A little opportunity popped up recently, in conjunction with element14, to test an energy harvesting module. As there seems to be a lot of development of low-powered sensors and radios which can run from coin-cells for extended periods, the whole idea of being able to eliminate batteries or extend the lifetime of these devices can be quite attractive. As I am a photovoltaics/solar energy engineer by qualification and an interested electronics hobbyist with some equipment, I decided to jump on the chance to see what was on offer and what it was capable of doing.

 

Introduction

Energy harvesting, at its core, is about trying to capture enough ambient energy and transforming it such that it can be used to achieve a purpose. Common ambient energy includes light, radio waves, heat or even the kinetic action of pressing a button. Note that this is not “free” energy as such (the laws of physics are conserved), but it could include energy which may not have performed much in the way of other useful tasks (e.g. lighting up an unused corner or a room). Assuming you can do this unobtrusively with sufficient efficiency and output, you could (in theory) make devices which can run on harvested energy alone, eliminating the need for a battery altogether.

 

This concept is not new, although many years ago, it was difficult to achieve in practice due to the (generally) low ambient energy available and the losses associated with quiescent draw of integrated circuits and limited conversion efficiencies. In recent years, however, better power converter designs, higher quality fabrication processes and the cost-effective availability of components like supercapacitors have increased the potential of energy harvesting to power modern IoT style devices especially when combined with greater power efficiencies available in modern microcontrollers in sleep mode and effective, robust, coded, low-powered digital silicon radios.

 

A key player in this field that I have been aware of for a number of years is EnOcean. They have a suite of sensors and controls, paired with their own proprietary radio technology which claims to operate from harvested power alone. While this was an interesting product, it had its limitations in cost and proprietary nature. A number of IC and component manufacturers have developed their own energy harvesting kits to allow for powering other devices – including solutions from STMicroelectronics and Texas Instruments. Recently, even Cypress/Infineon introduced their own evaluation kit for an energy harvesting BLE beacon based around their own power management ICs.

 

Amongst the available ambient energy options, visible light harvesting is perhaps the most common, owing to the almost continuous cost-reduction in commercially available solar cells and good efficiencies which commonly exceed 20%. While most common roof-top solar modules use crystalline silicon cells which offer high efficiencies and low prices, many of the lower power harvesting technologies have used thin-film silicon technologies including the amorphous silicon on glass cells (often brown in colour). These cells are not as efficient in part due to their thin structure limiting their absorbance and the quality of materials. They often suffer some light-induced degradation and shorter lifetimes, but are generally cheaper due to less silicon requirement and are claimed to have better performance with diffuse light sources as is usually the case in indoor environments. This is why you will often find this type of cell in a solar calculator, for example.

 

Silicon solar cells do have some limitations, however. Crystalline cells are rigid by their nature and cannot be flexed, while their processing involves high-temperature processes and the use of harsh chemicals such as hydrofluoric acid. Thin films can be deposited on more lightweight, flexible substrates, although many commercial options are quite thick and limited in flexibility. Thin films, however, still remain in the minority in terms of the market. Other semi-flexible options include the dye sensitised solar cell, inspired by plants, made with used exotic nanoparticles with a liquid electrolyte, offering slightly lower efficiencies and shorter lifetime than the best thin-film cells. There is also a risk that such cells would leak electrolyte. Despite that, it has seen application in the consumer space, for example, this backpack (although, it’s not particularly practical at just 0.5W).

 

To some extent, a natural evolution of this idea is organic photovoltaics. These cells use conductive organic polymers or organic molecules to produce polymer solar cells which can be made using low-temperature solution processes, potentially enabling high-throughput roll-to-roll processing. Such cells are highly flexible and lightweight compared to the other technologies and can be made at low costs. Unfortunately, these cells have had problems with photochemical degradation over time resulting in stability issues and lower efficiencies compared to conventional silicon cells. Laboratory record efficiencies have steadily improved, with energy harvesting applications perhaps being an ideal application due to the lower intensity of light (thus reduced degradation rate) and cost-sensitivity.

 

The Kit and its Features

The kit arrived in a branded cardboard box with a draft specification sheet. The package was so light that I thought they might have not put anything into that box – it weighed just 38 grams including everything inside the box!

