Half-time seminar Weining Song: Breaking Architecture Boundaries for Extremely Energy-constrained Internet of Things

  • Date: 19 December 2024, 15:00–16:00
  • Location: Ångström Laboratory, Ång 73121 (SoS Library, Hus 7 floor 3) Maze map: https://link.mazemap.com/slhVfl4z or digital by Zoom: https://uu-se.zoom.us/j/4561631543
  • Type: Seminar
  • Lecturer: Weining Song
  • Organiser: Department of Electrical Engineering
  • Contact person: Weining Song

Welcome to Weining Song's half-time seminar

Title: Breaking Architecture Boundaries for Extremely Energy-constrained Internet of Things

Speaker: Weining Song

Supervisors: Luca Mottola, Thiemo Voigt, Stefanos Kaxiras, Yuan Yao

Chairperson: Ted Johansson

Location: Ång 73121 (SoS Library, Hus 7 floor 3)

Map: https://link.mazemap.com/slhVfl4z

Zoom link: https://uu-se.zoom.us/j/4561631543

Time: Thursday, 19th of December, 15:00 – 16:00

Abstract

To enhance sustainability and reduce maintenance costs, ambient energy harvesting replaces traditional batteries as the only power source for Internet of Things (IoT) devices. However, ambient energy is erratic, leading to frequent energy failures, which causes devices to lose the program state. Then, program executions become intermittent, requiring the use of energy-hungry non-volatile memory (NVM) to persist program states. In this scenario, energy efficiency plays a critical role in ensuring forward progress. Most literature focuses on narrow optimization areas, such as memory operations on NVM. Our first paper also investigated this scope. Specifically, we explored eliminating silent stores—memory write operations that do not modify existing content—as a means to enhance energy efficiency. However, our findings indicate that resource limitations and modern NVM technology neutralize their impact on the energy figures in practice. Realizing the limited potential for further gains in this direction, we shifted our focus to the system architecture design, which offers broader opportunities to enhance energy efficiency. We designed TaDA (Task Decoupling Architecture), which enables efficient execution of IoT applications across multiple computing units. Compared with the single most efficient IoT device, TaDA saves up to 96.7% of energy and yields up to a 68.7x throughput improvement. Future research will focus on further enhancing the energy efficiency of TaDA and exploring other system-level design aspects, such as optimizing the wireless communication architecture, to improve the overall performance and sustainability of IoT devices.

FOLLOW UPPSALA UNIVERSITY ON

facebook
instagram
twitter
youtube
linkedin