Peking University's New Analog Chip Emer

Peking University's New Analog Chip Emerges: 1,000x Computing Power Leap, Ushering in a Triple Revolution in AI

 

On October 13, 2025, a research result published in the international top journal Nature Electronics dropped a bombshell on the global chip and artificial intelligence (AI) sectors. A team led by Researcher Sun Zhong from the Institute of Artificial Intelligence at Peking University, in collaboration with the School of Integrated Circuits, successfully developed a high-precision analog matrix computing chip based on resistive random-access memory (RRAM). This chip not only achieves an analog computing system with precision comparable to digital computing for the first time but also delivers a disruptive breakthrough in core performance: when solving a 128×128 matrix inversion problem, its computing throughput exceeds that of top-tier digital processors by more than 1,000 times—tasks that would take a traditional GPU 24 hours to complete can be finished in just 1 minute. Moreover, at the same precision level, its energy efficiency is over 100 times higher than that of traditional digital processors. This breakthrough is not merely a "performance upgrade" but will bring comprehensive transformations to the AI field.

 

AI Large Model Training: From "Time-Consuming Bottleneck" to "Minute-Level Breakthrough"

 

For a long time, the training of large AI models has been limited by computing power. Particularly, second-order optimization algorithms, which demand extremely high computational resources, have struggled to gain popularity because traditional GPUs cannot handle their complexity. However, the high-throughput feature of Peking University's new chip directly breaks this barrier. A 1,000-fold increase in computing power enables second-order optimization algorithms to move from "theoretically feasible" to "practical application." This not only compresses the model training cycle by several times but also supports the development of large models with larger parameter scales and more complex structures, providing core computing power for the breakthrough of general artificial intelligence (AGI).

 

Computing Power Cost and Energy: Solving the "High AI Consumption" Dilemma

 

As the demand for AI computing power grows exponentially, energy consumption and operational costs of data centers have become industry pain points. Traditional digital processors often come with extremely high energy consumption when delivering high computing power outputs. In contrast, the energy efficiency advantage of Peking University's new chip is a veritable "cost-reduction tool": with over 100 times higher energy efficiency at the same precision, it means that under the same computing power demand, the electricity cost of data centers can be reduced by more than 90%, while significantly lowering the construction and maintenance costs of cooling systems. This feature perfectly aligns with the global trend of "green computing power," allowing the large-scale application of AI technology to no longer be constrained by the dual limitations of energy and cost.

 

AI Application Scenarios: From "Cloud Dependence" to "Edge Autonomy"

 

In practical AI applications, edge devices (such as robots, drones, and smart sensors) often need to rely on cloud computing for data processing due to limitations in computing power and energy consumption, leading to issues like response delays and privacy security. The low-power, high-computing-power characteristics of Peking University's new chip bring a revolutionary breakthrough to edge computing—it can support complex signal processing and AI "training-inference integration" to run directly on terminal devices. This enables smart devices to complete model training and inference locally without relying on the cloud. For instance, industrial robots can process production data in real time and independently optimize their movements, while drones can make rapid navigation decisions in complex environments. This will further expand the application boundaries of AI in industries, healthcare, communications, and other fields.

 

Notably, the potential of this chip in the 6G communication field is particularly prominent. 6G base stations need to be equipped with hundreds or even thousands of antennas and process massive amounts of signals in real time. Existing digital chips either lack sufficient computing power or consume excessive energy, while analog chips can perfectly balance the contradiction between "high computing power" and "low energy consumption." They provide technical support for core signal processing in 6G communications and will also promote the in-depth integration of AI and communication technologies.

 

From an industrial perspective, the global high-end computing power market has long been dominated by companies like NVIDIA. The breakthrough of Peking University's analog chip not only provides a brand-new technical route beyond digital computing but also gives China a first-mover advantage in the field of analog computing. In the future, the dual-technical route of "digital + analog" may reshape the industrial pattern of AI computing power, helping China gain key discourse power in the global computing power revolution.

 

This new chip developed at Peking University is not just a technological breakthrough but also heralds new possibilities in the era of AI computing power. Faster, more cost-effective, and more flexible computing power support will drive AI from the "laboratory" to "all industries," truly making it a core driver of social progress.