AMBIENT-6G D2.2: Initial Report on Energy Neutral Device Hardware Design
Abstract
Deliverable D2.2 advances the results of D2.1 by translating the capability taxonomy of energy-neutral device (END) into implementable hardware. The report consolidates options for light, thermal, vibration, and radio frequency (RF) energy harvesting (EH), together with architectures for dedicated wireless power transfer (WPT). It details efficient RF front ends and charge pumps, and compares ambient and dedicated power paths. It then examines multisource power management, including combining strategies, cold start, prioritization, energy balance monitoring, and voltage limiting, with attention to low quiescent consumption. The document surveys representative low-power device platforms, backscatter communication (BC), and wake-up radios that align communication effort with available energy. A component-level life cycle assessment (LCA) supports responsible choices of materials and modules. The outcomes provide a traceable map from D2.1 classes to hardware building blocks, and prepare the implementation and validation work of D2.3, which will deliver prototypes, reference designs, and measurement-based energy budgets.
Full text
Towards standardized 6G connectivity for ambient-powered energy neutral IoT devices Deliverable D2.2 Initial Report on END Hardware Design AMBIENT-6G project has received funding from the Smart Networks and Services Joint Undertaking (SNS JU) under the European Union’s Horizon Europe research and innovation programme under Grant Agreement No 101192113. Date of delivery: 31th Oct, 2025 Version: 1.0 Project reference: 101192113 Call: HORIZON-JU-SNS-2024 Start date of the project: 1st Jan, 2025 Duration: 36 months Disclaimer: The content of the document has not been approved yet by the SNS JU.
D2.2 - Initial Report on END Hardware Design Document properties Document Number: D2.2 Document Title: Initial Report on END Hardware Design Editors: Riku Jäntti, Boxuan Xie, Masoud Kaveh (AAU) Authors: Priyesh Pappinisseri Puluckul, Dragan Subotic (IMEC) Riku Jäntti, Boxuan Xie, Kalle Koskinen, Masoud Kaveh (AAU) Osmel M. Rosabal, Onel. L. A. López, Amirhossein Azarbahram (OUL) Thomas Winiecki, Prasidh Ramabadran, Efstathios Katranaras (SEQ) Miltiadis Filippou, Aimilia Bantouna, Panagiotis Demestichas (WIN) Jona Cappelle, Lyssa Ramaut, Tijl Schepens, Jarne Van Mulders (KUL) Guus Leenders (QKS) Daniel Poehl, Ulrich Muehlmann (NXP) Contractual Date of Delivery: 31st Oct, 2025 Dissemination level: PU Status: Final Version: 1.0 Filename: AMBIENT-6G_2.2_v1.0 Revision History Revision Date Issued by Description v0.1 1 September 2025 AMBIENT-6G WP2 Internal review start. v0.2 15 September 2025 AMBIENT-6G WP2 External review start. v1.0 10 October 2025 AMBIENT-6G WP2 Submission to SteerCo. Page I
D2.2 - Initial Report on END Hardware Design Abstract: Deliverable D2.2 advances the results of D2.1 by translating the capability taxonomy of energy-neutral device (END) into implementable hardware. The report consolidates options for light, thermal, vibration, and radio frequency (RF) energy harvesting (EH), together with architectures for dedicated wireless power transfer (WPT). It details efficient RF front ends and charge pumps, and compares ambient and dedicated power paths. It then examines multisource power management, including combining strategies, cold start, prioritization, energy balance monitoring, and voltage limiting, with attention to low quiescent consumption. The document surveys representative low-power device platforms, backscatter communication (BC), and wake-up radios that align communication effort with available energy. A component-level life cycle assessment (LCA) supports responsible choices of materials and modules. The outcomes provide a traceable map from D2.1 classes to hardware building blocks, and prepare the implementation and validation work of D2.3, which will deliver prototypes, reference designs, and measurement-based energy budgets. Keywords: AMBIENT-6G, Energy-Neutral Devices, Ambient-IoT, Technology Selection, Hardware Selection, Initial Design, Life Cycle Assessment. Disclaimer Funded by the European Union. The views and opinions expressed are however those of the author(s) only and do not necessarily reflect the views of AMBIENT-6G Consortium nor those of the European Union or Horizon Europe SNS JU. Neither the European Union nor the granting authority can be held responsible for them. Internal reviewers Riku Jäntti (AAU) Efstathios Katranaras (SEQ) External reviewers Henrique Moura (IMEC) Jeroen Famaey (IMEC) Lieven De Strycker (KUL) Page II
D2.2 - Initial Report on END Hardware Design Executive Summary This document constitutes the issue of Deliverable D2.2: ‘Initial Report on END Hardware Design’, within the framework of the project titled "AMBIENT-6G-Towards standardized 6G connectivity for ambient-powered energy-neutral IoT devices" (Project Acronym: AMBIENT6G; Grant Agreement No: 101192113). This document advances the results of D2.1 by moving from a capability-centric taxonomy toward implementable hardware for energy-neutral device (END). It consolidates energy harvesting (EH) and wireless power transfer options, multisource power management units with protection and monitoring, and low-power communication paths based on backscatter and wake-up radios. In parallel, it surveys representative device platforms with measurable energy profiles and provides an updated view of environmental impact at component level. The result is a coherent map from capability needs to realizable building blocks that operate under intermittent energy while maintaining service quality and reliability. The contributions of this deliverable are fourfold. First, it provides a compact survey of light, thermal, vibration, and radio frequency (RF) energy sources together with front ends, rectifiers, and charge-pump circuits, and it clarifies when dedicated wireless power transfer (WPT) should complement ambient sources. Second, it describes multisource power management, including cold start, prioritization of inputs, energy balance monitoring, and voltage limiting, with an emphasis on low quiescent consumption. Third, it reports device platforms, backscatter options, and wake-up radio designs that align communication effort with available energy. Fourth, it updates life-cycle considerations so that design choices can reflect environmental impact of materials and manufacturing. The document serves as technical input to integration and validation activities and prepares the ground for more detailed implementation in the next phase. It should be emphasized that Deliverable D2.2 is an initial report on END hardware design. Consequently, the level of detail across chapters is non-uniform: certain topics, such as photovoltaic (PV) EH and RF charge pumps, already include experimental results, while others, such as unmanned aerial vehicle (UAV)-based wireless power transfer and energy-aware protocol switching, are presented in outline form with initial insights and results. This reflects the varying maturity of the work within WP2. Subsequent deliverables (D2.3 and D2.4) will expand the less developed parts with consolidated data, integration results, and validation outcomes. The present document should therefore be read as a first snapshot of progress and directions rather than a final account. In summary, the contents of this document are organized as follows: •Chapter 1presents the motivation, scope, objectives, and overall structure of the report, and positions the work with respect to the taxonomy of D2.1. •Chapter 2covers EH and WPT, including PV and thermoelectric options, RF rectifiers and charge pumps, comparison of ambient and dedicated RF EH, and multisource power management with protection and monitoring. •Chapter 3surveys low-power device platforms together with backscattering-type devices and wake-up radio front ends and processing, linking platform selection and communication choices to energy budgets. Page III
D2.2 - Initial Report on END Hardware Design •Chapter 4provides an updated life-cycle assessment at component level for boards, communication subsystems, processing, memory, energy storage, incoming energy interfaces, and casing, with guidance for sustainable design decisions. •Chapter 5concludes the document and outlines implications for the next deliverable. Page IV
D2.2 - Initial Report on END Hardware Design Contents 1 Introduction 1 1.1 Context from D2.1 and Link to the Present Deliverable . . . . . . . . . . . . . 1 1.2 Motivation..................................... 2 1.3 ScopeandObjectives............................... 2 1.4 Structure of the Document . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2 Energy Harvesting, Power Transfer, and Multisource Power Management Circuits 4 2.1 EnergyHarvesting................................. 4 2.1.1 Fundamentals, Types, and Characterization of Commercial Off-the-shelf Photovoltaic Cells for Light Energy Harvesting . . . . . . . . . . . . . 4 2.1.2 Fundamentals of Thermoelectric Generators for Thermal Energy Harvesting .................................. 13 2.1.3 Charge Pumps for Radio Frequency-Energy Harvesting . . . . . . . . . 16 2.1.4 Multi-Radio Access Technology Energy Neutral Device Modem Capability - Implications for Networking & Energy Harvesting . . . . . . . . . 18 2.2 Radio Frequency Wireless Power Transfer . . . . . . . . . . . . . . . . . . . . 20 2.2.1 Energy Transmitter Architectures . . . . . . . . . . . . . . . . . . . . 20 2.2.2 Energy Harvesting Circuits Design for Radio Frequency Wireless Power Transmission ............................... 22 2.2.3 Unmanned Aerial Vehicle-Based Wireless Power Transfer to Remote Energy-Neutral Devices . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.3 Power Management Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.3.1 Energy-Neutral Device Energy Harvesting from a Photovoltaic Cell . . 25 2.3.2 Power Management Circuits for Thermal Energy Harvesting . . . . . . 28 2.3.3 Multisource Power Management Units . . . . . . . . . . . . . . . . . . 28 2.3.4 Energy Balance Monitoring . . . . . . . . . . . . . . . . . . . . . . . . 30 2.3.5 Voltage Limiting Techniques for Energy-Neutral Devices . . . . . . . . 32 2.4 Summary ..................................... 35 3 Low-power IoT Devices, Backscatter Devices, and Wake-up Radio Design 36 3.1 Low-power Internet of Things Device Prototypes . . . . . . . . . . . . . . . . 36 3.1.1 DECT-2020Device............................ 36 3.1.2 Cellular Internet of Things Device . . . . . . . . . . . . . . . . . . . . 38 3.2 Backscattering Type of Devices . . . . . . . . . . . . . . . . . . . . . . . . . 41 3.2.1 Radio Frequency Identification Technology for Backscattering-Type EnergyNeutralDevices.............................. 41 Page V
D2.2 - Initial Report on END Hardware Design 3.2.2 Detailed Information on Backscattering-Type Energy-Neutral Devices . 43 3.2.3 Hardware Options for Backscattering . . . . . . . . . . . . . . . . . . 45 3.3 Wake-upRadioDesign .............................. 48 3.3.1 Light-Controlled Wake-up Front End Designed for Li2BC . . . . . . . 48 3.3.2 Wake-up Radio Processing Evaluations and Design Considerations in Cellular Internet of Things . . . . . . . . . . . . . . . . . . . . . . . . 49 3.4 Summary ..................................... 50 4 Life Cycle Assessment 51 4.1 Life Cycle Assessment of Typical Internet of Things Components . . . . . . . 51 4.1.1 Impact of the Printed Circuit Board . . . . . . . . . . . . . . . . . . . 52 4.1.2 Peripherals, Internal Memory and Processing . . . . . . . . . . . . . . 54 4.1.3 Communication.............................. 57 4.1.4 ExternalMemory............................. 60 4.1.5 EnergyStorage.............................. 62 4.1.6 IncomingEnergy ............................. 64 4.1.7 Radio Frequency Identifications . . . . . . . . . . . . . . . . . . . . . 66 4.1.8 Casing................................... 68 4.2 Summary ..................................... 69 5 Conclusion 71 Page VI
D2.2 - Initial Report on END Hardware Design Glossary 3D three-dimensional. 5G fifth-generation. ADC analog-to-digital converter. A-IoT Ambient Internet of Things. AC alternating current. AOA angle-of-arrival. AP access point. BC backscatter communication. BD backscatter device. BGA ball grid array. BLE Bluetooth Low Energy. BoM bill of materials. BPF bandpass filter. BS base station. Cat NB2 Category Narrowband IoT Release 14. CDTE cadmium telluride. CIGS copper indium gallium selenide. CMOS complementary metal oxide semiconductor. COTS commercial off-the-shelf. CPU central-processing unit. CSI channel state information. CSS chirp spread spectrum. DAC digital-to-analog converter. DAS distributed antenna systems. Page VII
D2.2 - Initial Report on END Hardware Design DC direct current. DECT Digital Enhanced Cordless Telecommunications. DMA direct memory access. DRAM dynamic random-access memory. DSP digital signal processing. E2E end-to-end. EC energy combiner. EDLC electrostatic double-layer capacitors. EEPROM electrically erasable programmable read-only memory. EF environmental footprint. EH energy harvesting. EMC electromagnetic compatibility. EN energy neutral. END energy-neutral device. EoL end of life. ESD electrostatic discharge. eSIM embedded subscriber identity module. ET energy transmitter. ETSI European Telecommunications Standards Institute. FDMA frequency division multiple access. FFT fast Fourier transform. FPGA field-programmable gate array. FSK frequency shift keying. GNSS global navigation satellite system. GPIO general-purpose input/output. GPS Global Positioning System. GWP Global Warming Potential. HASL Hot Air Solder Leveling. IC integrated circuit. IO input/output. Page VIII
D2.2 - Initial Report on END Hardware Design Chapter 2 Energy Harvesting, Power Transfer, and Multisource Power Management Circuits This chapter establishes the building blocks that turn intermittent energy into dependable service. It reviews ambient sources such as light, heat, vibration, and RF signals, and it introduces dedicated WPT when ambient supply is insufficient. For each source, we summarize conversion front ends, rectifiers, and charge pump options, together with link budget considerations. We then present multisource power management that combines harvesters, supervises cold start, prioritizes inputs, and protects storage and loads through voltage limiting and current control. We define interfaces between harvesters, storage, and the load, with emphasis on low-quiescent consumption and predictable transients, since those properties dominate service availability of low-power platforms. 2.1 Energy Harvesting In this section, several EH techniques are introduced, such as using light EH, thermal EH,RF EH, as well as an energy-aware protocol switching in END via harvesting potential predictions. 2.1.1 Fundamentals, Types, and Characterization of Commercial Off-theshelf Photovoltaic Cells for Light Energy Harvesting In this subsection, an overview of commercial off-the-shelf (COTS) PV cells employed for EH among END is presented. First, the principle of photovoltaic conversion and typical power output for moderate-sized panels are described. The main types of PV cells suitable for IoT applications are discussed in terms of materials, manufacturing processes, cost, market scale, and principal manufacturers. Subsequently, key performance parameters and related characteristics used to evaluate PV cells are introduced. The concept of simultaneous light information and power transfer (SLIPT) is then discussed, focusing on PV cell selection criteria and evaluation methods. Finally, the methodology adopted for experimental characterization of seven PV cells is detailed. Page 4 of 81
D2.2 - Initial Report on END Hardware Design 2.1.1.1 Overview and Principle of Photovoltaic Technology Photovoltaic conversion relies on the photovoltaic effect, which generates a voltage and electric current in a semiconductor junction upon exposure to light [1]. When photons possessing energy exceeding the semiconductor bandgap are absorbed, electron-hole pairs are produced and separated by the built-in electric field of a p-n junction, leading to direct current generation [2]. Typical PV cells employ silicon as the semiconductor material, although alternative materials such as cadmium telluride (CDTE),copper indium gallium selenide (CIGS), perovskites, and organic polymers have been developed [3]. In IoT energy-harvesting scenarios, PV cells convert ambient illumination, either sunlight or artificial indoor lighting, into electrical power to sustain low-power sensor nodes and wireless transceivers [4]. The recommended illumination level for different area can be found in [5]. Under indoor illuminance levels ranging from 100 lx to 1000 lx, advanced indoor PV harvesters can deliver power densities on the order of tens to hundreds of µW cm−2[4]. Under full-sun outdoor conditions (approximately 1000 W m−2), a moderate-sized panel with active area of 0.01 m2(for example, 10 cm ×10 cm) and an efficiency of 20 % can produce roughly 2 W of electrical power [6]. This amount is sufficient to charge batteries or supercapacitors that supply duty-cycled IoT devices. The power output of a PV cell generally depends on incident irradiance Gof light, cell temperature T, and cell efficiency η. The maximum output power of a PV cell can be expressed by Pmax =η G A , (2.1) where Ais the cell area. Cell temperature influences open-circuit voltage Voc negatively and short-circuit current Isc positively, yielding a net decrease in Pmax as temperature increases [3]. 2.1.1.2 Typical Parameters of PV Cells Electrical Parameters Photovoltaic cell performance is quantified by the current-voltage (I-V) characteristic under standard test conditions (STC: 1000 W m−2, AM1.5G standard spectrum, 25 ◦C). Key parameters include [3]: •Open-Circuit voltage Voc: the voltage at zero load current. •Short-Circuit Current Isc: the current at zero voltage. •Maximum Power Point (Pmax =Vmp Imp): the point on the I-V curve where the power is maximized, corresponding to voltage Vmp and current Imp. •Fill Factor (FF): defined as FF = Vmp Imp Voc Isc , indicating the “squareness” of the I-V curve. •Power Conversion Efficiency η: ratio of Pmax to incident power GA. Spectral and Responsivity Parameters Spectral responsivity R(λ)defines the photocurrent generated per unit incident optical power at wavelength λ. Quantum efficiency [1], [7] QE(λ) is related to responsivity by QE(λ) = λ R(λ) hc/e , where his Planck’s constant, cis the speed of light, and eis the elementary charge. Indoor PV materials are optimized for visible wavelengths (400 nm to 700 nm), whereas outdoor PV materials target the solar spectrum peak near 1000 nm. Page 5 of 81
