Yuriy,
You mentioned "we have to homogenize the dielectric".
-I disagree. I think one of the strongest part of FWE is the glass yarn
distance from the surface or from the copper trace.
This could be seen from cross section, although it is rougly related to
resin content percentage and glass yarn thickness. It is also related to
resin flow (affected by copper pattern on design) in lamination process,
which is explained in the research of Joe Dickson and Albert Chen. I am
trying to collect some raw prepreg sheet cross section images with the help
of the material fabricators. One I talked to did not have images like this
readily available. Someone will have to take these pictures.
My hunch is that this will be about measuring a lot of geometry. Even though
it is a lot of work and not seem practical, I will try to do it anyway. At
least for the newer materials. Then instead of calculating FWE exactly I
want to have a table of materials where anyone can compare any two prepregs
(or cores) and decide which one will be a better choice for his stackup. How
much better, that will not be analysed, only the relation.
The table would contain entries like: glass yard distance from surface,
glass pitch, width/pitch ratio. Ideally I would calculate the quality% for
each (as described below in my earlier email), but that would require the A
and B parameters determined from long series of measurements. I am not
equipped for such measurements and research, I design products (boards,
systems) directly.
The determination of the min/max DK as you propose seems to be another good
way of handling the issue. Either from simulation or from measurement with a
pin-like probe or trace. If you build a glass fabric model, then you can
"measure" DK at many points with low cost without having to build test
fixtures. But then the 3D model accuracy might be an issue. The glass fabric
dimensions have some kind of tolerance.
You mentioned that min/max DK would depend on the trace geometry. I suggest
to use thinnest trace possible (3.5mil manufacturing limit for typical
computing or networking production boards) and single-ended, that would
result in worst-case DK variation. All we need is worst-case values in
design engineering, not actual values. Worst-case, folks...
Don't forget DF variation on FWE, when the glass yarn is too close to the
trace, and when it is far! The provided DF values in manufacturer's tables
are probably not worst-case.
Regards,
Istvan Nagy
-----Original Message-----
From: Yuriy Shlepnev
Sent: Sunday, January 10, 2016 12:25 PM
To: buenoshun@xxxxxxxxx ; si-list@xxxxxxxxxxxxx
Subject: RE: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Hi Istvan,
You have started very interesting discussion here and raised a few good
points.
As many people on this list, I am interested in building a practical
solution for modeling or evaluation of the fiber weave effect (FWE). With
Chudy Nwachukwu from Isola we tried to build a "practical model" and
summarized it in Y. Shlepnev, C. Nwachukwu, "Modelling jitter induced by
fiber weave effect in PCB dielectrics", 2014 IEEE Int. Symp. on EMC, 2014 -
see #2014_04 at http://www.simberian.com/AppNotes.php
In that paper we propose to use dielectric mixing formulas and modulate the
volume fraction for glass in resin to build causal modes for skew, mode
transformation and resonances. The goal was to be able to apply the model
without experimental identification of the model parameters - we did not
reach this goal yet.
Theoretically, if you know broadband dielectric models for "glass" (that is
not really glass but a mixture of multiple components) and "resin" (also
complex mixture), know the mixing formula and can put the min/max values on
the volume fraction, you can estimate the mode transformation/skew for some
geometries.
The first problem with this approach, and with all other numerical
estimates, is the absence of models for the ingredients (glass and resin).
The second problem is what mixture equation to apply and how to compute the
volume fraction from the geometry. It may be possible to do it from a number
of cross-sections of the laminate.
Complexity of the problem comes from the comparable scale of the
inhomogeneity in dielectric and strip/trace size. To build a "practical"
model that includes FWE, we have to homogenize the dielectric either
entirely or in some areas. The goal of the homogenization is to find
effective permittivity that relates the averaged electric field and average
electric flux vectors as follows <D>=eps_eff*<E>. 3D EM analysis can be
used to evaluate the min and max of the effective permittivity (brute force
approach similar to experimental investigation). However, if you try it, you
will see that the effective permittivity value will depend on the geometry
(strip width and distance for differential strips). It will be also
different for all three field components (X, Y and Z). It may be not so
practical to do it for all materials in your extensive table :-)
To actually see what I am talking about, please attend our "A Material
World..." tutorial at the oncoming DesignCon -
http://www.designcon.com/santaclara/scheduler/session/a-material-world-modeling-dielectrics-and-conductors-for-interconnects-operating-at-10-50-gbps
Chudy, Scott McMorrow and I will talk about theory of broadband modelling of
composite dielectrics (including FWE) and conductor roughness, manufacturing
aspects and material model identification with many practical comparative
examples.
