Chips are a critical component of the AI industry. But new chips don’t hit the market with the same speed as new AI models and products.
Cognichip has a lofty goal of creating a foundational AI model that can help bring new chips to market faster.
San Francisco-based Cognichip is working to build a physics-informed foundational AI model that can be used by semiconductor companies to speed up the development process of new chips. The company is calling this approach “artificial chip intelligence” and hopes it can help speed up chip production times by 50% and reduce the associated costs as well.
This ambitious idea comes from semiconductor industry veteran Faraj Aalaei, who worked at various companies including Fujitsu Network Communications and Centillium Communications.
Aalaei told TechCrunch that his company’s origin story begins back in 2015. At the time, Aalaei was a member of the Silicon Valley Leadership Group, which would often meet to talk about what was ailing their industry.
He was growing concerned about what was happening in the semiconductor industry. He gave a presentation to the group about the stark drop in venture capital investing into semiconductor companies, which peaked at 200 deals a year in 2000, he said, and dropped to just one or two a year by 2015.
“I essentially warned the other CEOs that this cannot be good for us,” Aalaei said. “It cannot be good for the semiconductor industry in America, and that what we needed to do is to fundamentally change things about it. If that trend continues, then we’re going to lose our competitiveness. We’re going to lose the energy that brings in new ideas to the table.”
Considering how long it takes new chips to come to market, it’s not surprising these companies weren’t attracting venture investors, he said. Then he sat on the idea for almost a decade.
He went on to found Candou Ventures in 2016, and through that fund got to watch the rise in AI startups. When he realized that advancements in generative AI had gotten to a point where it could be used to potentially help solve some of the semiconductor industry’s existent challenges, he decided to launch Cognichip in 2024.
Cognichip has been operating in stealth ever since and has amassed a team of AI experts from places like Stanford, Google, and MIT to start building. Aalaei said it will take at least a few years to build the model to “ultimate performance” but said it should be able to help companies before it reaches that goal.
“When we get to that point, this artificial chip intelligence, we will be building a system that can actually act like an expert engineer,” Aalaei said. “Once we achieve that vision, then you can actually get the same work done with a fraction of the people and in much, much shorter time.”
Cognichip is now emerging from stealth with $33 million in seed funding in a round co-led by Lux Capital and Mayfield with participation from FPV and Candou Ventures.
Navin Chaddha, a managing partner at Mayfield, told TechCrunch that when he was introduced to Aalaei, he felt they were “cut from the same cloth.” The vast majority of work in the semiconductor industry is still being done by humans; he said he thinks the timing is right to bring AI into the mix.
“This is a major pain point, and the solution this company will provide will be a pain killer and not a vitamin for the semiconductor industry,” Chaddha said. “If you don’t have humans doing the job, can AI do it where there’s shortage of talent? Number one, great team, second, [they are] solving a real pain point in a massive, trillion-dollar industry.”
Aalaei said that he hopes Cognichip will also be able to help democratize access to building chips so that more semiconductor companies can get started and land investment. Easier access also means that smaller companies can build more specific chips for specialized or smaller models, too, he said.
All of this will depend on when, or if, the company can reach artificial chip intelligence.
“What we’re doing is not some incremental change,” Aalaei said. “We’re not building an [electronic design automation] tool, we’re not trying to tweak the process a little bit. We’re trying to set a new goal for our industry and bring some major change.”