D2.2 - Initial Report on END Hardware Design Table 2.1: Brief comparison of photovoltaic technologies relevant to energy harvesting. Technology Outdoor efficiency Indoor performance Flexibility Cost and maturity Typical uses Crystalline silicon 15–24 % Moderate at 200– 1000 lx Rigid Very low cost; very mature Utility, rooftop, outdoor IoT Thin-film 10–13 % Good at low lux Rigid or flexible Low cost; mature for CdTe/CIGS BIPV, façades, indoor IoT Perovskite and DSSC Evolving; fast progress High at 500–1000 lx Thin; potentially flexible Low-cost potential; stability limits Indoor harvesters, prototypes Organic Lower than cSi/CdTe 10–20 µW cm−2at 500 lx Very flexible; lightweight Low materials cost; emerging Wearables, conformal, indoor 2.1.1.3 Types of Photovoltaic Cells for IoT Devices Global PV module shipments exceeded 500 GW in 2024, dominated by leading manufacturers such as JinkoSolar, JA Solar, LONGi, Trina Solar, and Canadian Solar [8]. Tier-1 manufacturers typically produce modules at $0.20 W−1to 0.30 W−1in 2025, with large economies of scale reducing costs substantially [9]. Emerging IoT-specific PV suppliers include PowerFilm, Anysolar, and Ambient Photonics [10], [11]. A comparison of PV cell technologies are shown in Table. 2.1 and are introduced in the sequel. Silicon-Based PV Cells Crystalline silicon (c-Si) PV cells are the most prevalent, including monocrystalline and polycrystalline variants [3]. Monocrystalline cells exhibit higher efficiencies (19–24 %) than polycrystalline cells (15–18 %) due to superior crystalline quality [6]. Silicon wafer fabrication employs the Czochralski or float-zone processes, followed by doping, diffusion, and metallization steps. Module assembly involves encapsulation under ethylene-vinyl acetate and mounting on glass or polymer substrates. The levelized cost of silicon PV cells has decreased to below $0.15 W−1for large-scale production, with polysilicon feedstock and wafer manufacturing representing major cost components [12]. Leading manufacturers of c-Si modules include JinkoSolar, JA Solar, Trina Solar, LONGi, Canadian Solar, Hanwha Qcells, and ANYSOLAR [9]. Thin-Film PV Cells Thin-film technologies, such as amorphous silicon (a-Si), CDTE, and CIGS, offer lower production costs and mechanical flexibility at the expense of lower conversion efficiency. Amorphous silicon cells are deposited via plasma-enhanced chemical vapor deposition, while CDTE and CIGS are typically sputtered or co-evaporated onto glass or flexible substrates [12]. Typical efficiencies for thin-film cells range from 10–13 % in module form. First Solar and Solar Frontier are key manufacturers of such PV cells. Thin-film modules are used for buildingintegrated photovoltaics and indoor light harvesting due to superior low-light performance [13]. Perovskite and Dye-Sensitized PV Cells Perovskite solar cells utilize organic-inorganic halide perovskite absorbers (e.g., CH3NH3PbI3) deposited via solution processing [14]. Indoor perovskite PV devices with bandgaps near 1.75 eV can achieve power conversion efficiencies up to 25 % under 1000 lx illumination [15]. Perovskite solar cell fabrication employs spin coating, doctor blading (blade coating), or vapor deposition followed by low-temperature annealing, enabling flexible and low-cost modules. Dye-sensitized solar cells use mesoporous TiO2with adsorbed Page 6 of 81
D2.2 - Initial Report on END Hardware Design dye molecules and an electrolyte; their efficiency under indoor lighting can exceed 20 % in small modules [16]. Commercialization remains limited due to stability concerns. Organic PV Cells Organic PV cells employ semiconducting polymers and small molecules as active layers, fabricated by roll-to-roll printing on flexible substrates [4]. Indoor organic PV cell can deliver 10–20 µW cm−2at 500 lx, with power conversion efficiencies of up to 21 % under controlled indoor illumination [4]. Major organic PV cell suppliers include Heliatek and FlexEnable. Organic PV cells exhibit low cost and mechanical flexibility but remain susceptible to photochemical degradation. 2.1.1.4 PV Cells for SLIPT Apart from using PV cells for EH purpose, SLIPT technique integrates visible light communication with photovoltaic EH. This technique uses optical transmitters, such as a light emitting diode or laser diode, to modulate data onto a lightwave while simultaneously supplying energy to the receiver PV cell. In such configuration, the PV cell alternates between a photoconductive mode for information decoding and a photovoltaic mode for EH. Receiver architectures for SLIPT are generally classified as: (i) time-splitting, where time slots are alternated between EH and data reception; (ii) power-splitting, where a fraction of the optical power is diverted to a photodetector for information while the remainder is used by the PV cell for power generation; (iii) adaptive reception, which employs real-time adjustment of PV bias and load conditions to optimize the trade-off between communication and energy metrics. Hybrid designs may combine a high-bandwidth photodetector for data demodulation with a PV element tuned for maximum direct current (DC) conversion. The selection of PV cells for SLIPT applications should be guided by several key criteria. First, broad spectral responsivity in the visible band is required to ensure sufficient photocurrent for both EH and data detection. Second, high energy conversion efficiency is needed to maximize the harvested DC power under given illumination conditions. Third, low parasitic capacitance and resistance are imperative to achieve high-speed modulation and to raise the RC cutoff frequency. Fourth, rapid photodetection response, with nanosecond-scale rise and fall times, is essential for high data-rate applications. Finally, stability of open-circuit voltage and short-circuit current under high-frequency optical modulation should be ensured to prevent performance degradation during simultaneous operation. Evaluation of candidate PV cells is thus carried out through measurement of frequency response, bandwidth, alternating current (AC) amplitude transfer characteristics, and DC energy conversion under modulated illumination. 2.1.1.5 Experimental Characterization of COTS PV Cells Test Samples Seven models of COTS PV cells from ANYSOLAR are selected for characterization because of their widespread use in light energy-harvesting END prototypes, as shown in Fig. 2.1. The models are listed in Table 2.2 with key parameters provided by the vendor [17]. Throughout the table, it is observed that within the “K12L” family, increasing the unit-cell width from 10.8 mm to 21.5 mm nearly doubled the short-circuit current (from 21.0 mA to 41.9 mA) and correspondingly increased the maximum power from 132.3 mW to 263.0 mW, illustrating that wider unit cells capture more photons at the expense of larger package size and weight. Modules with different series counts (4, 9, or 12 cells) are found to enable nominal voltage Page 7 of 81
D2.2 - Initial Report on END Hardware Design Figure 2.1: Tested 7 PV cells samples (left) and characterization measurement setup (right). Table 2.2: Specifications of tested PV Cells provided by manufacturer. No. Model Voc (V)Isc (mA)Pmax (mW)Vmax (V)Imax (mA)FF (%) Efficiency (%)∆Vo c (mV)∆Is c (mA)Dimension (mm) Weight (g) Unit Cell (mm) Cells in Series 1KXOB12IK04TF 2.76 50.2 105.3 2.23 47.2 >70 25 −6.96 0.0227 23 ×25 ×1.2 1.2 21.5 ×5.65 ×1 4 2 SM500K12L 8.29 21.0 132.3 6.70 19.8 >70 25 −20.90.0095 24 ×32 ×1.8 2.5 10.8 ×4.71 ×1 12 3 SM710K12L 8.29 29.2 183.7 6.70 27.4 >70 25 −20.90.0100 32.5 ×33 ×1.8 4.5 15.0 ×4.71 ×1 12 4 SM850K12L 8.29 35.1 221.0 6.70 32.9 >70 25 −20.90.0159 38.5 ×33 ×1.8 4.3 18.0 ×4.71 ×1 12 5 SM101K12L 8.29 41.9 263.0 6.70 39.3 >70 25 −20.90.0189 45 ×32 ×1.8 4.8 21.5 ×4.71 ×1 12 6 SM141K04LV 2.76 58.6 123.0 2.23 55.1 >70 25 −7.00 0.0265 45 ×15 ×1.8 2.3 21.5 ×6.80 ×1 4 7 SM111K09L 6.22 46.7 220.4 5.02 43.9 >70 25 −15.70.0200 62 ×21 ×1.8 5.5 20.0 ×5.65 ×1 9 targeting for downstream power management: four-cell modules provide approximately 2.7 V, suitable for low-voltage converters, whereas twelve-cell strings yield about 8.3 V, appropriate for boost-converter inputs. Four-cell K04 modules achieved high short-circuit current (50–58 mA) in a compact footprint, producing Pmax values comparable to those of smaller-area twelve-cell modules due to reduced series resistance, although voltage step-up is required for loads above 2.2 V. The specified tolerances in open-circuit voltage (−6.96 mV to −20.9 mV) and shortcircuit current (0.0095–0.0265 mA) suggest that K12L cells experience slightly greater Voc drift under environmental variations, while all modules demonstrate robust current-output stability under moderate indoor lighting. A uniform 25 % efficiency rating across unit cells indicates a consistent semiconductor technology, but real-world performance will be governed by spectral mismatch, temperature coefficients, and series-resistance losses. From an SLIPT perspective, all cells feature fill factors above 70 % and substantial unit-cell areas that support both DC power harvesting and AC data reception; four-cell modules are expected to offer the highest modulation bandwidth due to lower total junction capacitance, whereas twelve-cell devices maximize harvested energy at the cost of reduced high-frequency response. Measurement Setup However, for above PV cells, key performance metrics related to SLIPT, such as AC response, DC response, and bandwidth are not provided by the vendor, which need to be systematically measured and analyzed. The measurement setup for the PV cell characterization is displayed in Fig. 2.1. The intensity modulation/direct detection method is utilized for light modulation. A functional generator (Tektronix AFG31000) produces a sine-wave electrical signal with an amplitude of 3 V, at frequencies ranging from 1–100 kHz, which modulates a light emitting diode (LED) driver biased by an 21 V DC power supply. A general purpose 7-LED module (OSRAM GW J9LHS2.4M-C0C7-2+35+13+35AW-1-100-R18) emits the modulated optical signal with the visible light spectrum from 400–800 nm. The optical output illuminated each PV Page 8 of 81
D2.2 - Initial Report on END Hardware Design 103104105 0 0.05 0.1 0.15 0.2 Benchmark PD 103104105 0 0.005 0.01 0.015 PV-1 103104105 0 1 2 3 4 5 610-3 PV-2 103104105 0 2 4 6 810-3 PV-3 103104105 0 0.002 0.004 0.006 0.008 0.01 PV-4 103104105 0 0.005 0.01 0.015 PV-5 103104105 0 0.005 0.01 0.015 PV-6 103104105 0 0.005 0.01 0.015 PV-7 Figure 2.2: Magnitude of PV-output AC components VAC for all PV cells in terms of the varying lightmodulating frequencies, at different distances between the LED and PV cell. 103104105 0 0.5 1 1.5 2 2.5 3 3.5 410-3 103104105 0 1 2 3 4 5 610-3 103104105 0 1 2 3 4 5 6 710-3 103104105 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 Figure 2.3: Comparison of the PV-output AC components across all tested PV cells, for different distances (0.5 m,0.4 m,0.3 m, and 0.2 m) between the LED and PV cells. cell under test, respectively. For each sample, both AC (peak-to-peak) and DC (mean) output of the photovoltaic-converted signal are recorded simultaneously by an oscilloscope (Rohde & Schwarz RTO64). All measurements are conducted in a darkened enclosure to eliminate ambient light interference. All measurement are repeated twice. Moreover, a ThorLabs PDA100A2 silicon photodetector is employed as a high-performance photodetector benchmark [18], featuring a bandwidth of 11 MHz. The subsequent paragraphs present and analyze the AC and DC response, bandwidth, and its correlation with manufacturer-rated output power. AC Response and Bandwidth Fig. 2.2 presents the measured magnitude of AC components VAC for all PV cells in terms of the varying light-modulating frequencies. For the measurement among each PV cell, different distances dbetween the LED and PV cell are adopted, ranging from 0.2 m to 0.5 m. A band-pass shape is observed for all PV cells: the magnitude rises from the lowest frequency, peaks between 3 kHz and 8 kHz, and thereafter decays with an approximate −20 dB dec−1slope towards 100 kHz. The trend is in close agreement with the small-signal RC model of a junction photovoltaic cell, where the dynamic resistance Rpv dominates at low frequency and the junction capacitance Cpv dominates at high frequency. The analytic corner Page 9 of 81
D2.2 - Initial Report on END Hardware Design 103104105 0 0.05 0.1 0.15 0.2 0.25 0.3 Benchmark PD 103104105 0 0.005 0.01 0.015 0.02 0.025 PV-1 103104105 0 0.002 0.004 0.006 0.008 0.01 PV-2 103104105 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 PV-3 103104105 2 4 6 8 10 12 14 16 10-3 PV-4 103104105 0 0.005 0.01 0.015 0.02 0.025 PV-5 103104105 0 0.005 0.01 0.015 0.02 0.025 0.03 PV-6 103104105 0 0.005 0.01 0.015 0.02 0.025 PV-7 Figure 2.4: Magnitude of PV-output DC components VDC for all PV cells in terms of the varying lightmodulating frequencies, at different distances between the LED and PV cell. frequency is fc=1 2πRpvCpv ,(2.2) and the experimentally extracted f3dB values follow this proportionality. Four-cell modules reach f3dB ≈20 kHz because the total junction capacitance is the smallest. Nine-cell devices attain 14–16 kHz, while twelve-cell strings exhibit the lowest bandwidth, from 2 kHz for the smallest cell area to 17 kHz for the largest area. Reducing the distance between the LED and PV cell from 0.5 m to 0.2 m raises the incident photon flux. It can be seen that VAC scales almost linearly with distance. For comparison, the benchmark photodetector produces a flat response up to 40 kHz and then rolls off, indicating that the experimental apparatus have small effect on the observed PV bandwidth. Fig. 2.3 compares the AC output of all tested PV cells at each LED-PV distance d. At d= 0.5 m, the midband amplitudes span approximately 0.5–3.7 mV, with PV-1 (four-cell) leading and PV-2 (twelve-cell smallest area) trailing. As dis reduced to 0.4 m and 0.3 m, the overall envelope of VAC increases in proportion to irradiance, yet the ranking of devices remains unchanged: four-cell modules always exhibit the highest amplitudes and twelve-cell modules the lowest, with intermediate performance from the nine-cell device. At the closest distance d= 0.2 m the peak AC outputs approach 15 mV for PV-1 and 12 mV for PV-7 (nine-cell), while the smallest twelve-cell cell remains below 6 mV. Across all distances, the peak response occurs between 3 kHz and 6 kHz, after which a monotonic roll-off is observed. This consistency confirms that device-intrinsic parameters, namely junction capacitance and series resistance, dictate the midband amplitude and high-frequency attenuation, whereas the optical intensity merely scales the magnitude of VAC. DC Response Fig. 2.4 shows that the output DC voltage VDC of each PV module is nearly invariant with modulation frequency. Across the entire 1–100 kHz sweep, the variation remains below five millivolts, much smaller than the step caused by distance. This confirms that energyPage 10 of 81
D2.2 - Initial Report on END Hardware Design 103104105 0.5 1 1.5 2 2.5 3 3.5 4 4.5 10-3 103104105 1 2 3 4 5 6 710-3 103104105 3 4 5 6 7 8 9 10 11 12 10-3 103104105 0.005 0.01 0.015 0.02 0.025 0.03 Figure 2.5: Comparison of the PV-output DC components across all tested PV cells, for different distances (0.5 m,0.4 m,0.3 m, and 0.2 m) between the LED and PV cells. 1234567 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 d=0.5 d=0.4 d=0.3 d=0.2 Figure 2.6: Measured 3-dB bandwidth of PV cells at different distances between the LED and PV cells. harvesting capability is rarely penalized by data modulation in the considered range. Furthermore, the output DC levels separate the cells into three groups. Four-cell devices harvest the highest voltage (up to 30 mV at 0.2 m) because each unit cell is large and the series string is short. Nine-cell devices harvest moderately less, whereas twelve-cell devices span a wide range that correlates strongly with the unit-cell area. The hierarchy is preserved over distance, proving that geometry rather than intensity governs the ordering. The PD yields a DC voltage that sits mid-range because it is optimized for responsivity rather than power extraction. Fig. 2.5 compares the DC output of all tested PV cells at different distances. In every subplot, VDC is almost invariant with modulation frequency for most of PV cells, confirming that carrier modulation up to 100 kHz does not perturb the harvested DC. At d= 0.5 m the DC voltages span approximately 0.8–3.1 mV, reflecting the series-cell count and unit-cell area. As d decreases, all curves shift upward in proportion to the incident irradiance (approximately 1/d2), yet their relative ordering remains unchanged: four-cell modules yield the highest VDC (e.g., for PV-6: VDC ≈30 mV at 0.2 m), nine-cell modules are next, and twelve-cell strings deliver the lowest voltages. This consistency demonstrates that EH is governed by PV cell geometry and illumination level rather than by the presence of high-frequency light modulation. Bandwidth Fig. 2.6 presents the measured 3-dB bandwidth f3dB of PV cells 1 through 7 at four LED-PV distances d= 0.5 m,0.4 m,0.3 m,and 0.2 m, respectively. At the lowest irradiance (d= 0.5 m), the smallest PV cell (PV-1) exhibits a modest bandwidth of approximately 1.2 kHz, Page 11 of 81
D2.2 - Initial Report on END Hardware Design 100 120 140 160 180 200 220 240 260 280 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2104 Cubic: y = 0.0149*x3 - 9.212*x2 + 1857*x - 1.066e+05 Measurement cubic curve fitting Figure 2.7: Measured 3-dB bandwidth of PV cells versus vendor specified maximum output power Pmax. while larger cells reach between 9 kHz and 16.5 kHz. As the distance is reduced to d= 0.2 m, the bandwidth of PV-1 increases markedly to about 13 kHz, whereas cells 2-7 show only minor changes (±1.5 kHz), indicating that above a threshold photon flux their intrinsic capacitance and series resistance govern the cutoff frequency. Among the larger-area modules, the nine-cell device (PV-7) and the largest twelve-cell module (PV-5) achieve the highest bandwidths (approx. 16–17 kHz), mid-sized twelve-cell modules (PV-3 and PV-4) reach 13–15 kHz, and the smallest twelve-cell module (PV-2) remains below 10 kHz. This ordering reflects the inverse relationship between total junction capacitance and achievable bandwidth. The data imply that PV-7 or PV-5 should be selected for SLIPT purpose requiring symbol rates above 15 kHz, whereas PV-2 or PV-6 may sufficient when EH is prioritized over data rate. The manufacturer specifies a maximum power Pmax under one-sun illumination for each PV device. When the experimental f3dB values are plotted against Pmax, as shown in Fig. 2.7, where a monotonic increase is visible. A cubic fit captures the empirical trend by f3dB = 0.015 P3 max − 9.21 P2 max + 1.86 ×103Pmax −1.07 ×105, where f3dB is in hertz and Pmax in milliwatts. The positive slope confirms that higher-power cells, which possess larger active areas and lower series resistances, inherently support wider communication bandwidth. Several insights are obtained from the above characterization measurements. The AC response of each PV module exhibits an RC-limited band-pass profile with peaks between 3 kHz and 8 kHz, indicating that the electronic time constant rather than the optical path governs the bandwidth. The measured 3-dB bandwidth is remarkably stable across a four-fold variation in irradiance (LED-PV distance from 0.5 m to 0.2 m), changing by no more than ten percent. Bandwidth is found to increase systematically with both reduced series cell count and higher manufacturerreported Pmax, and the empirical cubic fit derived herein provides a first-order predictor for untested modules. In parallel, the DC output remains unaffected by modulation frequencies up to 100 kHz, confirming that EH and data transfer can coexist without mutual degradation. These findings imply that no single PV module optimizes all performance metrics simultaneously; instead, selection should be guided by the specific data-rate and energy-harvesting requirements of the target application. Page 12 of 81
