Do's and Don'ts for EM Simulation Success

Dos

1. Make sure the answer is reasonable. Remember, the trouble with EM simulators is that they almost always give an answer (assuming you don't suffer a power outage, lack of RAM, or similar fate). It is up to you to decide if this answer makes sense. For example, if your interconnect is essentially a through connection, shouldn't you be concerned if the reflection coefficient is almost 1? Pay special attention to notorious trouble areas for simulators. For example, resonances should be reviewed carefully; are they real or merely unwanted by-products of the simulation method (for example, a box resonance). Also, non-static simulators typically have troubles at very low frequencies where the structure size is electrically small (for example, less than 1/100 of a wavelength).
2. Understand your simulator. This doesn't mean you have to run out and get a Ph.D. in electromagnetics. But, take the time to learn the basics. Most software companies have excellent training materials on their products. For example, understand what the operating assumptions are. Examples include: how does it work with ports and deembedding? How does it handle lossy materials? What does it mean by error? How does it mesh? I find it a useful exercise to try "breaking" the simulator by stress testing it on difficult examples. For example, if the simulator includes thick metals, try running examples with closely coupled lines. All simulators have limitations. The more you understand them, the more successful you'll be.
3. Start with simple cases of your problem. For example, you are trying to model transitions in a low temperature co-fired ceramic (LTCC) package technology. Start with a simple line. Then progress to one via transition. Then add loss and try again. In a short while, you can build up to a realistic problem, and you will have a good understanding of how the simulator is performing. If you start with a complicated problem, you may never know if your answer makes sense.
4. Know how good an answer you need. There will be error and uncertainty with the results from any simulator. How accurate an answer is required? Remember, your circuit will be built with variations in the manufacturing. It doesn't make much sense to ask the simulator to give you a load impedance to five digits, when the board manufacturer is saying the line impedance can vary ten percent. Perhaps your simulator works on a fixed grid. By changing the line width one mil, you might reduce the simulation time significantly, without affecting the results in any meaningful way. Figure 1 shows another helpful hint in setting up your simulation parameters.

Don'ts

1. Don't start a project by blindly inserting a mechanical drawing for your structure. This usually adds unnecessary detail, which can bring the simulator to its knees. Does it really matter if the corners of your package have the perfect, rounded, shape. Do you really need to include the thickness of the line? Simulators usually hate large aspect ratios in feature sizes. A 10 mil wide line with 2 micron thickness (125:1 ratio) is hard for any simulator to swallow. It requires very fine meshing for the line thickness, and a lot of elements for the wide line. Normally, all those pretty mechanical details are the enemy of efficient EM simulation.
2. Don't rely on automatic extraction methods without checking the geometry. In an extraction, the circuit's layout is automatically sent to the EM simulator. This is, of course, a powerful time-saving concept if properly used. But, always check the layout of the circuit that was simulated to make sure it's what you expected. There is usually a lot of configuration required to get the correct extraction. I've seen extractions where the lines end up on the wrong layers, bounding boxes the wrong size, and wrong materials, to name a few potential pitfalls.
3. Don't try to "hit a home run." This is what is called in military jargon: "a hero test." The engineer throws the complete, complicated structure in the simulator, fires it off and awaits (for a long time) an answer. The problems with this approach have already been mentioned: you won't have any idea if the result is correct; and if you do get an unexpected answer, what do you change in the circuit to correct the problem? Start small...understand... and work upward.
4. EM optimization is often not the answer. It is becoming popular to try to improve a design by placing an EM simulator in an optimization loop. The potential problems with this approach are many. First, there is the obvious problem that it could take a LOT of simulation time. Second, make sure you aren't trying to "design by optimization". This seldom leads to quality designs. I've actually had engineers say to me, "I don't have time to understand it. I'll just run a simulator for a week and build it!" I am an advocate of creating an electrical model of the structure. Then, the parameters in the model can be optimized, and verified with EM simulations. Of course, this approach requires the designer to take the time to create a good, physical based model.
5. Don't make EM simulation a substitute for measurement and experimentation. Rather, EM simulation is designed to complement measurements on a real prototype. There are always assumptions made in any EM simulation. Perhaps you left out something important, you didn't even realize. For example, material roughness, or a solder mask layer could matter.

Conclusion

EM simulation is a wonderful tool. Like any technology, it can be used and misused. There is no substitute for understanding your problem, your simulator, and knowing your goals. Happy simulating!

About the Author

Dr. John Dunn is a senior engineering consultant at Applied Wave Research. Before joining AWR, he was a professor of electrical engineering at the University of Colorado for 15 years. He received his Ph.D. in applied physics from Harvard University in 1984 and is a senior member of IEEE. He can be reached at: jdunn@appwave.com