[SI-LIST] Re: Diff.Pairs

  • From: "Scott McMorrow" <scott@xxxxxxxxxxxxx>
  • Date: Sun, 19 Oct 2003 19:58:53 -0700

Lee,
In the din of EMI discussions, the thread on differential pairs seems to 
have been disregarded.  I'm still waiting for you to clarify the 
following statement which you made:

    "It is time to stop representing differential signals as needing to be
    tightly coupled to each other in order to operate properly. It is simply
    not so. I have routed thousands of differential signal where each member
    of the pair is on a different layer. If this were not possible, 1 mm
    pitch BGAs with differential signals would be un routable. There are
    tens of
    thousands of such parts being shipped every month on PCBs where they are
    routed apart from each other. "

I am especially interested in the claim that differential pairs may be routed 
as uncoupled traces on different layers.  The unstated assumption is that this 
does not make any difference in the signal quality, noise profile of the board, 
or EMI.  I've shown through full wave electromagnetic simulations that there is 
significant noise injection from single ended vias into any reasonable plane, 
if ground vias for return path containment are not placed sufficiently close 
(<= 150 mils.)  With differential pair via transitions, a return path ground is 
not necessary to contain the differential mode energy.  One is absolutely 
required for single ended transition, if high frequency power system noise and 
edge degradation is to be kept to a minimum.
 
The claim that "tens of thousands of such parts are being shipped every month 
..." is not a proof.  I once analyzed a design for a very large processor 
company which showed conclusively that their currently shipped design would not 
work in the worst case, and in fact was an extremely sensitive and "twitchy" 
design.  The response of their management was that my analysis must be wrong, 
because they had shipped hundreds of thousands of PCB's containing this 
particular chipset and bus.  Six months later that company announced a total 
recall of all products that were designed with that chipset.  This was due to 
resonance conditions which my analysis had accurately detected.

So, I ask you to respond to the following question: 

    Do you truly believe that separating differential pairs into
    uncoupled single ended traces on separate layers, through single
    ended vias, with zero delay end-to-end delay skew,  is a good and
    acceptable design practice for all designs in this universe? 

If so, please explain?  If not, please state the necessary and 
sufficient requirements in order to validate such a design in all cases, 
and any assumptions that must be met.

regards,

scott


Scott McMorrow wrote:

