NAND Flash 記憶體已成為現代儲存系統的主流媒介，但隨著製程持續微縮，以及儲存單元在操作與資料保留期間逐漸老化，其可靠度也持續下降。當錯誤更正碼，例如: BCH、LDPC，達到糾錯上限時，SSD 控制器普遍會啟動 read-retry 機制，以補償閾值電壓的漂移。然而，傳統的讀取重試表在執行讀取重試時，僅依序套用所有讀取電壓參數，未考量區塊層級的差異性或參數間的內在相似性，因而導致過多的讀取重試次數，並在 3D TLC NAND 中造成顯著的讀取延遲。

本研究提出了一種自適應分組讀取重試策略 (AGRR)，一種結合參數相似度、LDPC 回饋與區塊行為的自適應群組式讀取重試機制，以提升讀取重試表的使用效率。我們提出群組相似性分數作為參數相似度指標，用以將讀取電壓參數分組並為每組選取代表參數。針對極端錯誤情況，AGRR 透過 LDPC syndrome 值選擇最適合的參數群組，以處理代表參數解碼失敗的情形。此外，基於歷史資訊的動態起始點機制可為每個區塊選定最佳起始參數，有效減少不必要的讀取重試次數。

在真實 3D TLC NAND 裝置的評估中，AGRR 可將讀取延遲最高降低 79%，且解碼成功率的下降低於 3%，在效率與可靠度間取得良好平衡，並僅需極低的區塊層級額外儲存成本。實驗結果顯示，AGRR 能有效提升現代 NAND Flash 系統的讀取效能與實用性。

NAND Flash memory has become the mainstream medium for modern storage systems, yet its reliability continues to degrade as technology scales and cells experience wear and retention loss. Read-retry mechanisms are widely adopted in SSD controllers to compensate for threshold-voltage shifts when error correction codes (ECC), such as LDPC, reach their correction limits. However, conventional read-retry table (RRT) applies read-voltage parameters sequentially without accounting for block-level variations or the intrinsic similarities among parameters, resulting in excessive retries and substantial read latency in 3D TLC NAND flash memory.

This work presents Adaptive Grouping-based Read Retry (AGRR), an adaptive group-based read-retry strategy that optimizes RRT usage by incorporating parameter similarity, LDPC feedback, and block-level behavior. We introduce the Grouping Similarity Score (GSS) to cluster read-voltage parameters and select a representative parameter (RP) for each group. To handle extreme error conditions, AGRR leverages LDPC syndrome values to guide group selection when multiple RPs fail. Additionally, a history-based dynamic start-point mechanism assigns an optimized initial retry point to each block, reducing unnecessary retry attempts.

Evaluated on 3D TLC NAND devices, AGRR reduces read latency by up to 79% while incurring less than a 3% loss in decoding success rate, achieving an effective balance between efficiency and reliability with minimal per-block state overhead. These results demonstrate that AGRR provides a practical and scalable solution for enhancing read performance in modern NAND Flash systems.

