Electrolytic vs Cermic Big Decoupling Caps

[gigavolt]

Hey, another newbie question here. I am wrapping up a design where I use both 10uF and 0.1uF caps to decouple some chips. Looking on digikey, they do sell 10uF chip capacitors and they are less money than electrolytics. Howevere I see other people still putting big electrolytic caps on their boards. The only other difference I can think of is ESR, but that should also be a win for the chip capacitors. Am I missing something and is there a reason you would go with electrolytics here?

[redkitedesign]

There can be a lot of reasons..

High capacitance ceramics have a large voltage derating, especially the smaller (0603) parts. This can be up to 90% at their rated voltage.
The lower ESR can also be a disadvantage; in combination with an inductance somewhere in the power they create very high Q LC-filters.

And it can be plain old conservatism. It used to work with electrolytics, so why change?

[reboots]

High-value MLCC caps typically use a Class III dielectric material with poor temperature stability, such as Y5V. Explanation of the different classes here:

https://www.kemet.com/en/us/technical-resources/heres-what-makes-mlcc-dielectrics-different.html

This means that the value will drift and degrade considerably with temperature, and with working voltage. A nominal 10uF chip capacitor might only be good for a few uF under many conditions.

The cheaper MLCC dielectrics are also piezoelectric, generating electrical noise from physical vibration.

For a simple bulk-capacitance application, where noise can be tolerated, these issues might not be a big deal. Simply add enough capacitance to derate for the worst case. But for more critical e.g. sensitive analog or digital applications, the performance might be unacceptable.

Electrolytics have their own issues, but don't require the same dramatic derating of capacitance as high-value MLCCs.

https://en.wikipedia.org/wiki/Derating

[redkitedesign]

High-value MLCC caps typically use a Class III dielectric material with poor temperature stability, such as Y5V.

Most high-value MLCC caps used nowadays are X5R or X7R, and dont have the ridiculous temperature dependence Y5V showed. There is really no reason anymore to use Y5V.

That doesn't remove the voltage derating however, which is really an issue for all MLCC dielectrics.

The cheaper MLCC dielectrics are also piezoelectric, generating electrical noise from physical vibration.

Not only the cheaper, ALL MLCC's are piezoelectric. However, there efficacy as microphone is limited, and totally irrelevant when used for power supply decoupling.

But the piezoelectric effect works both ways: If your power supply is oscillating between 20Hz and 20kHz, you will hear it!

[crgarcia]

Take a look at this "myth buster" video. It explains a lot of things on decoupling capacitors:

[David Hess]

Take a look at this "myth buster" video. It explains a lot of things on decoupling capacitors

And it propagates as many if not more mistakes.  The example circuit he has is not representative of how small ceramic decoupling capacitors are used in most applications, where inductance and *charge*, which yields the minimum capacitance value, are important.  Further he does not cover why bulk decoupling with electrolytic capacitors is used at all when large ceramic capacitors are available.

The application circuits he complains about use 0.1 microfarads as a standard value because it is the largest inexpensive value which will cover most cases, and not because it is the ideal value.  And it is usually a waste of time to optimize.  Look instead for an application note which specifically discusses decoupling and power distribution networks.

[radiolistener]

small smd capacitors probably have worse ESR, because they have less conductor and bad isolator with high losses. So, they have high power loss on heating.

But it depends on the construction and materials, for example very expensive porcelain capacitors with silver plated conductors are good for RF due to a very high Q-factor and ultra low ESR, but they are very expensive for using as a usual power filtering capacitors.

On the other hand, using low ESR polymer capacitors also can be bad for some circuit, because they have low loss so they can reflect pulses without damping and it can lead to unwanted ringing and oscillations in circuit which is not designed for low ESR capacitors.

So, it depends on the purposes. Some circuit needs low ESR polymer capacitor, some needs porcelain ulra low ESR, some needs usual electrolyte, some needs usual ceramic. Attempt to replace one type with another type can cause serious issues, it may lead to device malfunction, even if you try to replace for example electrolyte with polymer capacitors.

