NVIDIA

Semiconductor giant releases new research on using AI to improve chip designs

SAN FRANCISCO (Reuters) - Nvidia Corp, the world leader in computer chips used to create artificial intelligence, on Monday unveiled new research that explains how AI can be used to improve chip design.

The process of designing a chip involves deciding where to place tens of billions of tiny switches called transistors on a piece of silicon to create working circuits. The exact location of these transistors has a big impact on the cost, speed, and power consumption of the chips.

Chip engineers use complex design software from companies like Synopsys and Cadence Design Systems to help them optimize the placement of these transistors.

On Monday, Nvidia released a paper showing it could use a combination of artificial intelligence techniques to find better ways to place large groups of transistors. The article aims to improve on a 2021 publication by Google, whose findings later became the subject of controversy.

Nvidia's research took an existing effort developed by researchers at the University of Texas using what's called reinforcement learning and added a second layer of artificial intelligence on top of it for even better results.

Nvidia's chief scientist, Bill Dally, said the work is important because improvements in chip manufacturing are coming down with per-transistor costs in newer generations of chip-making technology now higher than in previous generations.

The idea runs counter to Intel co-founder Gordon Moore's famous prediction that chips would always get cheaper and faster.

"You're not getting savings of that scale anymore," Dally said. "To keep moving forward and deliver more value to customers, we can't get it from cheaper transistors. We have to get it by being smarter in design."