H19- Reliable Serial Backplane Data Transmission at 10 Gb/s. January 30, 2002 Slide 1 of 24

Transcription

1 H19- Reliable Serial Backplane Data Transmission at 10 Gb/s Slide 1 of 24

2 Evolution of the Interconnect F r e q u e n c y A c t i v e Channel Architecture Connectors Transmission Media Loss Properties Length Construction Board Interface Higher Performance Media Alternate Drilling Semiconductors Pre-emphasis Equalization Encoding Techniques Board Interface Higher Performance Media Alternate Drilling P a s s i v e Active P a s s i v e Length Slide 2 of 24

3 Transmission over Media Only FR4 ROGERS 4350 GETEK ARLON CLTE FR4: Jitter = 0.30 UI Opening = 238 mv GETEK: Jitter = 0.28 UI Opening = 268 mv ROGERS 4350: Jitter = 0.20 UI Opening = 426 mv ARLON CLTE: Jitter = 0.19 UI Opening = 520 mv -The output waveforms shown result from a 1-volt, 32-bit inverting K28.5 input bit pattern (10 Gbps, 60ps edges) that is applied to a 12 mil, 50 Ohm stripline trace that is 18 long. Slide 3 of 24

4 Transmission over Media Plus Two Connectors Slide 4 of 24

5 The Connector / Board Interface connector pin layer connection TOP layer connection BOTTOM layer connection Slide 5 of 24

6 Z-PACK HM-Zd Evaluation System System Environment using FR-4 (Nelco ) Backplane Line Card Different system lengths - 10, 22 and 30 Various trace widths No counterboring Slide 6 of 24

7 Performance Comparison Bottom Layer SDD21 (db) Daughtercard Trace Width Line Card Trace Width E E E+10 FREQ (Hz) 10 Inches 22 Inches 30 Inches Slide 7 of 24

8 Performance Comparison 30 Inches SDD21 (db) Daughtercard Trace Width Line Card Trace Width E E E+10 FREQ (Hz) Top Bottom Near-top Near-bottom Slide 8 of 24

9 0 Trace Length is NOT the Only Issue SDD21 (db) Daughtercard Trace Width Line Card Trace Width E E E+10 FREQ (Hz) Bottom Layer - 30 Inches Top Layer - 22 Inches Slide 9 of 24

10 Cable Insertion Loss (With/Without Connectors) SDD21 (db) Slide 10 of 24

11 Resonance Test System Similar to HM-Zd Evaluation System Backplane traces replaced with low loss cable Minimize loss in system Investigation of resonance effects Same daughtercard as HM-Zd Evaluation System Slide 11 of 24

12 Reducing The Top Layer Effect SDD21 (db) Driven by connector characteristics and design. No counterboring done on backplane. 1.0E E E+10 FREQ (Hz) Non-optimized Optimized Slide 12 of 24

13 Channel Conclusions Easy to see channel optimization in frequency domain Layer connection determines performance Top unpredictable Bottom predictable Magnitude of discontinuity effect increases with frequency Unpredictable losses at higher frequencies can negate improvements in channel performance Shorter system lengths media selection Optimization of unpredictable losses, i.e. top layer effect, appears possible Improvement of predictable losses gained through better transmission media selection Slide 13 of 24

14 Channel Attenuation Source Equalizer C ( f, l) = e k l (1+ s j ) f k d lf Belden 1694A Coaxial Cable 25m 50m 75m 100m 125m 150m 175m Slide 14 of 24

15 Adaptation Algorithm Signal transmitted with well-defined amplitude spectral content at specific frequency allows estimation of attenuation Attenuation estimate used to control inverse transfer function magnitude shape is fixed Same methodology can be applied to traces Slide 15 of 24

16 Adaptive Equalizer Advantages Can apply a lot of gain Have applied up to 30 db Wide range of attenuation characteristics with no adjustment Can handle wide range of data rates with no adjustment Fast adaptation with no training sequence Reliable convergence even with no eye present Slide 16 of 24

17 Post-equalizer Additive Jitter Sources Errors in adaptation Difference between trace attenuation characteristics and device frequency gain Above two sources can be reduced to 0.2UI additive jitter Ripples in channel attenuation due to reflections can be more serious Slide 17 of 24

18 Increased Jitter Due to Channel Loss Ripple Slide 18 of 24

19 Conclusions Equalizer will flatten channel ripple effects are independent of trace length up to equalizer maximum Generally ripple in frequency range below half the data rate is most significant ripple above fundamental frequency will have smaller effect Implication Good backplane design is required for equalizer work to maximum potential Slide 19 of 24

20 Z-PACK HM-Zd Evaluation System Slide 20 of 24

21 10.7 Gb/s Eye Diagrams after GN2001 Trace-only - 25 Bottom-Layer Connection - 22 Top-Layer Connection - 22 Slide 21 of 24

22 Summary of Testing Length Layer Section 1 Section 2 Section 3 Connection (8/12)* (5/12)* (5/8)* 22 inches 1 (top) 7 Gb/s 7 Gb/s 7 Gb/s 2 9 Gb/s 9 Gb/s 9 Gb/s 3 11 Gb/s 11 Gb/s 11 Gb/s 4 (bottom) 12 Gb/s 12 Gb/s 12 Gb/s 30 inches 1 7 Gb/s 7 Gb/s 7 Gb/s 2 8 Gb/s 8 Gb/s 7 Gb/s 3 9 Gb/s 8 Gb/s 8 Gb/s 4 10 Gb/s 10 Gb/s 10 Gb/s Note Speed based on observed 50% eye * - first number denotes daughtercard trace width in mils second number denotes backplane trace width in mils Slide 22 of 24

23 Active Cable Assembly in PT Module Slide 23 of 24

24 10.7 Gb/s over 10m of 22 AWG Differential Cable with HSSDC2 Connectors Slide 24 of 24

25 Test Conclusions For backplanes Ability to support 10 Gb/s data rates is layer connection driven. Rate ability degrades as layer connection moves towards top of board. Proper channel design can support 10.7 Gb/s 22 inches over FR-4 (Nelco ) For cables Ability to support 10 Gb/s data rates is limited by quality of cable assembly Refinement in connector design and equalization will increase transmission length Slide 25 of 24

26 Conclusion Successful transmission of data rates at 10 Gb/s and above is a synergistic issue between channel design (architecture, component selection and implementation) and active device techniques. Slide 26 of 24

27 For Further Information Demonstration at Tyco Electronics DesignCon Booth #942 John D Ambrosia (717) john.dambrosia@tycoelectronics.com Ken Lazaris-Brunner (905) , ext ken_la@gennum.com Michael Fogg (717) mike.fogg@tycoelectronics.com Bharat Tailor (905) , ext bharat_t@gennum.com Slide 27 of 24