Author Topic: Is this really a 100uF cap or marketing BS  (Read 2169 times)

0 Members and 1 Guest are viewing this topic.

Offline JustMeHereTopic starter

  • Frequent Contributor
  • **
  • Posts: 826
  • Country: us
Is this really a 100uF cap or marketing BS
« on: March 28, 2020, 12:16:36 am »
I bought some of these caps (see link) on the idea they were 100uF caps and would work well with the esp32.  After testing them, they are all in the 75uF area.  So I looked at the spec sheet (linked) and saw in the graph on page 2 that they are only 100uF around 400-500 kHz.  Is this common?  Should I really consider these 75uF caps?

https://ds.yuden.co.jp/TYCOMPAS/ut/download?pn=JMK316ABJ107ML-T++&fileType=Datasheet
 

Offline Prehistoricman

  • Regular Contributor
  • *
  • Posts: 216
  • Country: gb
Re: Is this really a 100uF cap or marketing BS
« Reply #1 on: March 28, 2020, 12:20:57 am »
Page 2 of a 1 page document?  :o

I also like how their graph stops at 500kHz but the scale continues to 10GHz.
 
The following users thanked this post: piguy101

Offline amyk

  • Super Contributor
  • ***
  • Posts: 8526
Re: Is this really a 100uF cap or marketing BS
« Reply #2 on: March 28, 2020, 12:32:08 am »
Ceramic caps have decreasing capacity with increasing applied voltage.
 

Online Alex Eisenhut

  • Super Contributor
  • ***
  • Posts: 3550
  • Country: ca
  • Place text here.
Re: Is this really a 100uF cap or marketing BS
« Reply #3 on: March 28, 2020, 12:57:59 am »
It's common for ceramic caps to be specified at 0V bias. You'll find you lose a lot of your uF rating with only a few volts bias. I have no idea why the value changes so much with frequency however. The kinds of dielectric to get these high capacitances in such small packages probably have all kinds of undesirable characteristics in the name of higher Dk.

To reassure you, they're all like that and Taiyo Yuden is a good brand.

There's debate about what's better for decoupling power supplies. I think lower value caps with more predictable specs are better than a capacitor tuned for maximum on only one spec.
Hoarder of 8-bit Commodore relics and 1960s Tektronix 500-series stuff. Unconventional interior decorator.
 

Online wraper

  • Supporter
  • ****
  • Posts: 17952
  • Country: lv
Re: Is this really a 100uF cap or marketing BS
« Reply #4 on: March 28, 2020, 01:25:21 am »
Pretty typical class II ceramic MLCC. Most likely will be used at something like 1-1.5V power rail anyway. And 80% drop of capacitance at rated voltage is far from the worst. Some are left with <10%.
EDIT, BTW if you look at capacitance/frequency chart, you may end up with effective capacitance much higher that rated.
« Last Edit: March 28, 2020, 01:31:12 am by wraper »
 

Online wraper

  • Supporter
  • ****
  • Posts: 17952
  • Country: lv
Re: Is this really a 100uF cap or marketing BS
« Reply #5 on: March 28, 2020, 01:39:37 am »
Quote
After testing them, they are all in the 75uF area
Heat them with soldering iron or hot air, let cool down and then measure again. Your measurement is not about DC bias but ageing which is reset when heated above certain temperature. Then capacitance will start to drop again with time.
https://www.murata.com/en-us/support/faqs/products/capacitor/mlcc/char/0006
« Last Edit: March 28, 2020, 01:41:44 am by wraper »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Is this really a 100uF cap or marketing BS
« Reply #6 on: March 28, 2020, 02:18:18 am »
Yes, measured value will be significantly below rating after aging, and under any condition of voltage or temperature.

Note that they're measured at some ACV, too.  C drops at low ACV, peaks at modest levels, then drops off again at higher voltages.

