Research On Characteristics Of 3D CT NAND Flash And Optimized Design Of Flash Translation Layer Algorithms

Compared with traditional hard disk drives,NAND flash memory is gradually becoming the mainstream storage medium in many fields such as mobile Internet terminals,video surveillance,data centers and aerospace,due to its high performance,small latency and low power consumption.However,on one hand,with the rapid development of Internet,big data and machine learning,users' demand for storage capacity is presenting an explosive growth.On the other hand,flash chip process technology has been reduced from 60 x nm to 10 x nm,which is facing physical limits of further scaling down,thus bringing challenges to the continuous growth of NAND flash storage density.3D NAND flash stores data in vertically stacked layers,thus effectively increases the storage density,and has received extensive attention from both academia and industry.NAND flash storage has reliability problems due to limited program/erase times and data retention issues.This paper researches on the performance variations,flash error distributions and storage reliability problems of 3D Charge Trap(CT)NAND flash on an ARM-and FPGA-based testing platform,studies the analysis and evaluation methods of performance and reliability during NAND flash lifetime,and proposes flash management algorithm optimization strategies based on the characteristics.In order to improve the storage reliability,NAND flash marks a whole block as bad block once any one of its pages fails,thus wasting the potential storage capacity.This paper proposes a space-aware fault-tolerant bad block management algorithm,SA-BBM,that only marks pages in a tier to failed state when a page fails and only migrates the valid data in that tier,which can reduce the overhead of data migration,improve storage utilization and effectively prolong the lifetime of NAND flash.In order to verify the effectiveness of SA-BBM,we build a simulation platform based on 3DSim,and run trace files collected from real workloads.Experimental results show that compared with the traditional bad block management algorithm,SA-BBM can prolong the lifetime of 3D CT NAND flash by 66% on average.