NVIDIA Details Physics AI, Agents, and Fab Digital Twins in Chip Manufacturing

NVIDIA outlined the deployment of accelerated computing, physics-informed AI models, engineering agents, and factory digital twins in semiconductor manufacturing. Multiphysics software provider COMSOL integrates the NVIDIA cuDSS library to accelerate semiconductor simulations by up to 10x. Applied Materials uses cuDSS to run its Ginestra materials simulation platform up to 10x faster than CPU-only systems, while using NVIDIA PhysicsNeMo to accelerate ACE+ topography simulations up to 35x. Siemens is integrating NVIDIA cuLitho into its Calibre lithography software, reporting up to 17x acceleration on key computational lithography workloads. For etch processes, Lam Research demonstrated using PhysicsNeMo to predict complex plasma behavior in milliseconds rather than running multi-day full simulations. On the test floor, Advantest applies NVIDIA NeMo Switchyard AI agents to automate model selection sequences, reducing operational response times from hours to minutes. In addition, Applied Materials utilizes NVIDIA Omniverse libraries to build physically accurate digital twins of complete fabs, enabling teams to evaluate layouts, material handling, bottlenecks, and operating strategies before physical deployment.

Key takeaways

• COMSOL, Applied Materials, and Siemens integrate NVIDIA libraries including cuDSS and cuLitho to speed up multiphysics and computational lithography simulations by up to 10x to 35x.
• Lam Research demonstrated that the PhysicsNeMo framework can predict etch plasma behavior in milliseconds compared with days required for full simulations.
• Advantest implements NVIDIA NeMo Switchyard AI agents to automate engineering model selection, lowering test-floor response times from hours to minutes.
• Applied Materials builds fab-level digital twins with NVIDIA Omniverse to model facility layouts, throughput, material flows, and operational bottlenecks virtually.