Cyborg dreams move closer to reality with low-power artificial neuron

UMass Amherst research promises better bioelectronic communication

Scientists affiliated with the University of Massachusetts Amherst have developed an artificial neuron that can communicate efficiently with biological neurons, a research advance expected to accelerate the development of bioelectronic devices and interfaces.

Neurons are nerve cells that transmit and receive electrical signals as part of the human central nervous system, which includes the spinal cord and brain. Various artificial neurons have been developed to facilitate connections between electronic and biological systems (like a neuromorphic perception system that can monitor the curvatures of fingers) or to make electronics more efficient. But they've required significantly more power than their biological counterparts.

Sign in to sound off

Register for The Register's Forums here.

The UMass Amherst artificial neuron works at power levels associated with biological neurons – it operates at 0.1 volts, which is about 10 times less voltage than prior artificial neurons and about 100 times less power, according to the UMass Amherst news service. It also matches the parameters of biological neurons in terms of signal amplitude, spiking energy, temporal features (neuronal firing patterns), and frequency response – characteristics that simplify communication.

Reducing the electrical inefficiency of semiconductors compared to the human brain has been a longstanding goal for computer scientists. At the Stanford Institute for Human-Centered Artificial Intelligence (HAI) conference last year, Surya Ganguli, an associate professor of applied physics at Stanford, emphasized the need "to rethink the entire technology stack from electrons to algorithms in order to really go from megawatts to watts."

The UMass Amherst researchers – Shuai Fu, Hongyan Gao, Siqi Wang, Xiaoyu Wang, Trevor Woodard, Zhien Wang, Jing Kong, Derek R. Lovley, and Jun Yao – have taken a step in that direction, as described in a paper published in Nature Communications.

"We demonstrate that an artificial neuron can connect to a biological cell to process cellular signals in real time and interpret cell states," the authors state in their paper. "These advancements enhance the potential for constructing bio-emulated electronics to improve bioelectronic interface and neuromorphic integration."

Jun Yao, associate professor of electrical and computer engineering, told the UMass Amherst news service that his team's artificial neuron could help make wearable electronic sensing devices less bulky by avoiding the need to amplify biological signals to a level that can be processed electronically.

"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," Yao explained.

The key to the power-efficient artificial neuron – a memristor – is the integration of a protein nanowire harvested from a bacterium called Geobacter sulfurreducens that can produce electricity.

Yao and his colleagues have used these nanowires in several other research projects, including a biofilm that can produce energy from sweat, a disease-sniffing electronic nose, and a technique for creating nanopores within materials to enable the harvesting of energy from the air. ®