Overcoming the SI Challenges in Designing Gb/s Backplane Channels

Transcription

1 Penn Sae Signal Inegriy Symposium Overcoming he SI Challenges in Designing b/s Backplane Channels Sephen B. Smih Madhumiha Rengarajan FCI USA LLC April 5 h 2014

2 Ouline of Presenaion Objecive of paper The connecor Discussion of resonance Daugher cards The firs backplane The second backplane Conclusions / Summary

3 The Objecive Design and build a passive backplane sysem ha successfully operaes a a daa rae of 25 b/s Doing i on he firs ry would be nice Wha defines success? Meeing he Opical Inerneworking Forum s CEI-25-LR Specificaion Meeing he IEEE 802.3bj Commiee s crierion Demonsraing success wih acual SerDes chips Figure ou how o eend his wisdom o daa raes higher han 25 b/s

4 The Connecor Required Feaures The EaMAX Connecor Ecellen Z 0 R.L. I.L. PSXT and mode conversion Zero in-pair skew Absence of resonance

5 Types of Resonance Sub resonance Typically avoided in pc boards by backdrilling Can occur in connecor separable maing inerfaces round-mode resonance Occurs when here are muliple grounds / reurn pahs Mied Combinaion of he wo Happens when here is a sub presen on a ground/reurn pah Migh behave like a sub and/or a ground-mode resonance

6 Sub Resonance Tradiional connecor inerface shown above lef has wo sub which resonae as shown in red. Inerface making wo poins of conac shown above righ does no resonae below 30 Hz.

7 Effecs of round-mode Resonance Ugly Inserion Loss

8 Effecs of round-mode Resonance Even Uglier Crossalk

9 Primer on round-mode Resonance Modal Analysis: S S

10 Primer on round-mode Resonance Modal Analysis: S S Time-Domain Response of Muliconducor Transmission Lines Anonije R. Djordjevic Tapan K. Sarkar Senior Member IEEE and Roger F. Harringon Fellow IEEE PROCEEDINS OF THE IEEE VOL. 75 NO. 6 JUNE 1987

11 Primer on round-mode Resonance Modal Analysis: Time-Domain Response of Muliconducor Transmission Lines Anonije R. Djordjevic Tapan K. Sarkar Senior Member IEEE and Roger F. Harringon Fellow IEEE PROCEEDINS OF THE IEEE VOL. 75 NO. 6 JUNE 1987 S S D V B j V d I d D I L j I R d V d v B v i i L i R v 0 0

12 Primer on round-mode Resonance Modal Analysis: Time-Domain Response of Muliconducor Transmission Lines Anonije R. Djordjevic Tapan K. Sarkar Senior Member IEEE and Roger F. Harringon Fellow IEEE PROCEEDINS OF THE IEEE VOL. 75 NO. 6 JUNE 1987 S S [Z = [R + j[l [Y = [ + j[b D V B j V d I d D I L j I R d V d v B v i i L i R v 0 0

13 Primer on round-mode Resonance Modal Analysis: Time-Domain Response of Muliconducor Transmission Lines Anonije R. Djordjevic Tapan K. Sarkar Senior Member IEEE and Roger F. Harringon Fellow IEEE PROCEEDINS OF THE IEEE VOL. 75 NO. 6 JUNE 1987 S S [Z = [R + j[l [Y = [ + j[b de { m 2 [U - [Z [Y} = 0 de { m 2 [U - [Y [Z} = 0 D V B j V d I d D I L j I R d V d v B v i i L i R v 0 0

14 Primer on round-mode Resonance Modal Analysis: Time-Domain Response of Muliconducor Transmission Lines Anonije R. Djordjevic Tapan K. Sarkar Senior Member IEEE and Roger F. Harringon Fellow IEEE PROCEEDINS OF THE IEEE VOL. 75 NO. 6 JUNE 1987 S S [Z = [R + j[l [Y = [ + j[b de { m 2 [U - [Z [Y} = 0 de { m 2 [U - [Y [Z} = 0 [ S I = [Z -1 [S v [ D V B j V d I d D I L j I R d V d v B v i i L i R v 0 0

15 Primer on round-mode Resonance Modal Analysis: Time-Domain Response of Muliconducor Transmission Lines Anonije R. Djordjevic Tapan K. Sarkar Senior Member IEEE and Roger F. Harringon Fellow IEEE PROCEEDINS OF THE IEEE VOL. 75 NO. 6 JUNE 1987 S S [Z = [R + j[l [Y = [ + j[b de { m 2 [U - [Z [Y} = 0 de { m 2 [U - [Y [Z} = 0 [ S I = [Z -1 [S v [ D V B j V d I d D I L j I R d V d v B v i i L i R v 0 0 Moral of he sory: All grounds are no creaed equal.

