advanced topics in circuit design high-speed electrical...
TRANSCRIPT
1
EE290C - Spring 2004Advanced Topics in Circuit DesignHigh-Speed Electrical Interfaces
Lecture #22Link Budgeting and BER AnalysisVladimir Stojanovic ([email protected])Stanford University and Rambus Inc.4/8/2004
2
Agenda
Backplane channel reviewLink system models and noisePerformance analysis
2
3
Backplane Environment - Recap
Line attenuationReflections from stubs (vias)
Back plane connector
Line card trace
Package
On-chip parasitic(termination resistance and device loading capacitance)
Line card via
Back plane trace
Backplane via
Package via
Back plane connector
Line card trace
Package
On-chip parasitic(termination resistance and device loading capacitance)
Line card via
Back plane trace
Backplane via
Package via
4
Backplane ChannelLoss is variable
Same backplaneDifferent lengthsDifferent 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
0 2 4 6 8 10
-60
-50
-40
-30
-20
-10
0
frequency [GHz]
Atte
nuat
ion
[dB
]
9" FR4, via stub
26" FR4,via stub
26" FR4
9" FR4
3
5
Inter-symbol Interference (ISI) - RecapChannel is low pass
Our nice short pulse gets spread out
0 1 2 3
0
0.2
0.4
0.6
0.8
1
ns
puls
e re
spon
se
Tsymbol=160ps
Dispersion –short latency(skin-effect, dielectric loss) Reflections –long latency(impedance mismatches –connectors, via stubs, device parasitics, package)
6
ISI
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 ISIAs a result middle symbol is detected in error
0 2 4 6 8 10 12 14 16 180
0.2
0.4
0.6
0.8
1
Symbol time
Am
plitu
de
Error!
4
7
CrosstalkDon’t just receive the signal you want
Get versions of signals “close” to youVertical connections have worse coupling
“Close” in these vertical connection regions
Far-end XTALK (FEXT)
Desired signal
Near-end XTALK (NEXT)
Reflections
Sercu, DesignCon03
8
Frequency View of Crosstalk
For this example:> 4GHz, noise is as large as the signal
0 2 4 6 8 10
-60
-50
-40
-30
-20
-10
0
frequency [GHz]
Atte
nuat
ion
[dB
]
FEXT
NEXT
THROUGH
5
9
Agenda
Backplane channel reviewLink system models and noise
Previous standard approachesStatistical modeling
Performance analysis
10
Parameter Definition for VT Based BudgetVoltage parameter definitions
Simultaneous switching output (SSO) noiseReceiver sensitivity (offset + overdrive)Channel lossCrosstalk Back-to-back read
Timing parameter definitionstQ: transmitter output timingtSH: receiver output timingtCE: channel timing errortJ: clock source jitter
6
11
Component of RAC to RDRAMTiming Budget ps %
Bit time 1250.0RAC tQ 500.0 40.0%
RDRAM tSH 400.0 32.0%tCE 290.0 23.2%tJ 60.0 4.8%
Margin 0.0 0.0%Total 100.0%
RDRAM VT Budget
Vterm=1.2V and 36D (channel)At 800Mb/s
tCE is at 23% of bit timevCE is at 62.5% Vswing (900mV)
12
Fiber Channel – Methodologies for Jitter SpecificationTotal 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 linearlyRJ: Gaussian and measured as an RMS valueRJ: Peak-to-peak jitter = 14 * RMS jitter for a BER of 10
-12
Total jitter = peak-to-peak DJ + peak-to-peak RJJitter measurement definitions
Jitter outputJitter transferJitter 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
7
13
Jitter measurement definitions
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
CDR Tolerance & XAUI Sinusoidal Jitter Tolerance Mask
