NeuCore: A Novel Neuromorphic Processor Architecture With On-Chip Event-Driven Learning

In this brief, we propose a novel resource-friendly neuromorphic processor architecture named “NeuCore” with fully event-driven on-chip Spike Timing Dependent Plasticity (STDP) learning. We focus on the partial clock-driven parameters and optimize them into the event-driven updating mode to reduce memory overhead. Several key techniques such as spike-driven updating values are proposed and well-scheduled to further improve the energy efficiency and implementation efficiency of on-chip learning. The experimental results show that the accuracy of NeuCore reaches 90.04% on DVS128 Gesture datasets. NeuCore achieves $4.44\times $ and $2.5\times $ speed-up in performance and learning throughput, respectively, when compared with the state-of-the-art trace-based work. NeuCore achieves an implementation efficiency of $1.37\times $ - $2.27\times $ compared to state-of-the-art counterbased work.