ReLANCE: A Resource-Efficient Low-Latency Cortical Neural Acceleration Engine
This new brain-inspired chip could make edge AI 2.85x faster and cheaper.
Deep Dive
Researchers have unveiled ReLANCE, a new cortical neural acceleration engine for edge AI. Its novel RCHH neuron model achieves a 35.2% speed improvement and a 24.5% reduction in hardware resources (LUTs) compared to current state-of-the-art designs, while also showing a 70% improvement in accuracy (NRMSE). The full Cortical Neural Pool architecture delivers 2.85x higher throughput (12.69 GOPS) than a comparable DNN engine with only a 0.35% accuracy drop on MNIST.
Why It Matters
It enables powerful, efficient AI directly on small devices, bypassing the cloud for faster, cheaper, and more private applications.