Is Neuromorphic Computing the Key to Power-Efficient Neural Networks: A Survey

Today, we are witnessing a technological revolution at a scale that was unimaginable just a decade ago. The advent of artificial intelligence (AI) in modern industry theoretically allows unlimited growth; however, in reality, the dreaded power-wall problem in the parallel computing paradigm limits us from exploiting the true potential of AI. Modern neuromorphic accelerators present a lucrative alternative to the traditional artificial neural network (ANN) accelerators for deep learning (DL) due to their promise of ultra-low-power operation. The Spiking Neural Networks (SNN) form the heart of neuromorphic accelerators; the SNN aims to replicate the highly energy-efficient process at work in our brains. In this chapter, we explore the current work and the limitations of neuromorphic computing in AI systems and our future together with this technology.