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 Istvan Nagy
Sent: Sunday, January 3, 2016 11:57 AM
To: si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Hi,
1067 seems to be spread only in one direction. This means you can only have
low skew traces in one direction. In every big high-density board I had we
needed traces on all layers in all directions.
A table in an isola presentation shows that the glass yarn is more than 2x
thinner for 1067 than it is for 2116, and the 2116 yard is almost as thick
as the whole prepreg/core. So the glass in the 2116 can get a lot closer to
the surface. This seemingly correlates to the 1067's higher resin content
(75% vs 55%), and to Lee's test results.
The 2116 has smaller gap and looks more uniform from the top, so contrary to
this if the 1067 still has lower skew, then it can only be explained with
it's thinner glass that is more away from the surface, or with the weave
pitch matching with the diffpair pitch. In your design maybe the diffpair
pich was luckily matching the 1067. My typical diffpairs have 10-11mil pich
due to high density design. So, for me maybe the 1067 (14mil pitch) would be
just as bad as 2116 (17mil pitch).
Instead of anecdotes, I would like to understand the mechanism more.
I thought the spreadness could be quantified bases on the glass
measurements. From the isola presentation the dimensions are x1=yarn
thickness, x2=yarn width, x3=pitch, then y1...y3 the same thing for the
other direction. From this I would calculate spreadness[%]= MIN(x2/x3,y2/y3)
. For 2116 it is 81.9%, for 1067 it is 61.8%. The 2116 is more spread, but
according to Lee Ritchey it still has bigger skew. Maybe we should not
bother with spread glass, just randomly test different fabrics? The diffpair
pitch used for the tests and its effect was not quantified. Were there any
tests with mismatching pitch conducted on 1067, or was the 1067 tested with
matching pitch while the 2116 was tested with unmatched pitch? Then its not
a fair comparison. Testing both with unmatched pitch like 10-11mil or both
with their matching values would be better. If matching gives better result
every case, then it has to be communicated as a pcb design rule.
Any way to quantify the effect of the cross section quality? Maybe we can
use this: The glass yarn distance from surface fluctuates along the weave.
The copper traces are on (or embedded into) the surface. The less it
fluctuates the better. Not in absolute terms (mils), but as percentage. Then
cs_quality[%] = min_distance / max_distance.
Then overall quality would be: quality[%] = (A*spreadness + B*cs_quality)/2.
A and B are weight constants, choose them to match test results (conducted
all with matched or all with unmatched pitch).
I have only seen the yarn thickness table, I still yet to see any cross
section photo. I want to get some cross section photos for each glass style
so I can compare them, too see what exactly is going on.
This above is my speculation, it is not from existing books or something.
I am just trying to be more scientific, rather than using trial&error.
Anyone interested in developing a scientific solution?
Testing is one thing, raw test results is not what I am looking for, but
rather a method that I can apply on any material in the future. The method
has to be based on geometry-data from material vendors. Tests have too many
variables at given values, that might be different in my designs (e.g
diffpair pitch).
regards,
Istvan Nagy
-----Original Message-----
From: Lee Ritchey
Sent: Sunday, January 03, 2016 10:29 AM
To: zhangjun5960@xxxxxxxxx ; si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Glass Weave effects and Cross sectioning
I have data, but it is not available for free. Spend large amounts of money
doing the testing.
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] On ;
Behalf Of jun zhang
Sent: Sunday, January 3, 2016 6:45 AM
To: Lee Ritchey <leeritchey@xxxxxxxxxxxxx>; si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Hi Ritchey,
Do you know how much extra skew they obtain? Do they have measurement datas
for the skew?
