Google AI Designs Computer Chips Faster Than Humans Can

Search engine giant Google has recently claimed that it developed a new artificial intelligence (AI) software capable of designing microchips much faster than humans can.

In a paper recently published in the journal Nature, the company claimed that it is utilizing machine learning in designing its next generation of Google tensor processing unit chips. According to Google, a work that takes months for humans to finish can now be accomplished by its AI in six hours or less.

For years, the company has been working to figure out how machine learning can be used to design and create chips. However, this latest effort appears to be the first one utilized in developing a commercial product.

Anna Goldie and Azalia Mirhoseini, two of the proponents of the new paper and co-heads of machine learning for systems at Google, shared some insights about this new AI. "Our method has been used in production to design the next generation of Google TPU," they said.

Google engineers also explained in the paper that this work has major implications for the chip industry. This new technology can unlock opportunities for co-optimization in the earlier stages of the chip design and development process. It will also enable companies to explore new possibilities when it comes to the architecture space for future designs. Additionally, customization of chips for particular workloads will be much easier. To put it simply, Google's AI will help accelerate the future of artificial intelligence development.

This research is considered an important achievement, according to an editorial piece published by Nature. In the said editorial, Nature said this kind of work could counteract the end of Moore's Law, which had been forecasted earlier. The publication also noted that Google's AI will "be a huge help in speeding up the supply chain."

Meanwhile, Yann LeCun, Facebook's chief AI scientist, praised the research in a series of tweets. "Very nice work from Google on deep RL- based optimization for chip layout," he said.

Essentially, Google's AI can draw the chip's floorplan, which includes plotting the placement of components like memory, CPU and GPU. It takes humans months to design these chip floorplans, but with Google's AI, the job can be done with less effort and time.

Google's AI won't necessarily solve the physical challenges of putting more transistors inside the chips. However, as The Verge said in a report, it can help find other approaches to boost performance at the same rate.