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.
Dependence of Electric field on mesh size in RF modeling
Posted 25.10.2012, 06:10 GMT-4 RF & Microwave Engineering, Mesh 7 Replies
Please login with a confirmed email address before reporting spam
We would like to model our laser in an RF module, solving for electrical field strength, to see the optical field distribution as a function of layer thicknesses, however we kept seeing unexplainable discontinuities in mode distribution when we would remove a layer or leave only a thin layer (~10nm)
One thing that changes with these thin layers is the mesh size in these thin layers. Therefore we checked whether for a fixed design (without thin layers) the results would change if we change our mesh size. The results can be seen in the attached file for 2 mesh sizes, for the exact same geometry (the three layers in the middle are 300, 250 and 300nm thick (with a step in the bottom one). As can be seen both the mode shape and the effective mode index calculated by Comsol do not change significantly. However, upon decreasing mesh size (the right of the figure) the calculated electric field strength keeps increasing, as can be seen by the encircled maximum E-field value. As can be seen this changes by a factor of almost 2 over this mesh-size variation, well outside numerical approximation range.
Plotting the number of mesh elements versus the Electric field yields the graph that is also shown in the attached picture.. Is there a reason we are missing that Comsol gives values this much higher for smaller grid elements? If so, what model settings could be involved?
One thing that changes with these thin layers is the mesh size in these thin layers. Therefore we checked whether for a fixed design (without thin layers) the results would change if we change our mesh size. The results can be seen in the attached file for 2 mesh sizes, for the exact same geometry (the three layers in the middle are 300, 250 and 300nm thick (with a step in the bottom one). As can be seen both the mode shape and the effective mode index calculated by Comsol do not change significantly. However, upon decreasing mesh size (the right of the figure) the calculated electric field strength keeps increasing, as can be seen by the encircled maximum E-field value. As can be seen this changes by a factor of almost 2 over this mesh-size variation, well outside numerical approximation range.
Plotting the number of mesh elements versus the Electric field yields the graph that is also shown in the attached picture.. Is there a reason we are missing that Comsol gives values this much higher for smaller grid elements? If so, what model settings could be involved?
Attachments:
7 Replies Last Post 16.04.2016, 06:31 GMT-4
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
you seem to have a SQRT(N) relationship between mesh number N with the field strength (in 2D), but what I did not get is the mesh density ratio w.r.t. your wavelenegth and geometrical size, for your small feature ? this would tells if its the denser or coarser mesh that is OK, and we talk about the wavelength IN the material, so you need to divide the vacuum wavelength by the respcetive "n" material index of refraction
--
Good luck
Ivar
you seem to have a SQRT(N) relationship between mesh number N with the field strength (in 2D), but what I did not get is the mesh density ratio w.r.t. your wavelenegth and geometrical size, for your small feature ? this would tells if its the denser or coarser mesh that is OK, and we talk about the wavelength IN the material, so you need to divide the vacuum wavelength by the respcetive "n" material index of refraction
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
looks like a problem with boundary conditions.
Ivar: from the attached pictures of electric field, I think he is meshing it small enough to resolve the wavelength.
Ivar: from the attached pictures of electric field, I think he is meshing it small enough to resolve the wavelength.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
I agree but not sure, COMSOL has still not implemented units on the scales ;(
But I do not see, like that, what else could give a SQRT(N) relation (in 2D), bizarre ...
--
Good luck
Ivar
I agree but not sure, COMSOL has still not implemented units on the scales ;(
But I do not see, like that, what else could give a SQRT(N) relation (in 2D), bizarre ...
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Thanks for the comment.
These devices were modeled at a frequency of 8.6e13Hz, corresponding to a vacuum-wavelength of 3.5um in a material with a refractive index of 3.5, so the effective wavelength in the material is about 1um.
For the 2 datapoints corresponding to the coarsest mesh your arguments make sense, in this case the grid spacing in (the coarser) horziontal direction is about 4 times smaller than the wavelength, which could account for odd values. For all the other datapoints however, the grid spacing is much smaller than the wavelength, from the 4th datapoint on there are over 15 mesh elements per wavelength, which should be more than sufficient.
The smallest feature, being the thin layer stretching to the sides is 150nm thick, This corresponds to 1 mesh element for the lowest two datapoints and 3 or more from the 4th datapoint onward.
So if only the first few datapoints would be off this would likely explain it, but I think there's more going on here. We are currently checking whether a similar model we have made before shows the same behavior.
These devices were modeled at a frequency of 8.6e13Hz, corresponding to a vacuum-wavelength of 3.5um in a material with a refractive index of 3.5, so the effective wavelength in the material is about 1um.
For the 2 datapoints corresponding to the coarsest mesh your arguments make sense, in this case the grid spacing in (the coarser) horziontal direction is about 4 times smaller than the wavelength, which could account for odd values. For all the other datapoints however, the grid spacing is much smaller than the wavelength, from the 4th datapoint on there are over 15 mesh elements per wavelength, which should be more than sufficient.
The smallest feature, being the thin layer stretching to the sides is 150nm thick, This corresponds to 1 mesh element for the lowest two datapoints and 3 or more from the 4th datapoint onward.
So if only the first few datapoints would be off this would likely explain it, but I think there's more going on here. We are currently checking whether a similar model we have made before shows the same behavior.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
The other test, on a model with very different geometry but the same physics yielded the same result. Even for extremely fine meshes the values of the E-field change when we refine the mesh. Therefore the thin layers cannot be the problem..
The modeling setup which we use is our sample, consisting of an upper and lower cladding at the top and bottom (n~3.35/3.4) and a thin active region (n~3.4) confined by two thin confinement layers (n~3.5).. The whole sample is in an air box surrounded by a perfectly matched layer to a perfect electric conductor. We don't see any problem with this setup (if description is unclear it's also attached), but then again, we don't see a problem anywhere yet there seems to be one.
The modeling setup which we use is our sample, consisting of an upper and lower cladding at the top and bottom (n~3.35/3.4) and a thin active region (n~3.4) confined by two thin confinement layers (n~3.5).. The whole sample is in an air box surrounded by a perfectly matched layer to a perfect electric conductor. We don't see any problem with this setup (if description is unclear it's also attached), but then again, we don't see a problem anywhere yet there seems to be one.
Attachments:
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hoi
well if nobody out here knows, then probably only "support" could tell, but pls report back, its interesting for us to understand too
--
Good luck
Ivar
well if nobody out here knows, then probably only "support" could tell, but pls report back, its interesting for us to understand too
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
I have similar question to ask. The size for my geometry domain is 1nm for incident wavelengths of 300 - 600nm. How should I decide accurate mesh element size for such a domain size??