EE290C - Spring 2004 Advanced Topics in Circuit Design High-Speed Electrical Interfaces Lecture #22 Link Budgeting and BER Analysis Vladimir Stojanovic (vlada@stanford.edu) Stanford University and Rambus Inc. 4/8/2004 Agenda Backplane channel review Link system models and noise Performance analysis 2 1 Backplane Environment - Recap Package On-chip parasitic Back plane trace Package via (termination resistance and device loading capacitance) Line card trace Line card via Back plane connector Backplane via Line attenuation Reflections from stubs (vias) 3 Backplane Channel Same backplane Different lengths Different stubs Top vs. Bot Required signal amplitude set by noise Need to architect the link to work over all channels Need tools to estimate link performance over all channels Attenuation [dB] Loss is variable 0 9" FR4 -10 -20 -30 26" FR4 -40 -50 9" FR4, via stub 26" FR4, via stub -60 0 2 4 6 8 10 frequency [GHz] 4 2 Inter-symbol Interference (ISI) - Recap Channel is low pass Dispersion – short latency (skin-effect, dielectric loss) 1 0.8 0.6 Reflections – long latency (impedance mismatches – connectors, via stubs, device parasitics, package) Tsymbol=160ps 0.4 0.2 0 0 1 2 3 ns 5 ISI 1 Error! 0.8 Amplitude pulse response Our nice short pulse gets spread out 0.6 0.4 0.2 0 0 2 4 6 8 10 12 Symbol time 14 16 18 Middle sample is corrupted by 0.2 trailing ISI (from the previous symbol), and 0.1 leading ISI (from the next symbol) resulting in 0.3 total ISI As a result middle symbol is detected in error 6 3 Crosstalk Don’t just receive the signal you want Get versions of signals “close” to you Vertical connections have worse coupling “Close” in these vertical connection regions Sercu, DesignCon03 Far-end XTALK (FEXT) Desired signal Reflections Near-end XTALK (NEXT) 7 Attenuation [dB] Frequency View of Crosstalk 0 -10 THROUGH -20 -30 NEXT -40 FEXT -50 -60 0 2 4 6 8 10 frequency [GHz] For this example: > 4GHz, noise is as large as the signal 8 4 Agenda Backplane channel review Link system models and noise Previous standard approaches Statistical modeling Performance analysis 9 Parameter Definition for VT Based Budget Voltage parameter definitions Simultaneous switching output (SSO) noise Receiver sensitivity (offset + overdrive) Channel loss Crosstalk Back-to-back read Timing parameter definitions tQ: transmitter output timing tSH: receiver output timing tCE: channel timing error tJ: clock source jitter 10 5 RDRAM VT Budget Component of Timing Budget Bit time RAC tQ RDRAM tSH tCE tJ Margin Total RAC to RDRAM ps % 1250.0 500.0 40.0% 400.0 32.0% 290.0 23.2% 60.0 4.8% 0.0 0.0% 100.0% Vterm=1.2V and 36D (channel) At 800Mb/s tCE is at 23% of bit time vCE is at 62.5% Vswing (900mV) 11 Fiber Channel – Methodologies for Jitter Specification Total jitter = Deterministic (DJ) + random jitter (RJ) DJ: Non-Gaussian, bounded in amplitude and has specific causes (duty cycle distortion, data dependent, sinusoidal and uncorrelated (power supply noise injection)) DJ is measured as a peak-to-peak value and adds linearly RJ: Gaussian and measured as an RMS value -12 RJ: Peak-to-peak jitter = 14 * RMS jitter for a BER of 10 Total jitter = peak-to-peak DJ + peak-to-peak RJ Jitter measurement definitions Jitter output Jitter transfer Jitter tolerance (ability of a CDR to successfully recover the data in the presence of jitter) Create a tolerance mask by examining the CDR lock at different frequencies vs. sinusoidal jitter magnitude 12 6 Jitter measurement definitions CDR Tolerance & XAUI Sinusoidal Jitter Tolerance