Ultra-Low-Power Edge Intelligence: Green AI Algorithms and Hardware Co-Design

Abstract This study investigates ultra-low-power edge intelligence as a co-design problem spanning learning algorithms, compilation/runtime systems, and hardware microarchitecture. The central claim is that energy-efficient edge AI emerges from joint optimization of (i) model structure and numerical precision, (ii) data movement across memory hierarchies, and (iii) accelerator-aware execution that minimizes stalls, redundant reads, and peripheral overheads. Accordingly, the proposed framework fo