Exploiting Steep Sub-Threshold Swing of Tunnel FET for Energy-Efficient Leaky Integrate-and-Fire Neuron Model

In this work, a single transistor (1-T) leaky-integrate-and-fire neuron based on band to band tunneling mechanism is proposed with significant improvement in energy consumption and integration density. For the first time, Tunnel FET's forward transfer characteristics with steep sub-threshold swing have been exploited to emulate neuronal behavior. Using calibrated simulation, it is validated that the proposed device is able to imitate the spiking behavior of a biological neuron accurately without using external circuitry. Since the underlying mechanism is the modulation of a tunneling barrier, the proposed LIF neuron shows significantly lower energy consumption of 750 fJ/spike, which is at least ∼10 folds lesser than previously reported 1-T neurons. To verify the applicability, the proposed neuron has been explored for reconfigurable threshold logic gates to implement various linearly separable Boolean functions including OR, AND, NOT, NOR, and NAND. Finally, a multilayer SNN has been designed that confirms the image recognition ability of the proposed neuron with an accuracy of 96.27%. The proposed neuron provides a solution to implement highly scalable and energy efficient threshold logic circuits for future neuromorphic computing.