[sdiy] What's all this decoupling cap business??

[Magnus Danielson]

From: "Paul Schreiber" <synth1 at airmail.net>
Subject: [sdiy] What's all this decoupling cap business??
Date: Wed, 26 Nov 2003 10:22:27 -0600
Message-ID: <000d01c3b439$86e50f40$0201a8c0 at speakeasy.net>

Paul,

> The 0.01uf decoupling versus 0.1uf decoupling is a "leftover" from about 1982. Before then, caps
> were the round, 10mm 'ceramic disc' types. However, about 1981 a company called AVX invented a
> technology called "monolythic multi-layer" capacitors. Instead of 2 parallel plates with
> dielectric in the middle (ceramic disc), the caps were made like stacking a deck of cards,
> alternating plates and dielectric.
> 
> The upside was the *physical size* of the caps (in this case, the *height*) was GREATLY reduced.
> Since the actual cap was tiny, you could make them *axial* leaded (like a resistor) and, (drum
> roll, please)..........use an auto-insertion machine without a special handler! If you were IBM,
> AT&T or Tandy, this was a HUGE deal.
> 
> Now, AVX had a good thing going, and they chared out the a** for them (the AVX sales rep at Tandy
> went from a Chevy to a Mercedes within 8 months), but....and this was a big but......the
> capacitance (at THAT time) was limited to 0.022uf.
> 
> So, let the battle begin! All the ceramic disc people could do is claim cheaper and larger
> capacitance. But, AVX published literally 50 "white papers" showing that 0.01uf was fine. Now,
> you don't suppose that had ANYTHING to do with the fact they couldn't produce 0.1uf caps, do you?
> 
> However, I still use 0.1uf although MOTM has Kemet caps, not AVX (basic patent has expired, Kemet
> cheaper).
> 
> Just is little historical perspective, showing how "standard engineering practice" can be
> initiated by MARKETING :)

Facinating, but not all of the facts straight. There are situations in
decoupling where a 100 nF cap is less suited than a 10 nF cap. This occurs at
higher frequencies. If you make a impedance measurement of a cap you will find
that as frequencies gets higher, the impedance lowers, until a certain point
when it starts to rise again! This point is the self resonance, and at the
self resonance it has stopped being capacitively reactive and become purely
resistive and at frequencies above the self resonance the capacitor has
inductive reactance, i.e. it acts like an inductor.

What has happend is that the stray inductance of the capacitor has become so
great that it compensates out the capacitance.

Now, the reason that we need to care about this problem is that the working
frequencies isn't as low as it used to be. It is everyday buissness to spew
frequencies at and above 100 MHz down the powerlines of a chip to the innocent
capacitor. Since the resonance frequency decreases with increased capacitance
and/or inductance, we want to keep both in the lower end for a chip bypass, and
we want to keep the reactance capacitive (so it spews out power). While we
might choose capacitors with low intrinsic inductance, and while we may route
a low-inductance path to the cap, in the end we also needs to lower the
capacitance too. At times will capacitors of different values even be
paralleled, like 10nf//100pF and stuff like that.

So, it is not only marketing that affects the choice of capacitor value, but
stray inductance and other non-ideal physics kick in as well.

Cheers,
Magnus - nor Chevy or Mercedes here...