LCD Display Inverter

Display Inverter / VGA Board / LCD Controller

China’s first 100 million neuron brain-like computer released: the number of neurons is equivalent to the mouse brain

my country has taken a new step in the field of brain-like intelligent computing technology!

On September 1, Zhejiang University and Zhijiang Laboratory released a brain-like computer, Darwin Mouse, in Hangzhou, which contains 120 million spiking neurons and nearly 100 billion synapses.

It is understood that this is my country’s first brain-like computer based on brain-like chips with independent intellectual property rights. The computer uses 792 “Darwin II” brain-like chips developed by Zhejiang University, supporting 120 million spiking neurons, Nearly 100 billion synapses, the number of neurons is equivalent to that of the mouse brain, and the typical operating power consumption is only 350-500 watts. It is worth mentioning that Darwin Mouse is also the largest brain-like computer with neurons in the world.

The research team has also developed a dedicated operating system for brain-like computers, Darwin’s Brain-like Operating System (DarwinOS), to achieve effective management and scheduling of brain-like computer hardware resources and support the operation and application of brain-like computers.

“Emergency rescue, listening to songs and understanding songs, and inputting ideas” are proficient in everything

What exactly does a brain-like computer do? In the on-site demonstration, multiple robots used the brain-like computer as the intelligent center to demonstrate the collaborative work in the flood-fighting and emergency rescue scene. The 3 robots with similar appearances undertake patrol, rescue and rescue tasks respectively, and their functions are different. They can be controlled by different brain regions and give different instructions to robots 1, 2, and 3.

The picture comes from the official WeChat account of Zhejiang University

Listening to songs and recognizing songs is also Darwin Mouse’s forte. The staff only needs to sing two songs in one song. The content of the song “sings” out. In addition to memory retrieval, a memory model architecture can also be established by drawing on the hippocampal network structure and neural mechanism. Through the pulse coding of memory, the same model can learn and memorize different types of data such as speech, songs, and texts.

The picture comes from the official WeChat account of Zhejiang University

With the help of brain-like computers, the researchers also realized the real-time decoding of steady-state visual evoked potentials of EEG signals, which can be “thought” typing input.

Brain-like computing: a new computing model that subverts traditional computing architecture

According to Pan Gang, the leader of the research team and a professor at the School of Computer Science, Zhejiang University, hardware and software are used to simulate the structure and operation mechanism of the neural network of the brain to construct a brand-new artificial intelligence system. This new computing model that subverts the traditional computing architecture is Brain-like computing. It is characterized by the integration of existence and computing, event-driven, highly parallel, etc. It is the research focus of international academia and industry, and it is also an important technology strategy.

Wu Zhaohui, academician of the Chinese Academy of Sciences and president of Zhejiang University, said that brain-like computers will become the main form and important platform for computing in the future, and will play an important and unique role in simulating brain functions, efficiently implementing AI algorithms, and improving computing power. Facing the future, interdisciplinary convergence will become a new method to solve major problems, and systematic innovation based on multi-disciplinary and multi-domain will become an effective form of developing brain-like computers. “We hope to learn from the structural model and functional mechanism of the brain, and apply the cutting-edge results of brain science to research fields such as artificial intelligence, so as to establish a new computer architecture that leads the future.”

According to OFweek Electronic Engineering Network, as early as 2015, Zhejiang University led the development of the “Darwin Generation” brain-like chip, which simulates the neuron LIF model and has stronger biological authenticity than traditional neural networks. This is also the first neural network in China. Mimic brain chip.

In 2019, the “Darwin II” brain-like chip was born. A single chip consists of 576 cores, supports 150,000 neurons and 10 million synapses, which is equivalent to the number of neurons in Drosophila. By cascading chips, a nervous system of tens of millions can be constructed, reaching a similar scale of the TrueNorth chip, but it can simulate synapses with higher precision than TrueNorth. The chip is also the largest single-chip neuron brain-like chip known in China.

In the exploration of brain-like intelligence, what is the difference between brain-like chips and traditional chips?

