STRAW: A Stress-Aware WL-Based Read Reclaim Technique for High-Density NAND Flash-Based SSDs

TL;DR

The paper addresses read disturbance in high-density 3D NAND by showing that disturbance impact is highly heterogeneous across wordlines, which makes traditional block-level read reclaim ($RC_{MAX}$) inefficient. It introduces STRAW, a WL-level read reclaim framework, comprising a new read-disturbance model that classifies WLs into groups and estimates per-WL tolerance using $ERC_{MAX}$ and disturbance rate $\alpha$, and a StrawFTL that uses Read-reclaim Parameter Table (RPT) and Resource-Efficient Counters (REC) with Space-Saving techniques to bound metadata. Empirical evaluation on real 3D TLC NAND data and MQSim-E demonstrates substantial reductions in RR-induced writes (up to ~92%) and notable tail-latency improvements (up to ~81%), with negligible storage overhead. The approach significantly improves SSD lifetime and performance in high-density NAND, enabling more reliable handling of read disturbance in modern storage systems.

Abstract

Although read disturbance has emerged as a major reliability concern, managing read disturbance in modern NAND flash memory has not been thoroughly investigated yet. From a device characterization study using real modern NAND flash memory, we observe that reading a page incurs heterogeneous reliability impacts on each WL, which makes the existing block-level read reclaim extremely inefficient. We propose a new WL-level read-reclaim technique, called STRAW, which keeps track of the accumulated read-disturbance effect on each WL and reclaims only heavily-disturbed WLs. By avoiding unnecessary read-reclaim operations, STRAW reduces read-reclaim-induced page writes by 83.6\% with negligible storage overhead.

Figures (9)

• Figure 1: Key factors for asymmetry in read disturbance across WLs.
• Figure 2: Heterogeneous disturbance impact under different read patterns.
• Figure 3: A comparison of the RR overhead between SSD$_{\text{real}}$ and SSD$_{\text{ideal}}$.
• Figure 4: Comparisons of the maximum read counts of each WL Group.
• Figure 5: Final Model of $RC_\text{MAX}$ and $\alpha$ under different PEC.