Emerging Neuromorphic Computing for Edge AI Application: A Systematic Literature Review

Abstract

Edge AI (Artificial Intelligence) and Neuromorphic computing are two intersected concepts that have significantly impacted in recent years. This is due to the ability of neuromorphic computing to mimic the functionality of the brain for processing information in an energy-efficient and highly parallel way. Likewise, Edge AI refers to deploying AI algorithms or models directly on an edge platform rather than relying on servers or cloud platforms. Thus, to accomplish this, neuromorphic computing and edge AI are combined due to the parallel processing ability of neuromorphic computing, which aligns well with edge AI applications' requirements compared to the traditional von Neumann architecture. Despite its promise for edge AI, very few studies have dealt with neuromorphic computing for edge AI due to limited resources and the unavailability of hardware and software tools, which hinder the progress of this realm of research. Therefore, the current SLR focuses on reviewing studies emphasizing neuromorphic computing for Edge AI, differences in conventional and neuromorphic computing, different chips used for neuromorphic computing, and applications of neuromorphic computing for edge AI. Moreover, the present SLR has scrutinized 25 papers based on inclusion and exclusion criteria. From the obtained 25 papers, different challenges are depicted, and future recommendations for overcoming these shortcomings are provided.