NVIDIA launches cuPotential CUDA-X library for AI-defined fabs
At SEMICON West, NVIDIA announced cuPotential, a new CUDA-X library developers can use to build AI agents that simulate how atoms and molecules interact; NVIDIA says early-access testing showed an additional 2–3x speedup on already GPU-optimized materials-simulation workflows, and the library is available in early access. Synopsys is the first adopter, citing a potential 30x speedup with cuEST on QuantumATK and a 12x speedup with cuDSS on S-Device benchmarks. Applied Materials, Cadence, KLA, Lam Research, Siemens and Synopsys are using NVIDIA's accelerated computing, Nemotron open models and CUDA-X libraries across chip design, manufacturing and inspection — for instance, Lam Research intends to integrate the open-source NanoVDB library into SEMulator3D for up to 30x faster wafer-etch modeling, and KLA reports its new NPP NCC kernel runs 50x faster than earlier CPU versions.
If these accelerated libraries and agent tools deliver as promoted, iteration speed across chip design and manufacturing could rise markedly — though most performance figures come from vendors' own early testing.