Spike-based dynamic computing with asynchronous sensing-computing neuromorphic chip

Neuromorphic computing, inspired by brain neurons and synapses and implemented via spiking neural networks (SNNs) on neuromorphic chips, promises energy-efficient machine intelligence. A central challenge is leveraging high-level brain dynamic mechanisms to realize energy advantages. This work presents an application-oriented algorithm–software–hardware co-designed neuromorphic system. We design and fabricate an asynchronous sensing–computing system-on-chip (SoC) named Speck, featuring a processor resting power of 0.42 mW and always-on behavior where no input consumes no running energy. We identify a common but overlooked “dynamic imbalance” in SNNs—similar firing responses across diverse inputs—and develop an attention-based dynamic framework to enable input-dependent energy variance. Together, the system demonstrates real-time operation with total power as low as 0.70 mW and ultra-low latency. Results highlight the potential of asynchronous, event-driven, sparse, and dynamic neuromorphic computing.

Man Yao, Ole Richter, Guangshe Zhao, Ning Qiao, Yannan Xing, Dingheng Wang, Tianxiang Hu, Wei Fang, Tugba Demirci, Michele De Marchi, Lei Deng, Tianyi Yan, Carsten Nielsen, Sadique Sheik, Chenxi Wu, Yonghong Tian, Bo Xu, Guoqi Li