Feasibility of 40/100G Heterogeneous System based on Channel Data

Transcription

1 Feasibility of 40/100G Heterogeneous System based on Channel Data Jan 2008 Technology Hiroshi Takatori 1

2 Outline Generalized methodology for feasibility analysis of heterogeneous (electro-optical) links is proposed in this contribution. The main benefit of the approach is its ability of comparative analysis between different PMD proposals using a unified methodology based on objective technical and economical metrics. Example analysis is made on the CDR-less Module introduced at the last HSSG meting for 100m-MMF [1]. The analysis is based on measured channel and IC package data [2] over various PCB trace lengths and SFP+ module electrical interface [3]. Same methodology would be equally applicable to the 100G PMDs analysis as well: copper and optical. [1] [2] High-speed package data [3] SFP+Channel Model 2

3 System ID Basics PRBS9, 511 data sequence polynomial x^9+x^4+1. Data Source DUT Link TX Copper / Fiber RX Data Software Adaptive FIR X(i) Error(i) SNR = 20*LOG10(X-rms/Error-rms) Noise Margin Definition Noise Margin = [Achievable SNR] SNRrequired SNRrequired is 17dB for 2PAM-NRZ, BER = 10 **-12 Std needs to define Noise Margin (~6dB) as a recommendation for a robust operation. 3

4 Heterogeneous Link, NRZ-10Gbaud Hyperbolic Tangent Model DFB/VCSEL Nonlinear Model IN OUT Pre-EQL EDC (FFE and DFE) Pre-EQL EDC (FFE and DFE) D PCB Trace 5, 15, 20, 30cm connector connector A Laser Driver AGC Laser Driver AGC TOSA B ROSA TOSA ROSA C 100mm MMF A B System Components Package [2] PCB and SFP+ Module [3] Laser Driver + Tosa 100m MMF A B C D C Rosa + AGC D PCB and SFP+ Module Host EDC 4

5 Package Frequency Response NEXT SD21(dB) NEXT SD21(dB) E+8 5E+8 100M 1E+9 5E+9 1E+10 2E+10 1G 10G Frequency (Hz) -80 1E+8 100M 5E+8 1E+9 1G 5E+9 1E+10 10G 2E+10 Frequency (Hz) Black: 35mm-21mm High BGA Blue:35mm-21mm Low BGA Red:55mm-33mm High BGA Green: 55mm-33mm Low BGA FEXT 35mm-21mm High BGA is used for the analysis. FEXT SD21(dB) E+8 100M 5E+8 1E+9 1G 5E+9 1E+10 10G 2E+10 Frequency (Hz) 5

6 PCB, Module Response Signal Path NEXT 5 0 SD21(dB) Black: MSTL_12mil Red: STL_5mil -40 1E+8 5E+8 1E+9 5E+9 1E+10 2E M G 10G Frequency (Hz) 5cm +4cm 5cm +4cm 15cm +4cm 15cm +4cm 20cm +4cm 30cm +4cm 30cm +4cm Black: BRCM_50mmMSTL_12mil Red: BRCM_50mmSTL_12mil E+8 5E+8 1E+9 5E+9 1E+10 2E M 1G 10G Frequency (Hz) Amplitude (V) , 15, and 30cm MSTL 5, 15, 20, and 30cm STL Including package Model TX source: 400mVop Pre-EQL 1-a*D A = 0~ Time (in UI), 10G baud Single Pulse Response (no TX pre-eql) Including package model Time (in UI) With one tap TX pre-eql used in the system (Prop. delay aligned) 6

7 Key Link Impairments IC Noise and TX Jitter added Crosstalk From other lane IC Noise added Nonlinear Model Pre-EQL EDC (FFE and DFE) Pre-EQL EDC D (FFE and DFE) Self Crosstalk Self Crosstalk connector connector A Laser Driver AGC Laser Driver AGC TOSA B ROSA TOSA ROSA C 100mm MMF Impairments Crosstalk at IC-Package and Module IC electronics noise Tosa Linearity and Rosa Noise TX & RX Jitter TX Peak to peak Jitter Uniform IC Noise and RX Jitter From other lane IC Noise RX Gaussian 7

