Vibeleaderboard
← All Intel
Intel / video

AI That Designs Its Own Chips: Ricursive's Anna Goldie and Azalia Mirhoseini

Source
youtube.com
Author
Sequoia Capital
Date

Context

Anna Goldie and Azalia Mirhoseini, co-creators of AlphaChip at Google, started Ricursive Intelligence to apply AI to chip design, and in a Sequoia Capital talk they explained their first bet. AlphaChip used reinforcement learning to generate chip layouts; they said it was published in Nature, used in the last four generations of Google's TPU chips, and adopted by outside firms such as MediaTek. Those adoption claims are theirs.

They target two slow steps, each taking up to a year with hundreds or thousands of experts by their account: physical design, which places and wires billions of components, and design verification, which checks the chip's logic. Because commercial tools can take days per iteration and AI learns best from fast loops, Ricursive first rebuilds the tools to run far faster. Mirhoseini showed a timing-analysis engine she said matches commercial tools closely at about a thousand times the speed. The talk offers a demo and company claims, not independent measurements.

Key takeaways · AI-distilled
  • Anna Goldie and Azalia Mirhoseini say their AlphaChip RL placement work was used in tape-outs of the last four TPU generations, Google's Axion CPUs, Pixel phones and autonomous-vehicle chips, and adopted by MediaTek.
  • Their new company targets the two longest poles in chip design, physical design and design verification, each of which they say can take up to a year and hundreds or thousands of experts.
  • Their approach is to rebuild EDA tools to run far faster so AI can iterate on them; they showed a static timing analysis engine they say correlates closely with commercial tools while running about 1,000x faster.
  • They describe AI-generated placements as curved and organic-looking rather than the regular layouts humans favor, which they say reduces wire length and improves performance.
Read the source www.youtube.com
More from Sequoia Capital
Recommended reads
Comments

Checking sign-in…

Loading comments…