In the "AI craze", chips led by GPUs grasp the "lifeline" of AI technology development and promote continuous iteration of technology. At the same time, the development of AI has also begun to feed back chip manufacturing, and now AI has generated an industrial scale RISC-V CPU core in just 5 hours.

Recently, the National Key Laboratory of Processor Chips and its cooperating units at the Institute of Computing, Chinese Academy of Sciences, designed the 32-bit RISC-V CPU "Enlightenment 1" using AI technology - the world's first fully automated CPU chip without human intervention.

This CPU adopts a 65nm process, with a frequency of 300MHz, and can run the Linux operating system. Its performance is comparable to that of the Intel 80486SX, and the design cycle is shortened to 1/1000.

With the help of AI technology, researchers generated 4 million logic gates in just 5 hours. Industry media pointed out that this is 4000 times the size of the circuit that GPT-4 can currently design.

The team stated that their training process takes less than 5 hours to achieve validation testing accuracy of>99.999999999%.

The "Enlightenment 1" is designed based on the BSD Binary Speculation Diagram algorithm. Researchers use AI technology to automatically generate CPU designs directly from "input output (IO)" without the need for engineers to provide any code or natural language descriptions.

In other words, it transforms the CPU automatic design problem into a "circuit logic generation problem that meets input-output specifications", where only test cases are needed to directly generate circuit logic that meets the requirements, eliminating the logic design and verification processes in traditional design processes.

It is worth noting that this method even independently discovered Feng? Neumann architecture (a computer design concept structure that combines program instruction memory and data memory).

The CPU was streamed in December 2021 and successfully ran the Linux operating system and SPEC CPU 2000 program after being sliced back. The related paper was released at the end of June this year.

Chip giants compete to apply AI to help chip design

Advanced chip manufacturing must go through over 1000 steps. Each stage requires complex calculations, and every step must be almost perfect, as is naturally the case with CPU design and manufacturing. In this challenging task, it usually requires a team of engineers to write code and then complete circuit design with the assistance of EDA tools. In this process, it is necessary to repeatedly test and verify the optimization.

Due to the extremely complex and labor-intensive process, a large chip design project can sometimes take months or even years to complete - taking the Intel K486 CPU as an example, the validation phase alone takes 190 days.

With the development of AI technology, more and more companies are beginning to introduce it into chip design and manufacturing.

Jensen Huang, the founder and CEO of Nvidia, emphasized the potential of Nvidia's accelerated computing and AI solutions in chip manufacturing. He believed that chip manufacturing was the "ideal application" of accelerated computing and AI computing.

Another chip giant, AMD's Chief Technology Officer Mark Papermaster, also revealed that AMD has started using AI in semiconductor design, testing, and verification stages, and plans to use generative AI more widely in the field of chip design in the future. Meanwhile, AMD has been experimenting with GitHub Copilot (developed in collaboration with GitHub and OpenAI) and researching how to better deploy this AI assistant.

Japanese semiconductor company Rapidus President Junichi Koike has stated that he will introduce artificial intelligence and automation technology to establish a mass production process with approximately 500 technical personnel. The company has the idea of having complete talent, equipment, and technology, and is expected to start mass production in 2027.

Indeed, in the chip design process, AI "does well" and can iterate indefinitely until the best solution is found. Not only that, while iterating, AI also learns and studies what patterns can create the optimal design. Therefore, AI actually accelerates the speed of chip design optimization layout and brings higher performance and lower energy consumption. In the verification and testing process, AI can also maximize test coverage and save time.

As for whether AI will take away the "job" of chip development engineers? The answer given by AMD Chief Technology Officer Papermaster is no. He believes that AI will not replace chip designers, as it will serve as an auxiliary tool with great potential to help accelerate design.