China’s tech researchers are claiming to have cracked a problem that’s stumped engineers for over a century. Scientists at Peking University say they’ve developed a new analog chip that can outperform even Nvidia’s most powerful processors by as much as 1,000 times – and it doesn’t even use the traditional ones and zeros that drive modern computing.
The breakthrough was published in Nature Electronics on October 13, and it centers on a throwback concept with a futuristic twist. The chip uses analog computing, meaning it performs calculations directly on physical electrical signals instead of converting everything into digital binary code. In other words, it does math by the movement of electrons rather than by shuffling around bits of data.
The team behind the chip says this approach solves two major bottlenecks that plague digital systems: power consumption and data transfer limits. When tested on complex communications problems, including wireless network equations used in massive MIMO systems, the analog chip matched the accuracy of standard processors while using about 100 times less energy. With further tuning, it reportedly blew past Nvidia’s high-end H100 and AMD’s Vega 20 GPUs, achieving performance up to 1,000 times faster.
What makes that even more surprising is that analog computing has been around longer than digital itself. The ancient Greek Antikythera mechanism, often called the world’s first computer, was fully analog, relying on gears to perform astronomical calculations. The problem has always been precision: analog systems are inherently noisy and prone to drift, which is why the tech world abandoned them in favor of reliable binary logic decades ago.
The Chinese researchers claim to have overcome that issue using a network of resistive random-access memory cells, or RRAM. Each cell stores and processes information by adjusting how easily electricity flows through it, effectively merging computation and memory. By configuring these cells into two loops – one for rough calculations and another for refinement – the chip can process massive amounts of data almost instantly while maintaining digital-like precision.
Because it doesn’t need to constantly move information between separate memory and processor units, the chip avoids one of the biggest energy drains in modern AI hardware. And perhaps most importantly, it was manufactured using standard commercial production methods, meaning large-scale production could be feasible.
The researchers believe this design could become a foundation for future AI processors and 6G network systems. If it lives up to its promise, Nvidia’s dominance in AI hardware might finally have a real challenger – and it’s coming from a place where the analog age never quite died, it just went into hibernation.