Cognichip has launched ACI Enterprise, a chip design platform built on physics-informed models rather than the large language models most rivals wrap around existing EDA tools.
The company argues that LLMs trained mostly on text produce design results barely better than random. It trained its own models on data from a real chip design project staffed with experienced engineers, then filled the gaps with synthetic data built by a team of AI scientists.
Manoher Bommena, a vice president of engineering at Renesas, described the platform as one that speaks the language of chip design rather than of text. Cognichip claims early results point to roughly a 100x speedup.
Santa Clara hosts the AI Infra Summit this week, where the company is walking through how the platform works. Funding so far totals $93M. The backer list includes Mayfield Fund, Lux Capital, FPV Ventures, Seligman Ventures, SBI Investment and Candou Ventures.
The pitch lands as chipmakers hunt for design productivity to offset ballooning verification and layout costs at leading-edge nodes.
