spiral inductor simulation
I′m simulating spiral inductors in HFSS and comparing the results with Momentum simulations. There are important differences between them, so, I think something is wrong in my HFSS structure.
I sent it attached, could anyone look at it and give me a hand????
thank u in advanced!
amaya
When simulating spiral inductor you need to do the following:
1) The most important is the return path. Use a loop of perfect electric around the spiral that will be the reference ground. Connect lumped ports from the Spiral edges to the ground.
2) Take a very big air box (in all axis) with radiation boundaries. If you think the air box is too big you can reduce its size and compare the results.
3) The most important thing is to MODEL THE GEOMETRY YOU ARE DOING MEASUREMENTS ON.
4) Check the "Solve Inside" of your metal objects. HFSS will give you the accurate Q factor, since the resistance calculation is perfect.
5) Accelerating the convergence: You might want to add a mesh operation on the inductor (I would use the width of the inductor as the maximum initial mesh element size). In addition you could change the refignment per pass (in solution setup) from 20% to 30%.
Regards,
Itai
There are at least "mistakes" in the project settup:
1. radiation boundary is touching the microstrip=> you need to either 1) draw a bigger air box surrounding your spiral structure, with about 1/4 lambda spacing between the boundary and any of the metal/dielectric; 2) you can remove your radiation boundary setting, and hence the spiral is housed inside a cavity.
2. the width of your wave port is less then the height(thickness) of the microstrip, the wave sent in will not be the quasi-tem wave supported by the microstrip, =>check the waveform calculated by HFSS=> to correct, follow an example HFSS project to setup the waveport (you need to several more things, for example, a metal cap).
Hey, guys,
I have the same problem, there're huge difference between the simulated one and the measured one. Normally, the simulated Q value is larger than the meas. one, but I have the opposite result: the HFSS simulated Q value is about 4~6 less than the measured one at the peak. I followed the example of the Ansoft HFSS offical training material, what's wrong with that??? I'm totally confused...
Another question, how to check the skin effect in HFSS?
Thank you,
Ruri
To check skin effect in any em analysis:
1) Make sure your mesh size in the metal is small compared to the skin depth.
2) Look at the resulting current distribution in 3 dimensions. Do you see the skin effect? Does the current distribution look smooth, with very strong high edge current?
Of course, at low frequency, you should see uniform current everywhere, no skin effect. All EM analyses I know of (including our own Sonnet) can give bad answers if you push them too far. Current distribution is very sensitive to error and is thus excellent for finding problems.
"How small of a mesh is small enough?" is a fair question. Just cut the mesh size in half in all three dimensions in the metal, and reanalyze. Hopefully, the difference will be small. If so, you are OK. If not, keep refining.
For line thickness less than about 5 or 10 skin depths, the above suggestion of "solve inside", or the equivalent in Sonnet of the multi-sheet thickness model, is a good one, provided the mesh is small enough.
For thicker lines, the tube-like model (in HFSS, this is "do not solve inside") that uses an equivalent surface impedance and surface currents should be plenty accurate enough (but always check these things for yourself, if it is important!). Agilent Momentum also has a tube-like model that can be used as well. Sonnet does not have a tube-like model. I don't think you would want to use the tube-like model at lower frequencies, where current flows through the entire body of the conductor, not just on the surface. An expert user can fudge the surface resistance to get the correct low frequency bulk resistance, but the inductance and capacitance will still be off. Unfortunately, it is difficult to check convergence (i.e., cut the mesh in half) with the tube-like model.
There is much more to this than what I have described above, but these should serve as initial talking points if anyone is interested.
Thank you, Rautio, it is so kind of you.
I have one more question, hope to get some advice from you. Since most of the spiral inductors provided by foundries have guard rings (tied to a.c. ground), how to deal with this in EM simulations?
Thanks again, and hope to have your response soon.
Ruri
Just put the gaurd ring in exactly the same way you are going to build it. The guard ring provides an alternate path for ground current to flow (probably lowering the first self resonante freuquency, depending on how close it is to the inductor) and also help keep ground return current from flowing over to adjacent components.
Hi, Rautio,
Thanks for your timely reply.
Actually I've already built the guard ring exactly the same way as the layout, but only leaving it floating (conneted to nothing), the simulated result did not give significant difference with the one without the guard ring. I'm wondering how to let the simulator know that the guard ring is intended to tie to the ground? In other words, how to deal with the guard ring except making it floating?
Best,
Ruri
Ruri -- If you intend to tie it to ground, then proceed to tie it to ground, presumably with vias. Be sure to put the vias in the same place in your EM analysis as you do in your actual circuit. If the gaurd ring makes no difference in the EM analysis, then it is not acting to decrease loss. The only thing it could do is decrease coupling to other nearby components. To check to see if it is doing that, analyze two inductors side-by-side with gaurd rings and without. If there is still no difference, then, best as I can tell, your gaurd rings are doing nothing but taking up space. All your ground return current is flowing in the substrate. In this case, throw the gaurd rings out and make your circuit smaller. Just be sure to check for unwanted component-to-component coupling via the substrate. Let us know what you find out, sounds interesting!
