The human brain is an incredibly efficient “computer”—capable of drawing inferences, integrating knowledge, and handling a wide range of tasks such as vision, hearing, language, learning, reasoning, decision-making, and planning. Brain-inspired computing applies the working mechanisms of biological neural networks to computer system design, creating intelligent computing systems that are low-power, highly parallel, and efficient—just like the brain.
On August 2, the State Key Lab of Brain-Machine Intelligence at Zhejiang University unveiled its latest achievement: Darwin Monkey (“Wukong”), a new-generation neuromorphic brain-inspired computer. It is the world’s first dedicated neuromorphic chip-based system of its kind to exceed 2 billion neurons, marking China’s advanced international standing in the field of neuromorphic brain-inspired computing.
Where Did “Wukong” Come From?
The complete “Wukong” system consists of 15 blade servers, each integrated with 64 Darwin III brain-inspired chips. These chips were developed early last year through a collaboration led by ZJU and Zhejiang Lab. A single Darwin III chip supports over 2.35 million spiking neurons and hundreds of millions of synapses, and is equipped with a dedicated instruction set for brain-inspired computation and an online neuromorphic learning mechanism.
Darwin III Brain-Inspired Chip
With these chips, “Wukong” reaches a neuron count close to that of a macaque brain—a significant step toward more advanced brain-like intelligence. Its typical power consumption is around 2000 watts. This achievement marks another major breakthrough for the ZJU brain-inspired computing team since September 2020, when they developed China’s first brain-inspired computer, Darwin Mouse (“Mickey”), which supported over 100 million neurons.
To overcome bottlenecks in inter-chip connection speed and energy efficiency, the team collaborated with the College of Integrated Circuits, Zhejiang University and developed DarwinWafer—a wafer-level ultra-integrated brain-inspired computing chip using 2.5D advanced packaging. This enabled the creation of a System on Wafer (SoW) blade server. The core of the entire SoW blade is a 12-inch wafer integrating 64 bare Darwin III chips, breaking free from the physical limits of traditional reticle-based manufacturing and achieving optimized interconnects at a micro-nano scale.
Neuromorphic Brain-Inspired Server with SoW Integration
What Can “Wukong” Do?
After more than two years of intensive R&D, the team made breakthroughs in several key technologies:
Building a large-scale neuronal interconnection and integration architecture that supports hierarchical, scalable inter-chip connections based on multi-dimensional grid topology;
Developing an adaptive timestep control method that enables task coordination among large-scale neurons;
Utilizing domestic wafer substrate processes and CoWoS-S 2.5D advanced packaging to realize wafer-level brain-inspired chips;
Proposing a hierarchical system resource management framework and designing data swapping strategies for multi-level memory systems to manage and schedule resources across large-scale neuronal systems.
To fully leverage the system’s capabilities, the team also developed a new-generation Darwin brain-inspired operating system. This system uses a layered resource management architecture and incorporates technologies such as load-aware scheduling and dynamic time-slice partitioning. These features enable efficient concurrent scheduling of neuromorphic tasks and dynamic optimization of system resources, taking into account communication bandwidth and task characteristics.
Brain-Inspired Operating System
The team has successfully deployed several intelligent applications on “Wukong.” It is capable of running brain-inspired large models like DeepSeek to perform tasks such as logical reasoning, content generation, and mathematical problem-solving. Leveraging its massive neuronal and synaptic resources, it has also preliminarily simulated animal brains at various scales—including those of C. elegans, zebrafish, mice, and macaques—offering new tools for brain science research.
Simulated Brain Models of C. elegans, Zebrafish, Mouse, and Macaque
The introduction of the “Wukong” brain-inspired computing system represents another breakthrough in neuromorphic computing by the ZJU team. Its large scale, high parallelism, and low power consumption offer a new computing paradigm for existing scenarios:
Providing a new computational foundation for AI development: Brain-inspired computing systems can address the high energy consumption and computational demands of current deep networks and large models, while their unsupervised online learning mechanisms may bring revolutionary progress to AI;
Serving as a natural platform for brain simulation and advancing brain science: “Wukong” can serve as a simulation tool for neuroscientists, offering new experimental methods to explore brain mechanisms and reducing the need for real biological experiments;
Promoting the development of artificial general intelligence (AGI): Human reasoning far surpasses current AI in both capability and efficiency. With its brain-like working mechanism and computational speed exceeding that of the human brain, “Wukong” will provide powerful support for future research in brain-inspired AI.