0.000.501.001.502.002.503.003.504.004.505.005.506.006.507.007.508.008.509.00
0.010 0.100 1.000 10.000 100.000
Jitter Frequency (MHz)
Jitt
er A
mpl
itude
(UI)
XAUI MaskLV at Vtt-Rx=1V & 3.125Gbps
XAUI Sinusoidal Jitter Mask
1.875MHz 20 MHz
0.1UI0.1UI
22KHz
14
Fiber Channel Jitter Specification for 1.0625 Gbps
8
15
Previous system modelsMostly non-existentBorrowed from computer systems
Worst case analysisCan be too pessimistic in links
Borrowed from data communicationsGaussian distributions
Works well near mean Often way off at tails
ISI distribution is bounded
Need accurate models To relate the power/complexity to performance
16
Comparison w/ Gaussian Model
0 25 50 75 100
-10
-8
-6
-4
-2
0
re sidual ISI [m V ]80 100 120 140 160 180
-10
-8
-6
-4
-2
0
40mV error @ 10-10
25% of eye height
4% Tsym bol
error @ 10-10
9% Tsym bol
log 10
pro
babi
lity
[cdf
]
log 10
Ste
ady-
Stat
e Ph
ase
Prob
abili
ty
phase count
Cumulative ISI distribution Impact on CDR phase
Gaussian model only good down to 10-3 probabilityWay pessimistic for much lower probabilities
9
17
A new model
Use direct noise and interference statistics
Main system impairmentsInterference
Voltage noise (thermal, supply, offsets, quantization)
Timing noise – always looked at separatelyKey to integrate with voltage noise sourcesNeed to map from time to voltage
18
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
10
19
Generating ISI Distributions
-3 -2 -1 0 1 2 -10
15
200
-10 10 0
0.25
0.5
0.75
1
-15 15 0
0.25
0.5
0.75
1
-25 -5 5 25 0
0.25
0.5
0.75
1
volta
ge [m
V]
sample #
[mV]
prob
abili
ty [p
mf]
prob
abili
ty [p
mf]
prob
abili
ty [p
mf]
[mV]
voltage [mV]
Convolution method
Tx Equalizedpulse response ISI distribution
20
Estimated Residual Error
-60 -40 -20 0 20 40 60 0
0.01
0.02
0.03
0.04
voltage [mV]
prob
abili
ty d
istri
butio
n of
resi
dual
ISI
-150 -100 -50 0 50 100 150 0
0.004
0.008
0.012
0.016
voltage [mV]
prob
abili
ty d
istri
butio
n of
resi
dual
ISI
Data sample distribution Edge sample distribution
5 Tap Transmitter Equalizer
11
21
Equalizer Related Error Sources
Residual ISI is the biggest source of errorQuantization error and equalizer estimation
Are significant for reasonable assumptions about accuracy
0 1 2 3
-60
-50
-40
-30
0 1 2 3
-60
-50
-40
-30data sample
residual ISI
Pow
er s
pect
ral d
ensi
ty [d
BV]
10%/tap max. eq. estimation error
10mV/tap max. eq. quantization error
frequency [GHz] frequency [GHz]
10mV/tap max. eq. quantization error
10%/tap max. eq. estimation error
residual ISI
edge sample
22
Random Voltage Noise
Thermal noiseResistor and Device noise
Quantization Estimation errorSupply noiseReceiver offset
12
23
Effect of Timing NoiseNeed to map from time to voltage
Ideal sampling
The effect is going to depend on the size of the jitter, the input sequence, and the channel
Jittered sampling
Voltage noiseVoltage noise when receiver clock is off
24
Effect of Transmitter Jitter
Decompose output into ideal and noiseNoise are pulses at front and end of symbol
Width of pulse is equal to jitter
Jittered pulse decomposition
ideal
noise
kb
kT
TXkε
Tk )1( +
TXk 1+ε
kT
TXkε
Tk )1( +
TXk 1+ε
+
kb−
kb
kb
1
2
V. Stojanović, M. Horowitz, “Modeling and Analysis of High-Speed Links,” IEEE Custom Integrated Circuits Conference, September 2003.
