NVIDIA Joins Forces with Multiple Semiconductor Companies to Advance AI Technology Applications in Fabs

en.Wedoany.com Reported - NVIDIA has announced that semiconductor companies including Applied Materials, KLA, Lam Research, Siemens, Cadence Design Systems, and Synopsys are adopting its CUDA-X accelerated computing software libraries and artificial intelligence technologies, with applications spanning chip design, materials simulation, wafer manufacturing, defect detection, and production testing. NVIDIA has simultaneously introduced cuPotential, an atomic-level simulation software library. Early testing shows that this tool can achieve a further 2x to 3x speedup on top of GPU-optimized computational workflows.

Synopsys has applied cuPotential, cuEST, and cuDSS to semiconductor simulation work. After adopting cuEST in its QuantumATK materials simulation software, preliminary testing shows computational speed can be increased by up to 30x; its S-Device simulation tool achieved a 12x speedup using cuDSS. Cadence Design Systems has introduced the related technologies into advanced packaging thermal stress analysis. After combining its AuraStack AI agent with the Celsius platform and cuDSS, packaging thermal and stress optimization computations achieved an 8x speedup.

Lam Research plans to integrate NanoVDB 3D data processing technology into the SEMulator3D semiconductor process modeling platform, with certain wafer etch simulation tasks expected to achieve computational speed improvements of up to 30x. The feature is scheduled for release by the end of 2026. Lam Research is also using the NVIDIA PhysicsNeMo framework for plasma process prediction, with related demonstrations reducing certain computational tasks from days to milliseconds.

KLA has adopted the NVIDIA NPP image processing software library to optimize semiconductor defect detection algorithms, with the relevant computational kernels achieving up to 50x speedup compared to earlier CPU implementations. Teradyne has applied NPP, RAPIDS cuDF, and CuPy to image sensor testing and probe maintenance analysis, achieving approximately 85x faster probe maintenance inference. Belgium-based research institute imec has used NVIDIA accelerated computing to process scanning electron microscope images and achieved a 30% improvement in computational lithography process window optimization.

Applied Materials has applied NVIDIA Omniverse, cuDSS, cuEST, and PhysicsNeMo to materials R&D, process development, and fab optimization, with certain process chamber simulations achieving up to 35x speedup, while also employing a fab digital twin system for facility layout and material flow simulation. Japan's Advantest has integrated NVIDIA AI models into its ACS real-time data infrastructure for semiconductor test equipment diagnostics, anomaly analysis, and corrective action recommendations.

NVIDIA's newly introduced cuPotential is currently in early access, NanoVDB has been open-sourced as part of OpenVDB, and NPP is available through the CUDA Toolkit. The partner companies are at various stages including production deployment, performance testing, software integration, and forthcoming releases.