Researchers fit tens of thousands of artificial brain synapses onto a single chip.
Researchers placed tens of thousands of tiny memristors (memory transistors)—silicon-based components that mimic the human brain’s information-transmitting synapses—onto a single chip which, when run through various tasks, was able to “remember” and reproduce stored images.
While artificial synapse networks currently exist only as software, the MIT team wants to build a hardware equivalent for portable artificial intelligence systems. The results, published this week in the journal Nature Nanotechnology, highlight the new memristor design—ideal for carrying out complex tasks on mobile devices that only supercomputers can handle.
Ultimately, scientists envision memory transistors requiring less chip real estate than conventional options, allowing for more powerful, portable computing devices that don’t even need Wi-Fi. “Imagine connecting a neuromorphic device to a camera on your car, and having it recognize lights and objects and make a decision immediately, without having to connect to the internet,” Jeehwan Kim, associate professor of mechanical engineering at MIT, said in a statement. “We hope to use energy-efficient memristors to do those tasks on-site, in real-time.”
The problem is, existing memristor designs are limited in their abilities. But Kim & Co. found a loophole in the form of metallurgy, the science of melding metals into alloys and studying their combined properties. Traditionally, experts add different atoms to strengthen materials. Kim, however, is not traditional. “We thought, why not tweak the atomic interactions in our memristor, and add some alloying element to control the movement of ions in our medium,” he said. The team settled on copper, which is able to bind with silver and silicon, acting as a sort of stabilizing bridge.
In their first test, engineers recreated a gray-scale image of Captain America’s shield, each pixel matched to a corresponding memristor on the chip, which was able to produce the same crisp image of the shield multiple times.
“We’re using artificial synapses to do real inference tests,” Kim said. “We would like to develop this technology further to have larger-scale arrays to do image recognition tasks. And some day, you might be able to carry around artificial brains to do these kinds of tasks, without connecting to supercomputers, the internet, or the cloud.”