A Highly Scalable Junctionless FET Leaky Integrate-and-Fire Neuron for Spiking Neural Networks

Abstract

In this article, a highly scalable and CMOS compatible double-gate junctionless field-effect transistor (DG-JLFET)-based leaky integrate-and-fire (LIF) neuron is presented for the sub-20 nm gate length which is the smallest reported until now. Using well-calibrated 2-D TCAD simulations, we demonstrated that DG-JLFET LIF is able to mimic biological neuronal behavior. The DG-JLFET LIF neuron shows a low threshold voltage ( -0.31 V) for firing a spike and requires 1.14 pJ of energy per spike which is ~ 32× less than partially depleted silicon-on-insulator (PD-SOI) MOSFET LIF neuron. The proposed neuron needs only 0.4 V of supply voltage that is ~ 7×, 7.5×, 5×, and 2× lower as compared to its counterpart PD-SOI MOSFET, FinFET, L-shaped bipolar impact ionization MOS (LBIMOS), and Si NIPIN Diode-based LIF neurons, respectively. In addition, at a gate length of 10 nm the DG-JLFET LIF requires 0.07 pJ of energy per spike, which is ~ 500×, ~ 642×, and ~ 86× lower than that of the PD-SOI MOSFET, single MOSFET and Biristor based LIF neurons, respectively. Moreover, the DG-JLFET LIF neuron shows spiking frequency in the range of megahertz, which is ~5 orders high compared to the biological neuron. The absence of metallurgical junctions in DG-JLFET eases the fabrication complexity and cuts down the thermal budget requirement.