If you want a better capacitor, consider aluminum polymer, or a lower-k grade ceramic.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline trobbins

  • Frequent Contributor
  • **
  • Posts: 826
  • Country: au
Re: Is this really a 100uF cap or marketing BS
« Reply #7 on: March 28, 2020, 03:28:50 am »
Perhaps it's a good lesson to learn - 'know' what you are buying.

Are you making up a pcb yourself, or wanting to replace an existing decoupling part on a pcb, and what is the operating voltage?
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 7454
  • Country: pl
Re: Is this really a 100uF cap or marketing BS
« Reply #8 on: March 28, 2020, 06:04:06 am »
I'm not sure if the increase of capacitance with frequency is real or some measurement artifact caused by rising impedance or whatnot.

It seems to imply that feeding the same amount of charge into the capacitor results in less increase of voltage if it's done faster. Why? :wtf:

edit
There is another thing. 100µF is relatively high capacitance for an MLCC. I have seen other "high density" MLCC whose size was very consistently 20% larger than normally, which just coincided with the tolerance specified in the datasheet. Is it possible that 100µF±20% also means 100µF-20%?
« Last Edit: March 28, 2020, 06:11:14 am by magic »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Is this really a 100uF cap or marketing BS
« Reply #9 on: March 28, 2020, 06:43:42 pm »
Well, it does; but if you restrict your thinking to instantaneous charge, you're going to have a hard time agreeing.  It's an AC steady state measurement, so needs to be approached with sine waves.

If we have a series resonant RLC circuit, we have the impedance:
\[ Z = R + j \omega L + \frac{1}{j \omega C} \]

If we equate this to capacitive reactance, well, we can't, reactance is the imaginary component of impedance and this is clearly a complex expression.  We have two obvious choices: throw out the real component, or take the magnitude as if it were reactance (ignore angle).

In the latter case, we have,
\[ X_C = \sqrt{R^2 + \left( \omega L - \frac{1}{\omega C} \right)^2 } \]
which doesn't really say much, so let's poke at it a bit.  For \$\omega \rightarrow 0\$, we have \$\omega L \rightarrow 0\$ and \$\frac{1}{\omega C} \gg R \$, so it reduces to:
\[ X_C(0) = \frac{1}{\omega C} \]
as we expect.

For \$\omega^2 \approx \frac{1}{LC}\$, things are more complicated.  Exactly on, it reduces to:
\[ X_C = \sqrt{R^2 + \left( \frac{1}{\sqrt{LC}} L - \frac{1}{\frac{1}{\sqrt{LC}} C} \right)^2 } \]
\[ X_C = \sqrt{R^2 + \left( \sqrt{\frac{L}{C}} - \frac{1}{\sqrt{\frac{C}{L}} } \right)^2 } \]
\[ X_C = \sqrt{R^2 + \left( \sqrt{\frac{L}{C}} - \sqrt{\frac{L}{C}} \right)^2 } \]
\[ X_C = \sqrt{R^2} \]
\[ X_C = R \]
Which gives an equivalent,
\[ C = \frac{1}{\omega R} \]
\[ C = \frac{1}{\frac{1}{\sqrt{L C_0}} R} \]
\[ C = \frac{\sqrt{L C_0}}{R} \]
\[ C = C_0 \sqrt{\frac{L}{C_0}} \frac{1}{R} \]
\[ C = C_0 \frac{Z_0}{R} \]
which tells us, at resonance, the equivalent capacitance can be higher or lower, depending on the ratio of characteristic impedance \$Z_0 = \sqrt{\frac{L}{C_0}}\$ to R.  (Note I'm using \$C_0\$ to denote the nominal capacitance, and C the apparent capacitance.)  Which sounds suspiciously like the Q factor of the circuit.

So you might next wonder: can there be a peak in capacitance?  Can we prove the plotted curve is realistic?  Under what conditions will it peak, or just roll off, or whatever?

Well, we can look for local maxima by taking the derivative with respect to \$\omega\$ and setting it to zero.

