异构处理器根据不同类型的应用定制处理引擎，能够有效提升系统效能，近年来成为了工业界的主流。针对异构处理引擎的研究普遍开展，但是对异构处理器的存储层次的研究还存在很大的空间。首先，不同类型处理引擎的访存模式存在差异，对存储子系统提出了不同的需求。其次，异构系统的任务调度对存储子系统的效率影响很大。再次，异构处理引擎共享存储子系统资源时，由于各自的访存模式存在明显差异，很可能出现相互干扰的情况。本研究针对异构并行处理器的存储层次设计和管理问题，选取移动、桌面和服务器三个主流应用领域的体系结构，根据各自应用的特点和业界的发展趋势从中提取最为关键的具体研究点，从存储层次的角度（以cache管理策略为研究的首要重点）来缓解异构处理器中的存储墙、功耗墙和编程墙。

Heterogeneous processors which incorporate specialized processing engines for different types of applications, can substantially improve system energy efficiency. Although heterogeneous processing engines have been investigated pervasively, memory hierarchy design for heterogeneous processors is not well explored. First, different types of processing engines have different memory access behaviors, which requires different memory hierarchy design. Second, heterogeneous task scheduling may significantly affect the efficiency of memory subsystem. Third, heterogeneous processing engines with different access patterns share the memory resources, and thus may interfere with each other. This work focuses on memory hierarchy design and management for heterogeneous processors. We pick representative architectures from three mainstream application domains: mobile, desktop and server, and focus on the key research issues according to the application characteristics and the trend of academia and industry. This work aims to alleviate the memory wall, power wall and programming wall of heterogeneous processors from the perspective of memory hierarchy (mainly the cache management scheme).