Google Uses AI to Design More Efficient Microchips in Less Than Six Hours

June 17, 2021
By: Laura Stotler

A new study from Google (News - Alert) reveals that artificial intelligence (AI) is capable of designing computer microchips at least as well as humans, and thousands of times faster. Google is using AI software to design microchips for its next generation of computer systems.

The study focused on floor planning, the process of designing the physical layout of a chip's parts. This development step is critical to a device's performance and typically requires months of planning from skilled workers. Google researchers participating in the study used AI software to treat the floor planning process as a game.

According to Azalia Mirhoseini, co-lead author of the study and a computer scientist at Google Research's Brain Team, the AI software viewed the chip's computation cores, memory controllers and other elements as the pieces of a game. The canvas they are mounted on acted as the board, and the AI produced winning results based on reaching optimal performance. That was calculated based on values like chip size and power usage.

The AI analyzed a greater number and variety of microchips as the study continued, becoming progressively faster and better at chip design. Previous attempts at automation were not successful because the software was not learning from each chip it analyzed, according to Anna Goldie, co-lead author of the study and a computer scientist at Google's Brain Team.

The Google researchers claim the new method can generate manufacturable chip floor plans that are comparable or superior to the ones made by human experts in less than six hours. The new chips are evaluated based on performance, energy consumption and chip area. The end result would potentially cut thousands of hours of human effort spent creating each generation of microchips.

Google is already manufacturing the AI-designed microchips and plans to continue to research the impact of using even more powerful AI software. Researchers added that the floor planning strategy used in chip design could also be applied to more complex issues like city planning and vaccine testing and distribution, where best uses of limited sets of resources are vital to success.




Edited by Luke Bellos


Original Page