D2.2 - Initial Report on END Hardware Design Heat source Ceramic plate Interconnect n-type p-type Interconnect Interconnect Interconnect n-type p-type Interconnect Interconnect Interconnect n-type p-type Interconnect Interconnect Interconnect n-type p-type Interconnect Interconnect Ceramic plate Heat sink (cold source) RL Vout Iout Figure 2.8: Basic structure of a TEG, showing four thermocouples connected electrically in series and thermally in parallel. A temperature gradient across the device induces a voltage via the Seebeck effect, causing a current to flow through the load resistor RL. 2.1.2 Fundamentals of Thermoelectric Generators for Thermal Energy Harvesting Thermal gradients present in industrial, environmental, or wearable contexts can be exploited as a continuous power source for low-power devices. 2.1.2.1 Working Principle of TEGs TEGs enable direct conversion of heat into electricity, leveraging the intrinsic transport properties of thermoelectric materials: thermal conductivity, electrical conductivity, and Seebeck coefficient. These materials can convert thermal energy into electrical energy and vice versa, depending on the direction of energy flow. •When electrical energy is converted into thermal energy, the phenomenon is called the Peltier effect, with applications in solid state heating and cooling. •When thermal energy is converted into electrical energy, the phenomenon is called the Seebeck effect, which is the primary mechanism exploited in the harvesting of thermoelectric energy. This effect will be the focus of this section. The Seebeck effect occurs when a temperature difference across a conductor or semiconductor induces a voltage between its ends. This voltage arises from the diffusion of charge carriers (electrons in n-type materials, holes in p-type materials) in response to the thermal gradient. The basic structure of a TEG is shown in Figure 2.8 [19]. A TEG consists of multiple thermocouples, each made up of a pair of n-type and p-type thermoelectric legs. These legs are connected electrically in series and thermally in parallel. The p-type leg has a positive Seebeck coefficient, while the n-type leg has a negative one. When these legs are joined by a conducting strip (typically copper), a voltage is developed across the device terminals. Connecting an electrical load to these terminals allows current to flow, thereby generating electrical power. ATEG continues to produce DC power as long as a temperature difference ∆T=Thot −Tcold is maintained across its surfaces. A higher ∆Tresults in a higher output voltage and thus power. Page 13 of 81
D2.2 - Initial Report on END Hardware Design in advance. The prediction-based approach has the benefit of providing to the END the time needed to activate the transceiver supporting the RAT that is both available in the geographical area of focus and, at the same time, less energy-consuming, taking into account the predicted EH opportunities in space and in time. In that case, the adaptation steps will be as follows: (i) the END or a network infrastructure entity (e.g., a prediction function within a MEC host) predicts ambient energy availability (across RATs regarding RF-EH) in the area and time of interest of the requesting END, (ii) the END monitors its current energy storage level, (iii) if both the predicted EH potential and current END energy storage level are low, the END switches to a low-power RAT, such as LoRa. On the contrary, if the predicted EH potential is high and current END energy storage level are sufficient, the END will switch to a higherthroughput RAT, such as NB-IoT. Figure 2.13 illustrates the proposed adaptive approach by means of activating/ deactivating END communication components. 2.2 Radio Frequency Wireless Power Transfer Radio frequency wireless power transmission (RF WPT) is a promising solution for energysustainable wireless networks, enabling continuous operation of ENDs without battery replacement or wired charging. Its end-to-end (E2E) efficiency, however, is constrained by propagation losses, rectifier non-linearities, and hardware limitations at both the transmitter and receiver sides. On the transmitter side, advanced energy transmitter (ET) architectures can deliver highly directive beams, extend coverage, and reduce hardware cost and complexity. On the receiver side, concurrent RF EH from both dedicated and ambient signals requires flexible circuit designs capable of handling wide input power ranges while maintaining high conversion efficiency. In addition, energy delivery through unmanned aerial vehicles (UAVs) offers a complementary solution for powering remote or hard-to-reach devices. Together, these advances are paving the way toward efficient, scalable, and adaptive RF WPT networks. 2.2.1 Energy Transmitter Architectures Physically large aperture arrays have become an attractive solution to boost power transfer end-to-end conversion efficiency and enable precise energy beam control in both depth and angular domains [32]. Fig. 2.14 illustrates emerging architectures based on lens antenna arrays, dynamic metasurface antennas, reflective intelligent surface (RIS)-equipped ET,movable antennas (MAs), and distributed antenna systems (DAS). These hold great potential to reduce the number of RF chains and minimize power consumption, complexity, and cost. For instance, lens arrays apply spatially varying phase shifts across their aperture to focus incoming electromagnetic waves from different angles onto distinct ports [33]–[35]. It transforms the spatial multiple-input multiple-output (MIMO) channel into a sparse beamspace representation, which therefore requires fewer active RF chains [36]. Moreover, true-time-delay lens architectures, such as Rotman lenses, inherently provide passive retrodirectivity by reversing the phase profile of an incoming signal without requiring active phase shifters or complex feed networks [37]. Notably, their discrete implementation can limit angle-of-arrival (AOA) estimation accuracy, which can be improved by exploiting phase differences across adjacent elements [38]. RIS-equipped transmitters comprised a collocated digital fed beamformer illuminating a reconfigurable aperture that transmits and reflects a phase-shifted version of the impinging signal [39]. It replaces the traditional analog network with an air interface. The added spatial flexibility of RIS can be Page 20 of 81
D2.2 - Initial Report on END Hardware Design DAS MAs RF chains dynamic metasurface antennas RIS feeder lens antenna Figure 2.14: ET architectures with physically large aperture arrays. leveraged to implement both single- [40] and multi-RF chains [41]ETs. These advantages can be further exploited through joint waveform-beamforming co-design [42] and even utilize more advanced apertures based on beyond diagonal RIS [43], [44]. Finally, dynamic metasurface antennas consist of densely packed, sub-wavelength-spaced reconfigurable metamaterial elements that, through individual tuning, perform analog signal-processing functions directly in the aperture, eliminating the need for dedicated circuitry. Beamforming design for such ETs has been investigated for both maximizing harvested energy [45] and minimizing power consumption in signal transmission [46], [47], while joint beamforming and waveform optimization is studied in [48]. Movable antennas is an alternative ET architecture, wherein radiation patterns are modified via large-scale physical array reconfiguration. Their value lies in that they do not need a large number of antennas but rather a reduced set of them that can transmit from different points of the transmitter aperture. In the context of power transfer, the spatial flexibility of MAs can help mitigating the double near-far problem [49], boosting the received power [50], and supporting diverse service requirements [51]. Notably, the varying aperture of MAs to adjust to the varying channel conditions/active ENDs positions can impact near/far-field operation [52]. Finally, these degrees of freedom can be added to traditional antenna arrays to boost the received power in non-line-of-sight (NLoS) scenarios with strict electromagnetic radiation constraints [53]. Finally, DAS coordinates the transmissions of multiple radio heads implemented as low-complexity ETs, which can be deployed either as conventional arrays or through the aforementioned scalable architectures, as shown in Fig. 2.14. Convenient deployment and beamforming optimization of DAS extends power transfer coverage by reducing the charging distance and avoiding blockage in the service area [54], [55]. Practical implementations of DAS, such as the 280-antenna testbed in [56], have facilitated real-world evaluation of synchronization requirements [57], infrastructure architectures [58], and waveform and protocol design [59]. Page 21 of 81
D2.2 - Initial Report on END Hardware Design 𝑓 1 𝑓2 𝑓N+ Figure 2.15: Advanced RF EH architectures based on (i) tunable (left), (ii) wide range dynamic range (center), and (iii) hybrid RF-DC combining (right) rectifiers. 2.2.2 Energy Harvesting Circuits Design for Radio Frequency Wireless Power Transmission By enabling a dual RF EH mode, ENDs can prioritize RF EH from ambient transmissions and only request dedicated power transfer services if needed; thereby, saving energy at the ETs. In fact, ambient RF EH can be seen as a backup, if available, to meet minimum service guarantees when power transfer momentarily degrades. Note that in both cases ENDs can benefit from highly reliable operation while saving energy at the ETs. However, concurrent RF EH from ambient and dedicated transmissions requires flexible circuit designs to account for the different demands on receiver operation. As shown in Fig. 2.15 (left and center), such advance control can be enabled by machine learning (ML)-driven central-processing units (CPUs) to intelligently switch between ambient and dedicated RF EH, optimizing overall performance [60]. Key differences of both services include channel state information (CSI) availability, directionality and timing of energy arrival, operating frequency, and input power range at the receiver [61], [62]. These requirements directly influence the antenna design [63], e.g., directivity, gain, and polarization, which may differ substantially from existing antenna technologies for RF EH circuits [64]. Moreover, to handle energy density variations over a wide spectrum one can resort to multi-band [65] or, more advanced, tunable [66]RF EH circuits, as shown in Fig. 2.15 (left). Boosting conversion efficiency while providing a broader beamwidth to harvest energy from multiple directions can be achieved with RF EH circuits comprising signal combiners in both RF and direct current domains, as shown in Fig. 2.15 (center). Notably, minimizing energy consumption during precoder optimization is crucial and has been tested through methods such as brute force, sequential testing, and codebook-based strategies [53]. Received power levels significantly vary between dedicated and ambient power transmissions, necessitating rectifier designs capable of operating efficiently across a broad dynamic range. To address this challenge, one can equip the RF EH circuit with multiple rectifiers, as shown in Fig. 2.15 (right), each optimized for a reduced set of input power [67]. Note that such an operation can be achieved with low-complexity distribution networks which automatically route the incoming RF signal to the optimal rectification path without requiring external control circuits [68]. Page 22 of 81
D2.2 - Initial Report on END Hardware Design 2.2.3 Unmanned Aerial Vehicle-Based Wireless Power Transfer to Remote Energy-Neutral Devices Many IoT devices operate in remote or harsh environments (e.g., forests, below bridges, underground locations), where EH from ambient sources is not always feasible or sufficient. Replacing or manually recharging batteries is labor-intensive and costly. UAVs can autonomously reach these devices, hover nearby, and wirelessly deliver energy, enabling continuous operation without human intervention. Therefore, energy delivery from a nearby mobile energy source, such as an UAV, offers a promising solution to overcome power limitations in the field [69]–[71]. The transfer from the UAV to the END can be supported by various WPT techniques [72]. In the literature, mainly RF systems (often combined with data transmission) and inductive systems have been investigated. Depending on the END energy requirements, which in turn depend on the related device class (as defined in D2.1), RF WPT may be preferred over inductive power transfer (IPT) or magnetic resonance coupling (MRC) systems. It is assumed that the UAV will often need to hover. This allows the distance between the transmitter and receiver to be reduced, thereby limiting energy losses in the WPT link. Given the power consumption of a hovering UAV, energy transfer should therefore take place as quickly as possible. The power density of RF-based systems is generally lower due to their low efficiency and the limitations on radiated power imposed by regulations such as those from European Telecommunications Standards Institute (ETSI). For small ENDs (class 1 devices), RF WPT can be considered a suitable transfer technology. However, when larger amounts of energy need to be transferred in a short time, a coupled WPT system is preferred due to its higher power density. Recent research has further examined magnetic resonance coupling (MRC) technology implemented on an UAV. [73] investigated the number of required UAV visits based on the size of the energy buffer and charging rate (C-rate), the daily energy consumption of the IoT node, and the power density of the WPT link. The overall energy efficiency for a specific UAV model was analyzed in [74]. Specifically, the UAV’s energy consumption for charging the IoT energy buffer and flying to his location compared to the stored energy in the IoT energy buffer. The implementation of a UAV with a proof of concept implementation was studied in [75]. 2.2.3.1 MRC Technology for Deployment on UAVs MRC is a wireless power transfer technique that uses magnetic fields and resonance to transfer energy between two coils over longer distances compared to IPT. Both the transmitter and receiver coils are tuned to the same resonant frequency using capacitors. When both sides resonate at the same frequency, energy transfer becomes more efficient, even if the coils are not perfectly aligned or are spaced farther apart than in standard inductive coupling. The mutual coupling factor is typically below 0.1. Due to this low coupling factor, the self-inductance of each individual coil changes only slightly, resulting in negligible impact on the resonance circuits. MRC still relies on magnetic induction as the underlying physical principle, but resonance enables looser coupling and longer transfer distances without a significant drop in efficiency. The inverter frequency on the transmitter side should remain constant, while the inverter voltage can be adjusted during energy transfer. This allows the efficiency of the entire WPT link to be controlled and optimized, especially under variable load conditions. On the UAV side, an MRC transmitter is required, consisting of a pre-regulator, inverter, and LC tank. The implementation on the UAV could be realized as illustrated in Figure 2.16. The Page 23 of 81
D2.2 - Initial Report on END Hardware Design Energy buffer IoTDevice UAV MRCreceiver Receivercoil Transmittercoil Flightcontroller Pre-regulator BrushlessDCmotor Bridgeinverter UAVbattery RCreceiver Figure 2.16: Render of an autonomous IoT charging system. The UAV battery powers the flight controller and pre-regulator, which drives the half-bridge inverter connected to the transmission coil. The receiver coil captures the magnetic field to charge the IoT energy storage [75]. details of the transmitter side are not further elaborated in this discussion, and the reader is referred to [75]. 2.2.3.2 MRC Receiver on the IoT Device Side This section focuses on the additional components required to make an END or IoT device compatible with the proposed UAV-based approach. This principle is primarily relevant for class3 devices with the presence of long-term rechargeable energy storage, such as the examples from D2.1. In addition to the application microcontroller and the sensors/actuators specific to the considered application, other components are needed to convert the alternating magnetic fields emitted by the UAV into energy. The components required for the energy receiver, as shown in Figure 2.17, include an LC tank tuned to the same frequency as the transmitter, a rectifier, and a wake-up and voltage detection circuit that brings the application microcontroller out of sleep or power-down mode via an interrupt pin. A protection circuit is included to disconnect the supply in case of high voltage spikes, while the energy buffer is charged using an efficient switched mode power supply (SMPS). If constant-current charging is required, a current sense circuit can be incorporated. Finally, a bidirectional load switch prevents reverse current from flowing from the energy buffer to the charging circuit. 2.3 Power Management Circuits This section introduces several power management circuits, such as EH from PV cells, EH from thermal sources and EH from multiple sources. The power management unit (PMU) forms the crucial part of a batteryless IoT design by ensuring regulated and managed power output to the device from an often unstable ambient energy source. Moreover, the energy balance monitoring and voltage limiting techniques for ENDs are also discussed. Page 24 of 81
D2.2 - Initial Report on END Hardware Design − +Energy buffer LC tank Rectifier FET Driver Digi. Pot. I2C GPIO1 ADC1 GPIO2 GPIO3 ADC2 Wake up Voltage sense Protection Switched mode power supply Current sense Bidirectional load switch Figure 2.17: Schematic representation of the MRC WPT receiver. GPIO1, GPIO2, and GPIO3 correspond to the wake-up input, protection-enable output, and buck-enable output, respectively. ADC1 and ADC2 are used to measure the input voltage and output current. The microcontroller unit (MCU) and IoT-specific components, such as sensors or actuators, are not shown, as these depend on the specific use case [75]. 2.3.1 Energy-Neutral Device Energy Harvesting from a Photovoltaic Cell Using PV cells to harvest energy shows a promising way to empower ENDs, which has been discussed in Section 2.1.1. The simplest way to harvest power from a PV cell is to connect the panel to a capacitor with a diode. The diode prevents current flow from the capacitor to the panel when the light is low. Even with a large panel, the output voltage will be low in low-light situations. A boost converter would be necessary to raise the voltage to a usable level. With a large panel, the maximum output voltage will be high, and either the storage must be able to handle the voltage or a converter is needed to lower the voltage to the storage level. Another way to harvest power from a solar panel is to use a harvester circuit that combines maximum power point tracking (MPPT) with the boost converter. MPPT can be implemented for example by measuring the open circuit voltage periodically and adjusting the harvester input voltage to a fraction of this, or by making small changes to the input voltage and measuring the change in the current. In situations when only a low level of light is available, the output voltage from a low-power solar panel can be less than 1 V. The harvester circuit must be able to operate in such situations. On the other hand, the maximum input voltage of the harvester sets an upper limit on the solar panel output voltage. The harvester circuit should also implement some form of MPPT. Voltage in the energy storage varies and can be over the voltage limits of the system components. Some form of voltage regulator is needed. A switching regulator has lower losses than a linear regulator, but requires an external inductor and may cause noise in other components. Integrating the regulator into the harvester simplifies the system. In addition, a primary battery can be used as a backup power source to keep the device operational for extended low periods of insufficient light. The harvester should be able to switch between harvested power and the backup automatically. As an example, the power management circuit of an END, Li2BC [76], is discussed in the following paragraphs. 2.3.1.1 Energy Harvester Circuit Energy harvester chosen for analysis in this section is Analog Devices ADP5091 [77], which is a part of the aforementioned Li2BC device. It collects energy from a PV cell using maximum power point tracking and uses the energy to charge the energy storage capacitors. It can also Page 25 of 81
D2.2 - Initial Report on END Hardware Design Figure 2.18: ADP5091 functional block diagram [77]. use a backup battery to provide power in case the storage voltage drops to an insufficient level. The operation of the harvester depends on the energy storage voltage VBAT. To prevent energy storage overvoltage, the charging is terminated when VBAT reaches the voltage VBAT_TERM. The storage capacitors are disconnected when VBAT is below the stop discharge voltage VSETSD. The supercapacitors are not damaged by undervoltage but at this storage voltage level the regulated output drops below the microcontroller requirement. Disconnecting the capacitors minimizes the leakage. The set backup voltage VSETBK is the storage voltage level that determines if the backup battery is used to power the device. The set power good VSETPG sets the voltage threshold to enable the microcontroller power. The VSETBK and VSETPG voltages have hysteresis. The actual threshold changes depending on if the storage voltage VBAT is rising (↑) or falling (↓). Voltage thresholds are collected in table 2.4. The thresholds have some variance that depends on the harvester’s internal voltage reference. The thresholds are set with external resistors. The harvester has a low-light indicator that can be used to indicate to the microcontroller when sufficient light for EH is not available. The harvester boost converter can also be disabled to reduce noise, for example in the case of sensitive measurements. 2.3.1.2 Energy Storage An END must have sufficient energy storage to be able to run in low-light situations, when the PV cells do not produce energy. The amount of energy storage and power consumption sets the device run time without external energy. Supercapacitors have significant internal leakage, which must be taken into account when implementing energy storage. The total stored charge in the capacitors is Q=nCV∆,(2.4) Page 26 of 81
D2.2 - Initial Report on END Hardware Design Table 2.4: Controller voltage thresholds. Threshold Reference Typical/VRange/V VBAT_TERM VBAT ↑5.04 4.77–5.31 VSETSD VBAT ↓2.63 2.48–2.77 VSETBK VBAT ↑2.98 2.81–3.14 VSETBK VBAT ↓3.12 2.94–3.28 VSETPG VBAT ↑3.04 2.87–3.20 VSETPG VBAT ↓2.78 2.62–2.93 where nis the number of capacitors, Cis the capacitance of a single capacitor, and V∆is the voltage over the capacitors. The current consumption Iof on END has three main parts: The working current Iwork, including the microcontroller, sensors, and communication, losses in the system Iloss, and internal leakage of the capacitors Ileak. The capacitor discharge time tis related to the charge with Q=tI. (2.5) Thus, when powered solely by the energy storage, the run time is then trun =Q I=nCV∆ Iwork +Iloss +Ileak (2.6) Supercapacitors with capacitance C= 0.1 F from Kemet FG series [78] were chosen for energy storage. 5 V supercapacitors from the series have voltage-holding characteristic of 4.2 V, or voltage drop V∆= 0.8 V over ttest = 24 h. Based on that, the internal leakage current Ilcan be estimated to be Ileak =CV∆ ttest = 0.9µA.(2.7) The fully charged voltage of the supercapacitors is 5.0 V. The microcontroller input and output voltage is equal to the supercapacitor voltage minus the LDO voltage drop (0.3 V). In order to keep the input-output voltage compatible with external digital components, it should be at least 2.5 V. Therefore, the lowest working voltage of the supercapacitors is 3.0 V, and then V∆= 2.0 V. On tests done with NUCLEO-L55ZEQ demo board, the working current Iwork was estimated to be 4µA. In addition, Iloss = 1 µAis estimated to be wasted on various component losses. Based on these numbers, the run time with different numbers of capacitors can be estimated using Equation (2.6). Table 2.5 shows the estimated run times. It can be noted that the run time does not increase linearly with the number of capacitors. As the number of capacitors is increased, the fraction of the work current from each capacitor diminishes. With these assumptions about the currents, with more than 5 capacitors the self discharge current is greater than actual work current per each capacitor, and additional capacitors do not significantly increase the run time. Page 27 of 81
D2.2 - Initial Report on END Hardware Design Table 2.5: Estimated runtime in hours with different number of supercapacitors. Capacitors 1 2 3 4 5 6 t/h9.4 16.3 21.6 25.8 29.2 32.0 2.3.2 Power Management Circuits for Thermal Energy Harvesting Similarly to solar energy harvesters, thermal generators require dedicated circuit techniques to operate efficiently. The low voltages and relatively high source impedance of TEGs make impedance matching and cold-start capability essential design considerations. Several power management integrated circuits (PMICs) have been developed specifically to address these challenges, as illustrated by the following examples. •Analog Devices LTC3108 [79], [80]– This ultra-low voltage step-up converter supports input voltages as low as 20 mV through the use of an external step-up transformer, enabling the operation of a TEG with temperature differences as small as 1◦C. The transformer provides impedance transformation, allowing the device to draw power efficiently from high-impedance sources. The LTC3108 is designed to present an effective input resistance in the range of 2–10 Ω, depending on the transformer turn ratio and the input voltage. As the input voltage decreases, the converter’s input resistance increases, which enables approximate impedance matching with the internal resistance of typical TEGs, thereby improving energy transfer efficiency under varying thermal conditions. •E-peas AEM20940 [81], [82] – This PMIC supports input voltages as low as 50 mV, with a cold-start capability from 50–380 mV, depending on the available input power. It integrates an adaptive MPPT algorithm that periodically samples the input voltage to dynamically match the source impedance. Unlike transformer-based solutions, the AEM20940 employs a fully integrated boost converter, which simplifies system integration and reduces overall size. The device features dual-regulated outputs and directly supports energy storage in supercapacitors or rechargeable batteries and includes autonomous load management based on state of charge (SoC). 2.3.3 Multisource Power Management Units Over the last decades, the state-of-the-art of PMU has substantially improved, both in academic and industrial research. There are numerous off-the-shelf PMUs available in the market, specifically tailored to various EH applications and use cases. However, a majority of the development on PMUs has been focused on a single source PMU that can harness energy from one source at a time. The nature of the deployment environment often determines the choice of the energy source of an ambient-powered device. In many situations, there can be more than one source available. However, the lack of PMUs that can accommodate more than one source at a time limits the exploitation of multiple sources for EH. A multisource PMU must simultaneously harvest energy from multiple sources without causing reverse energy flow between the sources and charge a storage element. However, this is not a trivial task, mainly due to the wide range of electrical characteristics possessed by different energy harvesters (Section 2.1). Therefore, harvesting energy from these varieties of energy sources implies the front end of the PMU should be capable of seamlessly tuning its electrical characteristics. Since the energy available to coldPage 28 of 81
D2.2 - Initial Report on END Hardware Design Source 1 Source 2 V = S1||S2 (a) Source Controller Switching Regualtor Source 1 Source 2 C VCOMB (b) Source Controller Switching Regualtor Source 1 Source 2 LVCOMB (c) Figure 2.19: Different energy combining topologies (a) diodes used to select the strongest source (b) capacitor or the (c) of a switching regulator shared and multiplexed between multiple sources. start a PMU is typically very limited, adding such luxuries can impair its overall performance if not carefully managed. A naive approach to multisource harvesting is to use an energy-OR circuit to add the sources together (Fig. 2.19). The energy OR-ing is the most basic and easy-to-use circuit in which diodes or MOSFETs provide isolation between the input sources [83]–[85]. However, such circuits provide the least efficiency since only the highest energy source is utilised at any time due to the diode/FET’s OR-ing property. A second approach is to use time-multiplexing circuits at the front end, which provides access to the source in a specific window. Time multiplexing circuits can be implemented using capacitors with voltage thresholds to transfer and combine energy from multiple sources [86]–[88]. These circuits overcome the limitation of energy OR-ing circuits by sequentially transferring energy from one source to the storage buffer based on voltage thresholds. Though they have superior performance over OR-ing circuits, they fail to support simultaneous EH. Additionally, their reliance on capacitors for energy transfer can impact the efficiency when using supercapacitors or capacitors as the main storage buffer due to the energy lost in energy transfer between capacitors. Another set of multisource front-ends uses the time multiplexing concept, but with a shared inductor between sources to combine energy [89]–[91]. The sources are sequentially connected to the inductor of a boost converter to enable energy transfer. On the industrial side, PMUs for EH have matured significantly over the past decade. Today, there exists a wide variety of PMUs, each optimised for different sources, power levels, and application requirements. Despite this progress, most commercially available EH ICs still manage only one source at a time. While many support different input types (e.g., solar cells, piezoelectric harvesters, or thermoelectric generators), they typically allow only a single source to be connected in a given configuration. For instance, PMU such as AKM AP4413, Texas Instruments BQ25570, or Analog Devices ADP5090 are optimized for single-source input and are highly efficient for PV or TEG operation. Nexperia’s NEH71×0 chips bring novelty to the market with their inductor-less architecture, which reduces footprint and bill of materials (BoM) cost. Though fabless companies such as e-peas and Trameto have made attempts at multisource PMUs, the field of multisource PMU itself remains in its early stages and is yet to reach Page 29 of 81
D2.2 - Initial Report on END Hardware Design Chapter 3 Low-power IoT Devices, Backscatter Devices, and Wake-up Radio Design This chapter examines END platforms for low-power IoT,backscatter device (BD), and wake-up radio designs, which succeeds the discussion of the proposed END concept in the deliverable D2.1. It links platform selection, sensing profiles, and compute needs to measured energy budgets and realistic duty cycles. We summarize criteria for microcontrollers, memory, clocks, and peripherals with deterministic wake and sleep behavior, together with firmware hooks for energy-aware scheduling. For backscatter, we outline reader and tag building blocks, describe impedance modulation and radar cross section, and frame range, throughput, and inventory latency as functions of reader density and regulations. For wake-up radios, we present front ends, correlators, and trigger logic that reduce idle listening while bounding false alarm probability and response time, with simple coupling to the main radio. 3.1 Low-power Internet of Things Device Prototypes In this section, two low-power IoT prototypes are introduced, including the Digital Enhanced Cordless Telecommunications (DECT)-2020 device and cellular IoT device. 3.1.1 DECT-2020 Device 3.1.1.1 DECT-2020 NR Overview DECT-2020 NR [98] is a noncellular 5G standard developed by ETSI that supports ultra-reliable low-latency communications (URLLC) and massive machine-typed communication (mMTC) in IoT applications. Using DECT-2020, devices can transmit device-to-device or broadcast to all devices in range. In a multi-device network, a single DECT-2020 device acts as a base station, allocating radio resources to other devices in the network. The base station acts as a gateway to other networks. DECT-2020 also supports the mesh network topology, where the base stations of different networks communicate with each other. DECT-2020 operates on a dedicated licence-exempt frequency band around 1.9 GHz, with actual frequencies depending on the market. It uses frequency division for network multiplexing and Page 36 of 81
D2.2 - Initial Report on END Hardware Design Figure 3.1: Left: Aalto DM51 DECT-2020 Module. Right: Aalto DM51 as Raspberry Pi add-on board. time division for device access multiplexing. DECT-2020 standard specifies the transmit power levels from −40 dBm to 32 dBm to optimise range, interference, and power consumption. 3.1.1.2 Aalto DM51 DM51 (DECT Module with nRF9151) shown in Figure 3.1, is a 65 mm by 50 mm module built around the Nordic Semiconductor nRF9151 [99]system-in-package (SiP). The module includes the nRF9151 SiP, a ceramic wideband antenna, and a Nordic Semiconductor nPM1300 [100] PMIC. There is also a provision for the use of an external antenna. The module can operate stand-alone, for example as a sensor node, or as an add-on module connected to a host computer such as Raspberry Pi. In standalone operation, the module is powered by a lithium polymer (LiPo) battery or via the USB port. The host provides power when the module is used as an add-on. The nRF9151 SiP integrates ARM Cortex-M33 processor, 1 MB flash read-only memory (ROM), 256 kB random-access memory (RAM), a multi-protocol modem (Long Term Evolution (LTE), DECT,global navigation satellite system (GNSS)), and various input/output (IO) interfaces. The universal serial bus (USB) port is used for module software updates and as a serial port. Seven SiP IO lines are routed to an expansion connector in the module, and their functionality can be adjusted with software. The nPM1300 PMIC handles the LiPo battery charging and power supplies for the rest of the module. Figure 3.2 shows the nRF9151 SiP power consumption when transmitting a DECT broadcast message with 0 dBm transmit power and 3.0 V supply voltage. The mean power consumption during a transmission is 234 mW, and approximately 100 mW when idle between transmissions. As of this time, low power operation of the module is not yet characterized. However, manufacturer specifications [100] can be used for estimations. In the nRF9151 SiP, the modem is the highest consuming component using 34 mA from 3.7 V supply when idle, and more when transmitting. The CPU core consumes 2.1 mA running a stress test application, and the CPU consumption is down to 2µAin a sleep mode. This indicates that for low power operation the modem should be turned off most of the time. Page 37 of 81
