We present hardware/software techniques to intelligently regulate supply voltage and clock frequency of intermittently computing devices. These devices rely on ambient energy harvesting to power their operation and small capacitors as energy buffers. Statically setting their clock frequency fails to capture the unique relations these devices expose between capacitor voltage, energy efficiency at a given operating frequency, and the corresponding operating range. Existing dynamic voltage and frequency scaling techniques are also largely inapplicable due to extreme energy scarcity and peculiar hardware features. We introduce two hardware/software co-designs that accommodate the distinct hardware features and function within a constrained energy envelope, offering varied tradeoffs and functionalities. Our experimental evaluation combines tests on custom-manufactured hardware and detailed emulation experiments. The data gathered indicate that our approaches result in up to 3.75× reduced energy consumption and 12× swifter execution times compared to the considered baselines, all while utilizing smaller capacitors to accomplish identical workloads.

We present ALFRED: a virtual memory abstraction that resolves the dichotomy between volatile and non-volatile memory in intermittent computing. Mixed-volatile microcontrollers allow programmers to allocate part of the application state onto non-volatile memory. Programmers are therefore to manually explore the tradeoff between simpler management of persistent state against energy overhead and possibility of intermittence anomalies due to nonvolatile memory operations. This approach is laborious and yields sub-optimal performance. We take a different stand with ALFRED: we provide programmers with a virtual memory abstraction detached from the specific volatile nature of memory and automatically determine an efficient mapping from virtual to volatile or non-volatile memory. Unlike existing works, ALFRED does not require programmers to learn a new language syntax and the mapping is entirely resolved at compile-time, reducing the run-time energy overhead. We implement ALFRED through a series of machine-level code transformations. Compared to existing systems, we demonstrate that ALFRED reduces energy consumption by up to two orders of magnitude given a fixed workload. This enables workloads to finish sooner, as the use of available energy shifts from ensuring forward progress to useful application processing.

We present the design and evaluation of a 3.5-year embedded sensing deployment at the Mithræum of Circus Maximus, a UNESCO-protected underground archaeological site in Rome (Italy). Unique to our work is the use of energy harvesting through thermal and kinetic energy sources. The extreme scarcity and erratic availability of energy, however, pose great challenges in system software, embedded hardware, and energy management. We tackle them by testing, for the first time in a multi-year deployment, existing solutions in intermittent computing, low-power hardware, and energy harvesting. Through three major design iterations, we find that these solutions operate as isolated silos and lack integration into a complete system, performing suboptimally. In contrast, we demonstrate the efficient performance of a hardware/software co-design featuring accurate energy management and capturing the coupling between energy sources and sensed quantities. Installing a battery-operated system alongside also allows us to perform a comparative study of energy harvesting in a demanding setting. Albeit the latter reduces energy availability and thus lowers the data yield to about 22% of that provided by batteries, our system provides a comparable level of insight into environmental conditions and structural health of the site. Further, unlike existing energy-harvesting deployments that are limited to a few months of operation in the best cases, our system runs with zero maintenance since almost 2 years, including 3 months of site inaccessibility due to a COVID19 lockdown.

The resource-constrained devices of the battery-less Internet of Things are powered off energy harvesting and compute intermittently, as energy is available. Forward progress of programs is ensured by creating persistent state. Mixed-volatile platforms are thus an asset, as they map slices of the address space onto non-volatile memory. However, these platforms also possibly introduce intermittence bugs, where intermittent and continuous executions differ.