Mask Jitter Amplitude (UI) 9.00 8.50 8.00 7.50 7.00 6.50 6.00 5.50 5.00 4.50 4.00 3.50 3.00 2.50 2.00 1.50 1.00 0.50 0.00 0.010 XAUI Sinusoidal Jitter Mask XAUI Mask LV at Vtt-Rx=1V & 3.125Gbps 1.875MHz 22KHz 0.100 20 MHz 0.1UI 0.1UI 1.000 10.000 100.000 Jitter Frequency (MHz) Jitter generation (jitter added by the PLL due to phase and supply noise) Jitter transfer (jitter at the output of the PLL due to refClk noise) Jitter tolerance (ability of a CDR to successfully recover the data in the presence of jitter) Create a tolerance mask by examining the CDR lock at different frequencies vs. sinusoidal jitter magnitude 13 Fiber Channel Jitter Specification for 1.0625 Gbps 14 7 Previous system models Mostly non-existent Borrowed from computer systems Worst case analysis Can be too pessimistic in links Borrowed from data communications Gaussian distributions Works well near mean Often way off at tails ISI distribution is bounded Need accurate models To relate the power/complexity to performance 15 Comparison w/ Gaussian Model -2 -4 -6 -8 -10 -1 0 40 m V erro r @ 1 0 2 5% of eye h eig h t 0 25 50 75 100 re sidu al IS I [m V ] 10 Steady-State Phase Probability 0 Impact on CDR phase 0 -2 9% T s ym bol -4 -6 -8 -10 log log 10 probability [cdf] Cumulative ISI distribution 4% T s ym bol e rro r @ 10 80 -1 0 100 120 140 160 180 ph ase cou nt Gaussian model only good down to 10-3 probability Way pessimistic for much lower probabilities 16 8 A new model Use direct noise and interference statistics Main system impairments Interference Voltage noise (thermal, supply, offsets, quantization) Timing noise – always looked at separately Key to integrate with voltage noise sources Need to map from time to voltage 17 Residual ISI Error Cannot correct all the ISI Equalizers are finite length EQ coefficients quantized Channel estimate error The error affects both voltage and timing Need to find the distribution of this error 18 9 voltage [mV] 200 0.75 0 1 2 sample # probability [pmf] -3 -2 -1 ISI distribution 0.5 0.25 0 15 -10 Convolution method 1 -15 15 [mV] 1 probability [pmf] Tx Equalized pulse response probability [pmf] Generating ISI Distributions 0.75 1 0.75 0.5 0.25 0 -25 -5 5 25 voltage [mV] 0.5 0.25 0 -10 [mV] 10 19 Estimated Residual Error 5 Tap Transmitter Equalizer probability distribution of residual ISI probability distribution of residual ISI 0.016 0.04 0.012 0.03 0.008 0.02 0.004 0.01 0 -60 -40 -20 0 20 voltage [mV] 40 60 Data sample distribution 0 -150 -100 -50 0 50 100 voltage [mV] 150 Edge sample distribution 20 10 Power spectral density [dBV] Equalizer Related Error Sources edge sample data sample -30 -30 residual ISI residual ISI -40 10% /tap max. eq. estimation error -50 -60 -40 10%/tap max. eq. estimation error -50 10mV/tap max. eq. quantization error 0 1 2 3 frequency [GHz] -60 10mV/tap max. eq. quantization error 0 1 2 3 frequency [GHz] Residual ISI is the biggest source of error Quantization error and equalizer estimation Are significant for reasonable assumptions about accuracy 21 Random Voltage Noise Thermal noise Resistor and Device noise Quantization Estimation error Supply noise Receiver offset 22 11 Effect of Timing Noise Need to map from time to voltage Jittered sampling Voltage noise when receiver clock is off Ideal sampling Voltage noise The effect is going to depend on the size of the jitter, the input sequence, and the channel 23 Effect of Transmitter Jitter Jittered pulse decomposition ideal bk bk ε kTX kT 1 kT ε TX k +1 (k + 1)T 2 + (k +1)T bk ε TX k +1 ε kTX − bk noise Decompose output into ideal and noise Noise are pulses at front and end of symbol Width of pulse is equal to jitter V. Stojanović, M. Horowitz, “Modeling and Analysis of High-Speed Links,” IEEE Custom Integrated Circuits Conference, September 2003. 