SI & EMI Control Design Tools....my thoughts..TFox


Power Delivery System Impedance, Ground Bounce, Simultaneous Switched Output Noise, Plane Noise

Developing a proper power delivery system is critical. The power delivery system impedance must present a suitably low value (generally around 0.1 ohm) from DC to the highest frequency of interest. For a 2ns rise time signal the highest frequency of interest could be over 1000Mhz depending upon the on-die capacitance. Poor power system performance in the lower frequencies can result in functional data / clock errors and Common Mode EMI. Poor Power system performance in the higher frequencies can result in EMI problems that are nearly impossible to correct. To calculate the Power System Impedance you must take into account everything including the DC source and its feed lines, the by-pass capacitors and their connection traces/via's as well as the effect of the power/ground planes, and how far capacitors are from the device to be bypassed. In the corner case of very closely spaced power ground planes, the actual location of the capacitors is not terribly critical. In the case of typical 4 or 6 layer boards where the power ground planes are fairly wide, the location of the capacitors is absolutely critical to a proper analysis.

If you are attempting to minimize layer count, capacitor count, or have oddly shaped split planes. I highly recommend simulating the problem using a sophisticated tool similar to HyperLynx Power Integrity from Mentor www.hyperlynx.com . There are other tools in this space, but this is the one I am familiar with and it makes some very complex interactions visible.

The "free" tool which is part of the Pro Tune Up class has limitations which are specified in the documentation page. It is not a "solve the mysteries of the universe" type of tool. It was written to help my students put some consistent rational discipline into power design. The ground bounce part of the tool will help alert you to the terrible price you will pay if there is inadequate power delivery pins in the package. If you are using very closely spaced power and ground planes, using best capacitor mounting practices and locations, the "free" tool will keep you in the ball park. If you have the board space, there is no cost having too many capacitors. If you want to reduce the cost, don't load the capacitor in manufacturing.

Free PS |Z| & Ground Bounce Calculator --> . PS|Z| & Ground Bounce Calculator DOWNLOAD


Ringing and Cross Talk

For board level work, this is the domain of the IBIS simulator. IBIS is like a spice simulator, but work at a more abstract level. Hence the typical IBIS simulator runs at about ten times the speed as a spice simulator. Spice simulators work well if you know what is going on inside the device package down to the driver level. Most device manufacturers do not want to publish a model of that detail because that would expose the secrets of their output drivers to their competition. Best of all worlds is to have a mixed mode simulation system that runs at IBIS level for speed and reverts to spice level when appropriate. In any case any topology other than a single driver and single receiver requires simulation to get a reliable answer. There are a number of good tools in this space. For my classes I need a tool that not only gets the right answer, but also is very easy to use. I do not have enough time in class to fool around with explaining how the tools work. For the combination of getting the right answer and being easy to use, it is hard to beat Hyper

EMI

Proper SI and PI minimize the energy available to drive the antennae and result in radiation. Signal Integrity and Power Integrity deal with of tenths of a volt. However, 20uV in an off board lead can easily result in FCC Class B failure. With the exception of a straight forward prediction of the radiated emission from a single routed trace, all other simulations of EMI are brutally complex requiring an unfathomable amount of data and precision well beyond the accuracy of the models provided. In other words, a solid simulation solution does not exist today and is unlikely to exist in my life time. The solution here is a fundamentally different approach. If there is energy available, there are certain structures that behave as effective antennae resulting in the EMI we see in testing. Hence, the effective technique is to seek out and eliminate those structures. This is where EMI engineers tend to go both blind and crazy looking at boards and trying to spot these problem structures.

We should always do our best to avoid generating the energy through rigorous SI and PI simulation techniques. We must also avoid creating accommodating antennae by understanding them and never creating them in the first place. However, we are still faced with the job of detecting and eliminating structures that may have been created by accident or by ignorance.

Examples are:

High frequency signals crossing a split in the reference plane.

Routing signals too close to the edge of a reference plane.

Illegal layer transitions which orphan return current.

Placing high speed clocks too close to off board I/O signals, etc.

Rule Checking Tools like IBM EMSAT www.MossBayEDA.com were built to relieve the human eye of the miserably tedious job of finding these problem structures.

Electrical Performance of Arbitrary 3D Structures

Microwave amplifier layout and the impedance / cross talk performance of connectors are both examples of problems that can only be solved using an arbitrary geometry finite element 3D field solver. These simulators are roughly 10 times slower than spice, but are absolutely necessary unless you are going to actually build the structure and then measure the performance using a vector network analyzer.

From a system designer's stand point using extremely detailed simulations are not practical either from a data entry or from a computational speed perspective. However, in the case of the connector carrying a set of high speed digital signals, we need to know how the total signal pat will actually perform. Hence, the ideal simulation system should be able to combine the output of an HFSS type simulator or HP VNA into the IBIS level simulation that runs at a reasonable speed for practical design work. HyperLynx has this capability, but I am suspect that other simulators also have that ability. My only point is that this type of capability will become more important as we continue to use faster technologies.

Electrical Performance of Very High Frequency Routing Structures

If you are doing over 5 GB/s layout, I highly recommend you investigate the Simbeor at www.Simberian.com. Dr. Yuriy Schlepnev has some excellent tools and a wonderful set of application notes. The output can be used as a liberary element in a HyperLynx IBIS mixed mode simulator.