I won't write out the algebra on this one, but Wolfram Alpha is happy to do it for free:

\[ \frac{d}{d\omega}  \frac{1}{\omega \sqrt{R^2 + \left( \omega L - \frac{1}{\omega C_0} \right)^2}} = 0 \]
is the setup, which evaluates to an all-positive solution,
\[ C_0 (2 L^2 \omega^2 + R^2) = 2 L \]

\[ 2 L^2 \omega^2 = \frac{2 L}{C_0} - R^2 \]
\[ \omega^2 = \frac{2 L}{2 C_0 L^2} - \frac{R^2}{2 L^2} \]
\[ \omega = \sqrt{ \frac{1}{C_0 L} - \frac{R^2}{2 L^2} } \]

And C measured at this frequency gives, ... well, it doesn't want to parse such a long expression for free.  Well, suffice it to say it'll give an answer higher or lower than Co depending on the Q, and that's why it rises in the curve.

A more hand-waving way to put it: the impedance is falling suddenly, due to a Q>1 resonance, and the phase is turning sharply towards resistive.  This is simply equivalent to a rising capacitance.  Past resonance, capacitance is negative (because the impedance is inductive).

Tim
« Last Edit: March 28, 2020, 06:47:33 pm by T3sl4co1l »
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 
The following users thanked this post: TiN, Siwastaja, magic

Offline JustMeHereTopic starter

  • Frequent Contributor
  • **
  • Posts: 826
  • Country: us
Re: Is this really a 100uF cap or marketing BS
« Reply #10 on: March 28, 2020, 09:39:27 pm »
Perhaps it's a good lesson to learn - 'know' what you are buying.

Are you making up a pcb yourself, or wanting to replace an existing decoupling part on a pcb, and what is the operating voltage?

I'm making the PCB myself.  Everything works.  I accidentally switched my power supply from 5v to 30v and made some magic smoke.

I was testing these caps to see if they were damaged and read them to be 75 uF at most frequency ranges on my LCR.   I then measured one fresh out of the packaging.   I saw the same thing.  That's when I went to the spec sheet to see what the C vs Hz was supposed to be.  That's when I saw the graph suggesting the cap only gets to 100uF in a very limited environment.  SMPS at 500k might work very well with these.

I got these to help with the sudden power demand the ESP32 can place.  There is also a 100nF cap with them to be the high frequency filter.
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 7454
  • Country: pl
Re: Is this really a 100uF cap or marketing BS
« Reply #11 on: March 29, 2020, 07:20:41 am »
Well, it does; but if you restrict your thinking to instantaneous charge, you're going to have a hard time agreeing.
I straight out refuse to agree on the grounds of pedantry ;D

It seems you just demonstrated that it is indeed a measurement artifact resulting from ignoring ESL and trying to blame its effects on capacitance and ESR. Perhaps it's a convenient simplification for some steady state AC applications, but that's it. It's not even a convenient way of modeling transient conditions in said "steady state" AC applications.

I suppose if you add that L into the equivalent circuit, most if not all of the apparent frequency dependence in C and R will go away.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Is this really a 100uF cap or marketing BS
« Reply #12 on: March 29, 2020, 07:40:57 am »
Well, it does; but if you restrict your thinking to instantaneous charge, you're going to have a hard time agreeing.
I straight out refuse to agree on the grounds of pedantry ;D

That is, a hard time in the sense of: you can do it in the time domain with differential equations, or you can use transforms as I implicitly did above.  The agreement is inevitable, as the results are equivalent; unless you're not much of a pedant after all and refuse on non-technical grounds. ;D


Quote
It seems you just demonstrated that it is indeed a measurement artifact resulting from ignoring ESL and trying to blame its effects on capacitance and ESR. Perhaps it's a convenient simplification for some steady state AC applications, but that's it. It's not even a convenient way of modeling transient conditions in said "steady state" AC applications.

I suppose if you add that L into the equivalent circuit, most if not all of the apparent frequency dependence in C and R will go away.

