Too Big to Fail? Active Few-shot Learning Guided Logic Synthesis
TimeWednesday, July 13th6pm - 7pm PDT
LocationLevel 2 Lobby
Event Type
Networking Reception
Work-in-Progress Poster
DescriptionLogic synthesis is a hard combinatorial problem relying on black-box approaches (e.g., simulated annealing) to generate sub-optimal synthesis transformation sequences (“synthesis recipe"). However, black-box optimization requires time-consuming synthesis runs (especially on large circuits) preventing fast exploration of possible synthesis recipes. In our work, we fine-tune a pre-trained model (on past synthesis data) to accurately predict the quality of a synthesis recipe for an unseen netlist. We show blackbox optimizer using fine-tuned model as evaluator can generate recipes of similar quality (compared to actual synthesis run) with an average 55\% lower run-time and achieve better quality-of-result(QoR) than state-of-art machine learning approaches.