NVIDIA announced on Oct. 14 that semiconductor companies are adopting its CUDA-X libraries, including a new cuPotential library, to develop what the company calls "AI-defined fabs."
The company says these tools allow engineers to use AI agents to optimize manufacturing processes, resolve production issues and improve efficiency by using digital twins to test changes before implementing them in physical production.
The new cuPotential library is intended for developers building AI agents that simulate atomic and molecular interactions. According to NVIDIA, the library accelerated already GPU-optimized workflows by an additional 2-3x in early-access testing. Synopsys is among the early adopters using the tool for materials simulations.
Other CUDA-X libraries mentioned in the announcement include cuEST for modeling electronic behavior and cuDSS for engineering simulations. Synopsys reported potential speedups of 30x using cuEST and 12x using cuDSS in early QuantumATK and S-Device benchmarks, respectively NVIDIA Newsroom.
Lam Research intends to integrate the NanoVDB 3D data library into its SEMulator3D platform. NVIDIA claims this should accelerate select wafer-etch modeling by up to 30x. This capability is targeted for release by the end of the year NVIDIA Newsroom. KLA used a new NPP NCC kernel to achieve performance 50x faster than previous CPU versions for chipmaking inspection NVIDIA Newsroom. Teradyne is using NPP and RAPIDS cuDF to achieve approximately 85x faster probe-maintenance inference NVIDIA Newsroom. Applied Materials is utilizing NVIDIA Omniverse and CUDA-X libraries to achieve up to 35x faster chamber simulations NVIDIA Newsroom.
cuPotential is currently available in early access. NanoVDB is available as open source within OpenVDB, and NPP is available through the NVIDIA CUDA Toolkit.