Benchmarking
Evaluation of Electromagnetic Software
Revised: June 4, 1997
© Copyright 1994, 1999 Sonnet Software, Inc. All Rights Reserved
Table of Contents
Chapter 1 - Introduction
Chapter 2 - Technical Overview
Chapter 3 - The Stripline Standard
Chapter 4 - The Stripline Standard and Triangular Subsections
Chapter 5 - The Microstrip Standard
Chapter 6 - The Coupled Microstrip Standard
Chapter 7 - Limit Tests
Chapter 8 - Lossy Ground Plane and Lossy Dielectric Tests
Chapter 2 - Technical Overview
Stripline Standard | Triangle Subsections | Microstrip Standard | Coupled Microstrip Standard |
Ultra-thin Dielectric | Zero Length Through | Short Length Vias | Loss Tests
This chapter provides a brief technical description of the entire suite of benchmarks. Benchmarks which are of particular interest can then be reviewed in detail in the referenced chapter.
Each benchmark is designed to detect and quantify specific error sources as described below. All test structures have very simple geometry and allow precise quantitative evaluation of error. (Complex geometries, which do not allow precise evaluation of error, are appropriate for the more common GABMAC validation as described in the previous chapter.) Given familiarity with the software being tested, it should be possible to complete the entire suite of benchmarks in one to two days.
1) The Stripline Standard (Chapter 3, "The Stripline Standard"). This standard consists of a simple through line. Since there is an exact solution for stripline, the width of the line can be set so that an exact 50 Ohm line results. The length of the line is set to exactly a quarter wavelength at 15 GHz. Total error is determined by simply adding the magnitude of S11 to the percent phase difference of S21 from -90 degrees. Use different subsection sizes to determine how error changes with subsection size. We recommend starting with a very short subsection length (512 per wavelength) and varying the number of subsections into which the width of the line is divided. Start with one subsection per line width, then two, then four, etc. Continue until analysis time becomes excessive. As subsection size becomes smaller, error should decrease and analysis time should increase. Plot or tabulate the error versus analysis time performance.
2) Triangle Subsections (Chapter 4, "The Stripline Standard And Triangular Subsections"). Some analyses use triangular subsections to allow representation of smooth curves. If improperly implemented, undesired error can result. To detect and quantify this error, repeat the above Standard Stripline, only now subsectioned with triangles, see Figure 6 in Chapter 2. Perform the analysis for the line subsectioned only one (triangle) subsection wide. Determine the amount of error due to triangles by comparing the result with the Stripline Standard subsectioned one (rectangular) subsection wide as performed above. As explained in the chapter, this test is not appropriate for Sonnet.
3) The Microstrip Standard (Chapter 5, "The Microstrip Standard"). There is no exact solution for microstrip, thus an approach similar to the Stripline Standard is not possible. Instead, the microstrip standard is a frequency independent lumped component, a series capacitor, embedded in a microstrip line. The dimensions and frequency of analysis are set so that the effect of fringing discontinuities (series inductance) are very small relative to dispersion in the connecting transmission line. To test an analysis for accuracy, simply analyze the capacitor as a function of frequency. The analysis should have reference planes set to the middle of the capacitor with results de-embedded to that point. The capacitance is independent of frequency. Any calculated variation of capacitance with frequency is error in the analysis of dispersion or error in the de-embedding. Since de-embedding is usually an integral part of an electromagnetic analysis, de-embedding error is appropriate to include in a quantitative benchmark. If the capacitance is not a standard output of the analysis under test, use the linearity of Y22 as an equivalent metric. Unlike the previous tests, this test is not intended to quantify error due to subsection size. Since some electromagnetic analyses have difficulty analyzing capacitors, the dielectric on this capacitor has been kept thick so that such analyses are not excluded from this benchmark.
4) The Coupled Microstrip Standard (Chapter 6, "The Coupled Microstrip Standard"). Because de-embedding works for single lines, does not mean it also works for coupled lines. This benchmark is a simple coupled line version of the Microstrip standard. Error in de-embedding coupled lines or in coupled line dispersion can be quantified using this standard. Some analyses are incapable of de-embedding coupled lines. This standard can not be used on such analyses.
5) Limit Tests - see the section "Ultra-thin Dielectric" in Chapter 7. The Green's Function (i.e., fields due to an "impulse function" of current) are very singular near the source of current. If two subsections are very close to one another and an analysis relies on numerical integration, very large error can result. This is especially true in thin dielectric capacitors. This test is similar to the Microstrip Standard except that the capacitor dielectric is now very thin, with dimensions and parameters similar to Metal-Insulator-Metal (MIM) capacitors common on GaAs integrated circuits. This test simply looks at the low frequency value of the capacitance and checks to make sure it is close to the DC parallel plate capacitance. When an analysis fails this test, it usually does so dramatically.
6) Limit Tests -- see the section "Zero Length Through" in Chapter 7. This test is a simple microstrip through line de-embedded to zero length. Any error left over is usually very small numerical noise due to finite precision. Any larger errors should be investigated carefully.
7) Limit Tests -- see the section "Short Length Vias" in Chapter 7. Most analyses have a low frequency limit where the small difference between two large numbers generates large errors and an analysis becomes unusable. For example, if there is any need to deal with integrated circuit dimensions at frequencies below 2 GHz, this test is absolutely critical. The test consists of an ultra short via. With the frequencies and dimensions suggested, if an analysis fails the test, it is likely to do so dramatically.
8) Loss Tests -- see the section "Lossy Ground Plane And Lossy Dielectric Tests" in Chapter 8. In many designs, accurate calculation of ground plane loss and dielectric loss can be important. These tests demonstrate precisely how much error there is in the calculation of these quantities.
Please feel free to investigate in detail any of the above tests that you find interesting. Results of these benchmarks as applied to Sonnet are included in this report. As they become available, we will provide results of these benchmarks applied to other electromagnetic software. Contact Sonnet for current information.
It is our hope that this document will initiate the transformation of the field
of electromagnetic software validation from a subjective hand-waving contest to one where
precise, quantitative evaluation of error is held in high regard.
Suggestions of additional benchmarks which can further this goal are welcome.