Dataflow-Driven Neuromorphic Architectures for Edge AI: Theory, Design, and Applications
Neuromorphic computing has emerged as a promising paradigm for energy-efficient and scalable machine learning (ML) at the edge. This work provides an integrated framework that connects theoretical foundations of neural networks with bio-inspired architectural and system-level realizations, emphasizing dataflow strategies as a critical determinant of efficiency and scalability. We analyze key architectural innovations, including memory-centric layouts and dataflow-optimized implementations, and evaluate their impact on resource consumption, latency, and robustness. At the system level, we examine edge AI applications that showcase how architecture-dataflow co-design enables high efficiency, real-time operation, and adaptability. By synthesizing theory, design, and application, this paper offers a unifying perspective on how dataflow principles shape the performance and scalability of next-generation neuromorphic and edge AI platforms.