24 12 Transmitter Jitter Noise Approximate the noise pulses with deltas Assuming jitter is much narrower than channel impulse response bk ε kTX+1 ε TX k +1 bk εkTX ≈ − bk − bk ε kTX Channel output Output with no jitter Response to the noise deltas 25 Jitter effect on voltage noise Transmitter jitter High frequency (cycle-cycle) jitter is bad Changes the energy (area) of the symbol No correlation of noise sources that sum Low frequency jitter is less bad Effectively shifts waveform Correlated noise give partial cancellation ε Rx k ≡ εkRx Receive jitter Modeled by shift of transmit sequence Same as low frequency transmitter jitter 26 13 Jitter Propagation Model TX precoder ak w bk ideal pulse response p( jT ) nvdd nin noise h jT + T PLL Hjit(s) sbE ∑b j = − sbS sbE ∑b j =− sbS âk xk x kjitTX ε kRX RX impulse response x ISI ( kT + φ i + ε kRX ) = x jitter (kT + φi + ε kRX ) = + 2 ε kTX,k +1 x kISI k− j k− j p ( jT + φ i ) T T RX TX RX TX h( jT + 2 + φi ) ε k − ε k − j − h( jT − 2 + φi ) ε k − ε k +1− j ( ) ( ) Channel bandwidth matters If h(T/2) is small, the noise is small 27 h(nT+1/2) not small, many pulses add White jitter Noise from Tx much larger than from Rx jitter From Rx jitter, noise is white From Tx jitter, filtered by the channel Y-axis is noise σ (in Volts) If the noise was white Power spectral density [dBV] Voltage Noise From Jitter Source – white jitter -30 Noise from Tx jitter -40 -50 Noise from Rx jitter -60 σ = 10mV => -40dBV 0 0.5 1 1.5 2 2.5 3 frequency [GHz] Bandwidth of the jitter is critical It sets the magnitude of the noise created 28 14 Jitter Source From PLL Clocks Noise sources Reference clock phase noise VCO supply noise Clock buffer supply noise RefClk +detector Phase Icp − Kpd Icp VCO R Kvco/s Clock buffer C ÷N Stationary phase-space model 29 RefClk Phase Icp − Kpd Icp +detector VCO R C ÷N Kvco/s Clock buffer Noise transfer functions [dB] Noise Transfer Functions 10 from input clock from clock buffer supply 0 -10 -20 from VCO supply -30 5 10 6 10 7 10 8 10 9 10 10 10 frequency [Hz] Low-pass from reference (input clock) Band-pass from VCO supply High-pass from clock buffer supply M. Mansuri, C-K.K. Yang, "Jitter optimization based on phase-locked loop design parameters," IEEE Journal Solid-State Circuits, Nov. 2002 30 15 PLL supply noise Deterministic noise still present. Where is this noise coming from? Total noise ~ 25mV peak-to-peak 3.7% of on-chip VddA (quiet PLL supply) E. Alon, V. Stojanovic, M. Horowitz “Circuits and Techniques for High-Resolution Measurement of On-Chip Power Supply Noise,” IEEE Symposium on VLSI Circuits, June 2004. 31 Noise Spectrum (1) 0 0 -20 -20 -40 -40 PSD (dBV) PSD (dBV) V dd noise -60 -80 -60 -80 Noise floor -100 10MHz V ddA noise 100MHz 1GHz Frequency Noise floor 10GHz -100 10MHz 100MHz 1GHz Fre quency 10GHz Deterministic noise frequency components: 200MHz - ASIC core operating frequency. Noise on link supplies due to ground bounce. 400MHz - reference clock, some link logic. 4GHz - link data rate. Data & edge clocks at 2GHz => 4GHz noise. Tail current modulation in diff. pairs. 32 16 Noise Spectrum (2) 0 0 -20 -20 -40 -40 PSD (dBV) PSD (dBV) V dd noise -60 V ddA noise -60 -80 -80 Noise floor -100 10MHz 100MHz 1GHz Frequency Noise floor 10GHz -100 10MHz 100MHz 1GHz Fre quency 10GHz Random noise mostly white. Low frequency peaking in Vdd noise due to underdamped impedance of distribution network. VddA distribution network more damped because of higher resistance. 33 2x Oversampled Bang-Bang CDR RX dn Slicer dn PD en data Clk edge Clk deserializer dataOut Phase control Phase mixer en (late) ref Clk PLL dn-1 Generate early/late from dn,dn-1,en Simple 1st order loop, cancels receiver setup time Now need jitter on data Clk, not PLL output 34 17 Data Clk Noise Model phase selector and PLL Base linear PLL jitter Add non-linear phase selector noise from CDR Model the CDR loop as a state machine The current phase position is the state State transitions are caused by early/late Jitter on input data and PLL means Possible to be late and get early PD result Often filter early/late to generate up/down A.E. Payzin, "Analysis of a Digital Bit Synchronizer," IEEE Transactions on Communications, April 1983. 35 Probability Transition Probabilities 1 p-up p-dn CDR loop 0.8 Accumulate-reset filter, length 4 Residual ISI 0.6 0.4 p-no-valid transitions p-late p-early At edge -30dBV Desired phase State = 133 p-hold 0.2 0 0 Example system: 50 100 150 200 250 Phase count On average move to correct position But probability of wrong movement is not small Need to find probability of at each phase location 36 18 Bang-Bang CDR Statistical Model pdn,i φ0 phold,i φi−1 φi pup,i φi +1 φL Need steady state probabilities of the states Have the transition probabilities Iteratively apply transition probabilities (Markov chain) Results will converge to a steady-state 37 Bang-Bang CDR Model log 10 Steady-State Probability 0 -5 -10 -15 0 50 100 150 200 250 Phase Count Gives the probability distribution of phase Which is the CDR jitter distribution 38 19 Noise and Interference Summary Many important sources of noise and interference ISI, crosstalk, quantization, estimation, etc. Largest error comes from ISI By factor of 10x Timing is noisy too High frequency transmitter jitter is bad CDR jitter needs to be considered Especially if the data input is noisy What is the impact on performance? 39 Agenda Backplane channel review Link system models and noise Previous standard approaches Statistical modeling Performance analysis 40 20 ISI and CDR Phase Distributions 300 2PAM - lin. eq -5 voltage [mV] 200 -10 100 -15 0 -100 -20 -200 -25 -300 50 60 70 80 90 time [ps] 100 110 -30 In ideal world, there would be only two dots This plot shows how these dots spread out Vertical slice – ISI distribution per time offset Horizontal weight – CDR phase distribution 41 Putting It All Together To compare different designs Compare the voltage margin at given BER Need to include all noise sources Accurate ISI distribution Transmit and receive jitter CDR jitter EQ quantization noise Receiver offset 42 21 BER Contours 5 tap Tx Eq 5 tap Tx Eq + 1 tap DFE 150 150 -5 -5 100 -10 50 -15 0 -50 -20 -100 -150 0 20 40 60 80 100 120 140 160 time [ps] margin [mV] margin [mV] 100 -10 50 -15 0 -50 -20 -25 -100 -25 -30 -150 0 20 40 60 80 100 120 140 160 time [ps] -30 Voltage margin Min. distance between the receiver threshold and contours with same 43 BER BER Contours PAM2 DFE 150 100 -5 150 -10 50 -15 0 -50 -20 -10 100 50 -15 0 -50 -20 -100 100 150 -25 200 250 0 200 -5 margin [mV] 200 250 margin [mV] 250 PAM4 linear equalization -150 -25 -200 40 80 120 160 time [ps] -30 -250 0 40 80 120 160 200 240 280 320 time [ps] -30 44 22 Model and measurements 0 log10(BER) -2 -4 -6 -8 -10 -12 -14 80 60 40 20 0 -20 -40 -60 -80 Voltage Margin [mV] PAM4, 3taps of transmit equalization, 5Gb/s, 26” FR4 channel 45 Attenuation [dB] Example channels 0 -20 (b) 26" NELCO, no stub -40 -60 -80 26" FR4, via stub -100 0 5 10 15 20 frequency [GHz] Legacy (FR4) - lots of reflections Microwave engineered (NELCO) V. Stojanović, A. Amirkhany, M. Horowitz, “Optimal Linear Precoding with Theoretical and Practical Data Rates in High-Speed Serial-Link Backplane Communication,” IEEE International Conference on Communications, June 2004 46 23 Capacity achieving bit loading 10 10 Multi-tone data rates with thermal noise 9 Capacity with thermal noise 8 Nelco 105Gb/s 7 Nelco 64Gb/s FR4 38Gb/s FR4 70Gb/s 6 #bits/Hz #bits/Hz 8 5 6 4 4 3 2 2 1 0 0 5 10 15 frequency [GHz] 20 0 0 25 2 4 6 8 10 12 14 frequency [GHz] Capacity is very big Practical rates lower low target BER<10-15 peak power constraint Thermal noise – the smallest noise source 47 Capacity with link-specific noise 140 therm al noise 120 100 therm al noise and LC PLL phase noise 80 FR4 Capacity [Gb/s] Capacity [Gb/s] NELCO 140 120 100 thermal noise 80 therm al noise and ring PLL phase noise 60 60 40 40 therm al noise and ring PLL phas e noise therm al noise and LC PLL phase noise 20 20 0 -25 -20 log10(Clipping probability) -15 -10 -5 0 0 -25 -20 log10(Clipping probability) -15 -10 -5 0 Effective noise from phase noise Proportional to signal energy Decreases expected gains Still, capacity is much higher than data rates in today’s links (3Gb/s) 48 24 NELCO 90 Data rate [Gb/s] Data rate [Gb/s] Multi-tone with integer bit loading a) NELCO 80 thermal noise 70 60 50 thermal nois e and LC PLL phase noise 40 80 70 60 thermal noise and LC PLL phase noise 30 20 10 10 -20 the rm al noise 50 20 0 -25 b) FR4 40 therm al noise and ring PLL phase noise 30 FR4 90 log10(Clipping probability) -15 -10 -5 0 therm al noise and ring PLL phase noise 0 -25 log10(Clipping probability) -15 -10 -5 0 -20 Peak-power constraint introduces large gap penalty to capacity (can go around with coding, but too expensive) Still pretty high data rates 49 Multi-level: Offset and jitter are crucial thermal noise 35 PAM16 30 PAM8 25 30 PAM8 25 20 PAM16 PAM4 15 20 25 20 PAM4 15 PAM2 PAM4 15 10 PAM2 10 PAM2 10 PAM8 5 5 0 0 30 Data rate [Gb/s] Data rate [Gb/s] Data rate [Gb/s] 45 40 thermal noise + offset+ jitter thermal noise + offset 2 4 6 8 10 12 14 16 18 20 Symbol rate [Gs/s] 0 0 5 2 4 6 8 10 12 14 16 18 20 Sym bol rate [Gs/s] 0 0 2 4 6 8 10 12 14 16 18 20 Sym bol rate [Gs/s] To make better use of available bandwidth, need better circuits PAM2/PAM4 robust candidate for next generation links 50 25 Full ISI compensation too costly thermal noise + offset 16 PAM4 14 12 20 20 18 PAM8 10 Data rate [Gb/s] Data rate [Gb/s] 20 18 18 16 16 14 14 12 PAM16 PAM2 PAM4 12 10 10 PAM8 8 8 thermal noise + offset+ jitter Data rate [Gb/s] thermal noise PAM2 8 6 6 4 4 4 2 2 0 0 0 0 6 2 4 6 8 10 12 14 16 Symbol rate [Gs/s] PAM4 PAM8 PAM2 2 2 4 6 8 10 12 14 16 Sym bol rate [Gs/s] 0 0 2 4 6 8 10 12 14 16 Symbol rate [Gs/s] Today’s links cannot afford to compensate all ISI Limits today’s maximum achievable data rates 51 Conclusions Backplane links limited by the channel ISI is large in baseband links Can’t completely compensate (At least not with reasonable area/power) Residual ISI also increases CDR jitter Generally have low BER requirements Accurate noise statistic important Many of large noise source are bounded Power constrained transmitter PAM4 and PAM2 with simple DFE are attractive solutions Still, capacity of these links is very big Smart multi-tone? 52 26