Efficient State Retention for Transiently-powered Embedded Sensing

We present state retention techniques to support embedded sensing applications on 32-bit microcontrollers whose energy provisioning is assisted through ambient harvesting or wireless energy transfer. As energy availability is likely erratic in these settings, applications may be unpredictably interrupted. To behave dependably, applications should resume from where they left as soon as energy is newly available. We investigate the fundamental building block necessary to this end, and conceive three mechanisms to checkpoint and restore a device's state on stable storage quickly and in an energy-efficient manner. The problem is unique in many regards; for example, because of the distinctive performance vs. energy trade-offs of modern 32-bit microcontrollers and the peculiar characteristics of current flash chips. Our results, obtained from real experiments using two different platforms, crucially indicate that there is no ``one-size-fits-all'' solution. The performance depends on factors such as the amount of data to handle, how in memory the data is laid out, as well as an application's read/write patterns.