D2.2 - Initial Report on END Hardware Design 0 200 400 600 800 0 2 4 Time (µs) Trigger (V) 0 200 400 600 800 0 40 80 120 Time (µs) Current (mA) Figure 3.2: nRF9151 SiP current consumption when transmitting a DECT message. 3.1.2 Cellular Internet of Things Device Cellular IoT devices often transmit and receive relatively small amounts of user data. For example, a smart meter unit will typically transmit a few hundred bytes of information per day. Between transmissions, these devices may enter a so-called power saving mode (PSM), where the terminal remains registered with the network but stops listening entirely for typically several hours. Most devices are configured to go to sleep mode for extended periods and wake up periodically, typically every few minutes, to listen for paging messages. For these devices, manufacturers still target battery lifetimes of 10-15 years. Finally, there are devices that need to be reachable at short notice, for example, to receive voice calls. These will negotiate sleep periods of no more than a few seconds with the network. Considering the emerging NR-based cellular IoT devices, the lowest power NR eRedCap devices like smart meters will likely be powered from a battery, specifically an A-size lithium thionyl chloride (LTC) battery which has exceptional voltage stability across a wide temperature range, a near-flat discharge profile, as well as low self-discharge, and can provide a storage capacity of about 3.6 Ah at 3.65 V (equivalent to 47 kJ of energy). To last 15 years, the device must consume less than 9 Joule per day or below 100 µW averaged in time. For cellular battery-based A-IoT devices, an ambitious target will be to power a device from a much smaller battery, e.g., a CR2032 coin battery. However, overall energy storage will be considerably lower (about 2.4 kJ considering 225 mAh capacity at 3 V). The lithium manganese oxide chemistry used provides a lower voltage, especially at low temperature, and lower peak currents. This section will first describe the impact of the various memory/storage options on the device design. Then, considering this memory impact and the different operational modes, we identify Low Power Management (LPM) strategies for power saving and we identify potential relaxations at receiver design. 3.1.2.1 Memory/Storage Options Any cellular IoT device will have to store code for controlling the device hardware and run the cellular protocol stack. It will also store and process data, for example, user TCP/IP packets, voice samples, or encoded IQ samples. Page 38 of 81
D2.2 - Initial Report on END Hardware Design Table 3.1: Memory options comparison Parameter External NOR External DRAM Internal SRAM Unit Data retention Non-volatile Volatile Volatile – Interface QSPI, DDR X8 SPI, DDR 32-bit, internal – Cost (excl. package) 3 2 12 ¢/MB Clock speed 100 250 500 MHz Write speed 2 500 2,000 MB/s Read speed 100 500 2,000 MB/s Write energy 7,000 60 3 µJ MB−1 Read energy 200 60 3 µJ MB−1 Retention power 0 5 200 µW MB−1 In a typical architecture, there are three principal choices for storing code and data. Firstly, there is flash memory (NOR or NAND flash), which is non-volatile and can store information for years without consuming power. However, access speed is low (especially for write) and the number of erase/program operations is typically limited to 100k cycles over the product lifetime. Secondly, dynamic random-access memory (DRAM) stores bits using cells made up of a single transistor and a single capacitor. As a result, these memories are compact and cost-effective. However, because of device leakage, period self-refresh is required to maintain the stored information. Finally, SRAM may be used to store information using flip-flops. Most implementations use six transistors per bitcell. While no refresh mechanism is needed, the flip-flops still exhibit some gate leakage, especially in smaller silicon process geometries. SRAM is nearly always integrated within the SoC which maximizes access speed and minimizes active power consumption. By contrast, we find flash and DRAM devices typically implemented outside the SoC but often integrated within the same package. This compromise allows each component to use the most suitable and cost-effective silicon process technology. Table 3.1 below compares the three storage options in terms of access speed and power consumption. The figures are based on low-cost but state-of-the art devices used in today’s cellular IoT products. The transfer energy figures include power required to drive the physical interfaces between the SoC and external memory chips. The table exposes drastic differences in cost, access speed and energy as well as data retention power. The following section will discuss different strategies to use these storage media for different sleep durations. 3.1.2.2 LPM Strategies Starting from PSM, it is clear that all code and data retention should use non-volatile flash memory. For example, context information established during cell attachment must be stored together with RF configurations such as channel frequency and bandwidth, so that at the next wake up the device can quickly reconnect, and resume operation as needed. For long sleep duration (minutes), it is still preferable to power down the entire SoC and store Page 39 of 81
D2.2 - Initial Report on END Hardware Design all code and context data in flash memory. The downside of this approach is that the reboot process will take longer due to slow access to the flash. Keeping the amount of code and data loaded to a minimum is key for minimizing battery consumption. For shorter sleep durations (seconds), retaining all code and context data in DRAM becomes more attractive. While leakage is non-zero, it is still low, and the energy required to re-boot the system is much smaller compared to re-booting from flash memory. Only for the shortest idle periods (milliseconds), it is beneficial to retain information in SoCinternal static random-access memory (SRAM) in order to avoid overheads for re-configuring the chip for the next active slots. In future work, we will develop detailed models to analyze the exact trade-offs involved. We will also outline the amount of code that is required for certain operations. For example, we will demonstrate that providing dedicated low complexity wake up signals (like the NR LP-SS and LP-WUS signals currently being defined in 3GPP Release 19) lower both the amount of code required as well as the modem processing time which reduces overall power consumption. Our goal is to provide a power model based on 5G NR technology that can provide an Asize battery lifetime of 15+ years and outline several enhancements fit for A-IoT technology and devices that make it possible to further lower power consumption to eventually eliminate batteries altogether. 3.1.2.3 LPM Receiver Depending on the mode that the device has to operate, the requirements for the various circuitry can be relaxed and operation may be acceptable at lower currents. We can generally consider, for example, three modes of cellular IoT device operation: 1) Mode-1: Active duplex traffic, requiring Tx and Rx blocks fully operational with high performance; 2) Mode-2: Active traffic reception and transmission at low power or inactive, requiring only Rx operational at optimum performance; 3) Mode-3: Paging monitoring, requiring Rx operational at low performance. The relaxed requirements due to absence of TX blocker and lower modulation complexity (including, e.g., relaxed in-band and out-of-band phase noise, RF linearity, ADC signal-to-noise ratio (SNR)) can be exploited to save power at the device. Some of the parameters that can be adjusted for maintaining the bare minimum performance and reducing power consumption when in low performance mode are: •Linearity of the receive chain. •Transistor current in the voltage-controlled oscillator (VCO) of the local oscillator, frequency dividers, and clock buffers. •Bit Resolution and/or Sample Rate of the ADC. •Transistor current in the oscillator of the clock generator for the ADC. •Number of active filtering stages in the baseband receive chain. •Load regulation of power supply modules. Table 3.2 indicates which relaxations are possible depending on the device operating mode. Page 40 of 81
D2.2 - Initial Report on END Hardware Design Table 3.2: Potential relaxations per device operating mode Device operating mode Distinct feature Reduce Linearity Reduce Oscillator Current/LO buffering Reduce ADC sample rate/resolution Reduce ADC clock generator current Bypass Active Filter Mode-1 Tx+Rx fully operational No No No No No Mode-2 Rx fully operational, Tx inactive or at low power Partially Partially No No Yes Mode-3 Rx only & operational at low performance Yes Yes Yes Yes Yes In a next deliverable, we will analyze and quantify the impact of the various parameters to understand the benefits of adjustable circuitry operation and where one can exploit similarly relaxed requirements to save power at ENDs devices. 3.2 Backscattering Type of Devices Backscatter communication (BC) enables ultra-low-power wireless connectivity by reflecting dedicated or ambient RF signals, eliminating the need for active transmission. Recent advances integrate analog and digital modulation, increase energy harvesting efficiency, and implement hybrid visible light communication (VLC)-RF architectures. These developments pave the way for scalable, energy-autonomous A-IoT systems operating in environments with minimal power budgets. 3.2.1 Radio Frequency Identification Technology for Backscattering-Type Energy-Neutral Devices Passive RFID systems are a cornerstone of ultra-low-power wireless communication, particularly suited for ENDs in A-IoT environments. These systems operate by reflecting incident electromagnetic waves, enabling both data transmission and EH without active RF components [22], [101]. A typical passive RFID setup consists of a reader and a tag. The reader emits an RF signal, which is received by the tag’s antenna. The tag modulates its input impedance to encode data onto the backscattered signal. This impedance modulation results in distinct states, which are detected by the reader. The tag’s chip switches between two impedance states, resulting in encoding binary information [102]. Figure 3.3 shows a representative RFID front-end architecture developed in 40 nm CMOS technology. It includes a differential RF-DC converter, voltage limiter, and load modulator. Zöscher et al. have demonstrated that such designs can achieve cold-start operation at input powers below −15 dbm, making them ideal for ambient-powered ENDs [21], [24], [25]. The architecture comprises several key blocks that enable backscatter communication: Page 41 of 81
D2.2 - Initial Report on END Hardware Design Figure 3.3: Differential UHF RFID front-end architecture in 40nm CMOS technology [25]. Receiving and Demodulating the Incoming Signal: The UHF antenna receives the incoming modulated signal at ports RF1and RF2. Electrostatic discharge protection is provided by ESD diodes. The signal is routed to the demodulator, which follows an envelope detector architecture to convert the received modulated signal into a digital data out bitstream. Backscattering via RF Modulator: A digital data in signal is fed into the modulator. The modulator dynamically alters its impedance in response to the incoming carrier wave. This impedance variation (shorting the RF front-end) causes a portion of the incident wave to be reflected, encoding the data in signal onto the backscattered wave. Recent advances in backscatter communication have extended the capabilities of RFID-based systems. Techniques such as harmonic backscattering, intermodulation, and ambient backscatter enable operation over longer distances and in complex environments [101], [103]. These methods exploit nonlinear components or ambient RF sources (e.g., Wi-Fi, bluetooth) to eliminate the need for dedicated readers for energy supply, enabling pervasive deployment of A-IoT devices. Ambient backscatter is particularly promising. Tags can reflect existing RF signals from infrastructure such as Wi-Fi access points or cell towers, drastically reducing the deployment cost and energy requirements. The analog front-end plays a critical role in enabling efficient EH and modulation. State-ofthe-art designs incorporate threshold voltage compensation, adaptive frequency selection, and differential architectures to improve conversion efficiency and robustness against process variations [21], [24]. These innovations are essential for reliable operation in low-field environments typical of ambient IoT deployments. Backscattering-type ENDs are particularly suited for applications requiring ultra-low power and long-term autonomy. These include smart infrastructure monitoring, wearable health sensors, asset tracking in supply chains, and environmental sensing. The passive nature of backscatter communication allows for scalable deployments, where a single reader or ambient RF source can support hundreds of tags simultaneously using multiple access schemes such as time divisionmultiple access (TDMA),frequency division multiple access (FDMA), or compressive sensingbased decoding [103]. In summary, backscatter communication is a foundational technology for energy-neutral wireless Page 42 of 81
D2.2 - Initial Report on END Hardware Design Ambient RF sources Carrier waves Receivers Backscattered signal Modulated light RF modulation Control signal Photovoltaic conversion BD Energy Sensors Data LEDs (a) (b) Figure 3.4: BD: (a) System model of LiBD, (b) schematic of LiBD. systems. Mostly used in the field of RFID, it is now a promising communication technology to enable A-IoT devices. Its passive nature, scalability, and compatibility with ambient RF sources make it a key enabler for the future of sustainable and ubiquitous IoT deployments. The integration of advanced analog front-ends, efficient modulation schemes, and ambient EH techniques continues to push the boundaries of what is possible with backscattering-type ENDs. 3.2.2 Detailed Information on Backscattering-Type Energy-Neutral Devices 3.2.2.1 UP-END, BD with VLC Relaying and Forwarding The ubiquitous presence of configurable lighting such as LED support sensors to transmit their data through optical wireless channels, known as VLC enabled IoT. We propose a combination of VLC and BC shown in Fig. 3.4a. In this scenario, the data originating from the sensors and sent using VLC, is converted to electrical signals at a BD, namely LiBD, and the BD then transmits the data by modulating and reflecting ambient RF signals. The design of BD shown in Fig. 3.4b with circuit diagram (left) and manufacturing footprints (right), comprising lumped elements mounted on a thin-film inkjet-printed circuit board and an inkjet-printed coplanar waveguide antenna. A solar cell BCSC241D4 harvests modulated light from the sensors and converts the light into the voltaic signal denoted by Vsc. The Vsc feeds the positive input of a comparator (CMP) TLV7031 and an RC low-pass filter (LPF) composed by a 100 kΩ resistor (R1) and a 0.1µFcapacitor (C1), respectively. This LPF has a cut-off frequency of 15 Hz at 3 dB, and outputs the DC component of Vsc, denoted by VLPF for feeding the negative input of the comparator. Hence, the comparator compares the AC and DC components of Vsc, setting the output voltage VCTRL to high(VH), if Vsc > VLPF, otherwise setting VCTRL to low (VL). The output signal VCTRL is then used to control RF switch ADG919 for backscatter modulation and reflection. The RF switch is connected to the coplanar waveguide antenna, switching the termination of antenna between two loads Zi∈ {Z1, Z2}. Assuming that the antenna is matched to characteristic impedance Za, the complex reflection coefficient [104] is expressed by Γi= (Zi−Z∗ a)/(Zi+Za), where ∗denotes the complex conjugation. The corresponding modulation factor M[104] is then M=1 4|Γ1−Γ2|2.The modulator achieves the maximum of M= 1, when Z1=∞and Z2= 0 corresponding to the switch either opening or shorting the antenna termination. The signal VCTRL controls the RF switch and is converted Page 43 of 81
D2.2 - Initial Report on END Hardware Design Figure 3.5: System model of Li2BC. from the modulated light carrying sensor data. Here, the binary frequency shift keying (FSK) baseband signal is implemented for demonstrating sensor data transmission, such that the switching speed varies between predefined frequencies f1and f2. When the ambient RF signal with a carrier frequency of fcarrives at the antenna, it gets mixed with the switch control signal VCTRL by the switching operation. The modulated backscatter signals with frequencies fc±f1and fc±f2will reflect and carry the sensor data. The reader detects and demodulates the backscattered signal and recovers the sensor data. 3.2.2.2 CO-END, BD with VLC Wake-up Front End and Control We present Li2BC, a novel backscatter device that integrates VLC receiving, processing, and backscatter modulation functionalities. By utilizing a low-power microcontroller, Li2BC can receive, amplify, demodulate, and store/cache VLC signals. This caching capability enables the device to store VLC data for subsequent control of backscatter operations, enhancing its configurability and flexibility. Unlike passive devices that simply relay incoming VLC signals, Li2BC empowers more sophisticated applications by combining light-based communication with signal processing and storage functionalities, paving the way for innovative IoT solutions. Figure 3.5 shows an overview of the proposed backscatter communication system enabled by the Li2BC. The LEDs are driven by a constant current LED driver controlled by a pulse width modulation (PWM) signal. Information can be coded into the frequency of the PWM signal. As long as the frequency of the PWM is above 2 kHz [105], the information encoding does not cause visible flickering of the light. The duty cycle of the PWM signal affects the LED light intensity, and it must be held constant to avoid any flickering. Li2BC is powered by a PV cell, which is also used to receive control data over light. As the information is encoded in a high-frequency component, it can be extracted by filtering. The filtered signal is then amplified and fed into the MCU for backscatter control. Internal capacitance and resistance of the PV cell form a circuity resembling an LPF. This sets an upper limit for the frequency of the light intensity encoding. Depending on the cell used, this limit is in 10–20 kHz range. The MCU communicates with BD-embedded sensors, acquiring sensing data periodically. To transmit the data with AmBC, Li2BC uses one of the MCU’s timers to generate frequencies for binary FSK modulation. The MCU changes the timer parameters for every transmitted symbol. The timer can run independently, the rest of the MCU is kept in a low-power state during symbol transmission to conserve energy. The modulated signal controls an RF switch Page 44 of 81