Inside the box is instructions for use (repeated from the spec sheet), a “welcome” message, a business card with the technical contact details and the device itself packed inside bubble wrap.

The device itself has a size a bit taller than a business card, as it features a 50x50mm 6-cell organic PV cell from their light energy harvesting range. This organic cell has a mottled deep aqua blue appearance with divisions between the cells clearly visible. The cell itself is quite thin – it feels just like a sheet of laminated paper that has been through a pouch laminator. Contacts are made to the right of the cell using crimps which appear to pierce through the substrate, connecting the cell to tabs which are soldered on the rear.

 

The underside has the two contacts for the organic PV cell, along with a logo and a warning that the kit is for evaluation only.

Straight out of the box, the kit is warped – but this is not a problem! In fact, it’s one of the key features of the cell – it is flexible and glued to a flexible flat PCB backing which neatly illustrates how well it can bend. It is also extremely thin and low profile, although the circuitry itself is not flexible so sits on a very thin fibreglass PCB.

 

That being said, while the cell can be flexed to conform to different shapes, the usefulness of this will really depend on the application as often this would lead to sub-optimal angles between the cell and the light source and potentially even self-shadowing which would reduce the harvested energy. Perhaps the flexibility will add reliability and robustness in some applications, but I suppose low cost and light weight are the more desirable properties. Because the stability of organic panels are a potential limitation, the application of these cells into low-intensity “indoor” energy harvesting situations appears to be a sensible applications.

The kit consists of Wago 2059 0.5mm2 connectors for input and output, accepting wires from 0.14-0.34mm2/22-26AWG stripped to a length of 4-5.5mm. Configuration of the output voltage is made with SW1 and SW2, while SW3 allows for switching in a primary battery back-up to supply any shortfall of energy from the harvesting solution. The harvested energy is stored in the CAP-XX GA230 0.4F 5.0V two-cell supercapacitor in the centre of the image. Energy is harvested by the e-Peas AEM10941 “Highly-Efficient, Regulated Dual-Output, Ambient Energy Manager for up to 7-cell solar panels with optional primary battery”, with configuration made by the three resistors underneath marked CGF. This harvesting solution features maximum power-point tracking (MPPT) to more efficiently extract energy from the cell, especially important for cells with poor fill factors or under variable lighting conditions. The boost inductor can be seen in the bottom left corner.

 

While the AEM10941 is capable of two rails of output, this does not seem to be used directly as the output currents are rather limited and perhaps insufficient to meet the peak current demands of higher-powered radios. Instead, a highly efficient buck converter IC from Texas Instruments TPS62740 “36-nA Iq Step Down Converter for Low Power Applications” is used to provide the output voltage. It seems there may have been an unrouted trace in the design, resulting in a copper “bodge wire” being seen just to the left of the IC. It appears this converter uses a chip inductor instead. As this is an early preview evaluation kit, the assembly appears to be hand-made with slightly inconsistent SMD soldering.

 

Test pads (rather than loops) are provided for some measurements, although they are all in the form of flat pads making it less convenient for attaching probes semi-permanently but does preserve the low-profile appearance. Unfortunately, there does not seem to be an easy way to access the organic PV cell for testing on its own (independent of the harvesting circuit) without desoldering it from the flat flexible PCB. This is something I probably won’t attempt – overheating the plastic substrate may lead to rapid and permanent damage.

 

The draft preliminary documentation supplied provides some key information about the kit and how it can be used. It could be very much improved, however, as it omits quite a lot of key information which can be useful to evaluators, for example, a product schematic. The purpose of TP2, TP5, TP6 and TP7 are not detailed and the existence of the Texas Instruments buck converter is not detailed either. While graphs are supplied of the output behaviour with the X-axis labelled in minutes, the load attached to the kit during those tests was not detailed. The specification of the organic PV cell is not provided either – specifically efficiency, I-V curve, operating irradiance range or lifetime. Even the information about the connectors and suitable wires was not provided. To round it all off, there is even a typo on the front page – “itslef” rather than “itself”. While it is possible to determine some of this with a bit of research on the user’s behalf, it’s often much appreciated if we are not left to “discover” these details on our own.

