I was recently asked to give an exact definition of signal integrity analysis, and to explain when signal integrity analysis should be performed. In other words, what is SI, and how do you know whether or not you need to run SI analysis?

What is Signal Integrity?

We will start with a definition. Signal integrity is typically described as relating to the analog effects of interconnects on pristine digital signals, or how interconnects can screw up the perfect signals from the chips. Another way of describing signal integrity is that it's because of signal integrity's effects that interconnects are not transparent to signals.

I like to group all of the various signal integrity problems into six families:  

1. Signal quality due to reflection noise
2. Crosstalk between line transmission lines
3. Ground bounce noise from return path discontinuities
4. Rise time degradation from losses in the interconnects
5. Voltage collapse on the power delivery paths
6. EMI.

The way to fix these problems is to identify their root causes and design them out from the very beginning. If you do an excellent job of designing these problems out, then you won't see signal integrity problems in your design. If the rise times of signals are all long, such as 2-5 nsec, then the signal integrity effects are typically pretty negligible and interconnects are transparent, no matter what you do--almost. The shorter the rise time, the more dominant signal integrity problems will become.

Of course, since you can never completely eliminate any of these problems, the goal in a cost-effective product is to pay the minimum extra to minimize these problems to an acceptable level. And, you also want your product to work the first time, because rarely can you afford the time or money for redesigns. You can probably add design margin to reduce the risk of failure, but this adds cost. How do you manage the risk and the cost tradeoffs?

This is where signal integrity analysis comes in.

To Analyze or Not to Analyze, That is the Question

If you want really low risk, and cost is not an object, then just overdesign your product: Use lots of layers, use packages with lots of power and grounds, use really thin dielectric layers between the power and grounds, use lots of decoupling capacitors, use wide signal paths, with Rogers Teflon-based laminates, and use microvia HDI substrates.

Even then, there is no guarantee signal integrity problems will be eliminated, but the risk will be lower. If you want to keep your risk low while minimizing the extra cost of unnecessary design margin, you can analyze the signal integrity effects before you build your product to establish confidence the product will work with the reduced margin, or redesign it before you commit to fab.

Signal integrity analysis is about evaluating system performance, before you build your product. It's not just a matter of using complicated simulation tools; it is using the appropriate balance between rules of thumb, analytical approximations and numerical simulation tools. All three of these tools are important in analyzing system-level performance. If you have confidence in your analysis tools, you can reduce your risk that the product won't work the first time, even with a small design margin.

Signal integrity analysis can save you time and money and let you move up the learning curve for a cost-effective design process. The more of the right kind of analysis you perform, the less design margin needs to be added to your product for the same risk level, and the potentially lower your product's cost can be.

So, when should you go for signal integrity analysis? If you use 5 nsec signals, you may be able to rely on luck to get your design to work the first time.

But remember, as clock frequency goes up, luck goes down. And signal integrity analysis can help you achieve a cost-effective design the first time.