Google claims to have developed artificial intelligence software that can design computer chips faster than humans.
The tech giant said in an article in Nature on Wednesday that a chip that would take man months to develop can be dreamed up of its new AI in less than six hours.
The AI has already been used to develop the next iteration of Google’s Tensor Processing Unit chips, which will be used to perform AI-related tasks, Google said.
“Our methodology was used in production to design the next generation of Google TPU,” write the paper’s authors, led by co-leaders of machine learning systems at Google, Azalia Mirhoseini and Anna Goldie.
In other words, Google is using AI to develop chips that can be used to create even more sophisticated AI systems.
Specifically, the new AI from Google can create the “floor plan of a chip”. This essentially involves plotting where components like CPUs, GPUs, and memory are placed in relation to each other on the silicon chip – their positioning on these tiny boards is important as it affects the power consumption and processing speed of the chip.
It takes people months to optimally design these floor plans, but Google’s deep reinforcement learning system – an algorithm trained to take specific actions to maximize its chance of a reward – can do it with relatively little effort.
Similar systems can also defeat people in complex games such as go and chess. In these scenarios the algorithms are trained to move figures that increase their chances of winning, but in the chip scenario the AI is trained to find the best combination of components in order to make them as computationally efficient as possible. The AI system was fed 10,000 chip floor plans to “learn” what works and what doesn’t.
While human chip designers typically arrange components in neat lines, Google’s AI uses a dispersed approach to design their chips. This isn’t the first time an AI system has gone rogue after learning to perform a task based on human data. DeepMind’s famous “AlphaGo” CI made a highly unconventional move in 2016 against Go World Champion Lee Sedol, who amazed Go players around the world.
Google engineers stated in the paper that the breakthrough could have “big impacts” on the semiconductor sector.
Facebook’s chief AI scientist, Yann LeCun, praised the research as “very nice work”. on twitter, adding, “This is exactly the kind of setting that RL shines in.”
The breakthrough was hailed as an “important achievement” that “will be of great help in speeding up the supply chain,” in a Nature editorial on Wednesday.
However, the magazine states that “the technical know-how needs to be disseminated widely to ensure that the ‘ecosystem’ of companies becomes truly global.” “The industry must ensure that the time-saving technologies do not drive away people with the necessary core competencies,” it continues.
Clarification: This story has been updated to reflect that Anna Goldie is a co-author of the article and the AI was used to create the next iteration of Google’s tensor processor chips.