 

Energy Harvesting Basics

In essence, the evaluation kit works as follows:

The ambient light energy is collected by the light energy harvesting cell, through the AEM10941 into the supercapacitor. Likewise, a primary battery can be connected which also charges up the supercapacitor in case of a shortfall of incoming energy. The supercapacitor is drawn by the TPS62740 to produce the output voltage as configured by the switches.

We can think of the supercapacitor like a “bucket” that stores our energy. The input will “fill up” our bucket from empty to up to 4.5V, just a little less than the rated 5V probably to protect the capacitor’s lifetime due to the need to ensure both cells are balanced and below their maximum voltages. The output converter is a buck converter, so it will draw down from this bucket as soon as it exceeds about 3.9V. Because it is a buck converter, it is only capable of stepping down the voltage, thus if the output is 3.3V, the converter will shut down once it falls below this level.

 

Because of the way it works, if your output draws more energy than the input, the “useable capacity” of the capacitor oscillates between the output voltage (1.8V – 3.3V) and the starting voltage (3.9V). If your output draws less energy than the input, then the capacitor can be used up to 4.5V.

 

The total energy stored in the capacitor can simply be calculated with E = C.V^2 / 2, although the available energy will also depend on the leakage current of the capacitor, quiescent draw of the output buck converter and its efficiency. The maximum usable stored energy is given by a draw less than the input and an output voltage of 1.8V - 0.4*4.5^2/2 – 0.4*1.8^2/2 = 3.402 Joules (or Watt-seconds). However, if the draw exceeds the input, the worst case is given by an output of 3.3V, resulting in a stored energy of 0.4*3.9^2/2 – 0.4*3.3^2/2 = 0.864 Joules.

 

The key of successful energy harvesting is energy budgeting as we cannot violate the laws of physics!

If the supply exceeds the load, the voltage of the supercapacitor will increase and hold nearly steady at the maximum voltage, maximising the storage available and ensuring continuous operation. At an even balance, the supercapacitor voltage will remain stable and the load will operate continuously although without any “margin” against changes in input or loss of efficiency of the cell over time.

 

In the case where the load exceeds the supply, the kit operates in an intermittent mode, where a more limited amount of energy storage exists in the capacitor. The advantage of this is purely that high power operations (e.g. sending a message) can be sustained intermittently with the energy being collected over long periods to make it happen (e.g. hours).

 

In many real-life cases, the loads are highly variable and can, at times, have a high peak to average ratio (e.g. burst transmission versus sleep). The energy in the supercapacitor and the 300mA-rated buck converter allows for meeting the needs of these bursts. So as long as (on average) the input power meets or exceeds the output and the storage is sufficient to tie-over these bursts, continuous operation can also be assured.

 

Performance Testing

Testing the performance of the energy harvesting kit in absolute terms is rather difficult. The energy levels we are dealing with are relatively limited and incorrect test setup can lead to perplexing results. Likewise, I do not have any radiometric equipment to measure absolute irradiance, standard xenon “solar simulator” light sources or the spectral distribution of the light sources.

 

Because very small currents are involved, regular power supplies and electronic loads would not be accurate enough to perform testing with. While I do have a Keithley 2450 SMU from the recent RoadTest, it is currently engaged in a long-term test and will soon have to go back for repair, so instead I decided to take a different approach to characterising the kit. Using my Rohde & Schwarz RTM3004 with 10:1 passive probes (10Mohm impedance), I monitored the voltage on the organic PV cell, the supercapacitor and the output. The high input impedance reduces the effect of loading on the circuit, almost to the point it is negligible – at 1.8V, this would represent 1.1664mJ, which is at least two orders of magnitude less than the values we are attempting to measure. Calculations of energy yield are made by observing capacitor voltage over time which makes a key assumption that the capacitor is “exactly” 0.4F and does not take into account the losses in the oscilloscope probes. This does, however, include any of the quiescent losses the unit may have but excludes the efficiency loss of the output buck converter. The output voltage was set to 1.8V for this testing as this maximised the voltage range of the capacitor that could be used for testing, reducing errors. A 100 ohm resistor was used as a "dump" load so the accumulated energy could be depleted for a fresh recharge cycle.

 

In testing dark quiescent consumption, I measured the capacitor voltage initially and after some time with a Keysight U1241B 4-digit handheld DMM, thus removing “ongoing” leakage from the equation. I placed the whole kit inside a disused fridge with the door closed to ensure darkness.

 

Instead, I will look at its performance based on use in typical scenarios in the home office and under fluorescent and LED desk lamps. To gain a gauge of the brightness of the sources, I used the digital light sensor in my Xiaomi Redmi Note 8T phone which reports the lux brightness up to 32,000 lux. This is unlikely to be accurately calibrated, while variations due to angle of source and field uniformity of the light are not guaranteed. Note that as lux is based on lumens (a wavelength-weighted measurement), it is not a radiometric unit and thus the spectral distribution is important! However, the results still provide useful information in terms of the range of power you can expect to extract with such a kit under various situations.

 

Testing was performed without making any modifications to the kit, which unfortunately seems to preclude testing the I-V characteristics of the organic PV panel.

 

Energy Yield

I begin with the most important question – how much energy can this kit extract from the environment. The first test was my office workbench at home, where I measured 220 lux from a cool-white LED fixture which is a fairly “modest” amount of light.

With no load connected, the voltage went from 3.9V up to 4.25V in the span of 5500 seconds. Using some basic math, this means that 0.5705J of energy was extracted in 5500 seconds, or about 103.7µW.

 

To verify my calculations, I decided to put a resistive load on the 1.8V output that would be very close to this power draw to see if the capacitor voltage would remain constant. The calculated resistance is about 31.2kΩ. Looking in my junk bin, the best I could muster was 36kΩ made of two 18kΩ in series.

Over the same period, the capacitor voltage is seen to rise, extremely slowly, thus validating our calculated power yield as being very likely to be accurate (assuming the capacitor size is exactly 0.4F). To be sure this was not “by chance”, I then cut the series resistors apart to make the load go open circuit – the output voltage can be seen to go “noisy” at this time (~900s) and the capacitor voltage rises more quickly, reaching the saturation voltage of 4.5V by 3300s resulting in no charge until 5100s when the voltage has fallen far enough to start charging again.

 

Fluorescent Lamp Tests

Under neutral-white fluorescent lighting of 500 lux, the capacitor went from 1.9V to 3.9V in 3380s, resulting in a power of 686.4µW. Under 1000 lux, this took 2830s for a power of 819.8µW. This brightness would be considered a very bright workbench.

At 2000 lux, it completed the cycle in 1050s, delivering 2.21mW of power. With 4000 lux and the tube almost touching the cell, this was cut to 555.5s delivering 4.18mW. But these cases are unrealistic in the case of sitting something on a bench, so having highly efficient circuitry and reasonable expectations is important.

 

Warm White LED Lamp Tests

Trying to push this further, I used a warm white 11W OSRAM LED bulb, providing a measured 30,000 lux which allowed the unit to cycle in 308 seconds and deliver 7.53mW. At the beginning of the test, it's clear that the energy harvesting chip had not identified the panel's MPP due to the rapidly varying lighting condition, but settles into the right region after some time. It seems that every five seconds, it makes a determination of the MPP and the voltage goes up to Voc momentarily. Pushing the LED light source to near-contact with the panel exceeded the capability of my phone’s sensor, but it shortened the time to 287.5s delivering 8.07mW. The level of brightness was almost painful to look at and may eventually lead to degradation of the cell.

 

A summary of the results is as follows:

  • 220 lux neutral white LED – 103.7µW.
  • 500 lux neutral white fluorescent – 686.4µW.
  • 1000 lux neutral white fluorescent – 819.8µW.
  • 2000 lux neutral white fluorescent – 2.21mW.
  • 4000 lux neutral white fluorescent – 4.18mW.
  • 30000 lux warm white LED lamp – 7.53mW.
  • >32000lux warm white LED lamp – 8.07mW.

 

As a result, for continuous operation ideally it would be used to power devices with an average consumption up to the 100µW range and for brighter environments, even into the 1mW range. At higher average consumptions, either intermittent operation with potentially lengthy charging periods (around an hour) or battery back-up will be necessary. It should be noted that compared to direct battery connection, the use of battery through the back-up port will suffer double losses from the energy harvester’s boost converter and the output buck converter, so significant opportunities to harvest energy without the need to draw on the battery would be necessary to make this tradeoff worthwhile. Likewise, the increased size, cost and complexity compared to a battery (primary or rechargeable) would need to be considered in the context of product lifetime and extended life enabled by such a solution.

 

Dark Quiescent Losses

Out of curiosity, I wondered how quickly the supercapcitor might discharge if left in a perfectly dark environment. I charged the unit to the maximum voltage using a high-powered LED torch, measured the capacitor voltage and then left it inside a sealed disused fridge to ensure a perfectly dark environment. About three and a half hours later, I removed the unit from the fridge and re-measured the capacitor voltage.

 

The unit spent 12780s in the dark, registering an initial voltage of 4.476V and a final voltage of 4.234V. By my calculations, a loss of 0.4215264 Joules occurred during this time, which points to a quiescent power loss of 32.986uW or a current of 7.574µA (at the average voltage of 4.355V). From the datasheet, it seems 0.5µA may be lost into the AEM10941 and another 0.46uA to the TPS62740 and another 1-2µA to the HA230. The measured loss is about twice as much possibly due to other unaccounted losses due but is still an impressively small number that means the capacitor itself would take over 34 hours to completely discharge assuming a constant quiescent power draw.

 

Usage Scenarios

To gauge its usefulness, I tried to apply the kit to a number of usage scenarios. Unfortunately, as I am not working on any low-powered designs myself and have given away my Omron 2JCIE-BL01 BLE Environmental Sensor board, I was left to test with an Adafruit Feather Huzzah (ESP8266) I received from an element14 care package in the past, the Cypress BLE Mesh Evaluation Kit and a theoretical calculation based on the Omron 2JCIE-BL-01’s power consumption I had previously measured.

 

Preliminary LED Test

Initial testing of the output focused on powering a low-intensity green LED with 100-ohm series resistor, using an output of 3.3V. The combination was kept running for 19.5 seconds with a calculated (approximate) current draw of around 12mA. This illustrates the power of the supercapacitor to provide energy storage sufficient to power higher current operation on an intermittent basis.

Unfortunately, this also illustrates a shortcoming of such a solution. While supercapacitors offer longer cycle life and more rapid charge acceptance/delivery compared to rechargeable batteries, under load, their voltage changes rather rapidly. When the voltage falls to within around 0.2V of the output, the buck converter is unable to maintain regulation, falling into an “unregulated” mode of operation until the voltage falls to about 0.2V below the specified output voltage before the output shuts off. While this variation in output voltage is unlikely to cause immediate damage to connected devices, a change in supply voltage can cause sensors to report values which are less accurate than if fed a constant supply voltage. The present design offers no connectivity to downstream microcontrollers to inform them of capacitor state or to “shut down” the output if unnecessary in the case of intermittent operation, thus relying on the load design to have a very low standby current if continuous operation is desired.

 

Modified Adafruit Feather Huzzah ESP8266 Wi-Fi (2.4V) + Bosch Sensortec BMP280

I started off with a more audacious plan – to try and make an intermittently-powered Wi-Fi temperature sensor using an Adafruit Feather Huzzah ESP8266-based development board and a Bosch BMP280 sensor. Unfortunately, in its original configuration even using deep sleep, it consumed close to 20mA in its lowest powered state. Removing the 3.3V regulator and the battery charger IC and operating the unit down at 2.4V was able to reduce the quiescent current to 800uA, which is still pretty high compared to the datasheet values but perhaps is due to the connected programming auto-reset circuitry. Operational current peaked at above 80mA.

 

The board was programmed with a simple set of code that connects it to my home Wi-Fi using a static IP, reads the sensor value, sends a UDP packet and puts itself to sleep for 60 seconds. Using it in intermittent mode, with the capacitor charged to its starting voltage, the circuit was capable of sending two or three datagrams before it depleted the capacitor to below 2.5V. Recharging the capacitor under ordinary ambient light in my room took more than an hour to receive another set of readings.

