Ricursive Intelligence, an AI startup, says chips designed by artificial intelligence could help ease the growing shortage of computing power. The company’s founders shared their plans on stage at TechCrunch Disrupt in San Francisco on Wednesday.
Co-founders Anna Goldie and Azalia Mirhoseini are building AI that can design chips, learn from each design, and improve the next one. Their goal is to automate and speed up the chip design process.
Earlier this year, the company raised $335 million at a $4 billion valuation in four months, according to TechCrunch.
Demand for Compute Expected to Soar
Goldie said demand for computing power could grow at least 10,000 times over the next five years. She added that this estimate may be conservative.
She said the answer is not to build 10,000 times more data centers. Instead, she believes chips need to become more efficient.
That means chips that make better use of bandwidth inside data centers. It also means making better use of the wafers and fabrication plants used to produce them.
Goldie joked that even building data centers in space would not solve the problem. “You won’t see the sun anymore,” she said.
How AI Is Changing Chip Design
Some people may worry about AI designing its own chips. Goldie argued that handing chip design to AI will free humans to focus on building applications and exploring new uses for computing.
“Humans come bring their great ideas and creative thoughts, and execution is handled by AI,” she said. She added that this shift is already happening in chip design, software development, and other fields.
The founders said AI is finding unusual chip layouts that can beat traditional designs. They pointed to AlphaChip, an AI system they built before starting Ricursive.
AlphaChip produced a circular, “donut-shaped” arrangement of chip parts. The design surprised engineers at first but ended up working well.
Ricursive is also taking a different approach from standard practice. Instead of splitting a chip into dozens of smaller blocks, the company optimizes the entire chip at once.
The founders said this lets their AI create complex patterns, including wave-like designs. They plan to share more details about this work in the future.
Goldie said faster design could open the door to many new kinds of chips. Today, designing a new chip can take six months to two years, so companies must be very selective.
“But if you can design a chip in like a week,” she said, companies could build custom chips for uses such as space, AR and VR, and biological computing.
She said this approach could make financial sense while also helping chips use less energy.
