Re: Forwarded: waveform testloads

Hi Sacha,

Here is ONE simulator's (Quad Design/XTK) response.

> When generating for instance [rising waveform] information I currently
> use a non-reactive load at high and low voltage - is that valid?

Yes. See our recommendation for test waveforms below.

> These waveforms are only correct at one point each, the simulator
> will have to interpolate somehow. So it is quite clear that the
> results may differ from simulator to simulator, but to hold the
> differences to a minimum it was helpfull if one would know how
> interpolation is done, are these algorithms known?

We are not planning to provide all our internal algorithms on the
reflector at this time. They differ among simulation vendors.
All the simulation vendor's collective customers must make the choice
that best fits their accuracy requirements.

> I followed Dileep Divekar's argumentation that an expected non-linear
> load dependence cannot be modeled uniquely with only one test load.
> If that is still true can I provide waveforms with different loads
> to give some interpolation points of load dependence of the output?
> (will simulators take this into account?)

We partially disagree with this argument. We claim that the recommended
test loads below will provide results meeting our customer's accuracy
requirements (our customers are very picky!).

Additional waveforms could be provided as a QA test, however they
are not presently used directly in the computation.

> I wonder how simulators derive the output under arbitrary load con-
> ditions if only one [rising/falling waveform] table is contained by the
> model. Do they perhaps use parametric output characteristics and
> "scale" them with the given waveform? - If so, how to deal with
> bipolar devices that surely have a different load dependence?

By default, we would like to see 2 waveforms for a non-single-ended device
(e.g. non- open drain) devices. Without two, only the risetime/falltime
is used rather than any waveform(s).

Bipolar devices DO require a differnt algorithm.

> In the december 12th 96 meeting minutes it is noted that Mr. Rokusek
> is collecting simulator vendors's suggestions concerning test loads.
> Are the results available? (I didn`t find any, sorry if I overlooked
> it)

My email is crokusek@qdt.com, please send any test load recommendations
to me or the reflector for discussion, I haven't received any up to
this point.

> I do not want to start the diskussion again (though is seems to be
> on, anyway) but just want to get as much information as possible to
> improve the models.
> Thank you very much for your support, I'm hopefully looking forward
> to reply

The "improvement of models" should be the ultimate goal, it should be
possible to make/keep everyone happy: model makers, simulation vendors,
and most importantly our mutual customers.

------------------------------------------------------------------------
Recommended waveforms:

     o Single-ended Devices: 1 rising, 1 falling

           If the driver requires a pull-up or pull-down device to function,
           the termination resistor and supply used for the design should
           be used to terminate the driver for this test.

     o CMOS: 2 rising, 2 falling
     
           50 ohms to a terminating voltage equal to the drivers negative
           supply voltage.

           50 ohms to a terminating voltage equal to the drivers positive
           supply voltage.

     o TTL: 2 rising, 2 falling

           500 ohms to a terminating voltage equal to the drivers negative
           supply voltage.

           7 ohms to a terminating voltage equal to the drivers positive
           supply voltage.
           

In terms of IBIS we are currently accepting the following...

       - C_fixture == 0.0pF
       - L_fixture == 0.0nH
       - V_fixture - (VCC or GND) <= 0.7V
       - 3.0 < R_fixture < 30.0 (TTL to VCC)
       - 200.0 < R_fixture < 1000.0 (TTL to GND)
       - 25.0 < R_fixture < 200.0 (CMOS)
      
     Of all waveforms meeting the above criteria, the two (VCC and GND)
     waveforms with R_fixture values closest to the recommended R_fixture
     values are chosen. For open-ended devices, only one (VCC or GND or
     Vterm) waveform is needed.

     If the waveforms are unusable, ramp data (edge rate only) is used.

     As stated in the IBIS V2.1 specification, if more than one rising
     or falling edge waveform table is present, then the data in each
     of the respective tables must be time correlated. In other
     words, the rising (falling) edge data in each of the rising
     (falling) edge waveform tables must be entered with respect to a
     common reference point on the input stimulus waveform.

---------------------------------------------------------------------------

The above is not GOLDEN or set in stone--only ONE simulator's present
opinion.

Another specific simulation vendor may want e.g. reactively loaded
waveforms, if so then the ibis forum can agree to a compilation of all
waveform requirements.

Just trying to get the ball rolling...

Best Regards,

Chris Rokusek
Quad Design/Viewlogic

(P.S. I don't know the answer to your Ccomp question.)
 
Received on Fri Feb 7 10:15:50 1997