[SOLVED] What does propagation delay mean?

By the definition of this webpage, ASIC-SoC-VLSI Design: Transition Delay and Propagation Delay, propagation delay is: the time required for a signal to propagate through a gate or net.

This definition of propagation delay implies the input signal and the output signal have a causal relationship.

Now, let's consider a case:
There is a 2-to-1 multiplexer with inputs in0, in1 and sel; an output pin out.
in0, in1 and sel transit from 0 to 1 at the same time, and then out transits from 0 to 1,
and we measure these delays by, for example, HSPICE.
(1) Measure in0 to out delay.
(2) Measure in1 to out delay.
(3) Measure sel to out delay.

Is (1) a valid "propagation delay" for in0 and out pair?
In the logical perspective, the 0 to 1 transition of out is due to in1 and sel, not in0.
Thus, could we say (1) is not a propagation delay?

Now, let's consider the second case:
in1 and sel transit from 0 to 1 at the same time t.
However, in0 transits from 0 to 1 before time t, so that out transits from 0 to 40% of supply voltage.
And, again, we measure the three kinds of delays, we call them (4), (5) and (6):
(4) Measure in0 to out delay.
(5) Measure in1 to out delay.
(6) Measure sel to out delay.

The delay time of (5) shall be less than (2), however, is it a valid propagation delay?

If the measurement of propagation delay requires a causal relationship,
how do we measure the propagation delays of a black box circuit?
We might not know whether a transited input causes the output transition while multiple inputs are also transiting.
Does this mean when we measure a propagation delay of an input and output pair, we have to keep other inputs no transition?

(1), (2), (3), (4), (5), (6) are all propagation delays. Depending on circuitry in between, some of those delays might be the same, and may be different for 0 -> 1 and 1 -> 0 transitions (at the output, usually). In most cases, the longer of those will be the most interesting (and that might be used as overall propagation delay for the circuit).

With respect to voltage levels, that just depends on what you pick as reference points. Common are some % of supply voltage, for example for a 0 -> 1 transition you look at time it takes for output to reach 80% of Vcc (or whatever arbitrary level). Or guaranteed levels like 0.8V and 2.0V for TTL-compatible logic. Under specified operating conditions (temperature range, supply voltage), with output loaded in a pre-defined manner.

zxvc said:

If the measurement of propagation delay requires a causal relationship,
how do we measure the propagation delays of a black box circuit?

Click to expand...

I suppose one would first have to find out / define a relation between input A and out B, before you can speak of propagation delay from A to B. But just knowing that B changes as result of A changing, might be enough for that.

Hi RetroTechie,

For (1), the causal relationship between in0 and out is ambiguous.
It is difficult to say the 0 -> 1 transition of out is due to in0, in1 or sel.
Then, could we say (1) is valid? Or is the answer also ambiguous?

Another question is that:
does the causal relationship require the time of trigger point is less than the time of target point for calculating propagation delay?
For an similar example as the previous multiplexer:
At time t, sel = 0 and in1 = 1 have no transition; in0 has a 0 -> 1 transition; and then out has a 0 -> 1 transition.
At time t+1, sel = 0 and in1 = 1 have no transition; in0 has a 1 -> 0 transition; and then out has a 1 -> 0 transition.
Then, is this measurement a valid propagation delay?:
(7) trigger = in0's fall near time t+1, target = out's rise near time t.

zxvc said:

For (1), the causal relationship between in0 and out is ambiguous.

Click to expand...

No it's not - propagation time is time need for an input change to propagate to an output. Read: for any single value of propagation time, you're looking at 1 input, and 1 output. Changing multiple inputs at once & then looking at an output gives you a 'composite' (read: more or less meaningless) value.

Another question is that:
does the causal relationship require the time of trigger point is less than the time of target point for calculating propagation delay?

Click to expand...

If you mean: t(output change) is later than t(input change) - yes of course, output changes as result of input change so the output always changes at a later time than the input.

For an similar example as the previous multiplexer:
At time t, sel = 0 and in1 = 1 have no transition; in0 has a 0 -> 1 transition; and then out has a 0 -> 1 transition.
At time t+1, sel = 0 and in1 = 1 have no transition; in0 has a 1 -> 0 transition; and then out has a 1 -> 0 transition.
Then, is this measurement a valid propagation delay?:
(7) trigger = in0's fall near time t+1, target = out's rise near time t.

Click to expand...

No that's 2 propagation delays - one for the 0 -> 1 transition (in0 change -> output change), one for the 1 -> 0 transition (in0 change -> output change). Usually propagation delays are defined as 'min, typical, max' so you'd combine those 2 values into 1 in terms of average, and worst case.

Don't know what you meant with (7)... if this is still not clear, I suggest you have a look at some datasheets for simple logic (inverter, AND / OR gate, something like that) in CMOS 4000 or 7400 series. Such datasheets usually come with timing diagrams that show how propagation delay is measured, and under what conditions.

Hi RetroTechie,

After reading your explanation, I understand more.

(7) is a bad example. Please forget it.

Thanks for your explanation.