相变内存耐久性研究

项目名称： 相变内存耐久性研究

项目编号： No.61303002

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 王海霞

作者单位： 清华大学

项目金额： 23万元

中文摘要： 内存系统已成为计算机系统发展的瓶颈，研制高速、大容量、低功耗的新型存储系统是计算机系统设计的一个重要挑战。相变内存的出现为该问题提供了一种解决方案，但是相变内存写次数受限的问题限制了它的使用寿命，严重阻碍了它的实用化。本项目以提高相变内存使用寿命为目标，首先拟建立相变内存使用寿命的理论模型，其次基于理论模型分析抗磨损机理，最后根据抗磨损机理从多核处理器体系结构设计的角度进行优化，提高相变内存使用寿命。本项目拟提出延迟替换频繁写数据的Cache替换策略、基于目的节点脏数据量评估的复制和迁移技术，通过优化多核处理器架构减少对相变内存的写操作；拟提出一种基于数据预取技术的恶意攻击检测机制，利用数据预取技术中重复Cache缺失预测机制来检测恶意攻击程序，避免相变内存受到频繁写攻击。本项目研究对于提高基于相变内存的计算机系统使用寿命和性能具有重要的理论和实践意义。

中文关键词： 相变内存；耐久性；Cache替换策略；伪随机编码；流预测器

英文摘要： Memory system has been the bottleneck of computer systems. Designing a high-speed, large-capacity, low-power consumption memory system is an important challenge for computer system design. Phase change memory appears to provide a solution, but the limited write endurance restricts its development. To improve the endurance of phase change memory, this project intends to establish a theoretical model for phase change memory endurance, analyze endurance improving techniques, and then explore suitable multicore processor architectures. A novel cache replacement policy and a cache replication and migration policy are proposed to reduce writes to phase change memory. The novel cache replacement policy keeps frequently modified datum in cache as long as possible and the cache replication and migration policy makes decision based on the amount of dirty datum on the destination node. Furthermore, a malicious wear-out detection mechanism for phase change memory is proposed, which reuses the repeated cache missing streams detection of data prefetching. In summary, the research has important theoretical and practical significance for improving the endurance and performance of computer systems with phase-change memory.

英文关键词： phase change memory；endurance；cache replacement policy；pseudo-random encode；stream predictor