Energy efficient artificial gustatory system for in-sensor computing

Abstract

In this work, a novel bio-inspired artificial gustatory system is proposed for in-sensor neuromorphic computing. The system is demonstrated using a novel material-engineered compound-semiconductor double-gate ferroelectric tunnel FET. The behaviour of the device as a biosensor and as a spiking neuron is verified individually. Using extensive simulation in Atlas TCAD, the biosensor functionality is validated with I_ON sensitivity of 10⁸. Likewise, the device shows excellent neuronal behaviour with energy consumption of 37 aJ/spike, without using any external circuitry. It also exhibits control over the spiking frequency through amplitude, frequency, and duty cycle of input synaptic current. By cascading the biosensor and neuron, a complete bio-mimicked gustatory system is designed to identify a separate pattern for different tastes. The proposed gustatory system integrates both the devices, and simultaneously performs sensing and spike encoding. The biosensor transduces the pH and dielectric constant of the target biomolecule, corresponding to gustatory neurons present in the taste buds of a biological gustatory system. The neuron performs spike encoding and acts as an input neuron in a classifying network, which corresponds to gustatory cortex of its biological counterpart. The system consumes an average power of 49.58 pW which is ∼1000 × lesser than the state-of-the-art artificial gustatory system, eliminating the need for complex hardware and exorbitant energy consumption. Therefore, the proposed gustatory system provides a highly efficient and compact solution for neuromorphic gustatory sensing and classification, with potential applications in portable, wearable, and implantable devices.