The Buzz at This Year's Hot Chips Conference: AI Is Supercharging Chip Design

AI is speeding up chip design. That was a key theme in conversations I had this week with engineers, researchers and a handful of startups pitching AI-driven design help at the annual Hot Chips semiconductor conference in Palo Alto, Calif. 

OpenAI’s Jalapeño chip, which it claims is better than Nvidia’s Blackwell chips at AI inference, is good in part because its Sol and Astra models helped with the design and software process, the OpenAI team said in a presentation. And Google’s tensor processing unit team credited its DeepMind researchers for using AI to help make TPU v8, its latest chips, 6% more power efficient and also 6% more powerful. Nvidia of course has also leaned into AI-driven chip design, with an array of software libraries that apply AI to everything from the stencil layout for how circuits are lasered onto a chip to how data moves between the millions of microscopic parts of a chip.

Design startup Agentrys, which on Wednesday announced it has raised $25 million, aims to help chip designers use AI agents to automate the process and teach chip designs to improve themselves. Its CEO Mark Ren, who led Nvidia’s design automation efforts for the past decade, believes the future of chip design will be driven by “very powerful agentic systems that can do chip design on their own nearly from start to finish,” he said.