Temp and voltage variation of ceramic caps, or why your 4.7-uF part becomes 0.33 uF

11/27/2012 3:41 PM EST

I once used a 4.7uf 0402 (had to be small) X7R capacitor in a timing circuit (relatively non stringent) and found that only over time the capacitance dropped (50% or so). It was 6.3v rated and my circuit circuit was run a on 1.55v silver oxide battery. The pulse circuit duty cycle, essentially, only impressed an average of about 0.1 volt DC bias. Funny thing is that the capacitor retained the lower capacitance when power was removed and it could be refreshed to the original capacitance by applying soldering iron heat. Its been a while since I stumbled over this and I do believe there is literature (maybe panasonic) regarding this particular affect. BTW: 50 % did not meet my "non-stringent" requirement so I switched to the AVX tancerum type. Good article and warning to all.

11/27/2012 3:45 PM EST

I like Kemet and AVX for that reason. There are other good vendors as you mentioned, but stick with the devil you know. At some point the capacitance tracks the voltage, causing nonlinearities from the delta-C. Any idea about the time frame of the voltage effects? At what frequency would the voltage coefficient kick in? Hz? Sub-Hz?

11/27/2012 4:33 PM EST

Thank you Mark for this article. We tend to forget these things, and as you say, as the demand for "smaller and smaller" goes up these things become more critical than just a cursory glance at the data sheet. A "shout out" too to the manufacturers like Murata, TDK, AVX (who provide Spice models and 3D STEP files for some parts!) and Kemet (who always come through in a pinch with samples) that make these tools available; anytime a manufacturer provides these kinds of tools they go straight to the top of my list for "check here first".

11/27/2012 7:36 PM EST

This gets everyone sooner or later. I once worked on a PLL circuit that worked fine until a capacitor change to a "Better" part made the loop go unstable. It turns out that with the original capacitor in the integrator was loosing capacitance at higher DC bias and this was almost perfectly compensating the increasing loop gain of the PLL loop! The "Better" Capacitor did not loose capacitance with DC bias and the loop broke out into oscillation. This will get everyone sooner or later also - the seemingly small change that can't go wrong ends up making the circuit fail. Humbling.... At least it didn't fail in our customers hands, just on our test benches.

11/27/2012 10:40 PM EST

On the positive side, a large capacitance variation (4uF). This could be used to tune a low frequency filter or oscillator merely by altering the DC bias conditions. Will keep this in mind if an application ever comes up. Great article.

11/28/2012 2:43 AM EST

Tim Orr

It is the case that surface mount ceramic capacitors have a lot of hidden problems. I have never measured a capacitance that was above the claimed value on the reel, for X5R and X7R. Also, if you use them to decouple power supplies you may only get 20% of their value.

Q: Why use Y5V capacitors? A: Y5V capacitors are as cheap as chips. However, I think that a bag of chips may make a better capacitor.

Products made in the 'East' may have X7R in the BOM, and something cheaper on the PCB.

C0G and NP0 capacitors are the nearest to a stable SM capacitor that you can purchase, if you are very rich and want high value parts.

When is a capacitor a resistor? I have had two units returned because they stopped working. It turns out that an AC coupling capacitor (100uF X5R) developed a DC leakage current that made it look like a 4k resistor! As far as I find out, mechanical shock can cause micro cracks in the ceramic material, which conduct. I removed the capacitor and looked at it under a microscope. It looked perfect, but the 1/2 kg metal box which housed the product was twisted and its front panel BNC was bent!

If you want to make a microphone or a loudspeaker, put ceramic capacitors on your PCB.

And so it goes on! Film capacitors still work well.

11/29/2012 2:23 AM EST

This is a great article with a lot of insight. Hardly any engineers know about the voltage coefficient of ceramic caps and indeed most cap measurement LCR meters or cap 'bridges' can only apply a DC bias voltage of a few volts to the cap wen measuring, so the effect is rather hidden. To this end we manufacture and supply an external bias unit which sits between an LCR meter and the capacitor under test allowing up to 5kV to be applied to investigate and confirm application suitability. It can be rather revealing!
http://www.appliedrelaytesting.co.uk/products/asy0360.html

11/30/2012 1:45 PM EST

Back I the day (as it were: about 1996) I was working on a 70Mhz amplifier with an outrageous gain (over 60dB.) I was worried enough about it that I was using some serious decoupling on the powersupply lines: 10uF distributed approximately 1" intervals along the main power and ground bus, 1u and .1u at the PS pins to the very-close-ground, using microstrip construction for signal lines to the amp and from there to the high-speed comparator which created the 70Mhz clock for the timing chain, proper transmission line impedance matching, the works.

The circuit oscillated at 900Mhz, almost exactly. As in your case, I tested everything repeatedly, but it wasn't until I checked the .1uF AVX ceramic disk caps that I discovered the source: these little buggers would oscillate on 5v bias whether they were connected to anything else or not. After testing 10 at random out of the bag of 100, I relabeled the bag "900Mhz Oscillators" with a schematic (+5vdc to cap terminal || other terminal to ground with any load from 5v to ground) and put them aside. I got some different .1uF ceramic caps, which didn't self-oscillate and carried on.

I never did have a need for 900Mhz one-element oscillators, nor did I have the luxury of finding out why they did that.

11/30/2012 2:18 PM EST

Good article.
I also ran into this issue early in my career. I was asked to investigate why a batch of battery chargers weren't working. I traced it down to an RC oscillator with the wrong frequency (too high). Like you, I checked the parts (correct), checked their values (also correct). When I scoped the waveform, it wasn't what I was expecting (if I remember, the standard RC waveform was inverted). Turns out a bad (wrong) batch of caps was used, and their capacitance varied widely over voltage. So as the voltage ramped up, the capacitance dropped, and the time constant dropped making the frequency too high.
Long story short, don't just rely on good manufacturers, but make sure your distributers are providing the correct parts.

11/30/2012 3:01 PM EST

If you absolutely positively must have a ceramic capacitor that does not have temperature and voltage degradation issues, call Novacap and ask them about type H dielectric capacitors.

12/14/2012 3:34 PM EST

Why your 220uF cap is really only 7uF EVEN AT ROOM TEMP.
Download the Kemet file KEM_T2009_T495.pdf and look at the plot on page 4 how the capacitance declines with frequency -- 220 uF up to about 20kHz and by 5 MHz it is down to 7 uF. No wonder so many switchers get made that make so much noise even after swamping them with ton's of Farads. Picking a cost effective and minimum size cap set is a pain. While this Kemet data sheet is pretty good, (it covers C vs Freq, C vs Temp, Voltage vs Temp, Dissipation factor, DC Leakage, ESR) it still misses specifications for microphonics (both piezoelectric and capacitance modulation seriously affect low noise circuits like a ref for an ADC or a resonant tank circuit for RF.

12/18/2012 9:52 PM EST

Any idea why I can't read page two of this article?

12/20/2012 8:20 AM EST

Dear Mark,
i fully agree to all your comments on capacitance value decrease vs (temp°/DCbias).
You also forgot to mention that the Type II MLCC's also show a capacitance drop vs lifetime ( and extra -5% to take into account, and just after 1000hours operating).
As of today, Type I MLCC had the utmost performances, but very small capacitance density, so if you wanted stable high caps, that would use a lot of board space...

Now check this one out :
After 15 years working in capacitor world, i've joined a SpinOff of Philips semiconductor, called IPDiA.
We produce High density Silicon capacitors.
In few words, we have same performances as NPO's ( capa stability, failure rates, no piezzo effect), but we have same density as X7R.
We can indeed do for example 100nF in a 0402 ( iso a 1206 in NPo's).

And that is not all...
We specify our Sicaps up to 250°C, and we can offer thickness below 100µm thin.
Last but not least, our leakage current is rated in pAmps !
Imagine : NPO performances, but with 1µF 1206 250°C 100µm thin. Not bad hé ?

As the product director at IPDiA, i am glad to inform the community that there is a solution to Capacitance stability AND reliability issues.

We have a limited range available at Mouser or Digikey if you want to check them out.
Our go on our website : http://www.ipdia.com/

Or, just drop me an email if you have any questions, i would be glad to help :
laurent.lengignon@ipdia.com

Best regards

1/31/2013 5:35 PM EST

Hey Mark,
You're right, ceramic caps are not nearly constant capacitance w.r.t. operating environments.

Another damning feature of ceramic caps is the problem of LVIR failures. That stands for Low Voltage Insulation Resistance. Cracks or natural fissures in the ceramic material can foster dendritic growths of the plate material to cause a plate to plate connnection of filamentary dimensions. This wrecks the dc blocking ability of the capacitor in high-impedance circuits. In low voltage applications this causes circuit failure, whereas at high voltages, the filament is zapped off when it comes close to reaching through. This arcing only causes a noise disturbance in the circuit that may not occur again for hours or even weeks.

I believe the capacitance you are reporting on might be better termed "small-signal capacitance", i.e., dQ/dV. Some capacitor applications, such as relaxation oscillators, exercize the capacitor through a wide voltage swing. In these applications the small-signal capacitance does not apply directly. The voltage effects on relaxation oscillator frequency are somewhat less severe because the amount of change in stored charge is the integral of the small-signal capacitance over the range of the voltage swing. Obviously, the circuit analysis calculations are becoming more complicated.

When you get this deep into correlating design calculations with laboratory results, you will find that many thin-film resistors also have a significant voltage coefficient of resistance. If the thin film resistor data sheet does not mention voltage coefficient, do not assume it is zero! You will find that the terminal-to-terminal resistance goes down with increased voltage. And again there is a temperature effect on the resistance due to the ambient plus the part's self-heating, which is retarded by the thermal inertia (heat energy storage)of the part.

Thank God for non-linear circuit simulators and complete component models.

I wish you the best....

2/8/2013 10:19 AM EST

It is disturbing that the capacitors do NOT have their rated capacitance at their rated voltage.

2/18/2013 11:50 AM EST

Thanks for a very informative article Mark, but now I'm wondering how I snuck by in the past, considering the number of times I've selected a capacitor voltage based simply on breakdown/withstand considerations or convenience (parts inventory on hand). Do you think this is largely a result of SMD component shrink, is it most pronounced with ceramic capacitors, and is it accounted for at all in the capacitance tolerance spec?

Regards