[JustMeHere]

The main thing you are doing with decoupling is trying to keep a ground loop as short as possible.  MLCCs have a very short lead. With a ground and power plane the ground loop of a MLCC can be 1 or 2 mm.

A ground loop is the path the current takes and the path the displacement current takes.  Displacement current moves in the opposite direction.  The two are magnetically coupled.  It takes time for these two currents to "set up."

When these currents are setting up there is a strong magnetic field pulling the currents together.   This you know as noise or cross talk. (You can see this on your scope when in AC, or changing current, mode)The best way to solve this is with a short circuit. That won't for DC.  We need something that blocks DC.  A capacitor does.

There are two components to a current change.  There is the rate, and the volume.    The rate how fast the signal changes.  The volume is the size of the change.  The maximum rate of the change is limited by the ground loop. 

The volume of the change drains energy from the capacitor, So small capacitors have a problem keeping up.

Bulk capacitors are good with solving the volume problem.   And the ground loop currents are already set up before the volume part is applied.  So their size is not a limiting factor.

Think of a square wave that changes once every half of a second.   This is a 1 Hz signal.  Be careful with this.   This is not the frequency we are interested in.  The time it takes for the signal to go from low to high is the frequency you are interested in.

If the rise time is 1 us (micro second).  The frequency is actually 1 MHz. In order for the ground loop to set up in 1 us, it must be very short.

If the change volume is too large the small capacitor runs out of energy.  This is where the bulk capacitor steps in.  Before it can step in, it needs to set up it's ground loop.  This takes longer than the smaller capacitor.  If the smaller capacitor can "hold on" long enough for the large capacitor to "get ready" the voltage won't drop when the big capacitor is ready to "hold on" too.

It's probably over simplified, but it's easy to wrap your head around and helps with layout decisions.   

[David Hess]

Bulk capacitors are good with solving the volume problem.   And the ground loop currents are already set up before the volume part is applied.  So their size is not a limiting factor.

Besides supplying a low impedance at low frequencies, bulk capacitors are good for suppressing ringing in the PDN by terminating it at its characteristic impedance.  This explains the mystery about why some combinations of bulk and decoupling capacitors cause unacceptable ringing on the PDN.

[nachus001]

Hay!

Just a suggestion, if you are going to design a power bus, you can place one or more series of capacitor and resistor in different parts
of the circuit between +vcc and ground planes .
Capacitor between 100n and 10u and resistor 1ohm onwards.
If you notice excessive ringing, you can populate those series in order to tame the ringing by
changing the values of resistor and capacitor

cheers
Nachus

[David Hess]

Just a suggestion, if you are going to design a power bus, you can place one or more series of capacitor and resistor in different parts
of the circuit between +vcc and ground planes .
Capacitor between 100n and 10u and resistor 1ohm onwards.
If you notice excessive ringing, you can populate those series in order to tame the ringing by
changing the values of resistor and capacitor

High reliability designs which eschew electrolytic capacitors do exactly that for bulk decoupling.

[crgarcia]

Just a suggestion, if you are going to design a power bus, you can place one or more series of capacitor and resistor in different parts
of the circuit between +vcc and ground planes .

Hi Nachus!

Do you think you could point us at an example?

Do you mean something like this:

[Jwillis]

The only other difference I can think of is ESR, but that should also be a win for the chip capacitors.

That's not necessarily true. Different chemistries in capacitors can exhibit different properties at different frequencies. Electrolytics have the highest efficiency at low frequencies where ceramics can drop off in efficiency at low frequencies, The ESR on a ceramic can be several hundred  times or more at 100 to 120Hz than an equivalent valued electrolytic. This creates heat, less efficiency and shorter life expectancy. You could use a tantalum but the cost goes up compared to an equivalent electrolytic. It's a balancing act when selecting capacitors for specific applications.

[Infraviolet]

Post #11:

The trick for producing a high capacitance decoupling with ceramic caps, without getting the turn-on/turn-off LCvoltage spikes due to the ceramic's low ESR (hence it take a big current for a very brief moment when power is first applied) if to add a mall resistor (1 to 10 ohms) in series with the ceramic cap. So one end of the cap goes to ground, where the other end of the cap goes to the small resistor which then goes to your positive power rail. Sometimes you can just add the resistor in series with one of the ceramic cap and leave other ceramic caps in parallel with the single RC series combination and still not get any spikes during turn-on/off.

