Let me try my hand on why position matter. A normal passive channel is reciprocal. e.g. S12=3DS21 It only says the = off diagonal elements are symmetic. It doesn't say the diagonal elements = have to be equal. I believe this was the basis of Jeff Loyer's = discussion a while ago. The presence of the discontinuity affects the S11 and S22 dramatically = different based on whether it is close to the Tx or Rx. In the presences of imperfect loading on the Rx side, it is the = interaction between the S22 and loading that matters.=20 Thus position makes a difference. i.e. we are tuning the S22 with the = non-ideal loading. QED -----Original Message----- From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]On Behalf Of steve weir Sent: Tuesday, September 25, 2007 1:35 PM To: Jory McKinley Cc: Stephen Zinck; Scott McMorrow; leeritchey@xxxxxxxxxxxxx; npatel@xxxxxxxxxx; si-list@xxxxxxxxxxxxx Subject: [SI-LIST] Re: AC Coupled Signals Jory, I think this is good example of where intuitively appealing=20 misconceptions can seduce one into translating correlation into=20 causation. If you have more ringing in one case than another, it means=20 that you have set up a resonance that is more severe in the one case. =20 This can easily happen as a result of any number of things going on: =20 suboptimal silicon to package launch, suboptimal IC to PCB, via stubs,=20 connector transitions, etc, etc. The very simple test is to take a VNA, a couple of sections of coax and=20 a DC block. Move the DC block between the transmit end, the junction of = the two cables, and the receiver and look at the behavior of that net=20 channel. With good coax and connectors the channel performance will=20 change almost immeasureably. Now go and add a coax T on one side of the = DC block. Move that whole thing around as a unit and again the channel=20 performance remains the same. Add a second coax T on the other side of=20 the DC block from the first, and again move the whole thing around. The = results will still remain uniform. Now if you go and move one of those = T's someplace else, then the pesky mole you're trying to whack moves and = the resonance will pop up somewhere else. The bottom line is that it's=20 resonance that we need to fight and resonance doesn't know left from = right. Regards, Steve. Jory McKinley wrote: > To add to this, I will ask for release of lab data that I took that=20 > shows RX_EYE clearly improves as the AC cap/term location is moved=20 > closer to the RX. The data indicates that even though overall channel = > loss may not be affected, the 50ps edge rates we are sending through=20 > the channel are affected (in terms of time domain ringing) by the AC=20 > cap/term placement. This kind of feels right. > -Jory > =20 > > ----- Original Message ---- > From: steve weir <weirsi@xxxxxxxxxx> > To: Stephen Zinck <signalintegrity@xxxxxxxxxxx> > Cc: Scott McMorrow <scott@xxxxxxxxxxxxx>; jory_mckinley@xxxxxxxxx;=20 > leeritchey@xxxxxxxxxxxxx; npatel@xxxxxxxxxx; si-list@xxxxxxxxxxxxx > Sent: Tuesday, September 25, 2007 12:52:41 PM > Subject: Re: [SI-LIST] Re: AC Coupled Signals > > Steve, as far as I know where we have agreement that capacitor = location > can only affect performance where the combined capacitor and mount > presents a discontinuity and that discontinuity is located such that = it > forms a resonant structure with another discontinuity in the channel. = I > fail to see where we have moved any closer to supporting your premise > that locating a greater proportion of fixed loss before the capacitor > changes end to end loss than placing that same fixed loss behind it. > > As for lab measurements, we have these as we have characterized many > links. We also have extensive simulations. > > Regards, > > > Steve. > > Stephen