{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,1,9]],"date-time":"2026-01-09T17:12:28Z","timestamp":1767978748561,"version":"3.49.0"},"reference-count":55,"publisher":"Association for Computing Machinery (ACM)","issue":"1","funder":[{"name":"National Science and Technology Council, Taiwan","award":["NSTC 113-2221-E-A49-188-MY3"],"award-info":[{"award-number":["NSTC 113-2221-E-A49-188-MY3"]}]}],"content-domain":{"domain":["dl.acm.org"],"crossmark-restriction":true},"short-container-title":["ACM Trans. Storage"],"published-print":{"date-parts":[[2026,2,28]]},"abstract":"<jats:p>Computing near the source of data has been proven effective in terms of energy conservation, latency improvement, and privacy preservation. With this, edge intelligence refers to local CNN inference in embedded edge devices. Because edge devices are highly resource-constrained, in practice they store CNN model(s) in external flash memory and load them (or part of) during runtime inference. However, flash memory is subject to time-related retention errors, and thus a dilemma is that without error correction, a CNN model in flash memory quickly deteriorates and becomes useless. On the other hand, strong error correction involves extra read sensing and read retrying. Due to the concern of reliability, the lineup of embedded serial flash memory offers low-density solutions only. In this study, we investigate graceful degradation of CNN models in uncorrected high-density flash storage for edge intelligence. Inspired by the observation that in popular CNN models, the majority of CNN weight parameters share a few monotonic bit patterns in their high-order bits, we propose mapping the frequent bit patterns to low flash-cell voltage levels for protection against retention errors. Furthermore, as CNN layers show different levels of tolerance to bit errors, we also propose using adaptive cell-bit density to provide non-uniform error protection among layers. We conducted experiments on popular CNN designs, including VGG16, ResNet50, and InceptionV3, under realistic effects of flash aging. Results show that with prior methods, the inference accuracy of ResNet50 degrades to 70.7% in just three months; by contrast, with our approach, the degradation is less than 1% and the accuracy remains at 91.6% after one year of retention.<\/jats:p>","DOI":"10.1145\/3747298","type":"journal-article","created":{"date-parts":[[2025,7,7]],"date-time":"2025-07-07T07:17:21Z","timestamp":1751872641000},"page":"1-25","update-policy":"https:\/\/doi.org\/10.1145\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Graceful CNN Model Degradation in Uncorrected Flash Storage for Embedded Edge Devices"],"prefix":"10.1145","volume":"22","author":[{"ORCID":"https:\/\/orcid.org\/0009-0008-4191-469X","authenticated-orcid":false,"given":"Hung-Yi","family":"Chen","sequence":"first","affiliation":[{"name":"Computer Science, National Yang Ming Chiao Tung University","place":["Hsinchu, Taiwan"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-5727-7535","authenticated-orcid":false,"given":"Jin-Wei","family":"Chang","sequence":"additional","affiliation":[{"name":"Computer Science, National Yang Ming Chiao Tung University","place":["Hsinchu, Taiwan"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0009-0002-2227-2966","authenticated-orcid":false,"given":"Hong-Ruei","family":"Lin","sequence":"additional","affiliation":[{"name":"Computer Science, National Yang Ming Chiao Tung University","place":["Hsinchu, Taiwan"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6543-2064","authenticated-orcid":false,"given":"Li-Pin","family":"Chang","sequence":"additional","affiliation":[{"name":"Computer Science, National Yang Ming Chiao Tung University","place":["Hsinchu, Taiwan"]}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"320","published-online":{"date-parts":[[2026,1,9]]},"reference":[{"key":"e_1_3_2_2_2","doi-asserted-by":"publisher","DOI":"10.1145\/1031495.1031516"},{"key":"e_1_3_2_3_2","volume-title":"Proceedings of the 4th USENIX Conference on File and Storage Technologies (FAST 05)","author":"Zeinalipour-Yazti Demtrios","year":"2005","unstructured":"Demtrios Zeinalipour-Yazti, Song Lin, Vana Kalogeraki, Dimitrios Gunopulos, and Walid A. Najjar. 