Advances in Measurement Based Transient Simulation

advertisement
Time Domain Simulation in ADS, Slide - 1
Advances in Measurement Based
Transient Simulation
Presented by GigaTest Labs
Gary Otonari and Orlando Bell
March, 2008
1
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 2
Transient Simulation in ADS
History
•
•
•
•
Microwave SPICE -- EEsof time domain simulator for microwave
Convolution and Multilayer Line Models – High Speed Digital design
ADS - A design framework for SI
2006 - Transient Simulation improvements
Microwave
SPICE
GigaTest
Founded
HP EEsof
Merge
Convolution
MLM Models
Usability
Improvements
ADS
ADS
2006 SI
DCA Frontend
2
www.GigaTest.com
Transient Simulation
1
Time Domain Simulation in ADS, Slide - 3
Convolution Simulation
• For many years, the only simulator that allowed transient simulation
on models with frequency dependence
• How does it work?
S4P
SNP3
File="CABLEandCONNECTORwithoutCABLE.s4p"
R
R5
R=50 Ohm
4
tdrmp
VtPulse
SRC2
Vlow=0 V
Vhigh=2 V
Delay=0 nsec
Edge=erf
Rise=(2.2*risetime) psec
Fall=(2.2*risetime) psec
Width=50 nsec
Period=50 nsec
t
1
TLIN
TL6
Z=50.0 Ohm
E=360
F=5 GHz
xnem
2
3
Ref
xfem
Term
Term6
Num=6
Z=50 Ohm
time-domain impulse
from S-parameters
Term
Term5
Num=8
Z=50 Ohm
Convolve with
input waveform
R
R12
R=50 Ohm
tdrmn
VtPulse
SRC5
Vlow=0 V
Vhigh=-2 V
Delay=0 nsec
Edge=erf
Rise=(2.2*risetime) psec
Fall=(2.2*risetime) psec
Width=50 nsec
Period=50 nsec
t
TLIN
TL9
Z=50.0 Ohm
E=360
F=5 GHz
www.GigaTest.com
Transient output WITH
frequency domain
effects
3
Time Domain Simulation in ADS, Slide - 4
Convolution = Key Feature
• Coupled with Agilent’s measurement and simulation depth, this tool
provided a key capability to the signal integrity designer, and
leverage for things like:
– ADS microwave and RF transmission line models
– Network analyzer and TDR measurements
– ADS Co-simulation
= CONVOLUTION
0.6
0.4
0.2
m1m2
0.0
-0.2
-0.4
-0.6
0
100
200
300
400
500
600
700
800
900
time, psec
ML2CTL_C
CLin2
Subst="Subst1"
Length=45 mil
MLRADIAL2
) mil
All GOOD?
W=ms_w mil
Radial1
S=69 mil
Subst="Subst1"
X_Offset=31 mil
Layer=2
Y_Offset=0.0 mil
W_Left=ms_w mil
RLGC_File=
ReuseRLGC=no
W_Right=ms_w mil
S_Left=(ms_pitch-ms_w) mil
Not exactly…..
4
www.GigaTest.com
Transient Simulation
2
Time Domain Simulation in ADS, Slide - 5
Agenda
The GigaTest “Wish-list” for Transient Simulation
•
•
•
•
•
Causality
Improved S-parameter data processing
Passivity
Long Structures
FUTURE Directions….
5
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 6
What’s all this Causality Stuff, Anyhow?
Response before 0
impulse_response
30
20
10
0
-10
-20
-2
Transient simulation starts from 0
0
2
4
6
8
10
12
14
16
18
20
time, sec
1.0
Simulated
Original
Spectrum
0.8
Poor accuracy
0.6
0.4
0.2
0.0
0
1
freq, Hz
Non-causal impulse response degrades simulation accuracy
6
www.GigaTest.com
Transient Simulation
3
Time Domain Simulation in ADS, Slide - 7
Causality Correction in ADS
var("CMP1_IMP(2;1).ImpResp"
var("CMP1_IMP(4;3).ImpResp"
• ADS v2006U1 introduced an improved convolution
simulation algorithm with causality correction
• Forces all impulse responses to be causal
• Implements improved adaptive sampling and
extrapolation of frequency domain response
0.12
0.10
0.08
• Actual Impulse Response
Calculated during
convolution
0.06
0.04
0.02
• Starts at t=0
-0.00
-0.02
0
1
2
3
4
5
6
7
8
9
10
time, nsec
7
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 8
Causality Check: Long Interconnects
Agilent 86100 / 54754 TDR and GTL Probe Station
• The best way to see this is to look at measurements on a long
interconnect
• Sample is a 1.5 meter long Infiniband Cable, measured separately for
– differential TDR and TDT response (below left)
– 4-port S-parameter response
8
www.GigaTest.com
Transient Simulation
4
Time Domain Simulation in ADS, Slide - 9
Non-causal Simulation Results
ADS 2003C Simulation Data
• Prior to ADS 2006, transmission line models were non-causal in ADS
• Convolution applied to measured S-parameter data was also not
generally causal
• Results shown below are Convolution simulation for the 1.5 meter
Infiniband cable with
The risetime should
“degrade” in a causal
response
– Stripline Microwave Model
– Multi-layer Line Model (2D EM)
2.0
– 4-port S-parameter data
ML2CTL_C
CLin1
Subst="cable"
Length=1.42 meter
W=wire mil
S=(pitch-wire) mil
Layer=3
RLGC_File=
ReuseRLGC=no
1.5
1.0
The pulse should
not “arrive” before
the interconnect
delay allows
0.5
0.0
-0.5
7.50
7.75
8.00
8.25
8.50
time, nsec
8.75
9.00
9
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 10
Non-causal Simulation Results
ADS 2003C Simulation Data
• The complex loss description over frequency, determines whether a
transmission line model is causal.
• For most commercially available simulators, this is determined by the
frequency dependent dissipation factor, or dielectric loss tangent
• Measured S-parameter data is causal if measured correctly, but
simulated results can become non-causal when convolution simulation
is applied to it
•2.0
•tdtsp_strip-tdtsn_strip
•tdtmp-tdtmn
•tdtsp-tdtsn
Prior to ADS 2006U1:
• The microwave t-line
models were the most
non-causal
• MLM models were better
• Measured S-parameters
best
•1.5
•1.0
•0.5
•0.0
•-0.5
•7.50 •7.75 •8.00 •8.25 •8.50 •8.75 •9.00
•time, nsec
10
www.GigaTest.com
Transient Simulation
5
Time Domain Simulation in ADS, Slide - 11
ADS 2008A: Causality Enforced
•Exact same models and data simulated in ADS 2008
• Stripline Microwave Model
• Multi-layer Line Model (2D EM)
• 4-port S-parameter data
tdtsp_strip-tdtsn_strip
tdtmp-tdtmn
tdtsp-tdtsn
2.0
1.5
1.0
0.5
0.0
-0.5
7.50
7.75
8.00
8.25
8.50
8.75
9.00
time, nsec
11
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 12
Implications: Model Accuracy
ADS 2003C vs. 2008A
•Eye Diagrams generated from the MLM Model for a 1.5 meter long
Infiniband cable with mated connectors by ADS 2003C and ADS 2008A
m5
time= 324.8psec
m5=0.007
index=57.000000
eye(voutp-voutn, 1.25 GHz)
eye(voutp-voutn, 1.25 GHz)
m6
time= 322.3psec
m6=0.008
index=43.000000
0.6
0.4
0.2
m6
m5
0.0
-0.2
-0.4
-0.6
m6
time=328.1psec
eye(voutp-voutn, 1.25 GHz)=0.0
index=52.000000
m5
time=322.1psec
eye(voutp-voutn, 1.25 GHz)=0.0
index=12.000000
0.6
0.4
0.2
m5
m6
0.0
-0.2
-0.4
-0.6
200
300
400
500
time, psec
600
700
800
200
300
400
500
600
700
800
time, psec
Datarate = 2.5 Gb/s
Risetime = 100 ps
12
www.GigaTest.com
Transient Simulation
6
Time Domain Simulation in ADS, Slide - 13
What About S-parameters?
• Properly measured / modeled S-parameter data should be causal
• Convolution Simulation can introduce non-causal response
• EXAMPLE: Simple RLC Circuit (Causal) ==> After band-limiting (Non Causal)
Term
Term1
Num=1
Z=50 Ohm
L
L1
L=1 H
R=
R
R1
R=1 Ohm
C
C1
C=1.0 F
Admittance simulated from 0 to 1 Hz
1.0
impulse_response
0.8
Spectrum
Band-limiting induces non-causal
response
30
0.6
0.4
0.2
20
10
0
-10
-20
0.0
0
1
-2
0
2
4
freq, Hz
6
8
10
time, sec
Causality
Bandwidth
13
Risetime
Non-causal response
introduces simulation error
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 14
S-parameter Data Causality
• ADS 2008 has a new adaptive sampling algorithm with extrapolation
• This improvement is in addition to the causality enforcement
• Very helpful where source parameters slightly exceed bandwidth of SnP
• Example: 1.5 meter Infiniband cable at 2.5 Gb/s with 50 ps risetime
4
1
2
Ref
S4P
SNP2
File="concatenate_cable.ds"
• Same 1.5m cable
Datarate = 2.5 Gb/s
Risetime is 50 ps
• Jitter Measured: 13.7 ps
• Jitter Sim: 12.8 ps
• VERY good agreement !
time 424.2psec
eye(voutn,1.25GHz)=0.024
index=226.000000
time= 434.2psec
eye(voutn,1.25GHz)=0.296
index=43.000000
0.4
eye(voutn,1.25GHz)
3
m5
0.3
0.2
0.1
m6
0.0
-0.1
0
100
200
300
400
500
600
700
800
90
time psec
14
www.GigaTest.com
Transient Simulation
7
Time Domain Simulation in ADS, Slide - 15
Transient Wish List #2
S-parameter Data
• Noisy, resonant and lossy structures have historically been difficult to
TDR using convolution
• Improvements to the S-parameter data sampling and extrapolation
have other implications beyond causality
0
0
-2
dB(S(4,3))
dB(S(2,1))
dB(S(3,3))
dB(S(1,1))
-10
-20
-30
-4
-6
-8
-10
-40
0
1
2
3
4
5
6
0
freq, GHz
1
2
3
4
5
6
freq, GHz
15
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 16
Example: Noisy and Lossy S-parameters
• Sample is an Infiniband PCB + Connector
• Measured with 2-sided probing using
– N5230A 4-port PNA calibrated from 50 MHz to 10 GHz
– 86100A / 54754A Differential TDR
16
www.GigaTest.com
Transient Simulation
8
Time Domain Simulation in ADS, Slide - 17
Demo: Two-sided Probing
• GTL 5050 Rotatating Stage Probing System
17
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 18
Example: Noisy and Lossy S-parameters
• Single ended S-parameters are shown including return loss at each
port (left) and attenuation for each complementary net in the
differential pair.
• NOTE the small resonances in band, as well as the rapid
increase in both return loss and attenuation above 9 GHz
18
www.GigaTest.com
Transient Simulation
9
Time Domain Simulation in ADS, Slide - 19
ADS 2003C – TDR on Bandlimited Data
ADS 20003C / Measured TDR
Risetime = 50 ps Î 9.1 GHz
19
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 20
ADS 2003C – TDR on Bandlimited Data
ADS 20003C / Measured TDR
Risetime = 45 ps Î 10.1 GHz
20
www.GigaTest.com
Transient Simulation
10
Time Domain Simulation in ADS, Slide - 21
ADS 2003C – TDR on Bandlimited Data
ADS 20003C / Measured TDR
Risetime = 40 ps Î 11.4 GHz !