Zinck wrote: > > Hi Steve, > > > > I understand your point but I actually thought Scott and I were > > getting close. I guess I still need him to explain his statement: > > "The only time position matters is in the face of discontinuities." > > because this runs counter to your assertion. > > > > It would be good to have some concrete lab measurement results to = back > > either of our points up. I am sorry I don't have any. > > > > We agree on TDR/VNA characteristics... > > > > Steve > > > > Stephen P. Zinck > > Interconnect Engineering Inc. > > P.O. Box 577 > > South Berwick, ME 03908 > > Phone - (207) 384-8280 > > Email - szinck@xxxxxxxxxxxxxxxxxxxxxxxxxxx > > Web - www.interconnectengineering.com=20 > <http://www.interconnectengineering.com> > > > > ----- Original Message ----- From: "steve weir" <weirsi@xxxxxxxxxx> > > To: <signalintegrity@xxxxxxxxxxx> > > Cc: "Scott McMorrow" <scott@xxxxxxxxxxxxx>; = <jory_mckinley@xxxxxxxxx>; > > <leeritchey@xxxxxxxxxxxxx>; <npatel@xxxxxxxxxx>; = <si-list@xxxxxxxxxxxxx> > > Sent: Tuesday, September 25, 2007 12:24 PM > > Subject: [SI-LIST] Re: AC Coupled Signals > > > > > >> Stephen, OK so when you say "lossy" or "nonlinear" you mean=3D20 > >> discontinuous. Discontinuities aggravate resonances based on > >> specific=3D20 > >> structure material and geometries, in other words the distance on = a=3D20 > >> centimeter or millimeter scale between discontinuities. We = have=3D20 > >> essentially the same opportunities for channel discontinuities at = and > >> in =3D > >> > >> the vicinity of the transmitter as the receiver. So I still do not > >> see=3D20 > >> a defensible basis for the offered position: that placing a = capacitor > >> at =3D > >> > >> one end of the line versus the other changes the end to end loss. =20 > >> What=3D20 > >> matters is if wherever I place one discontinuity that it sets up a > >> sharp =3D > >> > >> resonance with another discontinuity. That can happen equally well > >> at=3D20 > >> either end of the line. > >> > >> If one looks at a channel with only a TDR I might understand = the=3D20 > >> erroneous perception that placing a discontinuity down the line = is=3D20 > >> better than up front. But that is an illusion. TDR resolution=20 > falls=3D20 > >> with interconnect distance. This ia a result of the inherent loss=20 > of=3D20 > >> the interconnect that shelves bandwidth and hence resolution = versus=3D20 > >> distance for the instrument. This is one of the big limitations of = > a=3D20 > >> TDR for channel evaluation. A through measurement with a TDT or = VNA=3D20 > >> does not suffer that limitation, give true measure of S21 and so > >> report=3D20 > >> the real channel performance. Eric Bogatin spends some time on = the=3D20 > >> issue of bandwidth versus interconnect length in his book. > >> > >> Regards, > >> > >> > >> Steve. > >> Stephen Zinck wrote: > >>> Scott, > >>> We may have some nomenclature issues here...=3D20 > >>> > >>> When I say "lossy interface to the capacitor" I mean with = impedance > >>> dis=3D > >> continuities. So I think we are on a similar page given your = statement: > >>> > >>> "The only time position matters is in the face of = discontinuities." > >>> > >>> Again, most often, my role is to simulate the customers system at > >>> the 1=3D > >> 1th hour. I don't recommend this, I just work within the customer's > >> needs=3D > >> /requirements. I make real world recommendations from simulation > >> results =3D > >> for designs where these discontinuities you mention are a fact of > >> life. G=3D > >> ranted my customers are not doing 5+ Gbit/s designs (right now ;-). > >>> > >>> Above these data-rates, all you mention, capacitor transition = (pad, > >>> via=3D > >> s, etc) are of the utmost