AI speeds Nvidia engineering

Nvidia says it now uses artificial intelligence to finish one recurring graphics processor design job overnight instead of assigning it to eight engineers for 10 months. (videocardz.com) The task is not drawing an entire chip from scratch. It is porting a standard cell library — a catalog of 2,500 to 3,000 tiny prebuilt circuit blocks — to each new semiconductor manufacturing process. (tech.yahoo.com) Bill Dally, Nvidia’s chief scientist, said the company built a reinforcement learning system called NVCell for that work, and said it runs on one graphics processor and can match or beat human layouts on cell size, power dissipation, and delay. (tech.yahoo.com) Standard cells are the Lego bricks of a chip: logic gates, memory elements, and other basic parts that engineers reuse thousands of times. When a chipmaker moves to a new factory process, those bricks have to be rebuilt so the final design still meets speed, power, and area targets. (cacm.acm.org) That step has become more painful as modern manufacturing nodes add more design rules and tighter constraints. Nvidia and other chip companies have been pushing machine learning into placement, verification, and layout because those stages involve huge search spaces and repeated trade-offs. (cacm.acm.org) Nvidia has been laying groundwork for this inside its engineering teams for years. In 2023, Dally presented ChipNeMo, an internal large language model trained on company design data to answer chip-design questions, summarize bug reports, and write scripts for electronic design automation tools. (blogs.nvidia.com) Dally said at Nvidia’s March 2026 Graphics Technology Conference session with Google’s Jeff Dean that the company is also using artificial intelligence for design exploration, bug handling, and verification, while saying fully autonomous chip design is still far off. (nvidia.com) The claim fits a broader shift in the chip industry, where companies are trying to shorten design cycles as advanced processors grow more expensive and more complex to build. Communications of the Association for Computing Machinery reported in 2024 that Nvidia, Intel, Advanced Micro Devices, International Business Machines, Google, and Apple were all applying artificial intelligence to parts of chip development. (cacm.acm.org) What Nvidia is describing is narrower than “artificial intelligence designs chips.” It is a case of software taking over one labor-heavy step in a long engineering flow, with human designers still deciding architecture, checking results, and signing off on the final silicon. (videocardz.com) For Nvidia, the immediate payoff is time: if a new process library can be rebuilt in one night instead of 80 person-months, engineers can spend more of a project on the parts of chip design the company still says machines cannot do alone. (tech.yahoo.com)