Fast Eye-Diagram Analysis

吳瑞北, Ruey-Beei Wu
Rm. 340, Department of Electrical Engineering
E-mail: rbwu@ew.ee.ntu.edu.tw
url: http://cc.ee.ntu.edu.tw/~rbwu

S. H. Hall & H. L. Heck, High-Speed Digital Designs, ch. 13.

 Contents
• Introduction
• Peak Distortion Analysis
• Diffusion in RC Line
• Fast Lossy Line Characterization
• Arbitrary Tx-Line Systems

R. B. Wu
 Margin Calculation
Ideal sampling position
Timing skew

Jitter
Voltage Noise and
required comparator
input

Voltage offset

Ideal reference point

January 4, 2017 Peak Distortion Analysis

R. B. Wu
 Margin Calculation (zoomed)
Ideal sampling position Timing skew Jitter

Voltage Noise and

required comparator
Voltage
input margin

Voltage offset

Ideal sampling position

Time margin
January 4, 2017 Peak Distortion Analysis
R. B. Wu
 Worst-case eye calculation
• Eye diagrams are generally calculated empirically
– Convolve random data with pulse response of channel
– Pulse response is derived by convolving the impulse reponse with
the transmitted symbol
• For eye diagrams to represent the worst-case, a large set of
random data must be used
– Low probability of hitting worst case data transitions
– Computationally inefficient
• An analytical method of producing the worst-case eye
diagram exists
– Computationally efficient algorithm

January 4, 2017 Peak Distortion Analysis

R. B. Wu
 Peak Distortion Analysis

Ref.: J. G. Proakis, Digital Communications, 3rd

ed., Singapore: McGraw-Hill, 1995, pp. 602-603
(not much detailed info here)

January 4, 2017 Peak Distortion Analysis

 Interconnect Model
• Point to point differential desktop topology
• Differential, edge-coupled microstrip (10” @ 55Ω)
• 2 Sockets
• 2 Packages (2” @ 45Ω)
• 1pF pad capacitance
• 50Ω single-ended termination

pkg socket 10” μstrip socket pkg

January 4, 2017 Peak Distortion Analysis

R. B. Wu
 Differential S Parameters

January 4, 2017 Peak Distortion Analysis

R. B. Wu
 Eye diagram (100 bits @5Gb/s)
Random data eye (100 bits) ---
Random data eye (1000 bits) ---

January 4, 2017 Peak Distortion Analysis

R. B. Wu
 Sample pulse response
ISI+ ISI-

precursor cursor postcursor R. B. Wu

Worst-case 0

0110100100000

VWC 0   ISI 
R. B. Wu
Worst-case 1

0101100000

Wu
VWC1  cursorR. B. ISI 
 Worst-case Patterns

Worst-case 0  1 1 0 1 0 0 1
Worst-case 1  0 0 1 0 1 1 0 R. B. Wu
 Summary
• Given S Parameters and the corresponding
pulse response, the worst case eye shape can
be determined analytically
• Worst-case co-channel interference can also
be determined analytically
• Advantages – objective, exact, and
computationally efficient

R. B. Wu
Diffusion in RC Line
 Diffusion (RC line, i.e., L=G=0)
V ( z, t ) I ( z, t ) V ( z, t )
  R  I ( z, t );  C 
z z t
 2V ( z, t ) V
  RC
z 2
t

Diffusion equation
 2u 1 u z 4 Dt
  u ( z, t )  A   e  y2
dy  B
z 2
D t 0

where A and B should be determined by boundary conditions.

 
RC 4 t z
V ( z, t )  A   e  y2
dy  B
0
R. B. Wu
 Diffusion in a Semi-infinitely Long Line
V (0,0 )  V0 ;
V (, t )  0 for t  0
0
A e  y dy  B  V0  B  V0
2

  2V0
A e  y2
dy  V0  0  A  V0  0  A  
0 2 
2  
 
 
RC 4 t z

 0
 y2
V ( z, t )  V0 1  e dy   V0  erfc z RC 4t
 

R. B. Wu
 Pulse Response
• Input: a rectangular pulse
 2V ( z , t ) V
RC line   RC
z 2 t
V ( z, t )  V0 erfc
t t0
 RC
4t

z  V0 erfc  RC
4 ( t t 0 ) z 
2  RC 4(t to ) z  y 2
  RC 4t z
 V0  e dy

Features:
. Delay is proportional to t2
. Long tail causes significant
inter symbol interference

R. B. Wu
Ref.: N. N. Rao, Elements of Engineering Electromagnetics, 6th Ed., Sec. 7.5
Fast Eye-Diagram Analysis
for Lossy Tx-Line
Ref.: W.-D. Guo, J.-H. Lin, C.-M. Lin, T.-W.
Huang, & R.-B. Wu, “Fast methodology for
determining eye-diagram characteristics of lossy
transmission lines,” IEEE Trans. Adv. Packag.,
Feb. 2009.
(Best Paper Award)

19
 Matched Lossy Tx-Line Systems
Microstrip line Rin ≈ ∞
Tx ГS
( Z0 ≈ 50Ω, Length : l )
ГL
VO
PRBS : RS=50Ω 8mil (W)
Rx
2mil (T)
5Gbps, and
tr = 50ps.
+ VS 5mil (H) εr = 4.4
RL=
50Ω
tan δ = 0.02
Metal : Copper

High

Monotonic Step Response

Voltage

Step Input
Signal If the tail is too wide, serious inter-symbol
interference (ISI) will be induced.

