----------------------------------------------------- 8 April 2004 Attendance ahealey@agere.com kcheong@marvell.com john.dambrosia@tycoelectronics.com Steve.Anderson@xilinx.com Herman.Eiliya@analog.com cathyl@lsil.com m_oltmanns@comcast.net why@pmc-sierra.com glen@vitesse.com GOleynick@fciconnect.com michael.w.altmann@intel.com mike@mike-lerer.com Bill.Woodruff@velio.com hrahbar@lsil.com T.Yoshizawa@motorola.com pavle@Centillium.com michael.w.altmann@intel.com tomaz@force10networks.com luke.chang@intel.com mbarazan@vitesse.com sbardone@force10networks.com ---------------------------------------- straw polls: Consensus was reached on using Sdd21 as opposed to S21 No one objects to having DC blocking cap be part of the receiver at least 3 people disagree with placing TP4 between the cap and the chip-- make the capacitor part of the receiver take a straw poll anyone against mandating ac coupling 3 people against removing DC options Agreement to go to 20 GHz on frequency response based on 3rd harmonic of 8b10b no objection to 12.5 Gbaud to handle 8b10b contingencies-- worst case to continue with the channel work no initial objection to 800 mV differential peak to peak as the worst case transmit amplitude-- But then Jeff said he would like to leave it unspecified. important because of loss characteristics for the channel-- cannot have a zero or negative signal-- peak to peak amplitude-- to continue with worst case analysis-- at this point does anyone have an issue going to ±10%? No objection as a placeholder connectors fitting in that scheme-- 20% is what the chip guys like-- for chips but we are spec'ing a channel channel only-- then 20% is too loose-- ------------------------------------------ Raw notes ------------------------------------------ Attendance taken by e-mail Joel Goergen-- brian taking notes today-- minutes and attendance taken care of email last night had an outline can only recommend not going to vote or decide anything recommendations and straw polls Next conference calls will be on April 22 and May 5, same time same channel Face to face meeting scheduled for May 26 and May 27 in Long Beach for Backplane Ethernet Continue to fine tune the parameters of the channel-- Attachment-- pdf-- sent out to the group-- Also sent to David Law. In terms of operating rules, I will ask the following... - Will someone volunteer to record the minutes? - Please send me an email once you are on the call. I will email the list out ... and later again with the minutes. - I'd like to limit the call to 2hours. - We can only recommend. - We can straw poll. - We can talk cost in relative terms only. In terms of the call ... - State the next call day/time 22apr04 and 5may04 - All shared sims preferred to be matlab scripts - Discuss attachment Outline presented in PDF document "backplane proposal conf call 8april04.pdf" PDF cx4 specification adam healy and david law-- pdf on the specification-- if you don't have it you need to get it Slide 3 came from Steve Dwyer-- not exactly the same picture as in the document Steve's picture shows a lot more data-- Dj Rj return loss, FR4 loss, cable loss FR4 references should be considered to be improved FR4 FR4+ is to be interpreted as "improved" FR4 XAUI clause 47 has the FR4 losses-- 3.125 GHz limitation question at receiver-- AC coupling cap-- separate test point? or just shoved within the receiver block-- baseline for channel-- build 5 cards-- with an SMA connector that will be available for anybody within the backplane study group Adam starts doing measurements-- disagree with something written-- has one of these cards-- run tests and parameters- agree or disagree on what the test equipment is showing-- after .ae and .Z OIF and fibre channel work need to have these cards in place-- if there are any cards on these things Mike-- reference worst case or typical channels-- calibration mechanism for simulation s-parametres, frequency response- something we can all get back to-- measure and simulate-- on test cards in lab-- disagree with something tyco comes up with, how to agree-- some boards from tyco to share-- force10 will come up with a set as well way to get concensus-- difficulty getting concensus-- hard to share some of the data with jeff cane a cisco-- with common platform can go back to Correlation-- corroboration boards that facilitate corroboration of test equipment John D-- loss vs. frequency in the impedance-- of the FR4 loss phase and group delays need to be in there return loss FR4 only shows loss vs frequency and impedance-- need to add in the phase information joel will extend the NEXT FEXT Z Phase All of these are not strictly magnitudes-- by measuring next and fext we are using complex datapoints and numbers mean to delete the loss vs. frequency and z on the outside FR-4's. make it a complex value-- go from paren-- all 3 fr4's and both plugs and both jacks. loss vs frequency is assumed to be S21-- total Next and Fext means more than 4 ports-- basing on individual measurements-- then it is 4 ports 2 port equipment vs. 4 port equipment-- mathematical trick