Reproduce “Efficient Stimuli Generation using Reinforcement Learning in Design Verification” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Deepak Narayan Gadde, Thomas Nalapat, Aman Kumar, Djones Lettnin, Wolfgang Kunz, Sebastian Simon
An automated, design-agnostic framework leverages Reinforcement Learning to efficiently generate stimuli for hardware design verification, reducing the number of simulations needed to achieve maximum code coverage. It demonstrates that RL agents consistently require fewer stimuli than random simulations for coverage closure across various System-on-Chip designs, particularly with the PPO algorithm.
Key insights
- Reinforcement Learning can significantly automate and accelerate the process of achieving code coverage closure in hardware design verification by generating more targeted stimuli.
- A design-agnostic framework enabled by metamodeling reduces manual intervention, making RL-based verification solutions more scalable and practical across diverse designs.
- Different RL algorithms and reward schemes exhibit varying efficiencies, with PPO generally performing well for code coverage and A2C showing particular strength for FSM coverage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.