Neuromorphic sensory computing

Abstract

The number of sensory nodes in the Internet of everything continues to increase rapidly and generate massive data. The generated information from sensory nodes is much larger than the total collective human sensory throughput. It is quite challenging to send all of the data produced at sensory terminals to the “Cloud” computation center, especially for those time-delay sensitive applications. This situation demands a dramatic increase in the computation near or inside sensory networks. Inspired by biological sensory systems with a high data compression ratio, neuromorphic sensory computing provides a way to efficiently acquire and process a large volume of data from complex environments. Researchers have been investigating emerging materials, devices, circuits, and computing architectures to implement an artificial sensory system with high energy efficiency, speed, and density. Here we summarize the important features of biological systems and their hardware implementations. Electrons and photons are two representative information carriers, in which electron carrier allows high integration density for complex computing and photon carrier has high connectivity, high speed, wide bandwidth, and low power consumption. We overview the electronic and optical neuromorphic sensory computing and hybrid opto-electronic sensory computing, and present advances on multimodal sensory computing and their potential challenges.