> >what is the best way to build the clock tree ?
>
> It is best to be consistent (equal loading on all clock lines),
> and one load per line usually works best.
>
> >I connect the same clock to few consumers which are very close ?
>
> I have seen a case where a "T" junction to two loads (one trace
> to a point between the loads, then equal very short stubs to the
> loads) caused a bad glitch or plateau, exactly at the threshold
> voltage where you can't afford it.
>
The following can help avoid a shelf or non-monotonicity:
1. If the load capacitances on the stubs are different, the stub with less loading can be made longer
to compensate. In other words, the stubs should have the same electrical length.
2. If the stubs have an impedance of Zo, the main line should be Zo/2 for a "T" junction.
3. The impedance of the driver should be matched to the line with a series resistor.
> IN THIS CASE it was actually better to route first to one load,
> then from there with a short trace to the second. Except that
> the resulting waveforms had different rise times, they actually
> had little skew this way. The simulated waveforms both crossed
> threshold almost simultaneously, and more importantly, were
> glitch-free.
>
> Other cases I've tried, with different ICs, behaved entirely
> differently!
>
> It is important to get good SPICE models and simulate it.
> Vary the clock frequency too. Your clock tree may work
> great at one frequency but fail at another.
>
> Depending on your clock drivers, lots of series terminations
> or clamp diodes may be needed.
>
> Regards,
> Andy Ingraham
>
> --------------------------------------------------
>
> Part 1.2 Type: application/ms-tnef
> Encoding: base64
**** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****