NVIDIA Adds cuPotential to CUDA-X for AI-Defined Fab Push
NVIDIA introduced cuPotential, a CUDA-X library for AI agents that simulate atomic interactions, as partners including Lam, KLA and Synopsys adopt its tools.
What you need to know
NVIDIA announced cuPotential, an early-access CUDA-X library that lets developers build AI agents to simulate atomic and molecular interactions, which NVIDIA says accelerated already GPU-optimized workflows by an additional 2-3x in early testing. Semiconductor companies including Synopsys, Lam Research, KLA and Applied Materials are using CUDA-X libraries across design and manufacturing.
- NVIDIA announced cuPotential in early access as part of its CUDA-X library portfolio.
- NVIDIA said early-access testing added 2-3x speedup to already GPU-optimized workflows.
- Synopsys reported a potential 30x speedup using cuEST and 12x with cuDSS in early results.
Atomic simulation library
NVIDIA said cuPotential provides GPU-accelerated functions that developers can add to existing software to model electrical forces and other interactions across large numbers of atoms. The company positions the library as a way for AI agents to run, evaluate and iterate on molecular simulations, which it said can be time-consuming even in GPU-optimized tools.
Synopsys is among early adopters, according to NVIDIA. NVIDIA also said cuPotential is available in early access.
Partner use across design and manufacturing
NVIDIA said Synopsys is using cuPotential, cuEST and cuDSS for semiconductor simulations, with early QuantumATK results showing a potential 30x speedup using cuEST and S-Device benchmarks showing 12x using cuDSS. Cadence is applying CUDA-X libraries and NVIDIA models to agentic AI for electromechanical systems, with one example cited at 8x faster package optimization using cuDSS.
In manufacturing, Siemens is using cuLitho for chip-pattern refinement, and Lam Research intends to integrate open-source NanoVDB into its SEMulator3D platform, with NVIDIA expecting up to 30x acceleration in select wafer-etch modeling and targeting release by the end of this year. KLA cited an NPP kernel running 50x faster than earlier CPU versions. Teradyne reported roughly 85x faster probe-maintenance inference, and Applied Materials cited up to 35x faster chamber simulations.
What is NVIDIA's new cuPotential library?
It is an early-access CUDA-X library with GPU-accelerated functions that developers can use to build AI agents simulating how atoms and molecules interact. NVIDIA said it added 2-3x speedup in early-access testing.
Sources & context
- NVIDIA and Semiconductor Leaders Advance AI Across Chip Design and Manufacturing for AI-Defined Fabs ↗nvidianews.nvidia.com
Something needs correcting? Send a correction.