12

I have found that the capacitance of some capacitors like ceramic capacitors can actually change when a DC bias voltage is applied across them. The voltage need only be a few volts to make significant difference to the capacitance. Capacitors are used in filters and also decoupling capacitors, this means that it can cause change the cut-off frequency and the performance of noise filtering on the power supply rails.

Does this not mean that we should stop using ceramic capacitors of this type, or is there a way to mitigate this problem?

JRE
  • 67,678
  • 8
  • 104
  • 179
quantum231
  • 11,218
  • 24
  • 99
  • 192

4 Answers4

12

A good designer will understand the characteristics of the parts she or he specifies and will choose parts that are optimal, all things considered.

The billions of ceramic capacitors used are testament to the broad range of applications for which they are appropriate.

Voltage coefficient, temperature coefficient, aging, and initial tolerance should always be considered. In some applications they are more important than others. For bypassing, within reason, we usually care mostly about the minimum capacitance over all conditions and over the expected life.

There are ceramic dielectrics eg. COG, NP0 with more ideal characteristics than, say, X7R, and some that are much worse.

In filter applications you may also be concerned with dielectric absorption.

In any case, it’s best to deal with suppliers who can provide detailed performance characteristics if your application is sensitive to variations, or at least get voltage and temperature coefficient curves even for bypassing.

Spehro Pefhany
  • 376,485
  • 21
  • 320
  • 842
  • Can capacitance change with other aspects of the signal/voltage applied on them e.g frequency? – quantum231 Jun 24 '23 at 21:58
  • @quantum231 Yes. You may want to look into the concept of complex permittivity. – Hearth Jun 24 '23 at 22:00
  • 4
    The first sentence of this answer is a platitude that contains no actual specific information, while also being a bit preachy and patronizing. Please consider removing it. – Chad Branzdon Jun 25 '23 at 03:33
  • If the datasheet says something like use 1 100uF, 3 22uF, 3 10uF, 5 2.2uF, 3 1.0uF capacitors for decoupling the voltage rails; what does a person buy? Do we buy the exact value in 10% tolerance or do something else? I am not sure how far we can deviate from these values. Some of these values might not be available in ceramic capacitors. – quantum231 Jun 25 '23 at 17:55
  • 1
    @user342060 I completely disagree. It is a response to the implicit attitude in the question that parts should behave as ideally as possible, when suitability for purpose is the primary engineering concern. Feel free to downvote and/or appeal to mods if you feel that addressing this is inappropriate. I do not think so. – Spehro Pefhany Jun 25 '23 at 20:04
  • 1
    @quantum231 for Murata parts they have an online and a downloadable tool called “sim surfing” that will allow you to find the typical behavior over voltage and temperature. That’s used in addition to the tolerance and other ratings on the simple datasheet. You will find that parts of similar dielectric can have wildly different voltage coefficients, usually parts from the same manufacturer that are much smaller for the same ratings perform more poorly. As to what the real minimums are- you’ll have to divine that from the circuit that needs bypassing or whatever. For example.. – Spehro Pefhany Jun 25 '23 at 20:15
  • 1
    .. a regulator might need 1uF minimum to be guaranteed to be stable(from the regulator datasheet), but you might have to use a 4.7uF or 10uF part to have it guaranteed stable. Of course if you put a 1uF rated part in there (and it’s actually 0.3uF at operating voltage) most units might well be stable under most conditions, and you’ll see hobby modules from offshore ‘designed’ like that, but it’s not a suitable approach for a professional or even a solid hobbyist design. – Spehro Pefhany Jun 25 '23 at 20:18
9

Note that the severity of the described DC bias issue depends the class of ceramic dielectric used. Class 1 material like NP0 are very stable. Class 2 materials like X7R have moderate performance variations. Class 3 materials and some class 2 like Y5V have severe DC bias and other variations. A valid question is why are class 2 and 3 made if 1 is so much better? The reason is for a given package size and voltage rating, these classes offer greater capacitance.

If target application is a filter with a specific frequency response characteristic, then NP0 or similar are generally used. All performance characteristic are far better. Note that for class 2 & 3, not only will frequency response change with applied amplitude, but also this will slightly distort the signal. So applications like high performance audio or radio frequency circuits should avoid class 2 and 3 in the signal path.

For supply rail decoupling the exact capacitance is generally not an issue. So the DC bias issue is not too problematic. If necessary, specify a larger capacitor so when biased the desired capacitance is achieved. Values like 10uF can be found in 0603 size. Yes this might only be 3uF when biased on a rail, but in NP0, 10n might be the largest option. Class 2 materials are often the best choice for rail bypassing. Given values like 100uF can be reached they are taking over from electrolytic in switched mode supplies, smaller and much lower ESR and ESL.

I have concerns about class 2 materials starting in Y. The performance variations seem so wild they seem to have questionable merit. They may have valid applications but I suspect the range will be more limited.

RYR051
  • 685
  • 1
  • 6
  • Y5V is still class II; class III materials can't be made into multilayer capacitors and are obsolete. – Hearth Jun 25 '23 at 12:33
  • With so much research going on in the world, is it true that we have not come up with cheaper materials that give great advantages like high capacitance of capacitance over voltage, temperature and signal frequency? – quantum231 Jun 25 '23 at 17:44
  • @quantum231 Just because you do a lot of research doesn't mean you'll invent the perfect . And that's even before you include cost in the criteria. It's greatly taking things for granted to think this should be the case. – DKNguyen Jun 26 '23 at 00:47
  • Good call on Y5V being class 2. I have corrected my answer – RYR051 Jun 26 '23 at 12:22
  • @RYR051,so do we need to use NPO or COG caps for decoupling – Hari Jun 26 '23 at 13:01
  • NP0 or COG are not suitable for decoupling, unless you have both lots of space on your PCB, and lots of money in your wallet. X5R/X7R are what is normally used for decoupling. – Vladimir Cravero Jun 26 '23 at 13:03
8

Just compensate for it in the design, same as how you spec an inductor to operate under mild saturation. Definitely don't just stop using MLCCs; they're tiny and cheap and very convenient!

If you need to filter a 5 V power rail with at least 100 nF of capacitance, find a capacitor that's at least 100 nF at 5 V applied voltage. If it needs to be an accurate 100 nF for whatever reason, it might be harder, but it still should be possible. Maybe a 220 nF capacitor rated for 6.3 V might have -50% capacitance at 5 V, for instance. Capacitor manufacturers often show typical curves for C vs bias (Murata's "SimSurfing" web app (no affiliation) is particularly nice for this).

If you need the capacitor to be an accurate 100 nF over a broad range of applied voltages, though, that is when you have to use something other than a type II dielectric. Either go for a C0G or U2J ceramic capacitor (type I dielectrics that don't exhibit this behaviour), or get a film or electrolytic capacitor.

Hearth
  • 27,177
  • 3
  • 51
  • 115
4

Does this not mean that we should stop using cermaic capacitors of this type?

Four parameters that make ceramic capacitors desirable:

  1. Low ESR
  2. Low ESL
  3. Small size, especially surface mount.
  4. Very inexpensive

Or is there a way to mitigate this problem?

So if, by design, the capacitor volage is constant, then there is no reason not to use a ceramic. If 10\$\mu\$F is desired , but at the operating voltage the capacitor has 50% of its value, then use a 20 \$\mu\$F value or two 10\$\mu\$F in parallel.

Edit:

The variation of capacitance with voltages is more severe with smaller volumes. Example 0402 exhibits extreme dependance while 1206 is mild. If the board real estate allows, then the larger volume capacitor may be preferred.

RussellH
  • 12,496
  • 2
  • 9
  • 34
  • Right at this moment, the primary concern is power supply decoupling. If the datasheet recommends that for each power rail I use, 1 100uF, 3 22uF, 3 10uF, 5 2.2uF, 3 1.0uF capacitors, what components should I buy and use? – quantum231 Jun 25 '23 at 17:48
  • @quantum231: To motivate your decision, you can find charts with a google search showing capacitance variation with voltage for ceramics. The manufacturers don't provide this information in datasheets. If all else fails, set up a jig and make some measurements. Then increase the capacitance of the ceramic to compensate for the reduction due to voltage. If you can fit larger sizes like 1206, then the variation is smaller. – RussellH Jun 25 '23 at 18:17