TY - GEN
T1 - Chimera
T2 - 22nd International Conference on Intelligent Data Engineering and Automated Learning, IDEAL 2021
AU - Yu, Mang
AU - Huang, Sitao
AU - Chen, Deming
N1 - Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021
Y1 - 2021
N2 - 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.
AB - 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.
UR - http://www.scopus.com/inward/record.url?scp=85126480700&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85126480700&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-91608-4_52
DO - 10.1007/978-3-030-91608-4_52
M3 - Conference contribution
AN - SCOPUS:85126480700
SN - 9783030916077
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 524
EP - 536
BT - Intelligent Data Engineering and Automated Learning - 22nd International Conference, IDEAL 2021, Proceedings
A2 - Camacho, David
A2 - Tino, Peter
A2 - Allmendinger, Richard
A2 - Yin, Hujun
A2 - Tallón-Ballesteros, Antonio J.
A2 - Tang, Ke
A2 - Cho, Sung-Bae
A2 - Novais, Paulo
A2 - Nascimento, Susana
PB - Springer
Y2 - 25 November 2021 through 27 November 2021
ER -