CRII: SHF CSR: A High Performance and Reliable Non-Volatile Memory Framework for Handheld Platforms

CRII：SHF CSR：适用于手持平台的高性能且可靠的非易失性存储器框架

• 批准号: 1566158
• 负责人: Lide Duan
• 金额: $ 17.44万
• 依托单位: University of Texas at San Antonio
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-03-01 至 2019-02-28
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1566158&HistoricalAwards=false
• 关键词: CRII; SHF; CSR; Performance; Reliable

Handheld platforms, such as smartphones, tablets, and wearable devices, nowadays dominate the semiconductor market. Increasingly demanding programs are run by users on handheld platforms powered by batteries. Therefore, specialized hardware components, i.e., IP cores, are introduced to improve handhelds? energy efficiency. However, the memory access patterns of CPUs and IPs are dramatically different, making main memory a critical bottleneck for a handheld system. Non-volatile memories, particularly STT-MRAM, are emerging as replacements for the existing DRAM-based main memory, achieving a wide variety of advantages. Nevertheless, substituting DRAM with STT-MRAM in handhelds also results in new design challenges such as read disturbance. Consequently, two read disturbance mitigation schemes are proposed in this project, constructing a high performance and reliable memory framework for handheld platforms.In this research, a flexible STT-MRAM read scheme, namely FlexRead, is proposed to optimize handheld platform performance and energy efficiency under read disturbance. FlexRead combines the advantages of two existing data sensing schemes, dynamically issuing the best sensing scheme based on the runtime system state. Furthermore, an application flow pattern prediction scheme, namely FlowPaP, is proposed to identify and remove unnecessary memory restores originally required for preventing read disturbance. FlowPaP analyzes the data flow behavior in the applications running on handhelds, recognizing that the memory is in a consistent loop of being alternately written and read. These two proposed schemes optimize multiple handheld platform design aspects, thus having a transformational impact on the handheld research community, industrial products, and even society. The discovered read disturbance issue is an important and realistic design challenge for future handheld platform generations. Moreover, the proposed research can benefit the courses and research seminar taught by the PI, also advancing the education of students from highly diverse backgrounds, e.g., those from under-represented groups in the STEM field.

手持平台，如智能手机、平板电脑和可穿戴设备，如今主导着半导体市场。越来越多的要求程序由用户在由电池供电的手持平台上运行。因此，专用硬件组件，即，IP核，引入，以改善手持设备？能源效率然而，CPU和IP的内存访问模式有很大的不同，这使得主存成为手持系统的一个关键瓶颈。非易失性存储器（特别是STT-MRAM）正在作为现有的基于DRAM的主存储器的替代品出现，从而实现各种各样的优点。然而，在手持设备中用STT-MRAM替代DRAM也会导致新的设计挑战，例如读取干扰。因此，本研究提出了两种读干扰抑制方案，为手持平台构建了一个高性能、高可靠的存储框架。本研究提出了一种灵活的STT-MRAM读方案，即FlexRead，以优化手持平台在读干扰下的性能和能量效率。FlexRead结合了两种现有数据感知方案的优点，根据运行时系统状态动态发布最佳感知方案。此外，一个应用程序流模式预测方案，即FlowPaP，提出了识别和删除不必要的存储器恢复原来需要防止读干扰。FlowPaP分析在手持设备上运行的应用程序中的数据流行为，识别内存处于交替写入和读取的一致循环中。这两个方案优化了多个手持平台设计方面，从而对手持研究社区，工业产品，甚至社会产生了变革性的影响。所发现的读取干扰问题是未来手持平台的重要和现实的设计挑战。此外，拟议的研究可以使PI教授的课程和研究研讨会受益，也可以促进来自高度多样化背景的学生的教育，例如，那些来自STEM领域代表性不足的群体。

项目成果

Enabling NVM-based deep learning acceleration using nonuniform data quantization: work-in-progress

• DOI: 10.1145/3125501.3125516
• 发表时间: 2017-10
• 期刊: Proceedings of the 2017 International Conference on Compilers, Architectures and Synthesis for Embedded Systems Companion
• 作者: Hao Yan;Ethan C. Ahn;Lide Duan

Enabling reliable main memory using STT-MRAM via restore-aware memory management: work-in-progress

通过恢复感知内存管理使用 STT-MRAM 启用可靠的主内存：正在进行中

• DOI: 10.1145/3125501.3125517
• 发表时间: 2017
• 期刊: Architectures and Synthesis For Embedded Systems (CASES
• 作者: Aboutalebi, Armin Haj;Duan, Lide
• 通讯作者: Duan, Lide

RAPS: Restore-Aware Policy Selection for STT-MRAM-Based Main Memory under Read Disturbance

• DOI: 10.1109/iccd.2017.110
• 发表时间: 2017-11
• 期刊: 2017 IEEE International Conference on Computer Design (ICCD)
• 作者: Armin Haj Aboutalebi;Lide Duan