The Intel i486 microprocessor from 1989 compatible with the QiMeng-CPU-v1 designed using AI. | Image credit-Intel
China has uncovered an AI system called QiMeng that is used to speed up the development of chip designs. Large Language Models (LLMs) are used to design entire CPUs and thus far, this AI process has helped China create two processors. The QiMeng-CPU-v1, comparable to an Intel 486, and QiMeng-CPU-v2, which is said to rival an Arm Cortex A53 used on lower to mid-range devices.
Here’s the thing. The Intel 486 dates all the way back to 1989 although the v2, if indeed it does match the Cortex A53 in performance, would be a 23-year improvement since the Cortex-A53 dates back to 2012. What do we make of that? The project, run by the Chinese Academy of Sciences (CAS), should be able to use AI to help design a processor that is equal to one that has more current specs. Reportedly, what QiMeng can accomplish in days are tasks that take human teams weeks to do.
For example, a chip made for a automatic-driving vehicle could take weeks for a team made of humans to create. QiMeng could finish the same task in just days.
Two of the top chip designers in the Electronic Design Automation (EDA) industry, Cadence and Synopsys, have used AI in important chip design steps of design and verification. Synopsys goes much further than CAS. Synopsys.ai has helped in the design of over 200 taped out chips. A chip that has been taped-out has had its design finalized and is being sent to the foundry to be fabricated.
The LLMs represent the technology used by AI chatbots and QiMeng has appeared at a propitious time for China. With the U.S. pressuring leading EDA firms to stay clear of China, QiMeng’s ascension comes at the perfect time in China-when it is needed the most. The country has been seeking independence in the semiconductor industry. Still, EDA software suppliers, including the aforementioned Cadence Design Systems and Synopsys, and Siemens EDA have had new export controls placed on their products in China where the trio generated 82% of EDA revenue in the country.
While the goal of developing QiMeng was to reduce costs, speed up development cycles, and increase efficiency compared to using manual methods of chip design, QiMeng v3 will be the key. The next version of QiMengt will show whether the AI system can design a more modern processor and help the Chinese semiconductor industry make up for U.S. sanctions that will prevent leading American EDA firms from doing business in the country. Synopsys, Cadence, and Siemens EDA now require licenses from the U.S. Commerce Department to do any business in China.
