First, I do not work for Ansoft, but I do use there products.
Second, I don't know if I'm an expert, but I might have a couple of tips..
* If the program fails the initial mesh for the MAGNETIC solution chances
are there is something funny with the "conductors".
+ Very small gaps between ANY geometry will drive meshing programs crazy.
This also goes for "dummy" objects. Keep "dummy" objects as simple as
possible (i.e. rectangles). Dummy objects are only used to provide extra
mesh points (which result in lessening the amount of mesk objects that
might have acute angles. Acute angles are NOT your friend when dealing
with FEA).
+It helps if you do not allow the edges of any "third" object "touch" any
other obect. This can also make the mesh maker crazy.
+check for unique metalic features... Small "nooks and crannies" (to
borrow the terminology from a Thomas' english muffin comercial) will also
drive the mesh maker crazy. Depending on the desired modeling frequency
constraints, one can sometimes make approximations of the real geometry
that will have little or zero effect on the resultant analysis (i.e. fill
in small gaps). BE CAREFUL with adjustments... remember to keep in mind
wavelengths that the model will be evaluated.
+ With BGA's, I assume that you have many "spherical" objects. Ansoft will
break up a sherical object into a polygon with some number of surfaces. It
is best to keep the number of surfaces to a minimum, but too small. The
lesser the number of surfaces the quicker the solve time the more likely it
will converge). HOWEVER, the true field solutions WILL be different. I am
not familiar with spherical objects so I really don't have any rules of
thumb for accuracies of surfaces vs. object geometry.
+ Try troubleshooting" your problem. Make the simplest geometry problem,
such as maybe only two pins. Optimize your meshing and dummy placement
with this model. This will allow many "quick" iterations to test different
options. Then, once you found the best / quickest / correct solution..
try expanding the proble to the full conductor matrix and apply the same
rules that you found worked with the simple model.
+ Here's something that's not quite obvious... I have found that mesh
times in 3D solvers can be increase by a factor of 2 to 3 because of.......
2D dielectric surfaces. When using the Ansoft drawing tools, it is
pretty easy to create 3D objects by joining two 2D objects. If the
resultant object is a conductor , you need you keep the 2D objects for use
as ports, However... I have found (and I don't really know why) that if
these objects are used for dielectrics, the meshers (again) go crazy...
creating many acute angle objects. There is an easy solution.. remove any
2D object that you can (Make sure they are not needed, first).
Best Regards,
Gus Panella
[email protected]
----------
| From: Wai Yeung Yip <[email protected]>
| To: [email protected]
| Subject: Ansoft mesh generation
| Date: Friday, November 08, 1996 7:52 PM
|
| To the expert Ansoft users out there,
|
| I am doing a lot of BGA modeling using the Ansoft Maxwell 3D and Quick 3D
| software programs, and I routinely run into mesh generation problems. I
am not
| getting satisfactory support from Ansoft and so any tricks I can learn
from you
| would be very helpful. Usually the program fails to even generate the
initial
| mesh. With the Maxwell 3D program I cannot get into using seeded mesh
without
| first having the initial mesh generated. So I have to resort to putting
dummy
| objects in the solution space to help the meshing. Have you any
suggestions as
| to what objects to use? I usually enclose what I consider problematic
part of
| the conductors with boxes but this does not always work.
|
| Even if I succeed in getting the initial mesh, the magnetic field solver
may
| still fail. In this case I have to do seeded mesh. Do you have
suggestions as to
| what seed values to use? Based on my experience, with too large seed
values, the
| mesh generator may fail, or even it does not, the magnetic field solver
may
| still fail. With too small seed values, I end up with too many elements
and I
| run out of memory before I can do enough passes to get a converged
solution.
|
| Another problem I run into is when I have extrusions and I need to
compute a
| port inductance matrix for de-embedding, the port inductance solver may
fail.
| This happens despite the fact that I have already got a converged 3D
inductance
| solution and so the mesh is supposedly adequately fine. Any workarounds?
|
| Thank you very much in advance for your assistance.
|
| Regards,
| Wai-Yeung Yip
| LSI Logic