[ibis-macro] Re: Eye mask definition for IBIS-AMI
- From: "Muranyi, Arpad" <Arpad_Muranyi@xxxxxxxxxx>
- To: "ibis-macro@xxxxxxxxxxxxx" <ibis-macro@xxxxxxxxxxxxx>
- Date: Mon, 19 Mar 2012 18:56:07 +0000
Walter,
That's not too hard to do. We just need to add a "z-axis" number
to the eye mask definition which would tell the tool at what BER
level on the 3D eye density or BER plot the eye mask is supposed
to be applied at.
I tend to disagree with this statement: "the BER is not a function of the
model but a function of the BER requirements for a specific channel in a
specific
application", because the BER we report after a channel analysis is
with respect to a logic threshold level. To illustrate this, I
dug out an excellent presentation on this topic:
http://www.google.com/url?sa=t&rct=j&q=peak%20distortion%20analysis%20osu_pres_april2007_frank.ppt&source=web&cd=1&ved=0CCsQFjAA&url=http%3A%2F%2Fweb.engr.oregonstate.edu%2F~pchiang%2Fclasses%2Fece679%2Fosu_pres_April2007_Frank.ppt&ei=-8pjT4GaB87UiALFhrGiDw&usg=AFQjCNF2-A1MNPKCpU6_ZZtOISD-x7ZoNA
Starting on slide 42 they explain the methods of calculating BER.
First they go through an example without ISI, and after that with
ISI included.
Slide 43-44 show the impulse response without ISI in which the
logic '1' corresponds to a level of 9 (unit?) and all the ISI
corresponds to a level of 0 (unit?). In our discussion we can
assume the unit to be voltage.
On slide 45 they show what the PDF of this logic '1' looks like,
and on slide 46 they show what its CDF looks like along with the
equation for obtaining BER(X) (integrating left to right). Slide
47 concludes that the BER above level 9 is one (error all the time)
and below 9 is zero (no error ever).
Jumping to slide 61, where the same results are shown with ISI
included, we can see a smoother transition from BER = 0 to
BER = 1. But note on slides 52 and 59 that the peak of the
impulse response in this example is 16, not 9 as in the previous
example.
Reading the numbers on slides 60 and 61 we can see that the BER
is zero below 12 and one above 26 for this logic '1' signal.
So you are correct in a sense that the BER is determined by the
characteristics of the channel (ISI), but the readout of a certain
BER value is done at a certain level. This level is basically
the threshold voltage of the input that receives and evaluates
the waveform. Without knowing this reference level you can't
get a reading for BER.
In our case, the eye mask would provide the levels at which we
should be reading the BER from the eye density or BER plots.
As I said before, in Rx_Receiver_Sensitivity we already have
this for the vertical levels, but there is nothing in IBIS-AMI
today which could give us the horizontal (time) location for
reading what the BER is. I guess the silent assumption might
be that we read the BER at 0.5 UI, which is the sampling point
of the next stage, but as far as I remember, this input might
still need a little setup and hold for the sampling to work
properly...
I disagree with your comment that "By defining an eye mask, you are
now extending IBIS AMI modeling to include how EDA tools analyze the results
of simulation.", because this eye mask doesn't have anything to
do with what algorithms the tool uses when comparing the
waveforms (the eye diagrams) with the eye mask. The tool can
do that any whichever way it wants to. The eye mask provides
the tool information about the input characteristics (or
requirements) of the stage that follows the Rx AMI model,
so this information is actually model data. This is a
glorified equivalent of the good old IBIS Vinh and Vinl
parameters we have used for a couple of decades. (I might
add that we could/should have had Ts and Th in addition to
Vinh and Vinl, but that's another story)...
Thanks,
Arpad
================================================================
From: Walter Katz [mailto:wkatz@xxxxxxxxxx]
Sent: Friday, March 16, 2012 11:52 AM
To: Muranyi, Arpad; ibis-macro@xxxxxxxxxxxxx
Subject: RE: [ibis-macro] Re: Eye mask definition for IBIS-AMI
Arpad,
I do not think you understood my question. I assume your diagram is a time
domain eye, with blue indicate no "events", and defines an inner eye contour.
