Professor Yang Yuchao's research group, hailing from the School of Integrated Circuits at Peking University, has published a groundbreaking research paper in the esteemed journal Nature Electronics. In this paper, they propose an innovative hardware solution for Spiking Neural Networks (SNNs), leveraging memristor hybrid-dynamics synaptic integrated arrays. By strategically coupling interface-type volatile memristors with their non-volatile counterparts, the team successfully constructed hybrid synaptic units that support the Fatigue Spike-Timing-Dependent Plasticity (Fatigue STDP) learning rule.

Experimental results showcased that this hardware system operates with remarkable stability across a broad frequency spectrum, ranging from 10Hz to 500kHz. It achieves an impressive energy efficiency of 0.298 TOPS/W, effectively surmounting the technical limitations of traditional SNNs, which often suffer from low learning efficiency and a lack of frequency adaptability in high-frequency noise environments.

The research team further elucidated the fatigue dynamics mechanism, which is primarily governed by reverse oxygen ion diffusion. This discovery was made possible through in-situ transmission electron microscopy. Consequently, the device can exhibit cumulative inhibitory effects when subjected to high-frequency pulses, while still maintaining efficient signal transmission at lower frequencies.

This achievement lays a solid foundation for the development of highly robust and low-power hardware for edge brain-inspired intelligent systems. Moreover, it validates the system's cross-frequency adaptive learning capabilities in practical applications, such as visual tasks and speech recognition.