The most cutting-edge silicon chips have propelled the evolution of artificial intelligence. Can AI now reciprocate?

Cognichip is creating a deep learning model designed to aid engineers in the creation of new computer chips. The challenge it aims to address is one that the industry has grappled with for decades: Chip design is incredibly intricate, prohibitively costly, and time-consuming. High-end chips require three to five years to transition from idea to mass production; the design stage itself can take up to two years before physical layout commences. Consider that Nvidia’s latest GPU line, Blackwell, comprises 104 billion transistors — that’s a significant number to organize.

According to Cognichip CEO and founder Faraj Aalaei, the market can shift during the time taken to develop a new chip, rendering substantial investments futile. Aalaei aims to introduce the types of AI tools that software engineers have utilized to expedite their tasks into the semiconductor design realm. 

“These systems have reached a level of intelligence where, by merely directing them and specifying the desired outcome, they can generate excellent code,” Aalaei shared with TechCrunch.

He asserts that the company’s technology has the potential to lower chip development costs by over 75% and drastically shorten timelines by more than half. 

The company emerged from stealth mode last year and announced Wednesday that it secured $60 million in new funding led by Seligman Ventures, with significant contributions from Intel CEO Lip-Bu Tan, who invested through his venture firm Walden Catalyst Ventures and will join Cognichip’s board. Umesh Padval, a managing partner at Seligman, will also become a board member. In total, Cognichip has raised $93 million since its inception in 2024.

However, Cognichip has not yet revealed a new chip that has been designed using its system and did not disclose any customer collaborations it claims to have engaged in since September. 

The company claims its advantage lies in employing its own model trained on chip design data, rather than utilizing a general-purpose LLM. This required the acquisition of domain-specific training data, which is a significant challenge. Distinct from software developers, who openly share large volumes of code, chip designers closely protect their intellectual property, making the type of open-source dataset that usually trains AI coding assistants largely inaccessible.

Cognichip has had to create its own datasets, including synthetic data, and obtain licenses for data from partners. The company has also devised processes that enable chip manufacturers to securely train Cognichip’s models on their proprietary data without revealing it.

In instances where proprietary data is unavailable, Cognichip has relied on open-source alternatives. In one demonstration last year, Cognichip invited electrical engineering students from San Jose State University to experiment with the model during a hackathon. The teams managed to utilize the model to create CPUs based on the RISC-V open-source chip architecture — a design that is freely accessible for anyone to build upon.

Cognichip is up against established competitors like Synopsys and Cadence Design Systems, as well as well-funded newcomers like ChipAgents, which completed a $74 million extended Series A in February, and Ricursive, which secured a $300 million Series A round in January.

Padval noted that the current influx of capital into AI infrastructure is the largest he has observed in his 40 years of investing.

“If it’s a super cycle for semiconductors and hardware, it’s a super cycle for companies like [Cognichip],” he stated.