Re: Req for Info: Differential Buffers

Cass:

You raise a good question. The IBIS format does not support documenting
the [Rising Waveform] and [Falling Waveform] tables using
differential loads, nor does it support any possible interaction in I-V
characteristics between outputs. However, I believe there are still
some approximate approaches that will be very accurate.

Some suggestions are given below. You may need to configure the test
circuit given by s2ibis and use the [iterate] command to get an
adjusted extraction. You also need to compare the simulation
results.

For V-T tables:

1. Use "equivalent single-ended load on each buffer connected
to the mid-level voltage. This does not guarentee that
the inverting buffer sees the same current as the non-inverting
buffer - as you have noted. Nor does it capture any non-inverting
and inverting buffer interaction. It may be possible to adjust this
approach using controlled current sources to inject the same
output current into the non-inverting node. However, this
seems equivalent to the second alternative.

2. Use a true differential resistance in the Spice extraction of
the [Rising Waveform] and [Falling Waveform] tables. However, put
the equivalent single ended loads in the IBIS file - as an approximation.
This captures the voltage waveforms under correct Spice simulation
conditions using the correct differential load. The fact that these
are documented based on single-ended loads is incorrect since the
mid-point voltage is usually not constant. However, the IBIS
simultations should generate the same non-inverting and inverting
voltage waveforms when the device is terminated with a true
differential load. (The accuracy of doing this may be simulator
dependent.) I recommend this approach.

For I-V tables:

1. Do the Spice I-V extractions with the non-inverting output connected
to the mid-point voltage. A refinement to this would be to connect the
non-inverting output to the equivalent single-ended load, as an approximation.
The I-V table may not capture exactly the inverting and non-inverting
interactions, but it should be a good approximation. I recommend this
approach.

2. If there is significant interaction on I-V characteristics between the
non-inverting and inverting outputs, then this may work: Configure the
Spice source such that tbe non-inverting output has the same current as
the non-inverting output for I-V extractions by using a controlled current
source. This might provide a more accurate representation of the I-V
characteristics.

Bob Ross
Mentor Graphics

Cass Russett wrote:
>
> The typical construct for converting Spice models to IBIS -- namely,
> Spice deck fixtures for producing V/I and V/T curves, do not appear
> directly relevant for converting differential buffers.
>
> For example, if the non-inverting transmit buffer output is connected to
> the S2IBIS2 fixture, how should the inverting output be connected?
> Please note that the manner in which the inverting output is connected
> does have an impact on the response of the non-inverting output.
>
> If the non-inverting output is ignored, then the results do not
> correlate to the actual buffer behavior because of the tightly matched
> current sourcing and sinking of the respective outputs. If the
> non-inverting output is connected to something, what is it for both the
> V/I and V/T fixtures?
>
> In short, what is the established procedure for differential buffers?
>
> Thanks in advance for any assistance,
> Cass Russett
Received on Mon Nov 29 16:39:29 1999