decoupling capacitor size

Other Hardware General Electronics
1

Hi!

I'm making a project with a MAX31855 thermocouple amplifier chip. It has to have decoupling capacitors on the power lines, and also on the input lines. It is a very sensitive chip, and I found I have to increase and increase the decoupling capacitor to have a better behaviour. The spec sheet recommends 10nF, but I found even 1uF has some noise spikes every now and then.

This got me thinking, is there a "too high" value of decoupling capacitor, if my time scale is not too small (0.1-1s) ? Is there any reason why people recommend certain size of decoupling capacitors for power lines or signal lines? How could I think about these in general?

2

If the capacitor value is too high the the high frequency responce suffers. That is why you have to use both a small cap 10 to 100 nF in parallel with a big one 10 to 100uF.

3

This got me thinking, is there a "too high" value of decoupling capacitor, if my time scale is not too small (0.1-1s) ? Is there any reason why people recommend certain size of decoupling capacitors for power lines or signal lines? How could I think about these in general?

There isn't (in theory) and there is (in practice).

The decoupling caps are there to provide instant "energy" to the chip - important when the chip is operating at high frequencies and the power supply wire presents itself as an inductor - which prevents the energy being delivered to the chip as quickly as could have hoped / demanded.

So in theory, the bigger a decoupling cap, the better off you are.

However, bigger caps tend to be electrolytic capacitors that have high ESR / ESL, making them slow in delivering such charges. Smaller caps tend to have lower esr/esl, and different caps (tantalum for example) tend to have lower esr as well.

So what you see in practice is a compromise: low value capacitors that hold adequate amount of charge and have low esr.

If the capacitor value is too high the the high frequency responce suffers. That is why you have to use both a small cap 10 to 100 nF in parallel with a big one 10 to 100uF.

If too high capacitance causes frequency responce to suffer, why would adding more capacitance helps?

It is for something else.

4

BTW, I have used capacitors (some times electrolytics) of 4.7uf on high speed digital circuits with no ill effect.

There is no "red line" on this, so to speak.

5

Thanks, it pretty much confirms my understanding. In this particular case I was working on the frequency response is not an issue, so I'll just try some bigger caps than so far.

Cheers!

6

so I'll just try some bigger caps than so far.

It is not terribly critical. I usually use whatever I have and I tend to stay small. But starting with what the datasheet suggests helps. And for highspeed circuits, starting with a ceramic disc (monolithic or otherwise) helps as well. I usually avoid tantalum as they tend to fail short. 10nf - 110nf is fairly common.

7

dhenry:

so I'll just try some bigger caps than so far.

It is not terribly critical. I usually use whatever I have and I tend to stay small. But starting with what the datasheet suggests helps. And for highspeed circuits, starting with a ceramic disc (monolithic or otherwise) helps as well. I usually avoid tantalum as they tend to fail short. 10nf - 110nf is fairly common.

Tantalum are rubbish for decoupling. So are most electrolytics.

Here's Atmel's advice on decoupling capacitors: http://www.atmel.com/Images/doc0484.pdf

8

fungus:

dhenry:

so I'll just try some bigger caps than so far.

It is not terribly critical. I usually use whatever I have and I tend to stay small. But starting with what the datasheet suggests helps. And for highspeed circuits, starting with a ceramic disc (monolithic or otherwise) helps as well. I usually avoid tantalum as they tend to fail short. 10nf - 110nf is fairly common.

Tantalum are rubbish for decoupling. So are most electrolytics.

Here's Atmel's advice on decoupling capacitors: http://www.atmel.com/Images/doc0484.pdf

Thanks a lot, this is very useful in general!

9

There isn't (in theory) and there is (in practice).

There is no difference between theory and practice. If one is apparently seen then it is because the theory has been simplified to such an extent that it is wrong.

I have used capacitors (some times electrolytics) of 4.7uf on high speed digital circuits with no ill effect.

If this is decoupling capacitors then no you won't see any ill effects, it is just that there will be reduced / no beneficial effects.

So in theory, the bigger a decoupling cap, the better off you are.

No that is over simplifying it. Some components like low voltage drop out regulators can become unstable with too much decoupling capacitance. Also some DC / DC converters will not start up if there is too much decoupling. It also can produce a switch on current surge that can trip circuit breakers or blow fuses.

However, bigger caps tend to be electrolytic capacitors that have high ESR / ESL, making them slow in delivering such charges.

No the ESR is only of interest in cases of ripple current, it is not a figure you can apply to a DC discharge.

If too high capacitance causes frequency responce response to suffer, why would adding more capacitance helps?

Because you have two capacitors. At high frequencies the big one looks like an inductor and has a high reactance and so is doing nothing for you, but the small one has a low reactance and so is shorting out the interference. Where as at low frequencies the small capacitor does nothing for you but the big one can absorb the ripple.

tantalum are rubbish for decoupling. So are most electrolytics.

They are both good for bulk decoupling, the low frequency stuff. However, tantalum capacitors used in power supply decoupling are a fire hazard as when they fail they tend to fail short circuit and burst into flames. That is why you won't get UL approval on a system using tantalum capacitors on a power supply.

10

Grumpy_Mike:

tantalum are rubbish for decoupling. So are most electrolytics.

They are both good for bulk decoupling, the low frequency stuff. However, tantalum capacitors used in power supply decoupling are a fire hazard as when they fail they tend to fail short circuit and burst into flames. That is why you won't get UL approval on a system using tantalum capacitors on a power supply.

I'll stick to ceramic+polyester...

11

Because you have two capacitors. At high frequencies the big one looks like an inductor and has a high reactance and so is doing nothing for you, but the small one has a low reactance and so is shorting out the interference. Where as at low frequencies the small capacitor does nothing for you but the big one can absorb the ripple.

Great explanation, Reader's Digest style, :slight_smile:

12

Because you have two capacitors. At high frequencies the big one looks like an inductor and has a high reactance and so is doing nothing for you, but the small one has a low reactance and so is shorting out the interference. Where as at low frequencies the small capacitor does nothing for you but the big one can absorb the ripple.

Try that on a scope (and if you cannot afford a scope, try it in a simulator), and see for yourself.

13

dhenry:

If the capacitor value is too high the the high frequency responce suffers. That is why you have to use both a small cap 10 to 100 nF in parallel with a big one 10 to 100uF.

If too high capacitance causes frequency responce to suffer, why would adding more capacitance helps?

It is for something else.

The point is the series inductance to the electrolytic makes it much less effective at high frequencies - it won't be as "stiff" in controlling the voltage on short timescales (its fine in the longer term). Small ceramic placed close to the chip is there to do the initial decoupling after the step-change that would otherwise knock the supply rail down (or up). The combination covers all the timescales. The voltage regulator takes over at lower frequencies.

For many digital circuits you can probably get away with being a bit sloppy on decoupling, but if you mix analog and digital you immediately have to be much more systematic on decoupling to reduce noise injection via the supplies. The larger the currents being switched (LED arrays for instance) the more careful you have to be to decouple adequately. Follow best practice always and you don't have to worry though :slight_smile:

14

The point is the series inductance to the electrolytic makes it much less effective at high frequencies - it won't be as "stiff" in controlling the voltage on short timescales (its fine in the longer term).

Close, but not exactly there.

Take a typical capacitor model (C+ESR+ESL). Model the line from the power source to your load as a 0ohm resistor + inductance. Now, put a sudden load on the whole thing, and watch the i-t response.

Play with the ESR+ESL combination and you will see why people parallel small (highspeed) capacitors with large capacitors.

Here is a hint: the wire inductance functions as a high-impedance element for highspeed loads (aka a digital circuit). Without a local decoupling capacitor, you will see substantial voltage drop on the load as the current goes up suddenly (di/dt for an inductor).