Evolving a storage block endurance classifier for Flash memory: A trial implementation

Abstract

Solid State Drives (SSDs) have a number of significant advantages over traditional Hard Disk Drives (HDDs) but are currently far more expensive and have smaller capacities. These drives are based on NAND Flash memory devices, which have limited working lives. The number of times locations in such devices can be successfully programmed before they become unreliable is termed their endurance.There is currently no way to estimate accurately when a location within a Flash device will fail, so manufacturers give extremely conservative guarantees about the number of program operations their chips can endure. This paper describes a trial implementation of Genetic Programming (GP) used to evolve a Binary Classifier that predicts whether storage blocks within Flash memory devices will still be functioning correctly beyond some predefined number of cycles.The classifier is supplied with only the measured program and erase times from a relatively early point in the lifetime of a block. Using the relationships between these times, the system can accurately predict whether the block will continue to function satisfactorily up to a required number of cycles. Experiments on test sets comprised of unseen data show that our classifier obtains up to an average of 95% accuracy across 30 runs.

Estimating MLC NAND flash endurance

“…Since it has been shown by Hogan et al [12] that it is possible to classify blocks as being of a particular performance group, this type of partitioning of the data can also be performed. The pilot investigation into this approach focuses only on the initial training of the symbolic regression expressions, and did not attempt to train or test the system using the required classifiers.…”

“…Sullivan and Ryan [26] reported on their use of Genetic Algorithms to extend the endurance of NOR Flash memory by optimising some of the Flash device's internal control parameters. Hogan et al constructed a successful technique to perform binary classification on NAND Flash blocks according to their endurance [12,13], that is, predicting whether or not a block would survive past a particular number of cycles. This group also examined the potential use of retention classifiers [14].…”

“…The block level division of data approach is based on the use of both classifiers (such as those proposed by Hogan et al [12,13]) and symbolic regression. Firstly a collection of binary classifiers are required to group the data points, or test blocks, by endurance, using the timing information recorded early in the block's lifetime.…”

The Characterisation and Optimisation of TLC NAND Flash Memory using Machine Learning - A Position Paper

“…This study used GP, an extension of GAs (Koza, 1992), to evolve a mathematical function that, given the start-of-life values for read, write and erase times, would predict the useful life of the NAND Flash block. The model obtained up to 95% accuracy on unseen data, thereby proving that it is possible to use this implementation method to predict real endurance figures (Hogan et al, 2012a). A parallel study to predict retention limits of MLC chips was also carried out using GP (Hogan et al, 2012b).…”

“…Further research carried out during this time fo-cused on predicting end-of-life for a NAND Flash part by using start-of-life measurements, such as program and erase time (Hogan et al, 2012a). This study used GP, an extension of GAs (Koza, 1992), to evolve a mathematical function that, given the start-of-life values for read, write and erase times, would predict the useful life of the NAND Flash block.…”