13
25
Transmitter Jitter NoiseApproximate the noise pulses with deltas
Assuming jitter is much narrower than channel impulse response
Channel outputOutput with no jitterResponse to the noise deltas
TXkε
TXk 1+ε
kb−
kb
≈TXkkb ε−
TXkkb 1+ε
26
Jitter effect on voltage noiseTransmitter jitter
High frequency (cycle-cycle) jitter is badChanges the energy (area) of the symbolNo correlation of noise sources that sum
Low frequency jitter is less badEffectively shifts waveformCorrelated noise give partial cancellation
Receive jitterModeled by shift of transmit sequenceSame as low frequency transmitter jitter
εεεεkRx≡≡≡≡
εεεεkRx
14
27
Jitter Propagation Model
Channel bandwidth mattersIf h(T/2) is small, the noise is smallh(nT+1/2) not small, many pulses add
TXw )( jTp
)(sHjitPLL
ka kb
TXkk 1, +ε
inn
+
2TjTh
+ kxISIkx
jitTXkx RX
kε
kaprecoder
impulseresponse
pulseresponse
vddn
RX
ideal
noise
∑−=
− +=++sbE
sbSjijk
RXki
ISI jTpbkTx )()( φεφ
( ) ( )∑−=
−+−−
−+−−−++=++
sbE
sbSj
TXjk
RXki
TXjk
RXkijk
RXki
jitter TjThTjThbkTx 1)2
()2
()( εεφεεφεφ
28
Voltage Noise From Jitter
White jitterNoise from Tx much larger than from Rx jitterFrom Rx jitter, noise is whiteFrom Tx jitter, filtered by the channel
Y-axis is noise σσσσ (in Volts)If the noise was whiteσ = 10mV => -40dBV
0 0.5 1 1.5 2 2.5 3
-60
-50
-40
-30
Pow
er s
pect
ral d
ensi
ty [d
BV
]
frequency [GHz]
Source – white jitter
Noise from Tx jitter
Noise from Rx jitter
Bandwidth of the jitter is criticalIt sets the magnitude of the noise created
15
29
Jitter Source From PLL Clocks
Ref Clk Phasedetector
Kpd
Icp
Icp R
C
VCOKvco/s
Clockbuffer
N÷
+−
Noise sourcesReference clock phase noiseVCO supply noiseClock buffer supply noise
Stationary phase-space model
30
Noise Transfer Functions
105 106 107 108 109 1010
-30
-20
-10
0
10
frequency [Hz]
Noi
se tr
ansf
er fu
nctio
ns [d
B]
fromVCO supply
frominput clock
fromclock buffer supply
Ref Clk PhasedetectorKpd
Icp
Icp R
C
VCOKvco/s
Clockbuffer
N÷
+−
Low-pass from reference (input clock)Band-pass from VCO supplyHigh-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
16
31
PLL supply noise
Total noise ~ 25mV peak-to-peak3.7% of on-chip VddA (quiet PLL supply)
Deterministic noise still present.
Where is this noise coming from?
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.
32
Noise Spectrum (1)
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.
10MHz 100MHz 1GHz 10GHz-100
-80
-60
-40
-20
0
Frequency
PSD
(dBV
)
10MHz 100MHz 1GHz 10GHz-100
-80
-60
-40
-20
0
Frequency
PSD
(dBV
)
Noise floor Noise floor
Vdd noise VddA noise
17
33
Noise Spectrum (2)
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.
10MHz 100MHz 1GHz 10GHz-100
-80
-60
-40
-20
0
Frequency
PSD
(dBV
)
10MHz 100MHz 1GHz 10GHz-100
-80
-60
-40
-20
0
Frequency
PSD
(dBV
)
Noise floor Noise floor
Vdd noise VddA noise
34
Slicer
PD
deserializer
PLL
dataOut
ref Clk
Phasecontrol
Phasemixeredge Clk
data Clk
RX
2x Oversampled Bang-Bang CDR
Generate early/late from dn,dn-1,enSimple 1st order loop, cancels receiver setup time
Now need jitter on data Clk, not PLL output
dn-1
dn
en (late)
dn
en
18
35
Data Clk NoiseModel phase selector and PLL
Base linear PLL jitterAdd non-linear phase selector noise from CDR
Model the CDR loop as a state machineThe current phase position is the stateState transitions are caused by early/lateJitter on input data and PLL means
Possible to be late and get early PD resultOften filter early/late to generate up/down
A.E. Payzin, "Analysis of a Digital Bit Synchronizer," IEEE Transactions on Communications, April 1983.
