부산대학교 도서관

A storage medium is called nonvolatile if it retains its information content even when it is not powered on. The widespread applicability of devices with good data retention properties have made nonvolatile memory an area of active academic and industry research. A significant problem encountered in this field is that for many nonvolatile memories the process of writing to the medium is inherently uncertain. This phenomenon fundamentally limits the amount of information that can be stored on the medium thus reducing its storage capacity. But often, such storage media are rewritable, that is, previously written content can be read and rewritten if necessary. This creates a powerful feedback path which allows for an improvement in its storage capacity. In this lecture we describe a novel information theoretic model for rewritable media that has been recently proposed by the authors [6]. For a simple yet important special case we show that rewriting can increase the capacity logarithmically in the average number of rewrites. We emphasize the role that feedback can play in increasing storage capacity, and briefly indicate how methods from estimation theory and dynamic programming could be used to design feedback strategies that can approach this improved storage capacity.