Gary,
This topic would take a longer answer to fully address, but I'll
try to be brief -
I have long since gotten away from the philosphy of letting the
supplier have so many degrees of freedom; it isn't necessary and
it does and will continue to cause problems.
Your SI engineer is right: the stackup should be specified as to
the exact interplane thickness. The P/G separation can at certain
frequencies attenuate resonances - there is a growing body of
papers on this topic. In addition, as you found out (and we did
also) too little of the not right kind of prepreg fill when the
dielectric spacing is a fill and not a core, can also give you
heartburn. But I also insist on specifying signal to signal
distances even though they have only a nominal effect on impedance,
because they can have a drastic impact on crosstalk. So, in other
words, in my opinion, some designs are clearly in the region of
requiring all or most all controllable parameters to be defined.
It has been somewhat an uphill battle as the suppliers really don't
like to have the constructions pinned down and the engineers really
don't want to have to work out constructions ... but it has to
happen. Because if you bring up a second source, or even, move the part
around from facility to facility within one company (say from the
prototype facility to the volume facility) the part will not work
the same - it may not even work - and both parts would still meet
the specifications.
Yes, this does mean that you will need to get involved in tolerancing
those parameters like thickness. I use tolerance as a tool: for those
separations which are critical, I use the tightest customary
tolerance (uusally Class II); then for those which are not so
critical, I use a loose tolerance; that is where I expect the supplier
to make up or loose thickness and I know where that is and I still
have it banded.
As for the actual constructions, you do need to make sure that aren't
so specific that you are locked into one supplier -- so, for every
construction I have at least 2 suppliers do a DFM; then I close any
discrepancies usually by averaging a nominal and/or bumping up a
tolerance: now I know 2 can build it and I know what it will be and
I know they will be reasonably close to each other.
Yes Er can change material to material, but also separation to
separation, and, by % resin etc... and then there is the concept
of "effective Er" .... basically what I have been doing that seems
to be working so far, is to state on the print what the modeling
assumptions are: impedance based on material with dielectric
constant of x.x +/- x -- this at least gets them working with the
same base resin/material set and/or if their numbers are very
different for their models it should set off an inquiry.
Good luck,
Valerie
------------- Begin Forwarded Message -------------
X-MTHubFilter-1.6: mail-srv1
Date: Wed, 10 Jul 2002 12:29:59 -0600
From: gacrowell <[log in to unmask]>
Subject: [TN] Stackup specification, tolerance, power plane resonance?
To: [log in to unmask]
I'm looking for some guidance regarding stackup specifications. Its a long
story, I hope you'll bear with me.
We had a fairly complex board fabbed and working. Over a couple of earlier
versions, we had built about 60 of the boards. The stackup for these
versions of the board is listed below. You can see that it has 4 internal
impedance controlled routing layers, plus some heavy copper internal power
planes. The board consumes some pretty high current, and distributes quite
a few different voltages.
Note the 3 mil spacing between planes, which is placed between power and
ground, just because it was thought to be a good idea for the (patented,
unmentionable) interplane "C" word.
On a similar, but different board, where the stackup had also listed 3 mils
between planes, the stackup was implemented by the fabricator with a single
prepreg sheet, and this apparently resulted in resin starvation in some
areas and we got some very exciting interplane shorts. (Other fab houses
had implemented the same stackup with two thinner prepregs.) We haven't
desired to specify lamination material in the past, preferring to leave the
implementation to the board house.
So now it comes time to do a new 'production' version (our production
quantities are very low, never numbering more than a hundred boards or so).
A couple of last wires are incorporated in the layout, and just to be safe,
the stackup is changed to eliminate the 3 mil interplane spacings, changing
them to 4 mil. This stackup is also listed below. Other interplane
spacings have been reduced to maintain the thickness.
This new board was ordered from a contract assembler, and that assembler
chose the board house and placed the order, using our specifications. We
have no issue with the board house chosen, its a perfectly reputable house
that I knew of, and I worked with them resolving minor issues in the board
fab on this and other boards.
Now comes the problem. On the new board, one aspect, completely unrelated
to the minor wire changes, does not work. Same components (even swapped
components from a working, older version), same component placements and
routing. Impedance on signal layers seems to be within spec (more checking
to be done). Some of the boards can be gotten to work (sometimes
intermittently), by adding caps around the particular chip that is giving
the problem, some gotten to work by adding bulk capacitance on the board,
and some aren't working yet regardless of changes.
The board is liberally populated with 0603 .01uF, and 0508 .1uF caps, and
33uF tants.
The boards that work with the addition of bulk caps seems to suggest that we
may be shy in that area. But change in operation with other caps seems to
suggest that we have a power plane resonance problem. More testing is being
done.
