Lange coupler: mom and measurement difference.

[Grig]

Hi

Has anybody the design example of 3dB Lange coupler on lumina substrate.
I'm interesting what difference was between agilent momentum result and measurement result?

Regards

 

[mesfet]

Hi Grig,

I did it several years ago. It's build on FR4 substrate, I remember it was about 2mm in thickness. Frequency about 1.8GHz. ADS is pretty accurate, I used momentum. Care must be taken for the wire connection during modeling, make sure it is connected by simulate at DC for checking connectivity or you can simply extract the model from momentum and simulate it in schematic level.


Mesfet+

 

[Manjunatha_hv]

Hello Grig, & mesfet

I found
1. AWR Microwave Office is good for Lange coupler design & Optimization...
MLANGE1 - EM quasi-static model (more accurate) for fingers of the coupler only & use EMSight for Interface interconnects..(Circuit & EM Co-simulation)

2. Also Sonnet is is very good for accurate EM simulations of the complete lngcp...


Attached are the complete documents & Project files, MWO, Sonnet & ADS related to Lange Coupler....

 

[Grig]

Hi

Thanks, it's realy interesting .

Have You something more (may be)? For ex:
Lumina substrate 3 db coupling Lange: topology description, momentum result, measurement result...

the reason of my question:
Someone tell me, that momentum can not predict accurately Lange performance (Lange with strong coupled fingers)-the reason --- momentum not taken in to account metal thickness

Thanks fo reply

 

[mesfet]

Hi Grig,

I believe metal thickness can usually be neglected as long as substrate thickness is much thicker than the metal, which is usually the case. I have the experience that metal thickness had to be taken into consideration when designing LTCC, and HFSS was needed.


Mesfet+

 

[jian]

Hi, Grig:

Thickness is very critical in modeling LANGE coupler. The reason is that the strip thickness in the fingers is no longer small compared to the width. If you do not use thickness in the model, you may get quite big difference. IE3D from Zeland (where I am working) has thickness model for more than 10 years and it is reliable. You can give it a try. Regards.

 

[rautio]

I've heard a couple sources state that metal thickness compared to substrate thickness is what is important. I imagine it might be the case, however, it is a very rare case. And, it is not sufficient by any means.

There are two much more common cases that must also be considered. First, as Jian suggests, metal thickness is often important when thickness is about the same as line width. Second, thickness is important when thickness and the gap between lines is about the same.

However, these are just rules-of-thumb. Do not trust rules-of-thumb. If you have any doubt what-so-ever, analyze your circuit with thickness, then without thickness and note the difference. (All significant EM tools have thickness models.) Then, to be sure, increase the meshing density (including in the vertical direction) to see if you have convergence to the degree that you need. Now, you will have good solid numbers on the analysis error. No need to ask people for their opinions on the matter.

 

[flyhigh]

Hi,

just to add something about Lange couplers in general. To my experience (some but not impresive) wire bonds are the main source of difference between simulations and measurements. No matter how you model them in EM simulator they tend to be different in a real circuit. So, wire bonds could add some parasitics that are out of control.

flyhigh

 

[maxwellian]

Hi,

just to add something about Lange couplers in general. To my experience (some but not impresive) wire bonds are the main source of difference between simulations and measurements. No matter how you model them in EM simulator they tend to be different in a real circuit. So, wire bonds could add some parasitics that are out of control.

flyhigh

I have found that more of the differences between modeled and calculated are due to whether or not the model in the EM simulation took into account metal thickness, more so than how the wirebonds were modeled. I suppose that if you are using very high wirebonds or very wide (or multiple side-by-side) flat ribbon bonds for a design, you could add some amount of extra phase in them to throw the results off a bit, but I think in practice most people keep these short enough. There might also be some issues if you're not very accurate about the placement of the wirebonds.

