Researchers from Google have uncovered that artificial intelligence (AI) could potentially design computer chips more efficiently than human engineers.
In a research paper published in Nature, Google Research explained that designing a physical layout of a computer chip, known as floorplanning, is a complex and time-consuming task but essential.
To help achieve a more efficient outcome, the researchers developed a series of algorithms to treat chip floorplanning as a game, components as pieces, the canvas on which the components sit on as the board, and the winning results were based on the performance which was assessed using evaluation metrics based on reference datasets of 10,000 chip placements.
During a six-hour period, the algorithms were able to develop chips that were similar or more efficient that those designed by humans, who typically take months, the researchers uncovered.
“Our method generates manufacturable chip floorplans in under 6 hours, compared to the strongest baseline, which requires months of intense effort by human experts,” the researchers said.
Not only does the method save time, but the researchers also highlighted that the quality of the work also improved “because it is exposed to more instances of the chip placement problem”.