Again, more information. TDR/TDT is an integrated/filtered response. It is
secondary to the VNA measurements being made. How do you guarantee that the
model you achieve scales across structural size? Do you have experimental
collaboration of this for quasi-isotropic dielectrics such as ABF films used in
packaging and lower conductance Cu?
Scott McMorrow, CTO Signal Integrity Group
Samtec
Office 401-284-1827 | +1-800-726-8329
www.samtec.com
-----Original Message-----
From: C.C. Hwang [mailto:cchwang2013@xxxxxxxxx] ;
Sent: Thursday, May 25, 2017 1:05 PM
To: Scott McMorrow <Scott@xxxxxxxxxxxxx>
Cc: shlepnev@xxxxxxxxxxxxx; John Lin <johnlinc@xxxxxxxxx>;
jose.moreira@xxxxxxxxxxxxx; si-list <si-list@xxxxxxxxxxxxx>
Subject: Re: [SI-LIST] Re: Differential vs. SE for PCB Dk/Df extraction
Hi Scott,
We matched the de-embedded S-param (and TDR/TDT) directly.
Ching-Chao
On Thu, May 25, 2017 at 10:00 AM, Scott McMorrow <Scott@xxxxxxxxxxxxx> wrote:
C.C.------------------------------------------------------------------
You keep talking about Self-Verification. Please elaborate.
Scott McMorrow, CTO Signal Integrity Group Samtec Office 401-284-1827
| +1-800-726-8329 www.samtec.com
-----Original Message-----
From: C.C. Hwang [mailto:cchwang2013@xxxxxxxxx]
Sent: Thursday, May 25, 2017 12:51 PM
To: shlepnev@xxxxxxxxxxxxx
Cc: Scott McMorrow <Scott@xxxxxxxxxxxxx>; John Lin
<johnlinc@xxxxxxxxx>; jose.moreira@xxxxxxxxxxxxx; si-list
<si-list@xxxxxxxxxxxxx>
Subject: Re: [SI-LIST] Re: Differential vs. SE for PCB Dk/Df
extraction
Hi Yuriy,
It's probably a matter of semantics, but eigenvalue, idealized
insertion loss, modal propagation constant or GMS all refer to the
eigenvalues of T matrix (which are summarized in one equation in page
25 of http://www.ataitec.com/PDF/MPX.pdf). There's much to like about the
eigenvalue approach due to its simplicity and it may "even" work under
perfect conditions (known cross section, exactly the same launch and trace,
etc.). I wonder how you verify your results if you don't have good
de-embedded IL, RL, NEXT and FEXT to compare with.
Granted that it's harder to match all de-embedded S parameters and TDR/TDT
than just propagation constants, but there are also many things to like about
our approach (such as self verification and more tolerant to manufacturing
variation) so it would not stop us from pursuing it. It was hard to beat a
Go master, but it did not stop AlphaGo...
Regards,
Ching-Chao Huang
On Thu, May 25, 2017 at 7:22 AM, Yuriy Shlepnev <shlepnev@xxxxxxxxxxxxx>
wrote:
Ching-Chao,
You stated "Matching only idealized insertion loss (i.e., attenuation and
phase delay) from eigenvalue solution is a necessary but not sufficient
condition."
The sufficient conditions for the techniques based on the eigenvalues (and
for all other technique for that matter) is the identity of the
cross-sections of the real structure and in the model and the accuracy of
the field solver used in the identification. As soon as those conditions are
satisfied and the material models are identified with just 2 GMS-parameters
or 2 Gammas, all S-parameters will match - we proved it in multiple
projects. Yes, it is "hard to believe", but this is true :-) Note that the
results obtained with any de-embedding technique are also dependent on the
cross-section and the field solver accuracy. Matching all 10 complex
S-parameters in differential case (4 complex parameters in case of symmetry
instead of just 2 in GMS or Gamma technique), you may end up with totally
wrong dielectric or roughness models, if the cross-section in the model is
not correct or field solver is not accurate. Have you been in such situation?