The engineer on this board is understandably frustrated, and complains that
the stackup should not have been changed from the earlier version, and that
the stackup should be specified as to the exact interplane thickness. The
new boards are .005" thicker than the old, but within the .093 +/-.008
tolerance.
I have tried to explain that (to my knowledge) it is common to not specify
exact laminate material, and that recipes for stackups to achieve certain
thicknesses can change from fabricator to fabricator, Er changes from
supplier to supplier, and a dozen other factors can change with the wind and
phase of the moon. The final target as I understand it is that impedance,
overall thickness, and copper weights are requirements that must be met
within tolerance, but that the fabricator will adjust as required within his
process to hit those specs.
So the questions:
Is it time to specify laminate material in detail? Where do I start? Do I
get the laminate selection from each particular fabricator, or require that
they buy what I specify? Will this affect cost?
Should I get the exact stackup recipe from the fabricator and require this
be maintained for new versions or other board houses?
Should I be placing a tolerance on interplane thicknesses? Would this
affect cost?
What should have been done to avoid this particular situation?
Your comments and suggestions would be appreciated.
Gary Crowell
Micron Technology
(fixed pitch font is probably required to get these to line up correctly)
EARLY VERSION STACKUP:
Filename Layer Cu plating prepreg est trace trace
oz oz in. width in. D-Code
============= ========= ==== ======= ======= ========= ======
20060r1.TSP top paste
20060r1.TSK top silk
20060r1.TSM top mask
20060r1.BRD board outline
20060r1.L01 top .5 1.0 .0065
prepreg .004
20060r1.L02 plane .5
core .004
20060r1.L03 signal .5 .005
prepreg .005
20060r1.L04 signal .5
core .004
20060r1.L05 plane 1.0 .005
prepreg .003
20060r1.L06 plane 1.0
core .006
20060r1.L07 plane 1.0
prepreg .003
20060r1.L08 plane 1.0
core .003
20060r1.L09 plane 2.0
prepreg .006
20060r1.L10 plane 2.0
core .003
20060r1.L11 plane 1.0
prepreg .003
20060r1.L12 plane 1.0
core .006
20060r1.L13 plane 1.0
prepreg .003
20060r1.L14 plane 1.0
core .004
20060r1.L15 signal .5 .005
prepreg .005
20060r1.L16 signal .5 .005
core .004
20060r1.L17 plane .5
prepreg .004
20060r1.L18 bottom .5 1.0 .0065
20060r1.BSM bottom mask
20060r1.BSK bottom silk
20060r1.BSP bottom paste
finished thickness: 0.093" +.008"/-0.008"
LATEST VERSION STACKUP:
Filename Layer Cu plating prepreg est trace trace
oz oz in. width in. D-Code
============= ========= ==== ======= ======= ========= ======
20060R3.FAB fab
20060R3.TSP top paste
20060R3.TSK top silk
20060R3.TSM top mask
20060R3.L01 top .5 1.0 .0065
prepreg .004
20060R3.L02 plane .5
core .004
20060R3.L03 signal .5 .005
prepreg .005
20060R3.L04 signal .5
core .004
20060R3.L05 plane 1.0 .005
prepreg .004
20060R3.L06 plane 1.0
core .004
20060R3.L07 plane 1.0
prepreg .004
20060R3.L08 plane 1.0
core .004
20060R3.L09 plane 2.0
prepreg .004
20060R3.L10 plane 2.0
core .004
20060R3.L11 plane 1.0
prepreg .004
20060R3.L12 plane 1.0
core .004
20060R3.L13 plane 1.0
prepreg .004
20060R3.L14 plane 1.0
core .004
20060R3.L15 signal .5 .005
prepreg .005
20060R3.L16 signal .5 .005
core .004
20060R3.L17 plane .5
prepreg .004
20060R3.L18 bottom .5 1.0 .0065
20060R3.BSM bottom mask
20060R3.BSK bottom silk
20060R3.BSP bottom paste
finished thickness: 0.093" +.008"/-0.008"
--------------------------------------------------------------------------------
---------------------------------------------------------------------------------
Technet Mail List provided as a free service by IPC using LISTSERV 1.8d
To unsubscribe, send a message to [log in to unmask] with following text in
the BODY (NOT the subject field): SIGNOFF Technet
To temporarily halt or (re-start) delivery of Technet send e-mail to [log in to unmask]: SET Technet NOMAIL or (MAIL)
To receive ONE mailing per day of all the posts: send e-mail to [log in to unmask]: SET Technet Digest
Search the archives of previous posts at: http://listserv.ipc.org/archives
Please visit IPC web site http://www.ipc.org/html/forum.htm for additional
information, or contact Keach Sasamori at [log in to unmask] or 847-509-9700 ext.5315
---------------------------------------------------------------------------------
|