A recipe for cooking up more effective artificial neurons

As artificial intelligence, or AI, programs become more complex, they’re running into the energy and efficiency limits of standard, and even super, computers. For the next generation of AI, researchers are looking to neuromorphic computers, which are inspired by the brain’s system of neurons and synapses. But first, they need to develop more effective artificial neurons and synapses.

Bisquert studied the fundamental material and device properties essential to create spiking neurons, which transmit information only when an electrical charge threshold has been exceeded, much like the neurons in our brains. At a minimum, a neuron system requires a capacitor, a chemical inductor, and a negative resistance, but the components can be integrated naturally into the physical response of the device, instead of built from separate circuit elements, the author reported.

In identifying the working conditions for smooth oscillations in the response of spiking neurons, Bisquert also found the neurons can be built in many ways. He thus proposed a methodology for assessing the material and device characteristics required for creating a spiking neuron.

“I think this paper, which combines different approaches, will be of interest to researchers in many different communities, such as material sciences, computational sciences, and those studying complex dynamic systems,” the author Juan Bisquert said. “I hope this paper will give people a better idea of the underlying methods needed for creating spiking neurons and apply it to new materials.”

Bisquert is continuing to study artificial neurons through a 2.5 million euro ERC Advanced Grant through the European Union and is organizing a conference on the topic for February 2024.

Source: “Device physics recipe to make spiking neurons,” by Juan Bisquert, Chemical Physics Reviews (2023). The article can be accessed at https://doi.org/10.1063/5.0145391 .