Exceeding the Limits of Binary Data Transmission on Printed Circuit Boards by Multilevel Signaling Markus Grözing, Manfred Berroth INT, in cooperation with Michael May Agilent Technologies, Böblingen Prof. Dr.-Ing. Manfred Berroth 1
Outline Motivation Backplane channel characteristics Multilevel penalty versus multilevel benefit Transmitter & receiver equalization Simulation of PAM 2 / PAM 4 transmission Conclusions & Outlook 2
Motivation Computer mass market, i.e. processor memory interface DDR3 ~ 1 Gbit/s FBDIMM 4.8 Gbit/s tomorrow: 7 to 20 Gbit/s Cables transmission length: 2 to 20 cm Backplanes, multi-chip modules, i.e. i.e. HDMI 1.3 10 Gbit/s network cross connects, mainframes today: few Gbit/s tomorrow: > 10 Gbit/s transmission length: up to 1m Optical networks, i.e. LAN, MAN and WAN today: 2,5 to 10 Gbit/s tomorrow: 40 to 100 Gbit/s trans. length: 10 m to 1000 km 3
Backplane channel characteristics time & frequency domain 4
FR4 Backplane (Illustration similar to measured backplane) 10 GbE XAUI = 2x 4x 3.125 Gbit/s 2x 2in (daughter card) + 30in (backplane) = 34in = 86cm Backplane-Illustration from Analysis of Backplanes, Stephen Anderson, Xilinx, Minneapolis, OIF-Contribution Number: OIF2004.005.01, Working Group: PLL 5
12-Port S-Parameters: Port Definition Differential ports Single ended ports Single ended ports Differential ports 1 1 3 2 4 2 3 5 7 6 8 4 5 9 11 10 12 6 Illustration from: Minimizing Multiple Aggressor Differential Crosstalk in High Speed Interconnects using Measurement-based Modeling, Mike Resso, Agilent Technologies 6
6-Port differential S-Parameters of 86cm Traces Insertion Loss Return Loss FEXT NEXT 7
Transmission Characteristic of 86cm Trace m1: IL (1.67 GHz) = 8.9 db m2: IL (3.33 GHz) = 22.5 db m3: IL (5.00 GHz) = 27.1 db S 43 [db] m4: IL (6.67 GHz) = 54.6 db Frequency [GHz] 8
Unit Pulse (1V, 1T) -Response of 86 cm Trace T=300 ps 3.33 Gbaud voltage [mv] T=150ps 6.67 Gbaud T=75ps 13.33 Gbaud time [ns] 9
Multilevel penalty versus Multilevel benefit 10
Multilevel Penalty Example: 20 Gbit/s, 0.5 V pp 50 ps 1/2 V PAM Signal Power Eye Area 2 1 1 100 ps 1/6 V 4 Penalty 1 / 9-9.5 db 2 / 3-1.8 db Picture of PAM4-Eye-Measurement from Proposals for CEI25 Channels Based on lower-k Dielectric Materials for Backplanes and Daughterboards, Helmut Preisach, Alcatel Lucent, Stuttgart, OIF-Contribution Number: OIF2007.135.00, Working Group: PLL 11
Multilevel Benefit: Consider SNR @ f Nyquist 0-10 20 Gbit/s PAM4 = 10 Gbaud 20 Gbit/s PAM2 = 20 Gbaud S21 [db] -20-30 Overall SNR gain Multilevel SNR loss (9,5 db) SNR gain due to reduced symbol rate -40 Noise Level SNR new SNR old NEXT FEXT Thermal Noise -50 0,0E+00 2,0E+09 4,0E+09 6,0E+09 8,0E+09 1,0E+10 frequency [Hz] 12
Transmitter & Receiver Equalization FIR-filter (FFE) in transmitter FIR FIR-filter in transmitter DFE in receiver FIR+DFE 13
