Chinese Researchers Unveil Analogue AI Chip 1,000 Times Faster Than Nvidia’s

Peking University's analogue chip could process calculations 1,000 times faster than Nvidia's H100 GPU with 100 times better energy efficiency.

Penking University scientists have developed an analogue chip that could process calculations 1,000 times faster than Nvidia's H100 GPU while consuming 100 times less energy. Led by Professor Sun Zhong, the breakthrough solves analogue computing's century-old precision problem for AI and scientific computing tasks.

What if the future of computing isn't about making digital chips faster, but about bringing back analogue technology that was considered obsolete decades ago? Chinese scientists from Peking University just demonstrated that possibility with an analogue chip that could theoretically process calculations one thousand times faster than Nvidia's flagship H100 GPU – while using a fraction of the power.

The breakthrough, published in the peer-reviewed journal Nature Electronics on October 13, tackles what researchers call a "century-old problem" that has plagued analogue computing since its inception: achieving both high precision and scalability simultaneously, reported SCMP.

"Precision has long been the central bottleneck of analogue computing," the research team explained in their paper. "How to achieve both high precision and scalability in analogue computing, thereby leveraging its inherent advantages for modern computing tasks, has been a 'century-old problem' plaguing the global scientific community."

Study author Sun Zhong, an assistant professor at Peking University, emphasized the significance of their achievement. The team's analogue computing approach could potentially offer one thousand times higher throughput and one hundred times better energy efficiency than state-of-the-art digital processors while maintaining the same precision.

READ ALSO: https://www.modernmechanics24.com/post/coral-baby-food-boosts-survival

To understand why this matters, you need to grasp the fundamental difference between digital and analogue computing. Digital computing works like a conventional light switch – it's either on or off, represented by ones and zeros. Analogue computing, by contrast, operates like a dimmer switch with a dial that can be adjusted to different levels, processing information using values that vary continuously within a range.

The Peking University device uses memory chips made of resistive materials – specifically, resistive memory arrays that store data by changing the electrical resistance of material between electrodes. This architecture allows the chip to solve complex mathematical problems directly through physical quantities like voltage and current, rather than through the step-by-step algorithms digital computers use.

Digital computing dominates modern technology for good reasons: vast storage capacity, programmability, and the ability to perform precise calculations through algorithms. But digital systems struggle with certain tasks, particularly large, continuous calculations used in simulating natural systems like weather patterns or powering artificial intelligence that relies on matrix-based calculations, according to SCMP.

WATCH ALSO: https://www.modernmechanics24.com/post/spacex-11th-test-flight-starship-rocket

"With the rise of applications using vast amounts of data, this creates a challenge for digital computers, particularly as traditional device scaling becomes increasingly challenging," the team noted. This challenge stems from the fundamental architecture of digital systems, where processing and memory remain separated – a limitation that creates increasing time and energy consumption as we approach physical computing limits.

Analogue computing sidesteps this bottleneck by using physical quantities to directly solve problems, allowing it to simulate dynamic systems or perform AI calculations faster and with dramatically lower power consumption.

The concept isn't new. The Antikythera mechanism, an ancient Greek mechanical model used to predict astronomical positions and eclipses, represents the oldest known analogue computer. In 1936, Soviet engineer Vladimir Lukyanov built the "water integrator," an analogue computer using water flowing through tubes to solve mathematical problems.

READ ALSO: https://www.modernmechanics24.com/post/texas-am-metallic-gel-could-transform-battery-tech

Despite analogue computing's theoretical promise, development has faced persistent barriers including low precision and scalability issues, relegating it to "legacy technology" status. The Peking University team's breakthrough addresses these limitations head-on.

According to the university, their analogue device has achieved accuracy comparable to digital systems, potentially helping address computing challenges for AI and 6G communications. The team demonstrated this by using their device to detect wireless communication signals, finding that system performance rivaled that of digital processors.

The researchers reported that their device has already surpassed top-tier GPUs in solving medium-scale matrix equations. Further improvements in circuitry could enhance performance even more, suggesting this represents just the beginning of what's possible.

Solving advanced mathematical problems proves critical for fields including scientific computing, signal processing, and training neural networks. As AI models grow increasingly complex and data-hungry, the energy and time costs of digital computation become prohibitive. Analogue computing offers a potential path forward that doesn't rely on continuously shrinking transistors or increasing power consumption.

WATCH ALSO: https://www.modernmechanics24.com/post/boston-dynamics-reveals-secrets-behind-humanoid-robot-grippers

The broader implications extend beyond raw performance numbers. If analogue chips can deliver on their promise while maintaining digital-level precision, they could fundamentally reshape how we approach computationally intensive tasks. Training massive AI models, running complex simulations, processing real-time communications data – all could become dramatically more efficient.

Whether this specific chip becomes commercially viable remains uncertain. But the Peking University team has demonstrated that analogue computing's "century-old problem" might finally have a solution – and the future of computing might look surprisingly like its past.