My experience has been that the wirebonds are a secondary issue to the metal thickness modeling issue. Usually what happens when you model Lange couplers with "flat" metal models, you get pretty good results for the input reflection coefficients, but and the "through" port is usually OK as well, but the coupled port prediction is off--the EM simulator predicts an undercoupling.

This usually is corrected by modeling the coupled lines with finite metal thickness.

Modeling the actual metal thickness accurately in a 3D simulator leads to pretty long simulation times. By accurate simulation I mean meshing the sidewalls of the metal well. Some of the planar simulators now do a pretty good job at modeling metal thickness for this kind of circuit.

--Max

 

[Grig]

OK, Max

And what do You think --- Side walls , substrate edges can produce different coupling or not?

I know that for narrow band filters enclosure effect exists (and was published)...

What's about accuracy of ADS2004 momentum modeling of thick metal current distribution.

I have understending that multiple measurements on number of structures needed to have unswer, but have no possibility for such measurements. (Then I must ask anybody).

Thanks

 

[maxwellian]

OK, Max

And what do You think --- Side walls , substrate edges can produce different coupling or not?

I know that for narrow band filters enclosure effect exists (and was published)...

What's about accuracy of ADS2004 momentum modeling of thick metal current distribution.

I have understending that multiple measurements on number of structures needed to have unswer, but have no possibility for such measurements. (Then I must ask anybody).

Thanks

Grig:

Yes, sidewalls (for simulators that have them) can influence the results if they are too close. I (just my opinion) that these shielded analysis techniques work well as long as you keep the sidewalls at a distance away about equal to the distance to the ground plane; at least for higher dielectrics (like alumina, gallium arsenide, and other substrates with dielectric constants on the order of 8 or higher--rule of thumb). I think the sidewalls show more influence on weakly coupled structures than with tightly coupled structures as in Lange couplers.

For substrates with lower dielectric constants, you probably need to move PEC sidewalls a little further away (2xground distance or more) so that they don't "grab" field lines where this would give bad results.

Substrate edges? I think most planar solvers assume that either you have a PEC coating on the edge of your workspace, or they assume infinite substrate extents. So I don't think they model substrate edge effects. Unless a truncated substrate is very close to the circuit in question, I don't think a structure like a Lange coupler would be affected much by this.

And yes, I agree that for narrow-band filters the enclosure effects can be very important (as has been published by Matthei and Rautio, and others). I think these fall into the "weakly coupled" resonator cases that I mention above.

I don't know much about Momentum's thick metal model in 2004A. There are no publications to back up what they are doing, so it is hard to know if it really adds anything that we don't already see with existing planar simulators that handle metal thickness. That's the trouble with a lot of these EM vendors--they don't publish much on their theoretical bases, so we are left with having to evaluate things for ourselves (which we should do anyway). In the 2004A software it looks to me like they finally followed the rest of the crowd and use basically a 2-sheet model for metal thickness. It has to be better than using their single sheet modeling for thick metal, but is it good enough...? Probably depends on your application.

--Max

 

[flyhigh]

I have found that more of the differences between modeled and calculated are due to whether or not the model in the EM simulation took into account metal thickness, more so than how the wirebonds were modeled. I suppose that if you are using very high wirebonds or very wide (or multiple side-by-side) flat ribbon bonds for a design, you could add some amount of extra phase in them to throw the results off a bit, but I think in practice most people keep these short enough. There might also be some issues if you're not very accurate about the placement of the wirebonds.

--Max

This is generally true, however the acctual influence of wire bonds is dependant on operating frequency as well. Alumina is high dielectric constant substrate and the bonding effects at 2GHz might be neglected but at 25GHz they can play a role.

Of course, one can change the substate to lower dielectric constante to make these effects secondary again.

If you model (mesh) finite metal thickness and are sure that it is most significant cause of discrepancies, than you also might need to take into consideration the acctual shape of the edge (underetching, overetching, line width/slot width variations etc.) as this is obviously critical. Depending on the technology available and the price one can afford, edges could look really scary under the microscope! I think this should also get worse with increasing substrate dielectric constant.

flyhigh

 

[maxwellian]

This is generally true, however the acctual influence of wire bonds is dependant on operating frequency as well. Alumina is high dielectric constant substrate and the bonding effects at 2GHz might be neglected but at 25GHz they can play a role.

Of course, one can change the substate to lower dielectric constante to make these effects secondary again.

If you model (mesh) finite metal thickness and are sure that it is most significant cause of discrepancies, than you also might need to take into consideration the acctual shape of the edge (underetching, overetching, line width/slot width variations etc.) as this is obviously critical. Depending on the technology available and the price one can afford, edges could look really scary under the microscope! I think this should also get worse with increasing substrate dielectric constant.

flyhigh

flyhigh:

You make good points. Are you doing Lange couplers on alumina at 25 GHz? That must be fun. I usually think of these being done on MMICs at such frequencies, and the bonds (airbridges, actually) are usually pretty well controlled. But I can see how a long or poorly made wirebond at millimeter-wave frequencies could throw things a bit.

Yes, you're right on with concern about over-etch or under-etch. The question then becomes one of "how to I simulate a conductor that doesn't have a vertical sidewall?" Or worse, modeling side-by-side conductors (coupled lines) that have non-vertical sidewalls. On this, I have had reasonably good results by approximating with vertical side-wall conductors where the vertical sidewall is placed at about a mean distance between the max and min sidewall extent. Not perfect, but it seems to get the job done reasonably well. How would you handle this?

--Max

 

[flyhigh]

Hi Max,

no, I am not doing Lange for my 25GHz project, but it was considered as an option and discarded. The problem is that the common substrate is defined with the rest of the sub-system parts and "something" wideband to split power was needed for the job. We solve it finally with completely different structure, it is subject to patent pending so I will not elaborate on this any more.

Considering edge shape, I am not sure what to say. It is good to have EM tools life was horrible for MW engineers before they become handy. Sometimes, however, one have to look under the microscope to see how things really look like. I did some LTCC realted work in the past. Ellyptical conductor cross section, conical vias etc. Possible to EM model but too complex to work with this in real time. Could you imagine optimizing multilayer circuits with all the conductors and vias having this "real" shape? Furthermore, all vias were conical but not 2 of them were the conical in same manner!

Maybe my conclusion is that one should be realistic about the differences of simulated and really made circuit and never try to track out 0.2dB difference between measured and simulated. For some cases even 1dB is perfect match in the "pass band" of the circuit.

Also, I think there is a conclusion (think I red it in Besser's book) that insertion loss coming out of EM simulators is always optimistic compared to measured results. This may also come from numerous idealizations.

Just to make myself clear, this is not the disadvantage of any EM simulator! This is the philosophy of modeling the nature by humans.

regards

flyhigh

 

[Grig]

Hi, flyhigh

I'm agree "one should be realistic about the differences of simulated and really made circuit " but one should understand which factor gives 1dB which gives 0.2dB


Is there any book or article about such analysiz?

Can You propose some book about "philosophy of modeling the nature by humans"?


Thanks

 

[flyhigh]

Hi, flyhigh

I'm agree "one should be realistic about the differences of simulated and really made circuit " but one should understand which factor gives 1dB which gives 0.2dB


Is there any book or article about such analysiz?

Can You propose some book about "philosophy of modeling the nature by humans"?


Thanks

Hi

an engineer should try to get each component as close to desired performance as possible. However, even when this effort is done to the large possible degree, there is stil going to be some disagreement present. One should put an effort to understand where is the cause of the problem, this is something we always do, but even if you know where the disagreement comes from, you might not be able to quantify it exactlly or do anything about it to improve it. Thus you have to work with what you have on your disposal.

