China’s Non-Binary AI Chips Based on Hybrid Stochastic Logic

• 10 Jun 2025

In June 2025, in a world-first, China has started large-scale deployment of non-binary AI chips using Hybrid Stochastic Number (HSN) computing, developed by Professor Li Hongge’s team at Beihang University. These chips are being used across aviation, industrial control, and intelligent display systems.

Key Points

• Computing Breakthrough: HSN computing merges traditional binary logic (accurate but power-hungry) with stochastic logic (energy-efficient but slower), achieving both low power consumption and high computational reliability.
• Overcoming Power & Architecture Walls: The hybrid chip bypasses limitations of binary chips (high energy use) and integration issues of non-silicon chips with CMOS infrastructure.
• Real-World Applications: Used in touch displays, medical and industrial systems, and flight control, the chip improves signal detection, power efficiency, and fault tolerance.
• In-Memory Computing: Reduces data transfer bottlenecks between memory and processor, enhancing speed and efficiency.
• Built on Domestic Tech: Developed using mature 110nm and 28nm processes by SMIC, the chip avoids reliance on US-restricted semiconductor technology.
• Next Steps: Team is working on a custom instruction set architecture (ISA) to power AI model acceleration, speech/image recognition, and neural networks.