Robert, in addition to comments from Dennis and Steve Baldwin, it is
common to use parallel 0.1uf caps to provide more capacity and lower ESR.
This is a common technique when designing CPLD's and FPGAs. Search through
previous articles in the comp.arch.fpga news group for more info.

   As far as tantalum caps, there is a good reason that the market for them
is `huge'. They provide capacity and low ESR in a small space but they cost
more than monolithic solutions. While they do have a rather `dramatic'
failure mode, they are used in critical systems that operate in land, sea,
air, and space environments. It boils down to good engineering. If you
specify a 10uf/6V tantalum cap in a 5V environment where there may be large
transients, it's going to fail but so will the chips on the 5V rail...
I often specify a tantalum and a 0.1uf in parallel. The tantalum provides
the capacity and the other the fast response to transients. Again, this
depends on the environment.

   - Tom

At 10:59 PM 11/11/99 -0500, Robert A. LaBudde wrote:
>At 09:35 AM 11/10/99 +1100, Dennis wrote:
>> >So the output capacitor should be no smaller than 0.1 uF and no larger
than
>> >200 uF. Usually the breakpoint is in the cost of the capacitor. Since you
>> >can buy 0.47-1.0 uF tantalum caps for the same price as 0.1 uF, using a
>> >0.47 uF or 1.0 uF tantalum is the best choice.
>><snip>
>> >Larger capacitors provide no advantage at low frequency, since the output
>> >impedance of the regulator is very low (e.g., 1 mOhm). At high
frequencies,
>> >distributed bypassing works better in any even.
>>
>>Only if the CAPS are suitable for high frequency stuff. There is no need to
>>use a large cap (Say 1uF) that has a high impedance at RF. In this case a
>>small cap say 100pF ceramic would be better.
>
>Having found some more information, I thought I'd revive this issue.
>
>See the attached figure showing capacitor impedance vs. frequency.
>
>Consider the power regulation issue again:
>
>If the supply is, say, 5 V, and the ripple is to be kept less than 1% = 50
>mV, and the package is going to pull 100 mA, then an impedance of 50 /100 =
>0.5 ohm can be tolerated. If only a 10 mA load, then this increases to 5
ohms.
>
>A 1 uF tantalum has a flat impedance curve below 1 ohm from 100 kHz up to
>10+ MHz.
>
>A 0.1 uF monolithic ceramic has sharpness (Q) in its impedance curve, so
>its impedance is below 1 ohm only between 2 and 10 MHz. It acts as a notch
>filter at about 1 MHz.
>
>A 1000 pF poly film cap has < 1 ohm impedance only from 50 MHz to 1 GHz,
>and acts as a notch filter at about 100 MHz.
>
>If a PIC circuit operates < 20 MHz, why would you prefer a smaller ceramic
>over the 1 uF tantalum, when it's got a better bandwidth on low impedance?
>Particularly since at 20 MHz the tantalum is probably lower impedance than
>the smaller capacitance ceramic?
>
>If op-amps with frequencies below 1 MHz are driven by the supply, higher
>capacitance is a more important factor in impedance than type is.
>
>So I ask again: Why use a small capacitor to bypass power on these types of
>circuits? Why isn't the 1 uF tantalum a good solution, albeit a few cents
more?
>
>It still seems to me you only need a very large capacitor at the input of
>the regulator, with 1-100 uF at the output, and 1 uF tantalums at the chip
>level to control switching noise.
>
>As I said before, I'm not an EE, so perhaps I'm missing something in the
>spec sheets here.


------------------------------------------------------------------------
Tom Handley
New Age Communications
Since '75 before "New Age" and no one around here is waiting for UFOs ;-)