Tsinghua’s Tianjic Chip selected for 15 world-leading internet scientific and technological achievements

On October 20th 2019, 15 world-leading internet scientific and technological achievements of 2019 were acclaimed at the sixth World Internet Conference held in Wuzhen. Tsinghua University’s “Hybrid Tianjic chip towards artificial general intelligence” was successfully selected.

The Tianjic Chip is the world's first hybrid brain-inspired computing chip for the development of artificial general intelligence (AGI). It has a decentralized multi-core architecture, and integrates local memory access and computation within each processing core. Each chip contains 156 functional cores, composing about 40,000 neurons and 10 million synapses. It not only supports the implementation of computer-science-oriented machine learning algorithms and neuroscience-oriented neuromorphic computing models, but also supports their hybrid modeling. The Tianjic chip provides a general computing platform for the development of AGI.

Shi Luping, the director of the Center for Brain-Inspired Computing Research (CBICR), Tsinghua University, said: “AGI has always been the dream of mankind. There are currently two general approaches towards it: computer-science-oriented and neuroscience-oriented approaches. The two approaches are very different and highly complementary.”

The Tsinghua team combined the two to construct a highly promising complementary approach and propose a novel brain-inspired computing architecture—the hybrid brain-inspired computing Tianjic chip, which can simultaneously support the computer-science-oriented and neuroscience-oriented models, such as artificial neural networks and spiking neural networks, so that they can play their respective advantages.

On the basis of the first generation of Tianjic chip developed in 2015, after continuous improvement, the second generation Tianjic Chip was launched in 2017, featuring high speed, high performance and low power. Based on Tianjic chip, the team further developed an unmanned bicycle system that processes multiple prevailing artificial neural network models and brain-inspired computational models, and simultaneously implemented multimodal tasks including real-time target detection and tracking, speech recognition, obstacle avoidance, balance control, and autonomous decision making. Its applications can be extended to areas such as autonomous driving, drones, and intelligent robots.