Simulation of a multilayer interdigitated capacitor

时间：03-31 整理：3721RD 点击：

Hi, I am designing a multilayer interdigitated capacitor in a BiCMOS process with 6 metal layers. This capacitor is mainly for decoupling purpose, but some variants are also used for RF purpose in millimeter-wave range. Targeting capacitance falls in the range of 50fF to 300fF roughly.

I have made layout with cadence and extract it with Calibre RC. I have made em simulations with ADS Momentum and Sonnet EM. The resulting capacitor between the two electrodes is very similar with Calibre and Momentum. But the result from Sonnet is about 20% lower. I think I have taken care of deembedding as well as subdividing thick metal enough in Sonnet. But the result remains. Could anybody give me some suggestions about this? Thank you!

How did you generate the Sonnet and Momentum projects? Did you use one of their tools to automatically generate the projects, or did you do it by hand? If doing it by hand, I wouldn't be surprised if there is a difference in the stackup. Momentum and Sonnet handle the dielectrics associated with thick metal very differently, so it is easy to make a mistake in the stackup.

If you auto-generated the projects, let us know how you did it.

Yes, this can be an issue. If you use thick metal definitions in Momentum, you have to realize that Momentum will effectively increase the thicknesses of your dielectrics in order to create models that they can simulate. You have to read their documentation very carefully so you can adjust or compensate your dielectric layers properly for how they handle growing the metal thickness.

I think Sonnet assumes that your thick metals will extend upward from the base level, and they will actually extend your thick metals through dielectrics if the metal is thicker than one or more dielectrics immediately above the base level.

--Max

Thanks you guys. I did generate sub stacks by hand. I think I have taken care of these definitions in the beginning.
In Momentum I have taken thick metal expasion into account when setting dielectric layers. In Sonnet, dielectric layers are independent from metal stack.
Has anyone done these exercises and made any comparison before?

You may find the following related topics of interest for you:

https://www.edaboard.com/viewtopic.p...&highlight=mom

https://www.edaboard.com/viewtopic.p...&highlight=mom

https://www.edaboard.com/ftopic366048.html

Thanks Timof. That's exactly what I need, although some illusions about extraction tools(even some 2D simulators)are blown.

Wait ... if you are using thick metal in Sonnet, this will change the effective metal to metal distance, compared to a thin metal stackup. With increasing metal thickness, the distance to the metal above will decrease.

~~

One detail where things could go wrong with capacitance calculation: By default, Sonnet checks one layer above and one layer below for mesh alignment. If you have more than one dielectric layer between the capacitor plates, you should go to Analysis > Advanced Subsectioning > Polygon Edge Checking and increase the value as needed. Incorrect mesh alignment with big subsections can lead to capacitance values which are smaller than expected.

Would you like to have a reliable reference value for your capacitance (based on a random-walk solver)?

Possibly. Please send me a private message through Edaboard with some information.

That's strange.

From my experience, the solvers agree well on the extracted capacitance for MIM, which is a pretty trivial analysis task. The only issues that I have seen are mistakes in the stack up and errors from mesh alignment.
I am not sure if you mean interlayer capacitor(vertical coupling) by MIM? I used both vertical and horizontal coupling, maybe that's where the difference is.

Yes, that is different. Most of my work was with "normal" MIM capacitors with horizontal plates (except for this).

That's strange.

From my experience, the solvers agree well on the extracted capacitance for MIM, which is a pretty trivial analysis task. The only issues that I have seen are mistakes in the stack up and errors from mesh alignment. Yes, this is called "ease of use". If a user needs to spend lots of time defining and verifying the input (geometry, dielectrics, meshing, etc.), and if this user in a non-expert in this tool - then the tool is not easy to use, and there is a high probability of making a mistake in defining the input...

Regarding vertical or lateral capacitive coupling - once the 3D geometry (including metals and dielectrics) and material properties are defined, it should not be a problem for a field solver to calculate the capacitance (unless the solver is using some simplifying assumptions, and is not limited by the mesh size when simulation domain is large).

I would call it the "generic solver" vs. "specialized solver" choice.

If you do nothing but interdigitated capacitors for the rest of your life, a very specialized tool will be the right choice. But if you also do capacitors and inductors and transformers and interconnects, a generic solver will be the preferred choice.

Wait ... if you are using thick metal in Sonnet, this will change the effective metal to metal distance, compared to a thin metal stackup. With increasing metal thickness, the distance to the metal above will decrease. Smaller distance between metals would produce higher capacitance, while he is seeing ~20% lower capacitance...