Defence of Thesis 2 October 2026: Lukas Liedtke – Faculty of Information Technology and Electrical Engineering

Doctoral Candidate Lukas Liedtke at the Department of Computer Science will hold a trial lecture and defend his doctoral thesis for the degree of Philosophiae Doctor (PhD). The degree is administered by the Faculty of Information Technology and Electrical Engineering at the Norwegian University of Science and Technology (NTNU). Thesis title Optimizing Energy Efficiency in Battery-less IoT Systems Trial lecture Assigned topic: Energy Tradeoffs for Machine Learning on the Extreme Edge Assessment committee The Faculty of Information Technology and Electrical Engineering has appointed the following members to the assessment committee for the evaluation of the thesis: First opponent : Professor Marco Zimmerling, TU Darmstadt, Germany Second opponent : Research Scientist Emily Ruppel, Bosch Research, United States Internal member : Associate Professor Di Liu Associate Professor Di Liu, Department of Computer Science, NTNU, has been appointed as the administrator of the assessment committee. The committee has concluded that the thesis is worthy of public defense for the PhD degree. Supervisors Main supervisor : Professor Magnus Jahre, Department of Computer Science Co-supervisor : Project Manager Frank Alexander Kraemer, DeepOcean Co-supervisor : Professor Per Gunnar Kjeldsberg, Department of Electronic Systems Time and venue The PhD trial lecture and defence of the thesis are open to the public: Trial lecture: 2 October 2026 at 10:15 – Disputasrommet, Gløshaugen, NTNU Defence of the thesis: 2 October 2026 at 13:15 – Disputasrommet, Gløshaugen, NTNU Thesis abstract Internet of Things (IoT) devices are becoming ubiquitous, with their numbers projected to grow exponentially. Today, most of them are powered by chemical batteries, a practice that is neither environmentally friendly nor economically sustainable at scale because frequent battery replacements result in substantial CO2 emissions and maintenance costs. A promising alternative is to equip IoT devices with the ability to harvest energy from their surrounding environment. This thesis presents three key contributions that support developers in designing and evaluating energy-efficient, battery-less IoT systems. Our first contribution is EStacker, a hardware evaluation platform that allows developers to perform repeatable evaluations of hardware and software configurations across different battery-less IoT designs. EStacker features a comprehensive power measurement subsystem that provides detailed insights into the energy distribution among core system components and application tasks, enabling the identification of inefficiencies in hardware and software behavior. Furthermore, we introduce ST-SP, an optimization that reduces evaluation time by an average factor of 6.3× across a diverse set of benchmarks, while maintaining a low average throughput error of just 7.7%. Energy storage in battery-less IoT systems is commonly implemented using (super-)capacitors, which inherently creates a trade-off: small capacitors enable fast response times, while large ones are needed to bridge extended periods of energy scarcity. To address this, adaptive energy storage and voltage control units such as REACT and CapDYN have been proposed. Although they can offer high efficiency and responsiveness, they struggle to simultaneously minimize overhead and support a broad range of SoCs. Our second contribution, Coulombix, addresses this gap by combining good efficiency (up to 81.8%), low overhead (through a self-managed design and low-cost components), fast response time, and flexible output voltage configuration, making it compatible with a wide range of SoCs. While Coulombix focuses on hardware-level optimization, the fluctuating nature of energy environments also presents opportunities for software-based optimizations. Our third contribution, ECM, enhances energy efficiency in connection-oriented wireless communication. ECM uses a lightweight analytical model to select an efficient connection policy at runtime based on the predicted harvested power. It also dynamically adjusts the system’s sampling frequency to align power consumption with the available input power, aiming for energy-neutral operation. Extensive evaluation demonstrates that ECM increases throughput compared to static connection policies by up to 9.5×.

Sjekk arrangørens side før du drar. Tid, pris og tilgjengelighet kan endres.