22 Sept 2022, 12:42 (modified: 26 Oct 2022, 14:21)ICLR 2023 Conference Withdrawn SubmissionReaders: Everyone
Keywords: Design Space Exploration, Reinforcement Learning
TL;DR: Microarchitecture design space exploration via reinforcement learning for RISC-V processors
Abstract: Microarchitecture determines a processor's detailed structure, affecting the processor's performance, power, and area (PPA). Deciding on a microarchitecture to achieve a good balance between the PPA values is a non-trivial problem. Previous arts mainly require expert knowledge. The solution becomes inefficient as nowadays processors become increasingly complicated. Machine learning has solved problems automatically with high-quality results via reduced access to domain knowledge. In this paper, we formulate the problem as a Markov decision process and propose an end-to-end solution framework via reinforcement learning. Firstly, a dynamically-weighted reward design is proposed to accommodate the optimization of multiple negatively-correlated objectives. Secondly, local heuristic search is adopted in the action design with prior knowledge of microarchitectures. Thirdly, lightweight calibrated PPA models are incorporated to accelerate the learning process. Experimenting with electronic design automation (EDA) tools on famous RISC-V processors demonstrate that our methodology can learn from experience and outperform human implementations and previous arts' solutions in PPA and overall running time.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics
Submission Guidelines: Yes
Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)
5 Replies