Supercapacitor Balancing Circuit

After a work colleague gave me a scrap circuit board containing some 120 Farad supercapacitors (also known as ultracapacitors), I just had to embrace my inner geek and put them to good use... and I knew just what that use would be. When I designed my Touch Controlled Digital Clock the intention was to include some form of memory backup, but in the end I decided to leave it out. Maybe, I thought, these supercaps could hold enough charge to run the clock during a short power cut lasting just a few minutes, and so perform the backup function it was missing. First though, I needed to learn a little bit more about them. Probably the most important thing to know is that supercapacitors have a very low voltage rating of about 2.7 volts, which if exceeded, will damage the capacitor and possibly become a fire risk, so strict measures should be taken to prevent this from happening. The purpose of the circuit described here is to monitor the voltage across the capacitor terminals. If it rises much above 2.6 volts, a transistor will turn on and bleed off the excess voltage through a resistor to bring it back down to a safe level. Because of their inherent 2.7V limit, supercaps have to be connected in series if they are to be used at higher voltages, but the downside to this is that the overall capacitance value will be reduced. In this case I am using two 120F capacitors in series and so the value will fall to 60F. It's just an inevitable tradeoff. It does mean though that it is safe to put two of these capacitors in series across the 5 volt USB type charger that I use to power my clock (2 x 2.7 = 5.4). You would think that the voltage across each capacitor would be the same while charging but this is not always the case as usually one will charge faster than the other which could result in the voltage across it exceeding 2.7V. Once charged though, they should level off and be roughly the same (there may be a slight difference due to tolerances). If the two capacitors used are well matched and charge up equally then this balancing circuit may not even operate, but it's there if need be to hold back the voltage of the quicker charging capacitor while the slower one catches up!


A TLV431 shunt regulator IC is used to monitor the voltage across each capacitor. They can be thought of as a programmable zener diode, with its voltage being set by two external resistors R1 and R2. In reality though, they are more like a voltage comparator with its two inputs, one inverting and one non-inverting, except the inverting input is not accessible externally but is instead connected to an internal reference voltage. The non-inverting input is accessible externally and is labelled 'Ref'. This pin will always 'try' to become the same voltage as the internal reference and the cathode (strangely known as 'K') will do whatever it needs to do (within limits) to make this so. There are two versions of the 431. The TL431 which has an internal reference voltage of 2.5 volts, and the TLV431 which has an internal referance voltage of 1.24 volts. The TLV version is used here because its lower voltage is more suited to this application. Just like any zener diode, it still needs a series resistor to limit the current, but for the two resistors that set the voltage, there is a simple equation to work out their values which is R1 R2 + 1 x 1.24 (or x 2.5 for the TL version). So if for example R1 has the same value as R2, the regulation voltage will be twice the reference voltage (2.48V). In this circuit though, R1 is slightly higher than R2, so our equation will look like 11,000 10,000 + 1 x 1.24 = 2.6V which is the perfect voltage to protect our 2.7V supercaps. The potential divider formed by R1 and R2 effectively produces a voltage of 1.24 volts on the reference pin, only when the voltage across them is 2.6 volts. This satisfies the conditions required by the TLV431 which will vary its cathode voltage accordingly to maintain equilibrium. The 1.2V drop across the base emitter junction of the darlington transistor is automatically taken into the equation as it is part of the regulation loop. 11K isn't a standard resistor value for R1 and so is made up from two resistors of 10K and 1K in series. The 1K resistor can be increased or decreased in value to fine tune the voltage at which the circuit limits

Veroboard layout

Each capacitor in the series chain requires its own balancing circuit because it's not certain which one will 'hog' the volts during charging. Another factor to consider is that an uncharged capacitor is pretty much a short circuit when power is first applied to it, so some kind of inrush current limiting should be included, which in this case is provided by a 6.8 ohm wirewound resistor. Originally I thought I'd place it in series with the positive feed, but decided instead to put it between the two capacitors. It doesn't really matter where it goes as the whole thing is a series circuit anyway, but apart from being easier to place on the board (the circuit is built on a standard size 24 x 37 hole stripboard), putting it in this position also allows each section to be tested individually to make sure it limits at the correct voltage. During testing, if you don't want to wait ages for the supercaps to charge up, they can be temporarily replaced with normal 1000uF electrolytics. Although the circuit will perform a limiting function without any capacitors connected, the voltage will measure slightly off due to what I can only assume is switching noise caused by the shunt action, which the capacitors naturally smooth out

To check that the circuit is operating correctly, connect a 5 volt supply to the +5V and TP2 terminals (negative to TP2). Measure the voltage across the +5V and TP1 terminals (the first capacitor) which should rise slowly and then level off at around 2.6V. Repeat again for the second section but this time with the 5V supply connected to the TP1 and GND terminals (+5V to TP1). Measure the voltage across the TP2 and GND terminals (the second capacitor) which again should rise slowly and limit at around 2.6V. What we're doing here is running each half of the circuit on its own with the inrush limiting resistor in series, on the negative side of the first section and then on the positive side of the second section (it doesn't matter where it goes because it's a series circuit). 6.8 ohms was chosen as the value because in the worst case scenario of the capacitors being a complete short circuit at first switch on, 6.8 ohms across a 5 volt 1A power supply will draw 735mA, and as the clock draws less than 200mA the overall drain on the power supply will be less than 1A so all is fine. Once the capacitors start to charge, the current consumption drops quickly anyway so things only get better

In practice I found that there was a slight difference between the limiting voltage measured across each capacitor which I can only put down to tolerance variations between the TLV431 regulators, as I had made sure that all the resistors were matched in value and any mismatch between the transistors would have been taken into account regardless due to them being a part of the regulation loop. Although the reference voltage of a TLV431 should be 1.24V, the datasheet states that it can be anything from 1.222V to 1.258V so this could explain the difference. The datasheet also shows the cathode current to be quite low at around 15mA max, but this is not an issue as the measured current is well within this limit. One component I haven't mentioned yet is the 1N5817 diode placed across the limiting resistor. This bypasses the resistor when there is a power failure and the supercapacitor bank has now become the power supply. Because this is a schottky type diode it has a lower forward voltage drop than a standard diode and so more voltage will reach the clock. If it wasn't there, the supercaps would feed the clock through the resistor which would cut down the available voltage considerably. During charging, the diode is reverse biased and effectively out of circuit

So finally, how do these supercapacitors perform? Well I would have been perfectly happy if they gave me around 10 minutes of backup, as power cuts tend not to last very long anyway, but they actually achieve a clock runtime in excess of 1 hour which totally exceeded my expectations. Yes the display does go rather dim after a while, but most importantly the time remains accurate when power is restored. I could not have asked for more