importance. And I would absolutely agree > >> that t=3D > >> he more perfect you make these transitions, the less it matters = where > >> the=3D > >> y are placed... > >>> > >>> So I do believe AC coupling capacitor position does matter, as you > >>> stat=3D > >> e, for the bulk of the designs occurring these days where component > >> footp=3D > >> rint and via optimization, etc. is NOT occurring... > >>> > >>> Steve > >>> > >>> Stephen P. Zinck > >>> Interconnect Engineering Inc. > >>> P.O. Box 577 > >>> South Berwick, ME 03908 > >>> Phone - (207) 384-8280 > >>> Email - szinck@xxxxxxxxxxxxxxxxxxxxxxxxxxx > >>> Web - www.interconnectengineering.com=20 > <http://www.interconnectengineering.com> > >>> > >>> ----- Original Message -----=3D20 > >>> From: Scott McMorrow=3D20 > >>> To: Stephen Zinck=3D20 > >>> Cc: jory_mckinley@xxxxxxxxx ; leeritchey@xxxxxxxxxxxxx ; > >>> npatel@micro=3D > >> n.com ; si-list@xxxxxxxxxxxxx=3D20 > >>> Sent: Tuesday, September 25, 2007 11:08 AM > >>> Subject: Re: [SI-LIST] Re: AC Coupled Signals > >>> > >>> > >>> Steven, > >>> > >>> I would not agree with your following statements. > >>> > >>> "I agree in theory with all you state. Assuming a lossless > >>> interface =3D > >> to the capacitor, it shouldn't matter where you place it, given a > >> purely =3D > >> linear system. But the real world is lossy, even when one makes = great > >> 3D =3D > >> solved structures. Manufacturing and other tolerances tend to take > >> the tr=3D > >> ek towards perfection to task." > >>> > >>> "Would either of you agree that AC coupling capacitor location = may > >>> ma=3D > >> tter with a lossy interface to the capacitor?" > >>> > >>> > >>> Insertion loss in a flat impedance linear lossy system will be > >>> indepe=3D > >> ndent of capacitor location. Run the math and see. The only time > >> positi=3D > >> on matters is in the face of discontinuities. In fact, given a = low > >> loss=3D > >> interconnect with discontinuities and a high loss interconnect with > >> disc=3D > >> ontinuities, the low loss system, with it's higher Q, will often = have > >> wor=3D > >> se behavior. > >>> > >>> An improperly designed 0402 capacitor transition for a 50 ohm = line > >>> ca=3D > >> n easily exhibit a discontinuity of 35 ohms for 50 ps. If attached > >> to po=3D > >> orly designed via transitions, the discontinuity will be even = worse. =20 > >> Whe=3D > >> n this is coupled closely to a high capacitance receiver input, a > >> high ca=3D > >> pacitance transmitter output, a low impedance via stub = discontinuity, > >> or =3D > >> a low impedance connector discontinuity, it can form a 1/2 wave > >> resonant =3D > >> circuit. This is most likely the problem you are seeing. =3D20 > >>> > >>> If the interconnect has essentially flat impedance, position = does > >>> not=3D > >> matter. If the capacitor transition is properly designed, position > >> does=3D > >> not matter. All of the data we have on this is proprietary at this > >> time=3D > >> =3D2E Our understanding of the physics has been verified by full = wave > >> mode=3D > >> ling, simulation and measurement. > >>> > >>> > >>> > >>> regards, > >>> > >>> Scott > >>> > >>> > >>> > >>> Scott McMorrow > >>> Teraspeed Consulting Group LLC > >>> 121 North River Drive > >>> Narragansett, RI 02882 > >>> (401) 284-1827 Business > >>> (401) 284-1840 Fax > >>> > >>> http://www.teraspeed.com > >>> > >>> Teraspeed=3DAE is the registered service mark of > >>> Teraspeed Consulting Group LLC > >>> > >>> > >>> Stephen Zinck wrote:=3D20 > >>> Hi Scott and Steve, > >>> > >>> To answer both of your questions, it is the resulting Hspice > >>> (with =3D > >> S-parameters) differential eye patterns, as viewed at the receiver > >> die, t=3D > >> hat were used to make a comparison of source versus destination AC > >> coupli=3D > >> ng capacitor locations. The system was excited with a string of = ones, > >> fol=3D > >> lowed by a single zero, followed by a string of ones.