Nvidia Principal R&D Software Architect - Machine Learning VLSI 

United States, Texas 

445516679

What you’ll be doing:

• Invent and optimize new methods for chip- and block-level floorplanning with PPA (performance, power, area) optimization, across a suite of internal tools. These tools already outperform the industry's alternatives in high-capacity analysis and optimization and will advance even further with your contributions.

• Particularly focus on algorithms that apply your experience with LLMs (Large Language Models), GANs (Generative Adversarial Networks), GNNs (Graph Neural Networks), RL (Reinforcement Learning), and other modern machine learning strategies.

• Translate novel ML ideas into practical, complete, and useful solutions to physical design problems, particularly at architectural and RTL level.

• As with any software engineering team, we do write a lot of code, but this is broader than a typical CAD or EDA role. Instead, we as a team own the whole process from discovery and invention of new optimization opportunities, to developing solutions and working directly inside design teams to facilitate deployment.

What we need to see:

• PhD or equivalent experience in Electrical Engineering or Computer Science.

• Minimum 12+ yearsexperience in CAD software development with at least 3 years in applications of advanced machine learning.

• Proven prior success in developing and applying advanced ML models to chip design, including multiple examples using GNNs, GANs, and/or LLMs.

• Product focus: although publication has its place, our group most recognizes innovations that are deployable to future products. As such, understanding the challenges and corresponding strategies for full-scale deployment is important.

• Demonstrated ability in software development using C++.

• Thorough VLSI understanding related to PPA with particularly advanced knowledge in one or more PPA-related optimization algorithms.

• Familiarity with design implementation tools required to prove and debug ideas, such as Fusion Compiler, PrimeTime, ICC2, and Innovus, as well as exposure to implementation flows commonly in Python, Tcl, and/or Perl.

• Strong communication and interpersonal skills.

Ways to stand out from the crowd:

• C++14 or newer experience, such as lambdas and concurrency.

• Formal understanding of traditional optimization techniques and principles, including A*, simulated annealing, gradient descent, differentiability, stochastic processes, and how these compare and combine with various ML strategies.

• Advanced understanding of floorplan representation, timing modeling, and fast global placement and routing algorithms.

• In general, an obsession with performance and the practical skills to build highly innovative software for world leading hardware.