Lift up to the “High” level

Low
R. B. Wu
20
Bit
Period Time
 Worst-Case Eye-Diagram
0.4
30”-Long Line 0.4
Initial state 0

0.3 0.3

Volta ge (V)
Voltage (V)

0.2 0.2
Upper Bound
0.1 0.1

0 0
0 2 4 6 8 10 0 50 100 150 200
Time (ns) degraded by ISI p )
Time (ns

Initial state 1
0.4 0.4

0.3 0.3
Volta ge (V)
Voltage (V)

Lower Bound
0.2 0.2

0.1 0.1

0 0 R. B. Wu
21
0 2 4 6 8 10 0 50 100 150 200
Time (ns) Tim e (nps )
 Worst-Case Bits Pattern
Eye shape
Using only two anti-polarity one-bit
data patterns as the input signals can
Still lack of jitter simulate the worst-case eye diagram
Single information for the transmission-line system with
Bit a monotonic step response.

Worst-Case Eye Diagram

0.4
Voltage

0010000... 0.3
Voltage (V)
+ 1101111...
0.2
00010000...
+ 11101111...
0.1

Time 0 R. B. Wu
22
t1 t2 0 100 200 300 400
t3 Time (ps)
 System Transfer Function
R(w) jwL
Z0 (w) ,
Z0 (w) H (w) 1 L (w) G(w) jwC
VO (w) VS (w), where
Z0 (w) RS 1 H (w) S (w) L (w)
2
RS Z0 (w) RL Z0 (w)
S (w) L (w)
RS Z0 (w) , RL Z0 (w) .

With assumption of low-loss line Z0 (w) L/C 50 typically and RS RL 50 for

nearly matched termination,
H (w)
VO (w) VS (w).
2
( R jwL ) ( G jwC ) l
Generally, H (w)
gl
e e Low-loss assumption

RC GL RG
g (R jwL) (G jwC) RG jw (RC GL) jw 2 LC jw LC 1
jwLC jw 2 LC
RC GL RC GL R C G L
jw LC 1 jw LC 1 jw LC ( g1 g2 g3 )
jwLC 2 jwLC 2 L 2 C

g1 l g2 l g3 l
H (w) e e e h(t) h1(t) h2 (tR.
) hB.
3 (t )Wu
23
Fourier transform
 Impulse Response
(A). Time-delay term
1 g1 l ] 1 jw LC l ] δ(t) δ(t-t0)
h1 (t ) [e [e
1 b
lim (e jw LC l )e jwt dw t0
2p b b
d(t t0 ) , t0 LC l LC l time

(B). Conductor-loss term 0.9 W 0.16

0
RS (microstrip line)
R C RS jw C W T H
1 g2 l ] 1
l 1
l [8]
h2 (t ) [e [e 2 L ] [e 2 L ]

b RS jw C
1 l
L )e jwt d w
lim (e 2 7mil
2p b b
4mil
RS s C FR4
1 b l
lim (e 2 L )est ds
2p j b b A=1.6*10-6
RS l
A ( s) RS s C
pL l A
4 1
[e 2 L ] u(t )
R. B. Wu
C p A2
24
t3 e t
 Impulse Response
er ere 1
Gd C tan de , tan de tan d
(C). Dielectric-loss term ere er 1 [9]
G L Gd |w| L
l l
1 g3 l 1 2 C ] 1 2 C ]
h3 (t ) [e ] [e [e B=4.2*10-11 7mil

4mil FR4

b Gd |w| L
1 l
2 C )e jwt d w tanδ=0.02
lim (e
2p b b
Gd L B
B l (s)
2 C Non-Causal
p( B 2 t 2 )

(C-1). Kramer-Kronig Relation

In early days, K-K relation was proposed to solve the problem that the derived electric
susceptibility of dielectric material do not satisfy causality in time domain.

1 2 w ' Im G(w ')

g (t ) G(w) e jwt d w
Re G(w) P dw '
2p p 0 w' w
2 2
1 2w (w ')Wu
ReR.GB.
G(w) g (t ) e jwt dt Im G(w) P d25w '
2p p 0 w' w
2 2
 Impulse Response
x 1
1 H g (x ) 1 e
H3 ( ) e B
e B
e Hg ( ) p [1 t / B 2 ]
1
B h3 (t ) hg ( z)
p [1 z 2 ]
through K-K relation
1 0.7

0.6
0.75
hg(z) Gibbs phenomenon

Impulse Response (V/s)

Real part 0.5
0.5
Imaginary part 0.4
hg,K-K(z)
Hg(ξ)

0.25 0.3

0.2
0
0.1
-0.25
0

-0.5 -0.1
0 1 2 3 4 5 6 7 8 9 10 -5 -4 -3 -2 -1 0 1 2 3 4 5
x (w B) z (t / B)

