Difference between 2.5D and 3D EM simulators taking EMSight and Axiem as an example

1) what is the difference between 2.5D and 3D planar simulators? Are these two same or different?

Practical MMIC Design by Marsh gives the definition like this

2.5D planar electromagnetic simulators such as Momentum are primarily the same as 2D simulators except the extra half-dimension means that the solver can calculate currents in the vertical (z) direction (usually for through-substrate vias) as well as the x and y directions.

3D planar electromagnetic simulators are similar, but they are also able to calculate the effects of conductor thickness and resistivity. Examples of these include Sonnet Lite and EM3DS.

One more definition is given in this way

Refer Link - **broken link removed**

One more definition is

2.5D implies that RF currents are allowed in two directions only (X and Y), and fields are calculated in all three dimensions (X Y and Z). No vertical currents are allowed. Examples would include micro strip or strip line filters which do not contain vias to ground. Note: This term has also been used to refer to 3D planar solvers such as Sonnet (Sonnet Software product) and Momentum (Agilent product) even though they all allow vertical currents.


3D Planar implies that currents and fields are allowed in all 3 directions, but circuits are restricted to stratified dielectric media. Examples would include most MMIC, RFIC, and PCB circuits. Examples include Sonnet (Sonnet Software product) and Momentum (Agilent product).

Because, I am also able to see 3D view in either of them (currently using EMSight and Axiem Simulators). What is the difference?


2) One more question on specifically on EMSight (2.5D) and Axiem (3D Planar)

I know that EMSight is closed and Axiem is open. What does this mean? Is it got to do something with the boundary conditions, Also which is more accurate? Also, FYI - Axiem is non-gridded (use mesh) and EMSight is gridded. Big deal, both of them simulate well and 3d em layout view is available in both of them. kindly clarify.


Regards,
Avs

Sonnet prefers the "planar 3D" description to indicate that Method of Moments calculates the full 3D fields and is not approximate.

When people talk about "2.5D" or "planar 3D" they mean the same thing. All these tools use "Method of Moments" simulation and they all are designed for planar layered structures with possibly many layers and vias, but can not do arbitrary 3D structures.

Open boundary (Axiem, Momentum) vs. closed box (EMsight, Sonnet) are two possibilities to do Method of Moments analysis. The closed box approach allows to calculate the coupling functions in closed form, without using numerical integration, so it is numerically very robust with the highest possible dynamic range. The open boundary approach with arbitrary subsections can usually mesh a problem with less unknowns, but requires multi-dimensional numerical integration to calculate the coupling functions, so it is more sensitive to numerical precision issues.

volker_muehlhaus said:

"The closed box approach allows to calculate the coupling functions in closed form"

Click to expand...

This statement is not accurate. What is availabe in "closed form" is the spectral domain Green's function--this isn't the "coupling functions" or mutual impedances. The muti-dimensional integration in the open boundary codes is mapped into summation of infinite series in the shielded codes. So in the "closed box approach", the "coupling functions" is not exactly in closed form--it is an multi-dimension infinite summation. There is no standard on the convergence of this series--you can't say that by summing to a certain number of terms, the result would be accurate to within the precision of the floating point calculation.

volker_muehlhaus said:

Sonnet prefers the "planar 3D" description to indicate that Method of Moments calculates the full 3D fields and is not approximate.

When people talk about "2.5D" or "planar 3D" they mean the same thing. All these tools use "Method of Moments" simulation and they all are designed for planar layered structures with possibly many layers and vias, but can not do arbitrary 3D structures.

Click to expand...

I'm a bit puzzled by this too. If it can do an arbitrary number of layers, can that not be used to represent an arbitary 3D shape? Let's assume we start with a PCB 30 mm x 30 mm which is double sided, 1.6 mm tall (i.e. W x L x H = 30 x 30 x 1.6). Now would Sonnet break down if there were 1000 layers of the same thickness, making this have dimenstions of W x L x H = 30 x 30 x 1600?