[1] Daniel Nicolas Bailon, Sergo Shavgulidze, and Jürgen Freudenberger. Read reference voltage adaptation for nand flash memories with neural networks based on sparse his- tograms. IEEE Access, 11:56801–56811, 2023.
[2] Yingcheng Bu, Yi Fang, Guohua Zhang, and Jun Cheng. Achievable-rate-aware retention-error correction for multi-level-cell nand flash memory. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 41(10):3438–3451, 2021.
[3] Yu Cai, Saugata Ghose, Erich F Haratsch, Yixin Luo, and Onur Mutlu. Error characteri- zation, mitigation, and recovery in flash-memory-based solid-state drives. Proceedings of the IEEE, 105(9):1666–1704, 2017.
[4] Yu Cai, Saugata Ghose, Yixin Luo, Ken Mai, Onur Mutlu, and Erich F Haratsch. Vul- nerabilities in mlc nand flash memory programming: Experimental analysis, exploits, and mitigation techniques. In 2017 IEEE International Symposium on High Perfor- mance Computer Architecture (HPCA), pages 49–60. IEEE, 2017.
[5] Yu Cai, Erich F Haratsch, Onur Mutlu, and Ken Mai. Error patterns in mlc nand flash memory: Measurement, characterization, and analysis. In 2012 Design, Automation & Test in Europe Conference & Exhibition (DATE), pages 521–526. IEEE, 2012.
[6] Yu Cai, Erich F Haratsch, Onur Mutlu, and Ken Mai. Threshold voltage distribution in mlc nand flash memory: Characterization, analysis, and modeling. In 2013 Design, Au- tomation & Test in Europe Conference & Exhibition (DATE), pages 1285–1290. IEEE, 2013.
[7] Yu Cai, Yixin Luo, Erich F Haratsch, Ken Mai, Saugata Ghose, and Onur Mutlu. Ex- perimental characterization, optimization, and recovery of data retention errors in mlc nand flash memory. arXiv preprint arXiv:1805.02819, 2018.
[8] Yu Cai, Yixin Luo, Erich F Haratsch, Ken Mai, and Onur Mutlu. Data retention in mlc nand flash memory: Characterization, optimization, and recovery. In 2015 IEEE 21st International Symposium on High Performance Computer Architecture (HPCA), pages 551–563. IEEE, 2015.
[9] Hyemin Choe, Jeongju Jee, Seung-Chan Lim, Sung Min Joe, Il Han Park, and Hyun- cheol Park. Machine-learning-based read reference voltage estimation for nand flash memory systems without knowledge of retention time. IEEE access, 8:176416–176429, 2020.
[10] Hua Feng, Debao Wei, Yongchao Wang, Yu Song, Zhelong Piao, and Liyan Qiao. Lightweight read reference voltage calibration strategy for improving 3-d tlc nand flash memory reliability. IEEE Transactions on Device and Materials Reliability, 23(3):370– 379, 2023.
[11] Akira Goda. 3-d nand technology achievements and future scaling perspectives. IEEE Transactions on Electron Devices, 67(4):1373–1381, 2020.
[12] Jing He, Xiaolei Yu, Qianhui Li, Bo Zhang, Xianliang Wang, Qianqi Zhao, Xuhong Qiang, Qi Wang, Zongliang Huo, and Tianchun Ye. Dnn-based error level predic- tion for reducing read latency in 3d nand flash memory. Microelectronics Reliability, 147:115008, 2023.
[13] Ravi Iyer, Li Zhao, Fei Guo, Ramesh Illikkal, Srihari Makineni, Don Newell, Yan Soli- hin, Lisa Hsu, and Steve Reinhardt. Qos policies and architecture for cache/memory in cmp platforms. ACM SIGMETRICS Performance Evaluation Review, 35(1):25–36, 2007.
[14] Hyundong Jang, Chanyang Park, Kihoon Nam, Hyeok Yun, Kyeongrae Cho, Jun- Sik Yoon, Hyun-Chul Choi, Ho-Jung Kang, Min Sang Park, Jaesung Sim, et al. Bi- directional long short-term memory neural network modeling of data retention charac- terization in 3-d triple-level cell nand flash memory. IEEE Transactions on Electron Devices, 69(8):4241–4247, 2022.
[15] Bvunarvul Kim et al. 28.2 a high-performance 1tb 3b/cell 3d-nand flash with a 194mb/s write throughput on over 300 layers. In 2023 IEEE International Solid-State Circuits Conference (ISSCC), pages 27–29. IEEE, 2023.
[16] Qiao Li, Min Ye, Yufei Cui, Liang Shi, Xiaoqiang Li, Tei-Wei Kuo, and Chun Jason Xue. Shaving retries with sentinels for fast read over high-density 3d flash. In 2020 53rd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO), pages 483–495. IEEE, 2020.
[17] Yingge Li, Guojun Han, Sanwei Huang, Chang Liu, Meng Zhang, and Fei Wu. Ex- ploiting metadata to estimate read reference voltage for 3-d nand flash memory. IEEE Transactions on Consumer Electronics, 69(1):9–17, 2022.
[18] Han-Yu Liao, Wen-Ling Hsu, Jen-Wei Hsieh, and Hung-Pin Chen. Read retry mech- anism for 3d nand flash memory: Observations, analyses, and solutions. In 2024 13th Non-Volatile Memory Systems and Applications Symposium (NVMSA), pages 1–6. IEEE, 2024.
[19] Chun-Yi Liu, Yunju Lee, Myoungsoo Jung, Mahmut Taylan Kandemir, and Wonil Choi. Prolonging 3d nand ssd lifetime via read latency relaxation. In Proceedings of the 26th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, pages 730–742, 2021.
[20] Duo Liu, Kan Zhong, Tianzheng Wang, Yi Wang, Zili Shao, Edwin Hsing-Mean Sha, and Jingling Xue. Durable address translation in pcm-based flash storage systems. IEEE Transactions on Parallel and Distributed Systems, 28(2):475–490, 2016.
[21] Yixin Luo, Saugata Ghose, Yu Cai, Erich F Haratsch, and Onur Mutlu. Heatwatch: Improving 3d nand flash memory device reliability by exploiting self-recovery and temperature awareness. In 2018 IEEE International Symposium on High Performance Computer Architecture (HPCA), pages 504–517. IEEE, 2018.
[22] Zhen Mei, Kui Cai, and Xuan He. Deep learning-aided dynamic read thresholds design for multi-level-cell flash memories. IEEE Transactions on Communications, 68(5):2850–2862, 2020.
[23] Nikolaos Papandreou, Nikolas Ioannou, Thomas Parnell, Roman Pletka, Milos Stanisavljevic, Radu Stoica, Sasa Tomic, and Haralampos Pozidis. Reliability of 3d nand flash memory with a focus on read voltage calibration from a system aspect. In 2019 19th Non-Volatile Memory Technology Symposium (NVMTS), pages 1–4. IEEE, 2019.
[24] Nikolaos Papandreou, Haralampos Pozidis, Nikolas Ioannou, Thomas Parnell, Roman Pletka, Milos Stanisavljevic, Radu Stoica, Sasa Tomic, Patrick Breen, Gary Tressler, et al. Open block characterization and read voltage calibration of 3d qlc nand flash. In 2020 IEEE International Reliability Physics Symposium (IRPS), pages 1–6. IEEE, 2020.
[25] Zhifang Shi, Yi Fang, Yingcheng Bu, and Guojun Han. Convolutional neural network (cnn)-based detection for multi-level-cell nand flash memory. IEEE Communications Letters, 25(12):3883–3887, 2021.
[26] Johann-Philipp Thiers, Daniel Nicolas Bailon, Jürgen Freudenberger, and Jianjie Lu. Read reference calibration and tracking for non-volatile flash memories. Electronics, 10(18):2306, 2021.
[27] Debao Wei, Liyan Qiao, Mengqi Hao, Hua Feng, and Xiyuan Peng. Reliability predic- tion model of nand flash memory based on random forest algorithm. Microelectronics Reliability, 100:113371, 2019.
[28] Min Ye, Qiao Li, Yina Lv, Jie Zhang, Tianyu Ren, Daniel Wen, Tei-Wei Kuo, and Chun Jason Xue. Achieving near-zero read retry for 3d nand flash memory. In Pro- ceedings of the 29th ACM International Conference on Architectural Support for Pro- gramming Languages and Operating Systems, Volume 2, pages 55–70, 2024. [29] Min Ye, Qiao Li, Jianqiang Nie, Tei-Wei Kuo, and Chun Jason Xue. Valid window: a new metric to measure the reliability of nand flash memory. In 2020 Design, Automation & Test in Europe Conference & Exhibition (DATE), pages 109–114. IEEE, 2020.
[30] Min Ye, Qiao Li, Daniel Wen, Tei-Wei Kuo, and Chun Jason Xue. wearmeter: an accurate wear metric for nand flash memory. In 2024 29th Asia and South Pacific Design Automation Conference (ASP-DAC), pages 442–447. IEEE, 2024.
[31] Meng Zhang, Fei Wu, Qin Yu, Weihua Liu, Yifan Wang, and Changsheng Xie. Ex- ploiting error characteristic to optimize read voltage for 3-d nand flash memory. IEEE Transactions on Electron Devices, 67(12):5490–5496, 2020.
[32] Kai Zhao, Wenzhe Zhao, Hongbin Sun, Xiaodong Zhang, Nanning Zheng, and Tong Zhang. {LDPC-in-SSD}: Making advanced error correction codes work effectively in solid state drives. In 11th USENIX Conference on File and Storage Technologies (FAST 13), pages 243–256, 2013.