2005. MicroHash: An efficient index structure for flash-based sensor devices. In Proceedings of the 4th USENIX Conference on File and Storage Technologies (FAST 05). USENIX Association, San Francisco, CA. Retrieved from https:\/\/www.usenix.org\/conference\/fast-05\/microhash-efficient-index-structure-flash-based-sensor-devices"},{"key":"e_1_3_2_4_2","doi-asserted-by":"publisher","DOI":"10.1145\/2968478.2968486"},{"key":"e_1_3_2_5_2","doi-asserted-by":"publisher","DOI":"10.5555\/1602165.1602197"},{"key":"e_1_3_2_6_2","doi-asserted-by":"publisher","DOI":"10.1109\/JIOT.2016.2579198"},{"key":"e_1_3_2_7_2","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2017.2767561"},{"key":"e_1_3_2_8_2","article-title":"Convolutional neural network architectures for matching natural language sentences","volume":"27","author":"Hu Baotian","year":"2014","unstructured":"Baotian Hu, Zhengdong Lu, Hang Li, and Qingcai Chen. 2014. Convolutional neural network architectures for matching natural language sentences. Advances in Neural Information Processing Systems 27 (2014), 2042\u20132050.","journal-title":"Advances in Neural Information Processing Systems"},{"key":"e_1_3_2_9_2","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2013.312"},{"key":"e_1_3_2_10_2","unstructured":"Karen Simonyan and Andrew Zisserman. 2015. Very deep convolutional networks for large-scale image recognition. Retrieved from https:\/\/arxiv.org\/abs\/1409.1556"},{"key":"e_1_3_2_11_2","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2015.7298594"},{"key":"e_1_3_2_12_2","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"e_1_3_2_13_2","article-title":"Keras Applications","author":"Chollet Fran\u00e7ois","year":"2015","unstructured":"Fran\u00e7ois Chollet et\u00a0al. 2015. Keras Applications. Retrieved September 2024 from https:\/\/keras.io\/api\/applications\/. (2015).","journal-title":"https:\/\/keras.io\/api\/applications\/"},{"key":"e_1_3_2_14_2","article-title":"Sipeed Maix-II-Dock","year":"2021","unstructured":"Sipeed. 2021. Sipeed Maix-II-Dock. Retrieved September 2024 from https:\/\/wiki.sipeed.com\/hardware\/en\/maixII\/M2\/resources.html. (2021).","journal-title":"https:\/\/wiki.sipeed.com\/hardware\/en\/maixII\/M2\/resources.html"},{"key":"e_1_3_2_15_2","article-title":"Coral Dev Board Micro","author":"Coral Google","year":"2022","unstructured":"Google Coral. 2022. Coral Dev Board Micro. Retrieved September 2024 from https:\/\/coral.ai\/static\/files\/Coral-Dev-Board-Micro-datasheet.pdf. (2022).","journal-title":"https:\/\/coral.ai\/static\/files\/Coral-Dev-Board-Micro-datasheet.pdf"},{"key":"e_1_3_2_16_2","article-title":"SparkFun Edge Development Board - Apollo3 Blue","year":"2021","unstructured":"SparkFun. 2021. SparkFun Edge Development Board - Apollo3 Blue. Retrieved September 2024 from https:\/\/cdn.sparkfun.com\/assets\/c\/2\/6\/2\/9\/Apollo3-Blue-MCU-Datasheet.pdf. (2021).","journal-title":"https:\/\/cdn.sparkfun.com\/assets\/c\/2\/6\/2\/9\/Apollo3-Blue-MCU-Datasheet.pdf"},{"key":"e_1_3_2_17_2","first-page":"187","volume-title":"Proceedings of the 2021 IEEE 32nd International Conference on Application-specific Systems, Architectures and Processors (ASAP)","author":"Banerjee Jishnu","year":"2021","unstructured":"Jishnu Banerjee, Sahidul Islam, Wei Wei, Chen Pan, Dakai Zhu, and Mimi Xie. 2021. Memory-aware efficient deep learning mechanism for IoT devices. In Proceedings of the 2021 IEEE 32nd International Conference on Application-specific Systems, Architectures and Processors (ASAP). IEEE, 187\u2013194."