基于强化学习的NAND闪存垃圾回收算法

打开文本图片集

中图分类号：TP333；TP301.6 文献标识码：A 文章编号：2096-4706（2026）05-0056-04

A Reinforcement Learning-based Garbage Collection Algorithm for NAND Flash Memory

RANYong，YANHua （College ofElectronics and Information Engineering，Sichuan University，Chengdu 61oo65，China）

Abstract： NANDfash memory requires a garbage collection strategy due to its physical constraints， which significantly degradessystemefciencyPrecisehotandcolddataseparationiscritical.Inaddition，NANDflashmemorycanonlyedurea limitednumberoferaseoperations.Therefore，improvingwearlevelingiscrucialforextendingitslifespan.Existingmethodsfail tolearn dataaccess pattersonlineandcannotaccuratelyidentifyhotandcolddataunderdynamic workloads.Toaddresstese isues，aReinforcementLearing-based Garbage Colectionalgorithm （RLGC）is proposed.Itrealizesaurate clasificationof hotandcolddata throughonline Adaptive Leamingand integrating wearleveling mechanism.Experimentalresultsshowthat compared withtheexisting bestperformanceofthebaselinealgorithm，the proposedalgorithmreduces theblock erasetimes by 16.4% ， the effective page copy times by 50.0% ， and improves the wear leveling degree by 33.3% ，which significantly improves the garbage collection efficiency and prolongs the lifetime of flash memory.

KeyWords： NAND flash memory; garbage collection; Reinforcement Learning; wear leveling

0 引言

高密度、低功耗的非易失性存储介质NAND闪存，因数据容量大、访问延迟低等优势，被广泛应用于嵌入式系统、固态硬盘（SSD）及各类消费电子设备中。（剩余7357字）