> Again, I'm not an EE, so perhaps I'm misunderstanding this issue. But a
> square-wave has odd harmonics with power dropping as 1/f^2. If the
> impedance of a capacitor rises proportional to f, then the product still
> drops.

It's not a square wave but a series of current impulses that occur at
intervals of 1/f, the duration of which are a function of the
rise/fall times. The capacitor has to do its thing when the gates
change from one state to another. This is when the current is drawn
from the capacitor/supply.

> The graph I found was from Rohm, who was describing the capacitor product
> lines they make. What types of leads were testing, I don't know, but it
> wouldn't seem as if they would want to portray their own products poorly.

You are thinking along the right lines in that ceramic caps are
fastest, tantalums quite fast and electrolytics, relatively slow. The
balancing act is in cost and capacity.
A 10nF ceramic cap may suitable for an unloaded PIC and be able to
supply the needed current during transitions. But if you now put a
few loads on the pins (like track capacitance), the little 10nF runs
out of steam very quickly.
You may decide that you require 1uF to cover this load. A 1uF ceramic
cap is still fast but it is also bulky and expensive.
A Tantalum here gives fairly good high frequency response and will
give you the capacity. You could use a wet electrolytic and probably
get more capacitance for the same price. Problem is they are slow and
won't do the switching transitions so you can use the ceramic to take
up that slack.

Rohm have the advantage of their circuit only consisting of a
capacitor. More complex designs require compromise.

> If size is the premier issue, then small capacity surface mounts would seem
> a good start. But how do you prove that they will meet bypassing requirements?

If you bring surface mount vs. through-hole into it, you are adding
another twist. You can safely generalise that surface mount anything
is better from a high frequency viewpoint because leads are inductive
and you want the inductance to be as low as possible.

> If over-voltage survival is the key issue, then high voltage capacitors
> would seem the starting point. But won't the filter caps on the regulator
> blow too? Won't the chip the bypass is protecting blow instead? Where are
> these power spikes or surges post-regulation coming from? If we've got
> them, doesn't it mean the onboard regulation isn't working?

The main problem is the failure mode. Tantalums fail short circuit
with a great deal of heat. They sit there burning with a hot flame. So
when they do go, they take out the rest of your circuit and make a
mess of your board. Other types of electrolytics have some degree of
self healing although they will go bang if you hit them hard enough.

It's back to compromise and cost-effectiveness. Larger capacitors
cost more and the unwanted parameters (inductance, resistance,
physical size) also go up. Like everything else, we can only reduce
the probability and severity of external influences like surges. Do
you design to survive lightning 10 miles away, 1 mile or a direct
hit ?

> This simple issue is an opportunity for me to understand how engineering
> requirement analysis should be performed to down-select design decisions.
> In the past, I've just taking the word of others on what the standard
> practices are.

You can go through the calculations yourself if you like. It's
similar to selecting a value for a pullup resistor. You can calculate
till you are blue in the face and come to the conclusion that
absolute values are relatively unimportant but that this parameter
is more important than that one and a particular value will suffice
in most instances

> However, I've noticed over the years that the 'standard' has shifted. 25
> years ago, you were supposed to put a 0.01 uF ceramic across power on TTL
> packages that generated incredible noise and used incredible power. Then
> the recommendation shifted to 0.1 uF ceramic or tantalum. Now we should
> have a combination of a high value and a small value. People are including
> large filter capacitors on the outputs of regulators.
>
> If board space and cost are key issues, why include multiple capacitors? If
> package noise and power usage have dropped dramatically, why require more
> bypassing than 25 years ago? A 7805 regulator was only good to +/- 5% 25
> years ago, but now its +/- 1%. Why do we then need more, not less, bypassing?

CMOS switches from rail to rail very quickly, requiring a short pulse
of current. CMOS inputs are also referenced to the supply voltage so
if the supply drops because of something switching, so does the
logic0/1 threshold. TTL on the other hand, has thresholds set by
transistor junctions so it is less susceptable to this.

As the switching speeds increase, the need to get that current pulse
out of the cap becomes more important. Big caps just can't do it so a
compromise is required. The little caps can supply the needs of the
chip but also need to be charged up again and fairly quickly. The
next capacitor up the chain can be larger and a bit slower to provide
that function. Those caps in turn, need to be topped up and if the
regulator isn't fast enough, yet another cap can bridge the gap.

> Even if I wasn't a 'rocket scientist' I'd realize there's something wrong
> with these 'creeping requirements'.

That's because the technology is creeping and the decoupling rules
have to keep up.

Steve.

======================================================
Steve Baldwin                Electronic Product Design
TLA Microsystems Ltd         Microcontroller Specialists
PO Box 15-680, New Lynn      http://www.tla.co.nz
Auckland, New Zealand        ph  +64 9 820-2221
email: steveb@tla.co.nz      fax +64 9 820-1929
======================================================