The kit was not intended for use with such “heavy” radios and understanding the power consumption, it is no surprise that intermittent operation is the only possibility with this combination. However, it was still impressive to see the output buck was sufficiently beefy and the supercapacitor’s ESR was sufficiently low to enable Wi-Fi connection and transmission in a reliable manner.

 

Cypress BLE Mesh Evaluation Kit (1.8V) + PIR Sensor

My second use case focused on Bluetooth Low-Energy Mesh, which I had a tinker with in a previous RoadTest. Programming the PIR sensor demo configured with the LOW_POWER_NODE flag active, paired with an always-on mesh helper node and operating at its lowest voltage of 1.8V, I was hoping to see some possibility of continuous operation as PIRs generally are quite low current.

In this case, the kit was able to maintain surveillance for close to half-an-hour using the stored energy, before running out of energy and recharging over the space of around an hour. Unfortunately, despite opening the jumpers for the serial interface to prevent current leakage, the other sensors and components (thermistor, ambient light sensor, external flash) on the development board likely consumed too much current to allow for continuous operation.

 

Omron 2JCIE-BL01 (Calculated, Theoretical)

Unfortunately, I don’t have my Omron 2JCIE-BL01 board, but this is a BLE multi-sensor board that is commercially available and optimised to operate for long periods from a CR2032 battery. This type of device is perhaps most suited for powering from an energy harvester kit.

 

From my previous experiments, using a beacon interval of one second, the board averaged 0.84mA consumption with a peak of 11.33mA during transmission. The sleep current was a very tame 4.85µA.

 

Assuming a room with a modest lighting level of 200 lux offering 100µW of power, at 3V, this would correspond to a current of 33.3µA. This means that the sleep current of the 2JCIE-BL01 leaves around 28.48µA going into the super-capacitor. A back of the envelope suggests that transmission peaks of 11.33mA occur for about 74ms for each read-and-beacon-broadcast, which consumes a capacity of 838µAs of energy. Based on a surplus of 28.48µA, this implies a broadcast can be afforded as frequently as every 30 seconds providing values for temperature, humidity, pressure, light, noise and acceleration. That is quite a useful outcome, considering that for room comfort, high frequency monitoring may not be necessary.

 

Conclusion

The idea of harvesting ambient energy to power electronics that do not require a battery is certainly one that has merit, especially as advances have been made in smart, low-power sensors and microcontrollers, digital silicon radios, switching converter/power management integrated circuits and photovoltaic cells.

 

The Epishine light energy harvesting module evaluation kit provides a platform to demonstrate their highly flexible, lightweight and thin organic photovoltaic cells for this application. The use of organic photovoltaics for this application has other additional benefits including lower costs and better performance with diffuse sources, whilst possibly avoiding the drawbacks of instability and photodegradation which typically occur more rapidly under strong lighting. This allows the potential for designers of sensor products to integrate energy harvesting into their designs rather than relying on existing products which may have proprietary elements.

 

The early-access unit I received was supplied with a limited amount of documentation which could be much improved and its design limited testing in its unmodified form. The kit was able to demonstrate its ability to provide energy continuously and in intermittent mode using a supercapacitor as a buffer. The kit delivered energy from my bench in the 100µW range up to about 1mW given ordinary intensities of lighting you might encounter in a home, office or workshop. As the rules of physics are firm, in order to reap the rewards of ambient energy, a fully optimised energy-efficiency-first design is necessary to ensure the ability to operate continuously over a range of ambient lighting conditions – there is no free lunch.

 

The Epishine kit also offers the ability to use an external primary battery as a back-up source thus extending its lifetime, however, users should be aware that this does come with an efficiency penalty as the energy is boosted into the supercapacitor and then bucked on the output, resulting in a loss of energy. Unless energy harvesting can form a majority of the energy usage requirements of the circuit, the additional cost, space, weight and complexity may not be worth it compared to using a larger primary battery or a rechargeable secondary battery. Additionally, while the panels may be flexible, providing the ability to contort to different form factors, this usually results in the reduction of collected energy due to sub-optimal panel orientation to the optical flux and self-shadowing effects. Additional allowances will need to be made by designers to account for degradation of the organic PV panel over time as well.

 

Thanks to cstanton and jesni for supplying this preview unit for review.