AI Transforms Semiconductor Manufacturing

The demand for advanced chips is rising and production processes are becoming increasingly complex. For engineers, it is crucial to understand how changes in processes behave, how materials interact with new steps, and how defects can be traced back to a cause. These analyses can take days, and physical experiments on a factory line are both slow and costly. An incorrect assessment of yield or process quality has immediate consequences. AI is changing the way semiconductor fabrication (fab) operates and enables engineers to work faster.

Accelerated computing power shortens the simulation cycle, allowing engineers to evaluate more alternatives before proceeding to physical tests. The use of multiphysics simulations, which model the interactions between different physical effects, can be up to 10 times faster using the NVIDIA cuDSS library, leading to optimized equipment and processes. Additionally, Applied Materials and NVIDIA are collaborating to discover materials at the atomic level faster and accelerate process development. The integration of NVIDIA physics-based technologies, such as PhysicsNeMo, enables topographical simulations to be performed up to 35 times faster than traditional methods.

Moreover, AI enables engineers to predict process behavior without having to perform an extensive simulation each time. Using the NVIDIA PhysicsNeMo framework, plasma behavior, a key element in etch processes within semiconductor production, can be predicted in milliseconds instead of the days that were previously required. Digital twins, based on NVIDIA Omniverse libraries, assist teams in simulating optimizations and business strategies before proceeding to physical experiments, significantly reducing costs.

The future of semiconductor manufacturing lies in connecting these capabilities: using production data to inform models, having these models evaluate changes, and feeding the findings back to engineers. This embodies the promise of the AI-defined fab: an environment where AI helps people learn from each production cycle and act on it sooner.

Read the full article from Nvidia.