Issue link: https://iconnect007.uberflip.com/i/666328
April 2016 • The PCB Design Magazine 67 then pick the best one. When that process is done with simulation, it's something we call a "blind sweep". So, for example, if we have a Tx with 3 taps that have 10 settings each, that's 1,000 combinations. So, for each channel they want to optimize, users run 1,000 different simulations and pick the one that gives them the best results. Shaughnessy: That seems like a lot of simulation to optimize just one channel. Westerhoff: Absolutely, but that's really just scratching the surface, because no one has just one channel or runs simulations just once. In fairness, people usually don't analyze all com- binations of settings, because all those simula- tions would take too long and there are a num- ber of combinations that can be ruled out any- way. But, even if you reduced those 1,000 com- binations down to 100, you still have to mul- tiply those 100 simulations by the number of channels you need to equalize and the number of design changes that require re-running the analysis. And thus, the number of simulations required across the design cycle climbs into the tens or hundreds of thousands pretty quickly and that's the problem to be solved. Those simulations consume a lot of time and compute cycles today using blind sweep methods. I ran a case study for a customer the other day, where optimizing Tx settings for one channel involved 1,120 simulations and 624 CPU-hours (26 CPU-days) – and that was just for one uncoupled channel. If we can replace those 1,120 simulations with a single OptimEye simulation run, there's an enormous savings there. Even if that one OptimEye run takes 10 times as long as one of those 1,120 blind sweep runs, that's still a savings of about 100 to 1. Shaughnessy: So, how does OptimEye work? Westerhoff: OptimEye is an application-specific technology that combines three key things to determine the best combination of equalization settings: simulation results using the vendor's actual IBIS-AMI models, information about the equalization capabilities of the specific SerDes Tx/Rx, and knowl- edge of IBIS-AMI models and IBIS- AMI simulation methodologies OptimEye is able to take a set of simulation results using base- line Tx/Rx settings and determine whether or not those results are optimal. If they are not, Optim- Eye predicts what the best settings would be and re-runs the analy- sis to see if the vendor models behave as pre- dicted. If not, OptimEye adjusts its prediction and the process iterates until it converges. One of the best parts of this technology is that the SerDes vendors don't have to rewrite their mod- els for OptimEye to work with them—a separate control file describes the SerDes device so that OptimEye can predict the best settings and run simulations using the vendor model to verify those predictions. There's more to how this works than I can describe here, but readers can go to our website to get more information about how OptimEye works and watch videos on the subject. Shaughnessy: I saw that you're also presenting a paper with MathWorks on creating IBIS-AMI mod- els. What's that about? Westerhoff: We're presenting methods for le- veraging SerDes data for AMI model generation. This has long been considered the Holy Grail of IBIS-AMI model development, being able to take the schematics and code created during SerDes design and use them to generate IBIS- sisoft: optimizing the state of the art " OptimEye is able to take a set of simulation results using baseline Tx/Rx settings and determine whether or not those results are optimal. "