I've often wondered how what advantages HFSS's Integral Equation solver option (HFSS-IE) over NEC based software for a Yagi. Both are based on MoM. I believe one of the limitions of NEC is that it assumes that currents flow along the length of metal, so it can't consider the boom of a Yagi where the current flow across the boom, not along the length of it. I'm not sure if HFSS's HFSS-IE solver is any better or worst in this case and I never got any definitave answer when I asked an HFSS support engineer this question.

Dave

drkirkby said:

I'm a bit puzzled by this too. If it can do an arbitrary number of layers, can that not be used to represent an arbitary 3D shape?

Click to expand...

Dave, you could do this brute force staircasing of arbitrary geometries, as a last resort, but the editor is not designed to support such drawing method. There are other tools which can handle arbitrary 3D much better.

drkirkby said:

I've often wondered how what advantages HFSS's Integral Equation solver option (HFSS-IE) over NEC based software for a Yagi. Both are based on MoM. I believe one of the limitions of NEC is that it assumes that currents flow along the length of metal, so it can't consider the boom of a Yagi where the current flow across the boom, not along the length of it.

Click to expand...

From what I heard about NEC, it builds a model from wires only, and does not support surfaces (e.g. rectangle for patch antenna). For a yagi, a wire based model is appropriate.

This is my understanding of avins's post.

The major difference between 2.5D and 3D is actually the modeling capability, not the EM-MoM simulation code, I am sure, the EM code for 2.5D and 3D are the same, since both need a green's function and solve for currents within or on the surface of the body. If this can be done, it does not matter if the current flows only along x,y directions or x,y,z directions. Same MoM code for them.

Big thing, modeling arbitrary 3D is non-trivial, company has to pay a lot of money to get a 3D modeling engine, imagine 3D lofting, sweeping, union, subtract, intersect. This software coding is even larger than the entire EM-simulation community. However, 2.5D is way simpler, it restrict you to only build layered structure with via, which dramatically narrowed down modeler's ability and can ignore 80% of the full 3D modeling code. I believe this is the secret.

rodgerwxh said:

This is my understanding of avins's post.

The major difference between 2.5D and 3D is actually the modeling capability, not the EM-MoM simulation code, I am sure, the EM code for 2.5D and 3D are the same, since both need a green's function and solve for currents within or on the surface of the body. If this can be done, it does not matter if the current flows only along x,y directions or x,y,z directions. Same MoM code for them.

Big thing, modeling arbitrary 3D is non-trivial, company has to pay a lot of money to get a 3D modeling engine, imagine 3D lofting, sweeping, union, subtract, intersect. This software coding is even larger than the entire EM-simulation community. However, 2.5D is way simpler, it restrict you to only build layered structure with via, which dramatically narrowed down modeler's ability and can ignore 80% of the full 3D modeling code. I believe this is the secret.

Click to expand...

I'm not convinced it is the interface that is the problem, as there are low cost CAD packages that do 3D drawings. Just Googling I found an open source CAD package

https://sourceforge.net/apps/mediawiki/free-cad/index.php?title=Main_Page

which supports Boolean operations. So whilst I'm not saying writing a 3D interface is a walk in the park, I'm not convinced that should be a huge step for expensive commercial software.

I believe in NEC, which is only limited to wires, an assumption is made that the current flows along the direction of the wire. So NEC is fine for modelling Yagi-Uda antennas if you assume the elements are isolated. However, it not possible to model the effect of a round boom on a Yagi-Uda antenna, as the current in the boom will go at right angles to the direction of the boom.

I don't know if that issue with NEC's MoM is inherent to the MoM method, or is just a limitation of NEC.

Dave

rodgerwxh said:

I am sure, the EM code for 2.5D and 3D are the same, since both need a green's function and solve for currents within or on the surface of the body.

Click to expand...

I don't think that's true. Even in planar 3D MoM, the method and basis functions can be quite different (open boundary vs. closed boundary, Sonnet/emSight vs. Momentum/Axiem vs. IE3D).

