Storage Reliability of Multi-bit Flash Oriented to Deep Neural Network

• Published in: 2019 IEEE International Electron Devices Meeting (IEDM) pp. 38.2.1 - 38.2.4
• Main Authors: Xiang, Y. C., Kang, J. F., Huang, P., Yang, H. Z., Wang, K. L., Han, R. Z., Shen, W. S., Feng, Y. L., Liu, C., Liu, X. Y.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2019
• ISSN: 2156-017X
• DOI: 10.1109/IEDM19573.2019.8993508

The storage reliability of multi-bit flash is of vital importance for the flash based deep neural network (DNN). In this work, the critical concerns correlated with the storage reliability (I d distribution and data retention) of multi-bit flash and its impacts on the DNN are investigated for the first time. The key achievements include: (1) A dynamic drain-voltage (V d ) programming method is proposed to achieve adequately tight (error rate <; 0.5%) and spaced drain-current (I d ) distributions in a reasonable time, which leads to comparable accuracy with the software (only 0.25% loss) for the CIFAR-10 recognition. (2) The statistical I d evolution of 16 states over time at different temperatures is studied in a 1Mb flash array, and a physical model is developed to characterize the retention of multi-bit flash. (3) Leveraging the physical model, the device and system co-design is proposed to enhance the reliability of the flash based DNN significantly.