intra-pair “unaccounted” skew - url.t · • laminate rotation to reduce skew was modeled and...
TRANSCRIPT
Intra-pair “Unaccounted” Skew –
Effects and Suppression
Peerouz Amleshi, Molex Inc. [email protected] 1-(630)-527-4309
Cong Gao, Molex Inc. [email protected]
Davi Correia, Molex Inc. [email protected]
Lin Shen, Insieme Networks Inc. [email protected]
8-FR3
Contents
• Introduction
– Goal
– Methodology
• 3D Full-wave Modeling
• Skew Minimization techniques
• Channel Simulation
– S-parameter Models
– Link Simulation
• Conclusion
Introduction • Goal
• Characterize skew generated from glass weave structures in PCB laminates.
• Investigate possible techniques to reduce glass weave effect .
• Assess the effect of skew in transmission quality of a simple link of known
length/loss.
• Methodology
• Find skew through construction of glass weave unit cell and utilization of a 3D full
wave simulation tool.
• Run statistical eye algorithm on constructed channels of known loss and skew..
3D Full-wave Modeling
The definition of glass weave unit cell (blue geometry).
3D Full-wave Modeling- Glass weave Models
Trace Glass
(Parametric geometry provided in ANSYS HFSS software)
2116 glass weave based symmetrical differential trace laminates
2x1080 glass weave based symmetrical differential trace laminates
3D Full-wave Modeling- Trace configurations
Trace configurations for common mode impedances of 25 (single ended), 30, 35, and 40 ohms (from top to bottom respectively) in symmetrical laminates of Megtron6 and 1x2116 (Left) / 2x1080 (Right) laminates.
3D Full-wave Modeling- Results
Glass type 1x2116
Width/Spacing (mil) Loss per in (dB)
@ 12.5 GHz
Skew per inch (ps)
Comm25 4.6/13 0.71 4.9765
Comm30 3.8/5.5 0.76 4.8974
Comm35 3.0/3.6 0.85 4.5657
Comm40 2.4/2.8 0.94 3.55
Glass type 2x1080
Width/Spacing (mil) Loss per in (dB)
@ 12.5 GHz
Skew per inch (ps)
Comm25 4.6/13 0.71 8.8371 Comm30 3.8/5.5 0.76 9.0338 Comm35 3.0/3.6 0.85 8.8390 Comm40 2.4/2.8 0.94 7.6014
Trace width, spacing, loss/inch, and skew/inch for several common mode impedances of a 100 ohms differential strip lines embedded in Megtron6.
5mil core 5mil prepreg
Glass type
2x2116
Width/Spacing (mil)
Loss per in (dB) @ 12.5
GHz
Skew per in (ps)
Comm25 9.4/30 0.54 0.9093
Comm30 7.7/9.5 0.57 0.0006
Comm35 6.3/6.3 0.62
1.2515
Comm40 5.3/4.7 0.68
3.0047
2x2116 glass weave
3D Full-wave Modeling- Results
9.6mil core 9.6mil prepreg
Offset 0mil 1mil 2mil 3mil 4mil 5mil 6mil 7mil 8mil
Skew per inch (ps)
0.0006 0.0174 0.0606 0.0982 0.1171 0.1344 0.1250 0.0847 0.0171
Offset 9mil 10mil 11mil 12mil 13mil 14mil 15mil 16mil 17mil
Skew per inch (ps)
0.0636 0.1114 0.145 0.353 0.1181 0.072 0.0393 0.027 0.0104
Skew results of parametric sweep for differential trace of Zcomm=30 ohms as a function of translational position .
3D Full-wave Modeling- Results
Skew Minimization techniques- Glass Weave Rotation
Case
Glass type 1x2116
Skew per in (ps) Original
Skew per in (ps) Rotate 45 deg
Comm25 4.9765 2.5408
Comm30 4.8974 2.8129
Comm35 4.5657 2.5637
Comm40 3.5500 1.9253
Comparison of skew per inch for a 45 degrees angular rotation of prepreg glass type 1x2116.
Skew Minimization techniques- Offsetting
Glass to Glass Position
2x1080 glass weave peak and valley (locked) positions.
Case Glass type
2x1080
Skew per in (ps) Original
Skew per in (ps) With offset
average
Comm25 8.8371 0.0033
Comm30 9.0338 0.1203
Comm35 8.8390 0.4060
Comm40
7.6014
1.0298
Skew for symmetrical vs. offset glass positioning.
