Collaborative Research: DESC: Type 2: Delphi: Life-time aware design frameworks for sustainable edge devices

合作研究：DESC：类型 2：Delphi：可持续边缘设备的生命周期感知设计框架

• 批准号: 2324862
• 负责人: Vijay Janapa Reddi
• 金额: $ 50万
• 依托单位: Harvard University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-10-01 至 2027-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2324862&HistoricalAwards=false
• 关键词: Collaborative; Research; DESC; Type; Delphi

Edge devices like wearable watches and cameras, phones, and sensors in homes, factories, and farms, have become the foundation of our daily interactions with technology and the source of significant data for numerous critical sectors like health and manufacturing. Unfortunately, the exponential growth in sheer numbers of these devices comes at a significant environmental cost. User and edge devices alone account for over one-third of the 4% of global carbon emissions attributed to information and communication technologies. This ecological impact is projected to worsen as the number of edge devices surges into the trillions over the next few decades. This project develops Delphi, an end-to-end framework that prioritizes environmental impact while considering user experience, performance, and efficiency when designing edge devices. Delphi will enable sustainable technological growth by allowing for the design of environmentally conscious edge devices throughout their lifecycle. Via Delphi, software and hardware designers will be better able to reduce the potential environmental harms of computing while still offering valuable computational services for users and society. The project will lay the foundations for devices that can last for decades, potentially reducing e-waste and transient, short-lived devices designed to minimize environmental footprint. The project will integrate sustainability into college courses, provide broader online educational modules, and host a sustainability hackathon. Delphi tools themselves will be open-source, open hardware, and come with open datasets– all to facilitate further research and enable carbon-aware design to ensure a more sustainable future for computing. This project seeks to make carbon and sustainability a first-order design parameter for future edge computing devices that range from tiny, energy-harvesting Internet of Things (IoT) devices to higher-performance consumer electronics. Delphi is a suite of carbon-aware design tools that consider factors like energy, e-waste, and water usage from the manufacturing of computational devices, as well as operation carbon factors of machine learning and software lifecycles. The framework covers sensors, computing, communication, and power, accommodating various edge devices, and includes probabilistic analysis of machine learning workflows design and their operational and embodied carbon impact. The project's research has three major tasks. First, quantifying device environmental impact by collecting a first-of-its-kind dataset via a state-of-the-art academic clean room, the Cornell Nanoscale Facility (CNF), with architectural carbon models for salient device components (e.g., processor, memory and storage, energy harvesting modules). This task integrates data into new foundational carbon models, guiding all research tasks. Second, tools for the design of systems with sustainability as a first-order design target, alongside performance and quality of service. The task develops Electronic Sustainability Records for devices on the Pareto-frontier to maintain system-specific sustainability ledgers to track environmental telemetry across the operational lifetime of devices. Finally, the third task develops runtime sustainability managers, including humans in the loop, to reduce device obsolescence. The software will gracefully degrade and upgrade system performance based on user choices, static lifetime requirements, and environmental factors. The comprehensive framework's effectiveness is demonstrated through short-lived "Ephemeral devices" and lifelong companion health and wellness wearable devices, nicknamed the "Infinite Bit." These two device archetypes provide a mechanism for continuous validation as the project matures.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.

可穿戴手表、相机、手机以及家庭、工厂和农场中的传感器等边缘设备已成为我们日常与技术互动的基础，也是健康和制造业等众多关键领域重要数据的来源。不幸的是，这些设备数量的指数增长是以巨大的环境成本为代价的。仅用户和边缘设备就占全球信息和通信技术碳排放量的4%的三分之一以上。随着未来几十年边缘设备数量激增至数万亿，这种生态影响预计将恶化。该项目开发了德尔菲，这是一个端到端框架，在设计边缘设备时，优先考虑环境影响，同时考虑用户体验，性能和效率。德尔菲将通过允许在整个生命周期中设计具有环保意识的边缘设备来实现可持续的技术增长。通过德尔菲，软件和硬件设计人员将能够更好地减少计算的潜在环境危害，同时仍然为用户和社会提供有价值的计算服务。该项目将为可持续使用数十年的设备奠定基础，从而可能减少电子废物和旨在最大限度地减少环境足迹的短暂设备。该项目将把可持续性纳入大学课程，提供更广泛的在线教育模块，并举办可持续性黑客活动。德尔菲工具本身将是开源的，开放的硬件，并带有开放的数据集-所有这些都有助于进一步的研究，并使碳意识设计，以确保一个更可持续的未来计算。该项目旨在使碳和可持续性成为未来边缘计算设备的一阶设计参数，这些设备从小型的能量收集物联网（IoT）设备到更高性能的消费电子产品。德尔菲是一套碳意识设计工具，它考虑了计算设备制造过程中的能源、电子废物和用水等因素，以及机器学习和软件生命周期的操作碳因素。该框架涵盖传感器、计算、通信和电源，可容纳各种边缘设备，并包括机器学习工作流程设计及其操作和具体碳影响的概率分析。该项目的研究有三大任务。首先，通过最先进的学术洁净室康奈尔纳米级设施（CNF）收集首个同类数据集，量化设备对环境的影响，其中包括突出设备组件的建筑碳模型（例如，处理器、存储器和存储装置、能量收集模块）。该任务将数据集成到新的基础碳模型中，指导所有研究任务。第二，以可持续性作为一级设计目标的系统设计工具，以及绩效和服务质量。该任务为帕累托边界上的设备开发电子可持续性记录，以维护系统特定的可持续性分类账，从而在设备的整个使用寿命期间跟踪环境遥测。最后，第三个任务开发运行时可持续性管理器，包括循环中的人类，以减少设备过时。该软件将根据用户选择、静态寿命要求和环境因素适度降级和升级系统性能。综合框架的有效性通过短暂的“短暂设备”和终身伴侣健康和健康可穿戴设备（绰号为“无限位”）得到证明。“这两种设备原型提供了随着项目成熟而持续验证的机制。该奖项反映了NSF的法定使命，并且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。