CAREER: Information Storage in Flash Memories: From Devices to Networks

摘要：闪存已经成为目前使用最广泛的电子非易失性存储器类型。为了从根本上将闪存的性能推向极限，需要一种新的数据存储编码理论。闪存面临的主要挑战包括写入效率低、数据可靠性和有限的存储器寿命。这项研究研究新的信息论方法，以大幅提高闪存的写入速度、电源效率、可靠性和寿命。新的存储方案可用于多种移动设备、嵌入式系统和海量存储系统。本研究探索了闪存中数据存储的新编码理论。为了提高闪存的速度、功率效率和寿命，开发了一种被称为浮点码的新的码族，它们联合对数据进行编码并最小化块擦除操作。设计了纠错码和交织方案来纠正闪存中的特征错误，并将编码方案进一步扩展到连续单元级别的使用以实现最佳性能。针对闪存在设备级、系统级和网络级这三个极具代表性和重要的应用中的数据存储应用，研究了新的存储方案和算法。在这些应用程序中，代码被设计为支持基于页面的写入、数据聚合和网络内分布式存储。

Abstract:Flash memories have become by far the most widely used type of electronic non-volatile memories. To fundamentally push the performance of flash memories toward their limit, a new coding theory for data storage is needed. The main challenges facing flash memories include low efficiency in writing, data reliability and limited memory longevity. This research studies novel information theoretic approaches to substantially increase the writing speed, power efficiency, reliability and longevity of flash memories. The new storage schemes can be used by numerous mobile devices, embedded systems and mass storage systems.This research explores new coding theories for data storage in flash memories. A new family of codes called floating codes, which jointly encode data and minimize block erasure operations, are developed to improve the speed, power efficiency and longevity of flash memories. Error-correcting codes and interleaving schemes are designed to correct characteristic errors in flash memories.The coding schemes are further extended to the usage of continuous cell levels for optimal performance. New storage schemes and algorithms are studied for data storage applications in flash disks, databases and sensor networks, three highly representative and important applications of flash memories respectively at the device, system and network level. In those applications, codes are designed to support page-based writing, data aggregation and in-network distributed storage.