D2.2 - Initial Report on END Hardware Design f A Figure 3.6: Example of how a tag generates frequency modulation by adjusting the antenna load. to change an antenna’s reflection coefficient backscattering ambient signal. The backscattered signal can be received by commonly used RF receivers. 3.2.3 Hardware Options for Backscattering A backscatter device reflects incoming RF signals to carry information. The tag modulates data onto these signals by changing how much of the incoming signal it reflects back. This is determined by the complex reflection coefficient Γi[104]. A tag changes its reflection coefficient by adjusting the load impedance connected to the antenna, as depicted in Figure 3.6. The matching between the antenna and load impedance enables the tag to fully absorb (Γ = 0), fully reflect Γ = 1 or partially reflect (Equation (3.1)) the incoming signal. Here, Ziis the input impedance connected to the RF port going to the antenna and Zais the antenna impedance with ∗being the complex conjugate. Γi=Zi−Z∗ a Zi+Za (3.1) At its core, a backscatter device toggles or switches between loads using an RF switch. The resulting impedance mismatch attenuates the incoming signal Ain by a factor |Γ|and adds a phase shift θ=∠Γito θin. The reflected signal has an amplitude Aout and phase θout . Aout =|Γi|Ain (3.2) θout =θin +θ(3.3) Different techniques, described below, exist to control the RF switch providing different benefits. Every technique has several modulation techniques that are the most suitable for it. 3.2.3.1 VCO-based A mixed analogue-digital approach makes use of a VCO to control a RF switch [106]. A digitalto-analog converter (DAC) generates the VCO control voltage determining the backscattered frequency. This approach uses frequency based modulation techniques such as FSK and chirp spread spectrum (CSS). The resolution of the DAC is determined by the amount of symbols the hardware needs to transmit. For example, with FSK an 8-FSK scheme is able to transmit 3 bits by switching between 8 different frequencies. To achieve this the DAC needs to have an 8-bit resolution. With CSS,SF bits are transmitted per symbol resulting in 2SF symbols and thus Page 45 of 81
D2.2 - Initial Report on END Hardware Design 2468 0 10 20 30 40 15.918.8 30.6 36.6 Number of Layers GWP [gCO2eq/cm2] Figure 4.1: PCB impact with varying number of layers 4.1.1 Impact of the Printed Circuit Board PCBs typically only represent a small fraction of the volume of an electronic device, yet they account for an estimated 3–6% of the total weight of waste electrical and electronic equipment (WEEE) [116]. Given the electronics industry is responsible for around 4% of global greenhouse gas emissions, the contribution of PCBs to this total is far from negligible [117]. Rigid PCBs Rigid PCB are composed of a substrate, most commonly FR4 laminate, which is widely used in the electronics industry [116]. This substrate typically contains toxic brominated compounds to meet flame retardancy requirements. Depending on the PCB type, this substrate is copper-plated on one or both sides. More precisely, for multilayer PCBs, this substrate is then stacked with alternating copper and pre-preg layers, which are all compressed together. As the number of layers increases, so does the amount of copper and pre-preg material. This results in a greater environmental impact, as both the FR4 laminate and the copper layers require significant energy and solvents during production. The impact of PCBs is assessed for different layer configurations, using a functional unit of 1 cm2. The data used are based on a 2023 reference year, and modeled with a Hot Air Solder Leveling (HASL) surface finish. An attributional LCA approach is applied. As expected, the environmental impact quasi linearly with the area of the PCB and increases with the number of layers (as shown in Figure 4.1). Flexible PCBs Flexible PCBs are manufactured using additive processes, such as screen or three-dimensional (3D) printing of conductive inks (silver or carbon-based) on flexible substrates like thin polyimide or polyester sheets [116], [118]. Although their production process is more technologically demanding compared to rigid PCBs, they are made with thinner materials and consequently require less raw materials [119]. In contrast to the subtractive etching process used for rigid PCBs, the additive printing processes used for flexible PCBs generate significantly less material waste due and reduces the use of harmful chemicals [120]. In [121], a comparative environmental analysis is conducted on various PCB configurations, differing in both substrate materials and conductive elements. Table 4.1 outlines the evaluated scenarios, covering both rigid and flexible PCB types. Figure 4.2 presents the GWP per square centimeter for each configuration, broken down by four key contributors: electricity consumption, conductive ink, Page 52 of 81
D2.2 - Initial Report on END Hardware Design Table 4.1: Substrate and conductive material used in different scenarios. Scenario Substrate Conductive Material PCB Type S1 FR4 Etched-Copper Rigid PCB S2 FR4 Ag NPs Rigid PCB S3 PET Ag NPs Flexible PCB S4 PLA60%–GF40% Ag NPs Flexible PCB S5 Paper Ag NPs Flexible PCB substrate, and chemical processing. This confirms the impact improvement of additive printing processes compared to subtracting etching processes. Flexible designs additionally often require fewer layers and can be volumetrically folded, enabling smaller device casings and potentially lower power requirements through shorter interconnections and more efficient layouts. Furthermore, in applications where the PCB is subjected to vibration or mechanical stress, flexible PCBs can even outperform rigid ones in operational durability due to their ability to absorb and withstand mechanical deformation. On the other hand, considering the shelf life of flexible PCBs, they typically last 1–2 years, while rigid PCBs can last much longer under proper storage and use. For example a rigid PCB made with FR4 can last 2-5 years and probably even longer [118], [122]. After some time, the surface finish of PCBs (e.g., HASL, ENIG, OSP) degrades and affects solderability. In the case of flexible PCBs, this degradation is accelerated because their more sensitive materials and adhesives are especially vulnerable to moisture, heat, and oxidation. On the other hand, the usable lifespan of a soldered PCB is generally in the order of 10-20 years in use [123]. However, with appropriate storage conditions and by using protective surface finishes, their shelf life can be significantly extended up to even 50-70 years [124]. Biodegradable Materials Reducing the environmental footprint of PCBs starts with reconsidering the materials used in their production. As shown in Figure 4.2, substrate selection plays a critical role: over 33% of a traditional PCB’s carbon emissions are attributed to raw materials, particularly the substrate and conductive ink [121]. While using substrate materials like polyethylene terephthalate (PET) and paper already have a significant improvement, more and more attention is being given to biodegradable alternatives. The use of bio-based substrates requires less energy-intensive processing and involves fewer harmful chemicals, in contrast with traditional epoxy–fiberglass PCB substrates [116], [121]. As a result, overall environmental impacts can be reduced by as much as 60–80%, further supported by a simplified EoL process. One of the most promising candidates is polylactic acid (PLA), which has already demonstrated reliable performance in high-speed digital electronics, expecting it to be soon available in industry [116], [125]. However, while bio-based materials show great potential, most still lack the thermal and chemical stability required for harsh environments like automotive applications [116]. For now, their use will likely be limited to consumer electronics, where performance and lifetime requirements are less stringent. Although this may seem limited, it could already yield substantial environmental benefits due to the massive volume of electronics produced and discarded globally, contributing to Page 53 of 81
D2.2 - Initial Report on END Hardware Design s1 s2 s3 s4 s5 0 5 10 15 20 25 30 35 8.62 6.29 6.29 0.73 0.63 0.11 5.2 4.4 4.4 4.4 4.4 14.2 7.1 0.19 0.19 0.19 Number of Layers GWP [gCO2eq/cm2] Electricity Conductive Ink Substrate Chemicals Figure 4.2: Impact of PCBs with different substrates and conductive materials [121] a more sustainable and circular industry. Biodegradable PCBs typically do not last as long as their non-biodegradable counterparts. Disintegration is typically very minimal during the conditions where the PCBs are normally used. For EoL processing conditions, strongly influenced by factors such as humidity and temperature, degradation can proceed rapidly, ranging from several days for mycelium-based PCBs to a few weeks for physically large array (PLA)-based substrates [126], [127]. PCB Assembly Before a PCB can be incorporated into a final product, electronics components must be mounted onto it during assembly phase using surface mount device (SMD),throughhole technology (THT), or mixed technologies. For a representative mixed assembly process, the environmental footprint is estimated at approximately 0.5 gCO2eq/cm2of assembled board area. This estimate is based on a representative industrial production line configuration that includes a screen printer, a THT placer, a chip shooter, a wave solder oven, and a reflow oven. This reflects a typical setup for mixed-technology PCB assembly. Variations in complexity, or the use of only THT or only SMD assembly show only negligible differences in impact. 4.1.2 Peripherals, Internal Memory and Processing MCUs differ greatly in their internal memory, processing power, and peripheral integration. Basic MCUs, such as ARM Cortex-M0(+) cores, focus on low power consumption and have minimal features, while MCUs with RF front-ends include extra circuitry for wireless communication and processing thereof, increasing their complexity. Some MCUs incorporate specialized peripherals, such as liquid-crystal display (LCD) or USB interfaces, or feature dual-core architectures to Page 54 of 81
D2.2 - Initial Report on END Hardware Design support more advanced applications. In this section, the impacts of different MCUs are quantified in order to determine how varying characteristics and peripherals affect their respective environmental footprint. The environmental footprint of an IC is mainly determined by two factors: the CMOS process node technology and the physical die area of the chip. The environmental impact of the chip packaging can typically be neglected as they, e.g., are in the order of <0.1 gCO2eq/mm3for ball grid array (BGA) packages [128]). To accurately assess the die area, the IC’s are encapsulated leaving the bare silicon die exposed. The die area can then be measured with a microscope. The process node technology (e.g., 180 nm,90 nm,40 nm CMOS) is identified from manufacturer data when available. Otherwise, it is estimated based on the IC’s release date, comparison with similar chips of the manufacturer, and knowledge of the manufacturer’s fabrication capabilities. 4.1.2.1 Modeled Components For this study, a set of STMicroelectronics microcontrollers was modeled to capture different design characteristics relevant to environmental impact. The selection includes simple low-power Cortex-M0(+) devices (e.g., STM32L031K6U6TR, STM32U031C8U6), MCUs with additional peripherals such as an USB or LCD (e.g., STM32U073CCU6), and devices integrating RF front-ends for Bluetooth Low Energy (BLE) or LoRa communication (e.g., STM32WB09KE, STM32WL55CCU6, STM32WB15CC). A high-performance variant manufactured in a smaller process node, such as the STM32WBA52CGU6, is also included. Together, these components represent the range from basic low-power designs to dual-core RF-enabled MCUs, enabling a representative comparison of their carbon footprints, showing how choices such as adding RF, integrating peripherals, or adopting smaller process nodes directly affect environmental impact. Table 4.2 gives an overview of the die areas and CMOS technology nodes used for modeling. - STM32L031K6U6TR: Ultra-low-power ARM Cortex-M0+ This 32 MHz Cortex-M0+ microcontroller features 32 kB of Flash and 8 kB of RAM, optimized for ultra-low-power applications with multiple sleep modes. - STM32U031C8U6: Ultra-low-power ARM Cortex-M0+ The STM32U031C8U6 microcontroller features 64 kB of Flash and 12 kB of SRAM and is an ultra-low-power 56 MHz MCU based on the Arm Cortex-M0+ 32-bit RISC core. - STM32U073CCU6: ARM Cortex-M0+ with LCD and USB driver Equipped with a 56 MHz Cortex-M0+ CPU,256 kB of Flash, and 40 kB of SRAM, this MCU combines energy efficiency with integrated LCD control and USB 2.0 full-Speed support. - STMWL55CCU6: Dual-core ARM Cortex-M4/M0+ with LoRa front-end Featuring a dual-core 45 MHz Cortex-M4/M0+ architecture, this MCU offers 256 kB of Flash and 64 kB of RAM, and integrates a LoRa radio for low-power wireless communication in IoT applications. - STM32WB09KE: ARM Cortex-M0+ with 2.4 GHz RF front-end (BLE) The STM32WB09KE incorporates a 64 MHz Cortex-M0+ core with 512 kB of Flash and 64 kB of RAM. It is designed for BLE connectivity, featuring an integrated 2.4 GHz RF Page 55 of 81
D2.2 - Initial Report on END Hardware Design Table 4.2: Impact specifications of modeled microcontroller components, including die dimensions and CMOS process technology. Component Width [mm] Height [mm] Area [mm2] CMOS [nm] Source / Note STM32L031K6U6TR 2.174 2.475 5.381 90 Based on similar ST Cortex-M0+ [129] STM32U031C8U6 2.380 2.502 5.955 90 Given by [129] STM32U073CCU6 2.787 2.974 8.291 90 Given by [129] STMWL55CCU6 3.629 3.679 13.356 90 From [130] STM32WB09KE 2.732 2.957 8.072 90 Assumed based on [129], [130] STM32WBA52CGU6 2.723 3.053 8.313 40 From [130] STM32WB15CC 3.388 3.336 11.302 90 From [130] front-end supporting BLE 5.4. - STM32WBA52CGU6: ARM Cortex-M33 with Bluetooth LE 5.4 Ultra-low-power, Cortex-M33 Trust Zone MCU 100 MHz with 1 MB of Flash memory, BLE 5.4 - STM32WB15CC: Dual-core ARM Cortex-M4/M0+ with Bluetooth LE 5.3 The STM32WB15CC is an ultra-low-power dual-core microcontroller featuring a 64 MHz ARM Cortex-M4 and a 32 MHz Cortex-M0+, with 320 kB Flash memory, featuring BLE 5.3. 4.1.2.2 Impact Comparison and Conclusion Figure 4.3 illustrates the environmental impact of various microcontrollers, each featuring different specifications. The STM32L031K6, which is a simple low-power M0+ MCU without RF capabilities, exhibits the lowest environmental impact at 74 gCO2eq. The impact of the STM32U031C8U6 is very similar (81.9 gCO2eq), which is expected as they both are low-power 90 nm Cortex-M0+ MCUs. In contrast, the STM32U073CC, also a low-power Cortex-M0+ MCU, exhibits a higher impact of 114 gCO2eq, primarily due to its integrated LCD controller and USB 2.0 support peripherals. The RF-enabled microcontrollers exhibit notably higher impacts due to their integrated RF front-ends. A meaningful comparison can be made between the STM32L031K6U6TR and the STM32WB09KE, as both are based on the ARM Cortex-M0+ core and lack additional complex peripherals, except for the 2.4 GHz BLE RF front-end in the latter. Their impacts are 74 gCO2eq and 111 gCO2eq, respectively. In contrast, the STMWL55CCU6 (with 868 MHz RF frontend for LoRa communication) shows a significantly higher impact of 184 gCO2eq, even compared to the STM32WB09KE, despite both including RF front-ends. This increase can be explained by its dual-core architecture. Furthermore, comparing the STMWL55CCU6 and STM32WB15CC, both dual-core microcontrollers with integrated RF front-ends, demonstrates that using either aBLE or LoRa front-end results in comparable environmental impacts, with emissions of apPage 56 of 81
D2.2 - Initial Report on END Hardware Design STM32L031K6U6 STM32U031C8U6 STM32U073CC STM32WL55CCU6 STM32WB09KE STM32WBA52CGU6 STM32WB15CCU6 0 50 100 150 200 250 300 74 81.9 114 184 111 211 155 GWP [gCO2eq] Generic MCU MCU + RF frontend Figure 4.3: Comparison of total environmental footprint of different MCUs proximately 184 gCO2eq and 155 gCO2eq, respectively. Finally, the STM32WBA52CGU6 has the highest impact of 211 gCO2eq, as it is made out of 40 nm CMOS technology to ensure its high performance capabilities [130], [131]. 4.1.3 Communication Wireless communication forms a crucial part of IoT devices, enabling communication over diverse ranges, bandwidths, and energy budgets. Requirements may vary significantly from ultra-low power and long-range communication (e.g., LoRa) to higher data throughput cellular connectivity (e.g., NB-IoT). The SiP modules typically used to enable this connectivity represent a non-negligible share of its environmental footprint due to their semiconductor content and supporting circuitry. Two representative communication technologies are analyzed in detail based on COTS components: LoRa and NB-IoT. Different antenna configurations are considered as they represent an additional contribution to the footprint of wireless communication. LoRa Transceiver •HopeRF RFM95W: The LoRa transceiver includes an SX1278 LoRa transceiver, capacitors, resistors, transistors, and one LDO. The SX1278 chip is modeled using a 22 nm CMOS process with an estimated die area of 12.93 mm2. The LDO is modeled using a 65 nm technology node, based on an average of the process nodes used for ALDOs and DLDOs [132]. The PCB has a surface area of 2.5 cm2and is modeled as a 4-layer FR4 board with a HASL surface finish. Assembly is assumed to take place on a medium-volume line operating at 300 units per hour. Additionally, 1 g of SnPb34Ag2 solder paste is used in the assembly Page 57 of 81
D2.2 - Initial Report on END Hardware Design Total SX1278 PCB LDO Resistors Transistors Other 0 100 200 300 358 266 47.1 16.1 16.17.18 10.3 GWP [gCO2eq] Figure 4.4: RFM95W LoRa module impact process. The total estimated environmental impact of the module amounts to 358 gCO2eq, as shown in Figure 4.4. NB-IoT Transceiver •Nordic nRF9160: The 16 mm x10 mm nRF9160 SiP is a highly integrated, low-power solution for cellular IoT applications. It combines an ARM Cortex-M33 processor, Long Term Evolution for Machines (LTE-M)/NB-IoT modem, Global Positioning System (GPS), PMIC, memory, and embedded subscriber identity module (eSIM) support in a single compact package. With its built-in RF front-end and secure processing, it enables reliable, efficient, and connected edge devices. The SiP is encapsulated in a molding compound, which makes identifying individual components challenging. Nordic provides data claiming the carbon footprint of a nRF9160 is 564 gCO2eq per unit. This value represents a cradle-to-gate for chip production, including wafer manufacturing, assembly/test, housing, leadframe, solderpaste, wirebond and transport of semi-finished goods, and components and assembly for SiP production. Other such as packaging materials, downstream transportation, use of product and EoL are not included. The processing node is not disclosed publicly, but a process node of 22 nm -40 nm can be assumed, as these are respectively used for the nRF54 and nRF52 series [133], [134]. Since the nRF9160 was released around the same time as the nRF52 series, a process node of approximately 40 nm is a reasonable estimate. Based on previous results, the environmental footprint of a typical MCU with an integrated RF frontend, such as the STM32WBA52CGU6, is about 211 gCO2eq. This leaves approximately 353 gCO2eq attributable to the wireless communication and GPS module, which aligns well with the footprint of the Quectel module discussed below. To ensure proper functionality of this SiP, several additional components are required. The overall environmental impact, including these components, is illustrated in Figure 4.5. •Quectel BC660K-GL: This high-performance LTE Category Narrowband IoT Release 14 (Cat NB2) module, based on the Qualcomm QCX212 chipset, is specifically designed for applications in smart metering, bike sharing, smart city, agricultural monitoring and Page 58 of 81
