AI Just Helped Design A CPU, Here's Why That Matters

"Chip design is not rocket science. It's much more difficult," is a running joke in the industry. But in an age where AI is helping solve some of the biggest challenges across medical science and economics, it was only a matter of time before the first big breakthrough at the intersection of AI and semiconductors arrived. This time, it didn't originate in Silicon Valley, but came out of China. Experts at the Chinese Academy of Sciences (CAS) used an AI system called QiMeng to design a processor chip.

The novel system can "fully automate the design of an entire CPU starting from I/Os." At the heart of the system is a Large Processor Chip Model (LPCM) that powers dedicated Hardware Design Agent and Software Design Agent components, both of which are applied in real-world application scenarios. In a research brief, the experts note that QiMeng is a viable solution for hardware and software design, adding that it will boost efficiency and intelligence in a field that is notoriously complex and keeps getting more sophisticated.

The team is now hoping to achieve industrial-grade automated processor chip design and guide the "entire processor chip domain" toward automation. So far, the team has already produced two processors, QiMeng-CPU-v1 and QiMeng-CPU-v2, which have managed to match the performance of an Intel 486 processor (from the 1990s) and an Arm Cortex-A53 processor (from the 2010s), respectively.

Chip designing is a notoriously slow and expensive process, and let's not forget the sheer amount of expertise it requires to get things right, let alone make a breakthrough. "I realized that making them is the most fascinating and complicated manufacturing problem in human history," Chris Miller, author of "Chip War", explained to El Pais.

What the team at the Chinese Academy of Sciences accomplished was essentially to take a large language model (LLM) — the underlying tech behind AI chatbots such as ChatGPT – train it to process queries about performance specifications of a processor, and turn it into an end-to-end architectural plan for making the chip while simultaneously creating the software for it. It also entailed training the system on a vast trove of information regarding the ins and outs of modern-age semiconductors. At the end of the day, it worked. As per the team, QiMeng's automated design system is "capable of dramatically reducing manual intervention, significantly improving design efficiency while shortening development cycles and lowering costs."

Of course, the chip designed by QiMeng compares to processors that are vastly outdated — over a decade behind modern standards — but the scaling potential is huge. And so are the overarching benefits. But most importantly, they will help China circumvent the restrictions imposed on it by the U.S. and end its reliance on Western partners, just the way Huawei managed to stage a comeback in the smartphone segment using a self-designed chip and software.

Semiconductor supremacy, especially in the age of AI, is one of the biggest geopolitical bargaining chips right now. It, therefore, doesn't come as a surprise that the U.S. doesn't want China to get the upper hand, especially after seeing homegrown products like DeepSeek and Qwen challenge early leaders like OpenAI's ChatGPT and Google Gemini on frugal resources. Accordingly, the U.S. has imposed restrictions on the world's leading electronic design automation (EDA) suppliers, such as Cadence, Synopsys, and Siemens, from doing business with China. On the commercial side of things, Nvidia — the global leader in providing critical AI accelerator chips — is also facing trade restrictions to prevent China from taking a lead in the white-hot AI race. 

Breakthroughs like QiMeng are a clear sign that China aims to not only become self-sufficient but also speed up one of the most challenging aspects of computing progress. As part of the tests, QiMeng was able to complete the entire front-end design of a 32-bit RISC-V CPU in just five hours, which is quite impressive. 

AI tools like QiMeng could help China overcome fundamental bottlenecks and make a massive leap, but it won't be the lone player in the arena. Leading EDL players like Synopsys and Cadence are also exploring how to use AI to speed up the engineering work. As Moore's Law for defining semiconductor advancements nears its end and chip architecture moves into a more complex domain, AI assistants like QiMeng could very well prove to be the secret weapon for progress.