Strained bipolar charge plasma transistor as a high speed LIF neuron

Abstract

For Neuromorphic computing and bio realistic dynamics, distinct CMOS devices have been carried out. But most devices used for developing artificial synapses and mimicking the LIF neuronal dynamics suffer from high power dissipation, low operating speed, and high cost of hardware implementation. In this work, leaky integrate and fire neural are implemented using a strained bipolar charge plasma transistor based on a floating body mechanism. For the first time, strained BCPT neuron is demonstrated to mimic biological behavior which provides an inherently low-energy, cost-effective, and easy implementation of the neuron. The strained BCPT based neuron exhibits maximum spiking energy of 196 aJ which is 1.53 $\times$ 10${}^5$, 1.79 $\times$ 10${}^5$, 5.1 $\times$ 10${}^4$, 2.9 $\times$ 10${}^4$, 32.1, 16.4, 1.28 $\times$ 10${}^3$, 918, 5820, 14.8 times less as compared to phase change CMOS, SOI CMOS, PCMO CMOS, Biristor, Bulk FinFET, PD SOI CMOS based on BTBT, FBFET, LBIMOS, DGJLFET, and Silicon nanowire. This work also investigates the effect of temperature, base width, germanium mole fraction, and metal electrode work function. The collector potential of 0.30 V is enough to produce a threshold spike current of 8 $rm\mu$A/$\mu m$ which is 9.33, 10, 2.66, 5, 6.66, and 1.33 times less as compared to SOI CMOS, Bulk FinFET, Si NIPIN, PD SOI MOS, LBIMOS, and DGJLFET, respectively. Here, the strained BCPT floating body is responsible for the accumulation of holes generated by impact ionization (II). The proposed neuron device demonstrates the basic function of LIF dynamics at 12 GHz frequency using the SILVACO 2D TCAD simulation tool. Hence, this work provides the possibility of easy fabrication of highly-integrated SNN at high operating speed and low-energy consumption.