Tech

Chinese researchers debut world’s first AI-based processor chip design system

Share
Share
Chinese researchers debut world's first AI-based processor chip design system
Left: Multimodal architecture of LPCM capable of understanding, representing, and generating both text and graph data. Right: Feedback-driven inference of LPCM with a dual-loop mechanism, consisting of an outer performance feedback loop and an inner functional correctness feedback loop. Credit: arXiv (2025). DOI: 10.48550/arxiv.2506.05007

A team of engineers, AI specialists and chip design researchers at the Chinese Academy of Sciences has designed, built and tested what they are describing as the first AI-based chip design system. The group has published a paper describing their system, called QiMeng, on the arXiv preprint server.

Over the past several decades, integrated circuit makers have developed systems for developing processor chips for computers, smartphones and other electronic devices. Such systems tend to be made up of large teams of highly skilled people who can take design ideas (such as faster computing or running AI apps) and turn them into physical designs that can be fabricated in specially designed factories. The process is notoriously slow and expensive.

More recently, computer and device makers have been looking for ways to speed up the process and to allow for more flexibility—some may want a chip that can do just one thing, for example, but do it really well. In this new study, the team in China has applied AI to the problem.

The work involved using an LLM to take user requests regarding performance standards and turn them into architectural plans for a processor chip that would meet the specifications and also create the software that runs on it.

The initiative was reportedly begun as scientists in China faced pressure to ramp up their chip-making abilities as Western countries have become less willing to share technological gains with them.

The new system has three interconnected parts. The first is a domain-specific chip model. The second is the design agent responsible for most of the work in building a design. The third part is an assemblage of design applications that are made available to the design agent. As with other AI-based learning applications, the system learns what a processor looks like and how it works by being exposed to a large database of existing technology.

The system has proved itself capable of creating effective designs—it came up with chips similar in capability to Intel’s 486 and Arm’s Cortex A53. Both chips are notably quite dated and lag far behind what is in use today. Still, the researchers are optimistic that QiMeng will evolve as it learns and will eventually be capable of designing chips on a par with those now being developed in the West at far less expense and over a much shorter timeline.

More information:
Rui Zhang et al, QiMeng: Fully Automated Hardware and Software Design for Processor Chip, arXiv (2025). DOI: 10.48550/arxiv.2506.05007

QiMeng Project Website: qimeng-ict.github.io/

Journal information:
arXiv


© 2025 Science X Network

Citation:
Chinese researchers debut world’s first AI-based processor chip design system (2025, June 11)
retrieved 11 June 2025
from

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.

Share

Leave a comment

Leave a Reply

Your email address will not be published. Required fields are marked *