AnaCoNGA: Analytical HW-CNN Co-Design Using Nested Genetic Algorithms

Nael Fasfous, Manoj Rohit Vemparala, Alexander Frickenstein, Emanuele Valpreda, Driton Salihu, Julian Hufer, Anmol Singh, Naveen Shankar Nagaraja, Hans Joerg Voegel, Nguyen Anh Vu Doan, Maurizio Martina, Juergen Becker, Walter Stechele

Publikation: Beitrag in Buch/Bericht/KonferenzbandKonferenzbeitragBegutachtung

9 Zitate (Scopus)

Abstract

We present AnaCoNGA, an analytical co-design methodology, which enables two genetic algorithms to evaluate the fitness of design decisions on layer-wise quantization of a neural network and hardware (HW) resource allocation. We embed a hardware architecture search (HAS) algorithm into a quantization strategy search (QSS) algorithm to evaluate the hardware design Pareto-front of each considered quantization strategy. We harness the speed and flexibility of analytical HW-modeling to enable parallel HW-CNN co-design. With this approach, the QSS is focused on seeking high-accuracy quantization strategies which are guaranteed to have efficient hardware designs at the end of the search. Through AnaCoNGA, we improve the accuracy by 2.88 p.p. with respect to a uniform 2-bit ResNet20 on CIFAR-10, and achieve a 35% and 37% improvement in latency and DRAM accesses, while reducing LUT and BRAM resources by 9% and 59% respectively, when compared to a standard edge variant of the accelerator. The nested genetic algorithm formulation also reduces the search time by 51% compared to an equivalent, sequential co-design formulation.

OriginalspracheEnglisch
TitelProceedings of the 2022 Design, Automation and Test in Europe Conference and Exhibition, DATE 2022
Redakteure/-innenCristiana Bolchini, Ingrid Verbauwhede, Ioana Vatajelu
Herausgeber (Verlag)Institute of Electrical and Electronics Engineers Inc.
Seiten238-243
Seitenumfang6
ISBN (elektronisch)9783981926361
DOIs
PublikationsstatusVeröffentlicht - 2022
Extern publiziertJa
Veranstaltung2022 Design, Automation and Test in Europe Conference and Exhibition, DATE 2022 - Virtual, Online, Belgien
Dauer: 14 März 202223 März 2022

Publikationsreihe

NameProceedings of the 2022 Design, Automation and Test in Europe Conference and Exhibition, DATE 2022

Konferenz

Konferenz2022 Design, Automation and Test in Europe Conference and Exhibition, DATE 2022
Land/GebietBelgien
OrtVirtual, Online
Zeitraum14/03/2223/03/22

Fingerprint

Untersuchen Sie die Forschungsthemen von „AnaCoNGA: Analytical HW-CNN Co-Design Using Nested Genetic Algorithms“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren