A harvesting review that is really about the array power budget

What the work claims

This is a review, not a primary experiment: Lin, Garoli, Proietti Zaccaria and colleagues from IIT Genova, Cambridge, Modena, St Andrews, Linkoping, ETH Zurich, DTU, and Henan University of Technology synthesize recent progress in triboelectric nanogenerators (TENGs), piezoelectric and pyroelectric transducers, indoor photovoltaics, radio-frequency rectennas, and thermoelectric generators, and then argue that the materials-centric narrative misses the system.¹ The headline advances they credit: liquid-solid TENGs that eliminate mechanical wear from repeated solid-solid contact; nonlinear piezoelectric oscillators that broaden operational bandwidth by over 300 percent; machine-learning-accelerated discovery of high charge-density dielectrics; and multiband RF harvesting built on metamaterial antennas.¹

Their critical turn is the second half of the paper. A harvesting chain is only as good as its weakest link between transducer and load, and the authors identify three links that papers rarely quantify: power management integrated circuits that must cold-start from millivolt-level sources, the impedance-matching problem when hybrid sources feed one converter, and the persistent degradation of micro-supercapacitors and thin-film batteries under realistic field conditions rather than on the bench.¹ The central design criterion they advance is energy closure: a system works not when its transducer peaks are impressive, but when the full harvest-condition-store-load loop sustains the application indefinitely.¹

How it works

The review's source table gives the honest starting point, and the spread is wide. Contact and mechanical-motion harvesters (TENGs) deliver microwatts to milliwatts, with a reported peak above 10.8 watts per square meter under favorable conditions; piezoelectric transducers reach up to 140 milliwatts; thermoelectric generators sit at nanowatts to microwatts, with 7.4 milliwatts per square meter reported only in a solar-driven hybrid; indoor photovoltaics are spectrum-dependent, with 106.25 microwatts per square centimeter predicted and a 28.38 percent power-conversion efficiency demonstrated under white LEDs; and a single RF rectenna unit delivers 562 microwatts from an incident power density of only 40 microwatts per square centimeter.¹ Every one of those numbers comes from a different primary study under different conditions, which is itself one of the review's quiet warnings.

Between source and storage sits the power management IC, and here the review is most concrete. Cold start is the hard problem: the PMIC must energize its own oscillator and charge pumps before any efficient conversion can begin. The showcase result, from Bose and colleagues in IEEE Journal of Solid-State Circuits, is an integrated electrical cold-start that fires an inductive boost converter from 57 millivolts, reaching startup in 135 milliseconds while drawing only 90 nanojoules from the source.² Earlier thermoelectric interfaces managed startup from roughly 35 millivolts only with mechanical assistance.¹ Storage has improved on paper: one micro-supercapacitor reaches 80.7 microfarads per square centimeter with 98.3 percent capacitance retention after 100,000 cycles, and thin-film devices claim energy densities up to 15 microwatt-hours per square centimeter per micron of active thickness.¹ But the authors insist that retention after cycling is not the metric that matters for a device deployed for years; storage aging, self-discharge, and encapsulation degradation under field exposure are, and those are measured far less often.¹

The cardiac case study is the mechanism made flesh. Liu and colleagues built a battery-free intracardiac pacemaker: a catheter-deliverable capsule of 1.75 grams and 1.52 cubic centimeters carrying a POM/PTFE triboelectric harvester, rectification, capacitive storage, and pacing electronics.³ In vivo in swine it produced about 6.0 volts at 0.2 microamps, equivalent to 0.026 microjoules per cardiac cycle and a maximum output power of 0.039 microwatts.¹ That is a deeply sub-microwatt source, yet a 10 microfarad capacitor charged to 3 volts sustained pacing for nearly 40 seconds, and an atrioventricular-block model was paced at about 1.5 volts, with function maintained over a three-week follow-up in which week-three pacing raised heart rate from about 90 to 108 beats per minute.¹ A follow-up symbiotic transcatheter pacemaker using electromagnetic induction with a magnetic-levitation energy cache sustained simultaneous energy regeneration and therapy for one month in a porcine bradyarrhythmia model.¹ The trick in both is temporal decoupling: the harvester trickles energy into storage continuously, and the load draws it in short pulses. Peak power never has to match.