This inner eye contour will be a function of the number of bits simulated,
typically ~1e-6 to 1e-7. In Statistical, the center of the eye is filled in
with probabilities << 1e-7.
So an eye mask only makes sense if it is defined for a specific BER, and the
BER is not a function of the model but a function of the BER requirements for a
specific channel in a specific application. Is the target BER 1e-20, 1e-17,
1e-12, 1e-9 or 1e-5.
Way back when, we agreed that the "prime directive" of AMI modeling was to
assure that all EDA tools would get the same Impulse Response in statistical
and the same waveform in time domain at the slicer, and it was up to the EDA
tool on how it analyzed these results.
By defining an eye mask, you are now extending IBIS AMI modeling to include how
EDA tools analyze the results of simulation. I am not objecting to this
extension, but this opens up many issues that are much more complex than just
defining an eye mask.
Walter
From: Muranyi, Arpad
[mailto:Arpad_Muranyi@xxxxxxxxxx]<mailto:[mailto:Arpad_Muranyi@xxxxxxxxxx]>
Sent: Friday, March 16, 2012 12:22 PM
To: Walter Katz; ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: RE: [ibis-macro] Re: Eye mask definition for IBIS-AMI
Walter,
The eye should be generated the same was as we already do.
Just draw a diamond in the middle of it, instead of two
horizontal lines.
In this picture I show what I think the Rx_Receiver_Sensitivity
parameter means with the two dashed lighter blue lines, and what
I think the eye mask could be with the darker blue diamond in
the middle, and optionally the horizontal lines above and below
the eye on the top and bottom:
[Description: cid:image001.png@01CD0396.9F60C770]
I agree, this is probably not difficult to achieve, but it is
not in the specification. I think it may be worth adding it...
Thanks,
Arpad
===============================================================
From: Walter Katz [mailto:wkatz@xxxxxxxxxx]<mailto:[mailto:wkatz@xxxxxxxxxx]>
Sent: Friday, March 16, 2012 8:23 AM
To: Muranyi, Arpad; ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: RE: [ibis-macro] Re: Eye mask definition for IBIS-AMI
Arpad,
Defining a format to describe an eye mask is trivial. Can you explain exactly
how the eye should be generated in both statistical and time domain AMI
simulations, and how the eye mask should be applied to it?
Walter
From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx]<mailto:[mailto:ibis-macro-bounce@xxxxxxxxxxxxx]>
On Behalf Of Muranyi, Arpad
Sent: Wednesday, March 14, 2012 4:28 PM
To: ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: [ibis-macro] Re: Eye mask definition for IBIS-AMI
Greg,
I am not sure we are discussing the same thing (correct me if I
am missing something in your thoughts).
The Eye Mask I am talking about basically describes the input
characteristics of the stage that follows the Rx AMI model.
It would specify what its Vinh/Vinl and Setup/Hold times are
with respect to the clock tick that is used to latch the data.
Our current specification already has Rx_Receiver_Sensitivity,
which defines the +/- voltage value around the nominal threshold
level (or Vref, if you want to call it that). The way I
understand this is that:
Vref + Rx_Receiver_Sensitivity = Vinh
Vref - Rx_Receiver_Sensitivity = Vinl
for the input that "reads" the output of Rx AMI.
What we seem to be missing in the IBIS-AMI spec is a parameter
that does the same for the timing aspect for this input. What
is the Setup and Hold for this input to actually get the correct
logic interpretation for the waveform that it sees?
My point is that if we have "Rx_Receiver_Sensitivity", we should
also have a corresponding timing parameter. If we put all these
together, we basically arrived at an eye mask definition...
