====================================================================== IBIS INTERCONNECT TASK GROUP MEETING http://www.eda.org/ibis/interconnect_wip/ Mailing list: ibis-interconn@freelists.org ====================================================================== Next meeting: Feb. 27, 2013 8 AM US Pacific Time Agenda: Attendance Call for Patents Agenda and Opens Review Si2 Specification Status S-parameter Definition/Stimulus Assumptions (D. Banas) Proceed with EMD as an independent specification? Next Meetings’ Schedule/Agenda • Feb. 27 • Mar. 6 For international numbers, please contact Michael Mirmak. Note: in case of issues with Lync we will use the WebEx noted at the bottom of this message ......................................................................................................................................... Join online meeting https://meet.intel.com/michael.mirmak/QZ193W0C First online meeting? [!OC([1033])!] ......................................................................................................................................... <--- Reservationless Bridge – Do not edit or remove --- 916-356-2663, 8-356-2663, Bridge: 2, Passcode: 8625431 Speed dialer: inteldialer://2,8625431 -----------------------------------------------------------------> Note: in case of issues with Lync we will use the WebEx noted at the bottom of this message ====================================================================== Attendees, Feb. 27 Agilent Technologies Radek Biernacki* Altera David Banas* ANSYS [Luis Armenta], Steve Pytel Cadence Design Systems Brad Brim*, Ambrish Varma Intel Michael Mirmak* Mentor Graphics Arpad Muranyi* Micron Technology Justin Butterfield*, Randy Wolff* QLogic Jason Zhou Signal Integrity Software Walter Katz* Teraspeed Consulting Group Bob Ross* Minutes No patents were declared. Brad Brim reported that the second week in March is the approximate Si2 specification public release target. Brad will distribute it to this group once it is available. On the question of how the new BIRD and Interconnect relate, Walter Katz stated that there’s no impact from new BIRD on Interconnect work, including on-die interconnect or package interconnect. In legacy IBIS, the buffer is at the pin, while in the proposed IBIS 6.0 approach, there’s “something” between the pin and the buffer. Bob Ross asked whether the on-die interconnect would be absorbed by the package. Walter replied that it could be, but that’s not the intent; in general, the package model should be between the pin of the package and the die pad; on-die interconnect should be between on-die buffer and the pad. David Banas requested the technical rationale for the new BIRD. Walter responded that the on-die interconnect is generated by the IC tool, while the package models are from a multi-chip module or other package design tool, with a different database. David asked whether the output from both is a Touchstone file, which could be cascaded. Walter answered yes, but that wouldn’t support multiple packages for a single die, etc. If we had two different TS files for this purpose in our structure, would that work? Walter suggested not, as power delivery networks would cause difficulty here. David asked whether we could introduce into this group an ATM topic this week, the question of whether it’s OK to include a buffer’s AC characteristics into the on-die interconnect section. David noted that, according to Walter, many IC vendors are defining the AC models of their on-die interconnect using an ideal step function, because the makers of the Touchstone file have combined the device/buffer AC characteristics into the on-die interconnect. David would like everyone to explicitly understand this. This seems to violate the idea that S-parameters violates the assumption that S- parameters behave independently of how they are driven. Walter provided the example that IC vendors create Touchstone files such that T-coils are on the buffer side. Bob asked whether some vendors do it one way, others do it another way. David replied that what’s being proposed is different than current engineering education about S-parameter behavior. Michael Mirmak expressed some confusion about the distinction; isn’t this a bandwidth problem, where an infinite step in the time domain implies an infinite bandwidth in the frequency domain? David replied that, if you take an S-parameter block and drive it with an ideal step, you get a correct (accurate) result; alternately, if you drive it with a realistic step, you will get the wrong result (less correct or accurate than desired). Radek Biernacki added that we are touching the fundamental issue of whether the data is correct or not. Inaccuracy could be a result of the circuit being described not being LTI. Walter mentioned IBM as a proponent of the proposed method. If you have a 10 GHz channel (at Nyquist), the risetime is ~10x faster (~100 GHz) . If you talk about edge, there is curvature, and high- frequency