Energy-efficient fault tolerance approach for Internet of Things applications

Fault tolerance (FT) is essential in many Internet of Things (IoT) applications, in particular in the domains such as medical devices and automotive systems where a single fault in the system can lead to serious consequences. Non-volatile memory (NVM), on the other hand, is commonly used to improve system reliability due to its unique properties to retain data even if the power supply is lost. However, one of the most important drawbacks of NVM is that it imposes significant overhead regarding timing and energy. In this paper, we have proposed a unique technique with the use of NVM to create FT application specific architecture with almost no timing overhead and low energy overhead. We address the implementation of applications that are specified using synchronous data flow model of computation. We combine the use of NVM and classical CMOS transistors so that NVM judiciously stores selected complete states of the pertinent program. It allows the program to resume from the saved state in NVM when faults occur. The frequency of the state selection can be flexibly adjusted for an arbitrarily specified FT timing/energy overhead. Moreover, to find an optimal state selection (with low overhead), we have applied an improved min-cut max-flow algorithm. On a variety of typical benchmarks, the simulation results indicate that our approach incurs only a small overhead over lower bounds. It is also generic in a sense that it can be applied to a wide spectrum of underlying IoT architectures and computational models.