Re: [SI-LIST] : Driver Strength

D. C. Sessions (dc.sessions@tempe.vlsi.com)
Thu, 05 Mar 1998 13:59:02 -0700

Lfresearch wrote:

> the discussion on driver strength has prompted me to make an observation seen
> during EMC testing. One of the immunity tests fires rapid ( ns timeframe )
> voltage transients onto all the wiring interfaces. I have observed that when
> digital circuits respond to the transients, it is the devices with the weakest
> pull-up or pull-down that go first. By reversing the transient polarity,
> significant shifts in the level at which the circuit responds may be observed.
> Also, if the pulling circuit is not located close to the input pin, the trace
> isolates its effect somewhat.

Just to make sure: are you talking about logic upset or hard failure?
Most CMOS outputs have the same number of drains connected to the
pad regardless, but adjust strength by tieing off gates to keep the
devices OFF all the time. That way, ESD transients still have the
full device complement to share the energy.

For logic upsets, OTOH, the problem is that current into a
high-impedance
output will produce higher voltage responses than current into a
low-impedance
output (Duh!)

> The problem is that by using drivers with lower drive capability we end up
> with designs that could have immunity problems. However, devices with higher
> drive capability tend to be the cause of radiated emissions problems.
>
> I guess there needs to be middle ground ( from the EMC engineers point of view
> ), so that we can maximise immunity and minimise emissions. Anyone want to
> take a shot at a set of guidelines?

Well, the EMC problem isn't so much driver strength as Ldi/dt noise.
This can
be addressed by:
1) Reducing max I independently of dt and L (the weak driver approach)
2) Reducing L independently of dt and max I (packaging solutions)
3) Increasing dt independently of max I and L (custom drivers)
4) Isolating the supply paths for high-speed I/O from victim lines
(floorplanning)

As is almost always the case in EMI/susceptibility engineering a
combination approach works best. Drivers with no more static drive
than necessary should be used with predrivers as slow as the application
will stand, with lots of low-inductance supply connections split between
noise generating and noise susceptible I/O groups.

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