Transmitter & Receiver Equalization FIR-filter transmitter electrical channel FIR-filter receiver DFE receiver data in FIR FIR DFE data out clock FIR FIR DFE T T T T T T T T c 0 c 0 c -1 c -1 c 1 digital domain c -2 analog domain c -2 c 2 digital domain 14
1 Re f 2 Example of Simulation Setup in ADS TRANSIENT Tran Tran2 StopTime=10.0 nsec MaxTimeStep=1.0 psec TRANSIENT Tran Tran1 StopTime=30.0 nsec MaxTimeStep=1.0 psec VtLFSR_DT SRC7 Vlow= -e_1 V Vhigh=e_1 V Rate=10 GHz Delay=0.5 nsec Taps= bin("10000000000000100") Seed= bin("10101010101010101") Rout=1 Ohm VtLFSR_DT SRC4 Vlow= -e_0 V Vhigh=e_0 V Rate=10 GHz Delay=0.4 nsec Taps= bin("10000000000000100") Seed= bin("10101010101010101") Rout=1 Ohm VtLFSR_DT SRC13 Vlow= -e3 V Vhigh=e3 V Rate=10 GHz Delay=0.1 nsec Taps= bin("10000000000000100") Seed= bin("10101010101010101") Rout=1 Ohm VtLFSR_DT SRC14 Vlow= -e4 V Vhigh=e4 V Rate=10 GHz Delay=0.0 nsec Taps= bin("10000000000000100") Seed= bin("10101010101010101") Rout=1 Ohm Var Eqn VAR VAR1 e_1= 0 e_0= 4.0 e1= 0 e2= 0 e3= 0 e4= 0 DT DT DT DT DT DT R R1 R=50 Ohm Var Eqn VAR VAR2 e_1= -1.7593 e_0= 4.0 e1= 0 e2= 0 e3= 0 e4= 0 L L1 L=0.25 nh R= VtLFSR_DT SRC6 Vlow=-e1 V Vhigh=e1 V Rate=10 GHz Delay=0.3 nsec Taps= bin("10000000000000100") Seed= bin("10101010101010101") Rout=1 Ohm VtLFSR_DT SRC8 Vlow=-e2 V Vhigh=e2 V Rate=10 GHz Delay=0.2 nsec Taps= bin("10000000000000100") Seed= bin("10101010101010101") Rout=1 Ohm InOrg Var Eqn VAR VAR3 e_1= -1.7593 e_0= 4.0 e1= -1.6056 e2= 0 e3= 0 e4= 0 C C2 C=0.35 pf Var Eqn VtPulse SRC18 Vlow= -1 V t Vhigh=1 V Delay=0.0 psec Edge= cosine Rise=1 psec Fall=1 psec Width=99 psec Period=10 nsec VAR VAR4 e_1= -1.7593 e_0= 4.0 e1= -1.6056 e2= 0.6329 e3= 0 e4= 0 VCVS SRC5 G=1 R1=50 Ohm R2=50 Ohm Var Eqn In VtPulse SRC21 Vlow=0 V Vhigh=e_0 V Delay=0.4 nsec Edge= cosine Rise=1 psec Fall=1 psec Width=99 psec Period=10 nsec VtPulse SRC23 Vlow=0 V Vhigh=e2 V Delay=0.2 nsec Edge= cosine Rise=1 psec Fall=1 psec Width=99 psec Period=10 nsec VtPulse SRC24 Vlow=0 V Vhigh=e4 V Delay=0.0 nsec Edge= cosine Rise=1 psec Fall=1 psec Width=99 psec Period=10 nsec VAR VAR5 e_1= -1.7593 e_0= 4.0 e1= -1.6056 e2= 0.6329 e3= -0.2325 e4= 0 S2P SNP1 File= "DD_2N" Var Eqn Var Eqn VAR VAR7 e_1= 0 e_0= 4.0 e1= -1.1920 e2= 0 e3= 0 e4= 0 Out VtPulse SRC22 Vlow=0 V t Vhigh=e_1 V Delay=0.5 nsec Edge= cosine Rise=1 psec Fall=1 psec t Width=99 psec Period=10 nsec VtPulse SRC20 Vlow=0 V t Vhigh=e1 V Delay=0.3 nsec Edge= cosine Rise=1 psec Fall=1 psec t Width=99 psec Period=10 nsec VtPulse SRC25 Vlow=0 V t Vhigh=e3 V Delay=0.1 nsec Edge= cosine Rise=1 psec Fall=1 psec t Width=99 psec Period=10 nsec VAR VAR6 e_1= -1.7593 e_0= 4.0 e1= -1.6056 e2= 0.6329 e3= -0.2325 e4= 0.0738 VSum R SUM1 R2 R=50 Ohm Vback Var Eqn VDFE VAR VAR8 e_1= 0 e_0= 4.0 e1= -1.2698 e2= 0.3672 e3= 0 e4= 0 VSum SUM6 Var Eqn SampleHoldSML SAMP5 Fnom=0 Hz Vhold SampleHoldSML SAMP1 Fnom=0 Hz VSum SUM2 VSum SUM4 VAR VAR9 r1=0.174 r2=0.076 r3=0.040 r4=0 r5=0 r6=0 r7=0 r8=0 r9=0 OutHold Clock Comparator CMP2 Vlow=0.0 V Vhigh=100 V VSum SUM3 VSum SUM5 Var Eqn VAR VAR10 r1=0.174 r2=0.076 r3=0.040 r4=0.013 r5=0.022 r6=0.012 r7=0 r8=0 r9=0 V_DC SRC16 Vdc=0.5 V VMult MULT1 VMult MULT2 VMult MULT3 VMult MULT4 VMult MULT5 VMult MULT6 VtPulse SRC17 Vlow=0 V Vhigh=1 V Delay=81.1 psec Edge= cosine Rise=1 psec Fall=1 psec Width=50 psec Period=100 psec VCVS SRC15 G=2 V_DC SRC26 Vdc=-r1 V V_DC SRC27 Vdc=-r2 V V_DC SRC28 Vdc=-r3 V V_DC SRC32 Vdc=-r4 V V_DC SRC31 Vdc=-r5 V V_DC SRC30 Vdc=-r6 V VtPulse SRC29 Vlow=1 V Vhigh=0 V Delay=81.1 psec Edge= cosine Rise=1 psec Fall=1 psec Width=50 psec Period=100 psec Vback3 Vcomp Vback1 Vback2 t t SampleHoldSML SAMP2 Fnom=0 Hz SampleHoldSML SAMP3 Fnom=0 Hz SampleHoldSML SAMP4 Fnom=0 Hz SampleHoldSML SAMP8 Fnom=0 Hz SampleHoldSML SAMP7 Fnom=0 Hz SampleHoldSML SAMP6 Fnom=0 Hz 15
Transmitter Swing Limit Peak voltage at transmitter is limited T T T c 0 c -1 c -2 k c = i i= 0 V max Peak voltage at transmitter set to ± 1 V in all simulations (1-norm of FFE equalization vector = 2) 16
Simulation of PAM 2 / PAM 4 transmission with FIR / FIR+DFE 3.33 Gbit/s 6.67 Gbit/s 10.00 Gbit/s 13.33 Gbit/s 17
Eye Quality Measures Eye Area [in pvs] A ~ ½ * T open * V open µ H (~ integrated charge on decision latch input ) V open T open µ L Eye Quality factor (= signal-to-noise ratio at optimum sampling time) BER estimation from Q-factor (assuming Gaussian distribution of sampled voltage, is too conservative in most cases) Q = µ σ H H BER = µ + σ e L L Q 2 2 Q 2π 18
Transmission Characteristic of 86cm Trace m1: IL (1.67 GHz) = -8.9 db m2: IL (3.33 GHz) = -22.5 db Insertion Loss [db] m3: IL (5.00 GHz) = -27.1 db m4: IL (6.67 GHz) = -54.6 db Frequency [GHz] 19
86 cm 3.33 Gbit/s PAM2 (no equalization) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.67 500 50 3.6 1.9 x 10-4 20
86 cm 3.33 Gbit/s PAM2 FIR (0 Pre-, 1 Post-FIR-Tap) Preemphasis horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.89 670 89 13.3 1.4x 10-40 21
Summary 3.33 Gbit/s 86 cm Quality Measure V open T open A open Q EQ Taps PAM [mv] [ps] [pvs] - - 2 500 201 50 3.6 FIR 1 post 2 670 267 89 13.3 22
86 cm 6.67 Gbit/s PAM2 FIR (no equalization) horizontal eye opening vertical eye height eye quality factor Q estimated optimum BER UI mv - - - - 23
86 cm 6.67 Gbit/s PAM2 FIR (0 Pre-, 1 Post-FIR-Tap) Preemphasis horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.63 140 6.6 3.5 2.5 x 10-4 24
86 cm 6.67 Gbit/s PAM2 FIR (1 Pre-, 1 Post-FIR-Taps) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.48 130 4.7 4.3 1.1x 10-5 25
86 cm 6.67 Gbit/s PAM2 FIR (2 Pre-, 3 Post-FIR-Taps) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.58 130 5.7 10.3 4.7x 10-25 26
86 cm 6.67 Gbit/s PAM2 FIR+DFE (2 Pre-FIR-, 4 DFE-T.) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.65 290 14.1 9.7 1.6x 10-22 27
