RE: S2ibis2 -- inputs with pull-up resistors, bi-directionals, Rising Waveforms, Falling Waveforms, etc.

Roland:

You have raised some good questions. I have put some responses to your
questions in your text.

Best Regards,
Bob Ross
Interconnectix

> To: "'ibis-users@vhdl.org'" <ibis-users@vhdl.org>,
> "'ibis-info@vhdl.org'"
> <ibis-info@vhdl.org>
> Subject: RE: S2ibis2 -- inputs with pull-up resistors, bi-directionals, Rising Waveforms, Falling Waveforms, etc.
> Date: Wed, 28 May 1997 17:23:54 -0700

> Hello,

> I need some help clarifying the creation of models in IBIS.

> 1. Inputs with pull-up resistors?
> Are these treated in any special way? (ie: different than that of a
> normal input pin?)

Pullup resistors would appear in the Clamps. They are not separated
out. So there is probable that they will be double counted.
For Pullups, the contribution in the [Gnd Clamp] table should be
should be subtracted. The [Power Clamp] table should be extended to
from 0 to Vcc (and possibly linearly extrapolated to 2Vcc) to provide
the separate Pullup resistor component.

> 2. Bi-directionals?
> I have a pad structure that has the following nodes:

> Pad (the output)
> Oen (ouput enable)
> In (feeds the output when output enable is asserted, and Pad functions
> as an output)
> Cin (feeds the core when output enable is deasserted, and Pad
> functions as an input)

> What is the proper way of setting this up in s2ibis2?

I would need to see the circuit (and documentation, if any) to
understand better how to set this up. My first idea would be
to tie Cin to any load such as 50 ohms to ground. I would use In
as the Input node and model the Pad as an I/O.

> 3. Rising and Falling Waveforms.
> I understand that these are supposed to describe the rising and
> falling edge waveforms of a driver, yet they are not required. There
> is also a note in the IBISv2.1 specification that states: "...A
> maximum of 100 waveform tables are allowed per model. Note that for
> backwards compatibility, the existing [Ramp] keyword is still
> required. The data in the waveform table is taken with the effects of
> the C_comp parameter included..."

> >From this, I was wondering, what would having 100 waveform tables be
> used for?

I doubt if 100 waveforms is useful, but there was no reason to set a lower
limit which may conflict with an unanticipated need. In most cases one
or two waveforms is sufficient, and some others might be provided to document
the response for a particular set of loads. Several simulators will use at
most only two of the waveforms that are provided.

Further, are the Rising and Falling Waveforms somehow a -
> replacement - for the [Ramp] keyword?

The Waveforms replace the [Ramp] keyword. Sometimes waveform data is
bad or several waveforms are inconsistent. The [Ramp] keyword is
provided so that the user can get still use the model and get a response.

Are the Rising and Falling
> waveforms usually generated in the lab at the bench, or can they be
> simulated? When are they required, if at all?

Waveforms are never required. They can be obtained from simulation
or from the lab. There are some situations where the waveform
response gives a better model because it allows getting data that
is closer to the actual loading conditions of the buffer. GTL is an example
where the loading conditions are, for example 25 ohms to 1.2 V, even though
the Open_drain buffer is powered by a 3.3 V supply - Implying that the
Ramp load is connected to 3.3 V.

> 4. Analog and Digital VDD's and VSS's.
> In the spice netlists of the files that I am dealing with, I often
> multiple VDD's, for example, VDDac, and VDDdc, which are used to
> power up separately the I/O ring and the core. Has anyone else out
> there had to deal with multiple VDD's like this? (I'm not talking
> about multiple VDD's such as 3.3 and 5 V, such that can be solved
> using pin mapping)

The IBIS model does not go into this detail. I would apply the
same voltage on all Vdd's to derive the IBIS model.

> 5. Stimulus input voltage risetime and falltimes
> How are these parameters used? And for what curves are they used
> for?

The default is 0.1ns. For very fast devices, even this may be too
slow. In general, I have seen negligible variation in the
output response due to input rise and fall times. However,
someone may still want to generate a model response based on an
anticipated or data book specified input response time of 2 or 3 ns,
for example.

> Clarification of any of the above questions (1-5) would be greatly
> appreciated,
> Thanx!

> Best Regards,
> Roland

> ***********************************************************************
> **
> *Roland Chang | email: roland@sfu.ca
> *
> *Simon Fraser University | OR
> *
> *School of Engineering Science | chang@pmc-sierra.bc.ca
> *
> *3rd year Computer Engineering Option |
> *
> ***********************************************************************
> **
> ~

 
Received on Thu May 29 15:12:40 1997