Let me also elaborate on the "idealized insertion loss". The modal insertion
loss and phase delays in GMS-parameters are not "idealized" form of
S-parameters. It is exact matrix transformation into 2-diagonal form in case
of S-parameters, and into diagonal form in case of scattering T-matrix. The
diagonal form of a matrix is the simplest matrix form in the eigenvalue
basis - it is not an approximation. What makes it an approximation is the
assumptions about identity of the connector and launches and identity of the
cross-sections in two line segments. GMS-parameters are very tolerant to
those variations - see sensitivity investigation at #2011_03 and #2010_03 at
http://www.simberian.com/AppNotes.php This is no more assumptions as in the ;
de-embedding, where the test fixtures are assumed to be identical and the
consequences of non-identity on the identified material models are mostly
un-known. In general, a complete de-embedding is more error-prone, comparing
to the incomplete de-embedding (that is GMS-parameters) or to the Gamma
extraction.
Considering the theoretical background of GMS-parameters or Gamma
extraction, the “eigenvalue” formulation for the T-matrix of the middle
segment was first suggested by R. A. Soares et. al. in “A Unified
mathematical approach to two-port calibration techniques and some
applications”, MTT, 1989. Though, you can find the diagonal exponential form
of the T-matrix as early as 1975 in works of N. Franzen and R. Speciale from
Tektronix (they tried to solve the problem directly). The eigenvalue
formulation was extended to multi-conductor lines by Sequinot et. al in
1998. The math behind it simple and magic at the same time and is based on
the fact that T-matrix of a t-line segment is diagonal exponential in modal
space with the normalization to the modal characteristic impedance - no
assumptions or approximations. If converted to S-matrix, the result is
Generalized Modal S-parameters or GMS-parameters without reflection and mode
conversion exactly. If you take GMS insertion loss and divide by length, you
get the attenuation per unit length. That is also the real part of the
complex propagation constant or Gamma. The imaginary part of Gamma can be
computed through the phase. Getting GMS-parameters can be considered as
incomplete de-embedding – no information on the characteristic impedance to
re-normalize the matrix. But, it is not needed, because for the loss
evaluation or material identification only GMS-parameters or Gammas
extracted from them are needed. In general, all recently suggested
techniques for loss control or material identification are now converging to
the “eigenvalue” technique. It works for multi-conductor lines in general
(not just single-ended or differential).
Finally, the result of the identification should be material models
suitable for the analysis at least in some other EDA tools.
Comparison of numerical results obtained with different tools or
methods should be a part of any tool validation, that includes the material
models.
Though, that could be a subject for another discussion :-)
Best regards,
Yuriy
Yuriy Shlepnev, Ph.D.
President, Simberian Inc.
2629 Townsgate Rd., Suite #235, Westlake Village, CA 91361, USA
Office
+1-702-876-2882; Fax +1-702-482-7903 Cell +1-206-409-2368; Virtual
+1-408-627-7706
Skype: shlepnev
www.simberian.com
Simbeor – Accurate, Productive and Cost-Effective Electromagnetic
Signal Integrity Software
2010 and 2011 DesignVision Award Winner, 2015 Best In Design&Test
Finalist
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of C.C. Hwang
Sent: Wednesday, May 24, 2017 5:20 PM
To: Scott McMorrow
Cc: shlepnev@xxxxxxxxxxxxx; John Lin; jose.moreira@xxxxxxxxxxxxx;
si-list
Subject: [SI-LIST] Re: Differential vs. SE for PCB Dk/Df extraction
Hi Scott,
We matched all single-ended, common and differential S parameters (magnitude
and phase of IL, RL, NEXT and FEXT) PLUS all single-ended, common and
differential TDR/TDT. It is "self-consistent" because of built-in self
verification.
Matching only idealized insertion loss (i.e., attenuation and phase delay)
from eigenvalue solution is a necessary but not sufficient condition.
Return loss, for example, is affected by DK and cross-sectional geometry.