You never quantified the disagreement between the measurements and simulations, nor specified the frequency, substrate etc., so if you are well off the performance my points will be wrong. Maybe you can correct your simulations to account for conductor thickness but could you garantee it is not going to be yet again just less undercoupled or may become overcoupled? You know the couse, you have the tool to bring your design closer, but what you make is not what you have simulated. Why? Sounds like philosophy?

Try diffrent EM simulator(s), 5-6 of them. There are not going to be 2 of them giving you the same result.

Lange published his coupler design in 1969. He didn't have EM simulator available, however it did work well.

flyhgih

 

[Manjunatha_hv]

Hello all,

To understand Shielded/Closed box MoM any lysing circuit with minimal walls effect...one should understand & experiment "Cavity Size Affects the EM Results" concept...
Cavities are vertical metal walls and their effect is included in the simulation.
The presence of a conductive side wall affect the simulation results.
If the cavity is larger than a few times the wavelength of signal, then the cavity’s effects on the simulation results may be minimal.
Cavity side wall effects also depends upon how close the trace metal is to the walls.

Then, Why not just make the cavity very large?
The size of the cavity relative to the wavelength of signals affects the simulation speed....


---manju---

 

[jian]

Hi, Flyhigh: I am interested in your comments on the reality shapes of circuits under microscope. I reallly want to know more about it. As you know, I am a software guy and I always try to make the software closer to reality. For the IE3D I am working on, it can handle precisely conical vias, any shaped wire bonds, thick metal with non-vertical-edges etc. However, I have not heard any oval shape conductors. I would like to learn something on it so that we can make a better model for it. Thanks!

 

[flyhigh]

Hi jian,

this is not the best picture but think is illustrative enough. It is a soft board, 17um metalization, further increased by 20um copper growing during via plating process, than fleshed with 2um gold. The line/slot pitch should be 100um/100um.

The vertical shape is semi-oval, should be clear from the shining effect. The edge is clearly non-uniform. I had pictures of vias and oval shaped embedded LTCC but they are hard to find, it long time ago.

 

[rautio]

Hi Manju -- Actually, making the box large can allow box resonances. If present, the box resonances show up as "suck-outs" in the data. If you analyze the same circuit in an unshielded analysis, the suck-outs change into radiation (the data is now nice and smooth, but the radiation can still cause big problems). If you are going to operate the circuit unshielded and there is radiation, you could easily couple to nearby components. I often recommend that a circuit should be analyzed in both shielded and unshielded analyses. Differences between the two can indicate problems. In this case, the suck-outs in the shielded analysis indicate that the unshielded circuit is radiating. (This problem can be solved by making the substrate thinner, among other things.)

Also, the box wall can be used to indicate how close you can place another component without strong coupling. On most designs, as suggested by Max above, you can get within one or two substrate thicknesses before there is a problem.

A lot of discussion above is about error sources in EM analysis. This has been a major area of my research, I have multiple papers dealing primarily with error analysis. Very important: you do NOT have to make measurements to quantify most error sources. Just do a convergence analysis. For example, what is the effect of box size in a shielded analysis? Do an analysis with one box size, then do a second with another box size.

What is the error due to cell size? Do an analysis with one cell size, then cut the cell size in half and do a second analysis. On a well designed EM analysis, most of the time the error for the finer cell size is about equal to the difference between the two analyses.

What is the error due to de-embedding? Analyze a device (like spiral inductor) using connecting transmission lines of one length, then analyze a second time using connecting transmission lines of double or half the length and with the reference planes in the same place with respect to the inductor. The difference gives an indication of the magnitude of the de-embedding error.

What is the error due to numerical precision (the noise floor)? Analyze a tightly coupled line with reference planes from each end meeting exactly in the middle. This is the "zero length coupled line". The correct answer for all coupled and reflected S-parameters is -infinity dB. The analysis result will be the noise floor of the analysis. For shielded analyses, this is typically 100 to 180 dB down.

These are all nice simple numerical experiments that can reveal a lot of information. Give them a try!