NVIDIA announced on October 14 that several major semiconductor companies are using its software tools to bring artificial intelligence into chip design and manufacturing. The company also introduced a new library called cuPotential.
The announcement centers on what NVIDIA calls “AI-defined fabs.” These are chip factories where AI helps engineers improve processes, fix production problems and raise efficiency.
Companies named in the announcement include Applied Materials, Cadence, KLA, Lam Research, Siemens and Synopsys. NVIDIA said they are using its computing and AI tools across chip design, manufacturing and testing.
NVIDIA Launches cuPotential Library
cuPotential is a new part of NVIDIA’s CUDA-X software collection. Developers can use it to build AI agents that simulate how atoms and molecules interact.
These calculations help companies model chip materials and manufacturing steps. NVIDIA said that in early-access testing, the library made workflows already optimized for GPUs run an additional 2 to 3 times faster.
Synopsys is among the early users. Its early QuantumATK results showed a possible 30 times speedup using another library called cuEST, while S-Device benchmarks showed a 12 times speedup with cuDSS.
“The semiconductor industry made the AI revolution possible. Now, AI is transforming how chips are built,” said Tim Costa, vice president and general manager of industrial and computational engineering at NVIDIA.
Cadence said its AuraStack AI Super Agent optimizes advanced chip packages for heat and stress 8 times faster using cuDSS. COMSOL reported simulations up to 10 times faster with the same library.
Chipmakers Report Faster Simulation and Inspection
Lam Research plans to add NanoVDB, an open source library for 3D data, to its SEMulator3D modeling platform. The company expects this to speed up some wafer-etch modeling by up to 30 times.
Lam also said NVIDIA PhysicsNeMo has helped cut plasma-prediction simulation time from days to as little as milliseconds.
Siemens is using NVIDIA cuLitho to speed up its Calibre software. That software helps refine chip patterns and prepare masks.
KLA built a new image-processing kernel using NVIDIA NPP. It ran 50 times faster than earlier CPU versions and matched KLA’s own hand-tuned tools.
Teradyne said it achieved about 85 times faster probe-maintenance inference. This helps the company find worn testing probes before they hurt chip yield.
Research group imec reported a 30% improvement in the lithography process window using NVIDIA tools. It is also using AI to remove noise from microscope images used for tiny chip measurements.
Advantest has added NVIDIA Nemotron models to its ACS RTDI system, which runs on Grace Blackwell hardware. The company said AI agents have cut some response times from hours to minutes.
Robot makers FANUC and Foxconn are connecting their robot models to NVIDIA’s Mega Omniverse Blueprint. This lets manufacturers test robotic operations in digital factory models before using them on the floor.
Applied Materials is using NVIDIA Omniverse and CUDA-X libraries across materials discovery, process development and fab planning. The company reported chamber simulations running up to 35 times faster.
