# Capacitors: voltage-dependent capacitance?

While playing around with a capacitive power supplies, I observed some very weird behavior when passing a 60Hz sinusoidal current through a 1000V disk ceramic capacitor. The voltage curve was very distorted, not a sine at all! After making sure I still understand Circuits 101, I got this:

Capacitance of ceramic capacitors may also change with applied voltage. This effect is more prevalent in class 2 ceramic capacitors. The ferroelectric material depends on the applied voltage.

This makes perfect sense: the VxQ curve of capacitors can be different from a straight line because real capacitors have real dielectrics which do not necessarity get polarized in a linear way (unlike the parallel plate case). The capacitive power supply is just a case that really exarcebates this effect. (I was using a class 2 ceramic cap and crossing 0V a lot)

Ok, so now the question: in my experience, this VxQ (or, alternatively, CxV) relationship is very rarely mentioned most of the time. Aside from capacitive power supplies, is this effect relevant? How do these VxQ curves look like for different types of capacitors?

• I realised late in life that ceramics may show some piezoelectric effect so applying a voltage will change their shape and hence their capacitance. Commented Aug 5, 2021 at 16:38
• Yes, it's a very common issue with ceramics, expecially the high dielectrics. A 10µF nominal capacitor can work as a 5µF at full working voltage. Also, that's the reason that for precision work we still use film dielectrics! Commented Aug 6, 2021 at 7:34

When it comes to voltage dependence of capacitance, class II ceramic capacitors are considered exceptionally bad, while tantalum, polymer and class I ceramics are considered very stable (source 1, source 2). I won't comment on the other types since I couldn't find authoritative sources, but I believe nothing is remotely as bad as class II ceramics.

Class II MLCCs can exhibit very significant decrease in capacitance as the applied voltage increases - up to 90% decrease in capacitance (!!!) when near maximum voltage. This effect is referred to as VCC (Voltage Coefficient of Capacitance) and is more pronouced in capacitors with higher volumetric capacitance, so smaller footprints tend to be more affected. Class I ceramics are made from a different material that is not affected by this.

I found some sources from Kemet (here and here) that elaborate on that. The 1210 vs 0805 footprint graph in the second source is particularly scary:

This is a very big gotcha, if you ask me! I haven't seen VCC mentioned in datasheets at all, and its effects can be a lot larger than the effects of temperature and aging, which are mentioned (perhaps because, according to this source, VCC is not defined by the EIA). The nominal capacitance of class II caps is measured near 0V, so you could miss a 10x derating factor!

Understanding this effect explains several rules of thumb involving ceramic capacitor selection, such as:

• Use tantalum capacitors instead of ceramic ones when the DC bias level might vary significantly
• Never use ceramic caps near the maximum rated voltage
• Only use C0G (class I) ceramics in oscillators
• Class II ceramics should only be used for buffering and decoupling
• But if I use Class II ceramic for decoupling, I will still get more than (possibly) 50% variation in the capacitance of the ceramic capacitor. So why should I even use the ceramic capacitor to begin with? Commented Jun 26, 2023 at 21:02
• @quantum231 Several reasons. First, surface-mount ceramic capacitors are the "most ideal" capacitors in existence in terms of low parasitic inductance (under 5 nH) and low parasitic resistance (0.05 Ω) at high frequencies. There is almost no competition in this aspect. In comparison, other types of capacitors are at least an order of magnitude worse, electrolytic capacitors are almost useless in particular. This reason along is the biggest reason to use them for decoupling high-speed digital circuits operating at 10 MHz to 1 GHz. Commented Aug 18, 2023 at 13:37
• @quantum231 Next, although some low-grade ceramic capacitors have bad tolerance (looking at you, Y5V). But X5R and X7R capacitors actually have good tolerance under no bias, around 5%. Most of the time, a digital circuit's DC voltage rails are running at a fixed voltage, so the capacitance loss can be compensated during design. Commented Aug 18, 2023 at 13:42
• @quantum231 Finally, the solid-state construction without an electrolyte give them an extremely long service life (but beware of mechanical cracking due to mishandling or board flexing...) - another reason that some DC-DC converters prefer an all-ceramic design. Commented Aug 18, 2023 at 13:45

True, it is not mentioned very often, but on the other hand, it is mentioned when it is important and not mentioned when it matters very little.

It is relevant in many applications, especially in things that require a certain capacitance to work properly, such as voltage regulators, both linear and switch mode regulators.

Sometimes voltage regulators mention that at least some amount of capacitance is necessary, and datasheet or application note says that it is good enough to put a capacitor with certain value, so the voltage de-rating is already taken care of, assuming typical component properties.

Sometimes this is taken care of mentioning to use say 50V capacitor in a 5V circuit, or 10V capacitor with certain physical size in a 5V circuit, as the effect is also dependent on package size.

In some filters where the input signal might have or cause a DC bias would have varying cut-off characteristics due to the capacitance depending on voltage.

This can also have interesting effects, as varying a voltage over a capacitor can cause it to vibrate along the applied signal and emit sound at the frequency of the signal.

Same goes the other way, the capacitors can act as microphones, picking up vibrations and converting that to voltage signal, which may disturb sensitive measurements.

It's just something you need to aware of, and you typically don't have to deal with these phenomenon on a daily basis. But it can sometimes explain why a circuit does not work or acts weird.

• Thank you for the answer. However, it still leaves me paranoid when designing circuits because I have to know the unwritten rules ("yeah a X7R is technically in spec but this has to be a NP0"). Are there any general guidelines? (maybe "only use class 2 ceramics for snubbing or power supply smoothing", or "when using class 2 ceramics for filtering signals, overdo the voltage rating by 10x") Commented Aug 5, 2021 at 17:24
• Don't worry. Depending on what are you designing, most likely first and only thing that requires NP0 types are caps with sub-nanofarad values like the capcitors for MCU crystals, and most often capacitors of that values are more commonly NP0 already. Maybe if you are doing analog audio circuits with op-amps and filters, the those could be better with NP0. Snubbing a SMPS switch node could be NP0 as well, but X7R should work too. Commented Aug 5, 2021 at 17:32
• @FrancoVS If all else fails, film capacitors are relatively immune to this effect, too. They're a fair bit physically larger, though. Commented Aug 6, 2021 at 2:09