NVIDIA expands simulation tools for chip design and manufacturing

NVIDIA introduced cuPotential, a CUDA-X library that helps developers build AI agents to simulate interactions between atoms and molecules. Available in early access, the library accelerated already GPU-optimized workflows by an additional 2-3x in early testing. Synopsys is among its early adopters, using it to accelerate materials simulations.

The release supports NVIDIA’s push toward semiconductor factories where AI helps engineers optimize processes, investigate production problems and improve yield. Synopsys, Cadence, Siemens and other companies are adopting its accelerated computing tools across design and manufacturing. Synopsys reported early QuantumATK results showing a potential 30x speedup using cuEST, while Cadence is using cuDSS to optimize advanced packages for heat and stress.

Manufacturing applications include Lam Research’s planned integration of the open-source NanoVDB library into SEMulator3D, which is expected to accelerate select wafer-etch modeling by up to 30x. That capability is targeted for release by the end of this year. Applied Materials is using NVIDIA tools for materials discovery, process development and factory digital twins, enabling up to 35x faster chamber simulations.

Inspection and testing are also priorities. KLA contributed insights to an NPP image-processing kernel that achieved performance 50x faster than earlier CPU versions. Teradyne is applying GPU computing to image-sensor testing and probe maintenance, while Advantest is using AI models and agents for production diagnostics and corrective recommendations.