Why it matters to you
This artificial synapse could play a pivotal role in projects where computers need to quickly respond to audiovisual input.
From artificial intelligence to machine learning, there is no shortage of projects that use computers to mimic some aspect of the human brain. However, even the most advanced computers struggle to imitate the brain’s natural capacity to process information efficiently.
Now, a team of researchers from Stanford University and Sandia National Laboratories has managed to create an artificial synapse that mimics the way real synapses learn information from the signals they receive. Traditionally, computers would process information and store it into memory as two separate processes, whereas this device creates a memory by processing, which is a more efficient solution.
“It’s an entirely new family of devices because this type of architecture has not been shown before,” said Alberto Salleo, an associate professor of materials science and engineering at Stanford, who also served as the senior author of the artificial synapse paper. “For many key metrics, it also performs better than anything that’s been done before with inorganics.”
More: Meet SyNapse, IBM’s computer chip that can help the blind see, detect diseases, and much more
The artificial synapse is said to be based on a battery design. Three terminals are spaced across two flexible films which are connected by an electrolyte that consists of salty water. The synapse then works as a transistor, with the flow of electricity between two terminals being controlled by the remaining terminal.
It is thought that this technology could one day be used to create a computer that can better imitate the way a human brain responds to auditory and visual stimuli. This would be particularly useful for voice-controlled interfaces and self-driving vehicles that need to process information quickly and accurately.
Only one artificial synapse has been produced so far, but researchers were able to simulate how an array of artificial synapses would work in a neural network by using thousands of measurements taken from experiments on the prototype. The simulated array was able to recognize handwritten single-digit numbers with 93 to 97 percent accuracy, which is considered to be a very encouraging result.