Stanford's "Artificial Synapse" May Lead To Improvements In Brain-Machine Interfaces
New technology from Stanford University is moving beyond devices that simulate a neural network, attempting rather to recreate a neural network with a biocompatible artificial synapse, the part of a neural network that allows neurons to pass chemical signals to another neuron. In a new study, researchers demonstrate the structure’s ability to vastly improve energy efficiency, especially in tech that works with auditory and visual signaling.
A classic computer codes information in a system of 0s and 1s — or two states — but recent research from the Salk Institute in California suggests that a brain cell uses 26 different ways to code, using a fraction of the energy required to power a computer. A traditional computer capable of functioning on the same level as the human brain would require a nuclear power station to run it, said scientists.
Information travels through the brain using structures called synapses, which pass chemical or electrical signals from one neuron to the next. While traditional computing separately processes information and stores it to memory, a synapse learns the information, requiring less energy to pass signals with every use, and passing more information with minimal amounts of power. Recent research has attempted to emulate this structure using artificial components.
“Deep learning algorithms are very powerful but they rely on processors to calculate and simulate the electrical states and store them somewhere else, which is inefficient in terms of energy and time,” said Yoeri van de Burgt, lead author of a paper published in Nature Materials, to Stanford News. “Instead of simulating a neural network, our work is trying to make a neural network.”
Stanford scientists claim that the architecture of their device is unlike anything previously attempted, a design similar to a battery with two thin films and three terminals connected by a saltwater electrolyte. Working like a transistor, one terminal controls the flow of electricity between the other two. Materials used to make the device are all inexpensive and already exist in the brain’s natural chemistry, meaning it may be possible for the synapse to work with live neurons in improved brain-machine interfaces.
“It works like a real synapse but it’s an organic electronic device that can be engineered,” said Alberto Salleo, associate professor of materials science and engineering at Stanford and senior author of the paper. “It’s an entirely new family of devices because this type of architecture has not been shown before. For many key metrics, it also performs better than anything that’s been done before with inorganics.”
Alic Talin, a member of the technical staff at Sandia National Laboratories who participated in the study, told IEEE Spectrum that the technology has “huge implications” for future generations of brain-machine interfaces.
“More and more, the kinds of tasks that we expect our computing devices to do require computing that mimics the brain because using traditional computing to perform these tasks is becoming really power hungry,” said Talin. “We’ve demonstrated a device that’s ideal for running these types of algorithms and that consumes a lot less power.”