Workshop Proposal: Processing-In-Memory (PIM) Technology - Grand Challenges and Applications

研讨会提案：内存处理 (PIM) 技术 - 重大挑战和应用

• 批准号: 2027324
• 负责人: Yiran Chen
• 金额: $ 5万
• 依托单位: Duke University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-07-01 至 2021-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2027324&HistoricalAwards=false
• 关键词: Workshop; Proposal; Processing; Memory; PIM

Processing-In-Memory (PIM) is a promising technology that was proposed to eliminate the costly data transmission between memory and computing units in conventional computer architectures. An input data can be directly injected into the memory that stores a constant for a computation without loading the constant to the computing unit. Such a design is particularly suitable for many emerging computations, such as graph analytics and neuromorphic computing, where one of the operands is constant during the computation while only the input changes. By eliminating the data transmission between the memory and computing units, PIM can substantially improve the computational efficiency of these applications.This project organizes a workshop to gather leading researchers and experts who have spent considerable effort and time studying Processing-In-Memory (PIM) technology. The proposed workshop will provide a forum for leading experts to consider in a synergistic manner a view of PIM technology from the perspective of circuits, architectures, systems, and applications. These researchers will discuss their visions of the critical challenges that need to be addressed in the near future. With the help of the participants, the workshop will produce a comprehensive report on the current state of the art and the outlook of PIM technology.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

内存处理（PIM）是一种很有前途的技术，它被提出来消除传统计算机体系结构中存储器和计算单元之间昂贵的数据传输。输入数据可以直接注入到存储用于计算的常数的存储器中，而无需将常数加载到计算单元。这种设计特别适合于许多新兴的计算，例如图形分析和神经形态计算，其中一个操作数在计算期间是恒定的，而只有输入发生变化。通过消除内存和计算单元之间的数据传输，PIM可以大大提高这些应用程序的计算效率。本项目组织了一个研讨会，聚集了花费大量精力和时间研究内存处理（PIM）技术的领先研究人员和专家。拟议的研讨会将为领先的专家提供一个论坛，从电路、架构、系统和应用的角度，以协同的方式考虑PIM技术。 这些研究人员将讨论他们对在不久的将来需要解决的关键挑战的看法。在与会者的帮助下，研讨会将产生一个关于PIM技术的当前状态和前景的综合报告。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。