36
Transition Probabilities
0 50 100 150 200 2500
0.2
0.4
0.6
0.8
1
Accumulate-resetfilter, length 4Pr
obab
ility
Phase count
p-early
p-hold
p-late
p-no-validtransitions
p-up p-dn Example system:CDR loopResidual ISI
At edge -30dBVDesired phase
State = 133
On average move to correct positionBut probability of wrong movement is not smallNeed to find probability of at each phase location
19
37
Bang-Bang CDR Statistical Model
Need steady state probabilities of the statesHave the transition probabilities
Iteratively apply transition probabilities (Markov chain)Results will converge to a steady-state
iφ1−iφ 1+iφ0φ Lφ
iholdp ,
iupp ,
idnp ,
38
0 50 100 150 200 250
-15
-10
-5
0
Phase Count
log 10
Ste
ady-
Stat
e Pr
obab
ility
Bang-Bang CDR Model
Gives the probability distribution of phaseWhich is the CDR jitter distribution
20
39
Noise and Interference SummaryMany important sources of noise and interference
ISI, crosstalk, quantization, estimation, etc.Largest error comes from ISI
By factor of 10xTiming is noisy too
High frequency transmitter jitter is badCDR jitter needs to be considered
Especially if the data input is noisy
What is the impact on performance?
40
Agenda
Backplane channel reviewLink system models and noise
Previous standard approachesStatistical modeling
Performance analysis
21
41
50 60 70 80 90 100 110-300
-200
-100
0
100
200
300
time [ps]
volta
ge [m
V]
-30
-25
-20
-15
-10
-52PAM - lin. eq
ISI and CDR Phase Distributions
In ideal world, there would be only two dotsThis plot shows how these dots spread out
Vertical slice – ISI distribution per time offsetHorizontal weight – CDR phase distribution
42
Putting It All Together
To compare different designsCompare the voltage margin at given BER
Need to include all noise sourcesAccurate ISI distributionTransmit and receive jitterCDR jitterEQ quantization noiseReceiver offset
22
43
0 20 40 60 80 100 120 140 160-150
-100
-50
0
50
100
150
time [ps]
mar
gin
[mV]
-30
-25
-20
-15
-10
-5
BER Contours
Voltage marginMin. distance between the receiver threshold and contours with same BER
0 20 40 60 80 100 120 140 160-150
-100
-50
0
50
100
150
time [ps]m
argi
n [m
V]
-30
-25
-20
-15
-10
-5
5 tap Tx Eq 5 tap Tx Eq + 1 tap DFE
44
BER Contours PAM2 DFE PAM4 linear equalization
0 40 80 120 160250
200
150
100
-50
0
50
100
150
200
250
time [ps]
mar
gin
[mV]
-30
-25
-20
-15
-10
-5
0 40 80 120 160 200 240 280 320-250
-200
-150
-100
-50
0
50
100
150
200
250
time [ps]
mar
gin
[mV]
-30
-25
-20
-15
-10
-5
23
45
Model and measurements
-80-60-40-200 20 40 60 80
-14
-12
-10
-8
-6
-4
-2
0
log1
0(B
ER)
Voltage Margin [mV]
PAM4, 3taps of transmit equalization, 5Gb/s, 26” FR4 channel
46
Example channels
Legacy (FR4) - lots of reflectionsMicrowave engineered (NELCO)
0 5 10 15 20
-100
-80
-60
-40
-20
0
Atte
nuat
ion
[dB
]
frequency [GHz]
26" FR4, via stub
26" NELCO,no stub
(b)
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
24
47
Capacity achieving bit loading
0 5 10 15 20 250
1
2
3
4
5
6
7
8
9