Brain-like intelligence, also known as brain-like computing. In the late 1980s, American scientist Carver Mead first proposed the concept of brain-like computing. The idea of ​​brain-like computing gets rid of the traditional computing model, imitates the working principle of the human nervous system, and is eager to develop fast, reliable, and low-cost computing technology.

In essence, brain-like intelligence is the ultimate goal of artificial intelligence, but in the final analysis, the complexity of the human brain is that brain-like intelligence can never be fully replicated. Therefore, brain-inspired intelligence is more hoping to learn from the working mechanism of the human brain and simulate a computer that has the same thinking and learning ability as humans.

The picture comes from the official WeChat account of Zhejiang University

The emergence of brain-like computers has brought great assistance to the research road of brain-like intelligence. The way brain-like computers work is very similar to people’s brains, and it is mainly reflected in two aspects:

1. The neuron model, the brain-like computer establishes an accurate model by imitating the human brain;

2. The information transmission between neural networks is transmitted by means of pulses in the biological brain.

In the 2019 top ten events of future science and technology jointly released by China Science News and NetEase News in 2019, “brain-like chips” are impressively listed. As a key technology on the road to brain-inspired intelligence exploration, the emergence of brain-inspired chips is even more promising.

Brain-like chips are very different from digital chips that rely on binary computing. The working principle of brain-like chips is similar to the type and number of ions flowing through synapses to activate neurons. By exchanging gradient signals or weight signals, the purpose of simulating the human brain is achieved. ; And traditional chips follow Feng. It is designed by Neumann architecture. Since storage and calculation are separated in space, the computer needs to be called back and forth between the two areas of CPU and memory during operation. Frequent data exchange not only leads to low efficiency of massive information processing, but also It also causes serious power consumption when working.

Therefore, the brain-like chip is based on the combination of microelectronics technology and new neuromorphic devices, hoping to break through the traditional computing architecture, realize the deep integration of storage and computing, greatly improve computing performance, improve integration, and reduce energy consumption.

The research layout of the world’s top scientific research institutions in brain-like chips

At present, the field of brain-inspired intelligence is still in its infancy, and there is still a long way to go before it is officially commercialized. This is also where countless world-class scientific research institutions are actively deploying. Whoever comes out on top will be the first to win the right to speak in the future.

IBM, the first company in the world to research brain-like chips. In 2011, IBM simulated the structure of the brain and created a generation of silicon chip model “TrueNorth” with perception ability, which can learn and process information like the brain, and reorganize according to the corresponding neuron connection paths; In 2014, IBM invested in DARPA With the support of the $100 million “Neuromorphic Adaptive Plastic Scalable electronic System” project, the “TrueNorth” second-generation brain-like chip was developed. Using 28nm silicon process technology, the number of neurons has increased to 1 million, compared with the previous generation. It is 3906 times higher, the number of programmable synapses is 976 times higher, 46 billion synaptic calculations can be performed per second, and the total power consumption is only 70 milliwatts.

Qualcomm, people may be familiar with Qualcomm because of its Snapdragon mobile chips, but few people know that Qualcomm’s layout on brain-like chips is also very far-reaching. In 2015, Qualcomm officially launched a brain-like chip called “Zeroth”, which people can program with traditional programming languages, and use the “NPU training” terminal to achieve human-like movements and behaviors.

Google, in 2014, proposed a machine learning model: Neural Turing Machines (NTM). To put it bluntly, it is to develop supercomputers through core chips, integrating the advantages of traditional Turing machines and neural networks, which can learn new knowledge from information while storing information, and use new knowledge to perform logical tasks.

Intel, the big brother in the chip industry, has also not missed the research on brain-like chips. Intel has designed neuromorphic chips based on two technologies: lateral spin valves and memristors. The former can switch on and off according to the direction of electron spins passing through, and the latter works like neurons and can replicate the processing power of the brain. The chip was generally optimistic about the outside world at that time, and the growth potential was huge.

In addition to the above-mentioned companies, many academic institutions have also participated in the research and development of brain-like chips. For example, the “Neurogrid” brain-like chip launched by Stanford University in 2014 is about 900 times faster than ordinary computers. The product prototype consists of 16 custom chips. , capable of simulating 1 million brain neurons and billions of synaptic connections.

