What if we let computers decide what would be best for computers? This is the idea that Google had when asking an AI to design microprocessors itself. And the result is stunning.
In a recent article published in the journal Nature, Google engineers claim to have created an AI capable of surpassing all of the world’s best engineers in building microprocessors. A small feat that could have important consequences in the world of technology.
« Our method generates chip designs in less than 6 hours, where the strongest human organization requires months of intense effort. “, Immediately announces the study of Google. By training a neural network with 10,000 chip models already released, the research giant has been able to drastically reduce the time required for the technical development of this component.
“Co-design of the AI and the hardware”
The chips in our smartphones are small architectural miracles, which require finding a constant balance between power, complexity, size, heat, energy consumption and many other factors. In short, it is meticulous and complex work, which until then had always escaped the machines. The engineers behind the chips obviously used software to design their chips, but a single machine had never produced any valid engineering drawings. Until now.
There is always delay
To fully understand why this announcement has potential, we must understand how our technological devices are built today. Every phone, every tablet and every computer is loaded with microprocessors imagined years before their release. Since chip design by humans is particularly long, as we have seen, our new gadgets are always a little behind the best in the laboratory when they are released.
Google’s AI could therefore significantly reduce the time between the design of a chip and its time to market. ” Automating and speeding up the chip design process can also enable the co-design of AI and hardware. », Specifies the article in Nature. Clearly, rather than releasing an iPhone embedding a chip imagined two years ago, Apple could release a phone with a component embedding technologies from a few months ago. Enough to make software and hardware progress work a little more hand in hand and significantly accelerate innovation.
To prove that this case is serious, Google even explains that it has already used its AI to design certain chips present in its data centers.
Even manage vaccine distribution?
Better yet, Google’s AI could make it possible to test dozens of chips in parallel before they even arrive in a phone. This hobby was not possible until then, since the technical development of a single chip was already taking months. And to make sure nothing is spoiled, AI will continue to learn as and when the best choices are made in order to produce results that are constantly improving. This constant learning ” will allow chip designers to be assisted by AIs with more experience than any human can ever acquire »Explains Google.
Google even goes so far as to suggest that its AI would be able to improve ” urban planning as well as vaccine testing and distribution “. Problems that also combine limited resources, flow optimization and space constraints. A bit like the production of a chip, say engineers at Google.
In his editorial on this subject, Nature concludes by evoking a point not really addressed in the publication of Google: that of employment. ” It is important to take into account the wider consequences of this automation, in particular the need to offer opportunities for upgrading skills for those who are currently doing this work by hand. », Explains the newspaper. Ah yes, we would have almost forgotten.
The continuation in video
Source: Numerama by www.numerama.com.
*The article has been translated based on the content of Numerama by www.numerama.com. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!
*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.
*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!