日本語
한국어
Tiếng Việt
简体中文
Chinese researchers have unveiled a groundbreaking analog chip that promises to revolutionize computing, potentially outperforming high-end graphics processing units (GPUs) from industry giants like Nvidia and AMD by an astonishing 1,000 times. This innovative development from Peking University addresses a long-standing challenge in analog computing while simultaneously tackling critical energy and data bottlenecks in emerging fields such as artificial intelligence (AI) and 6G technology.
Published in Nature Electronics, the study details how this new device, built upon resistive random-access memory (RRAM) cells, marks a significant leap forward. Unlike traditional digital processors that operate using binary 1s and 0s, this analog chip processes information through continuous electrical currents within its physical circuits. This fundamental difference allows it to perform calculations directly on its hardware, eliminating the energy-intensive data transfer typically required between a processor and external memory sources.
Unprecedented Speed and Efficiency for AI and 6G
The chip’s performance is particularly impressive in complex communication problems, including matrix inversion challenges prevalent in massive multiple-input multiple-output (MIMO) wireless systems. In these demanding scenarios, the Peking University team reported that their analog chip matched the precision of standard digital processors while consuming approximately 100 times less energy. Further optimizations boosted its performance to a staggering 1,000 times faster than top-tier GPUs like the Nvidia H100 and AMD Vega 20 – chips crucial for training advanced AI models such as ChatGPT.
The researchers emphasize that digital computers are increasingly challenged by the vast amounts of data required by modern applications, especially as traditional device scaling becomes more difficult. Their analog computing approach offers a compelling alternative, delivering significantly higher throughput and superior energy efficiency for comparable precision levels.
Solving the ‘Century-Old Problem’ of Analog Computing
While analog computing boasts a long history, dating back over 2,000 years to devices like the Antikythera mechanism, its widespread adoption has been hindered by issues of poor precision and practicality. Digital systems, with their stable binary states, have historically offered greater control and accuracy.
However, analog systems inherently excel in speed and efficiency. By representing calculations as physical operations on the chip’s circuitry rather than lengthy binary code, they can manage large volumes of information concurrently with substantially less energy consumption. This inherent advantage is now critical for data- and energy-intensive applications like AI and future 6G networks, which demand real-time processing of massive, overlapping data streams.
The breakthrough lies in configuring the RRAM cells into a unique two-circuit system: one for rapid, approximate calculations and another for iterative refinement, enhancing precision. This innovative design successfully marries the inherent speed of analog computation with the accuracy typically associated with digital processing.
Mass Production Potential and Future Outlook
Crucially, this advanced analog chip was developed using a commercial production process, suggesting a strong potential for mass manufacturing. This factor could significantly accelerate its integration into various technological sectors.
Looking ahead, the Peking University team aims to further enhance the chip’s circuitry to boost its performance. Their immediate goal is to develop larger, fully integrated chips capable of tackling even more complex computational problems at unprecedented speeds. This development could herald a new era in high-performance, energy-efficient computing, paving the way for next-generation AI and communication technologies.