Well no, L remains in the expressions, as you can plainly see. :)

What's inconvenient about modeling it?  Also, what's being modeled?

For transient applications like SMPS, the simple fact that the impedance is low suffices.  Its angle doesn't matter at all.  I'm quite happy to use whatever values of capacitors, as long as they do what they do!

Put another way: a ladder network of inductors, if they have alternating large and small values, is still just as good a filter as for any other angle of impedance.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline magic

  • Super Contributor
  • ***
  • Posts: 7454
  • Country: pl
Re: Is this really a 100uF cap or marketing BS
« Reply #13 on: March 29, 2020, 09:19:30 am »
L shows up in your expression because you use the RLC model with constant R, L and C to derive a C or RC model with frequency-dependent R and C, which mimics the way naive capacitance/ESR meters work.

The C in your C model is not the same as the C in your RLC model.

Philosophically, we can argue which C is the "true" C of the capacitor ;) I'm going to say it's the RLC C because the RLC model more closely reflects the physical reality, which is plain magnetic inductance existing in the capacitor rather than some magical property of the dielectric or frequency-dependent variation in electrode spacing.

By transient conditions I mean startup, shutdown or abrupt frequency change, when resonance gets out of whack and sudden peaks or dips are possible due to the L. Yes, the models are theoretically equivalent so you could model it in frequency domain by integrating the variable capacitance over all frequencies present in the transient, but in practice it would be a rather masochistic approach.
« Last Edit: March 29, 2020, 09:21:04 am by magic »
 

Offline rsjsouza

  • Super Contributor
  • ***
  • Posts: 6106
  • Country: us
  • Eternally curious
    • Vbe - vídeo blog eletrônico
Re: Is this really a 100uF cap or marketing BS
« Reply #14 on: March 29, 2020, 09:32:47 am »
For a few additional insights about construction and a few prscticsl aspects.. Check free_electron's posts in this almost eight year old discussion

https://www.eevblog.com/forum/beginners/esr-meter/
Vbe - vídeo blog eletrônico http://videos.vbeletronico.com

Oh, the "whys" of the datasheets... The information is there not to be an axiomatic truth, but instead each speck of data must be slowly inhaled while carefully performing a deep search inside oneself to find the true metaphysical sense...
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Is this really a 100uF cap or marketing BS
« Reply #15 on: March 29, 2020, 11:45:50 am »
L shows up in your expression because you use the RLC model with constant R, L and C to derive a C or RC model with frequency-dependent R and C, which mimics the way naive capacitance/ESR meters work.

The C in your C model is not the same as the C in your RLC model.

Correct, I separated out C as the LF-asymptotic, nominal or rated value, and Co as the apparent (measured) value.

An RLC model is just an approximation to a real component, and we can apply higher order approximations if we so desire.  You are welcome to solve the algebra for such a case; the RLC case however proves that a more complicated model is not necessary to explain the phenomenon, and indeed it is the least complicated model possible to show this effect. :-+


Quote
Philosophically, we can argue which C is the "true" C of the capacitor ;) I'm going to say it's the RLC C because the RLC model more closely reflects the physical reality, which is plain magnetic inductance existing in the capacitor rather than some magical property of the dielectric or frequency-dependent variation in electrode spacing.

Again, yeah, C is nominal, Co is measured.  Apologies if that was not clear in the derivation (I thought I had added a paragraph to that effect, perhaps it was not well marked).


Quote
By transient conditions I mean startup, shutdown or abrupt frequency change, when resonance gets out of whack and sudden peaks or dips are possible due to the L. Yes, the models are theoretically equivalent so you could model it in frequency domain by integrating the variable capacitance over all frequencies present in the transient, but in practice it would be a rather masochistic approach.

A much simpler model applies to these cases:

Because C varies so much with signal level and bias, it would be, at the least not very meaningful, and at worst rather disingenuous, to apply a high-order linear model.

