The World's First Heterogeneous Fusion Brain-like Chip Is on the Cover of Nature
Today, a research team led by Tsinghua University developed the world's first heterogeneous fusion brain-like computing chip, "Tianji", which appeared on the cover of the latest issue of Nature. This AI chip can integrate the two methods led by computer science and neuroscience to develop a general platform with the advantages of both, and take a step further towards the development of general artificial intelligence.
It’s brain-like computing and AGI again, but this time, it’s on Nature Domestic chips on the cover of the magazine.
today,Tsinghua University Brain-Inspired Computing Research CenterA study led by him brought about a major breakthrough in brain-like chip research, pushing people's research on AGI a big step forward.
Collaborators from multiple research institutions have jointly createdThe first heterogeneous fusion brain-like computing chip,It links traditional machine learning and brain-like computing methods.
This is also a milestone moment, as a paper from China was published in Nature for the first time in the fields of chip manufacturing and AI.
Why is it on the cover of Nature?
This article is titled Heterogeneous Tianjic Chip Architecture for General Artificial IntelligenceThe paper introduces the production process and working mechanism of the chip "Tianji".
Paper Title:Towards artificial general intelligence with hybrid Tianjic chip architecture
Address: https://www.nature.com/articles/s41586-019-1424-8
The key point of the secret is reflected inFusion.
In the research of AGI, there are two schools of thought.Computer Science Orientation, and the other isNeuroscience oriented.
This has led to the development of two different approaches. One is the artificial neural network (ANN), on the one hand, there are brain-like related research, such as pulse neural network (SNN).
The two models developed independently, using different languages, computing principles, coding methods, and scenarios. However, the development of AGI needs to draw on the advantages of both models.
For a long time, the hardware of the two modes has relied on different platforms and has been difficult to be compatible with each other.
To solve this problem, the research team developed an architecture that heterogeneously integrates the two solutions and created this cross-paradigm computing chip, which perfectly solves this problem.
The Tianji chip adopts a multi-core architecture and reconfigurable functional core modules, and supports a data flow control mode of a hybrid coding scheme.
It can not only adapt to machine learning algorithms based on computer science, but also implement neural computing models and multiple encoding schemes inspired by brain principles.
What are the highlights of this study?
The main innovation of the chip is reflected inFunctional Core (FCore)On the surface, FCore includes axon, synapse, dendrite, cell body, and neural router building blocks.
Through the reconfigurable Fcore mode, flexible modeling configuration and topological connection are achieved, and the encoding method can be converted between ANN and SNN modes, thereby achievingHeterogeneous neural networks.
FCore also covers the linear integration and nonlinear transformation operations used by most ANNs and SNNs, and can perfectly support the operation of both.
A Tianji chip is made of 156 The FCore consists of approximately 40,000 neurons and 10 million synapses, is manufactured using a 28-nanometer semiconductor process, and has an area of 3.8×3.8 square millimeters.
The Tianji chip has also made great progress in performance. It can provide an internal memory bandwidth of more than 610 gigabytes (GB) per second and can achieve a peak performance of 1.28 TOPS when running ANN.
Compared to the performance of GPUs, the chip's throughput is improved 1.6-100 times, the power efficiency is improved 12-10000 times.
What else do you want a bicycle?
To demonstrate the applicability of the chip and system, they built aSelf-driving bicycle, deployed on a Tianji chip, and conducted operational tests.
This unmanned bicycle platform has the functions of voice recognition, target detection and tracking, motion control, obstacle avoidance, and autonomous decision-making.Small brain-like computing platform.
During the experiment, the bicycle successfully completed autonomous driving, verifying the feasibility of their solution and chip.
Researcher Deng Lei said that the speech recognition, autonomous decision-making, and visual tracking functions of the unmanned bicycle system use a model that simulates the brain;
The target detection, motion control and obstacle avoidance functions use machine learning algorithm models.
The seemingly incredible bicycle demonstration also allowed more people to see the possibility of the arrival of AGI. It is said that their next step is to move towards commercial use.
With the help of super-powered chips, will AGI come?
Recently, information about brain-like research and AGI has frequently entered the public eye.
First, Musk announced new progress in brain-computer interface, which brought a wave of attention to brain science research. Then, Microsoft's investment of 1 billion in OpenAI for AGI research caused a stir in the industry. Recently, Facebook also announced its results in brain-like research...
There is no doubt that this breakthrough chip that achieves heterogeneous fusion has added fuel to the research of AGI.
Will AGI, which is coveted by countless people, become a reality in our lifetime?
-- over--