Combined MLCC

[Koen]

Hello,

    if reference designs often call for 2.2uF bulk then 0.1uF bypass and as all smd parts tend to become tinier, is there such a thing as combined MLCC ? Both values in the same part. I noticed 'stacked capacitors' but these were gigantic.

Thank you,
Koen

[kony]

X2Y and feedtrough designs will do what you are after decoupling or filtering wise.

[T3sl4co1l]

Yes, it's called using the one part and forgetting about anything else!

Such appnotes never provide an actual explanation of what they are doing, with proof that it works.  I've never seen a PDN (power distribution network) analysis in one.  Sure, they'll wave their hand saying, oh, the distributed values also distribute the resonances, and something something it probably works -- but this isn't actually an explanation, because it is willfully ignorant of the fact that, when two resonators are linked together, a third antiresonance occurs, between the two, with the opposite type of resonance.  In this case, two capacitors together -- each one series-resonant at some frequency -- have a parallel resonance at the frequency inbetween.  Parallel resonance is high impedance == more ripple voltage for a given ripple current.

In short, stacking caps of different values and sizes is more likely to make things worse.

Better is simply putting one or a few, modest value, small size chip caps at the device pins, and lossy "bulk" caps nearby.  The loss dampens the resonance with connecting traces.

An easy way to play around with this, is to build a ladder network (alternating series L and parallel-to-ground C) in SPICE.  Set up reasonable source and load impedances, and adjust the inductances (corresponding to trace length: approx. 1nH/mm) and capacitor values and types to see what behaves better or worse.

Tip: an 0603 size 0.1uF X7R ceramic usually has around 100mOhm ESR and 2nH ESL.

Source impedance will be whatever the source is; a lot of sources have electrolytic caps, so that will dominate their output impedance (i.e., a Thevenin equivalent source with Rs = ESR).  Loads are typically capacitive (power pin capacitance), some with resistance (CMOS logic should be mostly resistive, with the resistance corresponding to its average load current at the supply voltage), some with constant current (ECL logic, and most analog circuits), some with negative resistance (DC-DC converters -- don't forget to include their input capacitance and parasitics, too!).

The most important, most general truth to discover: the resonant impedance is Zo = sqrt(L/C), and the resonant frequency is Fo = 1 / (2*pi*sqrt(L*C)).  The Q factor is ESR/Zo, or Zo/ESR, depending on if it's parallel or series resonant.  In short: for a well-damped supply, you need ESR on part with Zo, and you need both on par with the required ripple response (change in load voltage divided by change in load current = maximum supply impedance as seen by the load).

Tim

[dmills]

The other major takeaway is that some ESR is NOT a bad thing, to the point that you can actually buy MLCC parts with specified (and somewhat high) ESR precisely because they are often better behaved then the very low ESR parts that are more commonly seen.

An MLCC with say 0.1 ohms of ESR is going to cause you far fewer resonant problems then one with a few mOhms ESR simply because on aggregate the Q will be very much lower.

Regards, Dan.

[PointyOintment]

I guess it's worth pointing out that if you want to increase the ESR, but can't find a cap of the right value with higher ESR (or it's too expensive), you can just put a resistor in series with the capacitor.

[dmills]

But that adds a few nH of ESL....

Time was this sort of thing mattered only to those who were into designing ultra low phase nose sources and other such esoterica, that we are discussing it on a BEGINNERS forum says something deeply scary about this place. 

Regards, Dan.

[T3sl4co1l]

But that adds a few nH of ESL....

Time was this sort of thing mattered only to those who were into designing ultra low phase nose sources and other such esoterica, that we are discussing it on a BEGINNERS forum says something deeply scary about this place. 

Regards, Dan.

Blame two things: the ever-forwards march of technology; and the appnote writers, always trying to be helpful but never actually checking their work.

On the upside, the added ESL is usually not a problem: you're damping against trace ESL, which might be tens of nH, so the added few-nH isn't a big deal.

Tim

[Koen]

Thank you for your answers. I've been trying to extract the very last drops of a GNSS receiver by toying around C-FB-C filters and various combinations of bypass capacitors. About the question above, I simply thought I was missing the right keywords to Google.

[Koen]

Somewhat related question to the ESR/ESL discussion above : can a higher DCR on an SMPS inductor be better noise-wise ? Current inductor is Murata LQH3NPZ2R2MMEL 2.2uH, 0.078ohm DCR, 2A rating, 3x3x1.5mm but I only really need 50mA out of it. It is followed by 2.2uF and 0.1uF capacitors. Thank you !

[dmills]

There are some truly shockingly poor app notes out there, I have seen examples from the big players more then once where the suggested circuit had clearly never been actually built and tested.

And yea, things like decoupling are a common place for the stupid to come out, I especially love those composite graphs of a few different values of cap where they have clearly only considered the magnitude of each parts impedance and not the 180 degree phase shift....

Regards, Dan.

[T3sl4co1l]

2.2uH resonates with 2.3uF at 70kHz, with a resonant impedance of ~1 ohm.  With capacitors of ESR < 50mohm and similar for the inductor, you'll have a Q factor around 8, and a peak supply impedance around ~8 ohms.

If the noise is at 70kHz, it won't be particularly good, compared to, say, using a tantalum or electrolytic of 2.2uF and ~1 ohm ESR.

This is only a direct application of the relations provided above, in the hope that an example will further inspire the reader to apply them.

50mA is low enough that you might use a ferrite bead without worrying about saturation; a 100-680 ohm (at 100MHz) part in 1206 or larger size should do fine.  This will have a lower Q than the inductor proper, but it will still be better with the lossy cap.

Tim