DESIGN AND OPTIMIZATION OF NEUROMORPHIC HARDWARE ACCELERATORS FOR ENERGY-EFFICIENT EDGE AI INFERENCE A CROSS-LAYER CO-DESIGN APPROACH INTEGRATING SPIKING NEURAL NETWORKS, MEMRISTIVE ARCHITECTURES, AND HARDWARE-AWARE MODEL COMPRESSION

As Edge AI continues to grow, energy-efficient inference becomes increasingly essential for real-time applications in constrained environments. This paper presents a cross-layer co-design framework integrating Spiking Neural Networks (SNNs), memristive-based neuromorphic hardware, and hardware-aware model compression techniques to optimize edge AI performance. By unifying algorithm-hardware co-optimization, we achieve significant reductions in latency and power consumption without compromising accuracy. Experimental evaluations on representative workloads demonstrate that our framework provides up to 85% energy savings compared to traditional architectures, while maintaining competitive inference accuracy.