to go from 2 ports to 4 ports-- can get different measurements-- Joel-- most people have 2 ports not 4 ports-- 2 port was a very common analyzer-- 4 port just became commercially available recently-- loss vs. frequency S21 assumed, had not taken into account-- RL listed-- no thoughts yet for S11 or S22. John-- Differential S21 no mode conversions are included? Joel-- is assuming single ended S21 taken on a 4-port system. S21 vs Sdd21 can be different according to Graeme depending on what the couplings are Joel is calling out S21 on 4 ports john-- Sdd21-- if port 2 to port 1-- 1 to 2 hooked up 3 to 4 hooked up-- doesnt match Sdd21-- nomenclature issue mostly S21-- in terms of a 4 port system with a channel that is just as much single ended as it is differential S21 and the Sdd are going to be very similar S21 or Sdd not do 4 port analysis-- not the right thing to do to convert from a 2 port to a 4 port conversion-- someone has to start with something-- put out an S21-- if we need more than we need to have something else-- or say s21 will not work any more-- Graeme believes Sdd21 is necessary John D says we need all the 4 port information-- Joel-- all 16 matrices in the equation then you dont need any of the Z or the Phase-- if you put all the matrices up-- that is all the data there is-- John-- return loss or loss vs frequency-- it would all be in the 4 port measurement would have to come up with the 16 equations-- to normalize or minimize or maximize-- nominalize the channel measurement-- Graeme-- All 16 are a nicety, but if you will stick with equation based modeling then it is not realistic, must stay with Sdd21-- cannot throw 16 of them in. John-- equation based model-- normative model-- stick with Sdd21 informative-- use all of the m TP1 and TP2-- we seem to be going in two directions here Graeme-- identifying TP1 and TP2-- have we precluded-- Calibration must be done very carefully if 2 port equipment is used only informative annex-- Graeme will help math from 2 ports to 4 ports-- have to define the conversion so that we can agree-- Sdd only of S21-- if we include 2 port and the differential numbers differential and common mode-- ignore the 2 port for specing the channel-- differential signaling from one end to the other-- 2 port measurement-- does that add any value to the spec-- adds value informatively-- S21 should be listed informatively All through the IEEE relation to CX4 and to XAUI-- DD and CD-- not widely accepted within the IEEE yet S21 with differential terminology? implied by the use of differential in the specification. joel-- consider it striked-- Look in terms of Sdd21-- Steve anderson likes it as well-- equation to be talked about will not change any combination of trace length and daughtercards onthe backplane 40 inches with 2 connectors-- 1 meter with 2 connectors as defined in the PAR criteria one issue the length-- connectors and material will have more effect than the actual length between them extremely difficult to John D'Ambrosia and Jeff Cain-- XAUI equation set-- second order-- defines very well what the channel parameters are-- spending a lot of time with a 3rd order equation set get more of the positive and negative curvatures that exist in real time from TP1 to TP2 output end of the mated connector tp2 tp3 input end of the mated connector-- where are tp1 and tp4-- will concede for this conversation-- change slides-- next Fext and Z as opposed to loss vs frequency either list all16 data sets sDD scD Scc next fext TP1 and TP4-- straw poll-- ask the question-- is there anyone that is against-- looking at TP1 and TP4 as defined change the TP4-- where are the others-- xmit, chan, and rx-- package pins-- where the fr4 ends and the device begins-- at the package pins-- joel- for more clarification-- TP2 at the receive filter-- ac coupling cap-- AC coupling cap-- out of this discussion-- for this particular discussion-- in his own mind, which side of tp4 the ac cap goes on-- receive filter and -- receive serdes device-- had not thrown in at which side it should go on-- from the spec point of view if you are going to exclude that cap and it is external-- how do we know that the standard will work none of the past stds-- include the ac block-- xaui includes it if required eye template is measured with ac cupl 47.344-- AC coupling is part of the receiver-- table 47-3 receiver coupling value -- external cx-4 AC coupling is part of the receiver unless stated otherwise-- various methods max value 470 pF-- limit inrush currents when connected to higher voltage transmitters-- assumed to be philosophical perspective-- responsibility of the designer to pass the specifications-- rate change becomes more difficult-- john is alluded to-- do we mandate AC coupling-- joel would be torn -- mandated in xaui-- AC is required-- CX-4-- mandated-- assuming that the ac cap is part of the receiver-- is anyone uncomfortable with TP1 and TP4 as defined on page 10 no one uncomfortable-- no objections assuming we don't mandate AC coupling-- TP1 to TP4 with the AC cap being on the channel-- 2 