If you dig on any major capacitor manufacturers website - Samsung, Murata, TDK you can with some effort find a way to get graphs (and sometimes even exact values for a DC bias you specify!) showing how the capacitance changes with applied DC voltage. Even without looking up those graphs, you can probably guess that you get less capacitance at higher DC bias.. IMO TDK has both the best tool and the some of the best capacitors at high capacitance values and large physical sizes. Anyway, one way or another you can get a graph of that. Typically the graph will show a peak somepoint at or just above 0, and then fall down. How dramatically it falls however is not exactly something you can predict. The magnitude of this effect at the planned opperating voltage may be negligible, or you may discover that your cap has only 10% of it's nominal capacitance at the voltage you're using it at, even though you're not even using it at full rated voltage.
The first place is easily taken by battery capacity, as you probably knew (even the Duracell and Energizer batteries, with actual datasheets are a little misleading in their presentation Capacitance of MLCCs is either #2 or #3, with the other of the two being MOSFET maximum current (when spec'ed at Ta = 25C it's not misleading, Ta is Ambient Temperature. But on larger fets it is often specified at Tc = 25C. That means the case is being held at 25C, no matter how much heat that implies sinking, an obviously unphysical condition - but even with this magic heatsink, the device heats up so much at the junction due to the thermal resistance of the case and power dissipation at the high current that it will burn out anyway. Where ambient and case temperature maximum currents are given, the one at Tc is several times that at the same temperature spec'ed as Ta. But while MOSFETs have really one flagrantly misleading spec, you can easily calculate with pencil and paper a reasonable approximation of the maximum frequency, duty cycle, that a particular microcontroller could drive it at without a gate driver, how much heat it would dissipate, and so on.
Capacitance of class 2 MLCCs is a function of a number of factors:
This aging gets reset by heating the caps above an annealing temperature, and the speed of aging decreases logarithmically over time (so it initially drops relatively quickly, but rapidly slows down. The annealing happens during manufacture and typically during reflow too. Thus class 2 and 3 MLCCs may read high immediately after reflow. the nature of the logarithmic slowing of the capacitance loss renders this mostly not a problem for normal class 2 dielectrics, other than to be a reminder to not use the absolute minimum capacitance you can get away with
This is what the dielectric often listed is (eg X7R) is mostly about (you can look up on any resource with basic information of ceramic caps for the details on that). Certainly for class 2 MLCCs you want the second digit (the number) to be large, the last letter to be towards the end of the alphabet, and the first digit to be near the start of the alphabet.
This effect can be as severe of a capacitor at near it's rated voltage having less than a tenth of it's nominal capacitance. Some capacitors are very bad this way, others relatively good. What trends can we see in these data?
- Electric charge is physical, so capacitors are storing a physical thing, and it should come as no surprise that physically larger caps can store more energy. 0805 is barely okay up to 4.7uF from most manufacturers, and 0603 not beyond 1uF or so unless the voltage is very low. , you can do a 12V pretty well with a 1206. If you care more about the
- This includes the Z direction - usually thicker caps are better than thinner ones under applied DC bias
- The maximum rated voltage does NOT effect it! They'll usually follow the same line up until one of them is past it's maximum range that they stop reporting it.
- The dielectric does, but the correlation between "good" vs "bad" dielectrics and how well they maintain capacitance is pretty erratic.
- The product series/line does matter, to a huge degree. So does the company.
- 0402 and 0102 caps are only suitable for decoupling and low capacitance (class 1) applications, ie, for crystal loading caps and that sort of thing..
The capacitor maker gets to claim their capacitor's voltage is it's maximum voltage without breakdown + a safety margin - while knowing entirely well that it only has that capacitance at about 0 volts and at their stated maximum voltage, that 4.7 uf cap may be just a few hundred nf (if its an 0603)... There seems to be, and from a physics standpoint we would expect to see) an upper maximum capacitance in a given volume with given set of technology - If you fix the size of the capacitor at 0402, and then start dialing up the nominal capacitance, for very low values, the nominal and effective capacitance track eachother (Ceff == Cnom), so the value starts at unity, but as nominal capacitance gets larger, effective capacitance (under a dc bias, which is how most people plan to use a capacitor with a value of several to several hundreds of microfarads) starts to fall behind, and the effective capacitance approaches a limiting value, even as nominal capacitance increases.
You'll notice that many of my breakouts have a 1206 capacitor on the input and a 0805 for the output even though both are the same value. This is why - the regulator output is at most 5v on those boards, so you can get around 4.7uF from a 4.7uF capacitor. But if I want a whole 4.7uF from the input capacitor, and the input is much over 5v, 0805 ceases to get the job done. But I don't want to use the big cap on both sides of the regulator, because 1206 caps of high capacitance are not negligibly expensive like 0.1uF caps are.