Instead, the design should be guardbanded: assume minimum and maximum possible values for the nonlinear elements in question.  If the circuit works at both extremes, and can also be shown to work inbetween*, that's it -- you don't need any higher order models, don't need complicated impedances, it works for a whole space of variation.

*For example, solving for the Q of the PDN (power distribution network) and checking that its maxima is either one or the other endpoints, or no more than a definite maximum value inbetween.


This is somewhat aside, but incidentally -- frequency step change is easily modeled, for a linear system anyway.  Take this example:
https://www.seventransistorlabs.com/Images/Induction725.jpg
LLC (series driven parallel resonant) tank, driven with a FSK waveform.  The envelope seems to resonate at different frequencies.  It turns out, these are actually the difference between F_driven and Fo.

This is easily understood from diff eq -- an inhomogeneous LTI system (i.e., one with an independent driving term, and only linear time-invariant dependent terms) has two responses: the homogeneous solution (the diff eq's natural solutions, in whatever proportion follows from initial conditions), and the driven solution (proportional to the driven source).

Also, that we can separate the driving source into two components: a tone burst at F1, interleaved with a tone burst at F2.  These signals have complementary windows, so that their sum is apparently a continuous FSK signal.

When we evaluate the system for a transient start or stop of the driving waveform at F1, we observe basically the first envelope.  A signal which in turn is composed of the transient (homogeneous) solution, essentially the system's impulse or step response, plus the driven solution, essentially the system's filtering effect upon the driven signal (a phase shift and gain term).

When we evaluate the system at F2, we get the second solution.  Sum them together and you get the whole repeating waveform.

A less formal way to look at it, but just as valid (and accurate!): suppose there's a source, with frequency at F1, which simply stops, drops to zero completely.  Well, the tank will ring down some.  Over a time constant of Q / Fo, more or less.  Also, it had been driven at F1, but the tank doesn't know that, it just rings down at Fo, its characteristic frequency.  Suppose we superimpose that response on the response of F2 suddenly starting, which will be a ramp up at the same time constant, at the different frequency, and reaching a different final amplitude.  What does their sum look like?  Well, the tank rings down at Fo.  We take the envelope of the sum.  What's the envelope over two different frequency components?  It's a DC offset plus their difference frequency.  And what should the envelope look like, given that one of those components is decaying and the other is increasing?  Well, it'd be a bounce of some sort.  And this is exactly what we observe in the waveform!  Finally, we can calculate the expected magnitude of those peaks or dips, simply from the resonant parameters of the circuit.

The application for this image was induction heating, so, amplitude control is relevant; this shows just how challenging the control loop will be for a frequency-shift type control.  But it's also useful knowledge for PLLs, communications signals (e.g. the required bandwidth of a FSK channel) and so on.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline trobbins

  • Frequent Contributor
  • **
  • Posts: 826
  • Country: au
Re: Is this really a 100uF cap or marketing BS
« Reply #16 on: March 29, 2020, 11:56:34 am »
SMPS at 500k might work very well with these.
Yes I'd expect like many very high-density multilayer ceramics, there are quite a few material and performance tradeoffs in the quest for ever higher capacitance density, and material non-linearities start to really stand out.  SMPS may be one application, although the ability to put bulk caps right at IC/load device terminals, with no track meandering out to some convenient place to put a bulk cap, can be very beneficial.
« Last Edit: March 29, 2020, 11:58:24 am by trobbins »
 
The following users thanked this post: JustMeHere

Online Conrad Hoffman

  • Super Contributor
  • ***
  • Posts: 2078
  • Country: us
    • The Messy Basement
Re: Is this really a 100uF cap or marketing BS
« Reply #17 on: March 29, 2020, 01:53:48 pm »
IMO, the dielectric properties are terrible, but the real question is what can be accomplished in a given volume and is it adequate for what the circuit needs. For many low voltage applications the high capacity MLCC is the best available choice, even if not a "pure" and mathematically elegant component.
 
The following users thanked this post: Siwastaja, JustMeHere


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf