[snip]
Take an 'HC14, and DC sweep the input _slowly_ and observe I(VDD). You pay a price for that kind of hysteresis. The sort of thing that can kill a supply sucking only 3mA off-line.
See line above "so I added hysteresis"... which Larkin didn't like, because he didn't devise it. (I use that all the time on-chip as a noise blanker with hysteresis... think over it with a longer time constant, but no series R on the feedback cap, no ESD diodes internally... then you can see how it work.) My approach, while using more parts, has a robust pulse drive to the opto-couplers _at_ the zero crossing, within 26us, without any tweaking needed.
Larkin's Totem Pole is fine is you can tolerate the 120us offset (which is what you get when you run the simulation long enough to reach full equilibrium on the "tweak" cap. Keep in mind I ran the simulation with the Moto/ON-Semi model blessed by Cadence as accounting for all effects and strays. LTspice, on the other hand, has some model of unknown origin, which claims almost an order of magnitude higher coupling coefficient than the Moto/ON-Semi version.
I will grant you that my design philosophy differs from that of the conventional discrete designer. In an I/C design, it has to always work right out-of-the-box, no tweaking; and active devices are cheaper than resistors. On the other hand the discrete designer's cost is PER COMPONENT... so "tweaking" is part of their game.... and a "blue-wire" fix is no big deal.
[snip]Having been burned once ;-), a standard procedure in my design evaluation is to exercise power supply sequencing, and all sorts of set-up conditions, watching for gotchas and latch-ups.
For an integrator inside a chip, I'd have no compunction at all to adding a whole bunch of extras to ensure no wind-up; or to an OpAmp to prevent long recovery when banged against the rails.
Discrete guys trust data sheets, and use them to promulgate a "proof" of why it works, or not ;-) ...Jim Thompson