Six years ago, a research team's artificial eye could resolve about 100 pixels — roughly enough to make out a blurry shape. Their newest version packs 367,500 sensing pixels onto a curved surface shaped like an actual eyeball, published this week in the journal Nature Materials.

  • 367,500 sensing pixels on the new hemispherical sensor
  • 1,905 pixels per inch — the resulting image density
  • 160+ degrees of field of view, aberration-corrected
  • >99% reduction in data bandwidth, thanks to built-in motion detection
  • 100 pixels — what the team's 2020 proof-of-concept could manage

Why a curved sensor was the hard part

Every digital camera sensor made today, from phone cameras to telescopes, is manufactured flat, because standard lithography tools etch circuits onto flat silicon wafers. The human eye is curved for a reason: a spherical surface naturally focuses light across a wide field of view without the heavy, distorted lens stacks flat sensors require.

Engineers at the Hong Kong University of Science and Technology solved that manufacturing mismatch using a 3D laser-assisted fabrication process that builds sensing pixels directly onto a hemispherical substrate, according to a report on the research. That let them mimic the retina's actual geometry instead of flattening the design and losing the optical advantages that come with curvature.

Planar sensors need multiple lens elements stacked in a row to correct for distortion at wide viewing angles, which adds bulk, weight, and cost. A curved sensor that matches the eye's natural geometry can collect a wide, sharp field of view with far simpler optics on top of it, which is part of why this approach has been pursued, on and off, by vision researchers for more than a decade without reaching this level of resolution.

Full color, and a trick borrowed from biology

The new sensor doesn't just see more pixels — it sees color, through layered sensing elements tuned to blue, green, and red wavelengths across a 300-to-800-nanometer spectral range. That's a meaningful jump from earlier bionic-eye prototypes, which were often limited to grayscale or narrow spectral bands.

The team also built in event-driven motion detection directly at the sensor level, so the hardware only transmits data when something in the scene actually changes, rather than streaming every pixel continuously. That single feature cut outgoing data bandwidth by more than 99% in testing, which matters enormously for any battery-powered or wirelessly connected device trying to process vision in real time.

A sensor that only reports what moves, instead of everything it sees, mimics how biological retinas already triage visual information before it ever reaches the brain.

What this is actually for

The research team has pointed to compact robotic vision and autonomous navigation as the most immediate applications — systems where a wide field of view and low power or bandwidth needs matter more than raw resolution. A hemispherical sensor can theoretically see nearly all the way around an object with a single component, instead of needing multiple flat cameras stitched together digitally, which is how most robots and drones currently achieve wide-angle vision.

Medical applications, including future visual prosthetics for people with retinal damage, remain a longer-term goal for this broader class of research, though this particular paper focuses on the imaging hardware itself rather than a clinical device. Going from 100 pixels to 367,500 in six years suggests the resolution curve for curved artificial retinas is still climbing fast, and there's no obvious ceiling yet in sight for how far the approach can go.