Process-Voltage-Temperature Variability Estimation of Tunneling Current for Band-to-Band-Tunneling-Based Neuron

Compact and energy-efficient synapse andneurons are essential to realize the full potential of neuro-morphic computing. In addition, a low variability is indeedneeded for neurons in deep neural networks for higheraccuracy. Further, process (P), voltage (V), and temper-ature (T) (PVT) variation are essential considerations forlow-power circuits as performance impact and compensa-tion complexities are added costs. Recently, band-to-bandtunneling (BTBT) neuron has been demonstrated to operatesuccessfully in a network to enable a liquid state machine(LSM). A comparison of the PVT with competing modesof operation (e.g., BTBT versus subthreshold and abovethreshold) of the same transistor is a critical factor inassessing performance. In this work, we demonstrate thePVT variation impact on the BTBT regime and benchmarkthe operation against the subthreshold regime (SS) andON-regime (I-ON) of partially depleted silicon-on-insulatorMOSFET. It is shown that the ON-state regime offers thelowest variability but dissipates higher power, hence notusable for low-power sources. Among the BTBT and SSregimes, which can enable the low-power neuron, the BTBTregime has shown similar to 3xvariability reduction (sigma(ID)/mu(ID))compared to the SS regime, considering the cumulativePVT variability. The improvement is due to the well-knownweakerP,V, andTdependence of BTBT versus SS.We show that the BTBT variation is uncorrelated withmutually correlated SS and I(ON)operation-indicating itsdifferent origin from the mechanism and location perspec-tives. Hence, the BTBT regime is promising for low-current,low-power, and low device-to-device (D2D) variabilityneuron operation