0
\$\begingroup\$

In Weste and Harris's CMOS VLSI Design, they write

In Section 10.2.5 we will see that flip-flops may experience hold-time failures if the system has too much clock skew, i.e., if one flip-flop triggers early and another triggers late because of variations in clock arrival times. In industrial designs, a great deal of effort is devoted to timing simulations to catch hold-time problems. When design time is more important (e.g., in class projects), hold-time problems can be avoided altogether by distributing a two-phase nonoverlapping clock. Figure 1.33 shows the flip-flop clocked with two nonoverlapping phases. As long as the phases never overlap, at least one latch will be opaque at any given time and hold-time problems cannot occur.

Figure 1.33 is given below, and it's simply a D flip-flop implemented in a master-slave configuration (and this time with nonoverlapping clock phases):

enter image description here

I am confused about two things:

(1) Is the point about hold-time violations with respect to the internal latches within the same flip flop, or to different flip-flops in the larger digital system? That is, is there an issue with the standard master-slave configuration as below, in that it might misfire in going CLK-to-Q if we use the standard single-clock implementation?

NB that in the image below the complemented QM signal (output of master latch) is an error (I believe) but that is immaterial to the question.

enter image description here

(2) I suspect the answer to my (1) above is that the comment was with regard to hold-time violations in different flip-flops and, if so, I do not understand how the nonoverlapping nature of the clock phases helps us avoid all possible hold time problems, as claimed. Let's consider two flip-flops, F1 and F2, built as in the first picture. Is the point that F1 only launches when \$\phi_1\$ goes high so that if we pick the non-overlap time \$t_{overlap}\$ long enough then we'll have no hold time violation with F2 receiving this launch? F2 needs its D2 input stable for some hold time around the falling edge of \$\phi_2\$ but I can't quite see past this. Would an answerer be able to explain, for instance, a situation in which these two-phase flip flops don't have an issue but the standard single-phase flip-flops do?

\$\endgroup\$
2
  • \$\begingroup\$ 'completely obviate' is too strong. All non-overlapping clock phases do is make it far easier to design it adequately, even if you're a bit careless of min/max specifications. If you want it to go wrong with non-overlapping clocks, then you can design it that way. \$\endgroup\$
    – Neil_UK
    Dec 14, 2023 at 19:06
  • \$\begingroup\$ Understood, I think. If you get the chance would you be able to demonstrate a fixed, given situation where there is a violation when we don’t have nonoverlapping clocks which is removed when we add some amount of overlap? I think this is the crux of what you are saying overlap does? @Neil_UK \$\endgroup\$
    – EE18
    Dec 14, 2023 at 19:16

1 Answer 1

0
\$\begingroup\$

The claim is wrong. This technique is more generally known as a "synchronizer", and if hold times are violated, can still have errors due to metastability.

If the allowable error rate is very low, it will be necessary to have a synchronizer with more than two stages.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.