As I mentioned above, MoM itself does not care about 2.5D or 3D, its basic principle is the field at a arbitrary point is a superposition of all fields radiated by the source, where the source is J and M (depends on you are using EFIE, MFIE, or CFIE) over the entire structure. the radiation relation between J and M and E and H is governed by green's function, which is the pulse response of your entire system, Once this procedure is implemented, it is suitable for all 3D cases, you solved your current (surface or volume). You get your field anywhere.

However, to solve your currents is a different story, you have to rely on your basis functions, singular quadrature and apply BCs. to fill your matrix. This is highly depending on structures, and a lot of specialized techniques can be applied to simplify a lot of unnecessary operations, 2.5D is one of them. NEC's wire restriction is another, because the integration of currents can be simplified to a surface or a line only rather than generic 3D volume. Of course, building a surface basis like RWG and even a line basis is much simpler as well.

NEC is a very old code (more than 40 years old) with several limitations, since it was developed for the primitive computers available at that time. Many softwares have been developed during the subsequent years which use the old NEC engine. For this reason, NEC-based softwares (like 4NEC2, Eznec, MMana, etc.) will never have the option to simulate planar antennas. So, 2.5D and planar 3D problems are impossible to solve using NEC. So, I don't know why you are mentioning NEC in this thread.

Here a brief explanation about 2, 2.5, and 3D field solver:

Field solvers can be roughly grouped into three geometrical classes of field-solver code. The first takes the 2D cross-section and solves for the transverse field distributions. These tools are useful for calculating trace impedances and coupling coefficients in microstrip or stripline board routes, for example, or any geometry with a uniform cross-section in the longitudinal direction. They solve rapidly because only the 2D cross section needs to be discretized. .
The second class of code meshes the surface of planar metals. Although this is still a 2D problem, if vias are introduced to establish connection vertically between metal plane layers, then the code is sometimes called 2.5D. The via can then be handled by look-up tables or can be modeled as a lumped element, for example. These tools allow for an arbitrary number of homogenous dielectric layers with patterned planar metal on the conductive layers. In this example, the solve time was reduced by orders of magnitude, using the 2.5D solver without sacrificing accuracy. Thicker dielectrics may not be well suited to a 2.5D solver due to field non uniformity in the Z direction. For most PCB applications, this limitation does not pose a problem.
The final geometrical class meshes a 3D volume. This class of tools is useful for solving for vias transitions, discontinuities in multilayer PCBs, capacitor mounting structures, and so forth. The solution time is much longer compared to 2D and 2.5D solvers, but the 3D solver has the benefit that it can be used to analyze a broad range of problems.

Source "Frequency-Domain Characterization of Power Distribution Networks"

carlos.lopez said:

The second class of code meshes the surface of planar metals. Although this is still a 2D problem, if vias are introduced to establish connection vertically between metal plane layers, then the code is sometimes called 2.5D. The via can then be handled by look-up tables or can be modeled as a lumped element, for example. These tools allow for an arbitrary number of homogenous dielectric layers with patterned planar metal on the conductive layers. In this example, the solve time was reduced by orders of magnitude, using the 2.5D solver without sacrificing accuracy.

Click to expand...

Not quite exact. This simplistic approach does not apply to the widely used planar MoM solvers like Momentum, Sonnet, Axiem, emsight, IE3D etc. They solve the vias by placing vertical currents on the vias, and include these in the caclualtion of the 3D EM problem.

carlos.lopez said:

Thicker dielectrics may not be well suited to a 2.5D solver due to field non uniformity in the Z direction.

Click to expand...

Also not quite exact: you just need to partition the via into multiple segments if the via length is >1/20 wavelength or so.

In **broken link removed** you have a 3D microstrip antenna simulation software. Vias to ground can be added.

I want to continue in this post after the above discussions.

Besides all we talked about 2D, 2.5D and 3D solvers, here is the question that really matters.

