Thank you. I guess I incorrectly assumed V_ce of 2V meant, 2V high, 0V low on that waveform they show. Not 2.2V high 2V low. Seems a bit sneaky to me.
If you designed something that operated like that, by adjusting the forward current to limit I_c and keep the output transistor from going in to saturation to get the faster response time, I'm guessing that would be highly dependent on the CTR, and so would vary a ton part to part?
Somewhat. The other thing you get speed from is keeping it biased. The delay is basically due to B-E capacitance, and the B-E junction is the photosensitive part. That is, the equivalent circuit is like a photodiode in parallel with a BJT (B-E). When it's dark, B-E discharges towards zero, and when lit after being dark, it has to charge up to Vbe (even fairly low Vbe since we're talking relatively small Ic here), which adds a significant hard delay (i.e. no output change until after it charges up). If you keep it biased rather than completely off, or use a small-signal receiver, it can be a bit better.
By "small signal receiver", I mean, the opto's output voltage might be say 100 to 600mV, with an amplitude of 50-200mV. If you have an AC signal, i.e. the logic level is changing frequently, with some maximum high/low pulse rate guaranteed by line coding or nature of the data, then a peak-to-peak detector obtains the logic-high and logic-low levels and simply a divider halfway between, and comparator, recovers logic-level data.
But you get
much more speed by just using a better device, like 6N136 (or better yet SFH6345), 6N137 or other logic-type optos, or other digital isolators (usually monolithic capacitor- or transformer-coupled). The latter are available to 50Mbps and higher, and don't cost much more than optos do (indeed they may be less when considering the multichannel types).
Tim