The ST doesn't have a different chip, it has timer interrupt + fast cpu. I am yet to see an AY clone with real different feature.
@grauw, does the demo go with reloading the frequency register?
One doesnt need the usual interrupt rate to make digi, one just needs to reload frequency register to make different length arms.
The cpu load is much less than digi.
The nice thing is that the cpu can even be a bit late on this and jitter and it doesn't matter. The tone is perfect.
Because the bitflip is decided by AY counter / comparator (the frequency register).
The AY started counting from 0 and when you tell the new comparator value a bit later, it doesn't matter.
When timer granularity is too little, the wait time should be rather rounded up than down.
Poking to early means before AY has gotten to the flip, messes the result.
Still in the long run the timer interrupt must be exactly same hz.
When timer granularity is little, every soandso time there is some hiccup.
When starting a tone, make phase reset. When tone is short enough, no hiccup.
Well I again wonder whether 9938 sends interrupts with opened borders.
Then MSX2 could sound like the ST.
Another idea: the game for every of its loop makes a cycle estimation. With that simulate a timer interrupt.
Again, this method can stand cpu jitter and still sound clean.
Wait... it should take no more cpu to not use two arms, but say 4 different arms.
Then AY becomes synthesizer 
