Greg, If an AMI model does have a Sample_Per_Bit requirement, then the EDA tool must do one of the following: 1. Choose a samples per bit that satisfies all of the Tx and Rx samples per bit requirements - not always possible when different models have different samples per bit requirements. 2. Choose a samples per bit for a simulation, and then run the impulse response and waveforms through a Torque Converter (Gearbox) that converts these waveforms to the model's required Samples_Per_Bit before calling the model, and then run the output of the model back to the Torque Converter backwards. In the ideal world, if the internals of the model do have a Samples_Per_Bit limitation, the model should be doing the Torque Conversions described above. According to IBIS 5.0 (and the proposed 5.1), a model which does have a Samples Per Bit limitation is non-compliant. If we add a reserved parameter Samples_Per_Bit in IBIS 5.2, then these models can be made compliant to 5.2 if they specify Samples_Per_Bit in the .ami file. Ken Willis made in a separate e-mail branch: Good discussion. I agree it is good practice to document the model usage. If the issue is really one of documentation, then I don't know if it is something we have to put into the standard. Most of the IBIS-AMI model kits have some readme or pdf documentation with them. This is not really different than regular Spice or IBIS models that have been distributed for many years. With AMI, you can even put comments into the .ami files themselves. So there are ways to properly document the models, without adding more keywords to the spec. Unfortunately, model makers often do not document the Samples Per Bit limitation, and in some cases do not realize they have a Samples Per Bit limitation because they only test them at one Samples Per Bit. Also, it is difficult for software to programmatically read these .pdf files and extract Samples Per Bit limitations. We certainly can add a compliance test to IBIS-AMI that would use the AMI test benches to verify that a model works at any number of Samples Per Bit. Walter From: ibis-macro-bounce@xxxxxxxxxxxxx [mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Gregory R Edlund Sent: Thursday, September 29, 2011 9:32 AM To: Arpad_Muranyi@xxxxxxxxxx Cc: IBIS-ATM; ibis-macro-bounce@xxxxxxxxxxxxx Subject: [ibis-macro] Re: Samples per bit for AMI Arpad, Do you envision the new samples per bit parameter being used by the EDA tool? Or is it for documentation purposes? There is another way to handle this scenario: we could integrate samples per bit into the quality check list, i.e. the model maker uses the quality check list as a means of communication with the user. Greg Edlund Senior Engineer Signal Integrity and System Timing IBM Systems & Technology Group 3605 Hwy. 52 N Bldg 050-3 Rochester, MN 55901 Inactive hide details for "Muranyi, Arpad" ---09/28/2011 06:44:32 PM---Hello everyone, I ran across yet another model today whi"Muranyi, Arpad" ---09/28/2011 06:44:32 PM---Hello everyone, I ran across yet another model today which was misbehaving From: "Muranyi, Arpad" <Arpad_Muranyi@xxxxxxxxxx> To: IBIS-ATM <ibis-macro@xxxxxxxxxxxxx> Date: 09/28/2011 06:44 PM Subject: [ibis-macro] Samples per bit for AMI Sent by: ibis-macro-bounce@xxxxxxxxxxxxx _____ Hello everyone, I ran across yet another model today which was misbehaving in our tool because it turned out that it would only work with 64 samples per bit and the user unaware of this used a different value. When I wrote about this topic some time ago, I was wrestling with another AMI model that would only work with 32 samples per bit. Neither of these models came with any documentation on what samples per bit settings they will work with. The problem is that when such models are misbehaving, all kinds of "fun stuff" is starting to happen, anywhere from crashes to explosions :-). I believe that the most robust solution would be to add a reserved parameter for samples per bit to the IBIS-AMI specification so that the model makers would be forced to document in the .ami parameter file sampling rate(s) the model will work with. Making blanket statements in the spec is not going to guarantee that the model maker will actually do anything about it, they may not even read that part of the specification at all... Please look over the attached BIRD draft in which I attempt to solve this problem by adding a reserved parameter to the specification. Questions, comments are welcome. Thanks, Arpad ============================================================== [attachment "SamplingRateBIRD_01.pdf" deleted by Gregory R Edlund/Rochester/IBM]