Google is using artificial intelligence (AI) to more quickly design its next generation of machine learning chips. According to Google engineers, the manufacture of these chips by the algorithm is comparable or even superior to those created by humans. In addition, The company claims that AI is capable of doing the work that humans do in months in less than 6 hours.
According a study published in the journal Nature, Google has been studying during years how to use artificial intelligence to create chips. Now it seems that this effort is beginning to be applied to the production of a commercial device: its TPU (Custom Tensor Processing) chips.
Google engineers emphasize that this technology has “important implications”For the chip industry. With it, companies can more quickly design their chips and customize them more easily for specific workloads.
Nature dares to point that this research is a “important achievement”As it will help find a way to increase the performance of the chips.
To create this Google algorithm, the company uses something known as ‘floorplanning’. Typically, this process requires human designers working with the help of computer tools to find the ideal silicon die design for subsystems on a chip.
These components include CPU, GPU, and memory cores and must decide where to place each element to achieve the highest efficiency and ultimate speed of the chip. Google engineers explain that, generally, designing these plans involves “months of intense effort”. However, with new technology they plan to tackle this process more easily.
To train the algorithm, they employed reinforcement learning and provided the AI with a data set of 10,000 chip floor plans of varying quality. This technology used that data to distinguish between good and bad floor plans and generate its own designs..
Sign up for our newsletter and receive the latest technology news in your email.