},{"key":"e_1_3_2_18_2","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2022.3203719"},{"key":"e_1_3_2_19_2","doi-asserted-by":"publisher","DOI":"10.1109\/TII.2023.3254668"},{"key":"e_1_3_2_20_2","unstructured":"Hongyu Miao and Felix Xiaozhu Lin. 2021. Enabling Large Neural Networks on Tiny Microcontrollers with Swapping. Retrieved from https:\/\/arxiv.org\/abs\/2101.0874"},{"key":"e_1_3_2_21_2","doi-asserted-by":"publisher","DOI":"10.1145\/3081333.3081359"},{"key":"e_1_3_2_22_2","first-page":"1","volume-title":"Proceedings of the 2021 IEEE International Conference on Pervasive Computing and Communications (PerCom)","author":"Cox Bart","year":"2021","unstructured":"Bart Cox, Jeroen Galjaard, Amirmasoud Ghiassi, Robert Birke, and Lydia Y. Chen. 2021. Masa: Responsive multi-DNN inference on the edge. In Proceedings of the 2021 IEEE International Conference on Pervasive Computing and Communications (PerCom). IEEE, 1\u201310."},{"key":"e_1_3_2_23_2","first-page":"381","volume-title":"Proceedings of the 2018 IEEE 36th International Conference on Computer Design (ICCD)","author":"Wu Fei","year":"2018","unstructured":"Fei Wu, Yue Zhu, Qin Xiong, Zhonghai Lu, You Zhou, Weizhen Kong, and Changsheng Xie. 2018. Characterizing 3D charge trap NAND flash: Observations, analyses and applications. In Proceedings of the 2018 IEEE 36th International Conference on Computer Design (ICCD). IEEE, 381\u2013388."},{"key":"e_1_3_2_24_2","doi-asserted-by":"publisher","DOI":"10.1145\/3445814.3446719"},{"key":"e_1_3_2_25_2","first-page":"243","volume-title":"Presented as part of the 11th USENIX Conference on File and Storage Technologies (FAST 13)","author":"Zhao Kai","year":"2013","unstructured":"Kai Zhao, Wenzhe Zhao, Hongbin Sun, Xiaodong Zhang, Nanning Zheng, and Tong Zhang. 2013. LDPC-in-SSD: Making advanced error correction codes work effectively in solid state drives. In Presented as part of the 11th USENIX Conference on File and Storage Technologies (FAST 13). 243\u2013256."},{"key":"e_1_3_2_26_2","doi-asserted-by":"publisher","DOI":"10.1109\/TCAD.2020.3012646"},{"issue":"1","key":"e_1_3_2_27_2","first-page":"1","article-title":"Low-energy error correction of NAND Flash memory through soft-decision decoding","volume":"2012","author":"Kim Jonghong","year":"2012","unstructured":"Jonghong Kim and Wonyong Sung. 2012. Low-energy error correction of NAND Flash memory through soft-decision decoding. EURASIP Journal on Advances in Signal Processing 2012, 1 (2012), 1\u201312.","journal-title":"EURASIP Journal on Advances in Signal Processing"},{"key":"e_1_3_2_28_2","article-title":"2021 NAND Flash Updates from ISSCC: The Leaning Towers of TLC and QLC","author":"Tallis Billy","year":"2021","unstructured":"Billy Tallis. 2021. 2021 NAND Flash Updates from ISSCC: The Leaning Towers of TLC and QLC. Retrieved September 2024 from https:\/\/www.anandtech.com\/show\/16491\/flash-memory-at-isscc-2021. (2021).","journal-title":"https:\/\/www.anandtech.com\/show\/16491\/flash-memory-at-isscc-2021"},{"key":"e_1_3_2_29_2","unstructured":"Toshiba. 2019. Toshiba BiCS3 NAND (TH58TFG9T23BA4C) datasheet. (2019)."},{"key":"e_1_3_2_30_2","doi-asserted-by":"publisher","DOI":"10.1109\/JSSC.2019.2900866"},{"key":"e_1_3_2_31_2","doi-asserted-by":"publisher","DOI":"10.1016\/j.sysarc.2022.102400"},{"key":"e_1_3_2_32_2","doi-asserted-by":"publisher","DOI":"10.5555\/1960475.1960493"},{"key":"e_1_3_2_33_2","first-page":"1","volume-title":"Proceedings of the IEEE International Symposium on High-Performance Comp Architecture","author":"Pan Yangyang","year":"2012","unstructured":"Yangyang Pan, Guiqiang Dong, Qi Wu, and Tong Zhang. 