基于错误感知的闪存存储系统性能和寿命优化技术研究

Flash memory has been widely used in storage systems, such as embedded systems, personal computers and even data centers. As storage devices, flash memory has many advantages, including high random access performance, low power consumption, shock resistance, lightweight form factors and so on. However, bad write performance and limited lifetime are two key challenges in the development of flash memory, which hold back the further applications of flash memory. In addition, the write performance and lifetime are exponentially degraded with the development of technology scaling and density improvement. With the understanding of these two key issues, improving the write performance and lifetime become the most important work in the development and applications of flash memory. This project proposes to improve the write performance and lifetime of flash memory from the view of flash controller and application characteristics with the understanding on the basic reason of bad write performance and limited lifetime, flash error. First, based on the understanding the different error sources of flash memory, a flash error model is constructed, which provides the basic framework for flash error aware optimization techniques. With the construction of flash error model, flash controller based error management techniques are designed to be flash errors awareness and application characteristics based error optimization techniques are studied to reduce flash errors. With these techniques, the performance and lifetime of flash memory are improved. The results of this project will provide new directions for performance and lifetime optimization of flash memory in embedded systems, personal computer systems, and data centers and further improve the development and applications of flash memory.

闪存作为目前发展最为广泛的存储设备具有诸多优点，比如随机访问性能好、功耗低、无振动以及尺寸小等特点。然而，闪存仍然存在两个主要问题，写性能较差和寿命有限，导致闪存的进一步发展和新型应用受到严重阻碍。另一方面，随着闪存密度和尺寸的改进，写性能和寿命呈现指数下降趋势。优化闪存写性能和寿命成为闪存发展至关重要的研究课题。本项目将从闪存性能和寿命的深层次原因，闪存错误，研究如何从控制器和应用程序访问特征两个角度改善闪存存储系统的性能和寿命。首先根据闪存错误源特征，构建以闪存性能和寿命为优化目标的闪存错误模型，为闪存错误感知研究提供基础框架。然后在此基础上开展基于控制器的闪存错误感知研究和基于应用程序特征的闪存错误优化研究，最终达到闪存性能和寿命优化的目标。本项目的成果将为闪存存储设备在嵌入式系统、个人计算机甚至高性能计算机中的性能和寿命优化提供新思路，为各类闪存存储设备的广泛应用提供技术支撑。

闪存目前已经广泛应用于各类计算机系统作为存储设备，包括手机、个人电脑甚至计算中心中。最近的数十年闪存的发展方法主要是通过尺寸的缩小和增加单个存储密度实现存储容量的大幅度提升。2011年开始，3D闪存开始得到很好的发展。3D闪存通过将存储元采用纵向堆叠的方式实现存储元密度的进一步改善从而实现容量的再一次飞跃。然而随着闪存的发展，可靠性开始成为限制闪存的应用的关键。根据目前调研到的数据表明，闪存的可靠性随着存储元尺寸、单个存储元密度和堆叠层次的增加不断下降，最终导致其性能和寿命随之下降。为了保障闪存在未来发展过程中的合理使用，改善闪存的可靠性、性能和寿命成为关键。本项目在此背景下，展开了以下几个方面的研究工作。首先，由于闪存的可靠性是基于闪存物理本质特征而导致的。因此，我们首先构建了一个能够模拟闪存可靠性的模型，从而为我们后续的研究提供基础。通过模型的设计，我们进行了闪存多个可靠性特征的优化技术研究，包括纠错技术，特别是近几年广泛应用的基于低密度稀疏矩阵的纠错技术（LDPC）的优化，除此之外，还有包括读错误、写错误、制程差异以及保存时间等等的针对性优化技术研究。通过我们的研究使得闪存即使在可靠性较差的情况下，其性能也得到了重大的改善，从而延展了闪存的发展。然后在此基础上，我们在系统纵向层次上进一步展开了性能和可靠性的优化研究工作，通过系统IO调度、垃圾回收并行化、文件系统碎片优化以及存储压缩感知优化设计等多个方法实现了闪存存储系统性能和可靠性的改善。通过本项目的实施，我们实现了闪存可靠性和性能的改善，为闪存的进一步推广使用提供了重要的技术保障。