17 Jul
2007
17 Jul
'07
8 p.m.
Michael B. Brutman wrote:
Tony Duell wrote:
I need some books and a double E degree. One more question ...
Some of what I'm reading says put the capacitor near the power supply,
and other things are saying get it as close to the chip as possible.
(With a few rules, I'm simplifying.) If the bypass capacitor is being
used to filer the AC component, wouldn't the answer for a good design
be "do both"?
Tony is exactly correct; what degrades the performance of bypass
(decoupling) capacitors most directly is the series inductance of PCB
traces (and wires), not DC resistance. For good high frequency
performance, small bypass caps should be as close as possible to a
devices power/ground pins to minimize the inductance.
That being said, as you mention there is not just one function of bypass
capacitors in a design. Generally speaking, small bypass caps (eg,
10nF-100nF) across a devices power/ground pins serve the function of
tying both the DC power and ground rails together at high frequency,
shunting noise and providing a very low impedance path between the rails
for signal return currents.
Larger bulk capacitors (10uF and larger) provide short term charge
storage to handle the current transients of the loads that the power
supply cannot. Placement on the PCB of these larger bulk caps is usually
not critical, and most times is done near the power supply or its
connection to the boards (like near the fingers on DEC boards).
So to answer your question, usually 'both' are required; larger caps
near the power supply connection, and smaller caps per device
power-ground pin pair (as a general rule of thumb).
Get the
cap as close to the chip as you can: think of the trace as a
(really low resistance) resistor: the resistance PS-to-cap is is much
larger than resistance cap-to-chip and bingo! low-pass filter.
Actually, it's the inductive reactance of the trace/PSU connections
that matters here. With high-frequency signals -- this doens't mean a
high clock rate necessarily, it means a fast ris-time edge -- you can
get significant voltage drop from very small 'inductors', like the
connction from the chip to the PSU.
-tony