Bob,
Yes, this will require something in the spec. One implementation of DDR4
with AMI equalization at the Tx and Rx assumes an impulse response that is
created from the derivative of a step response, and that the “0V” of an eye
is VrefDQ. The EDA tool can calculate this offset and remember it. Others
may not agree with this implementation, so the standard should clarify what
the model maker will expect the EDA tool will generate for an impulse
response of the channel (e.g. start and end at 0.0V), and that the output of
Tx and Rx GetWave will be a waveform who’s 0.0V corresponds to the DC VrefDQ
that the EDA tool can remember when generating the impulse response from the
step response.
Bottom line is this needs to be clarified in the specification.
Walter
From: Bob Miller [mailto:bob.miller@xxxxxxxxxxxxx]
Sent: Monday, January 25, 2016 3:43 PM
To: Walter Katz <wkatz@xxxxxxxxxx>
Cc: jyan@xxxxxxxxxx; IBIS-ATM <ibis-macro@xxxxxxxxxxxxx>
Subject: Re: [ibis-macro] There are no changes to the IBIS Parser to support
AMI models in single ended buffers
I think there is an annoying issue in single-ended IBIS-AMI simulations in
that the impulse fed into the algorithmic model in AMI_Init has no DC
information (it is functionally the derivative of the step response, which
does have DC information.) In a purely differential receiver we implicitly
assume that both the Tx differential reference and the Rx differential trip
point is 0V and interpret the impulse in such a way that a step response or
data stream generated from this impulse is symmetrical about 0.
However, in a "single-ended" system, the Tx and Rx references are not 0V
(and may be different from each other) so the Rx cannot merely process the
impulse through the Rx front end; it probably needs to perform the
equivalent of integrating the impulse into a step/pulse/data waveform (which
if you recall calculus101 requires the restoration of the DC "constant").
This DC constant probably has to come from the Tx.
In AMI_Getwave, the Tx can supply the "single-ended" output waveform with
the DC component preserved; as long as the EDA platform properly preserves
this through the channel convolution all is well (I hope!).
As in the IBIS Open Forum, I may be blathering nonsense; feel free to advise
me of it (again), Walter...
Regards,
Bob
On Mon, Jan 25, 2016 at 12:44 PM, Walter Katz <wkatz@xxxxxxxxxx
<mailto:wkatz@xxxxxxxxxx> > wrote:
John,
I verified that the IBIS 6.0 parser generate no warnings or errors when an
[Algorithmic Section] (interface to IBIS AMI model) is added to a singled
ended Input, Output or I/O buffer.
On page 171 IBIS 6.1 says:
“The [Algorithmic Model] always processes a single waveform regardless
whether the model is single-ended or differential. When the model is
differential, the waveform passed to the [Algorithmic Model] must be a
difference waveform.”
The bottom line is that this is nothing in the IBIS specification that
limits AMI modeling to Inputs, Outputs, I/O, single ended or differential.
There is also no constraint in IBIS that the Impulse Response must start at
zero, and end at zero. A number of tools will generate an impulse response
by differentiating a step response. This will lose any DC offset in the
data, but this is a limitation of the implementation of AMI on that tool.
IBIS does say that AMI modeling is used for SerDes designs, but does not
limit it as such. It certainly would not hurt to add some language that AMI
modeling can be used for other than SerDes.
Walter
Walter Katz
wkatz@xxxxxxxxxx <mailto:wkatz@xxxxxxxxxx>
Phone 303.449-2308 <tel:303.449-2308>
Mobile 303.335-6156 <tel:303.335-6156>
