LMH6702: what is a good video buffer op amp with less than 2 ns propogation delay?

Well Tommy, that LMH6702 is a pretty good choice for an uncommitted current feedback amplifier, an SO-8 package by itself has 0.4nsec delay, so use the SOT23 for sure - another option with internal gain set resistors would be the OPA693. I am guessing you want a gain of 2 in the amp stage, 

Good note, but the OPA693 is gain of 2 or 1,   I need variable gain.   The OPA695 looks like a good alternative in the same family though it does not explicitly state propagation delay.   Rise time is 800 ps.  I am guessing prop delay at about 900 to 1100 ps, of course depending on gain loading etc...   Does this sound about right?

Thank you for the feedback.

Tommy

Yes Tommy, the OPA695 is another good choice, I would set up the sim and run a linear frequency sweep to 1GHz - the slope of the phase is the die prop delay, then add something for the package an layout. 

Not sure what I was thinking, but you can also just run a group delay sim to get the prop delay directly for the die, 

Here that is set up for a gain of 2 OPA695

The plot isn't scaling well, but you can put markers on it to read off the delay, add about 0.3nsec to this for the SOT23-6 package delay, 

This is showing <1nsec total which is of course small signal, if you push into slew limiting it will be longer of course, 

Here I forced to axis to zoom in, and added the LMH6702 prop delay to this, should be a little shorter in sim it seems, 

Just to take this a little further, here is a +/-0.5V input at 0.1nsec input edge rate at 100Mhz square wave, 

A quick estimate of the outputs peak dV/dT will be 2.85*Vstep*F-3dB (an AC sim here showed about 780Mhz F-3dB) so this suggests 4.45V/nsec peak dV/dT on an output 2V 2nd order step response (pretty much right at the edge of slew limiting)

Anyway the sim matches the group delay with 0.59nsec delay at the zero crossing. Again, I am thinking this is the die as the models are meant to match full designer Cadence sims - in those days, I am not thinking a full or good package model was included in the design sims (much better now). I had tried to measure the package delay way back in the Comlinear days for my filter design work, I recall 0.4nsec for an SO-8 but had not continued on to the SOT23-6 as that was 15years in the future when I was doing that original effort. 

And, since I have these files set up now in TINA, what you might want to do is overcomp the LMH6702 by scaling the Rf up a bit to more like a Bessel response with slightly lower F-3dB to make sure a 2V output 2nd order step does not exceed its available slew rate, 

Here is a possible design point, 

That 540MHz F-3dB would ask for a peak output dV/dT of 2.85*2V*540Mhz = 3000V/usec. Right at the edge of its specified 3100V/usece. 

Here is the resulting step response delay, 0.48nsec in sim with less overshoot, 

Incidentally, if your source can drive the input load, these parts will do slightly better on prop delay and slew rate if you can run inverting.