Retention Model Aware Reading (ReMAR)
Retention Model Aware Reading predicts NAND read-reference voltages from retention age and program/erase count using a calibrated model.
ReMAR learns voltage settings from sampled blocks and stores programming timestamps. Background calibration still consumes reads; restart requires a valid clock or host time synchronization. [1]
HeatWatch additionally models temperature and dwell history. Elapsed retention age alone is not its thermally adjusted time. [2]
Linux exposes read-retry through raw-NAND driver callbacks. That interface is not a portable host control for an NVMe SSD. [3]
Relations
- related
- Retention Optimized Reading (ROR), Early retention loss, NAND read-reference voltage, NAND read-retry, HeatWatch
Limitations
- The 2018 characterization used single-vendor 3D MLC at 20 degrees Celsius; it does not establish present-day QLC behavior. [1]
- Reported RBER reduction concerns errors before ECC, not the probability of losing a file. [1]
- Temperature and prior dwell intervals affect the modeled cell state; a timestamp is not a complete reliability description. [2]
Citations
- Luo et al. Improving 3D NAND Flash Memory Lifetime by Tolerating Early Retention Loss and Process Variation. POMACS 2018, sections 4.1, 5.2 and 6.3.
- Luo et al. HeatWatch. HPCA 2018, sections 3-5. DOI: 10.1109/HPCA.2018.00050.
- Linux MTD NAND driver API: nand_setup_read_retry and read_retries.
Further reading
Sources & caveats
- Primary-source synthesis, not a hardware reproduction. Model fitting, chip characterization and application-level outcomes are distinct evidence.