Ricursive Intelligence pitches AI that designs more efficient chips
At TechCrunch Disrupt, Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini explained their startup's goal of using AI to automate chip design. They argue that AI can discover unconventional layouts and optimize whole chips, improving efficiency as compute demand grows. The founders say this approach could let humans focus on creative applications while AI handles execution.
Ricursive Intelligence's founders, Anna Goldie and Azalia Mirhoseini, presented at TechCrunch Disrupt on Wednesday. Their startup aims to build AI that designs chips, learns from each design cycle, and improves later versions, automating and speeding up the process. They cited projections of at least 10,000x more compute demand within five years, calling that estimate possibly conservative.
They said AI can find nonstandard layouts, citing AlphaChip's unexpected circular, donut-like component arrangement, which engineers initially found surprising but later found successful. Rather than splitting chips into many blocks, Ricursive is optimizing whole chips and exploring intricate, wave-like patterns. Faster design could enable custom chips for space, AR, VR, and biological computing.
If AI-assisted chip design shortens timelines and improves energy use, it could affect chip engineers, data-center operators, and technology users by changing how hardware is planned and produced. Faster custom silicon may broaden access to specialized computing for fields such as AR, VR, space, and biological research. However, greater automation may also shift some design work