[AnalogTodd]

I have spent decades designing linear regulators and a lot of that time is making sure I understand the types of capacitors out there (since capacitors form a dominant pole for stability in almost every regulator). Good designs include both input and output capacitors and their characteristics depends on the requirements of the circuit.

Some notes:
1. Electrolytic capacitors: Construction of these capacitors usually lends itself to higher ESR (50-100milliohm minimum to several ohms maximum), though there are newer designs that can get single digit milliohm ESR. Capacitance remains constant across voltage unlike many ceramic dielectrics. These are polarized devices, so you can't plug them in backwards or subject them to excessive reverse voltages or they will blow up (failure mechanism depends on type of capacitor as well).
2. Ceramic capacitors: The multi-layer approach to construction on these capacitors offers very low ESR and ESL (5 milliohm is a typical value). Dielectric material used does affect temperature characteristics as well as voltage characteristics. For smaller values, the NP0 and C0G dielectrics are very stable both with temperature and voltage. Larger values using Z5U and Y5V have much larger temperature and voltage variations (the codes indicate initial tolerance and min/max temperature) while X5R and X7R tend to offer better temperature characteristics. What determines voltage dependency on these dielectrics comes down to physics; you can't infinitely increase capacitance without affecting breakdown voltage or the voltage coefficient of the capacitance, hence larger capacitors with the same voltage rating and dielectric type will have less capacitance loss versus DC bias.

So why would one use an electrolytic versus a ceramic? Electrolytics with their ESR are often used on the input of regulators together with smaller low ESR ceramics. Many people do not use just one type or the other but instead have both. From a transient perspective, ceramics offer the ability to handle fast transients--the low ESR means that energy gets pulled from the capacitance without a significant voltage transient. So why add an electrolytic? The ESR is actually useful. If there are long wires or traces feeding your regulator, the ceramic capacitor with its low ESR paired with the inductance of the wires/traces creates a high-Q tank circuit. These circuits want to ring when perturbed since there is nothing to damp them out. Adding an electrolytic capacitor kills this ringing. Multiple time I've seen circuits where a high-Q tank circuit resonance on the input of a regulator was actually close enough to the loop crossover frequency of a regulator and the whole thing oscillates even though the regulator would normally be stable otherwise.

Questions? Feel free to post them here.

[Jim from Chicago]

If the rise time is 1 us (micro second).  The frequency is actually 1 MHz. In order for the ground loop to set up in 1 us, it must be very short.

If the change volume is too large the small capacitor runs out of energy.  This is where the bulk capacitor steps in.  Before it can step in, it needs to set up it's ground loop.  This takes longer than the smaller capacitor.  If the smaller capacitor can "hold on" long enough for the large capacitor to "get ready" the voltage won't drop when the big capacitor is ready to "hold on" too.

So if I understand correctly, the important thing is not to place the smallest value (in Farads) closest to the output, rather you want to place your smallest package capacitor closest, and the fact that the smaller package happens to usually be the smallest value is just incidental. So in a perfect world which defied physics, you could just use one capacitor in a very small package with a very large capacitance value.

[David Hess]

So if I understand correctly, the important thing is not to place the smallest value (in Farads) closest to the output, rather you want to place your smallest package capacitor closest, and the fact that the smaller package happens to usually be the smallest value is just incidental. So in a perfect world which defied physics, you could just use one capacitor in a very small package with a very large capacitance value.

The package with the lowest ESL and ESR, or impedance, should be closest.

But I have designed and fabricated transmission line circuits, meaning the capacitors were all equally distant, where even with the same package ESL and ESR, multiple capacitors with multiple values had to be placed in parallel to achieve a low impedance over a wide frequency range.  If all of the capacitors were the larger value, then the impedance at higher frequencies was higher, presumably because self resonance was at a lower frequency.