A Hardware-Based Multi-Stage Dynamic Power Management Architecture for Autonomous Low-Light Operation DOI: XXXXXXX.XXXXXXX Conference: IEEE/ACM International Symposium on Low Power Electronics and Design; August 05–07, 2026; Chicago, USA ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computer systems organization Embedded systems CCS: Hardware Power estimation and optimization CCS: Hardware Sensor devices and platforms

Charalampos S. Kouzinopoulos, Marcel Louis Meli, Martin Schellenberg, Philip J. Poole, Mathieu Bellanger, Matthias Kauer, Julien De Vos, Dimosthenis Ioannidis, Dimitrios Tzovaras · 2026

Read A Hardware-Based Multi-Stage Dynamic Power Management Architecture for Autonomous Low-Light Operation DOI: XXXXXXX.XXXXXXX Conference: IEEE/ACM International Symposium on Low Power Electronics and Design; August 05–07, 2026; Chicago, USA ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computer systems organization Embedded systems CCS: Hardware Power estimation and optimization CCS: Hardware Sensor devices and platforms by Charalampos S. Kouzinopoulos, Marcel Louis Meli, Martin Schellenberg, Philip J. Poole, Mathieu Bellanger, Matthias Kauer, Julien De Vos, Dimosthenis Ioannidis, Dimitrios Tzovaras (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-19879.txt.

Introduction

Kouzinopoulos and colleagues present a hardware orchestrated power management architecture for battery sensor nodes in low light. Software sleep modes leak a persistent quiescent current that dominates energy use when harvested light is scarce. Their design power gates the microcontroller and all non essential peripherals completely, with wake up run by an ultra low power PMIC, real time clock and a novel latch circuit. The result is a minimal quiescent drain of four hundred fifty two nanoamps, far below software sleep modes, improving autonomy for solar powered sensor networks.

Catalog record

CREATORCharalampos S. Kouzinopoulos, Marcel Louis Meli, Martin Schellenberg, Philip J. Poole, Mathieu Bellanger, Matthias Kauer, Julien De Vos, Dimosthenis Ioannidis, Dimitrios Tzovaras
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30650
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.19879