=3D20 > >>> > >>> I have not specifically designed a test board that varies the = AC > >>> co=3D > >> upling capacitor location along a trace. > >>> > >>> I understand the "shades of gray" here and agree that one = can't > >>> mak=3D > >> e a "rule of thumb" generalization in our line of work these = days.=3D20 > >>> > >>> I agree in theory with all you state. Assuming a lossless > >>> interface=3D > >> to the capacitor, it shouldn't matter where you place it, given a > >> purely=3D > >> linear system. But the real world is lossy, even when one makes = great > >> 3D=3D > >> solved structures. Manufacturing and other tolerances tend to take > >> the t=3D > >> rek towards perfection to task.=3D20 > >>> > >>> Do either of you have real world measured results, that you > >>> could s=3D > >> hare, that show no marked difference in received signal > >> characteristics w=3D > >> hen the AC coupling capacitor position is varied through a 30 inch > >> backpl=3D > >> ane system (or similar)? > >>> > >>> I believe my experience with capacitor location may prove true > >>> if t=3D > >> he capacitor interface is lossy (which is the case). A lot of my > >> customer=3D > >> s are just looking for quick ways to maximize performance using > >> standard =3D > >> component packages and standard layout practices (in the end, I = don't > >> lik=3D > >> e to give anything away that is low lying fruit). Most of the time = I > >> am d=3D > >> oing my analysis after the board is in layout, where I have limited > >> abili=3D > >> ty to change the design (unless it is really broken). In a perfect > >> world,=3D > >> where I am involved early, the package optimization and layout > >> structure=3D > >> s can be optimized as you state, but only if the margins warrant it > >> (syst=3D > >> em performance issues are expected after initial "what-if" > >> simulations ha=3D > >> ve occurred). The right tool for the right job rules the day... > >>> > >>> Would either of you agree that AC coupling capacitor location > >>> may m=3D > >> atter with a lossy interface to the capacitor? > >>> > >>> All the best, > >>> Steve > >>> > >>> Stephen P. Zinck > >>> Interconnect Engineering Inc. > >>> P.O. Box 577 > >>> South Berwick, ME 03908 > >>> Phone - (207) 384-8280 > >>> Email - szinck@xxxxxxxxxxxxxxxxxxxxxxxxxxx > >>> Web - www.interconnectengineering.com=20 > <http://www.interconnectengineering.com> > >>> > >>> ----- Original Message -----=3D20 > >>> From: Scott McMorrow=3D20 > >>> To: Stephen Zinck=3D20 > >>> Cc: jory_mckinley@xxxxxxxxx ; leeritchey@xxxxxxxxxxxxx ; > >>> npatel@m=3D > >> icron.com ; si-list@xxxxxxxxxxxxx=3D20 > >>> Sent: Tuesday, September 25, 2007 9:44 AM > >>> Subject: Re: [SI-LIST] Re: AC Coupled Signals > >>> > >>> > >>> Stephen > >>> > >>> Define "better" and then relate your simulations and > >>> conclusions =3D > >> to linear system theory and measurements. =3D20 > >>> > >>> I contend that the only difference an AC coupling capacitor > >>> can p=3D > >> ossibly have due to position in a linear interconnect is a result = of > >> impe=3D > >> dance mismatch. I contend that the capacitor will form a 1/2 wave > >> resona=3D > >> nt circuit with other interconnect discontinuities (connectors, = vias > >> stub=3D > >> s, packages, Tx die, Rx die ... etc) and that this interaction is > >> system,=3D > >> chip, connector and package design dependent. I contend that it is > >> this=3D > >> 1/2 resonance that can cause differences that can be measured, but > >> that =3D > >> there is no "rule of thumb", since the position and