Re: BIRD61 - Input Model

Fred:

You raise some good points and have offered an alternative.

Working directly with tables is the only approach I know so far.
As you indicate, table lookup method has its potential problems
for arbitrary inputs.

While different input modeling approaches might be implemented
(and defended) by different vendors, we are seeking at least
one approach that is defendable. Without knowing any approach,
we do not know what minimum set of data is necessary.

You indicate that simulation model can be developed that would
respond to arbitrary inputs. The set of data for this is the
set of Vout(t) tables for a range of Input ramps with different
slopes. Can you indicate the minimal set of data needed to
implement this approach?

  How many Vout(t) tables do you need? Will minimum and maximum
  slope tables be sufficient?

  Do you need a range of Vout(t) tables with different Start_point and
  End_point values to capture overdrive/underdrive effects?
  What is this set?

Also I have some general question on Input models:

  Do we need Vout(t) to be extracted under defined load conditions?

  Does a Vmeas need to be specified as the point at which the delay
    is measured? If Vmeas is internal, does it need to be given?

  Since the Input model transfer may be impacted by overdrive/underdrive
    conditions, does the input timing reference point for arbitrary
    input slopes or waveforms need to be specified as Vth defined
    elsewhere?

Bob

Fred Balistreri wrote:
>
...
>
> Bob what the current Bird does is give information of delay versus
> slope. However as you mentioned slope is a linear ramp. Left as it is
> a vendor would have to determine slope, presumably during simulation
> in order to choose the right delay. However as you know during SI
> simulation the slope is anything but linear. You therefore are opening
> a can of worms. No doubt each vendor's answer could be different since
> the algorithm to determine slope (what ever that means) could be
> different. Now for the killer. With non-linear slope you don't know who
> is right, and with the data given there is no way to determine it. I
> actually don't care that each vendor is different. Yeah after all that's
> how we distingish ourselves, right? I don't know about you but one way
> I win business is to prove my answer is better than yours. But this
> Bird does not give me enough information to make a better model or
> even a determination of.
>
> I've said this before but I'll repeat it again. More information could
> be achieved by giving the devices voltage transfer curve(s). That's
> Vout as a function of Vin. This would be tables for rising and falling
> waveforms under different input slope test conditions. This type of
> information gives the vendor precise information (much like I/V curves)
> about the devices characteristics without revealing any structural
> or equations at all. Since Vil, Vih are known the delay numbers can
> be easily derived from simulation. Not only that the model can be made
> more dynamic and would better measure delay as a function of non-linear
> slope. In this way it's straight forward. The vendor does not need to
> determine slope and no equations (guess work) is needed.
>
> Anyway my two cents (what the hell yeah!) as Canadians say.
>
> Best Regards,
> --
> Fred Balistreri
> fred@apsimtech.com
>
> http://www.apsimtech.com
Received on Thu Sep 16 10:44:54 1999