Lifetime Reliability of Systems-on-Chip: Unified Modeling and Dynamic Reliability Management

未来的集成电路每片芯片将包含数十个、数百个甚至数千个内核。然而，可以实现这一点的技术缩减也可能使底层硬件的可靠性降低，因为缺陷和磨损机制的数量不断增加。因此，可靠性是多处理器片上系统设计面临的主要问题之一。由于内核或片上网络（用于内核之间的通信）都可能成为这些系统的可靠性瓶颈，因此必须以统一的方式解决可靠性问题。为了解决可靠性挑战，本研究开发了一种新的统一的理论寿命可靠性建模框架。该框架基于有效的蒙特卡罗方法，将多处理器片上系统视为计算单元和通信单元的组合。本研究的目标是开发基于动态电压和频率缩放和应用重映射的新型动态可靠性管理技术。这些技术基于控制理论的概念，主动提高了多核系统的寿命可靠性。提出的动态可靠性管理技术能够开发更可靠的多处理器片上系统，通过从娱乐和游戏到生物工程，军事和太空的应用对社会产生巨大影响。更广泛地说，该项目的结果通过提高对可靠性作为新设计关注点与功耗、性能和面积作为传统目标之间权衡的理解，对未来集成系统的设计产生重大影响。

Future integrated circuits will contain tens, hundreds, or even thousand cores per chip. However, technology downscaling that can make this possible may also make the underlying hardware less reliable due to an increasing number of defects and wear out mechanisms. Therefore, one of the major problems facing the design of multiprocessor systems-on-chip is reliability. Because either the cores or the network-on-chip (used for communication between the cores) can become a reliability bottleneck for these systems, it is imperative that the reliability be addressed in a unified manner. To address the reliability challenge, this research develops a novel unified theoretical lifetime reliability modeling framework. This framework is based on efficient Monte Carlo methods to treat multiprocessor systems-on-chip as a combination of computation and communication units. The goal of this research is to develop new dynamic reliability management techniques based on dynamic voltage and frequency scaling and application remapping. Based on control theory concepts, these techniques proactively improve the lifetime reliability of multicore systems.The proposed dynamic reliability management techniques enable the development of more reliable multiprocessor systems-on-chip, which have a dramatic impact on society via applications ranging from entertainment and gaming to bio-engineering, military and space. More broadly, the results of this project impact significantly the design of future integrated systems by advancing the understanding of the tradeoffs between reliability as a new design concern and power consumption, performance and area as traditional objectives.