Page Loader
Summarize
World's 1st non-binary AI chip enters production: What's so special?
The chip is designed for key industries

World's 1st non-binary AI chip enters production: What's so special?

Jun 09, 2025
03:52 pm

What's the story

China has begun mass production of the world's first non-binary artificial intelligence (AI) chip. The revolutionary technology combines binary and stochastic (property of being described by a random probability distribution) logic, overcoming traditional computing limitations. The project is led by Professor Li Hongge's team at Beihang University in Beijing, and integrates China's proprietary hybrid computing technology into key sectors such as aviation and industrial systems.

Technological breakthrough

Addressing power wall and architecture wall

The non-binary AI chip addresses two major challenges in today's chip technologies: the power wall and the architecture wall. The power wall is a contradiction where binary systems, while efficient at carrying information, consume a lot of power. Meanwhile, the architecture wall arises from new non-silicon chips' inability to seamlessly communicate with traditional systems based on complementary metal-oxide-semiconductors (CMOS).

Innovative approach

Enabling fault tolerance and power efficiency

To tackle these challenges, Li's team proposed a new numerical system called the Hybrid Stochastic Number (HSN). This innovative method combines traditional binary numbers with stochastic or probability-based numbers. The non-binary AI chip's unique architecture allows unprecedented fault tolerance and power efficiency in intelligent control applications like touch displays and flight systems, all while circumventing US chip restrictions.

Architecture

It has an SoC design

Besides HSN, the non-binary chip uses in-memory computing algorithms. These lower the energy-intensive data that shuttle between memory and processors in normal architectures, hence improving overall efficiency. It also flaunts a system-on-chip (SoC) design, integrating many types of computing units in order to handle multiple tasks in parallel. This breaks it free from the limits of traditional homogeneous architectures.