UC San Diego

NAND Flash memories have become a widely used non-volatile data storage technology and their application areas are expected to grow in the future with the advent of cloud computing, big data and the internet-of-things. This has led to aggressive scaling down of the NAND flash memory cell feature sizes and also increased adoption of flash memories with multiple cell levels to increase the data storage density. These factors have adversely affected the reliability of flash memories.

In this dissertation, our main goal is to perform detailed characterization of the errors that occur in multi-level cell (MLC) flash memories and develop novel mathematical channel models that better reflect the measured error characteristics than do current models. The channel models thus developed are applied to error correcting code (ECC) frame error rate (FER) performance estimation in MLC flash memories and to estimating the flash memory channel capacity as represented by the channel models. We also utilize the characterization of inter-cell interference (ICI) errors to evaluate the performance of constrained coding schemes that mitigate ICI and improve the reliability of flash memories.

In Chapter 5, which is self-contained, we propose and study modifications to adaptive linear programming decoding techniques applied to decoding polar codes. We also propose a reduced complexity representation of the polar code sparse factor graph, resulting in time complexity improvements in the adaptive LP decoder.