magnitude of > >> disconti=3D > >> nuities are different in every system. In some systems the = receiver > >> cons=3D > >> titutes a larger discontinuity than the transmitter. In other > >> systems th=3D > >> is is reversed. In yet other systems, connectors and vias = represent > >> larg=3D > >> er discontinuites than do either the transmitters or receivers. It > >> all "j=3D > >> ust depends". To state a specific rule is just plain incorrect. > >>> > >>> I contend that once you remove the magic and myths = surrounding > >>> AC=3D > >> coupling capacitors, analysis of the 3D structure shows that by > >> reducing=3D > >> the signal path discontinuity through the capacitor, you will > >> necessaril=3D > >> y improve performance. An AC coupling capacitor, with it's > >> associated vi=3D > >> a and pad transition design, can be viewed as a black box which has > >> inser=3D > >> tion loss and return loss, and can be modeled quite well using = either > >> lum=3D > >> ped element approximations or (my favorite) S-parameters. As such = it > >> wil=3D > >> l cascade in a simulation model just like any other linear element. = =20 > >> If w=3D > >> e start with a system with flat 50 ohm impedance from end to end, = it > >> can =3D > >> be easily shown that no matter what the position of the capacitor > >> along t=3D > >> he interconnect is, the insertion loss of the system is identical. = > >> It is=3D > >> only the return loss, as seen from each end that changes. > >>> > >>> I've been designing AC coupling capacitor mounting = transitions > >>> pr=3D > >> operly for quite a few years now and have some 0402 designs that = keep > >> S12=3D > >> above -0.2 dB up to 7.5 GHz, S12 below -20 dB @ 5 GHz, and below = -15 > >> dB =3D > >> @ 10 GHz. For all practical purposes, these designs are = transparent > >> and =3D > >> may be placed anywhere in an interconnect design where there is > >> space, si=3D > >> nce there is little resonance interaction with other devices and > >> structur=3D > >> es. > >>> > >>> > >>> Scott > >>> > >>> > >>> Scott McMorrow > >>> Teraspeed Consulting Group LLC > >>> 121 North River Drive > >>> Narragansett, RI 02882 > >>> (401) 284-1827 Business > >>> (401) 284-1840 Fax > >>> > >>> http://www.teraspeed.com > >>> > >>> Teraspeed=3DAE is the registered service mark of > >>> Teraspeed Consulting Group LLC > >>> =3D20 > >>> > >>> Stephen Zinck wrote:=3D20 > >>> Hi Scott, > >>> > >>> My simulations show that the capacitor is best placed at = the > >>> re=3D > >> ceiver end of the transmission-line. Do you disagree? If so, why? > >>> > >>> Steve > >>> > >>> Stephen P. Zinck > >>> Interconnect Engineering Inc. > >>> P.O. Box 577 > >>> South Berwick, ME 03908 > >>> Phone - (207) 384-8280 > >>> Email - szinck@xxxxxxxxxxxxxxxxxxxxxxxxxxx > >>> Web - www.interconnectengineering.com=20 > <http://www.interconnectengineering.com> > >>> > >>> ----- Original Message -----=3D20 > >>> From: Scott McMorrow=3D20 > >>> To: signalintegrity@xxxxxxxxxxx=3D20 > >>> Cc: jory_mckinley@xxxxxxxxx ; leeritchey@xxxxxxxxxxxxx ; > >>> npat=3D > >> el@xxxxxxxxxx ; si-list@xxxxxxxxxxxxx=3D20 > >>> Sent: Tuesday, September 25, 2007 8:30 AM > >>> Subject: Re: [SI-LIST] Re: AC Coupled Signals > >>> > >>> > >>> Stephen, > >>> > >>> I'm sorry, this is a linear system. Except for possible > >>> reso=3D > >> nances that are created by discontinuities and modal conversion > >> (which ha=3D > >> ve absolutely zero to do with signal rise time), there is no > >> difference i=3D > >> n the attenuation of a capacitor placed at the Tx as opposed at = the > >> Rx. =3D > >> W.R.T. the receiver, if it is "lost in the rise-time degradation of > >> the =3D > >> system", it will be lost wherever it is placed. > >>> > >>> > >>> > >>> Scott