Comments on hot news about HFSS V11
-------------------------------------------------------------------------
HFSS? v11 offers even greater accuracy, capacity and performance. HFSS is widely used for designing RF/microwave components, onchip passives, PCB interconnects, antennas and IC packages. New higher-order, hierarchical basis combined with an iterative solver provides accurate fields, smaller meshes and efficient solutions for large multi-wavelength structures. Complex geometries solve 2 to 5× faster using half the memory with the new v11 fault tolerant, high quality finite element mesher. Continued user interface refinement and data linking enables co-design of complex electronic systems.
-------------------------------------------------------------------------
I hope its iterative solver is going to be even much faster than the one in CST MWS 2006B tetrahedral solver. As you might know, CST MWS 2006B iterative solver for its tetrahedral FD solver is very fast so that most of the time it doesn't make sense working with HFSS direct solver. Although not quite sure, it kinda looks like they have not implemented their domain decomposition method in HFSS V11.
I hope it will be officially released very soon.
can't wait to take a spin with the 11th version. it was time hfss to get faster...
"Complex geometries solve 2 to 5× faster using half the memory" -> this is very good to hear and easy to say...
The problem is that 2 to 5× increase in speed is nothing compared to that of CST MWS 2006B tetrahedral iterative solver. As an example it takes CST MWS about 28 minutes to simulate an antenna with 650000 tetrahedron mesh (for each port and frequency) on an Intel Woodcrest workstation with a lot of memory. For HFSS 10.1.*, it takes for ever so that after a few hours, I just have to stop it.
It looks like it has been officially released:
h**p://www.ansoft.com/news/press_release/070605.cfm
h**p://www.ansoft.com/products/hf/hfss/
Looking at their press/demo pictures one has to note a simulation of a complete car in HFSS. Really impressive and nice going. Looks like v11 is a major step forward. CST be ware ;)
hi,
ok, a complete car but on what hardware? There's no pc/memory specs given are there? CST has been able to do these things for years on standard pcs in the time domain. Seems as though Ansoft don't think much about going down the Time domain route like most MW/RF vendors....
aw
I hear a lot of discussion about speed.
What about the accuracy? Speed means nothing if the results are insufficiently accurate to be good for design. I don't get paid to analyze big problems; I get paid to design stuff that works. A benchmark to measurement or some kind of known benchmark for these large problems would impress me more.
--Max
The accuracy depends on the user skills of the software. HFSS is well known to provide excellent results that really correspond to measurements, but, again, output results depend on the user...
rfmw:
I understand that. But we all know that as EM vendors try to solve larger and larger problems, they are sometimes cutting corners by using approximations. Using things like banded or sparse matrix techniques, or iterative matrix techniques introduce error to the simulation at a higher rate. That is true regardless of the skill with which you use the software. Even skilled users need to know what the limitations are in the software, and some techniques for faster solution or for solving larger problems introduce different or elevated error sources.
Yes, user skill is very important in using these tools properly. But skilled users don't trust new versions or so-called improvements implicitly. They have usually learned through painful experience that new versions often mean locating and avoiding new bugs and deficiencies. You have to know the limitation of your tool in order to use it well.
--Max
Well said and I agree with you 100%. Address your accuracy and error accumulation to Ansoft. I'm also interested in their answer in this regard.
I've been evaluating V11beta for 10 days now and this release is so buggy that 10+ hour simulation results got erased due to the nasty bug. Beside, I see no big difference in the memory consumption when compared to v10. My design is planar antenna on a dielectric volume (electrically big).
Use Agilent EMDS
ADS_RFMW: Are you a distributor for Agilent or something? You have posted this comment on three threads at the same time.
I am not aware that there is anything technically new in the Agilent EMDS analysis engine since Agilent sold their customer base to Ansoft and retired the old Agilent HFSS. From what I have seen, Agilent hasn't added anything new to the engine, and Agilent was behind Ansoft in technology even back then. Now they are problably 7 years behind Ansoft in analysis technology. So as I see it, EMDS has 2 advantages over Ansoft HFSS:
1. It is integrated with ADS (which is nice if it really works
2. It's cheaper.
It is strange to see Agilent compete on the basis of price. That has never really been their niche. But competition is always good, and hopefully it will drive Ansoft prices down, even if the engine and the interface aren't nearly as good. (But don't get me wrong, I don't think Ansoft has everything together, either.)
--Max
hi,
I don't work for any vendor but find this quite interesting. It just seems crazy to me to rely on FE methods for trying to perform large simulations with exotic higher order techniques or cluster. The memory scaling for FE means that it is relying on what could be seen as an archaic method for electrically large structues. Why stubbornly work on a method like this when other vendors are taking the Time Domain path (other vendors have performed large structure simulatoins besides cars such as large antenna systems, for example)? Of course, FE and HFSS etc can be used for certain types of problem but it should be supported by other solver types. There are, to my knowledge, two vendors, possibly three that offer muli-solver technology. My point here is that, although higher order FE may bring some gain, it really cannot take on almost-linearly memory scaling solvers such time-domain ones, especially considering that any mesh would probably only satisfy the basic requirements for a simlation without even talking about mesh refinement.
aw
I agree completely. And I would even go so far as to say that you need more than one vendor. I don't think any one vendor has the best 3D frequency domain solver, the best 3D time domain solver, etc. all under one package.
I wish the commercial EM vendors would seriously consider better interoperability. That way, we could port simulation models back and force easily and get more (and better) work done. But so far, this hasn't been in their best business interest. They would rather lock us into their own proprietary frameworks. Hopefully a universal interface or file exchange format will be developed and embraced by major EM software vendors.
--Max
Hi Max,
in fact I am using software from just one vendor that can do all of these which means that I am not personally concerned about interoperability with other vendors. It's quite effective since I simply select a different solver for the same model. It's quite unique in this field.
aw
aw:
Interesting. So what are your applications? I can see perhaps some narrow applications where you don't care about whether or not you can interoperate with other CAD tools (often times, antenna design doesn't require a lot of interaction with other software tools). However, if you are designing RFIC, MMICs, packages and antennas, I would be willing to bet that design tool interoperability is very important. But maybe not if you don't care about other parts of the system--just focus on your small part of the design?
It's not just about EM solvers, or just about circuit simulators. There are a wide range of other simulators and design tools required for successful system designs.
So what software tool do you use, aw?
--Max