based colour-splitting technology at the IEEE International Electron Devices Meeting in 2023.
Its research describes structures made from silicon nitride embedded in silicon dioxide and fabricated using standard back-end semiconductor processing on 300-millimetre wafers.
If those advantages can be reproduced economically in mass-manufactured camera sensors, device designers would gain new options.
They could use the additional sensitivity to improve images taken in dim conditions. They could pursue higher resolution without increasing the sensor area. Alternatively, they might build smaller cameras while attempting to preserve usable image quality.
For AI glasses, that could make the difference between a camera that merely fits inside the frame and one that remains effective after sunset. Smartphones, medical equipment, industrial inspection systems and security cameras could benefit from similar improvements.
None of this means eyeo is about to eliminate the smartphone camera bump or appear in the first Gemini glasses. The company is developing an enabling technology that image-sensor manufacturers and device makers must still integrate, qualify and produce at scale.
Other ways to improve miniature cameras
Camera manufacturers have several ways to compensate for small image sensors.
Better lenses and microlenses can direct more light towards the silicon. Backside-illuminated and stacked sensor architectures can improve light collection or create more room for supporting electronics. Cameras can combine several exposures into one picture, while increasingly sophisticated computational photography can reduce noise, correct movement and reconstruct missing details.
Device manufacturers could also accept thicker frames, shorter battery life or lower image quality in exchange for bringing products to market sooner.
These approaches are not mutually exclusive. A future AI-glasses camera could combine improved optics, stacked electronics, advanced image processing and nanophotonic colour splitting. eyeo’s proposition is that retaining more of the original light provides a stronger starting point for everything that happens afterwards.
Software can repair an imperfect image. It cannot fully recover information that the sensor never recorded.
From an imec laboratory to a commercial foundry
eyeo was created as an imec spin-off in 2024. Its founding team consists of Jan Genoe, Alden Carracillo, Jeroen Hoet and Gerd Van den Branden, with Hoet serving as chief executive. The technology is based on approximately seven years of research conducted at imec.
In May 2026, eyeo raised €40 million in a Series A round led by Dutch deeptech investor Innovation Industries. Existing investors imec.xpand, the Invest-NL Deep Tech Fund, QBIC, High-Tech Gründerfonds and the Brabant Development Agency also participated.
Together with its €15 million seed round in 2025, that brought eyeo’s total funding to €55 million. The company says its technology is protected by 26 patents.
The funding is not the reason AI glasses suddenly need better cameras. It does, however, give eyeo the resources to confront the part of deeptech innovation that receives less attention than the scientific breakthrough: manufacturing it reliably.
Its current priorities include building an integrated-circuit and systems-design team in Antwerp, developing colour-splitting image sensors using 3D-stacked complementary metal-oxide-semiconductor technology and expanding relationships with original equipment manufacturers.
The commercial test for eyeo’s camera technology
eyeo has demonstrated that its colour-splitting technology works and says it has successfully integrated the process at a commercial foundry. The next step is considerably more demanding: producing millions of sensors reliably, at a price customers are prepared to pay.
The company’s waveguides must be manufactured at extremely small dimensions and aligned precisely with the pixels beneath them. They must separate colours consistently under different lighting conditions and across the range of angles at which light enters a camera. Production yields must also be high enough to compete with conventional colour filters, which are inexpensive, predictable and supported by supply chains refined over decades.
Adoption will require more than proving the performance of the nanophotonic structures. Image-sensor manufacturers must qualify the technology, adapt their designs and demonstrate that it can withstand the demands of high-volume production. Device makers will then need convincing that the gains in sensitivity, size or image quality justify changing an established component.
The first commercial application may therefore be in industrial inspection, security cameras or medical imaging rather than AI glasses. Customers in these markets may be more willing to accept higher initial costs when improved low-light performance or smaller sensors solve a specific problem. Smartphones and wearables offer far greater volumes, but also impose some of the industry’s toughest requirements for cost, reliability and scale.
In short: eyeo has raised the capital, built the scientific foundation and begun working with potential customers. Its defining test now is whether nanophotonic colour splitting can become routine semiconductor manufacturing.
Liked this article? You can support our independent journalism via our page on Buy Me a Coffee. It helps keep MoveTheNeedle.news focused on depth, not clicks.