NVIDIA says AI is reshaping chip fabs, cuDSS accelerates simulation up to 35x
NVIDIA described an AI-defined factory, with COMSOL simulation up to 10x faster via cuDSS and Applied Materials ACE+ morphology simulation up to 35x faster.
NVIDIA published an article on November 3 arguing that AI is pushing semiconductor fabs into an AI-defined factory stage, accelerating design, test and production decisions through accelerated simulation, learned physics models, AI agents and digital twins.
Accelerated simulation compresses engineering simulation that previously ran on CPUs from hours or days down to minutes or even milliseconds, letting engineers compare process options and predict defect risk before committing to expensive physical experiments. COMSOL sped up semiconductor manufacturing simulation by up to 10x with cuDSS, while Siemens integrated cuLitho into Calibre for up to 17x faster key simulation loads. Applied Materials reported Ginestra up to 10x faster with cuDSS and ACE+ multiphysics morphology simulation up to 35x faster with PhysicsNeMo.
Plasma behavior modeling in etching is complex and costly, and Lam Research showed PhysicsNeMo predicting plasma behavior in milliseconds where full simulation can take days. NVIDIA also cited Advantest using NeMo Switchyard agents to pick the right AI model automatically, compressing tasks that took hours into minutes.
NVIDIA stressed that AI narrows the option space and surfaces high-risk candidates, while critical process changes still require high-precision simulation, experimental validation and engineer sign-off.