Zhejiang University's Brain-Machine Intelligence National Key Laboratory officially launched the world's largest brain-like computer "Darwin Monkey (Wukong)", marking a breakthrough that signifies a new development stage in brain-like computing technology. This computer has more than 2 billion spiking neurons and 100 billion synaptic connections, for the first time approaching the scale of a macaque monkey's brain in an engineering system, opening up new paths for artificial intelligence and neuroscience research.
Technical Architecture: Computing Revolution Driven by Darwin3 Chips
The core of Wukong computer is 960 Darwin3 brain-like chips jointly developed by Zhejiang University and Zhijiang Laboratory. These chips use a spiking neural network architecture, transmitting information through discrete spike signals, more realistically simulating the working mechanism of biological neurons. Each chip supports up to 2.35 million spiking neurons, far exceeding the processing capability of traditional computing chips.
The design philosophy of Darwin3 chips is based on the core views of modern neuroscience — neurons as independent functional units of the brain. Through precise simulation of artificial neurons and synapses, these chips can achieve decision-making, learning, and memory functions similar to the human brain, providing a new technical path for efficient information processing.
The chips use an event-driven architecture, activating relevant modules only when receiving spike signals or executing computing tasks. This design makes the energy consumption per synaptic operation only 5.47 picojoules, showing excellent energy efficiency. Compared with traditional computing architectures, this low-power characteristic makes large-scale neural network deployment possible.
System Integration: Advanced Packaging Technology for Efficient Interconnection
In terms of system integration, Wukong computer adopts innovative 2.5D packaging technology. The research team directly packaged 64 Darwin3 chips on a single 12-inch wafer, breaking the physical limitations of traditional packaging methods. This design allows all interconnections between chips to be completed at the micro- and nanoscale, greatly improving data transmission speed and reducing communication power consumption.
The entire system consists of 15 blade-style neuromorphic servers, each equipped with 64 Darwin3 chips. Through this modular design, Wukong computer achieves high scalability and flexibility, allowing configuration adjustments according to different application needs.
Despite its massive scale of 2 billion neurons, the entire system's power consumption is kept within 2000 watts. This power level represents a qualitative leap compared to traditional supercomputers, laying the foundation for the practical deployment of brain-like computing.
Software Ecosystem: Darwin Brain-Like Operating System
To fully leverage hardware potential, the research team developed the配套 Darwin brain-like operating system. This system uses a hierarchical resource management architecture, combining task load-aware scheduling and dynamic time slice allocation mechanisms to achieve efficient parallel execution of neuromorphic tasks.
The intelligent scheduling function of the operating system dynamically allocates computing resources and communication bandwidth based on the characteristics of different tasks and the system's resource status. This optimization mechanism ensures that the system maintains efficient operation when processing complex neural network tasks.
The system also supports various spiking neuron models and learning rules, providing researchers with a flexible experimental platform. Through standardized interfaces and development tools, users can easily deploy and test various brain-like algorithms.
Application Practice: Technical Validation from Theory to Reality
Wukong computer has already demonstrated powerful capabilities in multiple practical application scenarios. According to Professor Pan Gang, the laboratory director, the team has successfully deployed several intelligent applications, including DeepSeek, on this system, completing complex tasks such as logical reasoning, content generation, and mathematical problem solving.
More scientifically valuable is that Wukong computer can simulate animal brains of different neuronal scales. From simple nematodes to complex macaque brains, this system provides neuroscientists with unprecedented experimental tools. This cross-species brain simulation capability has the potential to reduce reliance on live animal experiments, opening up more humane and efficient research approaches for brain science.
By precisely simulating the neural network structures of different animals, researchers can gain a deeper understanding of how the brain works and explore the fundamental principles of neural information processing. This is significant for understanding human cognitive functions and treating neurological diseases.
International Status: Leading the Development Direction of Brain-Like Computing
The release of Wukong computer has placed China in an important international leading position in the field of brain-like computing. Previously, the largest neuromorphic computing system in the world was Intel's Hala Point, with 1.15 billion neurons. Wukong computer's scale of 2 billion neurons is almost twice that, achieving a significant technological leap.
From a technical development perspective, the Zhejiang University team has had continuous technical accumulation in the field of brain-like computing. The "Darwin Mouse (Mickey)" released in 2020 had 120 million neurons, and now Wukong computer has 2 billion neurons, representing an increase of nearly 17 times in neuronal scale, demonstrating a strong trend of rapid technological iteration.
This achievement not only demonstrates China's innovation strength in cutting-edge technology fields but also provides an important reference for the development of global brain-like computing technology. With the continuous maturation of related technologies, it is expected that more countries and institutions will invest in research in this area.
Future Prospects: Technological Innovation with Dual Value
Professor Pan Gang has given a clear positioning for the future applications of Wukong computer. This system will play an important role in two directions: one is as a new computing foundation for the development of artificial intelligence, providing an efficient hardware platform for brain-like AI algorithms; the other is as a powerful brain simulation tool for neuroscience research, helping scientists better understand the mechanisms of the brain.
In the field of artificial intelligence, Wukong computer is expected to promote the development of brain-like algorithms, enabling more intelligent and human-like AI systems. By simulating the way the human brain processes information, these systems may demonstrate better adaptability and generalization ability when handling complex cognitive tasks.
In the field of neuroscience, this system provides an unprecedented experimental platform for studying brain function. By precisely controlling neural network parameters, researchers can verify various theories of brain science and explore the neural basis of cognitive functions.
As technology continues to develop, Wukong computer is expected to play an important role in multiple application areas such as intelligent robots, autonomous driving, and smart healthcare, promoting the advancement of artificial intelligence technology to a higher level. The emergence of this achievement marks that brain-like computing technology is moving from the laboratory to practical applications, laying a solid foundation for building a more intelligent future society.
References: https://interestingengineering.com/science/china-world-largest-scale-brain-computer