10
frequency [GHz]
#bits
/Hz
Capacity with thermal noise
Nelco 105Gb/sFR4 70Gb/s
0 2 4 6 8 10 12 140
2
4
6
8
10Multi-tone data rates with thermal noise
Nelco 64Gb/sFR4 38Gb/s
#bits
/Hz
frequency [GHz]
Capacity is very bigPractical rates lower
low target BER<10-15
peak power constraintThermal noise – the smallest noise source
48
Capacity with link-specific noise
Effective noise from phase noiseProportional to signal energyDecreases expected gains
Still, capacity is much higher than data rates in today’s links (3Gb/s)
NELCO FR4
-25 -20 -15 -10 -5 00
20
40
60
80
100
120
140
Cap
acity
[Gb/
s]
log10(Clipping probability)
thermal noise
thermal noise and LC PLL phase noise
thermal noise and ring PLL phase noise
-25 -20 -15 -10 -5 00
20
40
60
80
100
120
140
Cap
acity
[Gb/
s]
log10(Clipping probability)
thermal noise
thermal noise and LC PLL phase noise
thermal noise and ring PLL phase noise
25
49
Multi-tone with integer bit loading
-25 -20 -15 -10 -5 00
10
20
30
40
50
60
70
80
90D
ata
rate
[Gb/
s] a) NELCOthermal noise
thermal noise and LC PLL phase noise
thermal noise and ring PLL phase noise
log10(Clipping probability)-25 -20 -15 -10 -5 00
10
20
30
40
50
60
70
80
90
Dat
a ra
te [G
b/s] b) FR4
thermal noise
thermal noise and LC PLL phase noise
thermal noise and ring PLL phase noise
log10(Clipping probability)
Peak-power constraint introduces large gap penalty to capacity (can go around with coding, but too expensive)Still pretty high data rates
NELCO FR4
50
Multi-level: Offset and jitter are crucial
thermal noise + offset
thermal noise + offset+ jitter
To make better use of available bandwidth, need better circuitsPAM2/PAM4 robust candidate for next generation links
0 2 4 6 8 10 12 14 16 18 200
5
10
15
20
25
30
Dat
a ra
te [G
b/s]
Symbol rate [Gs/s]
PAM16
PAM8
PAM4
PAM2
0 2 4 6 8 10 12 14 16 18 200
5
10
15
20
25
30
Symbol rate [Gs/s]
Dat
a ra
te [G
b/s]
PAM2
PAM4
PAM8
0 2 4 6 8 10 12 14 16 18 200
5
10
15
20
25
30
35
40
45
Dat
a ra
te [G
b/s]
PAM4
PAM16
PAM8
PAM2
Symbol rate [Gs/s]
thermal noise
26
51
Full ISI compensation too costly
0 2 4 6 8 10 12 14 160
2
4
6
8
10
12
14
16
18
20
Dat
a ra
te [G
b/s]
Symbol rate [Gs/s]
PAM16PAM4
PAM2PAM8
0 2 4 6 8 10 12 14 160
2
4
6
8
10
12
14
16
18
20
Symbol rate [Gs/s]
Dat
a ra
te [G
b/s]
PAM8
PAM4
PAM2
0 2 4 6 8 10 12 14 160
2
4
6
8
10
12
14
16
18
20
Symbol rate [Gs/s]
Dat
a ra
te [G
b/s]
PAM2
PAM4
PAM8
thermal noisethermal noise + offset
thermal noise + offset+ jitter
Today’s links cannot afford to compensate all ISILimits today’s maximum achievable data rates
52
ConclusionsBackplane links limited by the channelISI is large in baseband links
Can’t completely compensate(At least not with reasonable area/power)
Residual ISI also increases CDR jitterGenerally have low BER requirements
Accurate noise statistic important Many of large noise source are bounded
Power constrained transmitterPAM4 and PAM2 with simple DFE are attractive solutions
Still, capacity of these links is very bigSmart multi-tone?