Collaborative Research: CNS Core:Small:IMPERIAL: In-Memory Processing Enhanced Racetrack Inspired by Accessing Laterally

协作研究：CNS Core:Small:IMPERIAL：受横向访问启发的内存处理增强赛道

• 批准号: 2133267
• 负责人: Jingtong Hu
• 金额: $ 32万
• 依托单位: University of Pittsburgh
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2133267&HistoricalAwards=false
• 关键词: Collaborative; Research; CNS; Core; Small

Next generation mobile systems require memory and storage with unprecedented density and access speed that meets strict power/energy and reliability constraints. Moreover, these systems can benefit from application specific acceleration on data intensive workloads. For instance, Internet of Things (IoT) devices are tasked with acquiring, storing, and processing vast amounts of acquired information. Edge systems may slightly relax power/energy constraints, but can benefit from acceleration of machine learning, security, or other application specific tasks while maintaining quality of service on tasks from simultaneous disparate users. This project explores applying a new and understudied emerging memory technology called domain-wall memory (DWM) and its application to the needs of mobile and edge devices. DWM has properties that can be exploited to increase storage density, access speed, and to relieve the memory access bottleneck that exists in modern systems. The PIs will leverage their expertise to create a cross-layer design approach spanning the device/circuit- through system-level to develop a novel cross-DWM (XDWM) memory architecture with lateral read and write access capabilities. These innovations will revolutionize storage and processing for next generation mobile and edge devices by providing synergistic data storage and efficient processing-in-memory (PIM) with hooks for reliability. A cross-layer evaluation methodology will be adopted to cover prototype fabrication, device-level characterization, architecture-level simulation, and full system integration and emulation to explore the PIM. The transformative nature of this research is a disruptive new memory system that is dense, reliable, energy-efficient, ultra low latency with compute capability that can revolutionize the storage and processing capabilities of next generation computing systems. Such systems particularly include IoT, mobile and secure shared use edge systems but also apply to high performance computing and cloud systems. Further impacts of the proposed research include the integration of various education and advocacy activities based on the resources available to the two PIs such as (i) outreach for local K-12 students through Pitt's “Investing Now” summer school and USF's “Engineering Day” and Expo, where Engineering solutions are showcased to approximately 10,000 K-12 students/parents/teachers. (ii) inclusivity: Both PIs have a track record of including Under-represented Minority (URM) students.. They will continue to focus on URM representation in their team. (iii) curriculum: course integration of the research at both sites.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.

下一代移动系统需要具有前所未有的密度和访问速度的内存和存储，以满足严格的功率/能源和可靠性限制。此外，这些系统还可以受益于针对数据密集型工作负载的特定于应用程序的加速。例如，物联网(IoT)设备的任务是获取、存储和处理海量获取的信息。EDGE系统可以略微放宽功率/能源限制，但可以从机器学习、安全或其他特定于应用的任务的加速中受益，同时保持对来自不同用户的任务的服务质量。该项目探索了一种新的、未被研究的新兴存储技术--域壁存储(DWM)及其在移动和边缘设备需求中的应用。可以利用DWM的特性来提高存储密度、访问速度，并缓解现代系统中存在的内存访问瓶颈。PI将利用他们的专业知识创建跨越设备/电路到系统级的跨层设计方法，以开发具有横向读写访问能力的新型跨DWM(XDWM)存储器体系结构。这些创新将通过提供协同的数据存储和高效的内存中处理(PIM)以及挂钩实现可靠性，彻底改变下一代移动和边缘设备的存储和处理。将采用跨层评估方法，涵盖原型制造、器件级表征、体系结构级仿真以及全面系统集成和仿真，以探索PIM。这项研究的变革性本质是一种颠覆性的新内存系统，该系统密度高、可靠、高能效、具有计算能力的超低延迟，可以彻底改变下一代计算系统的存储和处理能力。这类系统特别包括物联网、移动和安全共享使用边缘系统，但也适用于高性能计算和云系统。拟议研究的其他影响包括根据两个私人投资促进机构的现有资源整合各种教育和宣传活动，例如(I)通过皮特大学的“立即投资”暑期学校和美国联邦的“工程日”和博览会，向大约10,000名K-12学生/家长/教师展示工程解决方案，向当地的K-12学生进行外联。(2)包容性：两个私营部门主管机构都有吸纳代表不足的少数族裔学生的记录。他们将继续专注于他们团队中的URM代表。(Iii)课程：两个地点研究的课程整合。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Sustainable AI Processing at the Edge

边缘的可持续人工智能处理

• DOI: 10.1109/mm.2022.3220399
• 发表时间: 2023
• 期刊: IEEE Micro
• 影响因子: 3.6
• 作者: Ollivier, Sebastien;Li, Sheng;Tang, Yue;Cahoon, Stephen;Caginalp, Ryan;Chaudhuri, Chayanika;Zhou, Peipei;Tang, Xulong;Hu, Jingtong;Jones, Alex K.
• 通讯作者: Jones, Alex K.

Pod-racing: bulk-bitwise to floating-point compute in racetrack memory for machine learning at the edge

• DOI: 10.1109/mm.2022.3195761
• 发表时间: 2022-11
• 期刊: IEEE Micro
• 影响因子: 3.6
• 作者: S. Ollivier;Xinyi Zhang;Yue Tang;C. Choudhuri;Jingtong Hu;A.K. Jones

Toward Comprehensive Shifting Fault Tolerance for Domain-Wall Memories with PIETT

利用 PIETT 实现域壁存储器的全面移位容错

• DOI: 10.1109/tc.2022.3188206
• 发表时间: 2022
• 期刊: IEEE Transactions on Computers
• 影响因子: 3.7
• 作者: Ollivier, Sebastien;Longofono, Stephen;Dutta, Prayash;Hu, Jingtong;Bhanja, Sanjukta;Jones, Alex K.
• 通讯作者: Jones, Alex K.