NVIDIA announced that semiconductor companies are adopting NVIDIA CUDA-X libraries, including the new NVIDIA cuPotential library, to support AI-defined fabs. Developers can use cuPotential to build AI agents that simulate interactions between atoms and molecules.
AI-defined fabs use AI to help engineers optimize processes, investigate production issues and improve manufacturing efficiency. As demand for advanced chips increases, manufacturers need to expand production while managing more complex processes. Engineers can use digital twins to model materials, analyze inspection data and evaluate process changes before applying them in production.
The expanded CUDA-X portfolio provides accelerated tools for semiconductor simulation, 3D data processing and image analysis. These tools support engineering applications and AI agents used in chip design and manufacturing to model process behavior and analyze yield-related data.
NVIDIA cuPotential supports atomic-scale simulations
Modeling interactions between atoms and molecules is relevant to semiconductor materials research and manufacturing processes. These calculations can require substantial processing time, including when run on GPU-optimized software.
NVIDIA cuPotential provides GPU-accelerated functions that developers can add to existing software. The library supports simulations of electrical forces and other interactions across large groups of atoms. Synopsys is among the companies using cuPotential for materials simulations.
The library can also be used by AI agents to run, evaluate and update molecular simulations. In early-access testing, cuPotential provided an additional 2–3x acceleration for workflows that were already optimized for GPUs.
Semiconductor companies use CUDA-X libraries in AI-defined fabs
AI-defined fabs require tools for modeling chip designs, materials behavior and manufacturing processes before changes are introduced into production. NVIDIA CUDA-X libraries include cuPotential for atomic-interaction simulations, cuEST for electronic-behavior modeling and cuDSS for engineering simulations.
Synopsys is using cuPotential, cuEST and cuDSS for semiconductor simulations. Early QuantumATK results showed a potential 30x speedup with cuEST, while S-Device benchmarks showed a 12x speedup with cuDSS. Synopsys’ AgentEngineer solutions interoperate with NVIDIA Nemotron open models and NVIDIA NeMo to support AI agents for chip design and verification.
Cadence is using NVIDIA accelerated computing, NVIDIA Nemotron open models and CUDA-X libraries for AI-supported design and simulation of electromechanical systems, including semiconductors, advanced packaging and printed circuit boards. Cadence reported that AuraStack AI Super Agent optimized advanced packages for heat and stress 8x faster using cuDSS with Cadence Celsius. Nemotron is also used with the ChipStack AI Super Agent for semiconductor design and verification.
COMSOL is using NVIDIA cuDSS to accelerate semiconductor multiphysics simulations. The company reports that the technology can provide simulations up to 10x faster. COMSOL’s planned MCP Server will use the NVIDIA Nemotron Ultra open model for AI-assisted modeling and design exploration.
Voltai uses NVIDIA cuDSS and CUDA-X libraries in Voltaire, its agent-based multiphysics platform for 3D-IC design and simulation. The platform combines AI agents with GPU-accelerated solvers to reduce the time required for electro-thermal-mechanical simulation workflows from months to a day.
ChipAgents uses NVIDIA AI infrastructure and model-training and optimization software to develop AI models for chip design. The company is also working with NVIDIA on Build-A-Chip Bench, an AI benchmark for evaluating agents across workflows from specification through graphic data system output. The benchmark is intended to measure progress in automated semiconductor engineering.
NVIDIA libraries support manufacturing workflows
Semiconductor fabs use process models to evaluate and validate manufacturing steps before production. NVIDIA cuLitho accelerates calculations for printing chip patterns, NanoVDB is an open-source library for processing 3D data and NVIDIA PhysicsNeMo is an open framework for training physics-based AI models.
Siemens is using cuLitho to accelerate Calibre software for chip-pattern refinement and mask preparation. Its Solido team is also developing agent-based characterization workflows using Nemotron open models and NVIDIA NeMo open libraries.
Lam Research plans to integrate NanoVDB into SEMulator3D, its semiconductor process-modeling platform within the Semiverse Solutions product family. The company reports that NanoVDB can accelerate selected wafer-etch modeling tasks by up to 30x. Lam Research has also demonstrated the use of NVIDIA PhysicsNeMo to reduce plasma-prediction simulation times to milliseconds instead of days.
Luminary is using NVIDIA PhysicsNeMo for semiconductor package design. The company trains physics-based AI models to predict temperature, pressure drop and thermal stress across package configurations. Luminary reports that engineers can evaluate 100x more designs within the same schedule without running a separate simulation for every option.
CUDA-X libraries support chip inspection and testing
AI-defined fabs need to process inspection images and test data to identify defects, maintain equipment and evaluate yield. NVIDIA NPP provides GPU-accelerated image-processing functions, while NVIDIA RAPIDS cuDF supports GPU processing of large datasets.
Imec is using NVIDIA AI and accelerated computing to remove noise from scanning electron microscope images used for sub-nanometer chip measurements. The organization is also using NVIDIA cuLitho for computational process-window optical proximity correction and reports a 30% improvement in the lithography process window.
KLA applied production-kernel techniques to an NPP NCC kernel and reported performance up to 50x faster than earlier CPU implementations. The results were comparable with KLA’s manually optimized implementations.
Teradyne is using NVIDIA NPP and accelerated computing for image-sensor testing. It is also using RAPIDS cuDF and CuPy for GPU calculations. Teradyne reports approximately 85x faster probe-maintenance inference, which can help identify degrading probes before they affect yield.
Advantest has integrated NVIDIA Nemotron and NVIDIA Kumo-Anomaly into ACS RTDI for semiconductor production analytics. Using NVIDIA Grace Blackwell systems, ACS provides real-time diagnostics and recommendations for test-cell corrective actions. Advantest is also using NVIDIA NeMo Switchyard AI agents to automate model selection and support semiconductor test workflows, reducing response times from hours to minutes.
NVIDIA technologies support fab operations and automation
In an AI-defined fab, equipment data must be transferred from instruments to AI systems quickly enough to support operational decisions.
Applied Materials is using NVIDIA Omniverse and NVIDIA CUDA-X libraries, including cuDSS, cuEST and PhysicsNeMo, for materials discovery, process development and fab optimization. The company reports chamber simulation speeds up to 35x faster and uses fab digital twins to evaluate layouts and material flow.
Analog Devices is combining Synopsys physics simulation with the NVIDIA Isaac Sim framework and Omniverse libraries to develop sensing simulations and robotic benchmarks. These tools help engineers evaluate sensor behavior within complete robotic systems.
FANUC and Foxconn are connecting robot models to the NVIDIA Omniverse Blueprint. This allows manufacturers to simulate and validate robotic operations in factory digital twins before physical deployment.
OMRON is using NVIDIA Cosmos open world models and the NVIDIA Metropolis Blueprint for video search and summarization in inspection equipment. The company is developing agents that combine visual data, equipment documentation and manufacturing information to help engineers analyze printed circuit board results, investigate possible defect causes and support troubleshooting and quality control.
Availability
NVIDIA cuPotential is available through an early-access program. NanoVDB is available as open-source software within OpenVDB, and NPP is available through the NVIDIA CUDA Toolkit. Lam Research’s NanoVDB-accelerated SEMulator3D capability is scheduled for release by the end of this year.
For more information, visit nvidia.com.
The post Semiconductor companies adopt CUDA-X libraries for AI-defined fabs appeared first on Engineering.com.