Ricursive Intelligence says AI can help design more efficient chips

In short: Ricursive Intelligence says it is using AI to design computer chips faster and more efficiently to help meet growing demand for computing power.

What happened

Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini spoke at TechCrunch Disrupt about their plan to use AI to design computer chips. Chips are the small parts inside phones, laptops, and data centers that do the actual calculating.

They said demand for “compute” is rising quickly. Compute is just a way of saying how much processing work computers can do, like how much horsepower an engine has. Goldie said demand could be 10,000 times higher within five years, and argued that the answer cannot simply be building 10,000 times more data centers.

Instead, the founders want AI to help design chips that do more work using less energy and fewer manufacturing resources. They described an approach where the AI designs the whole chip at once, rather than splitting the work into many smaller parts, which is a common method today.

As an example of how AI can produce unexpected results, they pointed to AlphaChip, an earlier system they worked on. It produced a circular, “donut-shaped” layout of chip parts that surprised engineers at first but later worked well.

Goldie also suggested that if chip design time drops from months or years to about a week, it could become practical to make custom chips for specific uses, including areas like space and AR and VR (headsets that add digital images to the real world, or replace it).

Why it matters

Most new AI services rely on large amounts of computing power, and that often means higher energy use and more expensive infrastructure. If chips can be designed to run the same tasks more efficiently, it could lower costs and reduce pressure to build more data centers.

Source: TechCrunch AI