Can Symbolic Regression of Boolean Functions Boost Logic Synthesis?
Keywords: Chip Design, Logic Synthesis, Symbolic Regression, Monte-Carlo Tree Search
Abstract: Logic synthesis, which aims to synthesize a compact logic circuit with minimized size while exactly satisfying a given functionality, plays an important role in chip design. Recently, symbolic regression (SR) has shown great success in scientific discovery to recover underlying mathematical functions from given datasets. However, we found from extensive experiments that existing SR methods struggle to recover an exact and compact boolean function for logic synthesis given a truth table, i.e., complete input-output pairs of the circuit. The major challenges include (1) the greater complexity of underlying boolean functions compared to mathematical functions, and (2) the complex objectives involving both exact recovery and expression optimization towards circuit minimization. To address these challenges, we propose a novel symbolic factorized boolean searcher (SINE) to recover exact and compact boolean functions towards logic synthesis. Motivated by the Shannon decomposition theorem, SINE proposes a factorized boolean function representation to decompose the underlying boolean function into multiple simplified sub-functions, significantly reducing their complexity and thus improving the recovery accuracy. Moreover, based on the key observation that, logical sharing is significant for circuit size minimization. SINE proposes a self symmetric sub-expression motif operators mining mechanism to enhance the monte-carlo tree search method for optimized boolean function learning. To the best of our knowledge, SINE is the first symbolic regression framework capable of exactly recovering optimized boolean functions for circuit optimization. Experiments on circuits across a wide range of inputs demonstrate that SINE significantly improves the recovery accuracy and decreases the size of synthesized circuits by up to 24.32\% compared to state-of-the-art methods.
Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)
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: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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.
Submission Number: 7267
Loading