Hi, Rautio,
Sorry for my late response.
Guard ring is a big issue in spiral inductor modeling, but as far as I know, few info. regarding it is availabe. You mentioned to use via (I think it is contact) to ground the guard ring, but how? how to connect one object to the "ground", where is the "ground" in EM softwares?
Guard rings in RF circuit is absolutely necessary because it can be used to relax the substrate coupling. You proposed a good idea to verify the guard ring by constructing two spiral side-by-side. However, I do not have time to do that right now, I may do it later, and let you know the results.
Best,
Ruri
Hi Ruri -- "Ground" is a purely human concept. You can take any point in an entire circuit and call it "ground". The flowing electrons do not care. All the electrons need is a complete circuit. Just imagine the current flowing out the + terminal of the battery and back in the - terminal. For RF, part of the circuit can be capacitance (displacement current), but the circuit must still be complete.
All parts of the circuit are important for RF analysis. That includes the part of the circuit that we decide to call "ground". This is often forgotten by today's RF desginer. So your job is to figure out where the current coming back to the battery (i.e., the two ground pads for your CPW GSG probe) is flowing. Is it flowing in the guard ring, or is it flowing in the silicon substrate? Keep in mind it can flow differently at different frequencies. Just because someone tells you the guard ring is working, does not mean it is working. Check it for yourself.
A via is a vertical conductor. It connects metal on one layer to metal on another layer. If the via penetrates the silicon, it will connect to the silicon. Then current in the silicon can flow to the guard ring, and current in the guard ring can flow to the silicon. Put a via in wherever your guard ring has an Ohmic contact to the silicon. If there is no Ohmic contact to the silicon, do not put in vias. The Sonnet manual tells how to add vias to your circuit.
[/i]
aiturri wrote:
Hi Amaya,
I took the file and HFSS solved with not problems but the resonant frequeny
is about 24GHz, but your sweep just got up to 10GHz. I do not see any problems
with the model but needs some refining like others said: maybe the ports are too
big for example, would create modes that you are not looking.
the model is really small, only 450 μm, that might be the reason for fo=24GHz.
I had the same problem whem comparing momentum and HFSS, but i find got it fixed..
it is simple:
increase the simulation accuracy by decreasing the maximum delta S from 0.01 to 0.005.
Good Luck
Now that you have a spiral simulation going to your satisfaction, you can now look at some really amazing stuff. Just view the current distribution in the spiral.
First, make sure you have a good, smooth current distribution with well defined high edge current. This is important if you want to get an accurate value for something like I2R loss. You have to get the I right in order to get the I2R right. Any EM analysis can fail and a ragged current distribution is a good indication that you are on the edge of a cliff about to fall off. If you want to be sure you have good results, always always check the current distribution. It is very sensitive to numerical error.
The high edge current is important because any time you constrict the current, loss goes up. At first you will notice that much of the current is restricted to both edges on the input and output lines of the spiral.
Then, as you look at the rest of the spiral, something really strange happens. The high current is restricted first to one edge, then to the other, as you go along the spiral! This is called "current crowding". This additional constriction of the current increases the loss.
The cause of current crowding is the magnetic field from the inductor (visualize a bar magnet) goes up through the center, and some of it penetrates the spiral as it comes back down the outside. This magnetic field pushes and pulls the current in the spiral, causing it to flip back and forth between edges.
There is a way to eliminate the additional loss caused by current crowding, I can describe it if there is interest.
If you are not sure that current crowding is included in your analysis, look at the current distribution. If you don't see it, it is not included. Check out the attached image from Sonnet.
so many hothearted guys!
Hi Rautio,
I am very interested in knowing how to eliminate the additional loss caused by current crowding. what I have done so far is only by making the metal thicker. Do you have any other way?
thanks,
wlcsp
Hi Wlcsp -- Look at the current distribution above. (BTW, this is a real inductor on Si from Motorola.) Let's say the spiral line width is 10 microns. To reduce loss, split it into two conductors, say, 4 microns wide. Now, start at the beginning of the outside turn and follow it along. At about 3/4 turn around, the current starts flowing only on the inside edge. That is because there is less reactance there. This is where you swap the two 4 micron lines. Bridge one over the other so that the inside one is on the outside and the outside one is on the inside.
The idea is to make it so there is equal reactance for both edge paths. I have been presenting the above figure for about 3 years to many customers. For most customers, this idea is new. However, I have had 5 customers stand up and tell me that they have tried it and it really works. I have not tried it myself. Feel free to experiment. With a little smaller inductor, you should even be able to do this exepriment with free SonnetLite. Good luck, and if you come up with some results, please post them for the rest of us.
P.S. Whatever EM analysis you use, be absolutely sure to view the current distribution. If error in the analysis overwhelms the current crowding, you will see no current crowding in the current distribution view and attempting this experiment will lead you to incorrect conclusions. You might throw away a good idea because you incorrectly think it does not work. The primary error source in Sonnet is error due to subsection size and this is quantified by doing convergence analysis. The Green's function is calculated to full numerical precision and the de-embedding is exact to within numercial precision provided port connecting lines are not over-moded. The error due to subsection size almost always dominates.