NVIDIA expands Agent Toolkit for AI-driven chip engineering

NVIDIA has expanded its NVIDIA Agent Toolkit with new AI physics and accelerated computing capabilities, aiming to accelerate the adoption of autonomous AI engineers for chip and system design. The update adds re-architected NVIDIA PhysicsNeMo libraries alongside new and enhanced CUDA-X libraries, giving developers access to AI-powered physics models, accelerated simulation tools and quantum chemistry capabilities within engineering workflows. NVIDIA says its expanded Agent Toolkit enables developers to build AI agents capable of reasoning with physics models, launching engineering simulations and generating high-fidelity design data. The company has split its PhysicsNeMo framework into agent-friendly libraries that can be called directly by AI assistants. At the same time, updated CUDA-X libraries add GPU-accelerated sparse solvers and quantum chemistry tools for production engineering workloads. Engineering has reached an inflection point. AI can now work with tools of physics, simulation and design, said Timothy Costa, vice president and general manager of computational engineering at NVIDIA. With NVIDIA Agent Toolkit, developers can build agentic engineers that reason using physics, run complex simulations and generate high-fidelity data to become a new engine for innovation in chip and system design. The company also highlighted NVIDIA Nemotron 3 Ultra, which it says leads open models for agentic register-transfer level (RTL) coding when used with the NVIDIA Research ACE-RTL agent. The model is designed to help enterprises build AI agents for chip design while allowing deployment on local infrastructure using proprietary datasets. Cadence is using Nemotron models, CUDA-X libraries and accelerated computing in its AuraStack AI Super Agent and Millennium M2000 platform, claiming up to 20x faster multiphysics performance for advanced packaging and PCB design. Synopsys is incorporating the Agent Toolkit, NeMo technologies and Nemotron models into AgentEngineer to automate chip and system design workflows, while Siemens is combining NVIDIA AI models and CUDA-X libraries with its Fuse EDA AI Agent to orchestrate multi-tool engineering processes. Samsung is applying PhysicsNeMo for chip-scale thermal-stress analysis and using CUDA technologies to accelerate computational lithography, while Keysight says it has achieved up to a 10x improvement in electromagnetic simulations using cuDSS. Samsung, Synopsys and TSMC are also integrating cuEST into GPU-accelerated quantum chemistry workflows, with NVIDIA claiming speed-ups of up to 50x for key workloads.