Hope to your reply
Regards
On Sun, Jan 3, 2016 at 3:31 AM, Lee Ritchey <leeritchey@xxxxxxxxxxxxx>
wrote:
I have a client who built 400 assemblies with 2116 glass that had
dozens of 10G pairs on them. All were scrapped due to excessive skew.
Enough said!
From: Istvan Nagy [mailto:buenoshun@xxxxxxxxx]
Sent: Friday, January 1, 2016 10:53 AM
To: Scott McMorrow <scott@xxxxxxxxxxxxx>
Cc: Lee Ritchey <leeritchey@xxxxxxxxxxxxx>; SI <si-list@xxxxxxxxxxxxx>
Subject: Re: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Hi,
I am aware of the 3 distinct regions. The ratio of those is better for
2116 than it is for 1067, according to the 2D top view. In
“Better†I mean the 2layer (double fiber region) portion is a lot
bigger than the other 2 (gap and 1-layer) combined. Area measurements
are sometimes provided by the material manufacturer, other times just
I qualitatively compare the 2D photos.
Although I don't know anything about the yarn cross section
difference. I thought the “spread†was to reduce the 0L and 1L
regions. Maybe it is also for yarn cross section uniformity.
“modern mechanically spread fabrics.â€
My problem is that the same glass style from different material
manufacturers (and even the same glass style for same manufacturer but
different material names) has different mechanical spreading. And this
spreading is not really quantified anywhere. Maybe that’s the
missing link.
For example 1080 is available with old style and new style. One vendor
offers 1080 in 3 different spread varieties (standard,
laser-drillable, ss-ultraspread), but the dielectric table does not
really explain which is which. One vendor has distinction like
NEA1080_N136 vs NEA1080_S013.
Maybe we should refer to glass styles as “1234-vendorA-spreadBâ€,
instead of “1234â€. But then the fabs would not be able identify it.
Istvan Nagy
From: Scott McMorrow <mailto:scott@xxxxxxxxxxxxx>
Sent: Friday, January 01, 2016 5:29 AM
To: Istvan Nagy <mailto:buenoshun@xxxxxxxxx>
Cc: Lee Ritchey <mailto:leeritchey@xxxxxxxxxxxxx> ; SI <mailto:
si-list@xxxxxxxxxxxxx>
Subject: Re: [SI-LIST] Re: Glass Weave effects and Cross sectioning
It's not just the gap that is important for skew reduction. What you
do not see in a 2D view from above the weave is the thickness profile
of the weave. It looks more like an elliptical bundle for older
fabrics, and like a mush more spreadout and flattened bundle for
modern mechanically spread fabrics. Because of this, there are 3
distinct regions of fabric that meld into each other, each with a
different dielectric constant when viewed down the z-axis, A high
resin region, or gap. A single fiber region, with higher Er. And a
double fiber crossover region, or "knuckle", that has even higher Er.
Er variation is pretty bad when compared to modern fabrics.
<http://teraspeed.com/images/logo-02.png>
Scott McMorrow
R&D Consultant
Teraspeed Consulting - A Division of Samtec
16 Stormy Brook Rd
Falmouth, ME 04105
(401) 284-1827 Business
http://www.teraspeed.com
On Thu, Dec 31, 2015 at 3:22 PM, Istvan Nagy <buenoshun@xxxxxxxxx <mailto:
buenoshun@xxxxxxxxx> > wrote:
Gap is the area that is not covered with glass, there is only resin there.
You can see on the photos (in my excel file, second tab) as rectangle
empty areas. The 1067 has bigger gaps than 2116, very visible on the
photos.
So, you are saying the glass thread cross section is worse for one
material than for another? Or is where the 2 diagonal threads cross
and they somehow flatten? Do you have some drawing that explains it?
-----Original Message----- From: Lee Ritchey
Sent: Thursday, December 31, 2015 9:42 AM
To: buenoshun@xxxxxxxxx <mailto:buenoshun@xxxxxxxxx> ; 'SI' ;
scott@xxxxxxxxxxxxx <mailto:scott@xxxxxxxxxxxxx>
Subject: RE: [SI-LIST] Re: Glass Weave effects and Cross sectioning
I don't know what you mean by "gap". The reason that 2116 is not good
is that the weave results in peaks and valleys where the valleys are
filled with pure resin and the peaks are almost pure glass.
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
<mailto:si-list-bounce@xxxxxxxxxxxxx>
[mailto:si-list-bounce@xxxxxxxxxxxxx
<mailto:si-list-bounce@xxxxxxxxxxxxx>
] On Behalf Of Istvan Nagy
Sent: Wednesday, December 30, 2015 7:20 PM
To: Lee Ritchey <leeritchey@xxxxxxxxxxxxx
<mailto:leeritchey@xxxxxxxxxxxxx>
; 'SI' <si-list@xxxxxxxxxxxxx <mailto:si-list@xxxxxxxxxxxxx> >;scott@xxxxxxxxxxxxx <mailto:scott@xxxxxxxxxxxxx>
Subject: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Lee,
How come those fabrics yield better result than the 2116?
I have a photo album of glass fabrics from different vendors. The
1067,
1035 seems more loose than the 2116.
http://www.buenos.extra.hu/download/PCB_MATERIAL_LIBRARY.xls
For one vendor:
Measured gaps and coverage:
Fabric Xgap% Ygap%
106 74.05405405 50.48543689
1067 38.11188811 9.48905109
...
3070 14.18918919 11.26760563
2116 18.02325581 16.1849711
...
You see the 1067 has much bigger gaps than 2116.
I followed my intuitive idea that the more dense the fabric on the
photo, the more uniform it is and the least skew we get. Is there
anything wrong with my concept? Am I missing something? The percentage
of 2layer/1layer/0layer areas would determine the uniformity and density.
Layer, I mean layer of glass thread on top of each other. The 0layer
gaps have dimensions of 1.3…3mil (somewhat smaller than trace
width), the weave pitch ranges from 14mil…26mil (several times the
trace width & diffpair gap), while the impedance controlled PCB traces
are 3.5mil…7mil (typically
4.5mil) wide and gaps are like 5-8mil (signal pitch 10mil). Typical
high-density high layer count designs I have seen are like this. So,
none of the fabrics have small enough dimensions to appear smooth to
the traces while having large % gaps. Therefore I relied on the gap
percentage.
Scott,
To achieve the 14mil-18mil signal pitch, you need thick prepreg/core
layers like 6-8mil thick, then the whole stackup is thicker and needs
more layers due to lower density (18mil diffpairs vs 10 mil ones take
more space). So my
18 layer 93mil stackup would become 26 layer and 200mil thick. Is this
what you are suggesting? pitch-matching would need 2x thick PCBs. Nice
theory, but I am not sure about the practicality. On the other hand...
if this is proven practice, I might consider it for my next designs.
Some form factors like ATCA have fixed board thickness, and increasing
layer count by 50% might be hard to justify. We have 200-300 10Gig
diffpairs on my latest board design, although we did not measure skew
on it, just on-die captured eye diagrams. Small company, no SI-team...
"know the fill direction of the laminate that... "
- I will keep that in mind. Somehow it has to go into the fab drawing
gerber file I guess.
So, gap-percentage, pitch-matching, documenting fill direction and
12+deg rotation...
Best regards,
Istvan Nagy
Principal Hardware Engineer
-----Original Message----- From: Lee Ritchey
Sent: Wednesday, December 30, 2015 2:05 PM
To: 'Istvan Nagy' ; 'SI' ; bbakshan@xxxxxxxxx
<mailto:bbakshan@xxxxxxxxx>
Subject: RE: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Scott listed my preferred weave styles in another reply. And, yes we
do very well when we control the weave type. Better than any of the
other methods that have been proposed.
Just had a PCB with 200+ 10 Gb/S paths with the worst skew being 2 pSEC.