1[ H (w)] 2B
h4 (t ) 3 u(t ), Causal
p (B t )
2 2

where h4 (0) 1 p B R. B. Wu
26
 Normalized Impulse Response

Conductor Impulse Response Dielectric Impulse Response

A 2B
h2 (t ) u (t ) h3 (t ) u(t )
2
t3e A / t p B t
2 2

0.08 0.08 0.7

0.06
tmax. 0.67p A2
0.07 0.6
0.04
0.06
0.02 0.5

0.05 0 unit (sec)

A2 * h2(t)

0.4

B * h3(t)
0 5 10 15 20 25 30
0.04
B = 10*10-12
unit ( sec ) B = 30*10-12
0.3
0.03
A = 5*10-7 B = 50*10-12
A = 10*10-7 0.2 B = 70*10-12
0.02 A = 15*10-7
0.01
A = 20*10-7 0.1

R. 200
B. Wu
0 0

0 50 100 150 200 250 300 0 50 100 150 27

250 300
Time (ps) Time (ps)
 Characteristic Charts
Eye-Diagram Characteristics vs. An, Bn Vo (t) Vin (t) h(t), where
Vin (t) 0.5 VS (t)
A 2B
h(t ) d(t t0 ) u(t ) u(t ) Microstrip line
p A2 p (t 2 B2 ) RS=50Ω ( Z0 ≈ 50Ω, Length : l ) VO
t3 e t
A RS l B Gd Z0 l 2mil W
An
UI 4Z0 p UI
, Bn
UI 2UI
+ VS H εr = 4.4
RL=
50Ω
tan δ = 0.02
Metal : Copper
UI : Unit Interval, tr = 0.25UI

R. B. Wu
28
0 0
 2mil W
H εr = 4.4
tan δ = 0.02
Effects of Risetime Length : l

Effects of Rising Edge and Bit Period UI = 200ps

tr 80

70
W=76mil, l = 20"

Eye Height / Vin (%)

UI W=8mil, l = 20"
-20dB/decade 60 W=18mil, l = 40"
V( f )

50

Error 3%
40
-40dB/decade
30

-1 0 20 40 50 60 80 100
(UI) Freq. Signal rise time tr (ps)
100

Eye Width / Bit Period (%)

95

Error 0%
S21 (dB)

90

85
W=76mil, l = 20"
80 W=8mil, l = 20"
W=18mil, l = 40"
75

70 R. B. Wu
29
(πUI)-1 (πtr)-1 Freq. 0 20 40 50 60 80 100
Signal rise time tr (ps)
 Numerical Verification
2mil 8mil 2mil 18mil
l = 20” 5mil εr = 4.4 l = 30” 10mil εr = 4.4
tan δ = 0.02 tan δ = 0.02
Metal : Copper Metal : Copper
0.4 0.4

HSPI 0.3 0.3

Voltage (V)
Voltage (V)

CE :
PRBS 0.2 217 mV 177 ps 0.2 170 mV 154 ps
5Gbps, and 0.1
0.1
tr = 50ps.
0 0
0 80 160 240 320 400 0 80 160 240 320 400
Time (ps) Time (ps)

Proposed 215 mV 172 ps 166 mV 152 ps

Algorithm

R. B. Wu
30
 Numerical Results (5/6)
Table I. Dimension settings and An -Bn parameters of microstrip lines (T=2mil).
Case # Length (inch) Height (H) Width (W) An Bn 2mil W
1 5 mil 8 mil 0.064 0.135 H εr = 4.4
tan δ = 0.02
2 20 10 mil 18 mil 0.029 0.138 Length : l
3 40 mil 76 mil 0.007 0.140
4 5 mil 8 mil 0.127 0.270
5 40 10 mil 18 mil 0.058 0.275
6 40 mil 76 mil 0.014 0.279
|Error| < 3%
Table II. Eye-diagram analysis by HSPICE’s and the proposed algorithm’s result.
Case Microstrip (T=2mil) Eye-Opening Height Eye-Opening Width
# Len. Height Width HSPICE’s Algorithm’s Error HSPICE’s Algorithm’s Error
1 5 mil 8 mil 219.3 mV 215.4 mV -1.8% 176 ps 172 ps -2.3%
2 20” 10 mil 18 mil 275.2 mV 271.5 mV -1.4% 190 ps 187 ps -1.6%
3 40 mil 76 mil 310.8 mV 308.3 mV -0.8% 194 ps 192 ps -1%
4 5 mil 8 mil 68.1 mV 66.4 mV -2.6% 97 ps 95 ps -2.1%
5 40” 10 mil 18 mil 159.3 mV 156.3 mV -1.9% 152 ps 149 ps -2.0%
6 40 mil 76 mil 228.6 mV 223.8 mV -2.1% 176 ps R.
172 ps B. -2.3%
Wu 31
 Maximally Usable Length

Here, l1  27", l2  36"  l  27"