The team led by Professor Shi Luping from the Brain-inspired Computing Research Center of Tsinghua University in my country also launched the world’s first heterogeneous fusion brain-inspired chip “Tianjic” last year. It is understood that this hybrid chip has multiple highly reconfigurable functional cores that can support both machine learning algorithms and existing brain-like computing algorithms.

Countries seize the commanding heights of the frontier, what can brain-like intelligence technology bring?

Due to the huge development potential and broad market prospects of brain-inspired chips, brain-inspired intelligence technology has become the focus of national science and technology strategies and the core scientific and technological development areas that are vigorously promoted. With the introduction of the development strategies of brain-like research in developed countries such as the United States, Japan, Germany and the United Kingdom, China’s brain-like scientific research projects have also been officially launched.

From a global perspective, Japan started brain science and education-related projects as early as 2003, conducting research on educational theory and practical applications; the United States has been committed to brain-like technology research since 2012, from the Institute of Health to the Bureau of Information Research Then to the Defense Research Agency, brain-like scientific research has begun in many fields and directions; the European Union launched the “EU Human Brain Project” project in 2013, aiming to establish a set of latest and revolutionary information communication based on neuroscience technology to build a chip that simulates the function of neurons and use this chip to build supercomputer systems. It is reported that the overall investment of the plan is 1.19 billion euros and will last for ten years.

Similarly, in 2013, China also launched the “Brain Science and Brain-Like Intelligence Technology”, referred to as the “China Brain Project”, which mainly has two research directions: to explore the secrets of the brain, to overcome Brain disease-oriented brain research and brain-like research oriented to the establishment and development of artificial intelligence technology. These technologies embody neural labeling and neural circuit tracing technology, brain imaging technology, neuromodulation technology, neural information processing platform and so on. According to the order from “research” to “application”, the research content of brain science and brain-like artificial intelligence can be divided into four parts: 1. Brain neural network analysis; 2. Cognitive mechanism computational simulation; 3. Brain-like intelligence Algorithm innovation; 4. Application innovation of brain-like intelligence technology.

What can brain-like intelligence technology bring? Some people would think that brain-like intelligence is artificial intelligence. In fact, the two are not suitable to be equated. Brain-like intelligence is only one type of artificial intelligence technology. As a new technology developed by imitating the structure of the human brain to store and process information, brain-like computing is the cornerstone of artificial general intelligence.

Due to its powerful computing power and information processing capabilities, brain-inspired chips have driven artificial intelligence algorithms, intelligent perception and other related technologies with brain-inspired computing as the core, and have been used in AI applications such as smart homes, smart cities, smart medical care, and smart robots. widely used.

Important challenges for brain-like research

Because brain-like research needs to be based on in-depth knowledge of the human brain, and the brain, as the most complex and mysterious organ in the human body, has hundreds of millions of neurons that continuously emit electrical signals to form a dense network, which has puzzled scientists for centuries. While knowledge about this mysterious organ has grown rapidly in the past decade, the ultimate mystery of the brain remains a mystery.

Therefore, the biggest problem currently faced by brain-inspired research is the lack of cognition of brain functions, and more advanced brain observation methods and synchronous regulation technology are needed; secondly, brain-inspired computing uses spiking neural networks to replace the neuron model information transmission of the brain The principle, but this is still at the primary perception level. In terms of more advanced brain processing capabilities, brain-like computing is still vague. It is very difficult to convert the complex information transmission and processing process of the brain into a computational model; subject to Due to the limitations of chip technology, material technology, power consumption and other conditions, it is difficult for brain-like computing chips to achieve large-scale neuron interconnection integration and efficient real-time transmission of neuron pulse information in hardware conditions. Learning ability, brain-like chips have made achievements in memory, storage, reasoning, etc., and “learning” is the core of intelligence, and it needs deeper code logic and algorithm mechanism to improve.

All in all, although there is still a big gap between the current brain-like chip and the real human brain in terms of scale and intelligence, it also has advantages beyond the reach of the human brain. Today, a new track for brain-like scientific research around the world has been formed, and it is believed that many subversive theories and revolutionary technological achievements will emerge in the future.