Chip fabs are using AI to cut simulation time
Applied Materials, Lam Research and Siemens are among the firms using AI and faster computing to shorten chip-making simulations, according to NVIDIA.
The facts
- According to NVIDIA, software firm COMSOL is using the cuDSS library to speed up multiphysics simulations for chip manufacturing by up to 10x.
- NVIDIA says Applied Materials' Ginestra platform runs up to 10x faster than CPU-only methods, and its ACE+ platform runs topography simulations up to 35x faster.
- Siemens is integrating NVIDIA's cuLitho into its Calibre lithography software, with up to 17x acceleration on key simulation workloads, per NVIDIA.
- Lam Research showed AI predicting plasma behaviour in milliseconds instead of days, and Advantest says AI agents cut model selection from hours to minutes. These are vendor-reported figures.
Case notes · Danny's take
Why it matters
Chip demand is not the only bottleneck. The slow part is working out what to change on the line without wrecking yield. If simulations drop from days to minutes, engineers can test more ideas before touching real wafers. Note who is writing this: the speed-ups are NVIDIA's own numbers, and the chips it sells are the tools doing the work. Not independent proof. A claim to check, not a result to cheer.
For builders
The money in this story is in the software layer around fabs: simulation, digital twins and agents that automate engineering chores. If you build industrial tools, learn how physics-based models stand in for full simulations. Treat the quoted speed-ups as vendor claims until customers publish their own.
Not financial advice. Do your own investigation.
This story was summarised with AI from NVIDIA Blog's reporting and checked against our editorial policy. Danny's take reflects his views. Disclaimer.
Danny Walter
Founder of Moon Republic, Volt Terminal and Moon Republic Academy. Author of The Market Is a Crime Scene and After the Applause.