Temperature-aware Data Allocation Strategy For 3D Flash Memory

NAND flash memory has the advantages of small size,large capacity,fast speed,small size,and hard to be damaged.It is widely used in enterprise data centers and personal media devices.As physical blocks have more physical pages in threedimensional flash memory based on charge extraction technology,its capacity is several times as big as traditional flash memory.Large capacity brings some problems,such as more garbage collection operations,increased writing amplification,making the temperature of physical block rise.Recent studies have shown that high temperature affects the stability of charge extraction technique and leads to charge loss.This thesis studies the temperature problem of three-dimensional flash memory based on charge extraction technology.First,this thesis proposes a data distribution strategy that senses spatial location.The strategy divides the physical blocks into different block bands according to the spatial position.The blocks in the same band have the same characteristic that their space positions are not adjacent.The allocation algorithm of the new block and the selection algorithm of the erase block are optimized.The continuous logical access is dispersed throughout the space,avoiding the heat accumulation caused by the access concentration.For garbage collection,a delay strategy is proposed to effectively reduce the temperature impact of the erased block on adjacent physical blocks.Compared with the baseline algorithm,the peak temperature is reduced by 33.7%,the number of uncorrectable errors is significantly reduced,and the total number of erasures is reduced by 67.8%.Second,this thesis proposes a data distribution optimization strategy for flash memory redundant array erasure code algorithm.Due to the random occurrence of bad blocks in flash memory and the characteristic of off-site update,the implementation of the erasure code algorithm in flash is more complicated than that in hard disk drives.For a large number of checksum's updates brought by the update of the data,an optimized cache strategy is proposed,which reduces the total number of I/O throughput for the flash array.For the read-write access concentration brought by page-level mapping,a spatial decentralized data allocation strategy and a sliding window strategy are proposed.With these improvements,the peak temperature of the 3D flash redundant array is reduced by 60.5%,and I/O throughput is reduced by 53.2%.This thesis focuses on the temperature problem of 3D flash memory,and the proposed data distribution strategy effectively reduces the temperature of 3D flash memory.The optimization method proposed in this paper can integrate more flash memory in the same physical space,improve the capacity of 3D flash memory products,and facilitate the use of 3D flash memory in a wider range of temperature conditions.