Nvidia Principal R&D Software Architect CUDA Physical Design EDA 

United States, Texas 

568313237

What you’ll be doing:

• Invent algorithms for rapidly estimating voltage IR drop, timing, and power in CUDA as part of a suite of internal EDA tools for gate-level performance optimization.

• Develop scalable cell current modeling techniques for detailed IR simulation on GPUs.

• Research and develop strategies to more efficiently deploy EDA algorithms on GPUs, such as using compact approximations, lossy compression, and dynamic partitioning.

• Collaborate with other developers to extend CUDA deployment into other VLSI domains, such as placement, routing, extraction, timing, logic simulation, SPICE, and DRCs.

• This is a broad, flexible, hands-on software development role that can evolve as new opportunities are discovered and your personal interests grow.
  

What we need to see:

• MS, PhD, or equivalent experience in Electrical Engineering, Computer Science, Physics, or Mathematics.

• 12+ years’ experience, including extensive use of C++ and parallel computing.

• Expertise with timing and power modeling techniques commonly used in VLSI, such as Liberty CCS (Composite Current Source) models, moment analysis, Bayesian networks, and frequency domain analysis.

• Strong understanding of VLSI physical design, including power grids, IR drop, place and route, static timing analysis, and dynamic power calculation.

• Experience with CUDA packages, especially NVIDIA’s cuDSS (CUDA Direct Sparse Solver) library, cuBLAS, and/or cuSPARSE.
  

Ways to stand out from the crowd:

• Familiarity with industry standard power integrity tools, such as Ansys RedHawk/SeaHawk.

• Background with machine learning for VLSI, especially use of GNNs (Graph Neural Networks) or PINNs (Physics-Informed Neural Networks).

• Familiarity with modern C++17/C++14 concepts and usage.

• Expertise in other EDA components well-suited for GPU acceleration, including computational geometry, detailed physical design, SPICE, and functional simulation.