IBIS Macromodel Task Group Meeting date: 26 January 2016 Members (asterisk for those attending): ANSYS: * Dan Dvorscak * Curtis Clark Avago (LSI): Xingdong Dai * Bob Miller Cadence Design Systems: * Ambrish Varma Brad Brim Kumar Keshavan Ken Willis Cisco: Seungyong (Brian) Baek eASIC: David Banas Marc Kowalski Ericsson: Anders Ekholm GlobalFoundries: Steve Parker Intel: Michael Mirmak Keysight Technologies: Fangyi Rao * Radek Biernacki Maxim Integrated Products: Hassan Rafat Mentor Graphics: John Angulo * Arpad Muranyi Micron Technology: Randy Wolff Justin Butterfield QLogic Corp.: James Zhou Andy Joy SiSoft: * Walter Katz Todd Westerhoff Mike LaBonte Synopsys: Rita Horner Teraspeed Consulting Group: Scott McMorrow Teraspeed Labs: * Bob Ross TI: Alfred Chong The meeting was led by Arpad Muranyi. -------------------------------------------------------------------------------- Opens: - None. -------------------------- Call for patent disclosure: - None. ------------- Review of ARs: - None. ------------------------- Review of Meeting Minutes: - Arpad: Does anyone have any comments or corrections? [none] - Arpad: Motion to approve the minutes. - Ambrish: Second. - Arpad: Anyone opposed? [none] ------------- New Discussion: Back Channel BIRD 147.1: - Discussion: Arpad asked if Ambrish had made progress in this area. Ambrish said that he had noted a lot of interest and positive feedback at DesignCon. He said he planned to restart work on the topic shortly. Arpad asked Walter whether he had any plans to submit his own optimization proposal, in light of SiSoft's recently announced optimization tool. Walter said that they were interested in co-optimizing Tx and Rx models, but their tools had thus far introduced techniques that worked with standard AMI Models, and thus were independent of this topic. He also noted that he had produced a proposal that allowed Tx and Rx models to advertise their capabilities and communicate. That proposal had been tabled until Cadence had a chance to respond. Bob M. noted that Avago had a lot of interest in getting Tx to Rx communication for co-optimization because they wanted a technique that would allow them to match what is done in the actual hardware. There was some discussion on how easy it would be for people to find the latest working versions of the BIRDs and proposals. Radek suggested that Ambrish check for the latest version of his proposal and make sure it is posted. Walter said he would do the same for his proposal. Discussion of language corrections regarding "ground": - Walter: [Sharing the "Device Under Test vs. Device In Action presentation he had sent to the ATM group mailing list]. - [referring to slide 2, which shows Figure 16 from the IBIS 6.1 spec] - Aside from AMI, [External Model], and perhaps some packaging information, IBIS is generally a specification for how to generate a behavioral model of a device from measurements. - This figure shows the test fixture for doing the measurements. - In this case, we are measuring the voltage at the Pin in a test fixture. - Every measurement has a reference, which is the reference for the test fixture, which is shown by the ground symbol. - For a classic IBIS model that has only the [Voltage Range] keyword, the assumption is that the reference for that Model is a ground Pin. From the spec, "For many buffers, only one power pin and one common ground pin terminal are used." - That terminology isn't very clear ("common ground pin terminal"). [Pin Mapping] actually tells you which GND [Pin] is the reference ground for the [Model]. - The standard says to hook the test fixture probe to the I/O pin and make sure the test fixture is grounded to the appropriate GND [Pin] of that device. They call that ground the "absolute ground" or the "absolute ground Pin." - That's the reference for the V_fixture voltage and the package model. - [referring to the next slide "Device in Action"] - Same buffer now "in action." - DUT is now a device in action (DIA). - We really now have a 3 terminal device, I/O Pad, VDD, VSS. - We don't know what VDD and VSS are at any given time, they are now provided by a PDN. - There is no "absolute ground" here. There is VSS, which is connected to the GND Pin for this buffer. - We actually only have two voltage differences. - IBIS only talks about a DUT. - The paragraph before figure 16 is confusing. - Attempt to clarify that with proposed new language: - "For [Model]s that do not specify [Pulldown Reference] or [GND Clamp Reference] the reference for the V_fixture voltage shall be a ground pin for that [Model] under test as specified in the [Pin Mapping] section or a common ground pin." - C_comp would also be connected to this. - When [Pulldown Reference] or [GND Clamp Reference] are specified, and they are non-zero, then we need to create a reference node for that device. - If you're measuring V_fixture, it has to be with respect to some reference node. - We could define a reference node with a resistor to either the [Pullup Reference] or [POWER Clamp Reference], and a resistor to either the [Pulldown Reference] or [GND Clamp Reference]. Create a voltage divider between the existing reference nodes. - Introduce 4 new parameters to define these resistance values. - You would also know where to connect C_comp based on which of the new parameters R_DUT_pdref or R_DUT_gcref is used. - We need a reference node when the device is in action. - We can't rely on the rails providing what the [Model] says. We assume they come from a PDN and can vary. - We need a reference node within the device, so when the device is in action we can compare the voltage at the pad relative to that node with threshold values like Vinl, Vinh. - None of these other references ([Pullup Reference], etc.) make any sense unless we have a reference node inside the device in action. - Bob R: I don't think we need to complicate this section. - With respect to extracting IBIS parameters, the external reference is already defined as ground for all voltages. - We already have a way to define V_fixture with respect to an external reference voltage, which need not be any of the rails. That is supported and can be used to generate the internal K table models. - We have a more general solution now without the additional parameters. - This proposal also introduces the notion that we are extracting the [Model] from the Pin interface. The extracted model is at the die. There would have to be a de-embedding if we were hooking up to a ground pin to power the device during measurement. I don't think that was intended. I think we can safely assume the references are rock solid, at least for SPICE extraction. - In general, I want five voltages pu, pc, pd, gc, pad. We shouldn't try to simplify this figure. - Discussion: Radek said he thought Walter's proposal was a step in the right direction. He said that for extraction, even if all the references are different from zero they are with respect to some local ground. That node exists and needs to be pointed out as Walter was attempting to do. He said we therefore need six nodes, not five. Arpad agreed and wondered if we might introduce an Int_ref similar to Ext_ref. Both felt the proposed bridge divider circuit for generating the reference node might be too limiting. Radek wondered if it wouldn't be easier to leave that to the package model maker so we didn't have to invent a bridge circuit interpretation. Bob pointed out that we have an inherent conflict with some measurements based on the die and some based on the Pin. Radek agreed and said we would need to revisit the topic. - Arpad: Thank you all for joining. AR: Ambrish to find and post the latest version of his Back Channel proposal. AR: Walter to find and post the latest version of his co-optimization proposal. ------------- Next meeting: 2 February 2016 12:00pm PT ------------- IBIS Interconnect SPICE Wish List: 1) Simulator directives