IE3D Benchmark of Matrix Solvers

SMSi: Default high performance advanced symmetrical matrix solver for IE3D.
AIMS II: Fast and robust advanced iterative matrix solver implemented on IE3D.
AIMS III (1): Memory efficient and robust advanced iterative matrix solver on IE3D.
AIMS III (2): Same as AIMS III (1) except it has pre-conditioning enabled.
Notes: The simulations are performed on a Pentium III 450 MHz machine with 512 MB RAM. The simulation time is minutes per frequency point. The "*"s indicate estimated values.

Example 1: An 8 by 2 patch antenna array: Total 2644 cells and 3828 unknowns created:

RAM (MB) HD Space (MB) Time (minute) Primary Matrix
SMSi 118 0 27.8 50%
AIMS II 59 20 11.5 6.3%
AIMS III (1) 25 10 27.0 3.2%
AIMS III (2) 31 12 12.0 3.2%


Example 2: An 8 by 4 patch antenna array: Total 5498 cells and 7866 unknowns created:

RAM (MB) HD Space (MB) Time (minute) Primary Matrix
SMSi 490 0 241 50%
AIMS II 210 40 82 3.2%
AIMS III (1) 51 20 122 1.6%
AIMS III (2) 62 25 51 1.6%


Example 3: An 8 by 8 patch antenna array: Total 11093 cells and 15829 unknowns:

RAM (MB) HD Space (MB) Time (minute) Primary Matrix
SMSi 2005* 0 1965* 50%
AIMS II 730* 96* 502* 1.6%
AIMS III (1) 100 41 424 0.8%
AIMS III (2) 122 48 216 0.8%


IE3D Accurate and Efficient Simulation of Large Structures

Shown in the following picture is an 8 by 8 patch antenna array with feed network modeled on the IE3D 8.0 pre-release version, using the AIMS III advanced iterative matrix solver. The simulation includes all the coupling between elements and the feed network and does not neglect anything. It takes about 424 minutes per frequency point on a Pentium III 450 MHz machine using about 100 MB RAM.

COMMENTS:

The figure above shows the solved current distributin of the 8 by 8 patch array including feed network. For the iterative matrix solvers implemented into the IE3D 8.0, all the coupling between the elements and the feed network is included in the simulation. No approximation is used in including the far coupling and the final solution is a precise one. The iterative matrix solvers require users to define the separation distance which has a default value of 10 cells. The AIMS solvers save much simulation time and RAM. The pre-conditioning in AIMS III (2) causes some more RAM than the regular AIMS III (1). However, it saves about 50% of the simulation time.