Advanced silicon chips have supercharged artificial intelligence. Now, AI is returning the favor. Cognichip is building a deep learning model to assist engineers in designing new computer chips. The goal is to tackle the industry’s longstanding problems: chip design is incredibly complex, expensive, and slow. Advanced chips can take three to five years from concept to mass production. For instance, Nvidia’s latest Blackwell GPU has 104 billion transistors. Cognichip’s CEO, Faraj Aalaei, warns that markets can shift during this long timeline, wasting huge investments. He aims to bring AI tools, similar to those used by software engineers, into semiconductor design. The company claims its technology can cut development costs by over 75% and reduce the timeline by more than half. Cognichip recently raised $60 million, led by Seligman Ventures, with Intel’s former CEO Lip-Bu Tan joining its board. Total funding now stands at $93 million. However, Cognichip has not yet revealed a chip designed with its system or named its collaborating customers. The company’s advantage comes from training its own model on specialized chip design data, not a general-purpose AI. This required creating proprietary and synthetic data sets, as chip design IP is closely guarded. The firm also uses secure methods for clients to train models on their private data. For public projects, it has utilized open-source architectures like RISC-V. Cognichip faces competition from established firms like Synopsys and Cadence, as well as startups like Alpha Design AI and ChipAgentsAI. Investors believe the massive funding flowing into AI infrastructure creates a major opportunity for such design tools.