Where a skeptic should push

The most load-bearing assumption is that the review's assembled numbers are comparable, and they are not. Power density figures for TENGs, piezo elements, photovoltaics, and rectennas come from different groups with different drive conditions, load matching, and reporting conventions; several harvesting papers quote open-circuit voltage where deliverable power is the quantity that matters. The authors know this and lean on the system-level framing to sidestep it, but any reader who lifts a single number from the source table is doing so against the review's own advice.¹

Second, the two strongest system demonstrations are both cardiac, and both succeed partly because a beating heart is an almost ideal energy source: rhythmic, high-amplitude, unceasing, and co-located with the load. The review's own in-pipe and smart-glasses case studies operate in far poorer energy environments with more conditional results.¹ Transferring the energy-closure lesson to a stationary, warm, dark, vibration-poor environment requires checking whether the source assumptions survive the move. Third, the degradation critique is asserted more than quantified in the sections this analysis verified: the review names storage aging, thermal management, and encapsulation failure as the unmeasured risks, and calls for accelerated-aging studies, but offers few field numbers of its own.¹ That is a fair gap for a review to flag rather than fill, but it means the reliability half of the story is a research agenda, not a result.

What energy closure means for untethered arrays

The non-obvious implication concerns what an autonomous microelectrode array node actually spends its power on. A modern high-density array front end, with thousands of low-noise amplifiers, digitizers, and serializers, consumes orders of magnitude more than 0.039 microwatts; nobody will harvest their way to continuous full-bandwidth recording from ambient energy. The honest opportunity is narrower and more interesting. The cardiac case proves that a sub-microwatt source can support a microwatt-class instrument if the architecture accumulates during silence and spends in bursts. For array hardware that is a recognizable pattern: duty-cycled acquisition, where the node sleeps at nanowatt quiescent power, wakes on a schedule or a trigger, records a burst of compressed events, and transmits opportunistically. The cold-start PMIC is the enabling device class here: startup from 57 millivolts in 135 milliseconds drawing 90 nanojoules is precisely the subsystem that decides whether a battery-free node can wake reliably at all, and its quiescent current, not the transducer, sets how long the node can sleep.²

The threat is equally specific, and it is electrical. The converter that boosts 57 millivolts is a switching circuit, and switching circuits are exactly what a microvolt front end cannot tolerate nearby. The review treats impedance matching and conversion overhead as first-order constraints at microwatt scale;¹ an array designer should read that as a layout and topology constraint: the harvest-condition block must be electrically and physically divorced from the analog input stage, with storage as the firewall between them, or the harvested supply will modulate the noise floor it powers. There is a second, subtler trap in storage. The micro-supercapacitor numbers are bench numbers; a node parked inside a 37-degree incubator at high humidity for months faces precisely the unmeasured aging the review warns about, and a swollen or leaking storage element over living tissue is a contamination event, not just a reliability failure.¹ Finally, the environment problem deserves plain statement: an incubator is warm, dark, still, and RF-shielded by its metal shell, close to the worst harvesting environment imaginable. Energy closure for array instrumentation is realistic for wearable, epicardial, perivascular, and flow-loop configurations long before it is realistic for the incubated dish. The vendors who understand that distinction, and design burst-mode nodes for the environments that actually have energy, will own the autonomous-instrumentation niche; those who promise battery-free incubated MEAs are selling the review's figures out of context.

The bottom line

Established, with the caveat that this is a synthesis of other groups' measurements: transducers spanning TENG, piezo, indoor photovoltaic, and RF rectenna sources can deliver microwatts to milliwatts; integrated cold-start power management now starts from 57 millivolts at 90 nanojoules; and a complete harvest-store-load chain has kept a cardiac pacemaker running in a living pig on 0.039 microwatts of average harvested power for weeks.¹³ Not established: field longevity of the storage elements those systems depend on, comparability of the quoted source power densities, and any demonstration of harvesting near a microvolt-resolution front end. The claim that energy closure is the right design criterion is confirmed every time a duty-cycled node survives a week in culture; it would break the first time a cold-start PMIC's switching noise is shown raising the input-referred noise floor of the amplifier it feeds. For MEA instrumentation the review's value is that it relocates the engineering problem from the electrode to the power chain: the autonomous array node is a power-management design that happens to carry electrodes, and the parts that decide whether it works, cold start, quiescent current, storage aging, and converter isolation, are the parts nobody puts on the brochure.