Prediction of Elapsed Time based Wear Leveling for NAND Flash Memory in Embedded Systems

In recent years, many storage systems use NAND flash memory increasingly as their secondary storages. NAND flash memory has non-volatile memory characteristics with low power, low latency and high reliability. On the other hand, NAND flash memory has different issue, compared to existing secondary storages, which is the characteristics such as erase-before-write, low endurance and different operation unit. These problems can be solved by using address translation table, garbage collection and wear leveling techniques. Unfortunately, previous works are difficult to directly be applied to embedded systems because they did not consider the memory requirement. In this paper, we propose PET-WL, which stands for Prediction of Elapsed Time based Wear Leveling, and it can be efficiently applied to embedded systems. Our policy has the characteristics to predict the elapsed time of each block using the difference of the number of invalid pages. In experimental results, PET-WL prolonged the lifetime of NAND flash memory up to 430% and reduced the page migration cost up to 39%, compared to other techniques.