Thanks,
Arpad
==================================================================
From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx]<mailto:[mailto:ibis-macro-bounce@xxxxxxxxxxxxx]>
On Behalf Of Gregory R Edlund
Sent: Wednesday, March 14, 2012 1:55 PM
To: Muranyi, Arpad
Cc: ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>;
ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
Subject: [ibis-macro] Re: Eye mask definition for IBIS-AMI
Let's ask the question another way: if jitter and noise limits are NOT stored
in the AMI model somewhere, how does the user get that information? By calling
an application engineer, probably. And the AE may need to check with the
circuit designer. It's easy for the information to get stale. The flip side
is that suppliers may be unwilling to disclose jitter and noise limits. Then
again, users often have to sign a non-disclosure agreement to get the AMI model
in the first place, so that may not be a strong argument.
Greg Edlund
Senior Engineer
Signal Integrity and System Timing
IBM Systems & Technology Group
3605 Hwy. 52 N Bldg 050-3
Rochester, MN 55901
[Description: Inactive hide details for "Muranyi, Arpad" ---03/14/2012 12:25:18
PM---Hello Everyone, I would like to solicit for some more co]"Muranyi, Arpad"
---03/14/2012 12:25:18 PM---Hello Everyone, I would like to solicit for some
more comments on this
From: "Muranyi, Arpad"
<Arpad_Muranyi@xxxxxxxxxx<mailto:Arpad_Muranyi@xxxxxxxxxx>>
To: "ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>"
<ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>>
Date: 03/14/2012 12:25 PM
Subject: [ibis-macro] Re: Eye mask definition for IBIS-AMI
Sent by: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
________________________________
Hello Everyone,
I would like to solicit for some more comments on this
question. So far I heard one general response that it
is a good idea, and a very specific response with a
syntax suggestion, but I would like to get a few more
responses on whether it would make sense to add an eye
mask definition to the IBIS-AMI specification. The reason
I am asking for more comments along these lines is because
last time I brought this topic up the majority of the
responses were along the lines of "not needed", "useless"
etc...
However, the more I think of this, the more it seems that
this would actually be useful, especially since we already
have vertical half of the eye mask definition present in the
Rx_Receiver_Sensitivity parameter. Is there a reason we
shouldn't include the horizontal (timing) aspects of the
eye opening requirement in addition to what we already
have for the vertical (voltage) requirement?
Thanks,
Arpad
=============================================================
-----Original Message-----
From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Thursday, March 01, 2012 11:43 AM
To: colin_warwick@xxxxxxxxxxx<mailto:colin_warwick@xxxxxxxxxxx>;
ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: [ibis-macro] Re: Eye mask definition for IBIS-AMI
Colin,
Thanks for the suggestion. However, before
discussing the format and syntax of the eye
mask definition, I would like to get comments
on the idea itself, such as do we all agree
that we should do this, do we agree on what
the eye mask means, etc... Once we agree,
we could hammer out the details of the syntax.
Thanks,
Arpad
===================================================
-----Original Message-----
From: colin_warwick@xxxxxxxxxxx<mailto:colin_warwick@xxxxxxxxxxx>
[mailto:colin_warwick@xxxxxxxxxxx]
Sent: Thursday, March 01, 2012 11:29 AM
To: Muranyi, Arpad; ibis-macro@xxxxxxxxxxxxx<mailto:ibis-macro@xxxxxxxxxxxxx>
Subject: RE: Eye mask definition for IBIS-AMI
Hello Arpad,
Our 'scope *msk files look like this:
/* Top Polygon */
/* Polygon Number */ 1
/* Number of vertices */ 4
+0.0, 0.6
+0.0, 0.5
+1.0, 0.5
+1.0, 0.6
/* Middle Polygon */ 2
/* Number of vertices */ 6
/* X1, Yt */ 0.300 , 0.000
/* X2, Y+ */ 0.500 , 0.073
/* 1-X2, Y+ */ 0.500 , 0.073
/* 1-X1, Yt */ 0.700 , 0.000
/* 1-X2, Y- */ 0.500 , -0.073
/* X2, Y- */ 0.500 , -0.073
/* Bottom Polygon */
/* Polygon Number */ 3
/* Number of vertices */ 4
+0.0, -0.6
+0.0, -0.5
+1.0, -0.5
+1.0, -0.6
I don't know if the spec is published or not nor whether there's IP in the spec
(I'm in the EEsof division). But if this format is of interest, I can ask my
'scopes colleagues to get involved.
