FeNN-DMA: A RISC-V SoC for SNN acceleration

Researchers have developed a new, programmable computer chip called FeNN-DMA, specifically designed to run Spiking Neural Networks (SNNs). These brain-inspired networks are highly energy-efficient but poorly suited to standard hardware like GPUs. The new RISC-V-based chip matches the efficiency of specialized accelerators while supporting more complex models, simulating up to 16,000 neurons and 256 million connections. It achieved top accuracy on tasks like digit and tactile recognition.