快速降低系统功耗对电池供电的智能终端设备尤为重要。存储器是这些设备的关键部分，但功耗开销较大。新型存储器工艺因静态功耗低和存储密度高等优点为存储系统提供了新的解决方案。在缓存体系中使用MLC STT-RAM存储工艺，有助于增大存储容量和降低功耗。然而，基于新型工艺的缓存具有写代价高和写干扰等问题。本项目将研究如何优化缓存体系结构来解决这些问题，拟构建一个智能终端环境中缓存数据分配、缓存数据压缩和缓存数据管理的理论框架，从缓存数据访问流的三个阶段设计高能效的缓存。主要研究内容包括：1）基于访问行为学习的MLC STT-RAM缓存数据分配方法；2）基于统计学习的MLC STT-RAM缓存数据压缩方法；3）基于概率分布的MLC STT-RAM缓存数据管理方法。本项目的预期成果将为非易失性缓存设计者提供功耗优化的新思路，为新型存储技术在智能终端及计算机系统中的高效应用提供理论支撑。

It is very important to reduce the system power quickly for the battery powered intelligent terminal equipment. Memory is the key part of these devices, but its power consumption is high. The emerging memory technology provides a new solution for the storage system because of its low static power consumption and high storage density. The use of MLC STT-RAM storage technology in the cache architecture helps to increase storage capacity and reduce power consumption. However, cache based on emerging technology has the problems of high writing cost and write disturbance. This project will study how to optimize the cache architecture to solve these problems. This project intends to construct a theoretical framework of cache data allocation, data compression and data management in the intelligent terminal environment, we will design energy efficient cache from three stages of cache data access flow. The main research contents include: 1) Data allocation scheme of MLC STT-RAM cache based on access behavior learning; 2) Data compression scheme of MLC STT-RAM cache based on statistical learning; 3) Data management scheme of MLC STT-RAM cache based on probability distribution. The research results of this project will provide a novel idea of power optimization for non-volatile cache designers, and provide theoretical support for the efficient application of emerging memory technologies in intelligent terminals and computer systems.