>mike
>
>Lee does not appear to be talking about broadside.  If he was, and had 
>said this explicitly, I would have no problem.  Broadside is a bit tough 
>to control, but it is a quite valid approach to differential signal 
>routing.  But it is almost always highly coupled.
>
>If you look at the context of his posting, he is advocating that the 
>individual lines of a diff pair do not need to be routed as coupled 
>traces.  He advocates decoupling them to reduce losses.  Then he follows 
>with an admonition that the not only can they be uncoupled, but that 
>they can also be routed on different layers. It would be impossible to 
>route broadside differential pairs that were uncoupled, unless the 
>layers were widely spaced apart, which would be terribly space inefficient.
>
>Lee, can you comment on the quote which I snipped from your posting below?
>
>
>If I have misinterpreted the intention of this posting, I do apologize.
>
>
>scott
>
>
>Michael Chin wrote:
>
>  
>
>>Scott,
>>
>>I've followed your emails exchanges with Mike Brown
>>closely.  This subject is a very important item as we
>>tackle high speed design using differential pairs.
>>Today, we have differential pair in CMOS logic,
>>LVDS, and differential HSTL (RLDRAM).
>>
>>Both of you have argued with very good points and I
>>appreciated your thoughts.  But in this case, I would
>>side with what Lee has stated.  He may not have explicitly
>>stated the stackup in his assumption, but it is commonly
>>referred as the the broad side coupled differential pair
>>vs. edge coupled stripline differential pair.
>>
>>There is no question about the benefits and level of
>>control from edge coupled differential where both tracks
>>are routed on the same layer.  But it is also very common
>>to have a dual stripline layers between two Ground layers
>>for broadside coupling.  This is mostly used in the backplane
>>side and have demonstrated success in many designs.
>>
>>An example of this stackup can be:
>>
>>=============   Ground plane for image return
>>                Core, thickness H
>>    =====       Diff Layer #1
>>                Prepreg
>>    =====       Diff Layer #2
>>                Core, thickness H
>>=============   Ground plane for image return
>>
>>I have been to a lot of seminar where SI experts argued
>>the Pro/Con of edge coupled vs. broadside coupled differential
>>pair. I have seen good results from both styles on frequency
>>under 3GHz.
>>
>>Just my two cents,
>>Michael Chin
>>
>>Scott McMorrow wrote:
>>
>>    
>>
>>>Mike
>>>That is my point.  This is an example of routing a differential pair 
>>>with matched skew, but on different layers. I do not recommend this 
>>>routing  at all.  It is, however an example of what Lee has advocated 
>>>in his post, which I quote here:
>>>
>>>"It is time to stop representing differential signals as needing to be
>>>tightly coupled to each other in order to operate properly.  It is 
>>>simply
>>>not so.  I have routed thousands of differential signal where each 
>>>member
>>>of the pair is on a different layer.  If this were not possible, 1 mm 
>>>pitch BGAs with differential signals would be un routable.  There are 
>>>tens of
>>>thousands of such parts being shipped every month on PCBs where they are
>>>routed apart from each other. "
>>>
>>> 
>>>This is an extremely bad practice and bad advice and should be shunned..
>>>
>>>
>>>regards,
>>>
>>>scott
>>>
>>>Mike Brown wrote:
>>>
>>>
>>>      
>>>
>>>>Scott, the stackup shown has image current flowing in all 4 Pwr/Gnd 
>>>>planes for any signal transition, differential or common mode.  The 
>>>>only place where they cancel is at the driver's power/ground pins.  
>>>>Noise on any of the planes will cause current flow in that plane 
>>>>pair; so to the extent that current in the plane can be considered 
>>>>common mode, you are correct.  I just do not see this stackup as 
>>>>being in any sense differential.
>>>>That it can be used to connect a differential driver to a 
>>>>differential receiver is true.  That it is in any way electrically 
>>>>optimal is false.  That it may be the only route available may or 
>>>>may not be true.  The schedule would have to be extremely pressing 
>>>>for me to accept a board routed in this manner.
>>>>
>>>>Mike
>>>>
>>>>Mike
>>>>
>>>>Scott McMorrow wrote:
>>>>
>>>>        
>>>>
>>>>>Mike
>>>>>
>>>>>see below
>>>>>
>>>>>
>>>>>          
>>>>>
>>>>>>Scott,
>>>>>>how did the difference  between the noise voltages get to be 
>>>>>>common mode?  Given identical noise voltages on both traces, the 
>>>>>>difference is zero - and that is differential noise.  The CM noise 
>>>>>>is the average of the two noise voltages (sum/2)     
>>>>>>            
>>>>>>
>>>>>I am talking about a case where Lee has suggested that the 
>>>>>differential pair can be split between two different routing 
>>>>>layers.  The stackup might be as follows:
>>>>>
>>>>>Ground
>>>>>
>>>>>diff layer 1
>>>>>
>>>>>Power 1
>>>>>
>>>>>Ground
>>>>>
>>>>>diff Layer 2
>>>>>
>>>>>Power 2
>>>>>
>>>>>
>>>>>The noise on each of these layers is totally different.  Thus at 
>>>>>any point along the differential pair there is a common mode 
>>>>>voltage differential that occurs, causing a net current flow in one 
>>>>>direction or the other.  You are correct for signals that are 
>>>>>referenced to the same layer.
>>>>>
>>>>>Oh, and your point is well taken.  -1.5 dB of loss would be 1.5 dB 
>>>>>of gain.  Oops!
>>>>>
>>>>>
>>>>>regards,
>>>>>
>>>>>scott
>>>>>          
>>>>>
>
>  
>

-- 
Scott McMorrow
Teraspeed Consulting Group LLC
2926 SE Yamhill St.
Portland, OR 97214
(503) 239-5536
http://www.teraspeed.com




------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field

List archives are viewable at:     
                //www.freelists.org/archives/si-list
or at our remote archives:
                http://groups.yahoo.com/group/si-list/messages 
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu
  

Other related posts: