I'm quite sure that no in this discussion is paying much attention to me but here goes anyway.
The schematic of the TTL version of the 74x86 shows that the basic function of the XOR is performed by a cross coupled pare of transistors. But the cross coupling does not create the regenerative feedback that creates the storage mechanism for a flip-flop. The bases would have to be cross coupled to the collectors to be positive or regenerative feedback. Metastability is caused by both transistors functioning in the linear region such that there is nothing to "regenerate" to force the pair to flip one way or another. The circuit will drift until there is enough of an offset to cause the regeneration to snap a decision.
Faster flip-flops are made, in part, by more closely matching the cross coupled pair - which causes them to take more time to recover from the meta stable state. Flip-flops also have an internal time-constant. This time constant is the loop propogation delay. Flip-flops don't recover from a metastable condition randomly. They recover at some multiple of their internal time constant. Most recoveries occure at 1T, with successively fewer recoveries at 2T, 3T, 4T, ...
I have seen the Fast(for TTL) JK flip-flops take longer that 10T to recover. It took a storage scope for me to see that.
I have mentioned runting before. Runting is defined as a short pulse, around 1/2T (half T) that puts both Q and notQ at the same logic state. If the runt pulse is on Q, Q switches to the same state as notQ and then back to its previous state.
Runting is much more likely to occur that metastability. Both of which are caused by the D or JK inputs violating the setup and hold times of the clock. That is what the setup and hold times are - the time that the D or JK inputs can be changed without causing runting or metastability.
It is rather interesting to me that flip-flops are not used in high speed synchronis CMOS designs.
They are actually master-slave latches, not flip-flops.
To flip a flip-flop, the regenerative feedback needs to be overcome. When the feedback is nearly overcome, runting occurs. When the feedback is just neutralized but not overcome, metastability occurs.
In a latch, the feedback is disabled casing the the latch to accept the state of the input. Once the state is stabilized, the feedback is restored and a regenerative state is maintained. The latch input is disconnected when the feedback path is reconnected.
Doug
