NVIDIA details how AI is transforming chip fabrication into an ai-defined factory

Nvidia recently detailed how artificial intelligence is ushering in an era of AI-defined factories within semiconductor wafer fabrication. By integrating accelerated simulation, learned physics models, AI agents, and digital twins, the technology is drastically speeding up design, testing, and production decision-making. Accelerated simulation leverages technologies like GPUs and AI agents to compress engineering simulations from days down to mere minutes. This allows engineers to test process variations and predict defect risks virtually before costly physical runs. For instance, Applied Materials' Ginestra platform achieved up to a 10x speedup using cuDSS, while its ACE+ platform reached a 35x acceleration in morphology simulation via PhysicsNeMo. Other notable advancements include COMSOL boosting simulation speeds by up to 10x with Nvidia's cuDSS, and Siemens achieving up to a 17x speed increase by integrating cuLitho into Calibre lithography software. Furthermore, Lam Research demonstrated PhysicsNeMo can predict plasma behavior in milliseconds, compared to days for full simulations. Nvidia also noted Advantest using NeMo Switchyard AI agents to reduce several-hour tasks to minutes, enabling faster iteration cycles for chip development.