A pseudo-memcapacitive neurotransistor for spiking neural networks

In this paper, we present a pseudo-memcapacitive neurotransistor by embedding a nonvolatile, abrupt-switching memristor at the gate of an NMOS transistor to emulate neuronal integrate and fire behavior. Neural networks, implementing spikebased computing paradigms on hardware platforms, integrating memristor crossbar arrays on underlying CMOS circuitry, operate similarly as neuronal networks in the human brain, which can significantly improve the time and energy efficiency of standard data processors.We demonstrate that also nonvolatile memristors can be considered to realize neuronal leaky integration and firing functionality including the neuron reset being performed intrinsically by a sufficiently discharged ’membrane’ potential at the gate of a transistor. A versatile, compact and abrupt-switching model of a nonvolatile memristor with built-in cycle-to-cycle variability is proposed, forming a pseudo-memcapacitance along with the gate capacitance and evoking conditional neuronal spike generation depending upon the properties of the input pulse train. The SPICE code of the pseudo-memcapacitive neurotransistor is applied to verify the design parameters that trigger firing. Finally, the envisaged circuit realization of the proposed design is discussed.