China Develops Chip 1,000 Times More Efficient for the Future of AI

Scientists from the prestigious Peking University have developed a high-speed analog chip with an unprecedented capacity to address highly complex mathematical problems, essential for advanced scientific tasks and the development of artificial intelligence (AI).

This milestone, which could change the landscape of global computing, is most notable for the new component's extraordinary ability to use significantly less energy than conventional processors.

The promising results of this research were confirmed and published in the renowned international journal Nature Electronics, attracting attention from the technology community and the sector's largest competitors.

Overcoming the Century-Old Challenge of Analog Computing

Achieving high precision and scalability in analog processing has been one of the most persistent challenges for experts for over a century. This technology operates by using continuous physical signals (such as voltage or current) to manage information, using its own hardware for calculation.

Controlling these continuous signals is inherently more difficult than handling the discrete states (the binary 1 and 0) upon which today's digital computers are based.

On the other hand, current digital chips consume vast amounts of energy because their operations require millions of transistors to constantly switch states.

Compounding this, the continuous transfer of data between memory and the processor often becomes a bottleneck that slows down the computations themselves, limiting AI's potential.

Exceptional Performance: 100 to 1,000 Times Superior

The extraordinary performance of the new Chinese analog chip places it in a league of its own. It has been proven to offer energy efficiency and computational capacity between 100 and 1,000 times superior to the most advanced digital processors available today, such as NVIDIA's H100 and AMD's Vega 20.

These processors have been key in the machine learning revolution and in catapulting NVIDIA to the status of the world's most valuable company.

The modern component from Peking University not only improves efficiency but also offers a direct solution to crucial problems. It could solve matrix inversion used in MIMO systems (a fundamental technology for large-scale wireless communications) and mitigate the serious energy and data limitations faced by digital chips in emerging ecosystems, such as next-generation Artificial Intelligence and the development of the 6G network.