SuperCapacitors Vs Batteries Again

Supercapacitors are definitely not the same as batteries, we all know that. They tend to have a very low operating voltage, and due to their operating principle of storing charge on parallel plates, their discharge curve is quite unfriendly for modern microcontroller devices. Energy storage efficiency per unit volume is also low compared with modern lithium polymer (LiPo) batteries so all in all they don’t look all that useful for many of our projects. However, as [Andreas Spiess’] latest video demonstrates, they do have some redeeming features that might make them useful for certain embedded applications.

The low operating voltage initially looks like an issue for devices operating at a typical 3.3V, and it’s tempting to simply wire a few in series and roll with it. But as [Andreas] explains in his typically clear manner, it would be necessary to have a complex power stage, operating in buck mode with capacitor voltage above the required level, and in boost mode when it heads below. Too complex – it’s much easier to simply stick with a low voltage bank of paralleled supercaps, and just operate always in boost mode. Even doing this, you’re not realistically going to get more than a handful of hours operating voltage with an always active device.

So why bother at all with supercaps, surely using a LiPo is so much easier and better? In many cases the answer is definitely a yes. But LiPo cells must not be charged in freezing temperatures (apart from certain special low temp products), else the cell can rapidly be destroyed due to lithium metal deposition at the anode. Also you need to be careful charging them, especially when they’re heavily discharged, as they are easily damaged without the proper treatment. LiPo cells operate based on chemical principles – lithium ions literally have to move around inside the structure, and eventually the battery will wear out.

Supercapacitors have the advantage of very long life (but sometimes, they do leak) much more aggressive charging and discharging behaviours and will operate down to very low temperatures. This makes them very useful when a large amount of power is available sporadically (for super fast charge cycles) or in places where temperatures stay persistently very low, such as up a mountain were solar will work, albeit slowly, but LiPo batteries will definitely not be suitable.

Other battery chemistries are available, such as Lithium Iron Phosphate which can tolerate the cold. Also you can always just insulate the battery with an integrated heater and preheat the battery to a safe charging temperature as well. So, just like everything with electronics, it’s important to choose the correct parts for your application, and it all starts with the power source. Supercapacitors might just hit an appropriate price/performance point for that special application you had in mind.

Supercapacitors aren’t really suitable for many applications, like powering an eBike or running your laptop, but hey, they did it anyway.


41 thoughts on “SuperCapacitors Vs Batteries Again

  1. An interesting topic and great potential to explore variations in a suitable lab, thanks for posting :-)
    It occurs to consider a hybrid conglomeration in a self charging tightly integrated solid state form might be on the horizon, for example we know some elements are radioactive and at low ‘safe’ levels like thorium & bismuth. Then its ‘just’ a matter of investigating efficient methods of charge separation. I’d suggest looking at tunneling behaviours perhaps combining aspects of Sandia Labs neutristor which produces a neutron flux proportional to electric current exploiting fusion in a cigarette packet sized unit I guess with with electrolyte topology where (contained) neutrons can help kick charges around potentially pulled into a potential well integrating rate earth magnet films to trap – magnetic self charging been speculated about for decades.
    Fwiw. I’ve a few supercapitor types, a couple seem to raise their voltage albeit slowly – noise, RF, magnetics, radiation who knows…

    I recall when supercaps first came out a guy put 3 – oriented in 3 axes with high resolution data logging and in a Faraday cage watching microvolts changes on the unloaded units oddly correlating with regional earthquake activity. One wonders what else can pull or push charges to & from the electrode surfaces – variant of Maxwell’s demon maybe ;-)

    1. I have looked into this. The amount of electrical power (enough to be useful for anything, including trickle charging a battery) requires a tremendous amount of material. Far more than can be possessed without NRC license in the US.

      As an example, Po210, a highly active alpha emitter (with a half life of 138 days, that is probably too short to be useful in this application), has an alpha energy of 5.3MeV and a specific activity of 166 TBq/g (4486Ci/g).

      1 Bq = 1 event of radiation emission or disintegration per second, and 6,200 billion MeV = 1 joule. 1 joule per second = 1 watt.