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Hardware-Based Multi-Stage Dynamic Power Management Architecture for Autonomous Low-Light Operation DOI: XXXXXXX.XXXXXXX Conference: IEEE/ACM International Symposium on Low Power Electronics and Design; August 05–07, 2026; Chicago, USA ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computer systems organization Embedded systems CCS: Hardware Power estimation and optimization CCS: Hardware Sensor devices and platforms Charalampos S. Kouzinopoulos, Marcel Louis Meli, Martin Schellenberg, Philip J. Poole, Mathieu Bellanger, Matthias Kauer, Julien De Vos, Dimosthenis Ioannidis, Dimitrios Tzovaras Abstract . The advance of autonomous Smart Sensor Networks and embedded systems for the Internet of Things, powered by photovoltaic energy harvesting, is severely limited by energy efficiency, especially in low-light environments. While Dynamic Power Management is essential for energy conservation, conventional software-based techniques that rely on processor-managed low-power states incur a persistent quiescent current drain. This current becomes the dominant energy sink in energy-scarce conditions, limiting autonomy. The work of this paper addresses this limitation by introducing a robust, hardware-orchestrated dynamic power management architecture that improves existing configurations for battery-based sensor nodes. The proposed architecture achieves a minimal quiescent drain of 452 ​ n ​ A 452nA , by completely power-gating the microcontroller and all non-essential peripherals, with wake-up orchestrated by an ultra-low-power PMIC, RTC and a novel latch circuit developed specifically for this work. Our evaluation demonstrates that the dynamic power management architecture is significantly more efficient than traditional software-based sleep modes. 1. Introduction The rapid development of Internet of Things (IoT) applications has led to the widespread deployment of Smart Sensor Nodes (SSNs) , embedded, autonomous devices for data-driven automation at the edge. The nodes integrate sensors for data acquisition with wireless communication capabilities, while recent advances have enabled the integration of edge AI ; the execution of AI models at the edge ( 2 ) ( 5 ) ( 6 ) for localized decision-making. To achieve autonomy, these systems are often powered by energy harvesting sources such as photovoltaic (PV) elements. In such energy-constrained settings, particularly in indoor or variable outdoor environments, energy budget is largely dictated by low-light performance. Dynamic power management (DPM) ( 3 ) ( 4 ) ( 11 ) is a fundamental approach to maximize operational lifetime, typically involving Dynamic Voltage and Frequency Scaling (DVFS) , clock gating, establishment of Voltage-Frequency Islands with dynamically adjustable Voltage/Frequency knobs, as well as event-driven sleep and low-power modes. However, these strategies share a fundamental limitation: they are typically software-orchestrated. The main system microcontroller (MCU) must remain in a deep-sleep or standby state to manage wake-up events. This active-sleep state, while low-power, incurs a persistent quiescent current (indicatively 2.57\mu A for an STM32L4 MCU in stop 2 mode). In ultra-low-light conditions where harvested power is minimal, this quiescent drain becomes the dominant energy sink, creating an operational floor for autonomous operation. To improve upon this gap, we build upon and extend recent work in battery-less systems ( 9 ) ( 10 ) to propose a hardware-orchestrated DPM architecture for battery-based sensor nodes that completely power-gates the MCU and most peripherals during sleep cycles, eliminating their quiescent current. This architecture is built around the always-on power domain of a Power Management Integrated Circuit (PMIC) , powering only a low-power Real-Time clock (RTC) and a capacitive touch sensor for periodic or on-demand state switching. These hardware components are responsible for orchestrating the wake-up and shutdown of the main compute domain. Contributions. The main contributions of this work are: • A robust hardware-orchestrated DPM architecture for battery-based sensor nodes that eliminates MCU and peripheral quiescent current by fully power-gating the compute domain, achieving a measured sleep-state drain of 452nA . • A new latch circuit for the capacitive sensor, forming a key part of the DPM architecture. This circuit, which enables the sensor to provide a persistent enable signal from a momentary trigger, was designed by the authors and integrated into the commercial Microdul MS8892 capacitive sensor. • The design of an autonomous, energy-harvesting SSN , integrating the proposed DPM architecture. • Experimental results demonstrating cold-starting capability and sustained, net-positive energy operation under very low illuminance levels. 2. Related Work Most prior work on power management optimization for energy-harvesting autonomous SSNs focuses on aggressive duty-cycling, where components enter idle, stop or deep sleep during periods of inactivity. While effective, these approaches still incur significant quiescent currents, limiting energy autonomy under low-light conditions (i.e. ( 1 ) ( 12 ) ). To further reduce leakages, several designs actively cut-off power to non-essential components during inactive periods (i.e. ( 7 ) ( 8 ) ). However, in these approaches the MCU typically remains powered, reducing system autonomy. Some works fully power down the main load, including the MCU and most non-essential peripherals, while keeping only a minimal set of ultra-low-power components active ( 9 ) ( 10 ) . These designs achieve significant energy efficiency under ultra-low-light conditions as low as 5 - 20 lux. However, they are typically battery-less and rely exclusively on small capacitive storage, which severely limits the maximum supported load and edge processing capabilities. For example, ( 10 ) presents a Bluetooth Low Energy (BLE) node operating down to 5 lux by duty-cycling between a 310 nA Deep Sleep state and a Normal mode either periodically using an RTC or based on a user command via a capacitive touch sensor. Unlike our proposed architecture though, this system lacks a dedicated PMIC and battery. Instead, it relies solely on a 100\mu F capacitor charged directly from PV , restricting its application to minimal, highly intermittent loads. Although a custom ASIC PMIC is included, with a reported operating current of 30nA at room temperature, only a single comparator of the Integrated Circuit (IC) is utilized to monitor the voltage of the capacitor; the system can be powered up as long as the supercapacitor’s voltage is at least 2.25V . The quiescent current consumption of the system in Deep Sleep can be approximated to 140nA based on the consumption of the comparator, RTC and touch sensor. Similarly, ( 9 ) reports a LoRaWAN node operating down to 20 lux using a dedicated PMIC and a 10mF supercapacitor to power a LoRa SX1276 transceiver, a BME680 sensor and a Cortex-M4 MCU , requiring 8mJ per 10 -minute active cycle. The limits of the PMIC are set in such a way that the system will start operating as soon as the charge of the supercapacitor reaches 3V and it will continue charging up to a maximum of 3.6V . After the 3V threshold is reached, the supercapacitor can be discharged down to 2V . A total load of up to 44.8mJ can be supported by this configuration. A larger supercapacitor can be used to support increased loads, at the disadvantage of requiring additional time to charge it at similar illumination levels. Table 1. State-of-the-art comparison Work Always-on components Always-on current ( 10 ) MA198 (Comparator), RV-3028-C7, MS8892 30nA+45nA +\,65nA ( 9 ) EM8502 - ( 7 ) ( 8 ) STM32L0 1\mu A ( 1 ) nRF8001, MSP430F2274 2\mu A + 1\mu A ( 12 ) QN9080, BQ22570, AEM30940, HDC2080 27\mu A This work PMIC based on AEM10941 , RV-3028-C7, MS8892 310nA (datasheet) 452nA (measured) An autonomous BLE node is presented in ( 7 ) . The node uses the BLUENRG-2 BLE

Read the complete text