China takes the lead in large-scale production of non-binary AI chips

According to the South China Morning Post, China has taken the lead in the world by applying non-binary AI chips on a large scale, integrating its proprietary hybrid computing technology into key areas such as aviation and industrial systems.

The report stated that this breakthrough was led by Professor Hongge Li's team from the School of Electronic Information Engineering at Beijing University of Aeronautics and Astronautics. By fusing binary and stochastic logic, they overcame fundamental obstacles in traditional computing, achieving unprecedented fault tolerance and power efficiency in smart control applications like touch displays and flight systems, while bypassing US chip restrictions.

As reported by Guangming Daily earlier, Professor Hongge Li's team developed an open-source RISC-V architecture high fault-tolerant anti-interference, high-efficiency disruptive computing chip - the Hybrid Probability Computing SoC chip with complete independent intellectual property rights. This chip achieved disruptive innovations in number system representation (disrupting binary numbers), computing algorithms (in-memory computing), and heterogeneous computing architecture (SoC).

The report noted that current computing chip technology faces two major challenges: the power wall and the architecture wall. First, traditional chips cannot escape the contradiction between the low information-carrying efficiency and high power consumption of binary numbers (power wall). Second, non-silicon-based chips cannot directly communicate with traditional CMOS chips and their computing architectures (architecture wall).

To solve these problems, Professor Hongge Li's team has been researching the computer principles and chips of non-binary numbers, proposing a completely new hybrid probability number (Hybrid Stochastic Number, HSN) combining binary numbers and stochastic probabilistic numbers. This research first analyzed and unified the mathematical relationships among binary numbers, traditional probabilistic numbers, and hybrid probability numbers, constructing mathematical representations among the three number systems and analyzing their high fault tolerance, anti-interference, and high energy efficiency characteristics, opening up a new paradigm for constructing intelligent computing chips, laying the foundation for solving the power wall and architecture wall problems of silicon-based computing chips.

Currently, this technology is undergoing transformation, mainly applied in intelligent control fields such as touch recognition, instrument display, and flight control calculation. Professor Hongge Li said that the team is studying exclusive extension instruction sets and microarchitectures for hybrid probability computing, developing functions such as speech and image processing, intelligent computing (including AI acceleration for large models), and various complex calculations. The delay of on-chip dedicated operators is at the microsecond level, meeting multiple demands for specialized hardware computing and flexible variable software computing.

In recent years, the US government has continuously tightened its export restrictions on Chinese chips, suppressing and curbing the development of China's semiconductor industry. Against this backdrop, China's semiconductor industry has accelerated its growth, significantly enhancing its self-research and development capabilities.

A June 5 report by The New York Times stated that ASML CEO Peter Wennink told reporters that although China still has a long way to go in catching up with ASML's technology, the suppression measures implemented by the US would backfire, making China "work harder to succeed".

On May 28, NVIDIA CEO Jensen Huang said that regardless of whether there are US chips or not, China's AI industry will continue to develop.

Original source: https://www.toutiao.com/article/1834426844198924/

Disclaimer: The article solely represents the views of the author.