Ceramic Capacitor Behavior

[mawyatt]

There's been some discussions on the bad behavior of ceramic capacitors and specifically how the high dielectric types have a strong bias voltage dependance.

Here's an example of how one might use a simple RC Low Pass Filter for PWM, Bandwidth Limit, or Noise Filter application. This LPF is a 1K resistor with a shunt 0.1uF cheap ceramic disc of unknown origin. Three filters were assembled on a Protoboard and two of the filters had the ceramic shunt capacitor biased at 10V and 30VDC, while the first had no DC bias and used as a reference for comparisons.

The 1st plot shows the Bode response for 10 to 1MHz without any DC bias applied to any filters, and the three filters behave similar as expected. Magenta(C2) is the 1st LPF, Blue(C3) the 2nd LPF and Green(C4) the 3rd LPF.

The 2nd plot shows the 10V DC bias applied to the 2nd filter (Blue) and 30VDC applied to the 3rd filter (Green), note how the low pass corner is pushed out as the DC bias reduces the effective shunt capacitance. The unbiased filter 1st trace (Magneta) has a -3dB corner of ~1.4KHz, while the 2nd (Blue) 10V bias is moved out to ~4KHz. The last plot shows the same result with the cursor moved to the 3rd trace (Green) with 30V bias and a -3dB corner of ~8.3KHz.

This shows quite a change in filter performance caused by capacitance dependance wrt to applied DC bias voltage. 

Best,

[RoGeorge]

Bad behavior maybe, but I see there a Voltage Controlled Filter for audio. 
Analog synth musicians should be thrilled by the opportunity.

Thanks for posting the measurements! 

[mawyatt]

They (high K Dielectrics) also have really bad hysteresis

Best,

[tggzzz]

I've also been able to eyeball this in the time domain. Not only was an RC risetime dependent on the amplitude, but as the amplitude increased the risetime became visibly non-exponential.

Replaced the (small allegedly C0G) capacitor with an air-gap trimcap, and order was restored to the universe.

[mawyatt]

Doubt that was a C0G you replaced, these are known for their excellent overall characteristics with almost no voltage dependance. They are so good that they are used as the main core integration capacitor in many high resolution leading DMMs

Best

[T3sl4co1l]

Neat thing about large-signal waveforms here, you can get something more like a linear, or with an inflection point anyway, than an exponential curve on your square wave.

Can you do it with a ferrite bead as well?  Current bias?  That would be of interest for related threads.   (Albeit no currently running threads, I think.)

Tim

[Harry_22]

Good job!

Below is some Microcap modeling according to your research.

[nctnico]

It has been known for a long time to not use ceramic capacitors in signal paths where low distortion (for example audio) is critical. Another pitfall with ceramic capacitors is leakage current which can be in the double digit nA range.

[T3sl4co1l]

Microphonic also; film and tantalum can be preferred for low-frequency signal purposes for these reasons.

Tim

[Harry_22]

Ceramic is used in some places of a guitar tube amplifier to color the sound.

[mawyatt]

Good job!

Below is some Microcap modeling according to your research.

Nice work yourself with theses models

The phase effects as shown with your L R model is due to the setup and not actually due to capacitor, sloppy setup using long leads and a Protoboard, good enough to show the DC Bias effects tho!!

Edit: Grabbed one of the cheap caps and our LCR Meter DC Bias Adapter with some custom plotting software using a TH2830 LCR Meter, SDP3303X Supply and AG34401A DMM to sweep Capacitor DC Voltage (need to work on the plotting routines, our software skills are not that good!!!)

https://www.eevblog.com/forum/projects/bias-network-for-lcr-meter/

https://www.eevblog.com/forum/testgear/lcr-meter-plot-software/

Best,

[mawyatt]

More bad ceramic capacitor behavior. Plot generated with TH2830 LCR, SPD3303X and SDM3065X for temperature reading. Heating source is ceramic resistor, see below for details.

https://www.eevblog.com/forum/projects/lcr-lab-grade-parameter-plotting-fixture/

Edit: 0.01uF STE Blue Disc Ceramic, 10nF Green Mylar Film and Yellow 0.1uF added, note the better Mylar performance, especially the popular Yellow Rectangular type. Also a 10nF Mica as a reference for stability.

Best,

[coppice]

More bad ceramic capacitor behavior. Plot generated with TH2830 LCR, SPD3303X and SDM3065X for temperature reading. Heating source is ceramic resistor, see below for details.

I've seen people do temperature tests on their new design, get widely varying results across the temperature range, and then try to blame every part of the design except the capacitors. 

People keep talking about the variation of capacitance of these capacitors with DC, but it varies with AC too. The capacitance can vary through each cycle of the signal, leading to some interesting non-linearities in the outcome.

[harerod]

Neat thing about large-signal waveforms here, you can get something more like a linear, or with an inflection point anyway, than an exponential curve on your square wave.

Can you do it with a ferrite bead as well?  Current bias?  That would be of interest for related threads.   (Albeit no currently running threads, I think.)

Tim

Thanks for the cue, I was going to post this link:
http://harerod.de/applications_eng.html#EmiFilTest
This contains a "test protocol with pictures" for download, which I attached to this post. I think I put that online, after one of our discussions about current bias of ferrite beads.

These little boards are really helpful to check real life performance against datasheets. The first board shown can be used to voltage bias a capacitor or current bias an inductor. More examples are in the pdf. I have several more that are not shown online.

[mawyatt]

More bad ceramic capacitor behavior. Plot generated with TH2830 LCR, SPD3303X and SDM3065X for temperature reading. Heating source is ceramic resistor, see below for details.

I've seen people do temperature tests on their new design, get widely varying results across the temperature range, and then try to blame every part of the design except the capacitors. 

People keep talking about the variation of capacitance of these capacitors with DC, but it varies with AC too. The capacitance can vary through each cycle of the signal, leading to some interesting non-linearities in the outcome.

Capacitors are far from ideal under almost all conditions, and folks should pay close attention to their application.

Since the problem with voltage sensitivity applies to the voltage across the capacitor (differentially), for a given value capacitance lower frequencies become more problematic as the capacitive impedance is higher and usually a larger signal differential voltage across the cap for the volt dependent non-linear effects to work with. Whereas at higher signal frequencies the capacitive impedance is lower and less differential signal voltage, however now ESR and ESL become more problematic.

So there seems to be no "Sweet Spot" frequency-wise for general capacitor use, and likely why we have so many different types!!

Best,

[T3sl4co1l]

I always found it funny that, many people in audio circles, malign for example, the use of electrolytic capacitors in the signal path (coupling, filtering); no, they're not particularly good components, at all, but that's exactly the point, you use a low enough cutoff frequency in the coupling network that you don't care...

Of course, that was also when I realized some people will form beliefs about things, independent of any rational argument for or against those things

Tim

[mawyatt]

Some of the electroyltics we've tested aren't too bad as long as the peak signal voltage doesn't venture so far as to reverse the instantaneous voltage across the cap to much, and most behave better than the ultra-high K ceramic types wrt to signal/bias voltage and temperature effects.

Agree, the "audio" stuff sometimes seems "magic" to certain folks and defies common sense

Maybe we can sell some of these negative noise resistors and anti-distortion capacitors that suck out the signal noise and distortion

Best,

[TimFox]

It has been known for a long time to not use ceramic capacitors in signal paths where low distortion (for example audio) is critical. Another pitfall with ceramic capacitors is leakage current which can be in the double digit nA range.

That opinion in audio circles is based on capacitors other than C0G/NP0.
Only recently have relatively large values (> 100 nF) been readily available in C0G/NP0, so film capacitors (PS or PP) were the appropriate choice.

[Smokey]

I finally got an LCR that can measure DC Bias and threw in the 16V, 47uF ceramics I've been using as DC-DC converter 5V output caps...

https://search.murata.co.jp/Ceramy/image/img/A01X/G101/ENG/GRM32EC81C476KE15-01.pdf
Funny that in the data sheet they give this generic "Example" plot that shows capacitance change at 5V being negligible (or even positive?).

But the actual product page shows the real plot for this specific part....
https://www.murata.com/en-us/products/productdetail?partno=GRM32EC81C476KE15%23

capacitance change of -42% at 5V for a 16V rated part.  Measurements confirm essentially this.
well... that's bigger than I expected.. I'm actually surprised the converter was stable.

[ArdWar]

You shouldn't take charts in whole family template datasheet like that as anything more than showing "yes, a C-V chart looks like this".
All (almost?^[1]) the bigger capacitor manufacturers do have their own online detailed parametric charts for each P/N. Use that instead if you have to.

[1]: Idk if they all got one. I'm only familiar with those from KEMET, Samsung, Murata and TDK.

[T3sl4co1l]

I finally got an LCR that can measure DC Bias and threw in the 16V, 47uF ceramics I've been using as DC-DC converter 5V output caps...

https://search.murata.co.jp/Ceramy/image/img/A01X/G101/ENG/GRM32EC81C476KE15-01.pdf
Funny that in the data sheet they give this generic "Example" plot that shows capacitance change at 5V being negligible (or even positive?).

Yeah, a lot of people misinterpret this.  I forget if it's Murata or someone else, but another datasheet shows several colored curves, which one might understand as including the linked part in question.  Nevermind that in both cases, you're looking at the family datasheet, mostly dull junk about packaging, specifications and methods, quality and so on.

This one at least is specific that it's the "Sample: X7R(R7) Characteristics 0.1μF, Rated Voltage 50VDC" part, clearly not just anything in the family.

It would probably be better if they didn't have a curve at all and just had verbiage to the effect of "characteristics vary with temperature and voltage; consult the characteristics database at xxx for more information" -- or, you know, they actually generated the fucking datasheets per-part with curves.  They can, Samsung does (sometimes..), but most still do not.

[1]: Idk if they all got one. I'm only familiar with those from KEMET, Samsung, Murata and TDK.

I would consider AVX pretty big.  Venkel perhaps as well.  Neither offers characteristics on anything last I checked.  Their loss; I've no problem buying their film caps or whatever, but type 2 ceramics are a strict no-no for me.

Tim

[Smokey]

This brings up an interesting question....

For all you DC-DC converter designers, what is your process for selecting specific part numbers for input and output capacitors? 
.. How much work do you do up front scrutinizing datasheets ..
.. vs buying some caps and characterizing them yourself on their own ..
.. vs putting them in circuit and seeing how the converter works and going from there?

[Someone]

This brings up an interesting question....

For all you DC-DC converter designers, what is your process for selecting specific part numbers for input and output capacitors? 
.. How much work do you do up front scrutinizing datasheets ..
.. vs buying some caps and characterizing them yourself on their own ..
.. vs putting them in circuit and seeing how the converter works and going from there?

Where models/parameters/characteristics are available online you can usually get it right in simulation and have no surprises on the first hardware revision.

[Smokey]

This brings up an interesting question....

For all you DC-DC converter designers, what is your process for selecting specific part numbers for input and output capacitors? 
.. How much work do you do up front scrutinizing datasheets ..
.. vs buying some caps and characterizing them yourself on their own ..
.. vs putting them in circuit and seeing how the converter works and going from there?

Where models/parameters/characteristics are available online you can usually get it right in simulation and have no surprises on the first hardware revision.

So you simulate with specific part numbers and matching specific manufacturer models for all the caps?  That still leaves the question of your selection criteria for picking caps to try in the simulation.

[Someone]

This brings up an interesting question....

For all you DC-DC converter designers, what is your process for selecting specific part numbers for input and output capacitors? 
.. How much work do you do up front scrutinizing datasheets ..
.. vs buying some caps and characterizing them yourself on their own ..
.. vs putting them in circuit and seeing how the converter works and going from there?

Where models/parameters/characteristics are available online you can usually get it right in simulation and have no surprises on the first hardware revision.

So you simulate with specific part numbers and matching specific manufacturer models for all the caps?  That still leaves the question of your selection criteria for picking caps to try in the simulation.

Once the characteristics of candidate parts are known simulation might be directly on that, or a degraded/toleranced version, or on some worst case of a set of possible caps. Depends on what the time/budget/constraints are.

Knowing which parts to look for in the first instance is either experience or time consuming (or somewhere inbetween).