Google researchers have found that artificial intelligence (AI) could potentially design computer chips more efficiently than human engineers.
In a research paper published in the journal Nature, Google Research explains that designing the physical layout of a computer chip is a complex and time-consuming task, but essential.
To achieve a more efficient result, the researchers developed a series of algorithms that treat chip layout planning like a game, components like parts, the canvas on which components like the board rest, and winning results are based on performance that was evaluated using benchmarks based on benchmark data sets of 10,000 chip implantations.
A proven method
Over a six-hour period, the algorithms were able to develop chips similar to, or more efficient than, those designed by humans, which typically take months, the researchers found. “Our method generates manufacturable chip floor plans in less than 6 hours, against the strongest baseline, which requires months of intense effort from human experts,” the researchers explain.
Not only does the method save time, but the researchers also point out that the quality of the work has also improved, “because it is exposed to more cases of the chip placement problem.”
According to the researchers, the method has already been adopted by Google, and they add that it has been used for the production of a latest generation tensor processing unit.
Google first revealed it was experimenting with AI to design its chips last year. Jeff Dean, head of AI research at Google, told at the time that it could cut costs and produce more efficient models: “We’re asking our designers to experiment with it and see how they start. to use it in their workflows ”.