Energy-Efficient System Architectures for Intermittently-Powered IoT Devices
Various industry forecasts project that, by 2020, there will be around 50 billion devices connected to the Internet of Things (IoT), helping to engineer new solutions to societal-scale problems such as healthcare, energy conservation, transportation, etc. Most of these devices will be wireless due to the expense, inconvenience, or in some cases, the sheer infeasibility of wiring them. With no cord for power and limited space for a battery, powering these devices for operating in a set-and-forget mode ( i.e., achieve several months to possibly years of unattended operation) becomes a daunting challenge. Environmental energy harvesting (where the system powers itself using energy that it scavenges from its operating environment) has been shown to be a promising and viable option for powering these IoT devices. However, ambient energy sources (such as vibration, wind, RF signals) are often minuscule, unreliable, and intermittent in nature, which can lead to frequent intervals of power loss. Performing computations reliably in the face of such power supply interruptions is challenging. Intermittently-powered IoT devices are an emerging class of embedded devices that operate on energy harvested from intermittent sources. These devices execute long running programs incrementally (in small steps each power-ON period) and across multiple power-ON periods. A prerequisite for operating in this manner is the need for some form of checkpointing of system state from SRAM to non-volatile memory when power loss is imminent. Traditionally, microcontrollers have employed Flash memory as the primary non-volatile storage technology. However, the energy (and latency) intensive operations of Flash make it inefficient for frequent checkpointing, and consume a significant amount of energy that could otherwise be used for executing meaningful application-related computations and tasks. This dissertation proposes system architectures to improve the energy-efficiency of intermittently-powered IoT devices while ensuring the reliability and forward progress of applications executing on them. First, to reduce the checkpoint overhead, we explore a unified memory architecture using an emerging non-volatile memory. Recent advances in memory technology has resulted in the emergence of non-volatile memory that combine the benefits of SRAM with the non-volatility of Flash. Memories such as Ferroelectric RAM (FeRAM), Magnetoresistive RAM (MRAM), etc., have superior power-performance characteristics, as compared to Flash. In this dissertation, we propose an in-situ checkpointing scheme using a unified-FeRAM architecture to reduce the checkpointing overhead and demonstrate that it enables the efficient usage of gathered energy. Second, we present an energy-aware dynamic memory mapping scheme for hybrid FeRAM-SRAM MCUs in intermittently-powered IoT devices to exploit both the reliability benefits of FeRAM and the performance benefits of SRAM. Even though FeRAM is non-volatile, it is slower than SRAM and have a higher power consumption. However, SRAM is volatile making it unreliable for intermittently-powered IoT devices. Hence, in this dissertation, we propose an intermediate approach in hybrid FeRAM-SRAM MCUs to benefit from the non-volatility of FeRAM and the speed of SRAM. Last, we architect a new low power mode for deeply embedded MCUs by performing sleep mode voltage scaling to enable SRAM data retention at ultra-low power consumption. Most IoT devices operate in an intermittent manner wherein they become active for a short duration of time to perform the intended task and then enter a sleep mode. However, present day sleep modes of MCUs are energy-inefficient due to the requirement of retaining state. Hence, we propose a new low power sleep mode that retains the SRAM data at ultra-low power consumption and demonstrate the powering of the proposed mode via harvesting minuscule amounts of ambient energy. We believe that the contributions made in this dissertation take a significant step in realizing set-and-forget IoT devices and in furthering the field of intermittently-powered computing.
Raghunathan, Purdue University.
Computer Engineering|Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our