16 Why Muliple Resonan Frequencies? L g

17 Why Muliple Resonan Frequencies? L g Poin of curren origin

18 Why Muliple Resonan Frequencies? L g Toal reurn pah elecrical lengh is wha deermines he frequency of resonance. Poin of curren origin

19 Why Muliple Resonan Frequencies? L g L 1 Toal reurn pah elecrical lengh is wha deermines he frequency of resonance. L 1 In his case L 1 +L g +L 1 Poin of curren origin

20 Why Muliple Resonan Frequencies? L g L 1 L 2 Toal reurn pah elecrical lengh is wha deermines he frequency of resonance. L 1 Poin of curren origin L 2 L g In his case L 1 +L g +L 1 or L 2 +L g +L 2

21 Minimizing round-mode Resonance When possible use ground planes raher han discree ground pins. Try o make he inducance of all discree ground/reurn-pahs similar if no equal. Preven inducance pinch poins i.e. paricularly high-inducive ground pahs.

22 In-Pair Skew Prevenion no cu wih cu air pocke beween signals and ground plane is offse o minimize skew signal pair

23 EON Sizes and Fooprin Dimensions round Press-fi Conac from AirMa VS Lengh = 1.60 mm Drill diameer = 0.60 mm Finish diameer = 0.50 mm Signal Press-fi Conac Lengh = 1.40 mm Drill diameer = 0.45 mm Finish diameer = 0.36 mm

24 EON Sizes and Fooprin Dimensions

25 Daugher Cards Single-ended 50-ohm rouing o SMA connecors 16 channels 32 races Eplici effor made o minimize race crossalk Megron-6 subsrae minimize loss Board hickness = 0.1 inch Backdrilling done such ha maimum sub lengh was 6 ±3 mils

26 Daugher Card Sack Up Traces: Widh = 9.8 mils Lengh = 5 inches Signals roued on layers 7 and 10 Unused connecor signal pins roued on layer 12 o 50- ohm erminaions

27 Wiring Assignmen N M L K J I H F E D C B A TER X TER TER X TER TER X TER TER X TER TER X 6 X TX8- TX8+ X TX7- TX7+ X RX8- RX8+ X RX7- RX7+ X TER 5 TER X TX6- TX6+ X RX5- RX5+ X RX6- RX6+ X TX5- TX5+ X 4 X TX4- TX4+ X TX3- TX3+ X RX4- RX4+ X RX3- RX3+ X TER 3 TER X TX2- TX2+ X RX1- RX1+ X RX2- RX2+ X TX1- TX1+ X 2 X TER TER X TER TER X TER TER X TER TER X TER 1

28 The Firs Backplane Differenial 100-ohm rouing Megron-6 subsrae minimize loss HVLP Cu roughness Board hickness = 0.23 inch Backdrilling done such ha maimum sub lengh was 6 ±3 mils same as daugher cards

29 Backplane Sack Up Traces: Widh and separaion: 10/11/10 mils 7.5/5.5/7.5 in fooprin Lengh = 17 inches OIF Signals roued on layers and 11 Dielecric layers were fairly hick 10 mils

30 Anipad Design Fairly large anipad designed o minimize impedance capaciive mismach in fooprin Diving boards implemened underneah races in anipad o reduce ground sarvaion

31 Summary of OIF Requiremens CEI-25-LR Specificaion places limis on channel Inserion Loss I.L. Inserion Loss Deviaion ILD ILD RMS mus be <0.3 Reurn Loss R.L. Inegraed Crossalk Noise ICN Ploed agains Inserion Loss a Nyquis frequency 12.9 Hz for 25 b/s Maimum channel lengh = 27 inches 686 mm

32 Channel Simulaion Resuls

33 The Backplane Sysem

34 Failure of Measuremens One quarer of he channels failed he ICN requiremen

35 Debugging of Simulaion Simulaion consised of obaining and cascading ogeher S-parameer Touchsone models of each of he channel componens Two ses of daugher card races Two ses of daugher card connecor fooprins Two connecors Two ses of backplane connecor fooprins One se of backplane races

36 Diagram of Link Daugher card races Daugher card fooprins Connecors Backplane races Backplane fooprins

37 Eplanaion of Componen Models Traces were frequency-dependen RLC models Connecors and fooprins models from CST MWS

38 Eplanaion of Misake The far-side backplane fooprin model was conneced backwards i.e. miswired plane prin Backplane Traces Backplane Fooprin Connecor Daug Foop

39 Correced Simulaion The correced simulaion failed jus like he measuremens. Hooray!?!

40 Failure Mechanism N M L K J I H F E D C B A 6 TER X TER TER X TER TER X TER TER X TER TER X 5 X TX8- TX8+ X TX7- TX7+ X RX8- RX8+ X RX7- RX7+ X TER 4 TER X TX6- TX6+ X RX5- RX5+ X RX6- RX6+ X TX5- TX5+ X 3 X TX4- TX4+ X TX3- TX3+ X RX4- RX4+ X RX3- RX3+ X TER 2 TER X TX2- TX2+ X RX1- RX1+ X RX2- RX2+ X TX1- TX1+ X 1 X TER TER X TER TER X TER TER X TER TER X TER TX4 TX3 RX4 Traces roued on signal layer 1 Traces roued on signal layer 2 Traces roued on signal layer 3 Traces roued on signal layer 4 Layer-o-layer crossalk in he fooprin

41 Ancillary Fi o Daugher Cards Asymmerical wiring around signal vias caused problems wih ILD.

42 The Second Backplane Mainaining 6-mil gap reduced beween-layer XT Zero skew was implemened in he FP rouing

43 The Second Backplane Anipad size was reduced in lower layers where backdrilling was done. Ye more beween-layer crossalk reducion

44 The Second Backplane

45 Compliance wih OIF Specificaion

46 Compliance wih IEEE Specificaion The backplane sysem included 16 channels of lengh 1 meer o es agains he requiremens of he IEEE 802.3bj Pos Draf 2.1v2 release Channel Operaing Margin COM Tool. Addiionally i worked error-free wih 25 b/s SerDes from hree differen chip manufacurers.

47 Unepeced Low-Frequency Crossalk Low-frequency crossalk ploed in red occurring hroughou he enire rouing of he backplane for some of he channels

48 Unepeced Low-Frequency Crossalk a b Reduce magneic coupling beween adjacen signal layers by: Increasing he hickness of he separaing ground plane o a leas 1-oz. Cu Saggering races on adjacen layers such ha hey are no in perfec regisraion wih each oher

49 Erraa In he paper he minimum margin requiremen for he IEEE 802.3bj Channel Operaing Margin COM ool when using PAM4 modulaion is lised as being 5 db. This is incorrec. The correc minimum margin requiremen is 3 db i.e. he same as for NRZ modulaion.

50 Beyond 25 b/s s s g g g g g g s s mils mm race widh race-o-race separaion race-o-anipad separaion signal via drill Ø signal via finish Ø signal via pad Ø small ground via drill Ø small ground via finish Ø large ground via drill Ø large ground via finish Ø anipad widh row pich 1.20 column pich 2.80

51 Beyond 25 b/s s s g g g s s mils mm race widh race-o-race separaion race-o-anipad separaion signal & small ground via drill Ø signal & small ground via finish Ø signal via pad Ø large ground via drill Ø large ground via finish Ø anipad widh row pich 1.20 column pich 2.80

52 Resonance Reappeared Resonance no seen in XT

53 Cause of New Resonance There was no a ground layer on he op of he new daugher cards. All of he ground vias formed lile subs. Fied by conducive pain on he surface of he board Lile ground subs

54 a b Beyond 25 b/s 40 b/s Simulaion of fooprin wih larger column pich Power-sum NEXT PAIR 6 Power-sum FEXT PAIR 6 Significan NEXT improvemen Significan FEXT improvemen

55 Beyond 25 b/s 40 b/s

56 Conclusions / Summary To build a channel ha works a 25+ b/s: Use an appropriae connecor ha has Ecellen Z 0 R.L. I.L. crossalk and mode conversion Zero skew Absence of resonance Design a connecor fooprin ha has Feaures o reduce ground sarvaion Zero skew rouing Adequae separaion beween race and anipad edges Smaller anipads for ground layers ha are backdrilled away

57 Conclusions / Summary con. To build a channel ha works a 25+ b/s: Design a backplane ha Implemens he aforemenioned connecor fooprin feaures Mees he required loss characerisics Prevens magneic coupling beween differenial races on adjacen signal layers Avoid having subs in boh connecor and all boards

58 References D. M. Pozar Microwave Engineering 2nd ed. John Wiley & Sons Inc.1998 pp S. Sercu and L. Marens Accurae de-embedding of he conribuion of he es boards o he highfrequency characerisics of backplane connecor Proc. of he 6h Topical Meeing on Elecrical Performance of Elecronic Packages San Jose Ca. Ocober pp S. Sercu and L. Marens Characerizing N-por packages and inerconnecions wih a 2-por nework analyzer Proceedings of he 6h Topical Meeing on Elecrical Performance of Elecronic Packages San Jose CA. Ocober pp S. Sercu and L. Marens N-por characerisaion echniques wih applicaions o muli-por connecors an IC-packages.28h EuMC Workshop Proceedings 1s EuMW Amserdam Ocober pp OIF Common Elecrical Inerface 25 b/s Long Reach CEI-25-LR Specificaion Firs Ediion Opical Inerneworking Forum CEI/IEC Hard Meric Specificaion Firs Ediion Inernaional Elecroechnical Commiee H. Johnson and M. raham High-Speed Digial Design A Handbook of Black Magic Prenice Hall PTR 1993 pp V. Balasubramanian S. Smih and S. Agili Comparison of S-Parameer Concaenaion o Full-Wave Simulaion for High-Speed Inerconnec Analysis Proceedings of DesignCon 2007 San Jose CA January 29 February