SHF: EAGER: Transactional Processors: Exploiting Hardware Transaction Processing for Reliable Computing

随着半导体技术进入纳米时代，CMOS器件面临的瞬时故障(如高能粒子撞击引起的软错误)的脆弱性急剧增加。这种软错误已经成为设计下一代微处理器的主要挑战。虽然在低水平优化可靠性的技术可能是准确的，但它们会招致相当高的硬件管理费用和昂贵的制造工艺。该方案的目标是通过利用硬件事务处理的语义来探索一种新的灵活的处理器体系结构，以实现高效可靠的计算。PI不是简单地增加现有处理器以获得可获得的可靠性，而是建议利用来自数据库管理系统和最近的事务存储器的事务处理的语义来设计和实现事务处理器体系结构，其中可靠的计算是固有的属性。事务处理器旨在为从检测到的错误中恢复提供高效和灵活的事务级验证和本机支持。PI将探索硬件事务处理和基于事务的可靠计算的设计空间，以及新的编程语言构造，以扩展当前的编程语言，以便在事务中高效地编写程序。该项目的成功可能会导致基于硬件事务处理的低成本可靠计算平台的设计。此外，拟议的活动将提供一个独特的渠道，吸引代表人数不足的群体和少数族裔的学生进入科学和工程专业。PI计划利用几个学院和大学范围的外展项目与高中生和老师互动，以激励他们学习计算机科学。

With the semiconductor technology entering the nano-scale era, CMOS devices are facing a dramatic increase in vulnerability to transient faults such as soft errors induced by energetic particle strikes. Such soft errors have become a major challenge in designing next generation microprocessors. While techniques for optimizing reliability at low levels can be accurate, they incur significantly high hardware overheads and costly manufacturing processes. The objective of the proposed EAGER proposal is to explore a new flexible processor architecture for highly effective reliable computing by exploiting the semantics of hardware transaction processing. Instead of simply augmenting existing processors for an attainable reliability, the PI proposes to exploit the semantics of transaction processing from database management systems and recent transactional memories for the design and implementation of the transactional processor architecture, where the reliable computing is an inherent property. The transactional processor aims to provide highly effective and flexible transaction-level verification and native supports for recovery from detected errors. The PI will explore the design space of hardware transaction processing and transaction based reliable computing, as well as new programming language constructs to extend current programming languages for writing programs efficiently in transactions. The success of this project may result in design of low-cost reliable computing platforms based on hardware transaction processing. In addition, the proposed activities will provide a unique channel to attract students from under-represented groups and minorities into science and engineering. The PI plans to take advantage of several college and university wide outreach programs to interact with high school students and teachers to motivate them in computer science.