Nvidia accelerates Feynman chip design, manufacture on Blackwell GPU
Nvidia is using its own Blackwell GPUs to accelerate the design and manufacturing of its next generation Vera, Rubin and Feynman chips.
EDA vendors including Synopsys and Siemens EDA have used libraries from Nvidia to accelerate a wide range of design tools.
Synopsys PrimeSim is projected to accelerate circuit simulation by up to 30x running on the GB200 Grace Blackwell superchip, and adding other AI models to its chip design co-pilot via the Nvidia NIMS microservice.
“Chip design is one of the most complex engineering challenges in human history,” said Jensen Huang, CEO of Nvidia. “With Blackwell and CUDA-X, Synopsys is cutting simulation times from days to hours—advancing chip design to power the AI revolution.”
Nvidia has also developed a library to accelerate computational lithography that is used to design masks for chip making that can enable ever smaller features on a chip. This helps to increase the performance of the chips and cuts the cost of mask sets, which run into the tens of millions of dollars for leading edge processes. Moving to the latest Blackwell and Blackwell Ultra GPUs will shorted the design times.
“Over four years we have taken the entire process of computational lithography,” said Huang, talking about the CuLitho library for its CUDA framework. The company have over 900 CUDA libraries that run on its GPUs for different applications.
“TSMC, ASML, Synopsys, Mentor [Siemens EDA}, they are all using this. This is now at its tipping point and in five years time every mask will be designed on CUDA,” he said.
- CEO interview: AI creates step change for computational lithography
- TSMC taps Synopsys for 2nm computational lithography
The Synopsys Proteus tool for optical proximity correction (OPC) and inverse imaging technology (ILT) to resolve challenges of designing masks will run on the B200 Blackwell GPU to computational lithography simulations by up to 20x.
Synopsys is also expanding support for the NVIDIA Grace CPU architecture and enabling more than 15 Synopsys tools in 2025. These will also see a boost running on the coming Vera CPU with 88 custom ARM Neoverse cores which ships next year.
For transistor level TCAD simulation, early evaluation of the Synopsys Sentauras TCAD process and device simulation tool expects acceleration of up to 10x. This is currently under development and is expected to be available to customers later this year.
Synopsys is also using the materials and quantum CUDA-X libraries for its QuantumATK for atomic-scale modeling for semiconductor and materials research and development, although this is currently running on the previous Hopper GPU architecture.
It is adding the NIMS microservice to its Synopsys.ai copilot, which will double the speed of answers for design problems and gives access to other AI models.
Synopsys will see further benefits from the Nvidia Blackwell GPUs through its proposed acquisition of Ansys.
A new library called CuDSS is a direct sparse solver for CAE simulation models. “This is one of the biggest things that has happened in the last year,” said Huang. “We enable every important CAE library to be accelerated.”
Companies such as Ansys, Cadelnce Design Systems and Dassault use the libraries to solve linear systems with very sparse matrices. Direct Sparse Solvers are an important part of numerical computing for real-time applications like autonomous driving and process simulation, where increasing complexity and high throughput demands a robust direct solver.
The library uses the CPU’s sequential computing and the GPU’s parallel computing in the Grace Hopper superchip to solve sparse matrices of any size with only a few non-zero elements per row. The result is significantly higher performance than CPU-only solvers. The next move will be to run the solver on the Grace Blackwell GB300 systems and then on the Vera Blackwell superchip.
www.nvidia.com; www.synopsys.com
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
