Something big is taking shape on a quiet residential street near Stanford University in Palo Alto, California. Two former Google researchers, Anna Goldie and Azalia Mirhoseini, are founding a startup called Ricursive Intelligence. Their goal is clear: to create software that automates the design of modern chips. This could mean that any company will be able to design its own chips from scratch, faster and more cheaply than ever before.
Anna Goldie and Azalia Mirhoseini, who is also a professor of computer science at Stanford, have raised CZK 728 million (USD 35 million) to launch the company. The money came from investors such as Sequoia Capital and Striker Venture Partners. As a result, Ricursive Intelligence is now valued at CZK 15.6 billion (USD 750 million). They plan to launch their first product next year.
Challenging chip design
Today, creating a custom chip is a demanding undertaking. Companies such as Amazon and Google develop custom chips for artificial intelligence and data centers. Apple, meanwhile, saved billions of dollars by designing its M-series chips for its MacBooks. These chips are cheaper, more energy-efficient, and smaller because they are tailored to specific tasks.
But the process is expensive and manual. It includes everything from architecture design and testing to preparation for mass production. The entire process usually takes two to three years. If a design flaw is discovered only at the end, it causes major delays and costs that are difficult to recover from.
Goldie and Mirhoseini believe their software will automate all of this. Instead of years, chip design could take only weeks or even days. This would enable custom specialized chips to flourish at many companies. "We think there will be a boom in custom chips," Anna Goldie said.
Ricursive currently employs five researchers and engineers based in an ordinary house in Palo Alto. They sit in front of large screens, dissecting scientific papers and going through lines of code. Two of them worked with Goldie and Mirhoseini at Google on AlphaChip software, which designed chip layouts for Google, including TPU chips for artificial intelligence.
Startups founded by former leaders of major AI labs are attracting enormous interest. Examples include Periodic Labs, founded by OpenAI's Liam Fedus, and Reflection, founded by former Google DeepMind researchers. More than 50 investors approached Ricursive Intelligence. The goal is to open up the field to people who can create chips tailored precisely to their needs."
Source: wsj.com