2mil 8mil
RS l
5mil εr = 4.4 An al Eye-diagram
tan δ = 0.02 4Z0 p UI Find usable
specification
Metal : Copper Gd Z0 l length
Bn bl
UI = 200ps; 2UI EH : 40%
l min(l1, l2 )
a 0.13; b 0.27 EW : 60%
tr = 0.25UI = 50ps

l2

l1

0 0 R. B. Wu
32
 Experimental Verification (1/4)
50 line A RS l B Gd Z0 l
An , Bn
2mil
0.72mm UI 4Z0 p UI UI 2UI
Case 1 0.4mm εr = 4.4
tan δ = 0.02 l 20 inch 30 inch 40 inch
Metal : Copper line Comp. Meas. Comp. Meas. Comp. Meas.
1.9mm H=0.4mm, An 0.01 0.014 0.016 0.022 0.021 0.029
2mil
Case 2 W=0.72mm Bn 0.14 0.17 0.21 0.27 0.28 0.35
1mm εr = 4.4 H=1mm, An 0.004 0.0061 0.006 0.0092 0.008 0.0122
tan δ = 0.02
Metal : Copper W=1.9mm Bn 0.14 0.16 0.21 0.25 0.28 0.33
H= 0.4 (m m ) S 21 H= 1.0 (m m ) S 21
0 0

Case 1 Case 2
-10 -10
(dB)
(dB)

)
)

S21(dB
S21(dB

-20 20 inch -20 20 inch

30 inch 30 inch
-30 20
40(in)
inch -30 20
40(in)
inch
30 (in) 30 (in)
Measured
40 (in)
Curve Fitting Measured
40 (in)
-40 -40
0 2 4 6 8 10 0 2 4 6 8 10
Frequency (GHz)
Freq. (GHz) Frequency
Freq. (GHz )R. B. Wu
(GHz)
33
(a) (b)
Agilent E8364B
 Experimental Verification (2/4)
Environment for eye-diagram Measurement

Anritsu Pulse Pattern Generator MP1763C

Agilent infiniium 54855A DSO
Voltage [100 mV/div]

DUT
30 ps 325 mV

100 ps
Agilent 11500E
Time [50 ps/div]
R. B. Wu
34
Without DUT
 Experimental Verification (3/4)

Length = 20” Length = 30”

Case 1
0.72mm
2mil
0.4mm εr = 4.4
tan δ = 0.02
Metal : Copper

EH = 220mV, EW = 155ps EH = 150mV, EW = 147ps

Case 2
1.9mm
2mil
1mm εr = 4.4
tan δ = 0.02
Metal : Copper

EH = 240mV, EW = 165ps EH = 186mV, ER. B. Wu

W = 155ps35
 2mil W
H εr = 4.4
tan δ = 0.02

Experimental Verification (4/4) Length : l

|Error| < 5%
Eye-Diagram Height (mV) Eye-Diagram Width (ps)
H=0.4mm, H=1mm,
W=0.72mm W=1.9mm Predicted Measured Error (%) Predicted Measured Error (%)
l
1 217.8 220.0 -1.36 % 156.4 155.0 +0.09 %
20" 1 4 2 156.0 150.0 +4.0 % 144.5 147.0 -1.70 %
3 113.8 118.5 -3.97 % 128.4 133.0 -3.46 %
30" 2 5
4 227.5 239.8 -5.13 % 160.0 165.0 -3.03 %
40" 3 6 5 182.0 185.7 -1.99 % 150.5 155.1 -2.97 %
6 143.0 149.5 -4.35 % 139.4 144.6 -3.60 %

3 3

2 2

1 1
6 6
5 5
4 4

R. B. Wu
36
 Summary
• Propose a much faster methodology that only uses two anti-polarity one-
bit patterns as input signal to simulate the worst-case eye diagram at the
receiving end of transmission-line system with a monotonic step response.
• Introduce Kramer-Kronig relations to resolve the non-causality of the dielectric
loss related impulse response.
• Construct two design graphs for eye-diagram characteristics vs. An and Bn of
lossy transmission line.
• The variation of signal rising/falling edge will cause the eye-opening difference of
at most 3%, and the maximally usable length of transmission line under a
certain signal specification can be evaluated.

0.7
0.4
0.6
hg(z)
Impulse Response (V/s)

0.3 0.5
hg,K-K(z)
Voltage (V)

0.4

0.2 0.3

0.2
0.1 0.1

R. B. Wu
0
0 100 200 300 400 -0.1
-5 -4 -3 -2 -1 0 1 2 3 4 5
Time (ps) z (t / B) 37
 Convergence Problem and Modified
K-K Relation
• Modified mag-phase K-K relation:
H    R    exp  j     ln  H     ln  R     j  
D: unknown const
2  B   0  2
    
B 
Re B  ln    B    ln   D  
  B0  B0 
Normalized h3
 
d   j  ln  j t  D d 
2
h t   H   e jt e  
0.4


 e
2 

2 2.31
D=0
D=2.31
0.3
t
  B , t 
B B *h 3 0.2

The unknown constant D can

0.1
be determined numerically.
0
R. B. Wu
-20 -10 0 10 20
t/B
R. K. AHRENKIE "Modified Kramers-Kronig Analysis of Optical Spectra," JOSA, Vol. 61, Issue 12, pp. 1651-1655 (1971)
 Comparison with Previous Solution
H3 Comparison
1
Re(H3) modified
2B 2
Re(H3) previous
B  h3  u t  or
0.5 Im(H3) modified
 B  t 
2 2
Im(H3) previous

