By AI for AI

TPU pod racks. © Google.

A Nature journal article reports that Google is using artificial intelligence to design its next generation of TPU (tensor processing unit) chips intended for machine-learning. In other words, AI accelerates the development of AI. Algorithms now handle the “floorplanning” stage, involving the optimal arrangement of the subsystems on a chip, which is a delicate process because each choice affects speed and overall efficiency. Whereas it used to take teams months of design, this machine-learning method generates chips in less than six hours that are “comparable or superior” to those created by humans, according to Google. An editorial also in Nature proclaims this research an “important achievement” to help mitigate the forecasted end of Moore’s Law (stating that the number of transistors on a chip doubles every two years), given that we are reaching the physical limits of what can be achieved in extreme miniaturization.

⇨ The Verge, James Vincent, “Google is using AI to design its next generation of AI chips more quickly than humans can.”

2021-06-10