8 Crosstalk NEXT/FEXT Scenario NEXT-ref1 (Measured by [2]) from other lanes FEXT-A Pre-EQL EDC (FFE and DFE) D NEXT-ref2 (Measured by [3]) FEXT-D from other lanes connector NEXT-A1 A Laser Driver NEXT-A2 Amplitude (mv) NEXT- ref [2], IBM NEXT- ref [3], BRCM Single Pulse Response Source, 400mV,o-p Time (in UI) Node FEXT-A and NEXT-A1,A2 should be considered. FEXT-A and NEXT-A1 are bigger than NEXT-ref2 by ~2dB for FEXT-A and ~4dB for NEXT-A1 NEXT-A2 is in package and/or IC internal. No data available at this time. Use NEXT-ref1 Node A D FEXT-D, NEXT-ref, and NEXT-ref1 should be considered. FEXT-D is bigger than NEXT-ref2 by ~2dB More than two disturbers depending lane assembly for both nodes. 8

9 Results-1: SNR at A Sensitivities to IC noise cm 15cm 30cm Other conditions are ideal. 40 SNR (db) dBm/Hz is used for the rest IC Electronic Noise (dbm/hz) -140dBm/Hz is equivalent to 32nV/sqrt(Hz) into 100ohm 9

10 30 Results-2: SNR at A Sensitivities to Host TX-Jitter (IC noise = -135dBm/Hz) SNR (db) UI-pp is used for the rest. 15cm-MSTL-12mil Peak to Peak Jitter (UI) Amplitude of the uniformly distributed jitter is varied and added to the fixed random jitter( 0.015UI-rms). Peak jitter is defined as the addition of uniform and random jitters. 10

11 Results-3: SNR at A Sensitivities to Crosstalk (IC noise = -135dBm/Hz,Host TX-jitter=0.12UIpp) SNR (db) Number of Disturbers 2 sets of crosstalk are used for the rest. SFP+ 15cm-MSTL-12mil 11

12 Results-4: SNR at C Sensitivities to Limit Amp Gain (IC noise = -135dBm/Hz,Host TX-jitter=0.12UIpp, 2 crosstalk) cm-MSTL-12mil SNR (db) AGC Hyperbolic Tangent Model Vout = Vs*Tanh(Gain*Vin/Vs) Vpp Vs=0.4V 0.8Vpp 21 Gain=2.25 is used for the rest Gain 15cm-MSTL-12mil 12

13 Results-5: SNR at D Sensitivities to RX Jitter (IC noise = -135dBm/Hz,Host TX-jitter=0.12UIpp, 2 crosstalk) SNR (db) RX Jitter =0.02UIrms is used for the rest RX Jitter Jitter (UI-rms) RX jitter spectral is limited to 20MHz. 13

14 Results-6: Final SNR Sensitivities to Number of Taps of FFE and DFE (IC noise = -135dBm/Hz,Host TX-jitter=0.12UIpp, 2 crosstalk) FEE with 32 DFE 22 SNR (db) DEE with 3 FFE Noise Margin Number of Taps 0.14 Amplitude (V) End to end link response Reflections Time in UI 14

15 Final Result, Noise Margin Conditions 1. IC noise 135dBm/Hz 2. TX Jitter 0.12ppUI (uniform + gaussian) 3. Crosstalk SFP+ Moudle equivalent (2 sets of crosstalk) 4. RX jitter 0.02UI-rms (gaussian) m MMF 6. EDC/EQL 3tap FFE + 32 tap DFE Host PCB trace MSTL, 12mils width STL, 5mils width 5cm 15cm 30cm 5cm 15cm 20cm 30cm Margin db 6.1 db

16 Summary Generalized methodology for the feasibility analysis was described using the example of 100m MMF PMD. SNR analysis of the heterogeneous link (electrical and optical mixed) was based on measured package, PCB, and module. Link SNR and Margin was estimated with key link impairments: IC noise, jitter, Crosstalk, and optical component non-linearity. The main benefit of the approach is its ability of comparative analysis between different PMD proposals using a unified methodology based on objective technical and economical metrics. Analysis shows 3~6dB margin with sufficient number of FFE and DFE taps at the host EDC for the end-to-end link. The method is applicable not only to 100m MMF but also to other PMDs such as 10m copper, 10/40km SMF. 16