Metastability in Flip-Flops
High-Speed Digital Design Online Newsletter: Vol. 3 Issue 15
In your book you talk about metastability in flip- flops, and you suggest that two flip-flops are a good way to avoid it. I'd like to know what happens if you have two flip-flops in series, both using the same clock, and the first one goes metastable. Does the second one clock out whatever the first one passes, even if it's junk? Or does it default to a LOW?
Thanks for your interest in High-Speed Digital Design.
On a good flip-flop, when you meet the setup and hold window, it always gives you a solid output within the guaranteed maximum clk-Q delay.
The essence of metastability is this: when the data input violates the setup and hold window, if it transitions close enough to the exact sampling moment within that window, the flip-flop may take LONGER than the normal clk-Q delay to make up its mind. As the flip-flop is deciding what to do it may stay LOW, and then later pop HIGH (or vice- versa), or it may jump one way, and then come back. Some very old parts would oscillate for a while before settling down (I haven't seen that in twenty years).
The extra delay may be ten or twenty times longer than the normal clk-Q delay. This extra time is called the metastable resolution time.
OK, now let's treat your question. Suppose we label the two flip-flops A and B. When an input transition hits within the metastable sampling window of the A, it causes an inordinately long delay in the output of A.
When this delayed output arrives at B, everything is usually alright UNLESS the delayed signal transition from A happens to hit RIGHT ON TOP OF the metastable sampling window for B. In that case B produces a long clk-Q delay as well, which could incapacitate the succeeding logic.
As you might imagine the probability of this scenario happening is extraordinarily small. In a practical circuit, cascading two flip-flops practically squares the probability of failure. Adding the second flip-flop can easily turn a circuit that used to fail one time out of a billion (with only one sampling flip-flop) into a circuit that fails only one time out of a billion-billion (with two sampling flip flops). It's a very powerful technique.
Dr. Howard Johnson