Bringing Dynamic Sparsity to the Forefront for Low-Power Audio Edge Computing: Brain-inspired Approach for Sparsifying Network Updates

Dynamic sparsity is intrinsic to biological computing and is key to its extreme power efficiency. Edge computing systems can improve their energy efficiency and reduce response latency by exploiting this neuromorphic principle. The neuromorphic approach for the extraction of acoustic features replaces conventional ADC and DSP with biological cochlea-inspired filters and event generators implemented in mixed-signal circuits. The resulting sparse feature events drive inference in dynamic-sparsity-aware neural network accelerators to reduce computational load and memory access. The demonstration of edge keyword spotting shows the dynamic savings in power. Exploiting dynamic sparsity at all levels will be the next step toward the design of intelligent devices for the edge.