Google is using artificial intelligence to design chip floor plans faster than humans

Google claims that it has developed artificial intelligence software that can design computer chips faster than humans.

The tech giant said in an article in the journal Nature on Wednesday that a chip that would take humans months to design can be devised by its new AI in less than six hours.

AI has already been used to develop the next iteration of Google’s tensor processing unit chips, which are used to execute AI-related tasks, Google said.

“Our approach has been used in production to design the next generation of Google TPUs,” wrote the article’s authors, led by Google’s co-directors of machine learning for systems, Azalia Mirhoseini and Anna Goldie.

To put it another way, Google is using AI to design chips that can be used to create even more sophisticated AI systems.

Specifically, Google’s new AI can make a “floor plan” of a chip. This basically involves mapping where components like CPU, GPU, and memory are placed on the silicon matrix in relation to each other; its position on these tiny boards is important as it affects the chip’s power consumption and processing speed.

It takes humans months to optimally design these blueprints, but Google’s deep reinforcement learning system, an algorithm that is trained to take certain actions in order to maximize its chances of getting a reward, can do so with relatively little. effort.

Similar systems can also defeat humans in complex games like Go and chess. In these scenarios, the algorithms are trained to move pieces that increase your chances of winning the game, but in the tile scenario, the AI ​​is trained to find the best combination of components to make it as efficient as possible from the point of view. computational. The artificial intelligence system was fed 10,000 chip floor plans to “learn” what works and what doesn’t.

While human chip designers often place components in neat lines, Google’s AI uses a more scattered approach to designing its chips. This is not the first time that an artificial intelligence system has gone rogue after learning how to perform a task from human data. DeepMind’s famous AI “AlphaGo” made a very unconventional move against Go world champion Lee Sedol in 2016 that astonished Go players around the world.

Google engineers noted in the document that the advance could have “major implications” for the semiconductor industry.

Facebook’s chief artificial intelligence scientist, Yann LeCun, praised the research as “very nice work.” On twitter, adding “this is exactly the kind of environment in which RL shines.”

The breakthrough was hailed as a “major achievement” that “will go a long way in accelerating the supply chain” in a Nature editorial on Wednesday.

However, the magazine said that “technical expertise must be shared widely to ensure that the business ‘ecosystem’ becomes genuinely global.” He went on to emphasize that “the industry must ensure that time-saving techniques do not alienate people with the necessary basic skills.”

Clarification: This story has been updated to reflect that Anna Goldie is a co-author of the article and that AI has been used to develop the next iteration of Google’s Tensor Processing Unit chips.

Add Comment