Reliability of Solid-State Drives Based on NAND Flash Memory

• Published in: Proceedings of the IEEE 2017-09, Vol.105 (9), p.1725-1750
• Main Authors: Mielke, Neal R., Frickey, Robert E., Kalastirsky, Ivan, Quan, Minyan, Ustinov, Dmitry, Vasudevan, Venkatesh J.
• Format: Article
• Language: English
• Subjects: Confidence; Data retention; Data storage; Design defects; Disk drives; Drives; endurance; Error analysis; Failure; Failure mechanisms; Failure rates; Firmware; Flash memories; Flash memory (computers); Integrated circuits; JEDEC; NAND flash memory; Qualifications; Radiation effects; reliability; Reliability analysis; Reliability engineering; Soft errors; Solid state devices; Solid-state circuits; solid-state drives (SSDs); Statistical tests
• ISSN: 0018-9219; 1558-2256
• DOI: 10.1109/JPROC.2017.2725738

This paper reviews the reliability of solid-state drives (SSDs) based on NAND Flash memory from the perspectives of failure mechanisms, design mitigations, qualification methods, and field failure rates. NAND reliability is dominated by gradual memory-cell degradation in late life and defects such as interconnect shorts earlier in life. Design mitigations exist for these mechanisms. Qualification methods standardized in JEDEC JESD218 are designed to evaluate the mechanisms and mitigations, over a full drive lifetime, in the laboratory. Full-lifetime qualification provides confidence in the long-term reliability of SSDs that cannot be achieved by the early-life qualifications performed on hard disk drives. If NAND mechanisms are sufficiently suppressed, field reliability will be dominated by non-NAND mechanisms such as firmware bugs, power-loss events, radiation-induced soft errors, and failures in non-NAND components. A wide range of design validation and qualification tests are necessary to evaluate these non-NAND mechanisms. Published field reliability statistics indicate that SSDs are more reliable on average than HDDs, but they are not immune to failure, and there is wide variation among models. The NAND and non-NAND mechanisms are illustrated through new case studies of SSD internal qualification and field reliability data.