Cognichip Inc., a startup working to make chip design teams more productive, launched today with $33 million in initial funding.

Lux Capital and Mayfield jointly led the seed round. They were joined by FPV and Candou Ventures.

Cognichip is led by co-founder and Chief Executive Faraj Aalaei. He was previously the CEO of Aquantia Corp., a network chip supplier that was acquired by Marvell Technology Inc. in 2019 for $452 million. Cognichip’s other co-founders earlier worked at major tech firms such as Apple Inc. and Google LLC.

Cognichip is building a foundation artificial intelligence model called Artificial Chip Intelligence, or ACI. It promises to reduce the amount of manual work involved in developing processors. Cognichip reportedly believes that the model will take several years to reach “ultimate performance,” but expects it to start benefiting chip designers earlier.

According to the company, ACI can both speed up chip development and reduce the associated costs. Cognichip sees its model cutting the price of processor design projects by up to 75%. Additionally, the company says that ACI can identify opportunities to fine-tune a chip’s performance and efficiency.

The semiconductor industry has already started using AI to automate some engineering tasks. One of those tasks is the process of fine-tuning chip blueprints to more closely align them with user requirements.

After engineers craft a processor design, it’s theoretically possible to create millions of new variations with slightly different parameters. Some versions might be faster than the original, while others may require less power or space. An AI model can automatically try every single variation to find the one that best meets project requirements.

Companies are also using AI to automate other aspects of the chip design workflow. 

In 2020, Google debuted AlphaChip, an internally developed neural network that helps optimize its TPU chips’ floor plans. A processor’s floor plan determines the location of the circuits it includes. Last year, Google detailed that AlphaChip has helped its engineers improve four generations of TPU chips. 

Other companies, meanwhile, are using AI to scan chips for bugs before they enter mass production. Engineers don’t test processor designs manually but rather write scripts to perform the task. AI models can be used to enhance those scripts and, in some cases, generate new ones from scratch.

“As a founder who has successfully taken two semiconductor companies public, and as a venture capitalist, I have seen firsthand how semiconductor startups are often deemed ‘unfundable’ due to high costs and long timelines,” Aalaei said. “With the rise of generative AI, we have a once-in-a-generation opportunity to rethink how chips are designed.”

Photo: Unsplash