A NAND Flash Endurance Prediction Scheme with FPGA-based Memory Controller System

Abstract

The endurance of NAND flash memory continues to decrease with process scaling, leading to a decline in the reliability of the storage system and a rise on risk of data corruption. To enhance the reliability of the storage system, we utilize a neural network model with high accuracy and full application, to predict how far each block of a NAND flash can be cycled before the uncorrectable data errors occur. The input to the model includes program-time, erase-time and raw bit error (RBE) measured by FPGA (Field-Programmable Gate Array) and its output is a specific numerical value of endurance. Based on this prediction model, we propose a FPGA-based scheme for real-time endurance prediction with an accuracy of over 90%.