For circuit board problems, when should we think 2D or 2.5D is sufficient since its way faster? When should we think that 3D solver has to be used to capture certain effects.
I know frequency is definitely a major factor, for circuit board problem, most chips are MHz range (correct me if I am wrong). in such frequency range, the board size is much much smaller than the wavelength (1 meter at 300MHz), thus this is really a E-small problem and quasi-static case. E and H are almost decoupled which means even circuit theory can solve it well without losing big accuracy. However, I am curious why engineers sometimes not satisfying the 2D 2.5D or circuit solvers for such kind of problem? Unless circuit goes to GHz, otherwise non-3D solvers should be good enough, am I right?

Second thing I can think of is non-linear components, since non-linear devices generate high-order harmonics, namely 500MHz generate 1G, 1.5G and 2G signals. This sometimes is bad since signal lines are too close and high-order modes breaks the 2D 2.5D range, since the high-order mode of the field is much more challenging to solve, this is where 3D comes in as a niche. Am I right?

Anyway, from the end user's point of view, I think we do not care much on the nature of 2D, 2.5D and 3D solvers, we care more on in what circumstances should we choose efficient 2D 2.5D solvers and in what kind of situation, there is no simple way but to use 3D. This is something more meaningful in my opinion.

Rodger, having worked with these tools for many years, I don't agree with your summary and conclusions.

rodgerwxh said:

I know frequency is definitely a major factor, for circuit board problem, most chips are MHz range (correct me if I am wrong).

Click to expand...

Today, there is a wide range of applications where we have true GHz signals on PCB. This can be fast digital signals or RF signals, both is quite common.

rodgerwxh said:

in such frequency range, the board size is much much smaller than the wavelength (1 meter at 300MHz), thus this is really a E-small problem and quasi-static case. E and H are almost decoupled which means even circuit theory can solve it well without losing big accuracy.

Click to expand...

Indeed, the PCB software companies have their quasi-static fast PEEC solvers, which are based on these assumptions, and can handle these cases (low frequency, quasi static, high layout complexity). But this is almost a different world of EM tools than the RF EM tools discussed above: different software companies, different methods, different applications and different way of working (low frequency PCB people don't like to deal with S-parameters ....).

Thanks volker for your comment:

Its getting clear to me, so when and why we need these full-wave tools (all Ds) for these circuit board problems? Is it a waste of resources? Or there is a critical reason that full-wave tools have to be applied in these problems. If there is, is it b/c of higher and higher working frequency? or Larger integration scale which brings multiple couplings and scatterings? or non-linear effects? or something else.

I just want to discuss with experts here what is the vision, challenge and trend of PCB, on-chip board analysis tools in the short future.

rodgerwxh said:

Its getting clear to me, so when and why we need these full-wave tools (all Ds) for these circuit board problems?

Click to expand...

Let's be precise, because the difference matters: quasi-static versus full wave is another topic than 2D (cross section), 2.5D (planar) or 3D (volume meshing).
But I understand what you mean: why would anyone use a 3D simulator like CST or HFSS for PCB design.

As we had discussed a couple of times before, it is not about non-linear effects. If you look at EM modelling concepts, allmost all EM simulation can and should be done linear (excluding something like saturated ferrites etc.). The interface to non-linear circuit simulation is simple and the methjod well established. No need to invent "non-linear EM" in PCB-land.

The reason why some people use full wave RF EM simulators for PCB is much more simple: these and only these methods include radiation effects. If you don't need to account for radiation effects, because the PCB can be considered small and all effects quasi-static, then you should use a quasi-static solver indeed.

Thanks volker, I think I see your point now. I agree with you that non-linear EM for PCB is not a big concern since almost nothing radiate.

But for something like RFIC where radiation, resonance, crosstalk, creeping, surface wave etc. have to be considered, it will fall within the full-wave tools range.

Rodger, the ratio of size/wavelength in RFIC is usually smaller than in PCB, so that radiation is even less an issue for many RFIC circuits (there are exceptions).