Scientists build artificial neurons that work like real ones

"Our brain processes an enormous amount of data," says Shuai Fu, a graduate student in electrical and computer engineering at UMass Amherst and lead author of the study published in Nature Communications. "But its power usage is very, very low, especially compared to the amount of electricity it takes to run a Large Language Model, like ChatGPT."

The human body operates with remarkable electrical efficiency -- more than 100 times greater than that of a typical computer circuit. The brain alone contains billions of neurons, specialized cells that send and receive electrical signals throughout the body. Performing a task such as writing a story uses only about 20 watts of power in the human brain, whereas a large language model can require more than a megawatt to accomplish the same thing.

Engineers have long sought to design artificial neurons for more energy-efficient computing, but reducing their voltage to match biological levels has been a major obstacle. "Previous versions of artificial neurons used 10 times more voltage -- and 100 times more power -- than the one we have created," says Jun Yao, associate professor of electrical and computer engineering at UMass Amherst and the paper's senior author. Because of this, earlier designs were far less efficient and couldn't connect directly with living neurons, which are sensitive to stronger electrical signals.

"Ours register only 0.1 volts, which about the same as the neurons in our bodies," says Yao.

There are a wide range of applications for Fu and Yao's new neuron, from redesigning computers along bio-inspired, and far more efficient principles, to electronic devices that could speak to our bodies directly.

"We currently have all kinds of wearable electronic sensing systems," says Yao, "but they are comparatively clunky and inefficient. Every time they sense a signal from our body, they have to electrically amplify it so that a computer can analyze it. That intermediate step of amplification increases both power consumption and the circuit's complexity, but sensors built with our low-voltage neurons could do without any amplification at all."

The secret ingredient in the team's new low-powered neuron is a protein nanowire synthesized from the remarkable bacteria Geobacter sulfurreducens, which also has the superpower of producing electricity. Yao, along with various colleagues, have used the bacteria's protein nanowires to design a whole host of extraordinary efficient devices: a biofilm, powered by sweat, that can power personal electronics; an "electronic nose" that can sniff out disease; and a device, which can be built of nearly anything, that can harvest electricity from thin air itself.

This research was supported by the Army Research Office, the U.S. National Science Foundation, the National Institutes of Health and the Alfred P. Sloan Foundation.