-----Original Message-----
From: Istvan Nagy [mailto:buenoshun@xxxxxxxxx
<mailto:buenoshun@xxxxxxxxx> ]
Sent: Wednesday, December 30, 2015 1:35 PM
To: SI <si-list@xxxxxxxxxxxxx <mailto:si-list@xxxxxxxxxxxxx> >;
bbakshan@xxxxxxxxx <mailto:bbakshan@xxxxxxxxx> ; Lee Ritchey
<leeritchey@xxxxxxxxxxxxx <mailto:leeritchey@xxxxxxxxxxxxx> >
Subject: RE: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Lee,
What is on your list of good enough weaves? Are those provide much
better results than 2116? The nicer weaves are not as available as
2116, i am talking about series production of networking and computing
boards, not test boards.
Do you expect the skew problem to be resolved mainly by glass style
selection?
Regards, Istvan, mobile
*** Sent from my BLU smartphone device *** On Dec 30, 2015 9:07 AM,
Lee Ritchey <leeritchey@xxxxxxxxxxxxx
<mailto:leeritchey@xxxxxxxxxxxxx> >
wrote:
I have had severe skew problems with 2116 at 10 Gb/S. It is not on
my list of good weaves.
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
<mailto:si-list-bounce@xxxxxxxxxxxxx>
[mailto:si-list-bounce@xxxxxxxxxxxxx
<mailto:si-list-bounce@xxxxxxxxxxxxx>
] On Behalf Of Istvan Nagy
Sent: Monday, December 28, 2015 10:53 PM
To: bbakshan@xxxxxxxxx <mailto:bbakshan@xxxxxxxxx> ;
si-list@xxxxxxxxxxxxx <mailto:si-list@xxxxxxxxxxxxx>
Subject: [SI-LIST] Re: Glass Weave effects and Cross sectioning
Hello,
First getting on this topic of spread glass, everyone is looking for
the miracle spread that has no skew issues.
You can choose the glass style as a number, or to take it further and
choose a material manufacturer who promises mechanical spread or
square wave or other features too.
In reality the traditional glass styles (106, 1080) were kind of
"bad", newer more denser weaves (2116, 1501) are "better", but there
is no "excellent" fabric.
I just try to use 2116 for layers with 2Gig+ signals. So you specify
glass style, and material name, that's mainly what you can do about
it, before building the stackup for design and manufacturing.
You need to have the detailed datasheets listing all available glass
styles of your chosen material. Get glass fabric photos from your
material vendor (like Bill from NanYa), so you can see how much
difference those different fabrics (numbers) have in covered/uncovered
areas. The board area consists of "no glass", "1-layer glass",
"2-layer glass" parts. If you imagine a simple cloth fabric, you would
know what I mean. The best material has the smallest percentage of "no
glass" area and the largest percentage of "2-layer glass" area, that
would be the most uniform (logically thinking).
"If not, how can I guarantee that the right type of weave was indeed
selected when building my board?"
- instead of "they" building your stackup and board, you have to
determine the stackup before they manufacture it. If you just let fabs
and who knows who manufacture whatever for your board designs, then
many things can go wrong.
Regards,
Istvan Nagy
-----Original Message----- From: Boris Bakshan
Sent: Sunday, December 27, 2015 10:16 PM
To: si-list@xxxxxxxxxxxxx <mailto:si-list@xxxxxxxxxxxxx>
Subject: [SI-LIST] Glass Weave effects and Cross sectioning
Hi experts,
I've been reading much on glass weave effects and its contribution on
skew.
If I encounter a skew problem, can I say anything on weave type
,whether it is mechanically spread or not by doing a cross section of
my traces?
If not, how can I guarantee that the right type of weave was indeed
selected when building my board?
Thank you all!
------------------------------------------------------------------ To
unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx <mailto: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 <mailto: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 Avast antivirus software.
https://www.avast.com/antivirus
------------------------------------------------------------------ To
unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx <mailto: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 <mailto: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 Avast antivirus software.
https://www.avast.com/antivirus
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx <mailto: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 <mailto: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 Avast antivirus software.
https://www.avast.com/antivirus
_____
<https://www.avast.com/antivirus>
This email has been checked for viruses by Avast antivirus software.
www.avast.com <https://www.avast.com/antivirus>
------------------------------------------------------------------
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
