Felix Heide, Assistant Professor of Computer Science. Photo by Sameer A. Khan/Fotobuddy
Aug. 30, 2022
An ultra-compact camera can produce crisp, full-color images on par with a conventional compound camera lens 500,000 times greater in size.
Micro-sized cameras have tremendous potential to spot problems in the human body and enable sensing for super-small robots, but previous approaches capture fuzzy, distorted images with limited fields of view.
This new system, the size of a coarse grain of salt, combines the camera’s hardware and computational processing to create powerful imaging capability.
While a traditional camera uses a series of curved glass or plastic lenses to bend light rays into focus, the new optical system relies on a technology called a metasurface, produced much like a computer chip. Just half a millimeter wide, the metasurface is studded with millions of cylindrical posts, called nanoantennas, each roughly 300 times thinner than a strand of human hair.
To design such a large array of nanoantennas, the research team developed an AI method that would efficiently handle the design and characterization of all of these elements.
The new lens could enable minimally invasive endoscopy with medical robots. Arrays of thousands of such cameras could create full-scene sensing, turning surfaces — such as the back of a smartphone — into cameras. These metasurfaces are more cost-effective to produce than conventional lenses, and they take up significantly less space.
Another exciting component is the potential to perform computations in the same system. This could also have applications in privacy, wherein the chip could take a photo containing optically coded information.
“By letting computers learn lenses, computer science can bring unprecedented possibilities and an entirely different design space into the field of optical design.” — Felix Heide
Co-inventors:
Arka Majumdar, University of Washington; Graduate Student Ethan Tseng, Princeton University; Graduate Students Shane Colburn, Luocheng Huang and James Whitehead at University of Washington; Seung-Hwan Baek, POSTECH Computer Graphics Lab
Development status:
Patent pending.
Funding:
National Science Foundation, Defense Advanced Research Projects Agency, UW Reality Lab,