specs-- channel model itself-- would have to include the cap and not include the cap-- it shall be ac coupled-- cap in the receiver not in the channel-- there are a few people that would disagree with not including the AC cap as Graeme disagrees as well-- it should be part of the receiver-- no built in capacitors-- --------------------------------------------------- assuming that on the test fixturing-- sucked into the receiver die or sitting outside the chip-- cap to the channel not from the cap to the receiver-- at least 3 people disagree with placing TP4 between the cap and the chip-- make the capacitor part of the receiver --------------------------------------------------------- OK so people are in acceptance with TP1 and TP4-- with AC cap included in the receiver-- on the right side of TP4-- Joel will republish the note-- slip in a tiny block showing that the AC cap shows up on the right side of TP4-- take a straw poll anyone against mandating ac coupling graeme is against it-- for long reach at these frequencies-- headroom at the transmitter could be an issue-- therer are sredes on the market that do not require capacitors-- from a serdes supplier standpoint-- there are some systems that have dc coupling-- it was a pain to build, but it was done-- joel's question-- 90 nm and smaller-- 2.5 volts or 3.3 volts vdd io's once 1.2 cml implementations-- issue if not ac coupling esp. common mode newer technologies-- meet on the xmit side-- to the old receiver-- may have some issues-- may come in handy if the old gernations hvae it the new cards have to support the OIF-- support optional DC mode, but requires compatible process technologies on both sides-- if different technologies have to go back to ac coupled 802.3 has always gone-- modular, connectivity basis-- confining to backplane and solid gorund plane, think that we will complicate things too much more if we have AC and DC coupling-- transmitter is the same-- receiver may optionally have a dc capability-- dc block-- if a card -- then the xmitter doesnt take all the heat-- not a true statement that the tx doesn't have to take more-- could be additional surge current requirements- and additiona lrequiments on the channel-- don't need to make the recommendation today-- need to think more about hte dc blocking issue-- need to spec tx and rx, also spec suppression issues iwth hte channel- keep in mind-- keep a dc current to flow on this channel-- have to look at the dc characteristics as well as the ac characteristics of hte channel agency issue-- UL issues onthe bakcplane higher frequencies-- some problems-- hard to do. top layer to run it into the chip-- emi issues at higher frequencies-- From a system layout perspective-- # of vias as well as microstrip to stirpline configuration. Based on page 10-- group is comfortable talking about tp1 and tp4-- are there some work been done that changing the source power actually changes Sdd21-- not willing to put them out there-- OIF and IEEE-- corroboration-- old 2 port equipment-- cannot go to 100 Hz-- 500 Hz zone may be better-- 20 GHz 2 port s parameters out the-- below 60 db of attenuation-- why bother-- 1mm long-- couple db loss at assuming modulation if you spec amplitudes-- making assumptions about what we are trying to present in modulation on the jitter side-- 18 go to 20-- 18. 6.25 18.75-- rise and fall time for impulse response-- perfect transmitter-- no objection to 12.5 Gbaud to handle 8b10b contingencies-- worst case to continue with the channel work no objection to 800 mV differential peak to peak as the worst case transmit amplitude-- important because of loss characteristics for the channel-- cannot have a zero or negative signal-- peak to peak amplitude-- to continue with worst case analysis-- Jeff Cain-- any other number that Cisco would suggest-- skip until later. No one uncomfortable with ± 20 ohms on 100 ohm diff impedance-- why do we need a spec at DC XAUI 100 ohms ± 10 47.3.5.1 10% is very common for PCB fab-- change it to 10% because that is very common-- at this point does anyone have an issue going to 10%? connectors fitting in that scheme-- 20% is what the chip guys like-- channel only-- then 20% is too loose-- 10% on the recommendation-- if anyone has an issue we can talk about-- joel-- cisco does not spec backplane connection with an impedance tolerance-- spec on fr4 is the connector in a board or connector in itself-- receiver coupling-- place holders-- 12.5 gigs for now diff input level does not apply to the channel next and fext-- terminate victim channels with 100 ohm terminations that you find at the receiver-- take a look at this equation profile how does it match sdd curves on our system-- get some feedback within the next week on how we feel it models the channel-- the last 2 things there-- jeff-- still on the call-- slide 19-- page 19-- stolen off CX4 in fundamental shape-- came from Cisco-- should be multiple-- magnitude-- and phase is in there on page 19 next measurement is not a smooth curve, but it represents the peak in those humps-- next call-- issues on the sdd--