 B   j B ln  B  
2
d  B 
H3

B  h3  t  
0
j  t  2.31 B 
e


e
2
-0.5

Normalized h3
-1
0 0.5 1 1.5 2 2.5 0.7
B *f tmax  0 Modified h3
0.6 Previous causal h3
H3 Comparison
Previous noncausal h3
1 0.5
modified
previous 0.4

B *h 3
0.8
tmax  1.885  B
0.3
Magnitude

0.6
0.2
0.4 0.1

0.2 e  B  2 f 0
-10 -5 0 5 10
t/B
0
0 0.5 1
B *f
1.5 2
R. B. Wu
2.5

• Proposed modified method keeps same magnitude in frequency domain.

 Verification w/ Circuit Simulator
No dielectric loss, B=0 No conductor loss, A=0
H verification, Zc = 51.1  H verification, Zc = 50.4 
0 0

H fitted -10
H ADS
Magnitude (dB)

Magnitude (dB)
H fitted, previous
-5 -20

-30
H fitted
-10 -40 H ADS
H fitted, previous
8mil -50

-15
0 5 10 15 20
2mil -60
0 5 10 15 20
εr=4.4
Frequency (GHz)
length=40inch 5mil Frequency (GHz)
Pulse Response Pulse Response
0.35 0.4
0.3 Modified
Modified Previous
0.25 0.3
Previous ADS
Voltage (V)

Voltage (V)
ADS
0.2
0.2
0.15

0.1
0.1
0.05

0 0
5.5 6 6.5 7 5.5 6 6.5 7
time (ns) time (ns)
R. B. Wu
 Eye Height/width Contour
Eye width /UI (%), tr /UI = 0.25
0.15 0
1
70 50 200
80

30
60
40
0
24
90

0.1 10
50
80 70
60
20
30
40
% Pulse Response
An

0.25
90 50 Modified
80 70
0.05 Previous
60 0.2 ADS

Voltage (V)
70 0.15
80 6%
90

0
0 0.05 0.1 0.15 0.2 0.25 0.3 31% 0.1
Bn
0.05
Veye /Vh (%), tr /UI = 0.25
0.15 4 0
5.5 6 6.5 7
0 30 0
time (ns)
20
50 10

0.1
60
40 30 0 4% 8mil
20
An

50 10
70
60
40 30 2mil
0.058
0 20 εr=4.4, tanD=0.02
70
50
length=40inch 5mil
90
40
80 60 30
σ=5.78e7 S/m
0
0 0.05 0.1 0.15 0.2 0.25 0.3

R. B. Wu
Bn
 IEEE Electrical Performance of
Electronic Packing and Systems

Enhanced Eye Height Estimation with

Mismatched Lossy Transmission Lines

Shih-Ya Huang , Yung-Shou Cheng , Bob Liu , and Ruey-Beei Wu

Department of Electrical Engineering and Graduate Institute of Communication

Engineering, National Taiwan Univ., Taipei, Taiwan.
e-mail: f00942001@ntu.edu.tw and rbwu@ew.ee.ntu.edu.tw

Oct 21-24, 2012 / Tempe, Arizona

 Peak Distortion Analysis (PDA)
Transmission line system ( TD > UI ) : Single pulse response :
UI UI UI UI UI UI UI
Vo

0 -2V Transmission line

s o ISI+

Vworst _ 1  VH  | ISI  | t
-
Vworst _ 0   ISI
TD
 ISI

EH ,wc  Vworst _ 1  Vworst _ 0

VH  Cursor
 VH   ISI    ISI  Sampled value at
main signal
ISI  ,ISI   Post- Cursor
[3] B. K. Casper, M. Haycock, and R. Mooney, “An accurate and efficient analysis
method for multi-Gb/s chip-to-chip signaling schemes,” in IEEE VLSI Circuits Symp.,
R. B. Wu
43
June 2002, pp. 54–57.
 
Ideal Tx-Line , TD> UI, TD  N UI
TD=N UI, TD >UI  S and O : same polarity

0 -2V Ideal Transmission Line


s o ISI
Only with
V0
V0 (1+Γ) o
V0Γo = VH
V0ΓoΓs  S and O :opposite polarity
VHΓΓ s o
2
V0Γs Γo
2 2 
V0ΓoΓs With ISI  and ISI
2 2
VHΓΓ s o

V0 Γo3Γs2 .
... ..

  S O 
EH ,wc  VH   ISI   ISI  VH  1 
 

 1   S O
 
R. B. Wu
44
It’s a good approximation for general TD  UI , even when TD  N  UI
 Lossy Tx-Line, TD＞UI
The reference values of the post-cursors
can be found at the main pulse.
 V      , V  V     
m m
2 2
VpcL ,m pL ,m S O pcR ,m pR ,m S O

  
M m
Only main EH ,wc  VH  VA0   VA0  V pL,m  V pR,m   2  S O
m 1
pulse and α
are needed!   2m  TD  %UI
There would be a time mshift
m : number of reflected pulses .