Note the improved accuracy at the higher bandwidth (which
exceeds the 10 GHz data)
21
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 22
ADS 2008A – TDR on Bandlimited Data
ADS 2008A / Measured TDR
Risetime = 50 ps Î 9.1 GHz
22
www.GigaTest.com
Transient Simulation
11
Time Domain Simulation in ADS, Slide - 23
ADS 2008A – TDR on Bandlimited Data
ADS 2008A / Measured TDR
Risetime = 45 ps Î 10 GHz
23
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 24
ADS 2008A – TDR on Bandlimited Data
ADS 2008A / Measured TDR
Risetime = 40 ps Î 10 GHz
24
www.GigaTest.com
Transient Simulation
12
Time Domain Simulation in ADS, Slide - 25
How Does ADS Do This…?
• ADS 2008 has an improved adaptive sampling algorithm when
converting the frequency domain data to time domain impulse
response to better handle complex time domain response typical of
noisy, lossy and resonant structures
• New impulse response calculation has “smart” extrapolation to DC
• The plot shows measured S11 vs. Extrapolated/ Re-sampled S11
• Convolution settings should be left at default values
0
-20
-40
-60
-80
-100
-120
0
1
2
3
4
5
6
freq, GHz
7
8
9
10
25
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 26
Summarizing
• Before and After vs. Measured
• @ 50 ps risetime
.
26
www.GigaTest.com
Transient Simulation
13
Time Domain Simulation in ADS, Slide - 27
Transient Wish List #3: Passivity
• Non passive data can come from EM field solvers, measurements
or user processing
Package w/ S21 > 0 dB
de-embedded PCB w/ S11 > 0dB
• ADS 2006 implemented a new passivity enforcement for Sparameter data files
• Two common sources of non-passive data in measurement
– Dissimilar probing surface between calibration and measurement
– De-embedding
27
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 28
Non-passive Measured Data
• Usually, measured data will be non-passive only on short, low-loss
structures
• Often due to probe pads and DUT interfaces that differ from calibration
structures
• Example: package measurements on solder balls or solder bumps
The probe has a tendency to dig
into the soft solder, resulting in an
contact point that is “behind” the
calibration reference plane
Flip-chip BGA bumps
Calibration Substrate
28
www.GigaTest.com
Transient Simulation
14
Time Domain Simulation in ADS, Slide - 29
Non-passive Measured Data
• Flip-chip BGA differential pair w/ +S21
• Reflection is only slightly affected
• Turning on the passivity enforcement
causes the data to change from Blue
(non-passive) to Red (passive) below
m2
m2
freq= 2.350GHz
dB(S(2,1))=0.027
29
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 30
Non-Passive Data: De-embedding
• De-embedding often results in non-passive data
• This can be made worse if:
– measured data is used to perform the de-embedding
– the structure to be de-embedded is electrically shorter than the DUT
it’s being de-embedded from
• This non-physical data can severely impact the transient
simulation accuracy
Example: Test Board & Package
• Measured from test port to BGA
device solder bump pads
• Measured from test port to BGA
pads
• Desired Output: Package deembedded from PCB
30
www.GigaTest.com
Transient Simulation
15
Time Domain Simulation in ADS, Slide - 31
De-embedding Results
PCB + Package / PCB Only / De-embedded Package
• De-embedding is accomplished in ADS with the schematic shown
• 2-port DEEMBED component used 10
0
• 4-port component available
-10
-20
1
Term
Term14
Num=1
Z=50 Ohm
Term
Term12
Num=4
Z=50 Ohm
2
Ref
-30
-40
4
1
-50
2
3
0
Ref
1
2
3
4
1
