Cell accuracy of IE3D simulations

Hi all,
I faced this problem while designing strutures on a CPW line using IE3D. The manual claims that the accuracy of simulations will be very good if its about 20cells per wavelength. I started at 15cells/λ to 50. But with every simulation i was getting different results, S-parameters were not converging. WHich results should i take as most accurate. One point i would like to mention is i was interested in 1-5GHz and the structure was 2mm x 4.5 mm, (very small than λ).

Thanks in Advance

you shouldn't appy the λ/20 rule of thumb when the geometry is much smaller than λ. try finer mesh (<λ/50) and see if the result converges.

Or, post your geometry for us to try.

Did you use the proper edge cells? They are very critical. You may also pay attention to the port extension to the typical dimensions (gap and stirp widths). From your posting, it seems to me you are trying to model a tiny structure using some regular meshing. Please send the file to us at: support@zeland.com

Thanks for replies, Well I increased meshing to λ/100 and it converged.
But I have some doubts, How do I decide on edge cells. Am using edge cell of 8.5%, and in the structure i have some slots of 10um and above width, so i can say the min resolution is 10um. Is there any limitation on the minimum size of structure? For such small dimensions what should be the cell per wavelength.
Moreover how to define the size of edge cell for such a structure?
One last thing is how do i calculate the cells for port extension? what is the right number. Currently am using a default of 5 cells. Am also using extension for MMIC port, and Negative ports to feed CPW ground planes.

Thanks In Adv.

The rule of thumb is about 5-15% of the cell size. However, if your structure has small gap such as the gaps for the CPW, you may want to make sure the edge cell size is smaller than the gap size. For the port extension, five cells are normally ok. How big it is? It is shown on the layout editor as grayed lines. However, if you increase the meshing density too much, the five cells may still be short and it may affect accuracy. On the IE3D 11 to be released this month, you will have the option for automatic adjustment so that you will not accidentally cause accuracy problem when you increase the meshing density.

Here is the geo file i was talking about. its gives quite different results when i used 20, 35, 50, 75, 100. Which one is accurate??

I have sent you the suggestions. Basically, if you use the edge cells smaller than the gap width, you will get about the same results using 20 cells per wavelength as the results using 50 cells per wavelength. If you use the 10% criteria instead, the edge cell will be too large compared to the gap width and it creates a difference of about 4-5 dB for S11 below -25 dB. Anyway, it is due to the fact that your structure is very small. You need to adjust the edge cell size to about the same as the gap width, they you can get good results.

Hi Jian

Can you tell us what are the changes in IE3D 11 when the mesh procedure
Will be full automated? Like hfss that make auto mashing until the difference in error between the cells is less then the error you choose.

PL

Hi, Plasma:

Implementing an adaptive meshing scheme is not difficult to us. There are a few reasons why we have not implemented it: (1) MOM is the most stable algorithm. Normally, when we follow the rules (appropriate meshing size with appropriate edge cells), we should get very good results. (2) The computational effort of MOM normally increases faster than FEM with increased meshing density. Adaptive meshing scheme may increase the meshing density substantially. It may reach the limit much faster than FEM. (3) Ansoft had an MOM code called Strata many years ago. They claimed they had adaptive meshing. Some IE3D customers asked me for it. I told them the same thing about the (1) and (2). I am not sure the adaptive meshing in Strata is good or not good. At least, the Strata code had disappeared from market.

On the IE3D 11, we have implemented a new meshng program "Contemporary Meshing". It can reduce the number of unknowns and make the cells more uniform to avoid numerical sigularity especially for general shaped strutures when AEC is applied. The new meshing also allows multiple layers of AEC (for single pass design of 40 dB couplers requested by customers), AEC on thick traces with enhanced accuracy (requested by customers), meshing alighment between closed coupled plates and traces (automating the process for high accuracy simulation of MIM capacitors and structures with finite dielectrics). I think we will consider implementing adaptive meshing in the next revision. Again, implementing it is not difficult. The quesiton is how much it can benefit the users. We are considering some other advanced meshing features may be even more helpful to users. I think we understand what the users need. By the way, please stay tuned. The IE3D 11 should be released in this month, and it has much features. New meshing scheme is only one of the major features. Other major features are (1) distributed EM simulation in mixed Windows and Linux environment. (2) IE3D engine seamlessly integrated into MWO 2004 of AWR. (3) Formula based geometry modeling with easy tuning and EM optimization; (4) Simulation of mixed inifinite substrates and finite dielectircs; (5) Much simplified procedure to create multi-layer structuers with BGA and wire bonds from GDSII and DXF file formats. Future extension will allow users to fully automate the process from GDSII (or other formats) to IE3D full-wave simulations. Thank you very much.

Hi Jian

Do you have in the IE3D 11. the ability to parameterize the structure for
making fast changes? it also can help us in optimization. and parametric study
of the structure.

Thank you v.m.

PL

Yes. We do have it since Ie3dLibray 9.2. However, Ie3dLbrary 9.2 to 10.2 were still are still limited to one dimension associated with one variable. The IE3DLibrary 11 will have formula based geometry modeling. A dimension of an object can be an arbitrary formula entered by the user. It is powerful. Since Ie3dLibrary 10.0, we also allow users to write programs to define their own objects. The users can write their own codes in any language they want. They just need to follow some rules so that the Ie3dLibrary can call the external .exe to get the polygons saved in some fixed format. This is also an extremely powerful command. I know some user design very complicated structures. They have their own ways to generate the structures. They can use the UserDefined Ie3dObject on Ie3dLibrary to generate and optimize the complicated structures.

I can say that IE3D 11 is giving more accurate results than 10.02 when we make comparisons between the the results of SDA and CMT.

Thanks to the Zeland for IE3D 11. We are waiting adaptive meshing to make comparison with AEC.

Added after 14 minutes:

In addition,

I agree with plasma, with need parameterize the structures. I am also using LinMIC interconnect, we can define variables for geometrical dimensions so it allows us to make easy change of the structure.

Thank you plasma.

Thank you for your comments on IE3D 11.

You may try the formula based geometry modeling feature of IE3DLibrary. In the appendix, we demonstrated how we use all symbols and formulas to define all the dimensions of a filter. You can change the dimenisons globally anytime you like. It is a very good and powerful feature even though we need to polish it more. However, it does serve the purposes.

Regards,