D2.2 - Initial Report on END Hardware Design Total nRF9160 Side components 0 200 400 600 565.08 564 1.08 GWP [gCO2eq] Figure 4.5: nRF9160 SiP (including components surrounding SiP) other related sectors. It is distinguished by its low power consumption and compact form factor [135]. The impact is estimated for both the BC660K-GL module (Figure 4.6) as the additional components required to make this NB-IoT module functional (Figure 4.7). For the former, the electromagnetic compatibility (EMC) casing was thermally removed, and the various components on this PCB were identified and desoldered when necessary to measure their weight (e.g., ferrite beads, oscillators, ...). Additionally, the bare die area of the QCX212 chipset was measured by mechanically removing its plastic encapsulation. For this LTE IoT modem, we deliberately chose a 45 nm process node. Furthermore, another smaller IC, probably a small MCU with some logic to encode/decode AT commands, is modeled using a 90 nm process node. Beyond the module, several external components are needed for proper operation. These include decoupling capacitors and Zener diodes to stabilize and protect the Vbat supply, as well as resistors, capacitors, and diodes to interface with the subscriber identity module (SIM) card connector. The impact of both the SIM card connector and a physical SIM card were also taken into account. For this SIM card, the environmental impact was estimated using data from a study conducted by Fraunhofer [136]. For all other components, the Sphera LCA tool was used to estimate the impact based on the quantity, weight, or die area of the components. The SIM card module was specifically modeled using copper, stainless steel, and thermoplastic polyurethane (TPU), based on material data commonly found in datasheets of SIM card connectors, such as the CSI-25406X-18. For a standard SIM card connector [137], these materials were carefully separated and each component was individually measured to ensure an accurate representation, yielding the following results: –TPU:0.059 g – Copper: 0.110 g – Stainless steel: 0.193 g Antennas Page 59 of 81
D2.2 - Initial Report on END Hardware Design Total EMC shield Ferrite beads ICs PCB Capacitors Resistors Oscillators 0 50 100 150 149.88 4.57 8.7·10−2 83.22 53.99 0.29 3.87 0.14 GWP [gCO2eq] Figure 4.6: Quectel BC660K-GL (only module itself) •Dipole antenna: a dipole antenna can be modeled using a material mix of 20% copper, 40% steel, and 40% ABS [138]. The environmental impact of such a whip-like antenna is then estimated at 3.18 gCO2eq per gram of antenna (so respectively 0.785 gCO2eq/g, 1.3 gCO2eq/gand 1.1 gCO2eq/gfor copper, steel and PET). •PCB antenna: As shown in Figure 4.1, the environmental impact of a two-layer PCB is estimated at 15.9 gCO2eq/cm2. This value can be used as an approximation for the environmental impact of a PCB antenna. •PCB-mounted THT antenna: As an example of such an antenna, the ANT-2.4-MSATH1 through-hole mounted stamped and bent steel antenna can be used. This is modeled using a basic LCA model, where stainless steel is processed through stamping and bending. This specific antenna has a mass of 0.644 g, corresponding to an estimated environmental impact of 1.32 gCO2eq. In general, the impact of such a stamped and bent stainless steel antennas can be estimated at approximately 2.05 gCO2eq per gram of antenna. 4.1.4 External Memory Although the use of external SRAM ICs is uncommon, their environmental impact has been calculated across various scenarios, including different operating voltages and memory sizes, to estimate how their impact might scale when integrated within an MCU. The impact of a more commonly used external electrically erasable programmable read-only memory (EEPROM) IC has also been evaluated. However, the process node values for these technologies (250 nm for SRAM and 350 nm for EEPROM) are approximate and uncertain. Additionally, the SRAM was simulated in a 250 nm node only because it is the largest available in Sphera. Therefore, a direct comparison of their environmental impacts is not meaningful. SRAM Memory Sizes: 64 kb/256 kb/512 kb The die sizes of different volatile SRAMs, with different memory sizes, are measured and depicted below: •23A640T (64 kb): die area is 1.113 mm x2.095 mm =2.332 mm2 Page 60 of 81
D2.2 - Initial Report on END Hardware Design Total BC660K-GL SIM card SIM card connector Diodes Capacitors Resistors 0 100 200 300 400 393.05 149.8 229 3.31 7.14 0.43 4.13 GWP [gCO2eq/cm2] Figure 4.7: Quectel BC660K-GL (including components around module) •23A256T (256 kb): die area is 1.199 mm x2.119 mm =2.371 mm2 •23A512T (512 kb): die area is 2.016 mm x2.936 mm =5.919 mm2 The results show that the 64 kB and 256 kB SRAM chips have nearly the same die size despite their capacity difference. This suggests that for capacities close to each other, binning techniques are used to maximize yield. In contrast, the 512 kB chip does have a much larger die, indicating that higher capacities require genuinely bigger silicon areas. It is possible that also here binning happens between capacities in the same range. SRAM Supply Voltages: 1.5–1.95 V/2.7–3.6 V 64 kB volatile SRAM with different supply voltage ranges. The die area of both the 23A64OT and 23K64OT were measured. They differ in their voltage supply range. While the 23A64OT works between 1.5–1.95 V, the 23K640 works between 2.7–3.6 V. Measuring their die area yields the following results: •23A640T (1.5–1.95 V): die area is 1.113 mm x2.095 mm =2.33 mm2 •23K640T (2.7–3.6 V): die area is 2.173 mm x1.205 mm =2.62 mm2 Based on the large die size and 64 kB memory, the estimated areas of one SRAM cell of the 23A650T and 23K650T external memories are approximately 17.776 µmand 19.989 µm, assuming that 50% of the total die area is occupied with SRAM, and the other 50% is logic for peripherals. According to [139], these values correspond to a 350 nm or larger process node, where SRAM cells range from 9.26 µmto 50 µm. This also aligns with [140], which states that Microchip Technologies Inc. still focuses on older, mature process nodes. Since Sphera’s largest available data is for 250 nm process nodes, we estimate their impact considering this technology. The impact of both memory ICs can be found in Figure 4.8. The size difference is therefore likely due to higher voltage requirements rather than a different process node, as the die area, and consequently the SRAM cell area difference, between these two memory modules is too small [139]. Higher voltage chips need larger transistors, I/O drivers, ESD protection, and voltage safeguards, all of which increase die area. Page 61 of 81
D2.2 - Initial Report on END Hardware Design IC Antenna PET 0 1 2 3 43.8 0.41 0.32 GWP [gCO2eq] Figure 4.12: GWP of the Avery Dennison RFID AD-227m5 Monza 5 xpSL3S1206 ST25DV04KC ST25DV16KC ST25DV64KC 0 10 20 30 3.82.81 30.432.432.3 GWP [gCO2eq] Figure 4.13: GWP comparison of different RFID chips •Weight aluminum is 0.0252 g •Weight PET is 0.115 g 4.1.7.5 Comparison Above described RFIDs can be compared based on their GWP impact. As can be seen in Figure 4.13, the ST25DV series have the largest impact, due to their larger die area compared to the Monza 5 and the xpSL3S1206 UHF RFID chip. 4.1.8 Casing This section compares the environmental impact of common materials used for electronic casings, expressed as global warming potential (GWP kgCO2eq) per kilogram of material. By evaluating conventional plastics such as ABS and PC, biodegradable bioplastics like PLA, and Page 68 of 81
D2.2 - Initial Report on END Hardware Design ABS PC PET PLA Steel Cold rolled steel Aluminum 0 2 4 6 8 3.28 3.95 2.85 2.34 2.37 3.25 8.69 GWP [kgCO2eq/kgmaterial] Figure 4.14: GWP comparison of casing materials metals including steel and aluminum, Figure 4.14 highlights their relative sustainability, aiming to support more environmentally conscious material choices. 4.2 Summary The LCA of typical IoT components shows that the overall environmental impact of IoT devices is largely determined by manufacturing hotspots. The main contributor to the total environmental impact of low-power IoT devices is in the manufacturing phase. ICs are highly optimized for energy-efficiency, but are resourceand energy-intensive to manufacture. The analysis highlights that the choice of IC type influences the impact, stressing the importance of selecting only a powerful IC when strictly needed. Additionally, wireless communication modules, especially long-range wireless communication ICs, come with a distinct impact, whereas simple RFID-based solutions can enable applications with very limited environmental impact. Similarly, minimizing the number of PCB layers and opting for renewable biodegradable PCB materials can reduce impacts. During the use phase, energy consumption is the main driver of impacts. Both energy storage (e.g. in batteries) and incoming energy (RF WPT and solar) play a role. Depending on the amount of required energy, an optimal storage solution, either lithium-based energy storage, supercapacitors, or even non-rechargeable batteries, must balance capacity, lifetime, and environmental footprint. RF WPT must be carefully considered, since efficiencies are typically low. In this case, low-carbon renewable energy sources can lower the operational footprint of edge devices. Solar panels for example can provide an effective and sustainable alternative when conditions allow. However, the required components for such energy harvesters and their storage systems also add to the manufacturing impacts. To limit the total environmental footprint of a device, energy efficiency in operation must be complemented by careful design choices. Overall, three strategies stand out. Extending product lifetime distributes the high productionrelated impacts of PCBs,ICs, and batteries over a longer period. Minimizing energy demand during operation cuts both the direct emissions and the need for large energy storage. Finally, Page 69 of 81
D2.2 - Initial Report on END Hardware Design circular design principles, including repairability, reuse, and recovery of valuable materials, help conserve primary resources. Together, these findings show why design decisions at the component level and operational efficiency measures must be considered jointly to achieve the lowest overall life-cycle environmental impact. Page 70 of 81
D2.2 - Initial Report on END Hardware Design Chapter 5 Conclusion Deliverable D2.2 has advanced the work of D2.1 from taxonomy and characterization to concrete hardware solutions that enable autonomous operation of low-power Internet of Things (IoT) devices. D2.1 established classes of energy-neutral devices and clarified the relationship between capabilities and operational modes. The present document operationalizes this foundation through an integrated view of harvesting, power transfer, multisource power management, device platforms, low-power communications, wake-up strategies, and sustainability analysis. The structure spans energy harvesting (EH) and power transfer, including charge pumps for radio frequency (RF) EH, transmitter architectures, and a comparison between ambient and dedicated RF; power management circuits with emphasis on multisource strategies, energy balance monitoring, and voltage limiting; as well as device platforms, backscatter options, wake-up radio design, and a life-cycle assessment of typical components. The deliverable is an initial report that consolidates state-of-the-art building blocks and clarifies their roles within an energy-neutral stack. In harvesting and power transfer, it documents efficient RF front ends and outlines transmitter architectures based on large-aperture arrays, lens-based beamspace processing, and reflective intelligent surfaces, which together improve end-to-end efficiency while limiting active RF chains. The analysis also explains when dedicated wireless power transfer should complement ambient sources to achieve reliable operation. On power management, the document addresses multisource units, energy combiners, balance monitoring, and voltage limiting. It highlights complementary circuit techniques, such as dynamic voltage scaling and adaptive body biasing, to keep ultra-low-power systems within safe and efficient operating regions across variable sources. For device-level communications, the deliverable surveys platforms and links hardware choices to energy budgets. Example platforms include DECT-2020 modules for 5G IoT scenarios. It details backscatter options and presents wake-up designs that exploit visible light control and recent cellular mechanisms with low-power synchronization and wake-up signals, which relax frequency and timing requirements and reduce processing effort. Finally, the life cycle assessment (LCA) provides component-level impact factors and a consistent methodology for global warming potential over 100 years using the EF 3.1 indicator. This supports responsible selection of materials and modules, including energy storage and incoming energy interfaces. Page 71 of 81
D2.2 - Initial Report on END Hardware Design The main contribution of D2.2 is a coherent map from capability needs to implementable hardware, together with selection guidance and protection strategies that are suitable for intermittent energy. This map will guide D2.3 and D2.4. The following deliverable will focus on implementation and technical details, including integration of selected harvesters and multisource power management with representative device platforms, validation of RF front ends and transmitter options in lab and field conditions, realization of wake-up paths based on visible light and cellular low-power signals, and measurement-driven energy budgets. Therefore, D2.3 will turn the consolidated options of D2.2 into concrete prototypes and reference designs that can be reused across use cases of the project. Finally, D2.4 will report a comprehensive performance evaluation of the prototypes. Page 72 of 81
D2.2 - Initial Report on END Hardware Design References [1] A. Luque and S. Hegedus, Handbook of photovoltaic science and engineering. John Wiley & Sons, 2011. [2] P–n junction: Utility solar power and concentration,https://www.eeducation.psu.edu/eme812/ node/606, Accessed June 2025, 2025. [3] M. Piliougine, A. Oukaja, M. Sidrach-de-Cardona, and G. Spagnuolo, “Temperature coefficients of degraded crystalline silicon photovoltaic modules at outdoor conditions,” Progress in Photovoltaics: Research and Applications, vol. 29, no. 5, pp. 558–570, 2021. DOI: https://doi.org/10.1002/pip.3396. [4] A. Chakraborty, G. Lucarelli, J. Xu, Z. Skafi, S. Castro-Hermosa, A. Kaveramma, R. G. Balakrishna, and T. M. Brown, “Photovoltaics for indoor energy harvesting,” Nano Energy, vol. 128, p. 109 932, 2024, ISSN: 2211-2855. DOI: https://doi.org/10.1016/j.nanoen.2024.109932. [5] C. Kofod, Indoor lighting in the public and private service sectors guidelines, Accessed: 03-03-2024. [Online]. Available: https://ec.europa.eu/research/participants/documents/downloadPublic? documentIds=080166e5b6d1bf85&appId=PPGMS. [6] How to choose a custom solar panel for iot device? https://www.wsl-solar.com/Industry_News/ 2025/0106/custom-solar-panel-for-iot-device.html, Accessed June 2025, 2025. [7] E. Lorenzo, Solar electricity: engineering of photovoltaic systems. Progensa, 1994. [8] PV Tech, Top 10 solar manufacturers 2024,https://www.pv-tech.org/top-10-solar-manufacturers2024, Accessed June 2025, 2024. [9] Josh Jackman, The 9 largest solar panel manufacturers in the world,https://www.sunsave.energy/ solar-panels-advice/solar-energy/largest-manufacturers, Accessed June 2025, 2025. [10] Flexible solar panels for electronics,https://powerfilmsolar.com/, Accessed June 2025, 2022. [11] Low-intensity ambient light photovoltaic solutions,https://ambientphotonics.com/, Accessed June 2025, 2025. [12] Solar manufacturing cost analysis,https://www.nrel.gov/solar/marketresearchanalysis/ solar-manufacturing-cost, Accessed June 2025, 2025. [13] A. Satharasinghe, T. Hughes-Riley, and T. Dias, “A review of solar energy harvesting electronic textiles,” Sensors, vol. 20, no. 20, p. 5938, 2020. DOI: 10.3390/s20205938. [14] S. B. Kang, P. Patil, G. W. Yoon, G. S. Han, H. S. Jung, and D. H. Kim, “Study of interface engineering on perovskite-based indoor photovoltaics for powering internet-of-things,” Chemical Engineering Journal, vol. 502, p. 157 973, 2024, ISSN: 1385-8947. DOI: https://doi.org/10.1016/j.cej.2024.157973. [15] S. Shcherbachenko, O. Astakhov, Z. Liu, L.-C. Kin, C. Zahren, U. Rau, T. Kirchartz, and T. Merdzhanova, “High-bandgap perovskites for efficient indoor light harvesting,” Advanced Energy and Sustainability Research, vol. 5, no. 5, p. 2 400 032, 2024. DOI: https://doi.org/10.1002/aesr.202400032. [16] M. Freitag, J. Teuscher, Y. Saygili, X. Zhang, F. Giordano, P. Liska, J. Hua, S. M. Zakeeruddin, J.-E. Moser, M. Grätzel, et al., “Dye-sensitized solar cells for efficient power generation under ambient lighting,” Nature Photonics, vol. 11, no. 6, pp. 372–378, 2017. DOI: 10.1038/nphoton.2017.60. [17] AnySolar, Ixolar high efficiency solarbit & solarmd,https://www.anysolar.biz/Show_all, Accessed June 2025, 2025. [18] ThorLabs, PDA100A2 - si switchable gain detector,https://www.thorlabs.com/thorproduct.cfm? partnumber=PDA100A2, Accessed June 2025, 2025. [19] D. Enescu, Green Energy Advances. IntechOpen, 2019. [20] P. Castrillo and J. M. Salgado, “Effective electrical resistance due to current-induced heat flow in thermoelectric generators,” in 2013 Spanish Conference on Electron Devices, IEEE, 2013, pp. 151–154. [21] L. Zöscher, J. Grosinger, U. Muehlmann, H. Watzinger, and W. Bösch, “RF Voltage Limiters for Passive Differential UHF RFID Front-Ends in a 40 nm CMOS Technology,” in 2015 IEEE MTT-S International Page 73 of 81
