RE: output impedance of a driver

From: Todd Westerhoff <twester@hhnetwk.com>
Date: Mon Apr 09 2001 - 14:11:14 PDT

Jason,

I prefer to measure (delta V)/(delta I) from the linear portion of the
transistor curves. I just eyeball the curves and pick points from the
largest portion of the curve I can perceive as "linear" (squinting is
optional). There are potential pitfalls as Chris points out, but for common
application the technique seems to work just fine. It's just a starting
point anyway, as you should use simulation to test your assumptions.

As others have pointed out, be sure the check the slopes for both the pullup
and pulldown devices, as you need to balance the needs of both edges.
Should you find yourself dealing with one transistor significantly stronger
than the other, parallel termination to either power or ground may be in
order, as was discussed last week.

Todd.

-----Original Message-----
From: Chris Rokusek [mailto:crokusek@innoveda.com]
Sent: Monday, April 09, 2001 3:17 PM
To: ibis-users@eda.org
Subject: RE: output impedance of a driver

Jason,

The curves shown in the VI viewer are identical (in most cases) to the
tables in the IBIS file. You can use the output XTK file directly just as
you did with the IBIS file using the following mapping.

        IBIS XTK
        ------ -------
        Pullup High
        Pulldown Low
       POWER_Clamp Shunt
       GND_Clamp Shunt_Neg

However...

It is somewhat dangerous to provide only a single static impedance value as
others have pointed out due to the large variance over the typical operating
range, current state, process variation, and dynamic switching nature. But
if you are trying to perform a first-order approximation then you must first
decide which voltage range(s) and which state(s) is of the most importance
to your application because Z will vary with both state and voltage range.

In some sense you should ask your colleague what he will use the Z value for
because that will determine which Z value you give him.

For instance if he wants to know what voltage will propagate down the TLine
during the initial rising edge then you might approximate the Z by
connecting two points on the HIGH VI curve in the middle of the operating
range. This is poor approximation for at least three reasons: 1) The VI
curve may be non-linear, 2) The transition from off to on may be non-linear,
3) The opposing state (LOW) may still be driving simultaneously and turns
off non-linearly.

Another example is that your colleague may want to use the driver Z to
insure that reflections are absorbed. In this case you would need to figure
out in which voltage range/state the reflections will occur. If they occur
above/below the operating range of the device then the shunt curves (diodes)
will come into play...

Another methodology would be to create a linear driver (one with VI curves
of constant slope) and that is tweaked until its response matches the
response of the driver you are trying to approximate.

My advice is to instead of providing an impedance, give him waveforms that
cover all of the driving/loading cases (TLine/Receivers/Terminators/Process)
you are expecting to see.

Good luck,

Chris Rokusek
Innoveda

> -----Original Message-----
> From: Jason Leung [mailto:jleung@cid.alcatel.com]
> Sent: Monday, April 09, 2001 11:17 AM
> To: Muranyi, Arpad
> Cc: ibis-users@eda.org
> Subject: Re: output impedance of a driver
>
>
> Hi Muranyi, Arpad:
> Thanks for your advice!
> one more question I would like to ask you: because I have been asked from
> my colleague to use
> XTK/XNS to find the output impedance of the driver, I know that in the
> spawn -> VI viewer ,
> we can import the corresponding .wvi curve to observe the waveform.
> Among all those waveform, you can just pick on one waveform, and then do a
>
> delta x and delta y function, then you will be able to find the impedance.
>
> However, there are so many curves I have to choose(shunt and high, shunt
> and low and many
> many more) , do u know which one I have to choose for an accurate
> approximation of the
> output impedance of the driver .
> The whole point here is , since I know I can find the output impedance of
> the driver in the IBIS model
> I also wanted to try and find the output impedance of the driver in the
> XNS (QUAD) model ,so that
> I am able to compare what we have before the simulation and after the
> simulation( i.e. how the XTK/XNS
> interpret the output impedance of that driver ).
>
> thanks alot
> Jason Leung
>
>
> > Jason,
> >
> > The DC impedance can be found by dividing V/I on the IV curve of
> > the buffer. You will find that you get a different value at
> > practically any of the voltages due to the non linear shape of
> > the IV curve. Plus, the process variation of a chip can give
> > you a variation on the order of 2-3x. So finding the impedance
> > and matching the T-line to it is like shooting at a moving target.
> >
> > The reason I said DC impedance above is because the capacitive
> > components of the buffer will also alter the impedance of the
> > buffer a higher frequencies.
> >
> > Arpad Muranyi
> > Intel Corporation
> > ================================================================
> >
> > -----Original Message-----
> > From: Jason Leung [mailto:jleung@cid.alcatel.com]
> > Sent: Monday, April 09, 2001 6:14 AM
> > To: ibis-users@eda.org
> > Subject: Re: output impedance of a driver
> >
> > Hi all:
> > Morning!
> > I am trying to obtain the output impedance of a driver, so that I can
> > match the driver with a transmission line and proper termination.
> > The way that I have tried is I used the IBIS model and then translate
> > the model into QUAD model, so that I can used the XNS to observe their
> > rising and falling waveform. Moreover I am trying to find its output
> > impedance in XNS.
> > My question is : Does anyone ever try finding the output impedance of a
> > driver? AND how did you achieve that?
> >
> > Thanks very much for your comment!
> > Jason leung
>

 
Received on Mon Apr 9 14:11:44 2001

This archive was generated by hypermail 2.1.8 : Fri Jun 03 2011 - 09:53:47 PDT