Hybrid Neuromorphic Systems: An Algorithm-Application-Hardware-Neuroscience Co-Design Perspective: Invited Special Session Paper

Abstract

Spiking Neural Networks (SNNs) are considered to be the third generation of artificial neural networks due to its unique temporal, event-driven characteristics. By leveraging bio-plausible spike-based computing between neurons in tandem with sparse on-demand computation, SNNs can demonstrate orders of magnitude power efficiency on neuromorphic hardware in contrast to traditional Machine Learning (ML) methods. This paper reviews recent developments in the domain of neuromorphic SNN algorithms from an overarching system science perspective with an end-to-end co-design focus from algorithms to hardware and applications. The paper outlines opportunities at designing hybrid neuromorphic platforms where leveraging benefits of both traditional ML methods and neuroscience concepts in the training and architecture design choice can actualize SNNs to their fullest potential.