Fully Depleted MOSFET Based Bio-Plausible Synapse for Ultra-Low Energy Applications

Abstract

The basic unit of most of the AI/ML based systems is the neuron. The biological neuron has intriguing capability to process mammoth data in a flash of seconds and that too at extremely low energy overhead in range of few femto-Joules. However, most of the previously proposed electronic synapse lacks this ultra-low energy consuming capability of the neuron. So, in this work, an ultra-low energy synaptic semiconductor device operating in subthreshold conduction region have been demonstrated for real-time artificial intelligence (AI) applications. The proposed device is a fully depleted (FD) metal-oxide-semiconductor field-effect transistor (MOSFET) with charge trapping and de-trapping capabilities for synaptic weight modulation. The proposed device was observed to be $\approx 10^{3}$ times more energy efficient than previous electronic synapses.