权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Energy Efficiency in Computing Logical Operations: Fundamental Limits with and Without Feedback

计算逻辑运算的能源效率：有反馈和无反馈的基本限制

基本信息

批准号：
1809194
负责人：
Murti Salapaka
金额：
$ 40万
依托单位：
University of Minnesota-Twin Cities
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2018
资助国家：
美国
起止时间：
2018-08-15 至 2022-07-31
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=1809194&HistoricalAwards=false
关键词：
Energy Efficiency Computing Logical Operations

项目摘要

The exponential increase in the ability to compute and store massive amount of data is fueling transformative changes in the human condition. It has led to a highly connected world with information being accessible at remote parts of the world. The current focus on interconnection of infrastructural systems such as the, power grid, financial markets, health care systems, and the Internet of Things (IoT), all bear a direct relationship to the advances in the integrated chips, where the complexity and the number of components has doubled almost every two years, following the Moore's law. Less known is the fact that the energy dissipation per computation has outpaced Moore's law which is at the heart of the mobile computation revolution, enabling, small form-factor devices packed with high computational capabilities that include, light laptops, tablets, and cell-phones. With such exponential increases, it is inevitable that limits of current approaches and practices for computation and memory related technologies are reached. This proposal addresses the basic operations of computation and memory with respect to the fundamental limits on the energy dissipation needed to perform the operations. It also studies the energy cost associated with measurement and feedback actions for performing computing and memory management operations. It provides proof of concept experiments and methods to demonstrate realizability of the computational and memory related operations with energetics tens of order smaller than the current practices. The associated analysis of fundamental primitives of computational operations and the proof of concept experiments will guide realizations of future energy efficient computational platforms and will continue the remarkable increase in speed and efficiencies with which computations are performed.The focus of the project is on the fundamental limits on the energetics of computations and memory. The project investigates primitives of computational operations both analytically and experimentally. It uses the abstraction of the state of a memory bit via the position of a particle in a double well potential in a thermal bath. Transfer of particle across the wells in a potential landscape is used to realize AND, NOT and erasure operations. Under the project goals, bounds on the minimum energetics of transferring the state of the particle in the double well potential will be analytically derived. Here, key aspect of the approach is to estimate the change in entropy from the initial to the final state, and, relating the change in entropy to the energy associated with the transfer of state. The transfer of state can be accomplished with or without measuring the state of the particle in the double well potential; here, the effect of feedback where the measurement guides the protocol for transferring the state will be understood and analyzed. The theoretical foundations that link computations, statistical mechanics, and feedback will be accompanied by a versatile experimental platform. Double-well potentials will be realized by controlling optical fields which have the scale of energetics suited for the study. The experimental protocols will accommodate the need for different speeds of transferring the state; important to emulate quasi-static processes. Accurate and precise measurements of the position of the particle will be used for the estimation of the energy required in moving the particle across wells. The measurement protocols will be precisely characterized that will enable the study of closed-loop strategies that alter the transfer protocol based on measurements. The project will provide both analytics and proof-of-concept experimental realizations of basic primitives of computations with orders in magnitude smaller energy footprint.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.

计算和存储海量数据的能力呈指数级增长，正在推动人类状况的变革性变化。它导致了一个高度互联的世界，世界上偏远地区都可以访问信息。目前对基础设施系统互联的关注，如电网、金融市场、医疗保健系统和物联网(IoT)，都与集成芯片的进步有直接关系，根据摩尔定律，集成芯片的复杂性和组件数量几乎每两年翻一番。不太为人所知的是，每次计算的能量消耗已经超过了摩尔定律，摩尔定律是移动计算革命的核心，使具有高计算能力的小型设备成为可能，包括轻型笔记本电脑、平板电脑和手机。随着这种指数级的增长，当前计算和内存相关技术的方法和实践将不可避免地达到极限。该提案针对执行运算所需的能量消耗的基本限制来处理计算和存储的基本运算。它还研究了与执行计算和内存管理操作的测量和反馈操作相关的能源成本。它提供了概念证明、实验和方法，以演示与计算和记忆相关的运算的可实现性，其能量比目前的实践低几十个数量级。对计算操作基本原理的相关分析和概念验证实验将指导未来高能效计算平台的实现，并将继续显著提高执行计算的速度和效率。该项目的重点是计算和内存的能量学基本限制。该项目从分析和实验两方面研究计算操作的原语。它通过热浴中粒子在双势垒势中的位置来抽象记忆位的状态。在潜在地貌中，颗粒在井中的传输用于实现AND、NOT和擦除操作。在该项目的目标下，将解析地导出在双势垒势中转移粒子状态的最小能量学的界限。这里，该方法的关键方面是估计从初始状态到最终状态的熵的变化，并将熵的变化与与状态转移相关的能量联系起来。状态的转移可以在测量或不测量双势垒势中的粒子的状态的情况下完成；在这里，将理解和分析反馈的效果，其中测量指导用于转移状态的协议。将计算、统计力学和反馈联系在一起的理论基础将伴随着一个多功能的实验平台。通过控制具有适合研究的能量学尺度的光场，将实现双势垒势。实验协议将适应不同状态传输速度的需要；这对于模拟准静态过程很重要。准确和精确地测量颗粒的位置将用于估计在井间移动颗粒所需的能量。测量协议将被精确地表征，这将使得能够研究基于测量来改变传输协议的闭环策略。该项目将提供基本原始计算的分析和概念验证实验实现，其量级较小。该奖项反映了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。