-- Colin Warwick
Agilent EEsof EDA
http://signal-integrity.tm.agilent.com
-----Original Message-----
From: ibis-macro-bounce@xxxxxxxxxxxxx<mailto:ibis-macro-bounce@xxxxxxxxxxxxx>
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Wednesday, February 29, 2012 11:43 PM
To: IBIS-ATM
Subject: [ibis-macro] Eye mask definition for IBIS-AMI
Hello everyone,
We had a few conversations on this topic in the past on this
email reflector and also in the IBIS-ATM meetings, but the
discussions didn't result in any actions as far as the
specification is concerned. I would like to bring this
topic up again because to me it seems that we have an
incomplete "solution" for this in the IBIS specification.
Currently, we have an AMI parameter in specification called
"Rx_Receiver_Sensitivity". Pg. 147 describes this parameter
like this:
| Rx_Receiver_Sensitivity can be of Usage Info and Out and of
| Type Float and of Data Format Value, Range and Corner.
| Rx_Receiver_Sensitivity tells the EDA platform the voltage
| needed at the receiver data decision point to ensure proper
| sampling of the equalized signal.
In other words, this is basically the Vinh and Vinl of the
core's input stage that is connected to the output of the Rx
buffer which contains all the fun stuff, the filter(s), DFE,
CDR, etc... Since the AMI simulation's output waveform goes
into this input, the eye opening we get from the AMI simulations
better be above and below +/-Rx_Receiver_Sensitivity, otherwise
this input stage will not be able to interpret the signals
correctly.
However, there is nothing in the AMI specification which would
help us to evaluate the timing relationship requirements between
these waveforms and the clock ticks. In other words, we have
no information about what the setup and hold times are supposed
to be around the sampling point for this input stage to recognize
the signal correctly.
To me it would make sense to define a complete eye mask which
would include a definition for the width of the eye (time) and
its time relationship to the clock ticks as well as for its
height (voltage). I would be willing to write a BIRD for this,
but before I begin, I would like to have a little brainstorming
session to collect some ideas on how people would deal with this
and what we should put into the specification.
As a bare minimum, I would like to propose a set of parameters
which defines the shape of a diamond, laying on its side:
*********
* *
* *
* *
* *
* *
*********
For the voltages we could have a high/low level and DC offset
and for the timing we could have a left/right corner and
the beginning and end points of the horizontal lines. This
could be achieved with seven parameters.
I could see a few additional parameters for a more elaborate
definition, but for now I will leave it at that.
Questions, comments welcome.
Thanks,
Arpad
================================================================
---------------------------------------------------------------------
IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/
IBIS Macro reflector: //www.freelists.org/list/ibis-macro
To unsubscribe send an email:
To: ibis-macro-request@xxxxxxxxxxxxx<mailto:ibis-macro-request@xxxxxxxxxxxxx>
Subject: unsubscribe
---------------------------------------------------------------------
IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/
IBIS Macro reflector: //www.freelists.org/list/ibis-macro
To unsubscribe send an email:
To: ibis-macro-request@xxxxxxxxxxxxx<mailto:ibis-macro-request@xxxxxxxxxxxxx>
Subject: unsubscribe
---------------------------------------------------------------------
IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/
IBIS Macro reflector: //www.freelists.org/list/ibis-macro
To unsubscribe send an email:
To: ibis-macro-request@xxxxxxxxxxxxx<mailto:ibis-macro-request@xxxxxxxxxxxxx>
Subject: unsubscribe
Other related posts:
