> The ringback here is the magnitude over the supply voltages(VCC and
> VSS). Therefore the ringback 0.8v for LVTTL (3.3V) is about 2.5V (=
> 3.3-0.8) for overshoot ringback. and 0.8V(=0+0.8) for undershoot
In that event, then 0.8V is much too conservative for overshoot
ringback with TTL and LVTTL.
For both TTL and LVTTL, Vih(min) = +2.0V. Anything from +2.0V to
at least Vcc is guaranteed to be interpreted as H. If you define
overshoot ringback relative to Vcc, then for TTL you could have at
least some 3.0V (=5.0-2.0) of ringback at nominal supply voltage,
before it could even approach being a problem. (But true TTL
rarely switches rail-to-rail, so maybe this is not the best way
to define ringback.) For LVTTL, it's 1.3V (=3.3-2.0).
At narrow pulsewidths, you could go even lower (greater ringback).
As for undershoot ringback, it would be 0.8V or more.
The true "threshold" is around +1.4V. As you see, there are sizeable
guardbands from +0.8V to +1.4V, and from +1.4V to +2.0V, to account
for variations and the finite voltage gain of the input stage. Newer
logic families such as GTL have narrower guardbands. In reality you
can impinge on these guardbands somewhat, especially when doing so for
only a brief time (a nanosecond or so). But this is not quantified
very well, and I doubt it is standardized.
5V optimized CMOS has different Vih, Vil thresholds, so its
ringback tolerances would differ. But I don't use 5V CMOS.
You are quite right about overshoot and undershoot themselves
being bigger problems.