Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
How to mesh this micro-mirror model
Posted 8 déc. 2010, 08:33 UTC−5 4 Replies
Please login with a confirmed email address before reporting spam
Recently I am doing MEMS research about micro-mirror which contains intrinsic stress.
I have already created the model, but there exists some problems in meshing.
The model consists of 3 layers:
The 1st layer(bottom) Si(c), called device layer:
chip 16*16mm square
the radius of membrane? 3.5mm
20 clamped beam suspended around the rim of the membrane length: 125um width: 0.28 degree
thickness: 5um
The 2nd layer(middle) SiO2, called oxide layer:
16*16mm square
the radius of the hollow circle: 3.625mm
thickness: 1um
The 3rd layer(top) Si(c), called handle layer:
16*16mm square
the radius of the hollow circle: 3.625mm
thickness: 380um
First I meshed the boundary of the 1st layer(bottom) using free triangular, then I swept the mesh to this domain, it worked.
Second I meshed the 3rd layer(top) using the same method and it also succeeded.
But when I tried to mesh the 2nd(middle) layer using swept, the error came out "Mesh error (swe5): Unsupported topology of domain."
I don't know the reason, so can anybody help me to solve this problem or if you can mesh this model using another
wise method to get good mesh quality, please tell me!!!
The model is attached below.
I'll appreciate your help very much!
I have already created the model, but there exists some problems in meshing.
The model consists of 3 layers:
The 1st layer(bottom) Si(c), called device layer:
chip 16*16mm square
the radius of membrane? 3.5mm
20 clamped beam suspended around the rim of the membrane length: 125um width: 0.28 degree
thickness: 5um
The 2nd layer(middle) SiO2, called oxide layer:
16*16mm square
the radius of the hollow circle: 3.625mm
thickness: 1um
The 3rd layer(top) Si(c), called handle layer:
16*16mm square
the radius of the hollow circle: 3.625mm
thickness: 380um
First I meshed the boundary of the 1st layer(bottom) using free triangular, then I swept the mesh to this domain, it worked.
Second I meshed the 3rd layer(top) using the same method and it also succeeded.
But when I tried to mesh the 2nd(middle) layer using swept, the error came out "Mesh error (swe5): Unsupported topology of domain."
I don't know the reason, so can anybody help me to solve this problem or if you can mesh this model using another
wise method to get good mesh quality, please tell me!!!
The model is attached below.
I'll appreciate your help very much!
Attachments:
4 Replies Last Post 7 janv. 2011, 02:18 UTC−5
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Can anybody help me? waiting for your help~
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
You have a tricky issue there because your 1um oxide layer is so thin that you need very many mesh elements to have a resonable mesh shape (form factor). Either you should simulate it as a thin surface film and put the pysics in the layer, or you mut expect to use far more elements (some 2-3um in X-Y extend to be comparable with the 1um Z directon. And ideally to simulate thine shells you should have 3-4 elementsin the thickness, hence <1um elements in the x-Y direction and this for both of youre thin structure). Yoi might well end up with a RAM issue there too thereafter.
The main reson its complaining are twofold:
1) when sweeping start in one end and move onwards, not from both sides, the middle layer canot be sweept as its mostly oversconstrained
3) your eoemtry is not correctly overlapping there are some silver parts due the the sizes nd the values you are using.
Other comments:
a) no nee touse a sweep if its for 1 element in thickness, you can as well use standard 3D free theths
b) you should construct 1/4 or 1/8 and mirror the 3D parts at the end, the extra internal boundaries will make your mesh more symmetric and the results will be more coherent.
Meshing is partly an art, and you should at least run 2 cases by refining the mesh once to see if are not too mesh dependent
--
Good luck
Ivar
You have a tricky issue there because your 1um oxide layer is so thin that you need very many mesh elements to have a resonable mesh shape (form factor). Either you should simulate it as a thin surface film and put the pysics in the layer, or you mut expect to use far more elements (some 2-3um in X-Y extend to be comparable with the 1um Z directon. And ideally to simulate thine shells you should have 3-4 elementsin the thickness, hence <1um elements in the x-Y direction and this for both of youre thin structure). Yoi might well end up with a RAM issue there too thereafter.
The main reson its complaining are twofold:
1) when sweeping start in one end and move onwards, not from both sides, the middle layer canot be sweept as its mostly oversconstrained
3) your eoemtry is not correctly overlapping there are some silver parts due the the sizes nd the values you are using.
Other comments:
a) no nee touse a sweep if its for 1 element in thickness, you can as well use standard 3D free theths
b) you should construct 1/4 or 1/8 and mirror the 3D parts at the end, the extra internal boundaries will make your mesh more symmetric and the results will be more coherent.
Meshing is partly an art, and you should at least run 2 cases by refining the mesh once to see if are not too mesh dependent
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi, I am having the similar challenge. It took long time to solve and converge.
Therefore I want to mirror the 3D part. But I do not know how to do it in COMSOL 3.4
My question is:
1: There is no button named 'mirror'.
2: How to set the boundary condition for 1/4 or 1/8 parts before solving it.
3: Does "Sweep" mesh give right answer in the direction perpendicular to the direction of the" sweep"? Or it is just accurate in the sweep direction.
Thank you very much.
Therefore I want to mirror the 3D part. But I do not know how to do it in COMSOL 3.4
My question is:
1: There is no button named 'mirror'.
2: How to set the boundary condition for 1/4 or 1/8 parts before solving it.
3: Does "Sweep" mesh give right answer in the direction perpendicular to the direction of the" sweep"? Or it is just accurate in the sweep direction.
Thank you very much.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
3.4 is really too old, haven't used it for years now, so I cannot really tell how to any longer ;)
but mirror a partial mesh (into a symmetric geometry) is still missing in 4.1 and it would be very, very handy for complex models ;)
Setting up boundary conditions for 1/4 1/8 are tricky and you must really learn what you are doing. Again its model type dependent. All symmetry, antisymmetry conditions are in COMSOL. But if you are doing eigenfrequency analysis of structures, you must run several runs with all combinations of symmetric/antisymmetric conditions to get out ALL modes, otherwise you are restricted to a limited set via the symmetric BCs
Last a sweep mesh is in my opinion correct in all direction, it's not the mesh sweep method, nor tri/quad but rather the mesh density and quality (size and size ratio) that matters (and the available RAM to handle all)
--
Good luck
Ivar
3.4 is really too old, haven't used it for years now, so I cannot really tell how to any longer ;)
but mirror a partial mesh (into a symmetric geometry) is still missing in 4.1 and it would be very, very handy for complex models ;)
Setting up boundary conditions for 1/4 1/8 are tricky and you must really learn what you are doing. Again its model type dependent. All symmetry, antisymmetry conditions are in COMSOL. But if you are doing eigenfrequency analysis of structures, you must run several runs with all combinations of symmetric/antisymmetric conditions to get out ALL modes, otherwise you are restricted to a limited set via the symmetric BCs
Last a sweep mesh is in my opinion correct in all direction, it's not the mesh sweep method, nor tri/quad but rather the mesh density and quality (size and size ratio) that matters (and the available RAM to handle all)
--
Good luck
Ivar