Researchers Unlock Missing Piece for an 'Artificial Brain'
New research coming from the National Institute of Standards and Technology might get computers several steps closer to functioning like artificial brains. The NIST team developed a superconducting switch (a "synapse") that 'learns' like its biological counterpart. This switch could ultimately connect processors and store memories within computer systems exactly how the human brain stores information.
Developing brain-like computers has been a staple of science fiction for nearly a hundred years, but neuromorphic engineering became a reality in the late 1980s. It's largely been seen as the future of digital brains for devices like smartphones, computers, and robotics. Coupled with artificial intelligence programming, neuromorphic chips could be the key to faster computing at significantly less power and in more efficient systems.
“The NIST synapse has lower energy needs than the human synapse, and we don’t know of any other artificial synapse that uses less energy,” NIST physicist Mike Schneider said in a statement.
The NIST synapse is a connecting switch between incoming electrical spikes and the signals being output. It works in the same way that a human synapse quickly switches between two brain cells. The NIST creation has a flexible internal design that can be altered based on its experience or environment. The more electrical spikes that fire between the processors, the stronger the connections made by the synapse, the researchers explained. And just like their real counterparts, the artificial synapses both maintain old circuits while creating new ones.
However, unlike a human synapse, the artificial one moves significantly faster than the human brain. A brain cell fires at 50 times per second. The NIST synapse fires 1 billion times per second and at one tenth of the energy needed by the human brain. The researchers measured the energy needed at less than 1 attojoule. That's less energy than what's naturally found at room temperature as background energy.
Ideally, these new synapses would be found in neuromorphic computers that heavily rely on superconductive materials. That would make the entire system more efficient than other electronics which rely on superconductors. Plus, as the researchers point out, superconductive devices already mirror human brain cells in how they transmit signals. But thanks these new synapses, teh missing piece for artificial brains is no longer missing.
The synapse also uses technology familiar to the NIST team called a Josephson junction. These junctions sandwich the superconducting matierals with an insulator as a filling. As Schnieder noted, these junctions include 20,000 manganese and silicon nanoclusters per square micrometer. They gave the researchers the control they needed.
“These are customized Josephson junctions,” he noted. “We can control the number of nanoclusters pointing in the same direction, which affects the superconducting properties of the junction.”
Ultimately, these synapses could play critical roles in making processing data simultaneously a reality. Neuromorphic computers could be the new wave of reality given the increasing need for faster computing at lower energy costs.
The full paper from the NIST can be found in a recent edition of Scientific Advances.
Find out about the cutting-edge world of "mushroom computing" and how it could be used in AI, robots, and monitoring the environment.