Summary
NVIDIA says semiconductor companies are adopting its CUDA-X libraries to accelerate simulation, inspection and other chip-manufacturing work. Its new cuPotential library is in early access, with NVIDIA reporting 2–3x acceleration in early tests of already GPU-optimized workflows.
NVIDIA says semiconductor companies are adopting its CUDA-X libraries to apply AI and accelerated computing across chip design, manufacturing and testing. The announcement includes cuPotential, a new library for building AI agents that simulate interactions between atoms and molecules. NVIDIA reports that early-access tests accelerated already GPU-optimized workflows by an additional 2–3x.
The tools are intended to help engineers model materials and manufacturing processes, analyse inspection data and evaluate changes in digital twins—virtual models used to explore equipment or processes before making changes on a factory floor. NVIDIA describes an “AI-defined fab” as a semiconductor factory where AI helps engineers optimise processes, address production issues and improve efficiency.
AI tools for simulation and inspection
cuPotential adds specialised GPU-accelerated functions to existing software for modelling electrical forces and other interactions among large numbers of atoms. That kind of modelling can help researchers study semiconductor materials and manufacturing processes. NVIDIA says making the capabilities available to AI agents allows them to run simulations, assess results and iterate on them.
The library sits alongside NVIDIA tools for other engineering tasks: cuEST for modelling electronic behaviour and cuDSS for solving engineering simulations. Synopsys is using all three for semiconductor simulations. NVIDIA reports potential speedups of 30x in early QuantumATK results using cuEST and 12x in S-Device benchmarks using cuDSS.
Other examples cover different stages of production. Lam Research intends to integrate the open-source 3D-data library NanoVDB into its SEMulator3D process-modelling platform; NVIDIA says the addition is expected to speed up selected wafer-etch modelling by up to 30x. KLA has applied its production-kernel insights to a new NVIDIA NPP image-processing kernel, which the announcement says performed 50x faster than earlier CPU versions, at a level comparable to KLA’s hand-tuned implementations.
These figures describe separate tools and workloads, not one measure of overall factory output. Their relevance is that faster simulation and image analysis could give engineers more opportunity to test process changes and inspect production data within existing workflows.
Deployment across fab operations
The announcement also describes uses beyond chip simulation. Applied Materials is using NVIDIA computing and simulation tools across materials discovery, process development and fab optimisation; NVIDIA reports up to 35x faster chamber simulations. At imec, NVIDIA says AI is being used to remove noise from scanning electron microscope images, while cuLitho supports computationally intensive optical proximity correction. The organisation reported a 30% improvement in the lithography process window.
For factory automation, FANUC and Foxconn are connecting robot models to an Omniverse blueprint so manufacturers can simulate and validate robotic operations in factory digital twins before physical deployment. NVIDIA also describes OMRON using AI agents to combine inspection video, equipment documentation and manufacturing knowledge to support defect analysis and troubleshooting.
cuPotential is available in early access. NanoVDB is available as open source within OpenVDB, and NPP is available through the NVIDIA CUDA Toolkit. Lam Research’s NanoVDB-accelerated SEMulator3D capability is targeted for release by the end of 2026.