Channel Simulation- S-parameter Models
Cross sectional view of a differential stripline
with separate effective Dk regions.
-105-100
-95-90-85-80-75-70-65-60-55-50-45-40-35-30-25-20-15-10
-50
0 5 10 15 20 25 30 35 40 45
IL, d
B
f, GHz
Insertion Loss, 25" TraceInternal versus External Skew
0ps, Internal
5ps, Internal
10ps, Internal
15ps, Internal
20ps, Internal
25ps, Internal
25ps, External
20ps, External
15ps, External
10ps, External
5ps, External
0ps, External
Insertion loss characteristics of a differential stripline model with internally (red) versus externally (blue) added skew.
Channel Simulation- S-parameter Models
IL (left) and delay (right) differences and equalization between internal and external skew models.
0
20
40
60
80
100
120
140
160
180
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
15", [email protected]
VEYE_Max
VEYE_-5ps
VEYE_+5ps
Channel Simulation- Link Simulation
Absolute and relative eye eight (VEYE) as a function of skew for 15” (top) and 20”(bottom) Megtron6 channels.
-45-40-35-30-25-20-15-10
-505
10
0 5 10 15 20 25 30
VE
YE
De
lta
, %
Skew, ps
15", [email protected]
VEYE_ Max, Delta%
VEYE_-5ps, Delta%
VEYE_+5ps, Delta%
0
20
40
60
80
100
120
140
160
180
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
20", [email protected]
VEYE_Max
VEYE_-5ps
VEYE_+5ps
-50-45-40-35-30-25-20-15-10
-505
1015
0 5 10 15 20 25 30
VE
YE
De
lta
, %
Skew, ps
20", [email protected]
VEYE_ Max, Delta%
VEYE_-5ps, Delta%
VEYE_+5ps, Delta%
25Gbps
8 DFE
Channel Simulation- Link Simulation
0
20
40
60
80
100
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
25", [email protected]
VEYE_Max
VEYE_-5ps
VEYE_+5ps
-50-45-40-35-30-25-20-15-10
-505
10152025
0 5 10 15 20 25 30
VE
YE
De
lta
, %
Skew, ps
25", [email protected]
VEYE_ Max, Delta%
VEYE_-5ps, Delta%
VEYE_+5ps, Delta%
0
20
40
60
80
100
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
30", [email protected]
VEYE_Max
VEYE_-5ps
VEYE_+5ps
-60-55-50-45-40-35-30-25-20-15-10
-505
1015202530
0 5 10 15 20 25 30
VE
YE
De
lta
, %
Skew, ps
30", [email protected]
VEYE_ Max, Delta%
VEYE_-5ps, Delta%
VEYE_+5ps, Delta%
Absolute and relative eye eight (VEYE) as a function of skew for 25” (top) and 20”(bottom) Megtron6 channels.
25Gbps
8 DFE
0
20
40
60
80
100
120
140
160
180
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
15", [email protected], NRZ
VEYE_Max
VEYE_-5ps
VEYE_+5ps
Channel Simulation- Link Simulation
Absolute and relative eye eight (VEYE) as a function of skew for 20” Megtron6 channel simulated with 5 DFE taps.
-45-40-35-30-25-20-15-10
-505
10
0 5 10 15 20 25 30
VE
YE
De
lta
, %
Skew, ps
15", [email protected], NRZ
VEYE_ Max, Delta%
VEYE_-5ps, Delta%
VEYE_+5ps, Delta%
0
20
40
60
80
100
120
140
160
180
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
20", [email protected], NRZ
VEYE_Max
VEYE_-5ps
VEYE_+5ps
-60-55-50-45-40-35-30-25-20-15-10
-505
1015
0 5 10 15 20 25 30
VE
YE
De
lta
, %
Skew, ps
20", [email protected], NRZ
VEYE_ Max, Delta%
VEYE_-5ps, Delta%
VEYE_+5ps, Delta%
0
10
20
30
40
50
60
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
35", [email protected] VS. PAM4 (25Gbps)
NRZ
PAM4
-45-40-35-30-25-20-15-10
-505
1015
0 5 10 15 20 25 30
VE
YE
De
lta
, %
Skew, ps
35", [email protected] VS. PAM4 (25Gbps)
NRZ, Delta%
PAM4, Delta%
0
10
20
30
40
50
60
0 5 10 15 20 25
VE
YE
, m
V
Skew, ps
35", [email protected] VS. PAM4 (30Gbps)
NRZ
PAM4
Skew effect in 35” channel operating at 25 Gbps (top) and 30 Gbps (bottom) with NRZ (black) and PAM4 (red).