      So, even if you had 1 gram of Po-210, and could capture every eV of decay energy and turn it directly into electricity (the decay over 1 second of time), you could have 37.35KW of electricity. But this is not possible.

      milligram amounts of this particular isotope have been created by neutron activation of Bismuth-209 in nuclear power reactors.

      There are many problems with this. Too many to get into here. Destruction of semiconductor junctions and the tremendous “heat of decay” which in this example , a single gram of Po210 can generate 140 watts of power (as heat). Enough to destroy its container or over-heat nearby electronics if not cooled.

      There are times when radioisotope voltage and current sources are useful, for example as fA or pA level current references, Or high voltage sources when extremely small currents are required.

      When you need more power, radio-thermal sources are used. The decay energy turns into heat, and that heat is used to create electricity by some means.

  2. An interesting application has been to use supercaps for regenerative braking in vehicles. They’re well suited for it. High charge and discharge rates just aren’t a problem. I don’t know how they compare to the flywheel setup I’ve seen on busses…

    1. Batteries are better. Why dedicate a large volume of the vehicle to a relatively low density energy storage medium? A battery of the same size has the same series resistance and current capacity, but also has enormously more storage capacity.

      To decelerate a car from highway speed requires you to absorb about 0.7 megajoule. A supercap to store that will take up 50 liters of space: the same size as a ICE vehicle’s gasoline tank. Why bother? Just use a tiny fraction of the capacity of the battery you’re already lugging around.

      1. Regenerative braking is highly overrated and misunderstood. The motors work poorly as generators, especially at lower speeds, which means you only get about 5-15% of the energy back overall. At low speeds, what you’d want is something low voltage and low impedance to dump the energy into, because pumping it back up to 200-400 Volts into the main battery bank consumes most of the recovered energy.

        1. Well you are in the “misunderstood” group as well. 50 to 70 percent efficiency is pretty attainable in current EVs. That is why thr break pads and rotors on them practically last foreever. So tired of seeing your BS.

          1. Braking hard from high speeds, high efficiency is attainable, but this happens infrequently compared to braking gently at low speeds, which puts the overall efficiency down to practically meaningless.

            Think about it: you don’t touch the brakes on the highway until you exit, whereas driving stop and go in the city you’re braking every few hundred feet.

        2. There’s a time and place for regen braking. During acceleration, about 80%-90% of the energy from the battery gets to the road. During braking, energy recapture (regen) is about as efficient, near 80%. The end result is that after an accel/decel cycle, about a third of the energy is lost ( 1.00 – (0.80 * 0.8 0) = 0.36) . Still, recapturing 64% is better than recapturing 0%. Not that you actually recapture 64%; there’s lots of friction losses (wind, tires, bearing, etc), and electrical losses (I^2R, switching losses, etc).
          Even so, in an electric car, this results in a significant boost of range.

          That being said, the same battery that provides accel power is equally capable of absorbing similar regen power. Inserting supercapacitors into the mix doesn’t reduce any of the other losses.

      2. I’ve only seen the supercap approach on city transit busses. Getting a couple hundred extra liters of space on there is NBD. I’ve also seen (though “heard” might be more appropriate) a flywheel setup on another bus. I have no idea how the two compare.

    2. Now, as if it won’t become obvious, I know nothing. But since you mentioned supercaps in regenerative braking in evehicles could they also be used in fast charging? What I mean is dump the power quickly into supercaps and then transfer it more slowly to the batteries while you’re driving, not waiting to recharge. Now I know there are probably a million reasons why that’s a stupid idea but could just give me then highlights?

      1. Reason 1: Batteries can already accept power as fast as any (reasonable) charger and power cable can deliver it. There is no reason to add supercaps.
        Reason 2: Any reasonable size capacitor can store only a few seconds worth of high-power charging.
        Reason 3: A battery of the same volume as a supercap can accept power just as fast.
        Reason 4: charging a supercap, then discharging it only to charge a battery, is two extra steps of power conversion, with the attendant inefficiency. Just charge the darned battery, fer pete’s sake.

          1. Supercaps enable high efficiency regenerative braking in all electro-drive technologies.

            Supercaps enable power averaging and high efficiency regenerative braking for ICE Hybrid Electric processes. Power averaging output significantly reduces prime mover mass and volume fractions.

      1. Except, that MTBF value is for it leaving its specs. Not failure.
        The biggest spec that is looked at is the ESR of the capacitor, capacitance tends to remain more or less the same throughout its whole life, just like regular electrolytic capacitors.

        For high power applications, then a 2-5x higher ESR might not be the best thing in the world. But for low power applications it doesn’t really matter in the slightest.

        The capacitor will still work for a lot longer than its MTBF, it is just a bit less good at its job. (For high power applications, it can be junk however since low ESR is typically important in this field.)

        A crude example could be a 20F capacitor with an initial ESR of 0.1 ohm. If it increases with a crazy 5x in its ESR after 2000 hours at 65 C (MTBF), then it is still only 0.5 ohms of resistance. If we have a low power micro with some RF capabilities, then it might pull less than a couple of mA normally, and spike up to a few hundred mA when it turns on the radio, then it really wouldn’t have too much issue operating, especially if one places in some smaller capacitors for handing that really short high current burst of the radio transmitting.

        Realistically though, one won’t be running continuously up at the rated temperature, and with about every 10 C one moves down bellow it one doubles the ESR. And most super capacitors I have seen regards “failure” as an increase in ESR by 1.5-4 times.

        So to a degree, don’t be scared by the 1-3 K hours at 55-80 C that most super capacitors are rated at. For low power applications it really doesn’t matter.

        Also, “low power” is practically any application that stays bellow 10% of its continuous current rating. For larger super capacitors, that is still many amps.

  3. I do the opposite: buck only, not boost. I use a series stack of supercaps as a 30 second backup for my octopi rig. 10 of them in series are fed from the 24V main printer supply, and in turn feed the Pi through a buck converter.

    When I hit the main power switch, the Pi sees the loss of main power via an i/o pin, and starts a shutdown. The shutdown is complete before the capacitor charge is depleted.

    Steering diodes and a charging resistor keep the charge current sane on powerup. There is a 15-second ‘iffy’ period after startup where the Pi script might see a power fail and begin the shutdown, but the capacitors don’t have enough charge to last long enough.

    1. Only downside with that approach is cell balancing.

      I would put some low forward voltage diodes over the capacitors so that if one gets reverse charged that it won’t make it past maybe 0.1-0.2 volts. This will help ensure that the electrolyte isn’t going to break down.

      One can opt for more intelligent safety features like transistors putting a bit extra load on cells that have a higher voltage during discharge. Or a switched capacitor implementation that moves charge from one cell to another.

      The big issue here is mainly that cell capacity won’t be the same on all capacitors. +/- 20% isn’t particularly tight tolerances after all. (And even +/- 5% is still fairly bad, but excessively tight tolerance for a capacitor of this type.)

      One might think that “surely if one just charges them in a series string it won’t matter.” except, leakage… It too varies from cell to cell. So long term the voltage will drift for each cell.

      On initial charge up it will have a voltage over each cell that is proportional to the variations in capacitance, over time, it will move to the variations in leakage. This is why a lot of capacitor banks have a crude resistor divider as a “balancer”, not that a resistor divider helps in the slightest on discharge…

      Going with an all parallel solution ensures that variations in cell capacitance and leakage doesn’t matter. Making it far harder to accidentally reverse charge one or more of the cells.

      Though, series strings have the huge advantage of lower currents and larger voltages, making it easier to deliver a lot of power with less conductive losses, especially in one’s DC-DC converter.

        1. A string of series resistors doesn’t help on discharge. And it is during discharge that the risk of reverse charging capacitors is most likely.

          This is the main reason I suggest having some protection diodes to prevent a reverse charge from happening.

          To actually balance during discharge, one needs a more active balancing system. At the very least a simple transistor that puts a bit of load on the cells with the highest voltage left. Or just sensing and terminating the discharge if a cell is about to be reverse charged.

          A more energy effective method is to switch a capacitor back and forth between two capacitors during discharge. A relatively trivial feat.

          Here is a simple simulation of such an active balancer in comparison to a passive “balancer”:

          This can be expanded to an arbitrary amount of series capacitors, but it is easiest to implement in practice if one keeps oneself bellow about 10-15 volts. (Since the Vgs of most N-fets aren’t typically higher than 20 volts.) But one can take a more involved approach and get much higher voltages if desired, just need to drive the transistors in an isolated fashion.

          Experiment a bit yourself and see. The resistive balancer won’t do much during a discharge. The switched capacitor one however will, though its effectiveness is dependent on a lot of factors, like the switching speed and series resistance.

        2. A passive set of resistors won’t balance the capacitors during discharge. And during discharge is where you risk reverse biasing your capacitors.

          Having reverse charge protection diodes is a wise move at the very least.

          But you can also attempt using a more active balancing circuit that switches a set of capacitor back and forth, moving energy from the capacitors with a large voltage to the ones with a smaller voltage. This can rather easily be built with a chain of N-fets and an inverter and some clock source like a 555 timer.

          Here is an example circuit you can import into the Falstad circuit simulator:
          $ 1 0.000005 10.20027730826997 50 5 50 5e-11
          c 832 368 832 240 0 0.000017 -2.2380464892904013 0.001
          c 832 496 832 368 0 0.000014999999999999999 -2.4795639395284907 0.001
          r 912 496 912 368 0 100000
          f 672 464 736 464 32 1.5 20
          f 688 400 736 400 32 1.5 20
          f 672 336 736 336 32 1.5 20
          f 688 272 736 272 32 1.5 20
          w 912 368 832 368 0
          w 832 496 912 496 0
          r 1136 112 1184 112 0 100
          R 1184 112 1232 112 0 0 40 7.5 0 0 0.5
          w 832 240 736 240 0
          w 832 368 736 368 0
          w 832 496 736 496 0
          w 736 256 736 240 0
          w 736 320 736 304 0
          w 736 304 736 288 0
          w 736 368 736 384 0
          w 736 368 736 352 0
          w 736 416 736 432 0
          w 736 448 736 432 0
          w 736 496 736 480 0
          g 832 496 832 512 0 0
          w 672 464 672 336 0
          w 688 400 688 272 0
          I 624 336 672 336 0 0.5 6
          w 624 272 624 336 0
          w 624 272 688 272 0
          R 544 336 496 336 1 2 100 2.5 2.5 0 0.5
          w 912 368 976 368 0
          w 912 496 976 496 0
          w 832 240 912 240 0
          p 976 240 976 368 1 0 0
          p 976 368 976 496 1 0 0
          w 736 432 784 432 0
          w 736 304 784 304 0
          w 784 432 784 384 0
          c 784 384 784 304 0 0.000001 -2.4795639479750284 0.001
          s 544 336 624 336 0 1 false
          r 912 368 912 240 0 100000
          w 912 240 976 240 0
          w 912 112 976 112 0
          r 912 240 912 112 0 100000
          c 784 256 784 176 0 0.000001 -2.218704905706834 0.001
          w 736 176 784 176 0
          p 976 112 976 240 1 0 0
          w 832 112 912 112 0
          w 736 176 736 160 0
          w 736 192 736 176 0
          w 736 128 736 112 0
          w 832 112 736 112 0
          f 688 144 736 144 32 1.5 20
          f 672 208 736 208 32 1.5 20
          c 832 240 832 112 0 0.000014 -2.7798442899825737 0.001
          w 736 240 736 224 0
          w 784 304 784 256 0
          w 672 336 672 208 0
          w 688 272 688 144 0
          s 1072 112 1136 112 0 0 false
          s 1072 112 1072 176 0 1 false
          g 1072 224 1072 256 0 0
          r 1072 176 1072 224 0 1000
          w 976 112 1072 112 0

          If only one there were a handy way to post a compressed link to this specific site on hack a day…. Would be wonderful.

          1. Some minor corrections on the simulation…
            I broke it when setting it up.

            The inverter needs to output about 9-10 volts. Not 6…
            And the clock source needs to output 5 volts with a 5 volt offset. (Higher than 100 Hz is also preferred to be fair.)

