China begins production of first industrial-scale non-binary AI chip | Hardware | Types of cpu | Gpu vs cpu reddit | Turtles AI
China has started mass production of the world’s first non-binary AI chip, developed by Li Hongge’s team at Beihang University. Based on the hybrid HSN system, it combines binary and stochastic logic, improving energy efficiency, fault tolerance and compatibility with mature processes.
Key points:
- Innovative "non-binary" AI chip combines traditional and probabilistic logic (HSN).
- Improves energy efficiency and computation robustness in noisy environments.
- Built on SMIC 110nm and 28nm processes, bypasses US restrictions.
- Used in touch displays, flight systems and advanced industrial control.
China, with Beihang University and led by Professor Li Hongge, has launched large-scale production of the world’s first non-binary AI chip, a solution that overcomes the challenges of the "power wall" and "architecture wall" typical of conventional computing. The core technology is the HSN (Hybrid Stochastic Number), a numerical system that integrates precise binary numbers with probabilistic stochastic values, enabling robust and low-power calculations. This hybrid structure not only reduces the power required for intensive operations, but also makes the chip more immune to errors and noise, particularly useful in touch displays and aircraft controls. The design leverages SMIC’s mature 110nm and 28nm processes, ensuring compatibility with existing CMOS infrastructure and circumventing US restrictions on advanced technologies. Furthermore, the chip implements in-memory computing and SoC design, which reduces data traffic between memory and processor and enables the execution of multiple tasks in parallel, critical features for industrial and aerospace applications. In real-world use, these chips are already mounted in aircraft navigation systems, smart displays and industrial automation, offering latencies in the microsecond range and balancing hardware performance with software flexibility; furthermore, specific ISAs and microarchitectures for audio, visual processing and acceleration of complex AI models are under development. This move marks a technological advancement that aims at China’s autonomy in the semiconductor sector, offering an alternative model to scalability based exclusively on increasingly smaller nodes.
A production that confirms the concrete launch of a hybrid AI platform, ready to move on an industrial scale.