5
6
7
8
9
10
7
8
9
10
freq GHz
20
2
Ref
Term
Term13
Num=2
Z=50 Ohm
Term
Term11
Num=3
Z=50 Ohm
0
-20
-40
-60
0
1
2
3
4
5
6
31
freq, GHz
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 32
Passivity Enforced
• ADS 2008 can enforce passivity on the S-parameter data
• The Transient simulation control (left) shows how this is activated
• Before and After plots are shown for S(1,1) and TDR Impedance
20
10
0
S(1,1)
-10
-20
-30
-40
0
1
2
3
4
5
6
7
8
9
10
175
150
DiffZ
125
100
75
50
1.0
1.5
2.0
2.5
3.0
time, nsec
3.5
4.0
4.5
5.0
32
www.GigaTest.com
Transient Simulation
16
Time Domain Simulation in ADS, Slide - 33
Transient Wish List #4
Long Interconnects
• Generating TDR on long structures requires low frequency s-parameter
data to avoid aliasing in the convolution FFT
• Limitations in network analysis hardware can make this low frequency
data difficult to obtain
• The adaptive sampling capability implemented in ADS 2008 includes
better extrapolation to DC, which greatly improves the accuracy of the
TDR response on long structures
33
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 34
36” Backplane / Flex Circuit
• This DUT consisted of two differential flex transmission lines connected to
each other through a backplane channel.
• Each “side” of the flex is attached directly to a backplane connector “wafer”,
then mated to backplane connector mounted on the multilayer PCB.
Flex probed
at these
points
2
1
3
4
34
www.GigaTest.com
Transient Simulation
17
Time Domain Simulation in ADS, Slide - 35
VNA Measurements
• The raw data from 25 MHz to 20 GHz is shown on the left with differential
S(2,1) and S(1,1)
• The data on the right is S(1,1) at low frequency with the ORIGINAL data in
Green, the ADS processed (extrapolated, re-sampled) data in Orange
0
0
-10
-8
-20
-16
-30
-24
-40
-32
-50
-60
-40
0
-70
0
2
4
6
8
10
12
14
16
18
25
50
75 100 125 150 175 200 225 250
20
freq, MHz
freq, GHz
• The improved extrapolation at low frequencies greatly improves the TDR
35
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 36
TDR Measurements
Diff Z
1.20E+02
1.15E+02
impedance (ohms)
1.10E+02
1.05E+02
diff Z
1.00E+02
9.50E+01
9.00E+01
8.50E+01
8.00E+01
5.00000E-08
5.20000E-08
5.40000E-08
5.60000E-08
5.80000E-08
6.00000E-08
6.20000E-08
time (ns)
• TDR measured using an Agilent 86100B Infiniium Oscilloscope with 54745A
TDR Plug-in
• The data was captured from the scope and re-plotted using EXCEL (left)
36 ps
Risetime ~ 100
www.GigaTest.com
Transient Simulation
18
Time Domain Simulation in ADS, Slide - 37
TDR Simulation
TDR Data (from S-parameter measurements)
2003C vs. 2008A
impedance (ohms)
125
1.20E 02
120
1.15E+02
115
1.10E+02
110
1.05E+02
105
1.00E+02
100
9 50E+01
95
1.5
3.5
5.5
7.5
9.5
11.5
time, nsec
ADS 2003C is much lower than the actual TDR profile, due to aliasing while
ADS 2008A agrees well.
Risetime = 100 ps
37
www.GigaTest.com
Time Domain Simulation in ADS, Slide - 38
Conclusions
• ADS 2006 and beyond, significantly improves what was already the
most advanced transient simulation tool for high frequency design
• Improves accuracy by addressing:
– Causality
– S-parameter data processing
– Passivity
– Aliasing
• Future possible improvements:
– Impedance peeling
– PNA measurements to DC
38
www.GigaTest.com
Transient Simulation
19
Time Domain Simulation in ADS, Slide - 39
Who is GigaTest?
PCB Probing Systems and other measurement accessories
Signal Integrity Engineering
Measurement
Simulation
Design
Training
39
www.GigaTest.com
Transient Simulation
20
Download