R. B. Wu
45
 Loss constant α
Matched Tx. Line :
2
0 -2V 50 ohm Lossy 0 -2V 50 ohm Lossy Transmission Line
Transmission Line

s= 0 o= 0 s= 0 o= 0

1.0

0.8
Voltage (V)

0.6
VH 2

0.4 VH 1
0.2
TD TD
0
0 1 2 3 4 5 6 7 8
Time (nsec)

46
R. B. Wu
 Eye-Height Determination
─ General Mismatched Tx Line
Step 1 : Matched Pulse response
Mismatched tx-line system
1
50 Ohm Microstrip line
0-2V
Γs Γo 2

4
3

Mismatched 4

2
Pulse Response
1

TD

Step 2 : Proposed formula

:
 match
 
m

M
EH ,wc  1   S 1  O   VH  VA0   VAmatch
match
0  V pL,m
match
 V pR,m
match
  2 S O 
 m 1 

R. B. Wu
M=10 would be adequate
47
 Contour Map for Best Eye Height
• For arbitrary Tx-line system, a Eye Height v.s. Γs and Γo
1
general solution space is given
0.8
to facilitate termination design.
0.6
• Two trend-lines : 0.2
0.4 0.4
Source-end matching 0.6
0.2
Load-end matching 0.8

s
0 1
1.6 1.8
• Best eye height region is marked -0.2
1.2
Drop
as hatched region. (RS<Z0<RO) 1.4
-0.4 rapidly
1.4
• Drop-off rapidly when Γs<-0.6, -0.6
1.2
1
0.8
0.4

Γo>0.4 -0.8
1.6 0.6
0.2

1.8
-1
-1 - 0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
30 cm o
50 Ohm Micro strip line
0-2 V 0.75 (W)
0.01 (T)

s 0.4 (H)
o R. B. Wu
，
Metal: Copper ， Unit=mm
48
Testing Environment

Anritsu Pulse 3
Pattern Generator 107 
MP1763C
5

Tektronix
CSA8000B
Communication
Signal Analyzer

80A03 Probe
Interface + P7313
Differential Probe Differential
Probe

R. B. Wu
49
 Measurement Results

0.238 V

0.584
V

Measure EH=0.238 Measure EH=0.584

Predicted
V EH= PredictedVEH= 0.590
Error = 2.94 %
0.245V Error =V1.03 %
R. B. Wu
50
Peak Distortion Analysis and
FIR Filter for Arbitrary Lines
Y.-S. Cheng and R.-B. Wu, “Direct eye diagram
optimization for arbitrary transmission lines using
FIR filter,” IEEE Trans. Compon., Packag.,
Manuf. Technol., vol. 1, pp. 1250-1258, Aug.
2011.

51
 Peak Distortion Analysis for Arbitrary Lines

• Downward Response: Step response drops to steady at late time.

Voltage
– sM1: 1st max. before it drops to steady
S
sm2: min. before it grows.
sM2: max. before it drops again, …
– Treat 1st min. at steady, and sm1 = s∞. 0 1 2 3 4 5 6 7 8
Time (UI)

• Upward Response: Step response rises to steady at late time.

– sm1: 1st min. before it grows to steady

Voltage
S
sM1: max. before it drops.
sm2: min. before it grows again, …
0 1 2 3 4 5 6
Time (UI)

R. B. Wu
Y.-S Cheng, and R.-B. Wu, “Direct eye diagram optimization for arbitrary transmission lines
using FIR filter,” IEEE Trans. Comp., Packag., Manuf. Technol., 2011. (accepted)
 Fast Eye Diagram Analysis
 lb k 1 k Voltage

VH   smi   sMi

Variation S

 i 1 i 1  EHwc  VHlb  V0ub  2VHlb  s VHub

V lb  s  V ub VHlb EHwc
 0  H
tJ 1
Timing
t J 2 Jitter
Vth
ub Eye Width
k 1 k V0

V ub
H   sMi   smi V0
lb

i 1 i 1
0

V 0
ub
 s  V lb
H
0 0.2 0.5 0.8 1 Time (UI)

Tap Coefficient Optimization System Step

response
s (t )
• The output of FIR filter
N N n-tap FIR filter nà n+1
Define obj. function
y (t )   bk  x(t - kT ) sr (t )   bk  s(t  kT ) n=1
k 0 k 0

• Tap coefficients of FIR filter

Direct Search
bk   b0 , b1 , bN  b[n]=[b0 b1…bn]

• Obj. function for the optimized No

Yes
FIR filter design by sr (t ) Worst-case eye
diagram simulation
Objective
function
obj (bk )  VHlb VHub
R. B.different
Wutap num.
Fast eye diagram Optimal b[n] for
analysis
 Optimal FIR Filter Design
Rin  
TX Z 0  50, Length: FIR filter as pre-emphasis
Vin VO
ZS Pre-emphasis Channel RX
7 mil
ZL
VS
1 mil

