Chinese Analogue Chip Breakthrough Claims 1,000 Times Faster Processing Than Leading Digital Processors

Chinese scientists have developed an analogue computing chip that could deliver processing speeds 1,000 times faster than leading digital processors, such as the H100 graphics processing unit (GPU), while consuming just a fraction of the energy.

The breakthrough, published in the peer-reviewed journal Nature Electronics on 13 October, represents the first time analogue computation has achieved precision comparable to digital systems.

Researchers from Peking University's Institute for Artificial Intelligence, led by Dr Sun Zhong, created the chip using resistive random-access memory (RRAM) technology.

Unlike conventional digital processors that rely on binary code, the analogue device performs calculations using continuous electrical signals directly within its circuits.

When tested on complex matrix problems, the chip completed in roughly one minute what would take a high-end GPU core an entire day, while consuming over 100 times less energy at the same precision level.

The device tackles what researchers describe as a century-old problem that has plagued analogue computing: achieving both high precision and scalability.

The team configured RRAM cells into two circuits—one providing fast approximate calculations and another refining results through iterative processing.

This hybrid approach enables the chip to match the accuracy of digital processors whilst maintaining analogue computing's inherent speed and efficiency advantages.

The breakthrough improves analogue precision by an astonishing five orders of magnitude, or nearly 100,000 times compared to traditional analogue systems.

Potential applications span next-generation 6G communication networks, where base stations could process massive antenna signals in real time with ultra-low power consumption, and AI model training, where the technology could drastically reduce both training time and energy costs for large language models.

The chip was fabricated using commercial semiconductor processes, making it potentially scalable for mass production.

Industry analysts suggest this could reshape AI hardware by offering an energy-efficient alternative to Nvidia's GPUs and AMD's processors in specific applications such as inference acceleration and wireless signal processing, particularly as digital computing approaches its physical limits in the post-Moore era.

Please click here to add Swarajya as your preferred and trusted news source on Google