Channel Simulation- Link Simulation
4 DFE 6dB CTLE gain
0
10
20
30
40
50
60
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
35", [email protected] VS. PAM4 (35Gbps)
NRZ
PAM4
-20
-15
-10
-5
0
0 5 10 15 20
VE
YE
De
lta
, %
Skew, ps
35", [email protected] VS. PAM4 (35Gbps)
NRZ, Delta%
PAM4, Delta%
0
10
20
30
40
50
60
0 5 10 15 20 25 30
VE
YE
, m
V
Skew, ps
35", [email protected] VS. PAM4 (40Gbps)
NRZ
PAM4
-20
-15
-10
-5
0
0 5 10 15
VE
YE
De
lta
, %
Skew, ps
35", [email protected] VS. PAM4 (40Gbps)
NRZ, Delta%
PAM4, Delta%
Skew effect in 35” channel operating at 35 Gbps (top) and 40 Gbps (bottom) with NRZ (black) and PAM4 (red).
Channel Simulation- Link Simulation
0
10
20
30
40
50
60
25 30 35 40
VEY
E, m
V
Bit rate (Gbps)
35 inches, 15ps skew
NRZ
PAM4
0
10
20
30
40
50
60
25 30 35 40
VEY
E, m
V
Bit rate (Gbps)
35 inches, 0ps skew
NRZ
PAM4
Eye height vs. data rate for 35” channel of 0 (top) and 15 ps (bottom) skew for NRZ (Blue) and PAM4 (red).
Channel Simulation- Link Simulation
VICTIM
AGGRESSOR# of Total layers: 20 # of Signal Layers: 8 Height: 3.9mm Drill Size: 0.45mm PCB material: Megtron6
Via and victim/aggressor configurations used for studying the effect of skew in crosstalk.
Channel Simulation- Link Simulation
26
28
30
32
34
36
0 5 10 15 20 25 30
VEY
E, m
V
Aggressor Skew, ps
FEXT WITH SKEW, trace Length=30"[email protected], 25Gbps NRZ
With_FEXT
W/O FEXT (Thru@0ps)
-80
-70
-60
-50
-40
-30
-20
-10
0
0 5 10 15 20 25 30
FEXT
, dB
f, GHz
Skew Induced FEXT
0ps
5ps
10ps
15ps
20ps
25ps
Skew induced excess FEXT (top) and resulting VEYE (bottom) for DCviaBPviaDC channel.
Channel Simulation- Link Simulation
DC: 5in BP: 20in
Conclusion
• “Unit cell” concept was used to model several configurations of stripline traces.
• Skew from unit cell was scaled up to obtain ps/inch directly from “group delay” values. No S-parameter cascading was necessary.
• Worst case placements of trace over glass laminates were studied.
• For all cases analyzed, a wide range of 0.0006 to 9 ps/inch values for skew were obtained.
• Higher Zcomm reduced skew unless it was countered by the match between trace spacing and weave pitch.
• The lowest skew value was resulted when trace spacing matched the glass weave repetition period.
• Laminate rotation to reduce skew was modeled and re-confirmed to be effective.
• Predicted skew values are far larger than what has ever been measured historically by the authors. This needs to be investigated further.
• 25 Gbps eye height and width simulations suggested that 20 ps of skew can be tolerated, assuming the channels consist of simple traces and low via crosstalk. Effect of CM noise injection, and the consequence of its conversion, was not considered.
• 5ps off max sampling of the eye did produce significant VEYE variations/degradations.
• At 25 Gbps, there may be an advantage in applying PAM 4 over NRZ with respect to the change in signal degradation but not absolute VEYE.
• Skew increased the crosstalk noise between two offset coupled differential via structures, but the eye degradation was insignificant for sufficient S/N margin.
Authors Peerouz Amleshi
Electrical Engineering Director
Molex Inc.
Email: [email protected]
Cong Gao
Signal Integrity Engineer
Molex Inc.
Email: [email protected]
Davi Correia
Signal Integrity Engineer
Molex Inc.
Email: [email protected]
Lin Shen
Principal Engineer
Insieme Networks Inc.
Email: [email protected]