NVIDIA Agent Toolkit gains PhysicsNeMo and CUDA-X for autonomous chip design

NVIDIA today announced a major expansion of its NVIDIA Agent Toolkit for engineering, now including NVIDIA PhysicsNeMo and CUDA-X libraries as agent-ready tools. PhysicsNeMo provides AI physics skills that allow agents to train and deploy customizable physics models for complex design and simulation tasks, effectively turning model architectures into callable tools. The updated CUDA-X libraries introduce three key components: cuISS (CUDA Iterative Sparse Solvers) accelerates large sparse linear systems for physics-based simulations; cuDSS (CUDA Direct Sparse Solvers) speeds up large complex sparse linear systems for EDA and scientific simulation; and cuEST (CUDA Electronic Structure Theory) brings high-accuracy quantum chemistry simulations at device-relevant scales using density functional theory. These capabilities enable agents to run accelerated simulations and integrate quantum chemistry into production workflows.

Additionally, NVIDIA highlights that its Nemotron 3 Ultra model leads among open models in agentic register-transfer level (RTL) coding, as demonstrated by the ACE-RTL agent from NVIDIA Research. This agent helps enterprises build customizable AI agents for chip design and verification, where RTL coding demands high accuracy and domain expertise. Industry leaders including Cadence, Siemens, and Synopsys are already using NVIDIA's accelerated computing and agentic AI to advance autonomous engineering workflows across chip design, verification, packaging, and systems. Timothy Costa, VP and GM of computational engineering at NVIDIA, stated: 'Engineering has reached an inflection point. AI can now work with tools of physics, simulation, and design.' The toolkit positions engineering teams to build agentic engineers that reason using physics, run complex simulations, and generate high-fidelity data, creating a new engine for innovation in chip and system design.