[sdiy] How to Determine Value of Power Supply Decoupling Caps?

[Magnus Danielson]

From: ASSI <Stromeko at compuserve.de>
Subject: Re: [sdiy] How to Determine Value of Power Supply Decoupling Caps?
Date: Wed, 26 Nov 2003 22:52:06 +0100
Message-ID: <03112621591100.00646 at Robert>

> On Wednesday 26 November 2003 16:56, john mahoney wrote:
> > I have seen the ".1 uF rule" mentioned in several places. Having
> > followed the links provided by Philip Gallo, however, I was left with
> > the impression that .01 uF SMD caps were the best "RTBC" (rule of
> > thumb bypass caps) to use on ICs. :-)
> 
> These are actually two rules of thumbs for two different situations and 
> from different times. One is from a time where the thought of running 
> something that was not called a radio at more than 10MHz would freak 
> out most folks and the other is from a time where running something 
> with almost 500 pins at 2.6GHz (please say hello to my microwave oven) 
> by sinking 80A into it elicits shrugs from passers-by.

This is another way of indicating what they bypass capacitors really are, they
are tiny pools of energy. One may naïvely think that "but I got my by honkin-
tonkin powersupply just down the lines here...", and indeed you do, but that is
the whole issue... down the lines... now, cables, PCB traces or whatever have
parasitic inductance, so at higher frequencies their impedance goes up and thus
will a quick change/high frequency not "feel" a low impedance voltage source
as it does down in DC. This is why we toss in bypass capacitors to start with,
since when the impedance of the lines back to the powersupply raises, the
capacitor takes over, so that the capacitor will loose charge as it delivers
current in place of the power-supply, and when the switch-impulse is finished
it will recharge. For quick designs, you have to put these small caps really,
really close (low inductance, small physical size, low values) to each of the
chips power-pins, then a little way back a larger one (to charge the little
ones and take some of the other heat) and on we go. Another thing one does is
to use the PCB as a big capacitor, since when you set up a pair of metal sheets
with in parallel, it does form a capacitor, and that is how the power-planes
act anyway. Power-planes also has a lower inductance than separately routed
traces.

So, decoupling for larger designs with high speeds is a little art in itself.
The old TTL style "a 100nF cap over each chip" was the old and lazy days.
Then again, I've seen some fairly complex designs which started out basically
capacitorless (PPG Wave's).

> If you find explicit power bypassing recommendations in a datasheet, 
> you can be sure that there is a good reason for it and you should 
> follow it to the dot unless you _know_ differently.

I.e. you built up a good understanding about the "black magic". I've seen
several engineers greatly fail on the "_know_ differently" part and sure
enought they have troubles with their designs.

> The recommendations on typical values also have (probably much) to do with
> the available types and ranges of values for the caps at the time when the
> rule was cast (meaning: commonly available and being as big as practically 
> possible). As with everything, bigger isn't always better: if chosing 
> the next value up means more series inductance (and/or having to lay 
> out longer traces) it may not be a wise choice for your application. If 
> inrush current from the caps kills the power regulator on power-up or 
> stored energy frys some unsuspecting IC on power down, then you've 
> overstepped the boundary as well.

Indeed. You can buy incredibly compact high-C caps these days (POSCAPS). It is
easy to scatter many of them around the board (seen it done). Good stuff used
correctly but deadly when used too much. This is another misstake people do,
when something is unstable, the try to toss more high-C caps on the problem
rather than trying mid-C or low-C caps at the right place. In the end, you
should apply the capacitance which is best for the job in rise/fall-time.

> In particular, the recommendation for using tantalum capacitors is not 
> really a good one for some types of modern tantalum caps. SMD caps are 
> now available in a much larger range, so you can easily find 100nF 
> types that are as good as the older 10nF ones (beware, the "old" series 
> are still around - they aren't bad, just different).

The dielectrum makes a difference, and the design of the cap itself, in
lowering parasitic inductance.

There is a reason modern LCR-meters of class uses sweeped sine and is really
a network analyser rather than the traditional bridge.

Hints: Use a network analyser and/or TDR.

Cheers,
Magnus