Winfield Hill said:
Bill Sloman wrote...
Good conservative advice for a novice I suppose, but talking
about "guaranteed maximum frequency" is rather misleading, and
not very accurate in my experience. For example the spec is
usually a maximum delay time or a minimum "maximum" frequency,
and it's measured in a test circuit with added capacitance
(e.g. 50pF) that's much higher than one usually encounters in
a real application, and at a lower voltage, (e.g. 4.5V instead
of 5V). As a result actual circuits usually can go _faster_
(not slower) than these conservative "specs." For example, TI
gives 21MHz for their 74hc191 counter, but I bet one would be
hard pressed to find a real part with such a slow upper limit.
I should hope so. The guaranteed maximum frequency ought to be the
2.5- or 3-sigma tail of the distribution, so fewer than one part in a
thousand should fall over at that frequency, at worst case temperature
and supply voltage.
In a production environment, that is what you design for, so that very
few of your boards fall over under worst case conditions.
At Nijmegen University, one of my colleagues designed a board around
the actual performance of 74F, and he got it back about once a year,
as one or other of his chips drifted back from its initially measured
performance.
Okay for a reasearch one-off, perhaps. But when the client started
complaining about systematic pattern-related clock jitter, I
redesigned the board using ECL for the critical clocks, and never had
to touch it again - going over from DIP packaging to surface mount for
most of the components to made enough room for the extra ECLinPS chips
(only available in surface mount anyway).