86 cm 6.67 Gbit/s PAM2 FIR+DFE (3 Pre-FIR, 9 DFE-T.) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.75 340 19.1 19.4 6 x 10-84 28
86 cm 6.67 Gbit/s PAM4 FIR (no equalization) horizontal eye opening vertical eye height eye quality factor Q estimated optimum BER UI mv - - - - 29
86 cm 6.67 Gbit/s PAM4 FIR (1 Pre-, 1 Post-FIR-Tap) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.41 170 10.5 7.1 7 x 10-13 30
86 cm 6.67 Gbit/s PAM4 FIR (2 Pre-, 3 Post-FIR-Taps) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.46 210 14.5 15.1 2 x 10-51 31
86 cm 6.67 Gbit/s PAM4 FIR+DFE (2 Pre-FIR-, 4 DFE-T.) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.43 240 15.5 11.6 4 x 10-31 32
86 cm 6.67 Gbit/s PAM4 FIR+DFE (3 Pre-FIR-, 9 DFE-T.) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.45 300 20.3 17.0 4 x 10-65 33
Summary 6.67 Gbit/s 86 cm Quality Measure V open T open A open EQ Taps PAM [mv] [ps] [pvs] FIR 1 pre 2 130 72 4.7 1 post 4 170 123 10.5 FIR 2 pre 2 130 87 5.7 3 post 4 210 138 14.5 FIR 2 2 290 97.5 14.1 +DFE 4 4 240 129 15.5 FIR 3 2 340 112 19.1 +DFE 9 4 300 135 20.3 Q 4.3 7.1 10.3 15.1 9.7 11.6 19.4 17.0 34
86 cm 10.00 Gbit/s PAM2 FIR (2 Pre-, 4 Post-FIR-Taps) horizontal eye opening vertical eye height eye quality factor Q estimated optimum BER UI mv - 67 2.6 5 x 10-3 35
86 cm 10.00 Gbit/s PAM2 FIR+DFE (4 Pre-FIR-, 9 DFE-T.) horizontal eye opening vertical eye height eye quality factor Q estimated optimum BER UI mv - 172 2.5 7 x 10-3 36
86 cm 10.00 Gbit/s PAM4 FIR (2 Pre-, 3 Post-FIR-Taps) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.29 110 3.2 12.8 4 x 10-38 37
86 cm 10.00 Gbit/s PAM4 FIR+DFE (3 Pre-FIR-, 9 DFE-T.) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.37 180 6.7 13.7 3 x 10-43 38
Summary 10.00 Gbit/s 86 cm Quality Measure V open T open A open EQ Taps PAM [mv] [ps] [pvs] FIR 2 pre 4 post 2 - - - FIR 2 pre 3 post 4 110 58 3.2 FIR 4 +DFE 9 2 - - - FIR 3 +DFE 9 4 180 74 6.7 Q 2.6 12.8 2.5 13.7 39
34 in 13.33 Gbit/s PAM2 FIR (19 FIR-Taps) horizontal eye opening vertical eye height eye quality factor Q estimated optimum BER ps mv - 10 2.3 1 x 10-2 40
86 cm 13.33 Gbit/s PAM2 FIR+DFE (4 Pre-FIR-, 9 DFE-T.) horizontal eye opening vertical eye height eye quality factor Q estimated optimum BER UI mv - 93 2.2 2 x 10-2 41
86 cm 13.33 Gbit/s PAM4 FIR (2 Pre-, 3 Post-FIR-Taps) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.13 20 0.2 4.5 3 x 10-6 42
86 cm 13.33 Gbit/s PAM4 FIR+DFE (3 Pre-FIR-, 9 DFE-T.) horizontal eye opening vertical eye opening eye area eye quality factor Q estimated optimum BER UI mv pvs 0.27 90 1.8 9.1 6 x 10-20 43
EQ FIR Summary 13.33 Gbit/s 86 cm Quality Measure V open T open A open Taps PAM [mv] [ps] [pvs] 19 2 - - - Q 2.3 FIR FIR +DFE FIR +DFE 2 pre 3 post 4 9 3 9 4 2 4 20-90 19-40 0.2-1.8 4.5 2.2 9.1 44
Summary PAM X / EQ versus Bit Rate Bit Rate [Gbit/s] 3.33 6.67 10.0 13.33 IL @ f bit /2 [db] -8.9-22.5-27.1-54.6 Modul. EQ-Meth. none +/- - - - PAM 2 FIR FIR+DFE + Overkill + + - - - - PAM 4 FIR FIR+DFE Overkill Overkill + + + + - + 45
Conclusion: Multilevel Benefit! Insertion loss @ f bit /2: ~ 10 db: PAM 2 w/o any equalization PAM 2 with FIR (1-tap, preemphasis) ~ 20 db: PAM 2 with FIR, PAM 2 with FIR+DFE PAM 4 with FIR PAM 4 with FIR+DFE ~ 30 db: PAM 4 with FIR PAM 4 with FIR+DFE ~ 40 db: PAM 4 with FIR+DFE Benefit 1: for low loss channels IL @ f bit /2 < ~25 db PAM 4 offers better eye opening with less equalization effort Benefit 2: for high loss channels IL @ f bit /2 > ~25 db transmission only possible with PAM 4 46
Summary Residual ISI after limited/non-ideal equalization is the most important limiting factor for the possible backplane transmission throughput Equalization is a MUST for PAM4-transmission, too, but the EQ-effort may be lower than for PAM2 PAM4 clearly offers the potential for an increase in transmission throughput FFE+DFE PAM4 offers the largest throughput Main challenges for real-world-implementation: PAM4-DFE-circuit implementation at high speed PAM4 needs transmission-amplitude dependant decision levels 47
48
BACK UP Decision Latch Sensitivity 49
90 nm CMOS Decision Flip-Flop Layout _V CLK I Bias V D Q D Q in,d A=1 V out,d CLK _CLK _V in V out V in _V out V CLK V SS 50
90nm CMOS Flip-Flop: Measured Phase Margin 360 200 mv Single-ended input voltage swing as parameter phase margin [degrees] 270 180 90 0 25 mv 50 mv 100 mv 35 mv decision latch requires ~ 2.5 pvs diff,pp input eye area for error-free operation 400 mv 300 mv 0 10 20 30 40 50 f Toggle = f bit @ 5, 10 Gbit/s. f Toggle = ¼ f bit @ 20, 30, 40 Gbit/s Phase margin 10 GHz: 324 @ 200 mv input bit rate [Gbit/s] 274 @ 50 mv 51
BACK UP Simulations with Crosstalk 52
Differential NEXT & FEXT w/4-port VNA Modified illustration from Minimizing Multiple Aggressor Differential Crosstalk in High Speed Interconnects using Measurement-based Modeling, Mike Resso, Agilent Technologies 53
86 cm 6.67 Gbit/s FFE PAM2 +NEXT w NEXT w/o NEXT horizontal eye opening UI 0.56 0.58 vertical eye height mv 181 183 eye quality factor Q 9.7 10.3 estimated optimum BER 1 x 10-22 5 x 10-25 54
86cm 6.67 Gbit/s FFE+DFE PAM2 +NEXT w NEXT w/o NEXT horizontal eye opening UI 0.73 0.75 vertical eye height mv 427 426 eye quality factor Q 17.5 19.4 estimated optimum BER 1 x 10-68 6 x 10-84 55
86cm 13.3 Gbit/s FFE+DFE PAM4 +NEXT w NEXT w/o NEXT horizontal eye opening UI 0.26 0.26 vertical eye height mv 140 140 eye quality factor Q 7.9 9.1 estimated optimum BER 8 x 10-16 6 x 10-20 56
Back UP Channel capacity limiting factors 57
Channel Capacity Limiting Factors 1. Residual ISI after limited, non-ideal equalization 2. Crosstalk near-end (NEXT) and far-end (FEXT) 3. Circuit deficiencies duty cycle distortion, resolution and offset of deciders 4. Timing jitter transmitter and receiver 5. Thermal noise transmitter & receiver circuitry, termination resistors 58