HMN 2025: What is the world’s first AI-based processor chip design system

A group of engineers, AI specialists and chip design researchers on the Chinese Academy of Sciences has designed, constructed and examined what they’re describing as the primary AI-based chip design system. The group has printed a paper describing their system, referred to as QiMeng, on the arXiv preprint server.

Over bygone days a number of many years, built-in circuit makers have developed techniques for growing processor chips for computer systems, smartphones and different electronic devices. Such techniques are usually made up of enormous groups of extremely expert individuals who can take design concepts (resembling quicker computing or operating AI apps) and switch them into bodily designs that may be fabricated in specifically designed factories. The course of is notoriously sluggish and costly.

More just lately, laptop and machine makers have been in search of methods to hurry up the method and to permit for extra flexibility—some could desire a chip that may do only one factor, for instance, however do it rather well. In this new study, the group in China has utilized AI to the issue.

The work concerned utilizing an LLM to take consumer requests relating to performance standards and switch them into architectural plans for a processor chip that will meet the specs and in addition create the software program that runs on it.

The initiative was reportedly begun as scientists in China confronted strain to ramp up their chip-making skills as Western international locations have change into much less keen to share technological beneficial properties with them.

The new system has three interconnected elements. The first is a domain-specific chip model. The second is the design agent accountable for many of the work in constructing a design. The third half is an assemblage of design purposes which can be made accessible to the design agent. As with different AI-based {learning} purposes, the system learns what a processor seems to be like and the way it works by being uncovered to a large database of present expertise.

The system has proved itself able to creating efficient designs—it got here up with chips comparable in functionality to Intel’s 486 and Arm’s Cortex A53. Both chips are notably fairly dated and lag far behind what’s in use at present. Still, the researchers are optimistic that QiMeng will evolve because it learns and can ultimately be able to designing chips on a par with these now being developed within the West at far much less expense and over a a lot shorter timeline.

© 2025