Nat. Commun. 9, 4661 (2018)

Neuromorphic networks built from artificial neurons offer a new form of energy-efficient brain-like computing. So far, however, the demonstrations of artificial neurons have been limited to integrate-and-fire behaviours falling far behind the capabilities of biological neurons in terms of computational complexity and dynamics. Now, Yi et al. report neuron circuits emulating single tonic, phasic and mixed-mode excitabilities and experimentally observe 23 types of biological neuron spiking behaviours.

As a first step, the researchers develop a scalable electroforming-free active memristor made of VO2 on a SiNx-coated silicon substrate, with high endurance and low device-to-device variation. Apart from the VO2 active memristors, the basic circuit topology of the artificial neuron consists of other two circuit elements — a capacitor and a load resistor. By customizing the passive circuit components and operation parameters, they can achieve all three classes of spiking behaviour, including tonic, mixed-mode and phasic spiking. In particular, the newly demonstrated phasic excitability can be realized simply by replacing the load resistor with a capacitor. Finally, the neurons exhibit input-noise-sensitive stochastically phase-locked firing, the behaviour commonly observed in biological neurons.