Publication:
QuFi: adaptive tiled Gustavson output reuse for edge sparse DNN accelerators

Simple item page
dc.contributor.authorNavarro, Adrián
dc.contributor.authorCano, José
dc.contributor.authorAbellán, José L.
dc.contributor.authorAcacio, Manuel E.
dc.contributor.departmentIngeniería y Tecnología de Computadores
dc.contributor.otherFacultad de Informática
dc.date.accessioned2025-09-24T11:41:45Z
dc.date.available2025-09-24T11:41:45Z
dc.date.issued2025-11-12
dc.description.abstractIn recent years, a myriad of Deep Neural Network (DNN) accelerator architectures have been proposed targeting efficient Sparse Matrix-Sparse Matrix Multiplication (SpMSpM) for the edge. Most focus on the dataflow, i.e., the order in which processing elements perform multiply-accumulate operations, while overlooking the influence of memory structures. For Gustavson-based dataflows, which have been proven to be the most efficient for the mid-sparsity scenarios exhibited by sparse DNN workloads, we find that the memory structure's organization storing partial results significantly affects performance. Different DNN models, even layers within the same model, impose diverse requirements on this structure. Rigid designs, commonly assumed so far, often cause frequent merging operations that degrade performance. In this work, we advocate for the necessity of providing adaptability at this memory structure and propose QuFi, a configurable merging memory structure designed for easy integration into any Gustavson-based accelerator and tailored to edge devices. Implemented as a collection of queues across multiple memory banks, QuFi supports reconfigurability through queue fusion and dynamic clustering while providing high bandwidth. Through a detailed evaluation that considers its inclusion into a state-of-the-art accelerator design for the edge, we show that QuFi provides an average speedup of 1.64x, up to 73% reduction in off-chip memory traffic, and leads to total accelerator area savings of 17%.
dc.formatapplication/pdf
dc.format.extent8
dc.identifier.urihttp://hdl.handle.net/10201/159689
dc.languageeng
dc.publisherIEEE
dc.relationThis work has been funded by the MCIN/AEI/10.13039/501100011033/ and the “ERDF A way of making Europe”, EU, under grant PID2022-136315OB-I00; by MICIU/AEI/10.13039/501100011033 and the ``European Union NextGenerationEU/PRTR’’, under grant RYC2021-031966-I. This work was also partially supported by the EU Project dAIEDGE (GA Nr 101120726) and the Innovate UK Horizon Europe Guarantee (GA Nr 10090788). Adrián Navarro was supported by grant 22294/FPI/23 from Fundación Séneca (Región de Murcia, Spain).
dc.relation.ispartofThe 43rd IEEE International Conference on Computer Design (ICCD 2025) November 10-12, 2025 Dallas, USA
dc.rightsAttribution-NonCommercial-NoDerivatives 4.0 International*
dc.rights© 2025
dc.rights.accessRightsinfo:eu-repo/semantics/openAccess
dc.rights.urihttp://creativecommons.org/licenses/by-nc-nd/4.0/*
dc.subject.odsNo relacionado con ningún objetivo de desarrollo sostenible
dc.titleQuFi: adaptive tiled Gustavson output reuse for edge sparse DNN accelerators
dc.typeinfo:eu-repo/semantics/article
dspace.entity.typePublicationes
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
ICCD_25___Camera_Ready.pdf
Size:
1.38 MB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
1.37 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Artículos