BIRD 42.3

As long ago promised, here's a talking case for BIRD42.3
For simplicity, I've restricted the case to a very simple
falling-edge behavioral model. The driver in question
is a 50-ohm switched resistor, with the following
time/conductance tables. The pulldown driver is common,
but there are two different pullup responses: All
listed conductances are piecewise linear.

  Time | Pulldown | Pullup #1 | Pullup #2 |
        | Conductance | Conductance | Conductance |
--------+-------------+-------------+-------------+
   0 ps | 0 mS | 20 mS | 20 mS |
 500 ps | 0 mS | 20 mS | 20 mS |
1000 ps | 0 mS | 10 mS | 20 mS |
1500 ps | 0 mS | 0 mS | 0 mS |
2000 ps | 10 mS | 0 mS | 0 mS |
2500 ps | 20 mS | 0 mS | 0 mS |
3000 ps | 20 mS | 0 mS | 0 mS |

As is typical for real-world drivers, these are
break-before-make.

*IF* the load had time to settle completely to zero
current before the next edge, there would be no
difference between them; however, in real high-speed
switching even unterminated lines take too long to
settle and thus the pullup is still sourcing current
when this schedule starts. Obviously, a terminated
line is even worse; either way the turnoff of the
pullup has important consequences.

IMNSHO, this effect is far more important than the
(extremely rare) case of major crowbar current,
and any solution to this will necessarily also take
care of those few instances where crowbar current
is actually significant.

-- 
D. C. Sessions
dc.sessions@tempe.vlsi.com