In a lab of the Department of Precision Instruments, Tsinghua University, a group of cameras recorded fast-changing pictures on a flat panel display. A nearby buzzing server was running a program to recognize the images.

“This is our brain-inspired computer running a fast visual recognition program. This computer is a system application formed by TianJic chips,” introduced Prof. Shi Luping, director of the Center for Brain-Inspired Computing Research.

The TianJic chip featured on the front cover of the journal Nature in August 2019. It is the world’s first heterogeneous fusion brain-inspired computing chip that supports both neuroscience-based pulse neural networks and computer science-based artificial neural networks and was rated as one of the top 10 sci-tech achievements in China in 2019.

In a corner of the lab stood a black bicycle, with a computer case on its rear rack. The TianJic-mounted bicycle can maintain balance, avoid obstacles, track a target, recognize voice commands and make decisions all by itself.

Shi returned from overseas to join Tsinghua University in 2013, focusing on brain-inspired computing. In 2014, Tsinghua University set up the Center for Brain-Inspired Computing Research based on seven departments including the Department of Precision Instruments.

As the director of the center, Shi began to explore the uncharted territory of artificial intelligence (AI) with an interdisciplinary team formed by researchers from the fields of brain science, computing, microelectronics, electronics, precision instruments, automation and materials.

“There is no existing theory, direction or path for brain-inspired computing. You have to start from scratch. But it is also an opportunity, because the exploration of uncharted territory has the potential to bring breakthroughs in key scientific and technological fields.”

According to Shi, the research and application of AI have made great progress, but at present most AI solutions are aimed at specific scenarios.

“With one solution for only one scenario, it is narrow AI. It will be challenged when dealing with uncertain and dynamic complex problems without sufficient data. One of the best ways to solve this problem is to develop general AI with human intelligence characteristics,” he said.

At present, there are two mainstream approaches to general AI research. One is from the perspective of computer science, based on the Von Neumann computer, artificial neural networks and the deep learning method.

The other is from the perspective of brain science research, using nano-scale devices to simulate the information processing characteristics of biological neurons and synapses, based on the non-Von Neumann architecture neural morphological chips and pulse neural networks.

Brain-inspired computing is a new computing technology based on neuromorphic engineering, using basic principles of brain science for the development of general AI. The research of brain-inspired computing is still in the initial stage, and there are no recognized technical standards in the world, he added.

When studying the earlier papers of Alan Turing and Von Neumann, Shi found that they began their research with the goal of achieving general AI. Limited by the hardware development at that time, however, their research chose to provide one solution for one scenario.

“However, with the development of advanced precision instruments, we have more and more understanding of the brain, which will usher in breakthroughs in brain science,” he said.

The development of supercomputers provides a better simulation environment for brain-like computing, while big data and cloud computing provide a “digital universe” whose complexity is getting closer to that of the brain. The brain and the digital universe can be regarded as mirror images, learning from each other and promoting mutual understanding.

Meanwhile, new nanotechnology could create devices that use as much energy as human neurons, paving the way for brain-inspired computing.

“Silicon and carbon are very similar in terms of their elemental properties. The human brain has realized general intelligence on the basis of carbon, so we have reason to believe that general intelligence can also be realized on the basis of silicon.”

Shi led the team to develop a complete brain-inspired computing platform from theory, chips, software to systems and applications, which is still being updated. The TianJic chip has been developed for generations, and the world’s first brain-inspired commercial chip has also hit the market with many different applications.

“In recent years, Tsinghua University has made innovative achievements in theory, chips and devices in the field of brain-inspired computing, reaching the frontier of the global research community,” said Shi.

Ma Songchen, a PhD student in charge of the chip system at the center, said the team had taken into consideration the fact that the production process can be handled by domestic manufacturers. The chip can achieve excellent computing power based on the current domestic manufacturing level, which is also one of the potential advantages of brain-inspired computing.

Before his PhD work in the center, Ma thought that scientific research meant doing whatever his supervisor assigned, without considering cooperation with other researchers or potential combinations with other fields.

“But in this team formed by multidisciplinary experts, everyone should think from the perspective of the whole team to solve problems,” Ma said. “For example, when I design chips, I must consider the needs of software and systems.”

“Whether it is the cooperation between departments or the work within the team, we support each other, share achievements and make progress together for a big goal,” said Shi.

“The focus of our research is not only chips, systems and software, but also the trend and direction of the future of humanity. I believe China can make even greater contributions,” he said.

by Xinhua writers Wang Chenxi and Zhao Xinhu