Zhejiang University and Zhijiang Lab recently developed China's first brain-like computer - the Darwin Mouse, which is also the world's largest brain-like computer in terms of neurons.
It contains 792 Darwin second-generation brain-like chips developed by Zhejiang University, supporting 120 million impulse neurons and nearly 100 billion synapses, which is the same size as the number of neurons in the mouse brain.
Its typical operating power consumption is only 350-500 watts.
The research team has also developed DarwinOS, an operating system for brain-like computers, which enables effective management and scheduling of brain-like computer hardware resources to support the operation and application of brain-like computers.
At today's press conference, Director of Zhijiang Lab and Deputy Secretary of the Party Committee of Zhejiang University said that the two research teams will develop a larger-scale neuronal brain-like computer based on China's proprietary brain-like chips.
The team will also research the basic brain-like software system to support its operation and development, and gradually realize open source and open source, to promote the development of new brain-like computing technology in China, he said.
Current computer development mostly chooses the John von Neumann architecture known for numerical computation, that is, the way to add, subtract, multiply, and divide numbers to carry out information architecture.
As Moore's theorem is gradually failing, the limitations of the von Neumann architecture are becoming more and more apparent, and problems such as storage walls, power walls, and intelligence enhancement are making current computer development face major challenges.
Scientists around the world have set their sights on mimicking the biological brain, the original dream, to develop new computing technologies that mimic the structure and computational mechanisms of the human brain in order to achieve high levels of computing efficiency and intelligence.
The biological brain is able to naturally produce different intelligent behaviors during interaction with the environment, including speech comprehension, visual recognition, decision-making tasks, and operational control, and consumes very little energy.
In nature, many insects with far fewer than a million neurons are able to do real-time target tracking, path planning, navigation, and obstacle avoidance.
Pan Gang, the leader of the research team and a professor at Zhejiang University's School of Computer Science and Technology, said that the hardware and software are used to simulate the structure and operation mechanism of the brain's neural network to construct a new artificial intelligence system, a new computing model that overturns the traditional computing architecture, namely brain-like computing. It is characterized by integrated storage and computing, event-driven, and highly parallel.
In 2015 and 2019, Zhejiang University developed Darwin 1 generation and Darwin 2 generation brain-like computing chips respectively, using the chip to simulate the structure and functional mechanism of the brain's neural network, which has advantages in the fuzzy processing of images, videos, and natural language.
The result is a powerful rack-mounted brain-like computer that integrates 792 of China's proprietary Darwin 2G computing chips into three 1.6-meter-tall standard server chassis.
The working mechanism of brain neurons is that the inflow and outflow of potassium and sodium ions lead to changes in the cell membrane voltage, thus transmitting information, "It can be simply understood that a neuron receives input pulses that lead to an increase in the membrane voltage of the cell body, and when the membrane voltage reaches a specific threshold, it sends an output pulse to the axon and passes through the synapse to the Subsequent neurons thus change their membrane voltages to enable the transfer of information."
The important point here is an asynchronous operation, which means that it starts when the signal comes and doesn't run without it. Brain-like chips work similarly to biological neuron behavior, transmitting signals through pulses, which allows for a high degree of parallelism and efficiency.
There are 150,000 neurons on each chip, and each of the four chips makes aboard, and several boards are then connected together to become a module. This is how the brain-like computer is built like a building block.
This DarwinOS is oriented towards a hybrid computing architecture of von Neumann architecture and neuromagnetic architecture, which enables unified scheduling and management of heterogeneous computing resources and provides an operation and service platform for large-scale pulsed neural network computing tasks.
At present, the Darwin brain-like operating system has a microsecond switching time for functional tasks and can support the management of billion-level brain-like hardware resources.
As a result, the value of brain-like computer research can really be realized - both in the life of intelligent task processing, but also can be applied to neuroscience research, providing neuroscientists with faster and larger-scale simulation tools, providing new experimental means to explore the workings of the brain.
Currently, researchers at Zhejiang University and Zhijiang Lab have implemented a variety of intelligent tasks based on the Darwin Mouse brain-like computer.
The researchers used the brain-like computer as an intelligent hub to realize the cooperative work of multiple robots in a flood rescue scenario, which involves the simultaneous processing of multiple intelligent tasks such as speech recognition, target detection, path planning, etc., as well as the collaboration among robots.
They also used a brain-like computer to simulate a number of different brain regions, built a neural network model of the lateral geniculate nucleus of the thalamus, and simulated the periodic responses of neurons in this brain region when visual stimuli were flashed at different frequencies.
They drew reference from the hippocampal neural loop structure and neural mechanisms to build a learning-memory fusion model, to achieve music, poetry, riddles, and other temporal memory functions; to achieve a steady-state visual evoked EEG signal real-time decoding, can "ideas" typing input.
At present, brain-like computing research is still in the early stages, Darwin Mouse brain-like computer, both in terms of scale or intelligence and the real human brain are still a big gap.
But its significance lies in being able to provide an important practical example of this technological pathway, providing researchers with a tool and platform to validate brain-like algorithms to solve real-world tasks with greater robustness, real-time, and intelligence.
Chinese team develops flexible all-solid-state sodium battery