Tolerance of Deep Neural Network Against the Bit Error Rate of NAND Flash Memory

Flash memory can store large number of model weights for a deep neural network on edge devices. In this paper, we study the effects of Flash's bit-error-rate on the neural network's weights and prediction accuracy, given that error correction engine is disabled due to power constraint in edge devices. The reliability experiments on MLC NAND Flash show that DNN prediction accuracy remains above the acceptable level in the presence of different NAND reliability issues for the following memory usage: less than 5 years of data retention, ~100 program-erase cycle, and less than 30k read stress.

[1]  Onur Mutlu,et al.  Error Characterization, Mitigation, and Recovery in Flash-Memory-Based Solid-State Drives , 2017, Proceedings of the IEEE.

[2]  Paul H. Siegel,et al.  Characterizing flash memory: Anomalies, observations, and applications , 2009, 2009 42nd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO).

[3]  Andrea L. Lacaita,et al.  Reviewing the Evolution of the NAND Flash Technology , 2017, Proceedings of the IEEE.

[4]  Biswajit Ray,et al.  Independent detection of recycled flash memory: Challenges and solutions , 2018, 2018 IEEE International Symposium on Hardware Oriented Security and Trust (HOST).

[5]  Onur Mutlu,et al.  Vulnerabilities in MLC NAND Flash Memory Programming: Experimental Analysis, Exploits, and Mitigation Techniques , 2017, 2017 IEEE International Symposium on High Performance Computer Architecture (HPCA).

[6]  Mark G. Karpovsky,et al.  Secure memories resistant to both random errors and fault injection attacks using nonlinear error correction codes , 2013, HASP '13.

[7]  Neal R. Mielke,et al.  Reliability of Solid-State Drives Based on NAND Flash Memory , 2017, Proceedings of the IEEE.

[8]  En Xia Zhang,et al.  State-of-the-Art Flash Chips for Dosimetry Applications , 2018, 2018 IEEE Nuclear & Space Radiation Effects Conference (NSREC 2018).

[9]  Angelo Visconti,et al.  Cycling-Induced Charge Trapping/Detrapping in Flash Memories—Part II: Modeling , 2016, IEEE Transactions on Electron Devices.

[10]  Biswajit Ray,et al.  Non-Invasive Detection Method for Recycled Flash Memory Using Timing Characteristics † , 2018, Cryptogr..

[11]  J. Kessenich,et al.  Bit error rate in NAND Flash memories , 2008, 2008 IEEE International Reliability Physics Symposium.

[12]  Minjae Lee,et al.  Fault tolerance analysis of digital feed-forward deep neural networks , 2014, 2014 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[13]  Ken Takeuchi,et al.  3D-NAND Flash Solid-State Drive (SSD) for Deep Neural Network Weight Storage of IoT Edge Devices with 700x Data-Retention Lifetime Extention , 2018, 2018 IEEE International Memory Workshop (IMW).

[14]  David Brooks,et al.  On-Chip Deep Neural Network Storage with Multi-Level eNVM , 2018, Design Automation Conference.