Software R&D Engineer, RTL Optimization Tools

About this role

NVIDIA's success builds on a foundation of industry leading hardware. A key strategy in achieving this is our combining of the best of external EDA with highly optimized, internal EDA tools. Our team develops these tools by fusing advances in parallel computing, machine learning, and novel algorithms in C++. We are seeking an innovative CAD Software Engineer with particular interest in strategies and algorithms for large scale RTL quality, timing, and power optimization. Such optimization usually includes a mix of graph-based algorithms, AI, and feedback from RTL designers, so having experience relevant to each of those areas would be ideal. In practice, techniques often depend on many related domains, so a solid understanding of DFT, clock distribution, power gating, and other SOC integration aspects is essential.

Developing software within a leading hardware company means getting to almost exclusively focus on the latest processes and most advanced designs. We're not bogged down by legacy support, niche roles, or convoluted approval processes. Our developers enjoy unusually high intellectual freedom and the ability to explore broad roles. If you like to work across many technical areas and see your successes directly realized in the world's best AI hardware, this is it!

What you’ll be doing:

• Invent new methods to enable parallel, graph-based RTL traversal, analysis, and manipulation.
• Devise strategies for rapidly analyzing the impact of RTL changes on data path latency, power, and impact to DFT, clocking, and power delivery.
• Explore use of LLMs (Large Language Models), GNNs (Graph Neural Networks), GANs (Generative Adversarial Networks), and Reinforcement Learning for suggesting or automatically implementing RTL modifications.
• Explore high performance algorithms for clustering, min cost tree covering (technology mapping), datapath implementation and other details of logic synthesis, especially that efficiently incorporate human insight.
• As with an...