The cross-sectional geometry in turn affects the surface roughness and DF
extraction. It's hard to imagine that matching only idealized insertion
loss to extract DK/DF will give the same original IL, RL, NEXT, FEXT and
TDR/TDT.
Regards,
Ching-Chao Huang
On Wed, May 24, 2017 at 1:39 PM, Scott McMorrow <Scott@xxxxxxxxxxxxx> wrote:
C.C.------------------------------------------------------------------
I agree with the need for FEXT and NEXT to dial in the characteristics of
the resin-rich layer between differential pair conductors. I published
this in my training years ago. Another simple way to perform the
separation of dielectric properties is to use the separation of
differential and common mode phase delay.
I'm surprised by your omission of phase delay. Everything you need is
contained in amplitude and phase. TDR/TDT is an integration and obscures
the primary information. Good for a final check but not the best way to
identify the primary material parameters. Of all the measurements you can
make, phase delay is also the one most free of noise.
Just sayin'.
Scott
Scott McMorrow, CTO Signal Integrity Group Samtec Office
401-284-1827
| +1-800-726-8329 www.samtec.com
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of C.C. Hwang
Sent: Wednesday, May 24, 2017 11:54 AM
To: dmarc-noreply@xxxxxxxxxxxxx
Cc: shlepnev@xxxxxxxxxxxxx; John Lin <johnlinc@xxxxxxxxx>;
jose.moreira@xxxxxxxxxxxxx; si-list <si-list@xxxxxxxxxxxxx>
Subject: [SI-LIST] Re: Differential vs. SE for PCB Dk/Df extraction
For accurate DK/DF extraction, we should measure differential traces and
match all IL, RL, NEXT, FEXT and TDR/TDT after causal de-embedding.
Because the glass/resin composite makes PCB stripline structures
inhomogeneous, the extracted DK/DF will depend on the cross-sectional model
being used. The extracted DK/DF can be considered "effective" values and
they are self consistent with the model being used when all IL, RL, NEXT,
FEXT and TDR/TDT are matched.
We showed that FEXT and its polarity, among others, can have profound
implication in DK/DF extraction in a DesignCon paper:
http://www.ataitec.com/PDF/Paper_AfullyautomatedSIPlatform.pdf and ;
http://www.ataitec.com/PDF/MPX.pdf
Regards,
Ching-Chao Huang
www.ataitec.com
On Wed, May 24, 2017 at 8:28 AM, Bert Simonovich
<dmarc-noreply@xxxxxxxxxxxxx> wrote:
John,------------------------------------------------------------------
To add to the discussion, the roughness of the copper used in the
fabrication of the core laminate and etching before final
lamination will affect total phase delay which translates into and
effective Dk (Dkeff). So the Dkeff you extract is only good for the
particular geometry measured. It is not the intrinsic Dk of the dielectric
material.
See my DesignCon2017 paper, "A Practical Method to Model Effective
Permittivity and Phase Delay Due to Conductor Surface Roughness".
http://bit.ly/2qWcHPm
Furthermore, since the dielectric is non-homogeneous, the glass
style, resin content, number of dielectric layers used and where
the traces are positioned relative to the glass weave pattern will affect
results.
Best regards,
Bert Simonovich
Signal/Power Integrity Practitioner | Backplane Specialist |
Founder LAMSIM Enterprises Inc.
Email:Lsimonovich @lamsimenterprises.com Web Site:
http://lamsimenterprises.com
Blog: http://blog.lamsimenterprises.com/
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of Yuriy Shlepnev
Sent: 24-May-17 10:37 AM
To: johnlinc@xxxxxxxxx
Cc: jose.moreira@xxxxxxxxxxxxx; 'si-list'
Subject: [SI-LIST] Re: Differential vs. SE for PCB Dk/Df extraction
Hi John,
Let's take a practical case and analyze possible outcomes - such as
FR408HR in the "Lessons Learned" project - paper #2014_01 at
http://www.simberian.com/AppNotes.php
The dielectric around the strips is mostly resin.