            I built it to balance 2 capacitors in series first and simply expanded it to 3 without second thought about changing the parameters… So that is why the active balancer doesn’t work at first.

            Generating a voltage 1.5-2 volts above the voltage in the capacitor bank itself is a trivial task in practice, just a small boost converter is sufficient, the balancer should realistically not draw a ton of power.

          2. Interesting switched balance approach. I’d like to see a real implementation that would work at 24V with real FETs.

            In practice it’s not necessary, at least when the caps were new: over the range this series stack operates, a single cap would need to be more than 25% different from its brethren to approach zero voltage, and they never did.

            But I did put diodes in there anyway — just 1N4148s (because I had literally thousands), but they wouldn’t see much current. I’m pretty sure they’ve never been exercised. It’s been a few years since I opened it up though. It would be interesting to see how the caps have aged.

          3. Getting it working at higher voltages isn’t too problematic.

            Just break it into groups, and feed each group with its own isolated DC-DC converter. Might though not be super cost effective at this point though.

            If I had a chip factory, I would make a sot23-5 device for the application with the following pinout:
            1. Clock in (A capacitively coupled signal from our reference clock. Preferably 40-150 kHz or lower for idle balancing.)
            2. Flying capacitors (The pin we tie a capacitor to the next stage and tie in the capacitor from the prior stage)
            3. Next stage in (The high voltage side of the next stage. Or a charge pump driven voltage booster for our last stage. (The charge pump booster is 2 caps and 2 diodes and an oscillator, trivial stuff))
            4. Power out (The low voltage side of our current stage)
            5. Power In (The high voltage side of our current stage)

            Even if a device like this might only be able to handle 100-200 mA continuously, it would provide quite a bit of balancing with fairly small losses. And if one needs more current, then put two or more in parallel. (A higher power version might though be needed at some point, peppering on 5+ sot23-5 packages per stage would be silly….)

            But a chip like this would likely be somewhat cheap, maybe a few tens of cents each in hundred off quantity, and would likely aid in the development of capacitor based UPS solutions for where one don’t need many minutes of runtime, but rather enough time to save and gracefully shut down. (or just ride through a brownout.)

          4. So even if you’re willing to accept the cost, complexity, failure modes and risk of implementing such a complex balancing scheme, does it even provide any benefit at all?

            The very nature of shuttling charge by switching capacitors from stage to stage is very inefficient: you lose at least 50% of that energy while charging the capacitor — it gets dissipated in the switch resistance or capacitor ESR. Only by using an inductor can you avoid this loss.

            Whereas if you simply short a capacitor once its voltage goes to zero during discharge (i.e., with a diode) you can extract all the energy from all the caps — no need to shuttle charge around and and lose half your energy. If you quibble about the power loss to the diode drop, use a switched FET “perfect diode” instead.

            In practice, you extract most of the stored energy from ta capacitor by letting the voltage drop just 30-50%. With modest care in making sure your capacitors are equal you will never have one see reverse voltage unless it has already failed.

          5. To be fair, there is always some degree of parasitic inductances, but nothing prevents us from adding a series inductor with our switching capacitors, so the efficiency can go beyond 50%.

            But shorting out the capacitors when it reaches a near zero voltage also technically works. Only problem with this is that super capacitors have quite extensive dielectric absorption. A trait also seen in regular electrolytic capacitors but to a lesser degree.

            I have poked about a bit at some 500 F capacitors, if I charge them up to 2.5 volts and leave them there for an hour or two, before discharging it back down to 0 volts in a few minutes and keeping it at 0 volts for a few seconds before then letting it sit open circuit, then it will recover back to a good 0.6-1 volts ready for quite a bit more discharging, it weren’t close to empty.

            This I fear would make the concept of just shorting out “empty” cells lead to a much greater overall energy loss compared to moving a bit of energy to keep all capacitors at a more even voltage throughout. After all, the balancing circuit’s inefficiencies only affects the unbalanced portion of our energy within the bank. And in a series parallel bank, the unbalanced part can be fairly small.

