Neuron device based on single transistor latch for highly scalable neuromorphic hardware

Abstract

Neuromorphic hardware, inspired by the structure and principles of the human brain, can greatly reduce the energy consumption of the AI functions by removing the von Neumann bottlenecks with massive parallel-ism. In particular, a spiking neural network (SNN), which combines spike transmissions with spatiotemporal processing, has received considerable attention due to its superior energy efficiency. To realize the SNN in hardware, an artificial neuron that performs a leaky integrate-and-fire (LIF) function is required. Up to now, CMOS circuit-based artificial neurons have been used, but they have limitations on chip density due to the use of heavy hardware components such as operational amplifiers (op-amps). In this study, a novel single transistor neuron (1T-neuron) based on a single transistor latch (STL) phenomenon is proposed with a com-plete neuromorphic system, by co-integration with artificial synaptic devices. It has great benefits in terms of CMOS-compatibility compared to other device-based artificial neurons. Furthermore, novel artificial sensory neuron devices, such as artificial visual, tactile, olfactory, and gustatory sensory neurons, are demonstrated for in-sensor computing to reduce the cost of the hardware and energy needed in the sensory system.