Research On A Hybrid Wear-leveling Algorithm For Solid State Disk

Because of the restriction of the peculiar physical characteristics of flash memory,the erase cycle of a block should not exceed a limitation, or the data stored in the blockwill be unreliable. In order to prolong the lifetime of flash based Solid-State-Disk (SSD),wear-leveling algorithm is required to evenly distribute the wear within the wholestorage space. But every wear-leveling algorithm brings in extra erase cycles, and theimpact of extra erase cycle stands out as the endurance of flash memory decreasing.This dissertation focuses on the problem of extra erase cycles introduced bywear-leveling algorithm in Solid-State-Disk controller. The main works andcontributions of this dissertation are as follow:(1)This dissertation specified the characteristics of flash memory andSolid-State-Disk, and then emphasized the importance of wear-leveling algorithm insidea Solid-State-Disk. This dissertation illuminated the research trend of wear-levelingalgorithm by presenting a review of wear-leveling algorithm and chose hybridwear-leveling to improve after analyzing several wear-leveling algorithms.(2)This dissertation presented an improvement on hybrid wear-leveling (HWL)algorithm to solve the problem of high overhead ratio. The presented improvementseparated the wear-leveling method inside a group from the wear-leveling methodbetween different groups. A random wear-leveling method inside each group and adeterminant wear-leveling between different groups are adopted respectively. The extraerase cycle produce by random wear-leveling method was reduced by increasing theaccuracy when choosing cold data. The wear-leveling method between different groupswas based on threshold control, which guaranteed the low overhead required by thewear-leveling operations between different groups. The improvement utilized thefeature that flash memories in Solid-State-Disk were access by parallel bus to reduce theRAM space required by wear-leveling algorithm.(3)A series of simulation and test experiments were designed to evaluate theperformance merits of improved hybrid wear-leveling algorithm. The experiment resultsshowed that the improved hybrid wear-leveling algorithm maintain standard deviationof erase count and RAM space consumption onto the quantity of hybrid wear-levelingalgorithm and reduce the over-head ratio of the algorithm from1.7to1.2.