            A slower overall discharge would give the capacitors more time to release this absorbed energy. To a degree one can argue that if dielectric absorption is a serious issue that one maybe have an undersized capacitor bank for the application regardless.

            There is however more reasons for why we need balancing. It can mainly be used to ensure that the voltage over our capacitors stay more or less equalized over multiple charge cycles. It is also considerably more energy efficient than passive equalization through resistors. Not to mention that active balancing has the advantage that we can run it only when balancing is actually needed, instead of all the time.

            But I agree that an active balancing circuit is a bit complex, and why I wouldn’t mind seeing the single chip solution postulated before.

            The postulated chip above would also be somewhat simple to use in practice, and together with a current sense resistor, comparator, a tiny bit of logic and a pair of oscillators we can have it switch between a low frequency idle mode when current consumption is low, and a high frequency mode when current consumption is high. Additionally saving in on power while also allowing for worse overall tolerances between our series stages in our bank. (or one can just use a micro and be even more fancy in picking a frequency.)

  4. > such as Lithium Iron Phosphate which can tolerate the cold

    No. They discharge just fine when cold, but charging must be done under 0.02C to avoid the same metal plating issue.

    New car batteries built on LFP have built-in heaters which switch on automatically and switch off after the battery is up to charging temperature, which means you have to run the motor a minimum time until the battery is warm – or else you get no charge.

    1. 100 pct agree, the charge rate is a function of anode composition, and that is still mostly graphite in cells these days. If you want charge capability at low temp and better Wh/kg then go with LTO… not as high energy density as graphite based, but fast charge capable and low temp applicable. 80Wh/g ish.

      1. >80Wh/g ish.

        If you go down that low, you might as well use NiMH – it can take even more abuse like complete deep discharge to zero volts and overcharging without catching on fire, and charging in freezing conditions. It’s probably cheaper as well.

  5. A very helpful application for supercaps is in model railways. Dirt on the tracks or junctions etc always lead to small power interruptions or circuit shortages. Motor power is immediately off, but the inertia of the train pushes the engine forward over the problematic part. That is how it was before the supercaps. But the microcontrollers in the trains did not like this at all.
    With the supercaps, microcontrollers and engine are continuously powered for smooth operation even at very low speeds.
    The circuits need high power for a very short time, where supercaps excel.

  6. I’ve sent a few mails to companies producing massive dump trucks telling them that those vehicles could become lighter and much more fuel efficient if they used supercapitors to recuperate the energy each time a load was dumped in the truck. Those things already use electric drive but have massive diesel engines to produce the electricity. Supercaps could grab that energy and charge a battery which, in turn would power electric motors. Would this work?

    There is a quarry near where I live. Trucks drive up a mountain where they are loaded with rock, then drive down again before delivering it. Imagine if they were to make a deep lake at the bottom and simply drop the rocks into it. The overflowing water could power a turbine and supercaps could grab the wave energy.

  7. Nitpick: it’s not “Energy storage efficiency”. It is “energy density”, either per volume unit or per weight unit.

    Efficiency is more like the ratio of what you get our vs. what you put in.

    Sorry for being that one :)

  8. The only time i’ve found them useful was to buffer the output of bench power supplies. Sometimes a load can have short term needs beyond the capability of even a meety PSU, and ride through for 1/2 a second or so can be handy.

    Best put together with a little thought about discharging spare energy after tests are complete.

  9. I would like to add Lithium Ion Capacitors (LICs) to the competition. I believe LICs combine many advantages of Li-ion batteries (LIB) and supercapacitors (EDLC) making it a perfect choice for batteryless IoT applications. Recently the price of LICs dropped significantly because new factories are built. Now you can get a 30F 3.8V LIC for 4 euros. LICs also charge in seconds, but have much lower self discharge and they have 6 times the capacity compared to supercaps.

    I am currently selling this solar harvesting into LIC board on Tindie
    It is based om E-peas AEM10941 and it uses MPPT to efficiently harvest the energy.

    But I am working on a new design, that adds a 3.3V/400mA buck-boost converter so it can make fully use of the LIC capacity.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.