4 mil εr = 4.4, tanδ = 0.02

Metal: Copper

b[n]=[1, -0.091, -0.158, -0.0325] b[n]=[1, 0.34, 0.453]

2

w/o FIR filter 1.6 w/ FIR filter

Voltage (V)
1.2 w/o FIR filter
w/ FIR filter

0.8 Step Reponse

w / o FIR filter

ZS = 120 Ω
Step Reponse ZS = 18 Ω w / FIR filter
w / o FIR filter 0.4 Sampled Voltages
w / FIR filter Local maximum

CL = 0.2pF Sampled Voltages CL = 0.2pF Local minimum

Local maximum 0
Local minimum 0 1 2 3 4 5 6 7 8 9 10
Time (UI)

281.4 mV
399.5 mV 1046.8 mV
166.9 mV
157.5 ps 172.0 ps 147.4 ps 192.1 ps

R. B. Wu
 Experimental Verification
Worst Case Eye Contour

Voltage [100mV/div]
0.4

0.2

Voltage (V)
0 Eye diagram closure
Eye diagram closure
-0.2

-0.4

-0.2 0 0.2 0.4 0.6

Time (ns) Time [125ps/div]
b[n]=[1, 0.34, 0.453]
1
Worst Case Eye Contour
w/o FIR filter

Voltage [100mV/div]
0.8 0.4
Voltage (V)

w/ FIR filter 0.2

Voltage (V)

182.7 mV
0.6
0 472 ps 193 mV
448.1 ps
-0.2
0.4 Step Reponse
w / o FIR filter -0.4
w / FIR filter
0.2 Sampled Voltages
Local maximum
-0.2 0 0.2 0.4 0.6 Time [125ps/div]
Time (ns)
Local minimum

0
0 1 2 3 4 5 6 7 8 9 10
Time (UI)
R. B. Wu
Y.-S. Cheng & R.-B. Wu, “Direct eye diagram optimization for arbitrary transmission lines
using FIR filter,” IEEE T-CPMT, 2011. (accepted)
 2-Tap FIR Filter
FIR: y(n) = b0 * x(n) + b1 *x (n-UI) |b0| + |b1| = 1
11 Data b0  1  b1  1
Cursor Main Output
x(n)
Driver (b0) y(n)

1 bit de-emphasis
Delay Driver
Post-Cursor
b1
Desired tap coefficients
Pulse response
Step response
 b0  b1   Vsat VFIR b0
VFIR b0
(b0  b1 ) Vsat

Early Settle
0 t
0 t 0 1UI 2UI
0 1UI 2UI
 b1
R. B. Wu
 Passive FIR Realization

Realization of 2-tap FIR

by single-stub tx-line

1UI 1UI VFIR

RS=Z0 X Vo
Z1 Z1 Zo = 50 Ω
Vin
VS ～ Zh 0.5UI 7 mil
1 mil RL
Design parameters:
A 4 mil εr = 4.4, tanδ = 0.02
RT
Metal: Copper Z1; Z h ; RT

Equivalent circuit representation

R. B. Wu
Y.-S. Cheng and R.-B. Wu, “Passive FIR filter design using reflections from
stubs for high speed links,” EDAPS, Hangzhou, China, Dec. 2011.
 Equalizer Design
1UI 1UI V
 VS : PRBS, RS= Z0 X FIR Vo
Z1 Z1 Zo = 50 Ω, l = 20 in.
tr/tf=30ps, 2V, RL= Z0
8Gbps
VS ～ Zh 0.5UI 7 mil
1 mil
A 4 mil εr = 4.4, tanδ = 0.02
 RS=RL=Z0=50 RT
Metal: Copper

 Z1  69,

 Z h  165,
 R  31
 T 100%

Voltage (V)
Voltage (V)

204mV Equalizer 415mV

113ps
69ps

64%

R. B. Wu
Time (ps) Time (ps)
 Equalizer Design 2 – Unmatched
Source
1UI 1UI V
RS
X FIR Vo
Vs : V p  p  2V , Z0 Z0 Zo = 50 Ω, l = 20 in.
RL= Z0
8 Gbps
VS ～ Zh 0.5UI 7 mil
1 mil
RL  Z 0  50 A 4 mil εr = 4.4, tanδ = 0.02
RT
Metal: Copper

Design parameters: Rs ; Z h ; RT
Simulated
 Mismatched RS Topology  RS  23, by ADS

 Z h  115,
 R  21
 T

Voltage (V)
Voltage (V)

309mV 565mV
111ps
72ps

Time (ps) R.(ps)

Time B. Wu
Application Examples
 Application to Via-Stubs in IPC
eq. ckt Top Top
+ +
Port 1 Gnd Port 2
- Via Via -