McMorrow > >>> Teraspeed Consulting Group LLC > >>> 121 North River Drive > >>> Narragansett, RI 02882 > >>> (401) 284-1827 Business > >>> (401) 284-1840 Fax > >>> > >>> http://www.teraspeed.com > >>> > >>> Teraspeed=3DAE is the registered service mark of > >>> Teraspeed Consulting Group LLC > >>> =3D20 > >>> > >>> Stephen Zinck wrote:=3D20 > >>> Hi Jory, > >>> > >>> I have simulated this at length and concur with your experience = that > >>> th=3D > >> e=3D20 > >>> capacitor is best placed at the receiver... > >>> > >>> In effect, the attenuation associated with the capacitor placement > >>> at t=3D > >> he=3D20 > >>> receiver (parasitics/pads/vias) is lost in the rise-time = degradation > >>> of=3D > >> the=3D20 > >>> system. > >>> The classic "don't break it until you have to" rule is = applicable... > >>> OK=3D > >> this=3D20 > >>> is my rule... :-) > >>> > >>> All the best, > >>> Steve > >>> > >>> Stephen P. Zinck > >>> Interconnect Engineering Inc. > >>> P.O. Box 577 > >>> South Berwick, ME 03908 > >>> Phone - (207) 384-8280 > >>> Email - szinck@xxxxxxxxxxxxxxxxxxxxxxxxxxx > >>> Web - www.interconnectengineering.com=20 > <http://www.interconnectengineering.com> > >>> > >>> ----- Original Message -----=3D20 > >>> From: "Jory McKinley" <jory_mckinley@xxxxxxxxx> > >>> To: <leeritchey@xxxxxxxxxxxxx>; <npatel@xxxxxxxxxx>; > >>> <si-list@freelists=3D > >> =3D2Eorg> > >>> Sent: Monday, September 24, 2007 5:31 PM > >>> Subject: [SI-LIST] Re: AC Coupled Signals > >>> > >>> > >>> I will elaborate a bit on what I have seen. I have measured = (time > >>> dom=3D > >> ain)=3D20 > >>> in the lab some effects that appears to be location specific in = the=3D20 > >>> placement of the AC coupling caps at the rcvr. Now this may be = due > >>> in =3D > >> part=3D20 > >>> to the fact that I am using 50-ohm resistor termination in each = lead > >>> as=3D > >> =3D20 > >>> well and the combination (cap plus rcvr reflection) is giving = some=3D20 > >>> imbalance depending on distance. The best rcvr eye that I am = seeing > >>> is=3D > >> =3D20 > >>> when I can move the AC/term as close to the rcvr as I can. By the > >>> way =3D > >> > >>> these are 5Gb/s signals. > >>> If I have time I will try and isolate what I am seeing and even > >>> simulat=3D > >> e=3D20 > >>> it, has anyone else seen or simulated this? > >>> -Jory > >>> > >>> ----- Original Message ---- > >>> From: Lee Ritchey <leeritchey@xxxxxxxxxxxxx> > >>> To: "npatel@xxxxxxxxxx" <npatel@xxxxxxxxxx>; si-list@xxxxxxxxxxxxx > >>> Sent: Monday, September 24, 2007 1:06:06 PM > >>> Subject: [SI-LIST] Re: AC Coupled Signals > >>> > >>> Nikil, > >>> > >>> I have made measurements on test PCBs and the location is not all = that > >>> important. In identical pairs, one with AC coupling capacitors = and > >>> the=3D > >> > >>> other without, the loss vs. frequency is virtually identical at = leas > >>> ou=3D > >> t=3D20 > >>> to > >>> 6 GHz. That would be 12 Mb/S. > >>> > >>> Lee Ritchey > >>> > >>> > >>> [Original Message] > >>> From: <npatel@xxxxxxxxxx> > >>> To: <si-list@xxxxxxxxxxxxx> > >>> Date: 9/24/2007 10:21:37 AM > >>> Subject: [SI-LIST] AC Coupled Signals > >>> > >>> Hi all, > >>> In case of AC coupled signals does anyone know of an optimum = placement > >>> for the caps? I mean should they be placed near the source, = receiver, > >>> middle of the transmission line? > >>> How much difference does it make in the opening of the eye? > >>> The signals are differential CML running at 3.0Gbps > >>> > >>> Thanks, > >>> Nikhil > >>> > >>> ------------------------------------------------------------------ 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 technical documents are available at: http://www.si-list.net 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