Ricursive Intelligence aims to use AI to get more out of data-center bandwidth, wafers, and fabrication capacity through chip design, instead of relying on additional data centers to satisfy rising compute demand. Its founders, Anna Goldie and Azalia Mirhoseini, see improved chips as a possible answer to the compute shortage. Getting more from silicon is still their proposal, rather than a proven solution.
At TechCrunch Disrupt on Wednesday, the founders outlined the system they’re building: AI that designs chips and uses what it learns from each attempt to improve the next. They want to make chip development faster and automate the process.
“There is this understanding that in the next five years we are going to have at least 10,000x more demand for compute, and that’s somewhat potentially conservative,” Goldie said on stage. “So we don’t want to build 10,000 times more data centers. We think that we need to help by making chips that are more efficient, that use the bandwidth in those data centers more efficiently, that use the wafers and fabs more efficiently, because even if we make [data centers] in space or something, I don’t know, you won’t see the sun anymore.”
The pair’s earlier AI system, AlphaChip, has already generated a layout engineers hadn’t necessarily anticipated: chip components arranged in a circular, “donut-shaped” pattern. The founders said engineers were surprised at first, but the arrangement worked. In their view, AI can discover unconventional layouts that perform better than those produced through traditional methods.
With Ricursive, the founders said they’re tackling whole-chip optimization, rather than breaking chips into dozens of smaller blocks as conventional design methods do. Working at that scale lets AI produce complex patterns, including wave-like designs that the pair intend to explain in more detail later. AlphaChip demonstrates successful AI-assisted component placement, but Ricursive is still working toward its wider goal of learning across designs and automating chip development.
As for concerns about AI designing its own chips, Goldie argued that handing design execution to AI could free people to work on applications and investigate new ways to use computing.
“I think the way everything is shifting is that humans come bring their great ideas and creative thoughts, and execution is handled by AI,” Goldie said. “And this is happening in chip design, software development, and other places. And that’s something that recursive does for chip design.”
Better versions of existing layouts aren’t the founders’ only aim. They believe faster design cycles could bring previously impractical chip types within reach by changing the economics of custom designs.
“If it takes six months or a year or two years to design a new chip, then you really have to be extremely selective about the chips that you design,” Goldie said. “But if you can design a chip in like a week, and then say for manufacturing, you can design custom chips for all sorts of different types of applications in like space and AR, VR, and you biological computing, and it makes sense financially to do that, and we can be more energy efficient.”
A design cycle of about a week describes a possibility, not a production result Ricursive has reported. Whether its iterative approach can deliver financially viable, more energy-efficient custom chips for space, AR, VR, and biological computing is still an open question.
