Scientists in China have developed a new kind of chip that abandons the binary world of 1s and 0s for an analog approach, performing calculations directly on physical circuits. According to the researchers, the device could outperform the latest GPUs from Nvidia and AMD by as much as 1,000 times, while consuming roughly 100 times less energy. Their study, published in Nature Electronics on October 13, demonstrates how this analog processor tackles longstanding limitations in digital computing and opens the door to far more efficient handling of massive datasets.
The breakthrough centers on arrays of resistive random-access memory (RRAM) cells that both store and process information by regulating electrical flow. Unlike conventional digital processors that shuttle data between computing and memory units, this analog design computes within its hardware, avoiding energy-intensive transfers and providing continuous, high-speed data processing. When tested on complex problems such as matrix inversion for massive multiple-input multiple-output (MIMO) wireless systems, the chip matched digital processors in accuracy while dramatically reducing energy use. With further optimization, the team reported performance gains exceeding 1,000 times that of high-end GPUs like Nvidia’s H100 and AMD’s Vega 20, both critical for AI model training.
Analog computing is far from new. Ancient devices like the Antikythera mechanism, built more than 2,000 years ago, relied on gears to perform calculations. For most of modern computing history, however, analog systems have been considered impractical because continuous signals such as voltages are far harder to control with precision than the stable 1s and 0s of digital processors. Yet the Peking University team argues that the limitations of digital scaling, combined with the growing demands of AI and 6G applications, make analog computing a viable and transformative alternative.
According to the study, this approach offers not only unprecedented throughput but also energy efficiency unmatched by contemporary digital hardware, achieving the same computational precision with far less power. As researchers continue to refine the design, this analog chip could mark a turning point in computing, bridging centuries of technological evolution from mechanical gears to modern AI-ready processors.
