MOPTIC-SM: Sleep mode-enabled multi-optimized intermittent computing for transiently powered systems

The proliferation of Internet of Things (IoT) based devices has led to a significant increase in their usage across a broad range of application areas. However, the replacement and/or periodic maintenance of the billions of conventional power supplies used by these systems is required on an annual basis, which raises serious environmental and economic concerns. To subjugate these issues, transiently-powered embedded systems (TPESs) make use of an ambient energy resource. However, the non-uniform availability of ambient energy results in frequent system reboots. This problem can be mitigated by utilizing state checkpointing in non-volatile memory, yet a high number of checkpoints can lead to excessive energy consumption. In this research, a novel sleep mode-enabled multi-optimized intermittent computing method is proposed that combines data sampling and memoization to reduce the number of checkpoints. The proposed method is validated using the Microchip SAM-L11 embedded platform for the implementation of the Canny Edge Detection (CED) algorithm. The results of the experiments indicate that the proposed method effectively reduces the number of checkpoints for a given application by 50 percent in CED and maintains 70 percent accuracy in comparison to conventional TPES checkpointing. It is believed that the proposed solution will have several exciting applications in both the consumer market and industry.

Automated summary

The proliferation of IoT devices has increased their usage across various applications. To address the environmental and economic concerns of replacing conventional power supplies, TPESs utilize ambient energy. However, the non-uniform availability of ambient energy leads to frequent system reboots, and excessive energy consumption due to a high number of checkpoints. This research proposes a novel sleep mode-enabled multi-optimized intermittent computing method that reduces the number of checkpoints by combining data sampling and memoization.