If we use single ended strip and identify one effective model for
dielectric, the result is usable for either single-ended strips or
for loosely coupled differential. Nothing else is needed to cover those
cases.
If we use tightly coupled differential traces, the complex
propagation constants for differential and common modes will be
different due to the resin around the strips (result of the
spatial "averaging" of E-field). To identify one dielectric model,
we can use either differential or common mode for the model identification.
The model identified with the common model will be close to the
single-ended case, but not accurate for the differential mode
analysis - does not matter what solver is used. The dielectric
model identified with the differential propagation will be
different, but usable only for the analysis of differential modes
in the line with similar geometry. It will give wrong result for
the common mode propagation. Simply put, isotropic one dielectric model
will not be sufficient for such case.
Two dielectric models are needed - one for layer around the strips
(resin) and one for the rest of the cross-section (three-layer model for
strips, anisotropic dielectric is an alternative).
Such model will be more accurately for all cases - single-ended,
loosely and tightly coupled differential for both differential and common
modes.
Dielectric models for such layered model of the cross-section can
be identified with the two modes in the tightly coupled differential
traces.
Best regards,
Yuriy
Yuriy Shlepnev, Ph.D.
President, Simberian Inc.
2629 Townsgate Rd., Suite #235, Westlake Village, CA 91361, USA
Office
+1-702-876-2882; Fax +1-702-482-7903 Cell +1-206-409-2368; Virtual
+1-408-627-7706
Skype: shlepnev
www.simberian.com
Simbeor - Accurate, Productive and Cost-Effective Electromagnetic
Signal Integrity Software 2010 and 2011 DesignVision Award Winner,
2015 Best In Design&Test Finalist
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of John Lin
Sent: Tuesday, May 23, 2017 9:27 PM
To: shlepnev@xxxxxxxxxxxxx
Cc: jose.moreira@xxxxxxxxxxxxx; si-list
Subject: [SI-LIST] Re: Differential vs. SE for PCB Dk/Df extraction
Thank all to shed light on my questions. I appreciate.
Hi Dr. Yuriy,
Thank you for your insightful explanation.
Based on it, can I conclude that the Dk/Df extraction for
Single Ended shall be close to the that of loosely coupling
differential but different from tightly coupling due to the E field
in between the tightly coupling traces and common mode convention?
If the extracted dk/df are used for simulator to model any
structures, ie SE, tightly/ loosely coupling differential pair
...etc., does Dk/Df extracted from SE makes more sense and more
accurate assuming the tool automatically calculate the E field in
between two tightly coupling differential traces?
Thanks again for helps.
John Lin
2017Е││5Ф°T24Ф?г 04:04О+-"Yuriy Shlepnev"
<shlepnev@xxxxxxxxxxxxx>Е│LИ│?О+
Hi Jose,
Technically, any structure on PCB with measurable parameters can be
used to "tune" the material parameters if we can build a model of
the structure. I agree with your statement that "A simple trace is not a
good structure".
However, use of Beatty standard with de-embedding may be not the
best approach either. First of all because of the de-embedding (as
I can see in your paper) - that is difficult and error-prone for
PCB materials with inhomogeneous dielectrics and large
manufacturing variations. Also, the number of parameters to match
simulations with measurements is also excessive - reflection and
transmission parameters of the de-embedded structure should be
matched simultaneously. This problem is common for all material
identification techniques based on de-embedding - too complicated
and error-prone for PCBs. So, what is better than a simple trace? -
it is two simple traces :-) S-parameters of two trace segments can
be used to extract either reflection-less generalized modal
S-parameters - the simplest form of S-parameters of a t-line
segment for any line with any number of traces (not an
approximation). It is just one complex transmission for single
ended case and two transmissions for differential case (reflections
and mode transformations are zero by definition, not by
approximation). It is easy to fit same reflection-less model for a
t-line segment, to find the material properties - the technique is
in practical use since
2009 - see
#2010_01 and all papers after that at
http://www.simberian.com/AppNotes.php.
The logarithm of the generalized modal transmission parameter
divided by the length difference is the modal complex propagation
constant or Gamma - that can be also used for the material
identification in the same way (sometime called eigenvalue
technique). Though, the techniques with Gamma, such as SPP light
with S-parameters (see
#2016_02 at http://www.simberian.com/AppNotes.php), requires one
additional step - taking the logarithms. Usually it is easy step
and produces the same result as the technique with GMS-parameters.
And, as Scott mentioned, both GMS or Gamma techniques allow easy
dielectric and conductor loss separation.
Considering the single-ended vs. differential, as Gert already
mentioned, the identification results can be very different. This
is because of the layered type of inhomogeneity of the PCB materials.
We always identify some effective permittivity averaged by the
applied electric field. Techniques based on a wide strip line
resonator have preliminary out of plane component of electric field
- the identified value of Dk are good for structures with primarily
out of plane electric field. The other extreme is techniques with E
parallel to dielectric surface - they identify in plane value of Dk.
It is consequence of the layered structure. A strip line with
regular width (close to target impedance) has both out of plane and
in plane electric fields - Dk identified with it will be between the out
or plane (min value) and in plane (max value). See more on that and
references in the "Material World..."
tutorial #2016_01 at
http://www.simberian.com/TechnicalPresentations.php The bottom line ;
is that the end result of the identification should be usable for the
modeling of traces within the actual interconnects width range.
Which value would be better for that? - the answer is obvious, the
values identified with the traces used as the actual interconnects.
Values identified with single-ended can be safely used for loosely
coupled differential, simply because of almost the same structure
of the electric field. However, tightly coupled differential traces
have more energy in the out of plane electric field. The space
between the tightly coupled traces with mostly in plane electric
field can be filled mostly with the resin and property of the resin
may be different from the rest of the resin-fiber mixture. That
changes the identified Dk. The effect shows up as difference in the
phase or group delay of the differential and common modes or as
NEXT on single-ended S-parameters. Layered dielectric model should
be used in such cases to improve accuracy for both differential and
common mode modeling. Use of differential traces is essential for
such cases - see examples is at the "Lessons learned..." paper
#2014_01 at http://www.simberian.com/AppNotes.php
Best regards,
Yuriy
Yuriy Shlepnev, Ph.D.
President, Simberian Inc.
2629 Townsgate Rd., Suite #235, Westlake Village, CA 91361, USA
Office
+1-702-876-2882; Fax +1-702-482-7903 Cell +1-206-409-2368; Virtual
+1-408-627-7706
Skype: shlepnev
www.simberian.com
Simbeor - Accurate, Productive and Cost-Effective Electromagnetic
Signal Integrity Software
2010 and 2011 DesignVision Award Winner, 2015 Best In Design&Test
Finalist
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of Moreira, Jose
Sent: Tuesday, May 23, 2017 1:36 AM
To: johnlinc@xxxxxxxxx; si-list
Subject: [SI-LIST] Re: Differential vs. SE for PCB Dk/Df extraction
In my opinion differential or single-ended makes no difference for
Dk/Df extraction. Important is the kind of structure you use. A
simple trace is not a good structure. I suggest a resonant standard like a
Beatty Standard.
Check the Designcon 2018 paper "Non-Destructive Analysis and EM
Model Tuning of PCB Signal Traces using the Beatty Standard"
Jose
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of John Lin
Sent: Dienstag, 23. Mai 2017 05:53
To: si-list <si-list@xxxxxxxxxxxxx>
Subject: [SI-LIST] Differential vs. SE for PCB Dk/Df extraction
Hi SI gurus,
While using VNA to measurement traces on PCB for loss and Dk/Df
extraction, my colleagues say the differential handhold probe head
can have wider bandwidth than single ended handhold probe head. Is it
true?
Can someone help to explain the reasons?
Also, for Dk/Df extraction, which is better structure, using
single ended or differential traces? Why?
Thank you for helps in advance.
Thanks,
John Lin
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
---
This email has been checked for viruses by AVG.
http://www.avg.com
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
