UCLA

The on-going data revolution demands storage systems that can store very large quantities of data while being fast, reliable and cheap. Emerging storage technologies such as flash and granular media offer improved densities, faster access times, and are more power-efficient than conventional hard disk drives. The primary drawback associated with these new devices is their high error rate, caused by difficulties in programming, voltage drift, and wear-out. Coding methods used in existing storage applications are based on symmetric, Hamming-type metrics. However, when used in new memory devices, these traditional approaches result in costly overprovisioning. In this work, we present advanced coding-theoretic techniques applicable to modern storage devices that exploit the asymmetries in the underlying physical operations for improved performance. In many cases of interest, the results in this thesis represent the state of the art. Taken collectively, our results can help enable all modern, data-intensive technologies that require reliably storing large quantities of data.