Re: [SI-LIST] : Re: Is feedback really EVIL?

Scott McMorrow ([email protected])
Thu, 14 May 1998 10:36:47 -0700

Feedback isn't evil ...

... only a bit chaotic in reflective wave systems.

Feedback can be a very good thing in well terminated transmission
line environments. However, then feedback is being used to stabilize
operating points. If this is the case, then a behavioral model will nicely
describe a well behaved feedback system ... within limits. It's those nasty
reflections that hurt feedback systems.

I once looked at a design of a pipelined sram where the output used feedback
to lower the drive current on the first reflection. THe design assumption with
that device was that the reflection came from the end of the line. ANd that they
could lower the drive strength to decrease power dissipation . In my case,
the sram sat on a leg of a bus with a star topology. The first reflection came
from the low impedance center of the star, causing the driver to lower it's
output drive strength .... arrrgh! ... when it should have been increasing it's
drive strength to compensate for the lower apparent impedance. Needless to
say, that sram vendor was disqualified due to poor output buffer design.

The problem, of course, is understanding why feedback is being used in the
first place. What are the design assumptions of the feedback driver? Does
your system match the design assumptions? If not then a complete Spice
stability analysis of the system is in order. Unfortunately, how many vendors
will give outside SI engineers full access to Spice models for their proprietary
feedback buffers? Not many in my estimation. I have trouble getting models
for PLD's and FPGA's from some vendors worried about theft of intellectual
property (i.e. buffer and process data).

Scott McMorrow
Principal Engineer

mailto:[email protected]