Theory, methods, and tools for cross-layered design of uniquely efficient failure-resistant systems

独特高效的抗故障系统的跨层设计的理论、方法和工具

• 批准号: 1255951
• 负责人: Sandeep Gupta
• 金额: $ 18万
• 依托单位: University of Southern California
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-04-01 至 2018-03-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1255951&HistoricalAwards=false
• 关键词: Theory; methods; tools; cross; layered

In the era beyond Moore's law, it will be imperative to consider improving resistance of integrated circuits to failures due to fabrication defects and variations, soft errors, and wearout as first-order considerations, to ensure economic viability of the semiconductor industry and robust operation of digital systems. This research will develop and demonstrate the benefits of a new multi-layered approach - spanning circuit, micro-architecture, instruction-set architecture, system, and user layers - for improving resistance to failures caused by fabrication, soft errors, and wearout. An approach to characterize user-layer impact of failures within each module will be developed to dramatically reduce overdesign and overtesting in future failure resistant systems. This research will also develop the first infrastructure to support systematic development of multi-layered approaches for failure resistance, including the first comprehensive models of layers, the first systematic method to identify candidate approaches and characterize their performance and overheads, and to derive globally optimal designs. This approach will identify efficient designs that will achieve failure resistance at uniquely low overheads. The utilitarian gains to society should be immediate and and are likely to be substantial. The methods and tools developed will provide significant improvements in yield, performance, robustness to soft errors, and system lifespan. This project will train students and industry experts in the science and art of of efficient failure resistance, and prepare them for the era beyond the end of Moore's law. The principle investigator will also distribute to the community of researchers and industry experts the models, framework, and tools to be developed as a result of this project.

在超越摩尔定律的时代，必须首先考虑提高集成电路对由于制造缺陷和变化、软错误和磨损而导致的故障的抵抗力，以确保半导体行业的经济可行性和半导体行业的稳健运行。数字系统。这项研究将开发并展示一种新的多层方法的好处——跨越电路、微架构、指令集架构、系统和用户层——以提高对制造、软错误和磨损引起的故障的抵抗力。将开发一种表征每个模块内故障对用户层影响的方法，以显着减少未来抗故障系统中的过度设计和过度测试。这项研究还将开发第一个基础设施，以支持多层抗故障方法的系统开发，包括第一个综合层模型、第一个识别候选方法并表征其性能和开销的系统方法，并得出全局最优设计。这种方法将确定有效的设计，以独特的低开销实现故障抵抗。社会的功利收益应该是立竿见影的，而且可能是巨大的。开发的方法和工具将显着提高产量、性能、对软错误的鲁棒性和系统寿命。该项目将培训学生和行业专家有效抵抗故障的科学和艺术，并为他们迎接摩尔定律终结后的时代做好准备。首席研究员还将向研究人员和行业专家社区分发该项目要开发的模型、框架和工具。