How AI Could Build Computer Chips Faster

Computers making themselves?

Key Takeaways

  • A new method of designing chips using AI could save thousands of hours of human effort.
  • Google recently announced it’s developed a way to design chips with AI that will be used in a commercial application.
  • Some observers say the AI-design process will mean better chips at lower prices for users.
"AI" letters sitting on a 3D model of a computer chip

zf L / Getty Images

Researchers are using artificial intelligence to build computer chips faster. Industry insiders say the effort is likely to lead to better chips at lower prices for users. 

Google recently announced it’s using AI to help design its next generation of machine learning chips. After years of research, the company’s AI efforts are paying off and will be used in an upcoming chip meant for AI computation, according to a paper published in the journal Nature

"The beauty of autonomous chip design is that it significantly reduces the barrier to entry for companies to access the power of AI chips because fewer designers are needed to produce a high quality and application-optimized design," Stelios Diamantidis, a senior director of Synopsys Artificial Intelligence Solutions, which produces AI software for chip design, said in an email interview. 

"Ultimately, it will result in more convenience, safety, automation, and seamless communications across just about every aspect of our lives at a lower cost and in a wider variety of applications."

Computers Building Computers

Google is using AI to build better versions of AI by planning a chip’s design. The software finds the best place to put components like CPUs and memory, which is challenging to do at such tiny scales. 

"Our method has been used in production to design the next generation of Google TPU," wrote the authors of the paper, led by Google’s co-heads of machine learning for systems, Azalia Mirhoseini and Anna Goldie.

"Ultimately, it will result in more convenience, safety, automation, and seamless communications across just about every aspect of our lives."

Google researchers claimed that AI design could have "major implications" for the chip industry. According to the scientists, the new Google method can generate manufacturable chip plans in less than six hours that are comparable or superior to those made by experts in all essential details, including performance, energy consumption, and chip area. The method could save thousands of hours of human work for every generation of microchips.

Facebook’s chief AI scientist, Yann LeCun, praised the paper as "very nice work" on Twitter, saying "this is exactly the type of setting in which RL shines."

Like a Game of Chess

Designing a chip can take humans weeks of experimentation, Diamantidis said. He likened the process to a game of chess, an area where AI already has beaten humans. 

"To give you a sense of the complexity of a typical modern integrated circuit (IC) design, consider the following comparison," he added. "In the game of chess, there are roughly 10 to the 123rd [power] number of states or potential solutions; in the placement process of designing a current day chip, it’s 10 to the 90,000th."

"The beauty of autonomous chip design is that it significantly reduces the barrier to entry for companies to access the power of AI chips."

Diamantidis predicts AI designs could push chip performance and energy efficiency to more than 1,000 times current levels. 

"Searching this vast space is a very labor-intensive effort, typically requiring many weeks of experimentation and often guided by past experiences and tribal knowledge," he added. "AI-enabled chip design introduces a new, generative optimization paradigm that uses reinforcement-learning (RL) technology to autonomously search design spaces for optimal solutions."

AI design of chips is rapidly growing, Diamantidis said. Synopsys is a leading supplier of AI-enabled chip design tools, and its customers are every major semiconductor and electronics company in the world, he claimed. These companies are either supplying chips to or developing mobile devices, high-performance computing systems and data centers, telecommunications equipment, and automotive applications.

AI concept showing a picture of a human brain on a computer chip

Yuichiro Chino / Getty Images

"We can't name specific customers, but just in the past few months, adopters of our AI tools have been able to set, and then immediately beat, world records in design productivity, being able to achieve with a single engineer in weeks what it used to take entire teams of experts months," Diamantidis said. 

Ultimately, users will be the ones to benefit from better chip designs, Diamantidis said. He added that "all of this is being driven by our desire to process more data, automate more functions in the products we use, and integrate more intelligence in almost everything that touches our lives."

Was this page helpful?