areas are therefore attenuated. This doesn’t affect engineering decisions. Computational assumptions are simplifications. Arpad suggested thinking about an RC circuit, with a certain time constant, an ideal step response input, and an exponential output. If the input is not ideal, and the slope is faster than the RC time constant, you get a highly similar result; slower, you get a waveform dominated by the input slope. Bob asked why anyone would use a non-ideal step. Michael again asked about the infinite bandwidth problem: can’t we just specify the bandwidth limitation from input to output? An extra sharp waveform implies an infinite slope implies an infinite bandwidth. David asked whether the Touchstone file in the proposal is a representation of the physical t-coil and any other physical structures, or do we actually state that some passive *and* some active element components are included? He added that he is worried that half of us are riding along on a train without understanding what we’re signing up for. The proposed template calls out for an ideal edge. Walter replied that one could have a new reserved parameter or use BIRD116 to express an explicit edge rate. Radek added that it’s up to the model maker to specify the point of stimulus that actually feeds the rest of the buffer. The S-parameters may include more than passive behavior. Brad spoke, noting that his background is in RF and microwave design. The S-parameters in this context are linear but not necessarily passive; linear, small-signal response. RF transistor S-parameters are measured at a bias point; change the bias point and the S-parameters will be different as well. The more stuff you put in it, the more apt recipients are to misuse them under a different bias condition. Radek replied that IBIS-AMI usage of Touchstone doesn’t rely on the same definitions with the same bias points. We are making assumptions about what is inside. We are calling it an S-parameter, linear and small-signal by definition; we have to make it really clear that we are using it this way. LTI is the fundamental assumption. David suggested that all participants attempt to obtain answers on S-parameter assumptions, including test conditions, from everyone. Who decides? EDA tool vendors? Model makers? IC vendors? A specification/cookbook definition? David will discuss the specifics with his IC design colleagues and document presumptions, reporting them here. Bob stated that, per Walter’s statement, the new BIRD doesn’t apply; BUT the output of the TX and the input of the RX is called the “channel”; does it include the redistribution layer or not? Walter stated that it does not in 5.0/5.1, but does in 6.0. There is not explicit on-die interconnect object in 5.0/5.1; it will be explicit in 6.0 and included in the channel. Michael noted that the group is “in hold mode” regarding EMD and Si2; the new BIRD doesn’t affect this. Walter added that believes EMD is “on the table” for consideration by this group. He proposed that this body verify EMD as direction and confirm parameter tree structure as the direction for the EMD specification. Michael noted that EMD can point to IBIS, but IBIS doesn’t have to point to EMD; EMD can be developed as an independent specification without having IBIS mention it. Bob added that EMD could point to Sections 1-3 of IBIS to define syntactical rules. Walter strongly recommended keeping it a separate document, just as the same as Touchstone. Walter moved that the group take up the question next time of proceeding with EMD, using parameter trees, as separate specification. The team will discuss and, assuming no issues, vote on this question next time. ====================================================================== In case of Lync issues only, we will switch to WebEx as noted below. Meeting Number: 732 940 715 Meeting Password: IBIS ------------------------------------------------------- To join this meeting (Now from mobile devices!) ------------------------------------------------------- 1. Go to https://sisoft.webex.com/sisoft/j.php?J=732940715&PW=NNWY2NmRmZTY0 2. If requested, enter your name and email address. 3. If a password is required, enter the meeting password: IBIS 4. Click "Join". 5. Follow the instructions that appear on your screen. ------------------------------------------------------- Audio conference information ------------------------------------------------------- Call-in toll number (US/Canada): 1-650-479-3208 Access code:732 940 715 http://www.webex.com IMPORTANT NOTICE: This WebEx service includes a feature that allows audio and any documents and other materials exchanged or viewed during the session to be recorded. By joining this session, you automatically consent to such recordings. If you do not consent to the recording, discuss your concerns with the meeting host prior to the start of the recording or do not join the session. Please note that any such recordings may be subject to discovery in the event of litigation.