Discussion Forum

Hi

You have some documentation on thin film structures modelled as surfaces, basically you must add the physics so that the thickness remains a parameter you enter (it's very close to the way COSMOL implements 2D as internally 2D is in fact the same 3D, but with a fixed "depth" mostly 1[m] by default).

The difficulty starts when you have a stack of lets say 2-3 or more thin layers, as you need a surface per layer they should not overlap, on the other side they cannot be the same. Some FEM programmes have specific layer elements that take this into account, but their physics is mainly limited to structural, sometimes with a thermal DoF. Comsol does not have this, and rely on the physics you write.

This would be an interesting development: how to make a multilayer surface, with surface offset etc, but to write that so it's compatible with All Physics (as COMSOL is implemented) is not trivial, still would be very usefull formany of us.

Your error message is probably linked to the high shape ratio 10 mm/10nm it's 1E6:1 (I use 1E4:1 as an absolute maximum limit) yours is about the precision of the binary floating point representation, your matrices become awfully skewed, and probably singular. Furthermore, with such small features you must check your geoemtry "repair" tolerance, that usually is set to 10[um] by default, if you import geometry it will truncate these thin films, and hence further errors.

FEM is nice, and software like COMSOL makes it easier, but still you must check your demand w.r.t. general hypothesis such as RAM available, floating point representations etc

Have fun Comsoling
Ivar

Hi,

thanks for your quick reply, it was reassuring to hear that this seems to be an already established but as yet not solved problem. Do you think that reconstructing the model by simply magnifying it say by 10e4 would possibly be helpful? I am neither a physicist nor a thermal expert, but the PDEs should be linear to the best of my knowledge, and so scaling up the model should still give an insight into the thermal situation. All I want is to see how the Joule heating of my filaments on the thin film spreads through the filaments, and through the films, so if I scale up the power supplied as well as the dimensions, it should still be enough to see roughly what happens. Of course, this neglects the effects that arise due to the nanoscale setting (such as quantum effects that change thermal conductivity at small scales etc...)
In the long term, being able to simulate using the actual dimensions would however be very nice to do.
Thanks again!

Hi

you still have the "scale" during mesh feature, this allows you to make easily some highly uneven mesh elements, still could give you some clues by running your model as 3D and keeping the number of mesh elements low, but you will probably be required to use linear element shape functions as the mesh quality will be very low.

Perhaps worth a try. If you have essentially a thermal question you could perhaps also try to write out the equations directly and simplifiy the layers. But not sure how to start that, someone else out there with some good ideas ?

Good luck
Ivar

Hi Ivar,

to follow up on my last posts: recently, I thought I might try modeling my thin films with faces instead of solids. When I included them in my model as faces, however, I got "Error 2100" again. I assume this is because Comsol can't handle a face just on its own (i.e. without a solid belonging to the face), or is there any way in which I could get itl to do this?

Regards,

Kevin

Hi Kevin
I have run into the exact same problem. I am trying to estimate the temperature distribution arising from Joule heating of a tiny wire (I have elements as thin as 10nm in my device) on a thin film. When I am making the thin film wider than 0.1mm, COMSOL throws an internal error in geometry decomposition. However, when I make 0.1mm wide substrate, everything computes fine.

Could you please tell me how you solved your problem?

Regards
Naga