D2.2 - Initial Report on END Hardware Design Microwave Symposium, ISSN: 0149-645X, May 2015, pp. 1–4. DOI: 10.1109/MWSYM.2015.7166839. [Online]. Available: https://ieeexplore.ieee.org/document/7166839/?arnumber=7166839 (visited on 09/03/2024). [22] D. Dobkin, The RF in RFID - Passive UHF RFID in Practice, English. Elsevier Inc., 2008, ISBN: 978-07506-8209-1. (visited on 09/15/2024). [23] D. Poehl, “Design and Analysis of a 2.45 GHz RF Energy Harvesting Device,” English, Master Thesis, TU Graz, Apr. 2025. [Online]. Available: https://doi.org/10.3217/mdmdj-nq927. [24] L. Zoscher, P. Herkess, J. Grosinger, U. Muehlmann, D. Amschl, and W. Bösch, “A Differential Threshold Voltage Compensated RF-DC power Converter for RFID tag ICs,” in 2017 Integrated Nonlinear Microwave and Millimetre-wave Circuits Workshop (INMMiC), 2017, pp. 1–3. DOI: 10.1109/INMMIC.2017.7927297. [25] L. Zöscher, P. Herkess, J. Grosinger, U. Muehlmann, D. Amschl, and W. Bösch, “Passive Differential UHF RFID Front-Ends in a 40 nm CMOS Technology,” in 2017 47th European Microwave Conference (EuMC), Oct. 2017, pp. 105–108. DOI: 10.23919/EuMC.2017.8230810. [Online]. Available: https: //ieeexplore.ieee.org/document/8230810. [26] D. Poehl, U. Muehlmann, F. Amtmann, P. Thueringer, and J. Grosinger, “Design and Analysis of a 2.45 GHz RF Energy Harvesting Device,” en, in 2025 IEEE Wireless Power Technology Conference and Expo (WPTCE), 2025. DOI: 10.1109/WPTCE62521.2025.11062318. [27] N. Soltani and F. Yuan, “A Step-Up Transformer Impedance Transformation Technique for Efficient Power Harvesting of Passive Transponders,” en, Microelectronics Journal, vol. 41, no. 2-3, pp. 75–84, Feb. 2010, ISSN: 18792391. DOI: 10.1016/j.mejo.2009.12.010. [Online]. Available: https://linkinghub. elsevier.com/retrieve/pii/S0026269209002316 (visited on 11/02/2024). [28] B.-R. Heo and I. Kwon, “A Dual-Band Wide-Input-Range Adaptive CMOS RF–DC Converter for Ambient RF Energy Harvesting,” en, Sensors, vol. 21, no. 22, p. 7483, Nov. 2021, ISSN: 1424-8220. DOI: 10. 3390/s21227483. [Online]. Available: https://www.mdpi.com/14248220/21/22/7483 (visited on 01/25/2025). [29] M. Basim, D. Khan, Q. U. Ain, K. Shehzad, S. A. A. Shah, B.-G. Jang, Y.-G. Pu, J.-M. Yoo, J.-T. Kim, and K.-Y. Lee, “A Highly Efficient RF-DC Converter for Energy Harvesting Applications Using a Threshold Voltage Cancellation Scheme,” en, Sensors, vol. 22, no. 7, p. 2659, Mar. 2022, ISSN: 1424-8220. DOI: 10.3390/s22072659. [Online]. Available: https://www.mdpi.com/1424-8220/22/7/2659 (visited on 01/25/2025). [30] W. X. Lian, H. Ramiah, G. Chong, K. K. Pakkirisami Churchill, N. S. Lai, S. Mekhilef, Y. Chen, P.-I. Mak, and R. P. Martins, “A Fully-Integrated CMOS Dual-Band RF Energy Harvesting Front-End Employing Adaptive Frequency Selection,” IEEE Access, vol. 11, pp. 74 121–74 135, 2023, ISSN: 2169-3536. DOI: 10.1109/ACCESS.2023.3296145. [Online]. Available: https://ieeexplore.ieee.org/document/ 10185045/ (visited on 01/25/2025). [31] J. K. Yong, W. X. Lian, H. Ramiah, K. K. P. Churchill, G. Chong, N. S. Lai, Y. Chen, P.-I. Mak, and R. P. Martins, “A Fully Integrated CMOS Tri-Band Ambient RF Energy Harvesting System for IoT Devices,” IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 70, no. 12, pp. 4705–4718, Dec. 2023, ISSN: 1549-8328, 1558-0806. DOI: 10.1109/TCSI.2023.3304664. [Online]. Available: https: //ieeexplore.ieee.org/document/10224309/ (visited on 11/17/2024). [32] B. J. B. Deutschmann, U. Muehlmann, A. Kaplan, G. Callebaut, T. Wilding, B. Cox, L. V. der Perre, F. Tufvesson, E. G. Larsson, and K. Witrisal, Physically Large Apertures for Wireless Power Transfer: Performance and Regulatory Aspects, 2025. arXiv: 2503.06807 [eess.SP]. [Online]. Available: https: //arxiv.org/abs/2503.06807. [33] Y. Sun, F. Dang, C. Yuan, J. He, Q. Zhang, and X. Zhao, “A beam-steerable lens antenna for Ku-band high-power microwave applications,” IEEE Transactions on Antennas and Propagation, vol. 68, no. 11, pp. 7580–7583, 2020. DOI: 10.1109/TAP.2020.2979282. [34] E. Ahamed, R. Keshavarz, D. Franklin, and N. Shariati, “Software-defined programmable metamaterial lens system for dynamic wireless power transfer applications,” IEEE Transactions on Antennas and Propagation, vol. 72, no. 7, pp. 5840–5851, 2024. DOI: 10.1109/TAP.2024.3410643. [35] A. Anand, “A novel long range wireless power transmission system using Fresnel lenses for S-band EM waves,” in IEEE Radio and Wireless Symposium (RWS), 2025, pp. 116–118. DOI: 10.1109/RWS62086. 2025.10904830. [36] S. Zhong and X. Wang, “Wireless power transfer by beamspace large-scale MIMO with lens antenna array,” IEEE Transactions on Wireless Communications, vol. 18, no. 2, pp. 1390–1403, 2019. DOI: 10.1109/ TWC.2019.2892473. Page 74 of 81
D2.2 - Initial Report on END Hardware Design [37] S. Oh, D. H. Kim, W. Lee, J. Yang, C. Park, H. Lee, and S. K. Hong, “Millimeter-wave retrodirective wireless power transfer applicable to small electronic devices,” IEEE Journal of Microwaves, vol. 5, no. 2, pp. 322–332, 2025. DOI: 10.1109/JMW.2024.3518555. [38] K. Wu, W. Ni, T. Su, R. P. Liu, and Y. J. Guo, “Efficient angle-of-arrival estimation of lens antenna arrays for wireless information and power transfer,” IEEE Journal on Selected Areas in Communications, vol. 37, no. 1, pp. 116–130, 2019. DOI: 10.1109/JSAC.2018.2872363. [39] G. S. Lipworth, J. A. Hagerty, D. Arnitz, Y. A. Urzhumov, D. R. Nash, R. J. Hannigan, C. T. Tegreene, and M. S. Reynolds, “A large planar holographic reflectarray for Fresnel-zone microwave wireless power transfer at 5.8 GHz,” in IEEE/MTT-S International Microwave Symposium - IMS, 2018, pp. 964–967. DOI: 10.1109/MWSYM.2018.8439197. [40] O. M. Rosabal, O. A. López, S. Montejo-Sánchez, R. D. Souza, H. Alves, and M. Latva-Aho, “Average local EMF exposure and power consumption of a RIS-assisted WET system,” in 19th International Symposium on Wireless Communication Systems (ISWCS), 2024, pp. 1–6. DOI: 10.1109/ISWCS61526.2024.10639138. [41] O. M. Rosabal, O. A. López, V. D. P. Souto, R. D. Souza, S. Montejo-Sánchez, R. Schober, and H. Alves, “Wireless energy transfer beamforming optimization for intelligent transmitting surface,” arXiv preprint arXiv:2507.06805, 2025. [42] Z. Feng, B. Clerckx, and Y. Zhao, “Waveform and beamforming design for intelligent reflecting surface aided wireless power transfer: Single-user and multi-user solutions,” IEEE Transactions on Wireless Communications, vol. 21, no. 7, pp. 5346–5361, 2022. DOI: 10.1109/TWC.2021.3139440. [43] A. Azarbahram, O. L. A. López, B. Clerckx, M. Di Renzo, and M. Latva-aho, “Beyond diagonal reconfigurable intelligent surfaces for multi-carrier RF wireless power transfer,” in IEEE Wireless Communications and Networking Conference (WCNC), 2025, pp. 1–6. DOI: 10.1109/WCNC61545.2025.10978386. [44] A. Azarbahram, O. L. A. Lopez, B. Clerckx, M. D. Renzo, and M. Latva-Aho, Beamforming and waveform optimization for RF wireless power transfer with beyond diagonal reconfigurable intelligent surfaces, 2025. arXiv: 2502.19176 [eess.SP]. [Online]. Available: https://arxiv.org/abs/2502.19176. [45] A. Azarbahram, O. L. A. López, R. D. Souza, R. Zhang, and M. Latva-Aho, “Energy beamforming for RF wireless power transfer with dynamic metasurface antennas,” IEEE Wireless Communications Letters, vol. 13, no. 3, pp. 781–785, 2024. DOI: 10.1109/LWC.2023.3343563. [46] H. Zhang, N. Shlezinger, F. Guidi, D. Dardari, M. F. Imani, and Y. C. Eldar, “Near-field wireless power transfer with dynamic metasurface antennas,” in 23rd International Workshop on Signal Processing Advances in Wireless Communication (SPAWC), 2022, pp. 1–5. DOI: 10.1109/SPAWC51304.2022.9833917. [47] H. Zhang, N. Shlezinger, F. Guidi, A. Guerra, D. Dardari, M. F. Imani, and Y. C. Eldar, “Near-field beam focusing for wireless power transfer with dynamic metasurface antennas,” IEEE Internet of Things Journal, vol. 12, no. 12, pp. 18 596–18 605, 2025. DOI: 10.1109/JIOT.2025.3555113. [48] A. Azarbahram, O. L. A. López, and M. Latva-Aho, “Waveform optimization and beam focusing for nearfield wireless power transfer with dynamic metasurface antennas and non-linear energy harvesters,” IEEE Transactions on Wireless Communications, vol. 24, no. 2, pp. 1031–1045, 2025. DOI: 10.1109/TWC. 2024.3503908. [49] V. K. Papanikolaou, G. Zhou, B. Kaziu, A. Khalili, P. D. Diamantoulakis, G. K. Karagiannidis, and R. Schober, “Resolving the double near-far problem via wireless powered pinching-antenna networks,” arXiv preprint arXiv:2505.12403, 2025. [50] Y. Li, J. Wang, Y. Liu, and Z. Ding, “Pinching-antenna assisted simultaneous wireless information and power transfer,” arXiv preprint arXiv:2505.06240, 2025. [51] P. Chen, Y. Yang, B. Lyu, Z. Yang, and A. Jamalipour, “Movable-antenna-enhanced wireless-powered mobile-edge computing systems,” IEEE Internet of Things Journal, vol. 11, no. 21, pp. 35 505–35 518, 2024. DOI: 10.1109/JIOT.2024.3437201. [52] O. M. Rosabal, O. A. López, M. Di Renzo, R. D. Souza, and H. Alves, “Movable antennas-aided wireless energy transfer for the Internet of Things,” arXiv preprint arXiv:2506.21966, 2025. [53] O. L. A. López, H. Alves, S. Montejo-Sánchez, R. D. Souza, and M. Latva-aho, “CSI-free rotary antenna beamforming for massive RF wireless energy transfer,” IEEE Internet of Things Journal, vol. 9, no. 10, pp. 7375–7387, 2022. DOI: 10.1109/JIOT.2021.3107222. [54] O. L. A. López, D. Kumar, R. D. Souza, P. Popovski, A. Tölli, and M. Latva-Aho, “Massive MIMO with radio stripes for indoor wireless energy transfer,” IEEE Transactions on Wireless Communications, vol. 21, no. 9, pp. 7088–7104, 2022. DOI: 10.1109/TWC.2022.3154428. Page 75 of 81
D2.2 - Initial Report on END Hardware Design [55] A. Azarbahram, O. L. A. López, P. Popovski, and M. Latva-Aho, “On the radio stripe deployment for indoor RF wireless power transfer,” in IEEE Wireless Communications and Networking Conference (WCNC), 2024, pp. 1–6. DOI: 10.1109/WCNC57260.2024.10570703. [56] L. Van der Perre, E. G. Larsson, F. Tufvesson, L. D. Strycker, E. Björnson, and O. Edfors, “Radioweaves for efficient connectivity: Analysis and impact of constraints in actual deployments,” in 53rd Asilomar Conference on Signals, Systems, and Computers, 2019, pp. 15–22. DOI: 10.1109/IEEECONF44664. 2019.9048825. [57] G. Callebaut, J. Van Mulders, B. Cox, B. J. B. Deutschmann, G. Ottoy, L. De Strycker, and L. Van der Perre, “Experimental Study on the Effect of Synchronization Accuracy for Near-Field RF Wireless Power Transfer in Multi-Antenna Systems,” in 19th European Conference on Antennas and Propagation (EuCAP), Stockholm, Sweden, Mar. 2025, p. 4.94. [58] B. Deutschmann, C. Buyle, B. Cox, D. Delabie, M. Graber, T. Wildnig, K. Witrisal, and U. Mühlmann, System design study for energy-neutral devices interacting with the radioweaves infrastructure, version 2.0, Sep. 2023. DOI: 10.5281/zenodo.10548394. [Online]. Available: https://doi.org/10.5281/zenodo. 10548394. [59] J. Van Mulders, B. J. B. Deutschmann, G. Ottoy, L. De Strycker, L. Van der Perre, T. Wilding, and G. Callebaut, “Single versus Multi-Tone wireless power transfer with physically large arrays,” in 1st International Workshop on Energy Neutral and Sustainable IoT Devices and Infrastructure (EN-IoT 2024), Paris, France, Oct. 2024, p. 5.98. [60] J. C. Kwan, J. M. Chaulk, and A. O. Fapojuwo, “A coordinated ambient/dedicated radio frequency energy harvesting scheme using machine learning,” IEEE Sensors Journal, vol. 20, no. 22, pp. 13 808–13 823, 2020. DOI: 10.1109/JSEN.2020.3003931. [61] A. C. C. Chun, H. Ramiah, and S. Mekhilef, “Wide power dynamic range CMOS RF-DC rectifier for RF energy harvesting system: A review,” IEEE Access, vol. 10, pp. 23 948–23 963, 2022. DOI: 10.1109/ ACCESS.2022.3155240. [62] O. M. Rosabal, O. L. A. López, H. Alves, and M. Latva-Aho, “Sustainable RF wireless energy transfer for massive IoT: Enablers and challenges,” IEEE Access, vol. 11, pp. 133 979–133 992, 2023. DOI: 10.1109/ ACCESS.2023.3337214. [63] M. Wagih, A. S. Weddell, and S. Beeby, “Rectennas for radio-frequency energy harvesting and wireless power transfer: A review of antenna design [antenna applications corner],” IEEE Antennas and Propagation Magazine, vol. 62, no. 5, pp. 95–107, 2020. DOI: 10.1109/MAP.2020.3012872. [64] M. A. Ullah, R. Keshavarz, M. Abolhasan, J. Lipman, K. P. Esselle, and N. Shariati, “A review on antenna technologies for ambient RF energy harvesting and wireless power transfer: Designs, challenges and applications,” IEEE Access, vol. 10, pp. 17 231–17 267, 2022. DOI: 10.1109/ACCESS.2022.3149276. [65] S. Shen, Y. Zhang, C.-Y. Chiu, and R. Murch, “A triple-band high-gain multibeam ambient RF energy harvesting system utilizing hybrid combining,” IEEE Transactions on Industrial Electronics, vol. 67, no. 11, pp. 9215–9226, 2020. DOI: 10.1109/TIE.2019.2952819. [66] S. Shen, Y. Zhang, C.-Y. Chiu, and R. Murch, “An ambient RF energy harvesting system where the number of antenna ports is dependent on frequency,” IEEE Transactions on Microwave Theory and Techniques, vol. 67, no. 9, pp. 3821–3832, 2019. DOI: 10.1109/TMTT.2019.2906598. [67] D. Khan, S. J. Oh, K. Shehzad, M. Basim, D. Verma, Y. G. Pu, M. Lee, K. C. Hwang, Y. Yang, and K.-Y. Lee, “An efficient reconfigurable RF-DC converter with wide input power range for RF energy harvesting,” IEEE Access, vol. 8, pp. 79 310–79 318, 2020. DOI: 10.1109/ACCESS.2020.2990662. [68] S. Trovarello, G. Paolini, D. Masotti, and A. Costanzo, “Cascaded rectifiers for energy harvesting with a wide dynamic power range,” IEEE Journal of Radio Frequency Identification, vol. 7, pp. 64–73, 2023. DOI: 10.1109/JRFID.2023.3234805. [69] O. Cetinkaya, D. Balsamo, and G. V. Merrett, “Internet of mimo things: Uav-assisted wireless-powered networks for future smart cities,” IEEE Internet of Things Magazine, vol. 3, no. 1, pp. 8–13, 2020. DOI: 10.1109/IOTM.0001.1900064. [70] T. Ojha, T. P. Raptis, A. Passarella, and M. Conti, “Wireless power transfer with unmanned aerial vehicles: State of the art and open challenges,” Pervasive and mobile computing, vol. 93, p. 101820, 2023. [71] A. Tiurlikova, N. Stepanov, and K. Mikhaylov, “Wireless power transfer from unmanned aerial vehicle to low-power wide area network nodes: Performance and business prospects for lorawan,” International Journal of Distributed Sensor Networks, vol. 15, no. 11, p. 1 550 147 719 888 165, 2019. Page 76 of 81
D2.2 - Initial Report on END Hardware Design [72] J. Van Mulders, D. Delabie, C. Lecluyse, C. Buyle, G. Callebaut, L. Van der Perre, and L. De Strycker, “Wireless power transfer: Systems, circuits, standards, and use cases,” Sensors, vol. 22, no. 15, p. 5573, 2022. [73] J. Van Mulders, G. Leenders, G. Callebaut, L. De Strycker, and L. Van der Perre, “Aerial energy provisioning for massive energy-constrained iot by uavs,” in ICC 2022-IEEE international conference on communications, IEEE, 2022, pp. 3574–3579. [74] J. Van Mulders, J. Cappelle, S. Goossens, L. De Strycker, and L. Van der Perre, “Uav-based servicing of iot nodes: Assessment of ecological impact,” Sensors, vol. 23, no. 4, p. 2291, 2023. [75] J. Van Mulders, S. Boeckx, J. Cappelle, L. Van der Perre, and L. De Strycker, “Uav-based solution for extending the lifetime of iot devices: Efficiency, design and sustainability,” Frontiers in Communications and Networks, vol. 5, p. 1 341 081, 2024. [76] K. Koskinen, B. Xie, K. Ruttik, and R. Jäntti, “Li2BC: From visible light communication to ambient RF backscatter,” in Proc. 16th Ger. Microw. Conf. (GeMiC), Mar. 2025, pp. 346–349. DOI: 10.23919/ GeMiC64734.2025.10979106. [77] Analog Devices, “Ultralow Power Energy Harvester PMUs with MPPT and Charge Management,” ADP5091 / ADP5092 datasheet, May 2017, Rev. A. [78] KEMET Corporation, “Supercapacitors - FG Series,” datasheet, Dec. 2022. [79] S. A. Afghan and H. Géza, “Modelling and analysis of energy harvesting in internet of things (iot): Characterization of a thermal energy harvesting circuit for iot based applications with ltc3108,” Energies, vol. 12, no. 20, p. 3873, 2019. [80] Analog Devices, Inc., LTC3108 - ultra low voltage step-up converter and power manager, Datasheet, Rev. J, 2023. [Online]. Available: https://www.analog.com/media/en/technical-documentation/datasheets/ltc3108.pdf. [81] C. Lenz, S. Vostrikov, P. Mayer, and M. Magno, “From heat to power: Assessing thermoelectric energy harvesting for self-sustainable sensors,” in 2023 IEEE International Workshop on Technologies for Defense and Security (TechDefense), IEEE, 2023, pp. 384–389. [82] e-peas S.A., Aem20940 - energy harvesting power management ic for teg and pv sources, Datasheet, Revision 1.10, 2022. [Online]. Available: https://e-peas.com/documents/AEM20940/DS-AEM20940. pdf. [83] D. Carli, D. Brunelli, L. Benini, and M. Ruggeri, “An effective multi-source energy harvester for low power applications,” in 2011 Design, Automation & Test in Europe, IEEE, 2011, pp. 1–6. [84] Y. K. Tan and S. K. Panda, “Energy harvesting from hybrid indoor ambient light and thermal energy sources for enhanced performance of wireless sensor nodes,” IEEE Transactions on Industrial Electronics, vol. 58, no. 9, pp. 4424–4435, Sep. 2011, ISSN: 02780046. DOI: 10.1109/TIE.2010.2102321. [85] X. Cui, J. Zhang, H. Zhou, and C. Deng, “Powerpool: Multi-source ambient energy harvesting,” in 2020 6th International Conference on Big Data Computing and Communications (BIGCOM), IEEE, 2020, pp. 86– 90. [86] T. Kang, S. Kim, C. Hyoung, S. Kang, and K. Park, “An energy combiner for a multi-input energyharvesting system,” IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 62, no. 9, pp. 911– 915, 2015. [87] J. Colomer-Farrarons, P. Miribel-Catala, A. Saiz-Vela, and J. Samitier, “A multiharvested self-powered system in a low-voltage low-power technology,” IEEE Transactions on Industrial Electronics, vol. 58, no. 9, pp. 4250–4263, 2010. [88] P.-D. Gleonec, J. Ardouin, M. Gautier, and O. Berder, “Architecture exploration of multi-source energy harvester for iot nodes,” in 2016 IEEE Online Conference on Green Communications (OnlineGreenComm), IEEE, 2016, pp. 27–32. [89] S. Saggini and P. Mattavelli, “Power management in multi-source multi-load energy harvesting systems,” in 2009 13th European Conference on Power Electronics and Applications, IEEE, 2009, pp. 1–10. [90] S. Chamanian and P. P. Mercier, “Mipsimo: A multi-input piezo-adaptive single-inductor multi-output energy harvester achieving using a shared inductor with an integrated analog computer achieving 95% mppt efficiency,” IEEE Solid-State Circuits Letters, vol. 5, pp. 222–225, 2022. [91] G. Shi, J. Chang, Y. Xia, X. Wang, H. Xia, and Q. Li, “A multisource collaborative energy extraction circuit for vibration, ambient light, and thermal energy with mppt and single inductor,” IEEE Transactions on Industrial Electronics, vol. 70, no. 6, pp. 5819–5829, 2022. Page 77 of 81