Demand for advanced chips is rising while the margin for error in their production shrinks. Semiconductor manufacturing processes are becoming increasingly complex, making it difficult for engineers to predict how materials will interact with new steps or identify the root cause of defects. These analyses traditionally take days, and physical experiments on a fab line are both slow and expensive. A single wrong call on yield or process quality can have significant financial consequences.
NVIDIA is addressing these challenges by introducing an AI-defined fab model. This approach integrates accelerated computing, physics-based AI, AI agents, and digital twins to help engineers optimize processes and make better decisions across the full manufacturing lifecycle. The goal is to reduce the time engineers spend waiting for answers, allowing them to direct work based on their expertise rather than computational bottlenecks.
Accelerated Computing and Physics-Based AI
Accelerated computing allows engineers to evaluate more alternatives before committing to physical trials. Multiphysics simulations, which model interactions among heat transfer, electrical behavior, and mechanical stress, are critical for this stage. NVIDIA’s cuDSS library is helping software provider COMSOL accelerate these simulations for semiconductor manufacturing by up to 10x. This speed enables engineers to compare alternatives and decide which changes to test physically.
Applied Materials is collaborating with NVIDIA to accelerate atomic-scale materials discovery. By integrating NVIDIA cuDSS, Applied’s Ginestra simulation platform runs up to 10x faster than CPU-only approaches. Additionally, Applied’s ACE+ multiphysics platform, powered by NVIDIA PhysicsNeMo, delivers topography simulations up to 35x faster. These gains are not just about speed; they enable calculations that were previously too intensive to complete at production scale.
Siemens is integrating NVIDIA cuLitho into its Calibre lithography software, delivering up to 17x acceleration on key simulation workloads. Computational lithography defines how circuit patterns are printed onto silicon. The acceleration expands what engineers can verify before a wafer goes to print, allowing researchers to explore material behavior and reaction pathways that influence next-generation processes.
Physics-based AI further reduces the need for full simulations. Modeling plasma, the electrically charged gas used in semiconductor etch processes, requires solving complex equations that capture electrical, chemical, and thermal interactions. Lam Research demonstrated that the NVIDIA PhysicsNeMo framework can predict plasma behavior in milliseconds, compared with days for a full simulation. This gives engineers a faster way to explore process conditions without rerunning the full model each time.
AI Agents and Digital Twins
Engineers must turn analyses into decisions, such as revising a design or adjusting a production process. AI agents help carry out the sequential steps between these decisions. Evaluating which AI model to apply to a given task is a manual process that can consume hours on the test floor. Advantest is using NVIDIA NeMo Switchyard AI agents to automate that selection, reducing response times from hours to minutes.
Digital twins allow teams to simulate equipment configurations, layouts, and process changes in a virtual environment before committing to physical trials. This reduces experimentation costs, especially as fabs become more complex. Applied Materials uses NVIDIA Omniverse libraries to build physically accurate digital twins of entire fabs. Teams can simulate material movement, identify bottlenecks, and evaluate layouts and operating strategies. This extends virtual experimentation beyond individual processes to how the factory works as a whole.
The future of semiconductor manufacturing lies in connecting these capabilities. Production data informs models, and models evaluate changes, bringing findings back to engineers making production decisions. This cycle allows fabs to learn from each production cycle and act on that knowledge sooner, creating a more responsive and efficient manufacturing environment.
Source: NVIDIA
