Issue link: https://iconnect007.uberflip.com/i/241303
feature new year's resolution: validate your model data continues ticular semiconductor vendor's simulator. There What's a user to do? While there are numerous programs that claim to "detect and correct" are other types of models as well, but these are issues with S-parameter data (causality and pasthe types that most users have to contend with. sivity being two of the most popular issues), we Each type of model has its own unique charmaintain that users should take a close look at acteristics, limitations and typical quality probS-parameter data before letting a computer prolems. If you spend enough time working with a gram change it. This means developparticular class of model, you'll probing techniques to evaluate the ably develop techniques for veriquality of S-parameter data. Dr. fying those models before use Michael Steinberger's article in system simulations. A prototype board TDR: Reading the Tea Leaves Why should you care? outlines the use of TDR techWell, if you're using the rerun for a high-end niques to do just that. Designsults of system simulations to system can cost ers can use these techniques make design decisions, you should care a lot. A prototype $100,000, plus the to understand the topology that the S-parameter data board run for a high-end systime it takes to describes and its key charactem can cost $100,000, plus teristics. If those characteristhe time it takes to get assemget assembled tics don't reasonably match bled boards back and running boards back and the circuit being modeled, in the lab. Bad design decireally not sions that result from poor running in the lab. then there's worrying much point in about models can delay a product's Bad design decisions whether the data is passive ship date by months and and causal and how it might the lost sales could probably that result from be corrected. pay off your house. So, model The same is true for quality matters, but the real poor models can semiconductor models. Baquestion is whether users are delay a product's sic characteristics of the I/O in a position to do anything about it or not. ship date by months circuitry—output and termination impedance, voltage For example, let's look at and the lost sales swing, ESD capacitance, slew S-parameter models, which are useful for all sorts of apcould probably pay rate, non-linearity, equalization and more—can be obplications. In practice, there off your house. served by exercising a model are limits as to how accurate under the right test condimost S-parameter models are. tions. If the model meets reasonSince they are frequency-based able expectations, great—it's time models, they should specify how ento start running system design simulations. If ergy is transferred all the way down to 0 Hz, or not, it's time to figure out what's going on with DC. S-parameters can be created either analytithe model before running simulations that will cally (often through the use of 3-D field solver have to be rerun later anyway. software) or empirically (derived from measureAs I write this, I can't help but think that ment). Both techniques often have trouble genI'm coming across as a spoilsport. I imagine erating the DC portion of the S-parameter file the grumpy old man sitting on his front porch, accurately. Why does it matter? It matters beyelling at kids who walk across his lawn, excause many simulators perform a steady-state cept that he's saying, "Mind your S-parameters, (DC operating point) solution before beginning young man! Test your I/O models to make sure transient (waveform) analysis. If the DC charthey accurately represent return loss!" acteristics of the circuit are off, the operating As I've already conceded, testing simulapoint will be off, and the quality of the trantion models is neither quick nor enjoyable. And sient analysis may be compromised as a result. " " 12 The PCB Design Magazine • January 2014