This thesis presents a lifetime management technique for NAND flash memory systems, which extends the lifetime of NAND flash memory and reduces overhead of write and erase operations in wear leveling and garbage collection. The proposed lifetime management technique for NAND flash memory consists of dynamic wear leveling, static wear leveling, hybrid wear leveling and decision technique of cold blocks.
First, this thesis proposes Migration Cost Sensitive Garbage Collection (MCSGC), which is a dynamic wear leveling. The proposed MCSGC can separate into hot/cold data using recency information and update frequency. In proposed MCSGC, victim blocks in garbage collection are selected based on elapsed time from last page invalidation to current time on blocks. To improve the efficiency of garbage collection, MCSGC also separates cold data from valid pages in victim blocks according to update frequency. In experiment results, MCSGC prolongs the lifetime of NAND flash memory up to 82% and reduces the overhead of erase operations up to 29% compared with other dynamic wear leveling techniques.
Second, this thesis proposes Recency based Wear Leveling (RbWL), which is a static wear leveling. The proposed RbWL can reduce write and erase operations by an adaptive wear leveling according to the remaining lifetime of NAND flash memory and erase count variation of hot/cold blocks. This overhead reduction can help the lifetime prolongment of NAND flash memory. The simulation results show that RbWL prolongs the lifetime of NAND flash memory up to 33% and reduces the overhead of erase operations and write operations up to 46% and 25% respectively, compared with other static wear leveling techniques.
Third, this thesis proposes Hybrid Wear Leveling (HWL), which combines with MCSGC and RbWL. HWL can solve cold block problem of dynamic wear leveling which results from non-updated blocks, while it has similar overhead of dynamic wear leveling techniques. In experiment results, HWL can prolong the lifetime of NAND flash memory up to 29% and reduce the overhead of erase operations and write operations up to 29% and 46% respectively, compared with MCSGC and RbWL.
Finally, this thesis proposes Decision of Cold Block (DCB) which defines hot blocks to perform dynamic wear leveling and cold blocks to perform static wear leveling in an environment using HWL. DCB defines cold blocks based on the number of valid pages in blocks so that these blocks are not selected as victim blocks in dynamic wear leveling technique. In this way, DCB can reduce the overhead of page migration for wear leveling. Experimental results show that DCB can prolong the lifetime of NAND flash memory up to 6% and reduce the overhead of erase operations and write operations up to 6% and 7% respectively, compared with Non-DCB systems.