Skip to main navigation Skip to search Skip to main content

RecFlash: Fast Recommendation Inference on NAND Flash-Based In-Storage Computing with Embedding-Optimized Data Mapping

  • Sogang University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Recommendation systems are widely used for personalized suggestions, but the growing amount of user data makes real-time processing difficult. NAND flash-based in-storage computing (ISC) is a favorable solution due to its large capacity, but random memory access patterns in recommendation systems cause underutilized internal bandwidth and degraded performance. This paper proposes RecFlash, a fast recommendation inference accelerator using a data remapping algorithm with NAND flash-based ISC. Experimental results show that it improves latency by up to 81 % over the existing ISC architectures.

Original languageEnglish
Title of host publicationProceedings - 2025 21st IEEE Asia Pacific Conference on Circuits and Systems, APCCAS 2025
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798331589073
DOIs
StatePublished - 2025
Event2025 21st IEEE Asia Pacific Conference on Circuits and Systems, APCCAS 2025 - Busan, Korea, Republic of
Duration: 12 Oct 202515 Oct 2025

Publication series

NameProceedings - 2025 21st IEEE Asia Pacific Conference on Circuits and Systems, APCCAS 2025

Conference

Conference2025 21st IEEE Asia Pacific Conference on Circuits and Systems, APCCAS 2025
Country/TerritoryKorea, Republic of
CityBusan
Period12/10/2515/10/25

Keywords

  • data remapping
  • hardware accelerator
  • in-storage computing
  • NAND flash memory
  • Recommendation system

Fingerprint

Dive into the research topics of 'RecFlash: Fast Recommendation Inference on NAND Flash-Based In-Storage Computing with Embedding-Optimized Data Mapping'. Together they form a unique fingerprint.

Cite this