Chimera: A Hybrid Machine Learning-Driven Multi-Objective Design Space Exploration Tool for FPGA High-Level Synthesis

Mang Yu, Sitao Huang, Deming Chen

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In recent years, hardware accelerators based on field programmable gate arrays (FPGA) have been widely applied and the high-level synthesis (HLS) tools were created to facilitate the design of these accelerators. However, achieving high performance with HLS is still time-consuming and requires expert knowledge. Therefore, we present Chimera, an automated design space exploration tool for applying HLS optimization directives. It utilizes a novel multi-objective exploration method that seamlessly integrates active learning, evolutionary algorithm, and Thompson sampling, which enables it to find a set of optimized designs on a Pareto curve by only evaluating a small number of design points. On the Rosetta benchmark suite, Chimera explored design points that have the same or superior performance compared to highly optimized hand-tuned designs created by expert HLS users in less than 24 h. Moreover, it explores a Pareto frontier, where the elbow point can save up to 26% of flip-flop resource with negligible performance overhead.

Original languageEnglish (US)
Title of host publicationIntelligent Data Engineering and Automated Learning - 22nd International Conference, IDEAL 2021, Proceedings
EditorsDavid Camacho, Peter Tino, Richard Allmendinger, Hujun Yin, Antonio J. Tallón-Ballesteros, Ke Tang, Sung-Bae Cho, Paulo Novais, Susana Nascimento
PublisherSpringer
Pages524-536
Number of pages13
ISBN (Print)9783030916077
DOIs
StatePublished - 2021
Event22nd International Conference on Intelligent Data Engineering and Automated Learning, IDEAL 2021 - Virtual, Online
Duration: Nov 25 2021Nov 27 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13113 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference22nd International Conference on Intelligent Data Engineering and Automated Learning, IDEAL 2021
CityVirtual, Online
Period11/25/2111/27/21

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Chimera: A Hybrid Machine Learning-Driven Multi-Objective Design Space Exploration Tool for FPGA High-Level Synthesis'. Together they form a unique fingerprint.

Cite this