Retention Leveling: Leverage Retention Refreshing and Wear Leveling Techniques to Enhance Flash Reliability with the Awareness of Temperature

Wear leveling cannot entirely mitigate retention errors, which have recently become $a$ substantial reliability issue due to cell shrinking. This paper focuses on preserving accurate data by addressing retention errors rather than uniformly distributing erases across blocks, as maintaining accurate data is the true goal for enhancing reliability. We propose the concept of retention leveling, considering $a$ crucial factor, i.e., temperature, and compare it to traditional wear leveling. Experimental results demonstrate that retention leveling can significantly improve flash memory reliability by limiting the number of retention errors to merely 0.0019% of the conventional approach, while substantially alleviating the write amplification problem.