On Tuesday, September 9, 2014 9:05:04 AM UTC-5,
David Riley wrote:
They're
usually intended as a drop-in for standard aluminum
electrolytics, so the form factor will probably be all wrong
if you're trying to replace tantalums, but they're a lot
closer to tantalums in terms of nominal ESR without a lot of
the associated problems (particularly regarding derating of
the operating voltage).
I wouldn't worry too much about ESR for the stereotypical
decoupling. I just toss in 0.1 uF ceramics as per tradition
(not per engineering) and it always works anyway. Or
electrolytics if its an RC for a one shot or something.
Digital components are a lot tougher than people give them
credit. As an example where analog/RF types pull hair out, is
the stereotypical 0.1 we use for decoupling has a self
resonance below our system clocks (figure like 4 to 8 MHz
depending on exact type) and WAY below the gate switching
current spikes freqs. This drives the RF guys nuts because I
spend like $3 each on exotic MLCC chip caps with crazy high
self resonances for my microwave work, because I prefer my
capacitors to act as capacitors in the ckt rather than little
inductors or resonant cavities.
I have a nearly completely populated console IO board on my
workbench and I'm tempted to hook it up with RF gear in place
of the regulator and sweep the board and see where it
resonates with no (or few) actives installed. Intuitively
somewhere around 300 MHz it'll make a nice quarter wave
antenna, and thats so far above the resonant freq of the
decoupling caps that plus or minus their ESR who knows what
it'll act like. I should bring a working board into work and
sniff a working system with a spectrum analyzer sometime.
I'll hook up a board to an antenna analyzer too, for some
laughs. Maybe I could use a board with no actives populated
as an ultra QRP 70 cm antenna LOL. If time permits etc etc
I'll report back with some RF data.
This is tangentially related to the issue where the frequency
aka reciprocal of duration of spike, depends solely on logic
family and has nothing to do with system clocks (other than
here comes a 100 MHz gate switching noise spike, although it
is at a 4 MHZ pulse repetition rate).
Digital stuff is tough and although it doesn't have the power
supply CMRR that an op amp can have, you can feed some nasty
power in and it'll still work. Of course for when it doesn't,
that's why they invented schmidt triggers and stuff like that.
I know the actual history and reasons, but I've often thought
one of the best things about the S100 bus in on board
regulation, with decent regs, means excellent power noise
immunity between boards that bus designs with a large central
regulator don't have. Given the cost and availability of
TO-92 sized 7805 I've occasionally daydreamed about board
designs where you ignore power noise because every chip has
its own dedicated regulator... That would be interesting
indeed. And a pain to solder up. Still, TO-92 regs are
cheaper than the higher quality machine pin sockets, so its
not entirely insane as far as daydreams go.