# End-to-end Automatic Logic Optimization Exploration via Domain-specific Multi-armed Bandit

**Authors:** Walter Lau Neto, Yingjie Li, Pierre-Emmanuel Gaillardon, Cunxi Yu  
**arXiv:** [2202.07721](https://arxiv.org/abs/2202.07721)  
**Date:** 2022-02-15  
**Summary page:** https://www.summarizepaper.com/en/paper/2202.07721v1/

## Key points

- Electronic design automation (EDA) is using decision intelligence to reduce manual efforts and improve the design closure process.
- FlowTune is a generic end-to-end sequential decision-making framework for synthesis toolflow optimization.
- FlowTune addresses various optimization problems related to Boolean optimization, logic synthesis, and technology mapping problems.
- The proposed domain-specific multi-stage multi-armed bandit (MAB) approach is highly extensible and generalizable.
- FlowTune outperforms both hand-crafted flows and ML explored flows in terms of quality of results while being orders of magnitude faster compared to ML-based approaches.
- FlowTune is fully integrated with several industrial tools including ABC, Yosys, VTR, LSOracle, OpenFPGA.

## Summary

The field of electronic design automation (EDA) has seen a growing use of decision intelligence in recent years, with the aim of reducing manual efforts and improving the design closure process. To address these challenges, this paper presents FlowTune, a generic end-to-end sequential decision-making framework for synthesis toolflow optimization that utilizes a novel high-performance domain-specific multi-stage multi-armed bandit (MAB) approach. FlowTune addresses various optimization problems related to Boolean optimization such as And-Inv-Graphs (# nodes), Conjunction Normal Form (CNF) minimization (# clauses) for Boolean Satisfiability, logic synthesis and technology mapping problems such as post static timing analysis (STA) delay and area optimization for standard-cell technology mapping, and FPGA technology mapping for 6-in LUT architectures. The proposed domain-specific MAB approach is highly extensible and generalizable, demonstrated through its evaluation at the post-routing stage with two different FPGA backend tools (OpenFPGA and VPR) and two different logic synthesis representations (AIGs and MIGs). FlowTune is fully integrated with several industrial tools including ABC [1], Yosys [2], VTR [3], LSOracle [4], OpenFPGA [5]. The experimental results conducted on various design stages demonstrate that FlowTune outperforms both hand-crafted flows [1] and ML explored flows [6], [7] in terms of quality of results while being orders of magnitude faster compared to ML-based approaches. Overall, this paper presents an innovative solution to EDA optimization challenges that can significantly reduce manual efforts while improving the design closure process.