Measurement Based SI Modeling

Multi-gigabit communication links present many signal integrity challenges. One major challege is that of modeling the transmission channel. Accurate simulation at multi-gigabit signaling rates requires models that are accurate out to a very high frequency. In many transmission channel structures full wave 3D effects must be taken into account. Examples are vias, connectors and sockets. Developing the models with the use of a 3D field solver such as Ansoft HFSS is one possibility. I have not had good results with this yet. These types of tools have a very steep learning curve and it is difficult to create 3D geometries that are close enough to the real thing and can have ports assigned in reasonable ways to get good results. I think this type of simulation technology is promising, but it is still beyond the capabilities of a typical digital engineer to use and takes so much time to get good correlation that it is beyond the schedule constraints of a typical digital design project. Another drawback is very long simulation times.

One alternative is to use measurement based modeling. A network analyzer can be used to fully characterize part of transmission channel with frequency dependent S-parameters. These S-parameter models can then be simulated with directly in tools such as Synopsys HSPICE and Cadence Allegro SI 630 or converted into behavioral spice models for use in any spice based simulator. An online webcast, Measurement-based Modeling and Simulation for Gigabit Channels by Cadence and Agilent, describing this process is available at Techonline. The challege in this approach is that actual hardware is needed to make the measurements on. In most designs that I have been involved in, this is not available and test boards and fixtures would have to be designed. This has large schedule implications.

I really don't know what the best answer is for doing SI simulation on multi-gigabit communication links. I do know from direct experience that simulating these links before releasing designs is highly desirable. You don't want to be in a situation where an eye mask test is failing and not have a simulation to go back to for testing out theories for why it is failing and how to fix it.