wavelength for cell density

I am trying to simulate a on-chip resonator using IE3D and ADS momentum. I used the default 20cells/wavelength. But the simulaiton results(both) totally not agree with measurement result. The resonator is on a Si substrate. The design kit has substrate model, so there should be no problem with the substrate model. The size of resonator is very small (1.5mmx1.5mm), the simulation is from 0GHz to 20 GHz. what is the normal cell density I should use for small on-chip structure? Thank you.

danda821: For onchip structure or any electrically small structures with fast current variation, you can not use the default 20 cells per wavelength on IE3D. You will need to increase it. As I mentioned before, you would need to use at least 2 segmetns for each arm of a spiral in order to be accurate. You need to increase the meshing frequency (Fmax) or the cells per wavelength (Ncells) until you see the inner most arm of a rectangular spiral is divided into 2 cells, then you will get good results. Another important factor is the strip thickness. For onchip structure, the thickness is no longer small compared to the width, you will need to use the thickness model on IE3D to solve it.

Regards.

Hi Danda821 -- Since we do not know what your accuracy requirements are, it is not possible to say (with certainty) what mesh size you need to get results that meet your accuracy requirements. For example, if you need +/- 5%, or +/- 1% pr +/- 0.1%, you will need different mesh densities for each level of analysis error.

Rather, when high accuracy is required, you should always do a convergence analysis. This is true for any and all EM analysis tools you might use.

A convergence analysis is easy to do. Analyze your circuit with a coarse mesh density. Then analyze it with a second mesh density. Note the difference between the two results. If they are close together (compared to your requirements, whatever those requirements are), then you likely have good results. If the two results are far apart, analyze using an even finer mesh density.

Keep repeating this process until the difference between the results is small compared to your requirements. When accuracy is very important, never never go by a rule of thumb, unless you also check it using a convergence analysis. Or unless you are willing to risk design failure!

This problem should be easy to do with free SonnetLite (www.sonnetsoftware.com), if you would like a third answer. But don't trust Sonnet (or any other EM tool), be sure to do a convergence analysis!

Thank you. I tried to increase the mesh density from 20cells/wavelength to 50, 100, 120cells/wavelength. But there is not much difference. I also consider the metal thickness, not much change. The resonance frequency is totally different (15GHz simulation, 6 GHz measurement)

Hi, Danda821: Please send me the file (jian@zeland.com) and I will help you. Thanks!

Hi, Jian

I have just send you the geo file. Thank you.

In my structure, there are many small metal square fillings, and small holes in large size metal, which are for local density rule. These are not easy to model in IE3D or ADS momentum. I am wondering if there is any easy way to model these in EM tools.

Hi, Danda: You can build the holes on IE3D. Certainly, if there are many of them, it may slow down the simulation. I have sent you the reply on your e-mail. The holes may change your structure's resonant frequency slightly. If the physical length of the 2 terminals of your port is about 3 mm only, the resonant frequency will be around 16 GHz as predicted by IE3D. You mentioned that your measured resonance is 6 GHz. It should not be possible if the length of the loop is only 3 mm with the Ereff about 7-8.