Neuromorphic Engineering for Hardware Computational Acceleration and Biomimetic Perception Motion Integration

Abstract

In the tide of artificial intelligence evolution, the demand for data computing has exploded, and the von Neumann architecture computer with separate memory and computing units require cumbersome data interaction, which leads to serious degradation in performance and efficiency. Biologically inspired neuromorphic engineering performs digital/analog computations in memory, with massive parallelism and high energy efficiency, making it a promising candidate to get out of the woods. Memristive device-based artificial synapses and neurons are building blocks to form hardware neural networks for computing acceleration. In addition, it enables the implementation of integrated bionic perception and motion systems to mimic the human peripheral nervous system for information sensing and processing. Herein, the biological basis and inspiration are described first, and the memristive synapses and circuit-emulation neurons used for neuromorphic engineering are addressed and evaluated as well as the mechanisms. The computational acceleration and bionic perception motion integration of neuromorphic systems are discussed. Finally, the challenges and opportunities for neuromorphic engineering to accelerate computation and enrich biomimetic perception motion functions are prospected, and it is hoped that light is shed on future advances.