Artificial synapse network demonstrates plasticity behaviors

Neuromorphic computer chips, which use artificial neurons and synapses designed like those in our brains, could help revolutionize future autonomous computing and artificial intelligence systems. Research groups have developed and demonstrated single artificial synapses, but no one has tested a device that combines multiple synapses.

Sakai et al. have demonstrated an artificial synapse device combining multiple synapses for the first time. The researchers connected multiple single synaptic devices created from previously developed gold nanogap electrodes into an array. The simple design of the electromigrated electrodes was chosen for their ease of fabrication.

The researchers activated the artificial synapses by applying a field emission current to emulate synaptic functions before testing. Testing the connected device, which was structured like an axon linked by two synapses, showed conductance changes in response to a stimulation voltage, mirroring how memory functions in biological synapses.

“In the study we showed that a synaptic device of gold nanogaps can implement significant synaptic functions, notably short-term, long-term, and spike-timing-dependent plasticity,” said author Jun-ichi Shirakashi. “This shows that synaptic learning behavior can be emulated not only by a single gold nanogap, but also by a multiple connected gold nanogap.”

The authors’ results show such devices could be integrated in large-scale systems, which could enable the processing of more complex neuromorphic functions including pattern classification and image recognition. The authors plan to use such gold nanogap devices to continue studying tasks like Pavlovian conditioning and axon-multi-synapse networks with multiple connected nanogaps.

Source: “Multiple connected artificial synapses based on electromigrated Au nanogaps,” by Keita Sakai, Mamiko Yagi, Mitsuki Ito, and Jun-ichi Shirakashi, Journal of Vacuum Science and Technology-B (2022). The article can be accessed at https://doi.org/10.1116/6.0002081 .

This paper is part of the Neuromorphic Materials, Devices and Processing Collection, learn more here .