NVIDIA has introduced cuPotential, a new CUDA-X1. CUDA-X librariesNVIDIA’s CUDA-X libraries are accelerated software components used to speed up compute-intensive semiconductor workflows.2. CUDA ExtendedCUDA-X is NVIDIA's extended family of libraries and software for accelerated computing tasks. library for building AIartificial intelligence agents that simulate atomic and molecular interactions. In early-access testing, the library added another 2-3x speedup on top of workflows that were already optimized for GPUgraphics processing unitA GPU is a parallel graphics processor widely used for AI acceleration.s. The company positions the tool as part of a broader push to make semiconductor fabs more AIartificial intelligence-driven by accelerating materials modeling and other compute-heavy engineering tasks.
Alongside cuPotential, NVIDIA is promoting NanoVDBNVIDIA Accelerated NanoVDBNanoVDB is an open-source library for working with 3D data and accelerating simulation workflows. for 3D data processing and NPPNVIDIA Performance PrimitivesNPP is a GPU image-processing library used to accelerate inspection and test workflows. for GPUgraphics processing unitA GPU is a parallel graphics processor widely used for AI acceleration.-based image processing. NanoVDBNVIDIA Accelerated NanoVDBNanoVDB is an open-source library for working with 3D data and accelerating simulation workflows. is already being used to speed chipmaking simulation, with Lam Research achieving 30x faster simulation, while NPPNVIDIA Performance PrimitivesNPP is a GPU image-processing library used to accelerate inspection and test workflows. is helping accelerate inspection workloads for fabs. NVIDIA also points to cuLitho for chip-pattern calculations and RAPIDSRAPIDS Accelerator Program for Data ScienceRAPIDS is NVIDIA’s GPU data-processing software stack used for large-scale analytics. cuDF for large-scale dataset processing.
The announcement ties these libraries to a wider AIartificial intelligence-defined fab model, in which AIartificial intelligence helps engineers optimize processes, handle production issues and improve efficiency before changes reach the production line. NVIDIA says this requires faster modeling of chip designs, materials behavior and manufacturing processes, as well as digital twins for testing process changes before deployment.
Several semiconductor companies are using the stack across design, manufacturing and testing. Synopsys is using cuPotential, cuEST and cuDSS; early results cited include a potential 30x speedup with QuantumATKQuantum Atomistic ToolKitQuantumATK is used in electronic-structure and materials simulation workflows. using cuEST and a 12x speedup in S-Device benchmarks with cuDSS. Cadence is applying NVIDIA accelerated computing and CUDA-X1. CUDA-X librariesNVIDIA’s CUDA-X libraries are accelerated software components used to speed up compute-intensive semiconductor workflows.2. CUDA ExtendedCUDA-X is NVIDIA's extended family of libraries and software for accelerated computing tasks. libraries to agentic workflows for electromechanical systems, including advanced packaging. COMSOL is using cuDSS for multiphysics simulation, and Voltai says its Voltaire platform compresses electro-thermal-mechanical workflows from months to a day.
On the manufacturing side, Siemens is using cuLitho to speed up Calibre-based pattern refinement and mask preparation, while Lam Research plans to integrate NanoVDBNVIDIA Accelerated NanoVDBNanoVDB is an open-source library for working with 3D data and accelerating simulation workflows. into its SEMulator3D platform. NVIDIA says that addition is expected to accelerate selected wafer-etch modeling by up to 30x, and PhysicsNeMoPhysics Neural Operator ModelPhysicsNeMo is a framework for training physics-informed AI models that predict physical behavior. has been used to reduce plasma-prediction simulation time to milliseconds instead of days.
Inspection and test are also part of the rollout. KLA used NVIDIA NPPNVIDIA Performance PrimitivesNPP is a GPU image-processing library used to accelerate inspection and test workflows. to create a new kernel that reached performance about 50x faster than earlier CPUcentral processing unitMain processor that executes program instructions and performs data operations. versions. Teradyne is using NPPNVIDIA Performance PrimitivesNPP is a GPU image-processing library used to accelerate inspection and test workflows., RAPIDSRAPIDS Accelerator Program for Data ScienceRAPIDS is NVIDIA’s GPU data-processing software stack used for large-scale analytics. cuDF and CuPyCUDA PythonCuPy is a Python library for array and numerical computing on GPUs. for image-sensor testing and probe-maintenance inference, while Advantest is integrating NVIDIA Nemotron and Kumo-Anomaly into ACSAdvantest Cloud SolutionsACS RTDI is the production analytics system Advantest integrates with NVIDIA tools. RTDIreal-time data integrationRTDI refers to real-time diagnostics and inferencing for real-time diagnostics and production analytics on Grace Blackwell systems.
NVIDIA also cites fab-operation use cases beyond inspection. Applied Materials is using Omniverse and CUDA-X1. CUDA-X librariesNVIDIA’s CUDA-X libraries are accelerated software components used to speed up compute-intensive semiconductor workflows.2. CUDA ExtendedCUDA-X is NVIDIA's extended family of libraries and software for accelerated computing tasks. libraries including cuDSS, cuEST and PhysicsNeMoPhysics Neural Operator ModelPhysicsNeMo is a framework for training physics-informed AI models that predict physical behavior. for materials discovery, process development and fab optimization, with up to 35x faster chamber simulations and digital twins. imec is applying NVIDIA tools to remove noise from SEM1. scanning electron microscopyImaging technique that scans a focused electron beam across a sample.2. specific equipment modelComputational method or mathematical model for simulating, describing or optimizing technical systems.3. scanning electron microscopeSEM images are used for high-resolution inspection and measurement. images for sub-nanometer measurements and to improve the lithography process window by 30%.
cuPotential is available in early access, NanoVDBNVIDIA Accelerated NanoVDBNanoVDB is an open-source library for working with 3D data and accelerating simulation workflows. is available as open source within OpenVDB, and NPPNVIDIA Performance PrimitivesNPP is a GPU image-processing library used to accelerate inspection and test workflows. is available through the CUDA Toolkit. Lam Research’s NanoVDBNVIDIA Accelerated NanoVDBNanoVDB is an open-source library for working with 3D data and accelerating simulation workflows.-accelerated SEMulator3D capability is targeted for release by the end of the year.