Advances and Trends in Neuromorphic Computing Chips: A Case Study of the Darwin Series

The traditional von Neumann architecture faces significant bottlenecks known as the “memory wall” and “power wall”. In contrast, neuromorphic computing, by emulating the topological structure and information processing mechanisms of biological neural networks, constructs an integrated architecture that fuses memory and computation. This approach supports an event-driven sparse computing and communication mechanisms, providing an innovative technological pathway to overcome the limitations of conventional computing paradigms and achieve energy-efficient general-purpose intelligence. This article systematically reviews the technological advancements in neuromorphic computing chips, with particular emphasis on analyzing the innovative architectural breakthroughs in Zhejiang University’s Darwin series chips (especially Darwin 3). Furthermore, it provides insights into future development trends in this field.