Researchers from the University of Central Florida (UCF) have built a device for AI that replicates the retina of the eye. The research could lead to cutting-edge AI technology that would be able to immediately recognize what it sees, such as automatic descriptions of photos taken with a camera or cell phone. The technology could also be used in robots and self-driving cars.
A recent study published in ACS Nano describes the technology, which also performs better than the eye in terms of the range of wavelengths it can perceive, from the ultraviolet to the visible to the infrared spectrum.
Its ability to blend three different operations into one further contributes to its uniqueness. Currently available smart-imaging technologies such as those found in self-driving cars require separate data processing, memory, and sensing.
By integrating these three processes, UCF designed a device that is much faster than existing technologies, according to the researchers. With hundreds of devices on a one-inch-wide chip, the technology is also quite compact.
“It’s going to change the way AI is implemented today,” said Tania Roy, principal investigator of the study’s paper and an assistant professor in UCF’s Department of Materials Science and Engineering and Center for Nanoscience and Technology, “Today, everything is discrete components and runs on traditional hardware. And here we have the ability to use a single device for in-sensor computing on a small platform.”
The technology extends previous work by the research team, which created brain-like devices, which could enable AI to work in remote areas and spaces.
We have devices that behave like synapses in the human brain, but we still don’t provide images to them directly,” Roy said. Now, by adding image-sensing capabilities to them, we have synapse-like devices that function like ‘smart pixels’ in a camera by simultaneously sensing, processing and recognizing images.”
Molla Manjurul Islam ’17MS, first author of the study’s paper and a doctoral student in UC’s Department of Physics, noted that for self-driving vehicles, the device’s versatility will make it safer to drive in a range of conditions, including at night.
“If you’re riding in your self-driving car at night and the car’s imaging system only works at a specific wavelength, such as a visible wavelength, it won’t see what’s in front of it. But in our case, with our device, it can actually see the whole situation,” Islam described, “There’s no reported device like this that works in both the ultraviolet range and visible wavelengths as well as infrared wavelengths, so that’s the most unique selling point of the device. “
The key to the technology is nanoscale surface engineering made of molybdenum disulfide and platinum ditelluride, which can help enable multi-wavelength sensing and memory. The work was done in close collaboration with YeonWoong Jung, an assistant professor in UCF’s Center for Nanoscience Technology and the Department of Materials Science and Engineering.
The researchers tested the device’s accuracy by having it sense and recognize images of mixed wavelengths – an ultraviolet number “3” and an infrared portion, the mirror image of the number, put together to form an “8. They demonstrated that the technology can distinguish these patterns and identify them as a “3” in ultraviolet and an “8” in infrared.
Adithi Krishnaprasad ’18MS, a doctoral student in UCF’s Department of Electrical and Computer Engineering and co-author of the paper on the study, said, “We got 70 to 80 percent accuracy, which means they have a very good chance of being able to implement it in hardware.”
The researchers say the technology could be available within the next five to 10 years.
