March 2, 1998

SIGNAL INTEGRITY:  EMI simulation tools

Howard Johnson, PhD

You can't model everything. It's too complex a job to tackle.

A number of vendors have recently an-nounced software-based EMI-simulation tools. Many are embellished with flashy demonstrations, which, like the smell of coffee brewing or the sound of bacon frying, promise more than they can possibly deliver.

Let me give you some advice: Real-life EMI problems are much too complex for even the best software tools. As much as I wish the situation were untrue, at this point the best tool is still experience. Many aspects of the EMI problem make prediction a difficult task, especially when working at the system level.

First, the process involves simulation of 3-D wave patterns over a rather large area. That is, your computer will spend a lot of time grinding numbers to get even the most rudimentary results.

Second, every bit of metal in the product matters: every via, trace, pad, bonding wire, connector, and cable. Many times, the system parts you choose to exclude from the model turn out to be the very ones that create the worst EMI headaches. That's the nature of the problem. You seldom know beforehand what parts of a system will turn out to be the worst EMI offenders.

Third, EMI is a strong function of switching speed, data patterns, and precise data timing. That's right, data patterns and timing. If you don't believe me, find out what kind of software people ship to the EMI test range--some companies send several versions to see which works best.

The EMI problem is so complex, it's naive to think you can just type in a few parameters and get meaningful results. For example, how should you model the split-plane zone on a mixed 3.3/5V processor board? Obviously, the board stackup, the shape of the 3.3 and 5V regions, and the trace layout matter. Did you realize that placement and layout of bypass capacitors markedly influence results? How about power-supply noise? You'd have to model it, including phase, at all frequencies from 30 MHz through several GHz. Oops, that would be a function of the software running on the board, wouldn't it? Guess you have to model that, too. Now add little features such as power-supply wiring. It is well-known that enclosing your power-supply wiring in a good metal conduit can reduce radiation from the wiring. How would you like to spend your weekends modeling the contact resistance and inductance of the screws used to hold the conduit in place?

Get the picture? You can't model everything. It's too complex a job. You can't leave anything out, either, because you can never fully know what will matter. So, what can you do? Apply the rules you've always used:

1. Limit your rise times where practicable.

2. Use your experience. If last year's design was x, and today you plan to do 2x, you need another 6 dB of protection.

3. Run an EMI preview scan as early as possible.

The most promising new EMI tools will be more like expert systems than like simulators. They will give you advice and provide reference information, but they won't make rash predictions. If these tools are worth their salt, the first three bits of advice they will provide are listed in 1, 2, and 3 above.

| EDN Access | Feedback | Table of Contents |