Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [BP] Follow-up to Friday's ad-hoc call - SDD21 curve



Joe,

Let's put the reflector to good use.

 

I strongly recommend that you follow my lead with withdrawing my proposal, for a very simple reason - it doesn't meet our needs.  I stated at the meeting, that when I came up with my proposal, the thought was to use what Steve had accomplished at the lower frequencies, while keeping more in step with the original model at the higher frequencies.  This was not done in review with data, merely with review of the channel models.  If you look at my initial email with the proposal, I clearly stated that I would be looking at the proposed models with real data.  When you look at real data against the proposed channel models, both mine and Steve's models had failures at the low frequencies, while Joel's model worked the best. 

 

I have a few other questions and comments.

a) a continuous channel model is necessary, and is more than just aesthetic.  Furthermore, the step at the 3 GHz frequency would do a pitiful job of representing all discontinuities.  Steve actually brought this issue up in his email in regards to the stubs, and since sharing my data I have actually been contacted by people who are asking "What about the stub?"  We both know if a stub effect were to be included the mere 1 to 1.5 db delta that joel's model presents at 5 GHz would be a dream.

 

b) Please quantify how much power you feel Joel's new proposal would add.

 

c) I am talking to many system vendors now, and the assumption that they will go to "improved materials" is being over-estimated.  I am talking to several vendors now, and am actually amazed at how their new designs will be using 4000-13, and these are designs they intend to take up to 10G serial.  Furthermore, there was concern expressed at the May meeting that the lowest cost "improved FR-4" should be used on the daughtercards (which is why we actually fabed daughtercards with both -13 and -13SI).  Furthermore, while the prices of some materials are coming down, the price of the older materials is also coming down, making them still to be more attractive.  

 

d) You make mention of "legacy" channels in your email.  Please clarify what legacy backplanes have been used to justify the model.  I frown on the use of "legacy" as it is not descriptive enough.  Greenfield channels could be just as bad as legacy channels.  Just because better materials or different techniques are available, it does not mean that they will be implemented. 

 

e) Please note that the definition of "improved FR-4" does include a reference to temperature up to 70C.  Also note that in Joel's presentation to the committee, it is stated that a call is needed to address the effects of temperature and humidity on "improved FR-4".  The data that has been presented to the group to date, clearly shows that this will be a detrimental effect to the channel, so therefore, I disagree with your suggestion for the channel model.

 

Finally, and arguably most importantly, per the minutes, I do see system vendors voting for Joel's model.  This is a very key point- they need to define the problem that needs to be solved.

 

Thanks!

John

 

 

 

-----Original Message-----
From: owner-stds-802-3-blade@listserv.ieee.org [mailto:owner-stds-802-3-blade@listserv.ieee.org] On Behalf Of Joe M Abler
Sent: Sunday, September 19, 2004 10:09 PM
To: STDS-802-3-BLADE@listserv.ieee.org
Subject: [BP] Follow-up to Friday's ad-hoc call - SDD21 curve

 


Joel,
As a follow-up to Friday's channel ad-hoc, I'd like to reiterate my concern with lowering the SDD21 curve beyond what's needed to hit the task force objectives.  Those objectives are of course to go across 40" of "improved FR4."  It was set primarily for cost and system configuration reasons and I absolutely agree with the objective.  From the July meeting it's been my understanding that the main area of concern with the existing curve is at frequencies below 3GHz.  There were 3 proposals from yourself, John D., and Steve A. to address this.  There were 2 straw polls taken, the first favoring Steve's proposal and the 2'nd favoring yours.

The primary advantage of your proposal is that it provides the most relief at the lower frequencies, but the primary disadvantage is that it pulls the curve down more than necessary at higher frequencies.  This is done to maintain the definition of the curve with a single equation, which I consider an aesthetic objective.  The main disadvantage however is that lowering the curve will increase transceiver power and potentially area (although power is clearly the main concern).  The attenuation is increased and the SNR relative to xtalk is decreased.  The changes won't break the bank, but they curves define a minimum criteria a transceiver must design to and it will increase power regardless of the signaling method chosen.  Power is just as significant a concern for system vendors as cost, but I'm concerned it's taken too much of a back seat in the channel analysis.  What's needed in a channel model is the capabi! lity to manufacture a cost effective system (with improved FR4 materials) under the assumption that well skilled engineering resources are applied to the design.  It would be desirable to pickup a reasonable percentage of legacy backplanes, but that's not a driving objective.  A system vendor has many levers to pull to stay above the SDD21 curve - trace length, trace width, via design, etc.  Many vendors, particularly early adopters, will go to even better materials to get additional margin above the line or to go additional distance.  However all channels above the line pay the power that is set by transceivers designed to the defined standards curve.  The curve should therefore be set to the minimum needed to meet the objective, without addition for margin, to capture legacy channels, or for aesthetics.  Transceiver designers can target higher limits, and offerings will certainly be available that do, but the standard should not go beyond the! minimum needed or it forces a cost/power tradeoff beyond what! 's neces sary.

I would propose the following options to obtain the additional margin at the lower frequencies with minimal impact to higher frequencies:

·         Use John's curve.  He withdrew his proposal, I'll reinstate it here.  His curve provides additional margin at the lower frequencies while maintaining a single equation to define the curve.

·         Use Steve's curve at the lower frequencies until it intersects the existing curve (somewhat above 3GHz), and use the existing curve above that point.  That provides slightly more relief at the lower frequencies than John's curve and also has a relatively smooth transition point between curves.

·         Take your new curve below 3GHz and the existing curve above that.  This results in the most relief at the lower frequencies but does have a step at the 3GHz point.  One could argue that the step isn't a serious problem because the curve is defined to include all discontinuities so real channels will not follow a smooth line along it anyway.  However, I consider that much of a step to be highly undesirable and fully expect most others will as well.



Thanks,        Joe


Joe Abler                                                             abler@us.ibm.com
IBM Microelectronics Division                          919-254-0573
Technical Marketing & HSS Applications    919-254-9616 (fax)
3039 Cornwallis Road                                                                
Research Triangle Park, NC  27709