2012. Quasi-nonvolatile SSD: Trading flash memory nonvolatility to improve storage system performance for enterprise applications. In Proceedings of the IEEE International Symposium on High-Performance Comp Architecture. IEEE, 1\u201310."},{"key":"e_1_3_2_34_2","doi-asserted-by":"publisher","DOI":"10.5555\/2208461.2208487"},{"key":"e_1_3_2_35_2","volume-title":"Proceedings of the 2nd Workshop on Hot Topics in Storage and File Systems (HotStorage 10)","author":"Mohan Vidyabhushan","year":"2010","unstructured":"Vidyabhushan Mohan, Taniya Siddiqua, Sudhanva Gurumurthi, and Mircea R. Stan. 2010. How I learned to stop worrying and love flash endurance. In Proceedings of the 2nd Workshop on Hot Topics in Storage and File Systems (HotStorage 10). USENIX Association, Boston, MA. Retrieved from https:\/\/www.usenix.org\/conference\/hotstorage-10\/how-i-learned-stop-worrying-and-love-flash-endurance"},{"key":"e_1_3_2_36_2","first-page":"113","volume-title":"Proceedings of the 2022 IEEE 28th International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA)","author":"Hsu Hsiang-Sen","year":"2022","unstructured":"Hsiang-Sen Hsu and Li-Pin Chang. 2022. Exploiting binary equilibrium for efficient LDPC decoding in 3D NAND flash. In Proceedings of the 2022 IEEE 28th International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA). IEEE, 113\u2013119."},{"key":"e_1_3_2_37_2","first-page":"1","volume-title":"Proceedings of the 2018 IEEE International Symposium on Circuits and Systems (ISCAS)","author":"Mizushina Keita","year":"2018","unstructured":"Keita Mizushina, Toshiki Nakamura, Yoshiaki Deguchi, and Ken Takeuchi. 2018. Layer-by-layer adaptively optimized ECC of NAND flash-based SSD storing convolutional neural network weight for scene recognition. In Proceedings of the 2018 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 1\u20135."},{"key":"e_1_3_2_38_2","doi-asserted-by":"publisher","DOI":"10.1145\/3195970.3195997"},{"key":"e_1_3_2_39_2","first-page":"601","volume-title":"Proceedings of the 2019 42nd International Conference on Telecommunications and Signal Processing (TSP)","author":"Queralta J. Pena","year":"2019","unstructured":"J. Pena Queralta, Tuan Nguyen Gia, Hannu Tenhunen, and Tomi Westerlund. 2019. Edge-AI in LoRa-based health monitoring: Fall detection system with fog computing and LSTM recurrent neural networks. In Proceedings of the 2019 42nd International Conference on Telecommunications and Signal Processing (TSP). IEEE, 601\u2013604."},{"key":"e_1_3_2_40_2","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2021.3056082"},{"key":"e_1_3_2_41_2","doi-asserted-by":"publisher","DOI":"10.1145\/3479599"},{"key":"e_1_3_2_42_2","doi-asserted-by":"publisher","DOI":"10.1145\/3297858.3304011"},{"key":"e_1_3_2_43_2","first-page":"551","volume-title":"Proceedings of the 2015 IEEE 21st International Symposium on High Performance Computer Architecture (HPCA)","author":"Cai Yu","year":"2015","unstructured":"Yu Cai, Yixin Luo, Erich F. Haratsch, Ken Mai, and Onur Mutlu. 2015. Data retention in MLC NAND flash memory: Characterization, optimization, and recovery. In Proceedings of the 2015 IEEE 21st International Symposium on High Performance Computer Architecture (HPCA). IEEE, 551\u2013563."},{"key":"e_1_3_2_44_2","first-page":"312","volume-title":"Proceedings of the 2019 Design, Automation & Test in Europe Conference & Exhibition (DATE)","author":"Liu Weihua","year":"2019","unstructured":"Weihua Liu, Fei Wu, Meng Zhang, Yifei Wang, Zhonghai Lu, Xiangfeng Lu, and Changsheng Xie. 2019. Characterizing the reliability and threshold voltage shifting of 3D charge trap NAND flash. In Proceedings of the 2019 Design, Automation & Test in Europe Conference & Exhibition (DATE). IEEE, 312\u2013315."},{"key":"e_1_3_2_45_2","doi-asserted-by":"publisher","DOI":"10.1109\/TIT.1962.1057683"},{"key":"e_1_3_2_46_2","doi-asserted-by":"publisher","DOI":"10.1145\/3477016"},{"key":"e_1_3_2_47_2","first-page":"51","volume-title":"Proceedings of the 2018 IEEE 36th International Conference on Computer Design (ICCD)","author":"Wu Fei","year":"2018","unstructured":"Fei Wu, Zuo Lu, You Zhou, Xubin He, Zhihu Tan, and Changsheng Xie. 2018. OSPADA: One-shot programming aware data allocation policy to improve 3D NAND flash read performance. In Proceedings of the 2018 IEEE 36th International Conference on Computer Design (ICCD). IEEE, 51\u201358."},{"key":"e_1_3_2_48_2","first-page":"122","volume-title":"Proceedings of the 2023 IEEE International Symposium on High-Performance Computer Architecture (HPCA)","author":"Lv Yina","year":"2023","unstructured":"Yina Lv, Liang Shi, Qiao Li, Congming Gao, Yunpeng Song, Longfei Luo, and Youtao Zhang. 2023. MGC: Multiple-gray-code for 3D NAND flash based high-density SSDs. In Proceedings of the 2023 IEEE International Symposium on High-Performance Computer Architecture (HPCA). IEEE, 122\u2013136."},{"key":"e_1_3_2_49_2","unstructured":"Texas Instruments. 2020. MSP430FR58xx MSP430FR59xx and MSP430FR6xx Family User\u2019s Guide. (2020)."},{"key":"e_1_3_2_50_2","doi-asserted-by":"publisher","DOI":"10.1145\/3065386"},{"key":"e_1_3_2_51_2","first-page":"1","volume-title":"Proceedings of the 2018 IEEE High Performance Extreme Computing Conference (HPEC)","author":"Arechiga Austin P.","year":"2018","unstructured":"Austin P. Arechiga and Alan J. Michaels. 2018. The robustness of modern deep learning architectures against single event upset errors. In Proceedings of the 2018 IEEE High Performance Extreme Computing Conference (HPEC). IEEE, 1\u20136."},{"key":"e_1_3_2_52_2","unstructured":"Song Han Huizi Mao and William J. Dally. 2016. Deep compression: Compressing deep neural networks with pruning Trained Quantization and Huffman Coding. Retrieved from https:\/\/arxiv.org\/abs\/1510.00149"},{"key":"e_1_3_2_53_2","doi-asserted-by":"publisher","DOI":"10.1109\/RTCSA58653.2023.00018"},{"key":"e_1_3_2_54_2","unstructured":"2021. Pyldpc 0.7.9. Retrieved September 2024 from https:\/\/pypi.org\/project\/pyldpc\/. (2021)."},{"key":"e_1_3_2_55_2","volume-title":"https:\/\/www.micron.com\/products\/solid-state-drives","unstructured":"Inc. Micron Technology. 2017. NAND Flash Memory - FortisFlashTM Datasheet. Retrieved from https:\/\/www.micron.com\/products\/solid-state-drives. Accessed: 2024."},{"key":"e_1_3_2_56_2","first-page":"1","volume-title":"Proceedings of the International Conference on Massive Storage Systems and Technology (MSST)","author":"Du Yajuan","year":"2017","unstructured":"Yajuan Du, Deqing Zou, Qiao Li, Liang Shi, Hai Jin, and Chun Jason Xue. 2017. LaLDPC: Latency-aware LDPC for read performance improvement of solid state drives. In Proceedings of the International Conference on Massive Storage Systems and Technology (MSST). 1\u201311."}],"container-title":["ACM Transactions on Storage"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3747298","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,1,9]],"date-time":"2026-01-09T13:04:40Z","timestamp":1767963880000},"score":1,"resource":{"primary":{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3747298"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,1,9]]},"references-count":55,"journal-issue":{"issue":"1","published-print":{"date-parts":[[2026,2,28]]}},"alternative-id":["10.1145\/3747298"],"URL":"https:\/\/doi.org\/10.1145\/3747298","relation":{},"ISSN":["1553-3077","1553-3093"],"issn-type":[{"value":"1553-3077","type":"print"},{"value":"1553-3093","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,1,9]]},"assertion":[{"value":"2024-09-20","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2025-05-24","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-01-09","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}