基于卷积神经网络的闪存信道检测技术研究

PDF下载HTML

Title:

Research on flash memory channel detection technology based on convolutional neural network

作者:

胡海华, 韩国军, 张孝谊

广东工业大学 信息工程学院，广东 广州 510000

Author(s):

HU Haihua, HAN Guojun, ZHANG Xiaoyi

School of Information Engineering, Guangdong University of Technology, Guangzhou 510000, China

关键词:

NAND闪存; 信道检测; 卷积神经网络; 单元间干扰; 信道先验信息; 复杂度; 可靠性; 阈值检测

Keywords:

NAND flash memory; channel detection; convolutional neural network; cell to cell interference; channel prior information; complexity; reliability; threshold detection

分类号:

TN919.5

DOI:

10.11992/tis.202010029

摘要:

NAND闪存信道检测技术直接影响数据存储的可靠性，本文针对NAND闪存信道检测过程中因缺乏信道先验信息而导致检测性能显著降低的问题，提出了一种基于卷积神经网络(convolutional neural networks, CNN)的信道检测器。该检测器通过学习存储单元阈值电压随应用场景的变化特性，来初始化网络参数，并通过在系统空闲时间段优化网络参数来实现与信道的匹配。仿真实验结果表明：在信道先验信息未知的情况下，CNN检测器可获得比最优阈值检测器更好的检测性能；与现有的循环神经网络检测器相比，CNN检测器具有更低的复杂度，从而能获得更低的检测延时。

Abstract:

Channel detection technology for NAND flash memory directly affects the reliability of data storage. Lack of prior information of channel states can lead to significant reduction of detection performances. Aiming at addressing this issue, this paper proposes a channel detector based on convolutional neural networks (CNN). The detector initializes network parameters by learning the varying characteristics with application scenarios of the threshold voltages of the storage cells, and realizes channel matching through optimization of the network parameters during the idle time of the system. The simulation results show that when the channel prior information is unavailable, the CNN detector can achieve better detection performance than the optimal threshold detector. Compared with the existing cyclic neural network detector, CNN detector has lower complexity and therefore reduced detection delay.