CU4HDD Backplane Channel Analysis Presenter: Peter Wu, Marvell 1
Outline Analysis of 54 SAS backplane channels (www.t10.org) Channels are from connector to connector (TP1 <-> TP4) IL - Insertion loss ICR ( Insertion loss to crosstalk ratio ) Simulation results using COM model ( Annex 93A 802.3bj with updated configurations) Simulation results with Stateye V4.2.3 Performance analysis for PCS coding and equalizations PCS coding: 8B/10B vs. 64B/66B encoding TX EQ 3 tap TX FIR 2
COM Channel Operating Margin ( Annex 93A) Simulation Configurations 2.5G-BASE-X 8B/10B Baud rate: 3.125G. TX EQ: No DFE: No 5G BASE-X 8B/10B Baud rate: 6.25G With/Without: TX EQ of 3-Tap FIR DFE: 6-Tap 5G BASE-R 64B/66B Baud rate: 5.15625G With/Without: TX EQ of 3-Tap FIR DFE: 6-Tap 3
IL (db) Channel Analysis of Insertion loss 0-5 -10-15 -20-25 -30-35 X: 1.56 Y: -8.52 X: 2.58 Y: -14.87 Worst IL for 2.5G with 8B/10B Encoding Insertion Loss Worst IL for 5G with 64B/66B Encoding X: 2.81 Y: -19.25 X: 3.13 Y: -17.24 Worst IL for 5G with 8B/10B Encoding IL @ Nyquist for 5G with 8B/10B Encoding Scheme 2.5G 8B/10B Worst IL (db) 5G 64B/66B 5G 8B/10B 8.52 14.87 19.25-40 0 1 2 3 4 5 Frequency (GHz) HP24, HP25, HP26 are with worst Insertion loss Large insertion loss deviation @ ~2.8GHz. 4
db Channel Analysis of ICR 70 Insertion to Crosstalk Ratio 60 50 Worst ICR for 5G with 64B/66B Encoding Scheme 2.5G 8B/10B 5G 64B/66B 5G 8B/10B 40 30 20 10 0 X: 1.56 Y: 29.27 Worst ICR for 2.5G with 8B/10B Encoding X: 2.58 Y: 20.41 X: 2.8 Y: 15.68 Worst IL (db) 29.27 20.41 15.68-10 -20 Worst ICR for 5G with 8B/10B Encoding -30 1 2 3 4 5 6 Frequency (GHz) Some channels have ICR dip @ ~2.8GHz 5
COM (db) Simulation Results with COM - 2.5GBASE-X 2.5G 8B/10B 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 No DFE 6-Tap DFE 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 Channel ID 3dB COM Limit 2.5Gbps with 8B/10B encoding 3dB COM reserved for loss from implementations All channels pass without DFE or TX EQ 6
COM (db) Simulation Results with COM -5GBASE-R 5G 64B/66B 13 12 11 10 3 Tap TX FFE NO TX FFE 9 8 7 6 5 4 3 2 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 Channel ID 5Gbps with 64B/66B encoding All channels pass without TX EQ 7
COM (db) Simulation Results with COM -5GBASE-X 5G 8B/10B 12 11 10 9 3tap TX EQ 3 Tap TX FFE No TX EQ NO TX FFE 8 7 6 5 4 3 2 1 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 Channel ID 5Gbps with 8B/10B encoding 20% of channels fail even with TX EQ Non-optimal DFE shape due to 8B/10B idle patterns may result in further performance loss ( Lo_802.3CU4HDD_01_0915) 8
IL (db) ICR (db) Performance Analysis with COM IL dip Insertion Loss Insertion Crosstalk Ratio -5-10 X: 2.58 Y: -11.71 HP25 Worst IL with 64B/66B Encoding 50 40 HP25-15 30 X: 2.58 Y: 23.58 Worst ICR with 64B/66B Encoding -20-25 Worst IL with 8B/10B Encoding X: 2.83 Y: -19.18 20 10 Worst ICR with 8B/10B Encoding X: 2.81 Y: 17.24-30 0-35 -10-40 1 2 3 4 5 6 Frequency (GHz) -20 1 2 3 4 5 6 Frequency (GHz) HP25 has IL dip @ around 2.8GHz and ICR dip @ around 2.8GHz Out of band for 64B/66B scheme (Nyquist at 2.578125GHz) In band for 8B/10B scheme (Nyquist at 3.125GHz) 9
Simulation Results Using Stateye v4.2.3 For additional margin - target BER 10-15 No TX EQ added 6-Tap DFE Jitter model added Results of eye opening for all 54 channels Minimum Required Eye Opening 0.2V Observations: 5GBASE-X Mode(8B/10B), 10 channels fail the required eye opening. 5GBASE-R Mode(64/66B) and 2.5GBASE-X(8B/10B) all channels pass The results align with that of COM analysis 10
Simulation Results Using Stateye 5G mode Channels fail 11
Simulation Results Using Stateye 2.5G mode 12
Conclusions 2.5Gbps 5Gbps ICR are good for all channels. All channels pass without TX EQ or DFE. 8B/10B is feasible All channels pass with 64B/66B encoding and no TX EQ required Some channels fail with 8B/10B encoding 64B/66B is a better choice than 8B/10B for both performance and implementation considerations 13
BACKUP SLIDES 14
List of Channels : 54 channels 15 channel ID channel XTALK 1 'HP01' 'HP19' 2 'HP01' 'HP15+HP16+2HP17+2HP18' 3 'HP02' 'HP19' 4 'HP02' 'HP15+HP16+2HP17+2HP18' 5 'HP03' 'HP19' 6 'HP03' 'HP15+HP16+2HP17+2HP18' 7 'HP04' 'HP19' 8 'HP04' 'HP15+HP16+2HP17+2HP18' 9 'HP05' 'HP19' 10 'HP05' 'HP15+HP16+2HP17+2HP18' 11 'HP06' 'HP19' 12 'HP06' 'HP15+HP16+2HP17+2HP18' 13 'HP07' 'HP19' 14 'HP07' 'HP15+HP16+2HP17+2HP18' 15 'HP08' 'HP19' 16 'HP08' 'HP15+HP16+2HP17+2HP18' 17 'HP09' 'HP19' 18 'HP09' 'HP15+HP16+2HP17+2HP18' 19 'HP10' 'HP19' 20 'HP10' 'HP15+HP16+2HP17+2HP18' 21 'HP11' 'HP19' 15
List of Channels 22 'HP11' 'HP15+HP16+2HP17+2HP18' 23 'HP24' 'HP19' 24 'HP24' 'HP15+HP16+2HP17+2HP18' 25 'HP25' 'HP19' 26 'HP25' 'HP15+HP16+2HP17+2HP18' 27 'HP26' 'HP19' 28 'HP26' 'HP15+HP16+2HP17+2HP18' 29 'long_board_to_drive_oldconn' 'long_board_to_drive_oldconn_next' 30 'short_board_to_drive_oldconn' 'short_board_to_drive_oldconn_next' 31 'long_board_to_board' 'long_board_to_board_fext' 32 'short_board_to_board' 'short_board_to_board_fext' 33 'b1_thu' 'b1_next' 34 'b2_thu' 'b2_next' 35 'c1_thu' 'c1_next' 36 'c2_thu' 'c2_next' 37 'd1_thu' 'd1_next' 38 'd1_thu' 'd1_lcc' 39 'd2_thu' 'd2_next_hdd' 40 'd2_thu' 'd2_next_lcc' 41 'a2_thu' 'a2_next' 42 'a2_thu' 'a2_lcc' 43 'Intel_HDD_BP_C_MB_03_thru' 'Intel_HDD_BP_C_MB_03_FEXT' 44 'Intel_HDD_BP_C_MB_04_thru' 'Intel_HDD_BP_C_MB_04_FEXT' 45 'Intel_HDD_SC_MB_11' 'Intel_HDD_SC_MB_11_FEXT' 46 'Intel_HDD_SC_MB_12' 'Intel_HDD_SC_MB_12_FEXT' 47 'Intel_MB_C_BP_HDD_01_thru' 'Intel_MB_C_BP_HDD_01_FEXT' 48 'Intel_MB_C_BP_HDD_02_thru' 'Intel_MB_C_BP_HDD_02_FEXT' 49 'Intel_MB_LC_HDD_05' 'Intel_MB_LC_HDD_05_FEXT' 50 'Intel_MB_LC_HDD_06' 'Intel_MB_LC_HDD_06_FEXT' 51 'Intel_MB_LC_HDD_07' 'Intel_MB_LC_HDD_07_FEXT' 52 'Intel_MB_LC_HDD_08' 'Intel_MB_LC_HDD_08_FEXT' 53 'Intel_MB_SC_HDD_09' 'Intel_MB_SC_HDD_09_FEXT' 54 'Intel_MB_SC_HDD_10' 'Intel_MB_SC_HDD_10_FEXT' 16