In-1
25 cm
0.7 mm Via Stub Via Stub 0.7 mm

L L
|H(ω)|
2Rt
High-Pass Response
• QPI (CPU to CPU) : 0dB
Heq
Length: 26~29cm
Lossy Response
trise: 53.5ps(10~90%) Equalized Response
(Low-Pass)
(Low-Pass)
32.8ps(20~80%) Hchannel
Data rate: 6.4Gbit/s f(GHz)
0
fopt 71 of
R. B. Wu
Y.-S. Cheng, et al., “SI-aware layout and equalizer design to enhance performance
high-speed links in blade servers,” EPEPS, San Jose, CA, pp. 199-202, Oct. 2011
 Eye Diagram Comparison
Data Rate=10 Gbit/s, tr= 33 ps
Measurement
w/o equalization Simulation
Voltage (V)

279 mV
75 ps

Time (ps)

w/ equalization

Voltage [120 mV/div]

Voltage (V)

321 mV
343 mV
90 ps
82 ps

Time (ps) Time [20 ps/div]

R.72 B. Wu
 Next Generation Wide I/O Memory on
3D IC
 3D IC is a 3-dimensional integration of the hetero or homogeneous
chips by connecting them vertically in one package.
On-chip On-chip
DeCAP DeCAP
On-chip DeCAP :
(tens of pF)
Memory
Digital IC RF Micro-Bump

Silicon Interposer TSV

Bump
Package/PCB
• Reduced RC Delay
System Performance
• Reduced Power Consumption &
• Reduced Area Consumption Packaging Density
R. B. Wu
73
 Signal/Power Integrity Design
Signal integrity issue:
Signal will be distorted when propagating through TS
Eye diagram:
Case 1:

Tx:

Eye Height: 0.164V

Rx: Jitter: 0.7ps

TSV: diameter 10 μm bump:diameter 20 μm

pitch 40 μm pitch 40 μm
height 100 μm height 20 μm
R. B. Wu74
• TSVs, micro bumps and ideal RDL are included in the simulation.
• Transmitted signal: Pseudo-Random Binary Sequence (PRBS) at 20 Gb/s with rise/fall time 5ps.
 Eye Diagram Analysis
± Vin: 0.5V
Frequency:10Gbps
210-1 PRBS
Rise/Fall time:20ps
1 stacked TSV
2 stacked TSV 4 stacked TSV 8 stacked TSV

R.75 B. Wu
 TSV Parameters
• Electrical modeling

Structural parameters
• TSV diameter: 2a
• TSV height: h
• TSV-to-TSV pitch: 2d
• Insulator thickness : b-a

76
R. B. Wu

 SiSiO
Si 2

Simplified Circuit Model

0
1-stacked TSV

-2
|S21| (dB)

Full eq. ckt.

-4 Simplified eq. ckt.

Oxide TSV
TSV TSV
thickne DIAMET
10-stacked TSV PITCH length SiO2 Si Si
-6 ss ER
(P) (LTSV)
(Tox) (D)

20
0.2 m 100 m 50 m 100 m 3.9 11.9
S/m

-8
0.1 1 10
Frequency (GHz)
R.77 B. Wu
 RC Equalizer
Z02
Req 
RSi _ t
( Z0  RSi _ t ) RSi _ t
Ceq  Cox _ t
Z02

0 0

-2 Compensated result -2
RC equalizer Compensated result
RC equalizer
|S21| (dB)

TSV

|S21| (dB)
TSV
-4 -4

Single TSV 10-stacked TSV

-6 -6

-8 -8

0.1 1 10 0.1 1 10
Frequency (GHz) Frequency (GHz)

78equalizer design for through silicon vias

R. B. Wu
R.-B. Sun, C.-Y. Wen, and R.-B. Wu, “Passive
with perfect compensation,” T-CPMT, pp. 1815-1822, Nov. 2011
 0
Equalized Simulation Results 0
-1 -1
-2 -2
4 stacked
-3 -3 TSV
-4 Before equalizer -4
|S21|(dB)

|S21|(dB)
After equalizer
-5 Equalizer response -5
-6 -6
-7 -7 Before equalizer
-8 -8 After equalizer
1 stacked TSV Equalizer response
-9 -9
-10 -10
0.1 1 10 0.1 1 10
Frequency(GHz) Frequency(GHz)

0 0
-1 -1
-2 -2
-3 -3 Before equalizer
After equalizer
-4 -4
|S21|(dB)

|S21|(dB) Equalizer response

-5 -5
-6 Before equalizer -6 8 stacked
After equalizer
-7 Equalizer response -7 TSV
-8 -8
-9 2 stacked TSV -9
-10 -10
0.1 1 10 0.1 1 10
Frequency(GHz) Frequency(GHz)

R. B. Wu
Improved eye height, & nearly ZERO jitter!
 Conclusions
• Freq.-dependence in loss incurs “long tail” response,
thus ISI and eye diagram deterioration.
• Peak distortion analysis and analytic derivation gives
universal design curves for eye height/width.
• RL/RC equalizers are proposed for conductor/
dielectric loss dominant tx-lines.
• Generalization to tx-lines with unmatched load,
crosstalk, or faster edge is done by PDA & FIR filter.
• Topologies for next generation interconnects need
further study, e.g., RF, AC-coupled, optical,
nanowire, ... R. B. Wu
80