| Process improvement and technological innovation have brought higher storage density to NAND flash,while also making its data reliability worse.Particularly the uncertainty and difference in endurance not only reduce the lifetime of flash memory but also cause the devices to be replaced prematurely based on the nominal value far below the minimum actual value.By accurately predicting flash memory endurance and using the results to optimize wear-leveling strategies and implement bad block warnings,we can effectively extend the lifetime of devices and avoid severe losses due to sudden failures.To this end,a multi-class endurance prediction scheme based on the SVM algorithm is proposed,which can predict the remaining P-E cycle level and the raw bit error level after various P-E cycles.Feature analysis based on endurance data is used to determine the basic elements of the model.By analyzing the structural features and error mode,a variety of targeted optimization strategies are adopted,such as extracting the numerical features closely related to the endurance and reducing the noise interference of transient faults through short-term repeated operations.Besides,based on the ZYNQ-7030 chip,a high-parallel flash test platform supporting multiple protocols a feature preprocessing module is constructed,as well as a pipelined module of SVM decision model,which can complete a single prediction within 37 us.The multi-label and multi-class model can achieve accuracy of up to 95.8%.All its indicators are in the leading echelon and the increase in class types is conducive to expand the scope of model application.Besides,the RBF-kernel SVM model has the best prediction results.The optimization strategy can increase accuracy by 7% to 10%.The model can still achieve the same prediction effect on the additional test set.Finally,the bad block warning strategy based on this model can successfully realize early warning for 96.9% of the blocks.This solution can provide accurate endurance levels for the optimization of solid-state drive management strategies and the screening of flash memory devices. |
