Well, first, metastable states can last for more than one clock cycle. There is no absolute limit to the duration of the metastable state. Granted, it is very unlikely to last more than one clock cycle, BUT, if this hazard is happening on each input signal transition (because the propagation delays are stable to within picoseconds), and there are 1 million or more input signal transitions per second, you could easily have a failure every few days even if the probability of a single event is astronomically small. Secondly, if you have an input which is looking for an edge on the input signal by comparing the signal delayed by 2 clock cycles versus 3 clock cycles (i.e., comparing the input and output of a third FF after the synchronizer), then this circuit could see multiple edges where only one actually happened. On Fri, Jul 1, 2011 at 7:47 AM, Dave Tweed wrote: > Sean Breheny wrote: >> The serious problems are things like internal signals within the destina= tion >> device (which are dependent on the input signal) staying in an indetermi= nate >> state for several clock cycles. > > No. Like I said, as long as the input signals stay in the same state for = at > least two clocks at a time, no metastable state can last for more than on= e > clock cycle -- if the signal failed setup/hold on one clock edge, it will > meet it on the next. > > In the end, the only ambiguity as far as the receiving logic is concerned > is whether the transition happened on CPU clock edge N or clock edge N+1. > Most protocols don't care about this kind of jitter, especially synchrono= us > protocols like SPI or I2C, which use a data clock which is a fraction of = the > CPU clock. > > As far as the PCB layout goes, you already mentioned using a clock > distribution chip -- as long as each output goes to just one CPU, and the > lines are terminated properly, I don't see any problems there. > > -- Dave Tweed > -- > http://www.piclist.com PIC/SX FAQ & list archive > View/change your membership options at > http://mailman.mit.edu/mailman/listinfo/piclist > --=20 http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist .