When should decoupling capacitors be used.

Hi all,

A quick question when and where should i use decoupling capacitors.
Circuits that i have built at uni have always been giving to us and
there are often these decoupling capacitors but it is never be
explained why they are used and when they should be used. Could anybody
shine a light on this subject?

Thanks

Wayne

 

While there are meny exceptions and special cases, in general,
decoupling capacitors are added on each power pin on each IC. Their
purpose is to provide "local burst power" for the chip. Example: if a
chip switches its output in 1nS, there's going to be a sharp spike of
current needed by the chip to do that change. At those speeds, the
wires leading to the chip act like inductors, which block those kinds
of spikes. The capacitor provides the short burst of current needed
to accomodate that current spike. It also keeps the spike from
"spreading" to other parts of the circuit (it's like an LC low-pass
filter).

So, in general, an 0.1uF cap at each power pin of each chip. Keep
them as close to the pins as possible, keep lead length as short as
possible. Sprinkle a few 1uF caps around the board to decouple the
board itself, and a few 10uF caps near where the power comes into the
board to act as "bulk" capacitors.

For particularly fast chips, you'd put an SMD 0.01uF cap right at the
power leads, too.

 

By sprinkle you mean put them anywhere as well as the source pin on the
ICs. Will this not affect the circuit in other ways.

 

The whole point of decoupling is to decouple noise and droop. This is a
simple subject that has complicated issues and is covered regularly.

Search s.e.b, s.e.d and comp.arch.embedded for the subject for a number
of threads.

Cheers

PeteS

 

Well, I did say "in general" but yeah, I meant in addition to power
pin caps.

In my case, for example, with two-layer boards I use large copper
fills on the back for the power and ground nets. I like to put one
"sprinkled" 1uF cap for each copper region, so if there's only a tiny
bridge connecting it, it will still have its own bulk decoupling.

The idea being that, as you group things together into larger units,
you decouple that unit with the next size up cap. 0.1uF per pin, 1uF
per group, 10uF per board, etc. Using different size caps also avoids
problems with ringing, since you're avoiding having a "tuned circuit"
due to everyting having matched capacitances.

Each step decouples a higher frequency than the previous, and provides
power to the next.

 

Gotcha, didnt know about grouping and using larger capacitors. Thanks
for that

 

There is a pretty good article on that at
http://www.seattlerobotics.org/encoder/jun97/basics.html
As stated earlier, sometimes it is used as a current pool. When current
passes through a pc trace, it will generate a voltage (due to it's
impedance). You use the decoupling capacitor to short this out (or bypass it
to ground), so that it won't inadvertently toggle another circuit. Some
circuits are more sensitive than others to these types of noise. Also, some
types of circuits will generate more noise than others. There are rules of
thumbs to go by, but sometimes it is more of an art (developed through years
of experience).

Brian Ellis

 

Here are a few things that will help decoupling problems:

Ground returns are very important. Keep the ground returns for power
circuits separate from the low level sensitive circuits ground returns. The
lower the number of components on each ground return, the better. The wider
the ground return, the better. When possible, have each ground return go
back to the ground of the main filter capacitor. The shorter the ground
return, the better.

Counters and low level logics are usually more sensitive to noise, so
generally need more decoupling and better ground returns. Counters will also
generate more noise on the ground return.

Keep high power, analog and digital circuits as far away from each other as
possible. If they are all on the same circuit board, then their grounds must
be run very carefully (usually with a lot of decoupling).

I'm sure the others can add a lot to this.

Brian Ellis

 

prolly gonna get shot to bits on this one but here goes...

on the power rails of *anything* switching... as close to the power
pins of whatever it is you are decoupling as possible.

In a perfect world, logic is a nice clean 0-1-0 swing, but in reality
they are still analogue device and with TTL especially, the inrush
currents when the transistor saturate can cause dips and spikes on the
supply rails... very short, but they can get picked up by other
devices and cause malfunctions of your circuit. the correct D-cup
ahould remove normal glitches like this - you can still get rogue
chips or batches that will wallop the power whatever realistic
capacitor you use - these should be located and junked but this is
usually as part of test dammit! >(

OK.. I know this is a tad extreme, but you can't really have too many
and with 10nf at $0.002 (in suff. qtys) if you have room you should do
it.

Old style TTL was really horrible and used to spike the rails all the
time. I rememeber s100 bus boards with rows and rows of the stuff and
every one had a 10nF at the end - you can get capacitors that fit the
standard 14 & 16 pin (more?) DIL footprint and fit underneath chips...
not sure if this is a good idea - obviously the density goes up but
it's more hassle if you get a failure.