Issue link: https://iconnect007.uberflip.com/i/666328
66 The PCB Design Magazine • April 2016 ing with leading-edge customers to develop and prove new capabilities on working projects be- fore bringing those capabilities to market. We use our own software on complex real-world applications every day, because that's how you make products great. Anyone can come up with a canned demo that makes their software look good, but creating software that works well day- in and day-out on real world problems is an- other thing entirely. Shaughnessy: You have a couple of new things you're showing here at DesignCon. Why don't you tell us about what's new at SiSoft? Westerhoff: We're introducing a new technology called OptimEye that automatically co-optimizes transmitter and receiver settings in AMI simula- tions to maximize eye height. This lets users re- place the hundreds or thousands of "blind sweep" simulations they use to figure out equalization settings today with a single OptimEye run. Shaughnessy: Is this kind of capability new to the industry? Westerhoff: EDA simulation tools have had different types of optimization capabilities be- fore, but none of them have ever been specifi- cally designed for SerDes equalization. When people talk about a simulation tool having an "optimizer," they're typically talking about a generic set of algorithms that vary a set of con- trol inputs to affect some output, or goal. Those algorithms run different simulations to deter- mine how different settings affect the output and then attempt to adjust the control inputs to achieve some goal. But—and here's the im- portant point—those generic algorithms don't really have any understanding of the technol- ogy they're trying to optimize. They approach every problem the same way, with no applica- tion knowledge. SerDes links and AMI analysis, however, rep- resent a very specific class of problem. If you focus on how signals propagate from a SerDes transmitter to a receiver and how equalization affects those signals, you can do a much bet- ter job of developing algorithms that will ad- dress that specific problem. This is what SiSoft has done with OptimEye: developed a new co- optimization technology specifically targeted at IBIS-AMI simulations. This technology is some- thing that SiSoft has been working on for a long time, and OptimEye is actually the third gen- eration of that technology. Shaughnessy: What design problem does it solve? Westerhoff: Designers need to determine which combination of Tx and Rx equalization settings produces the biggest eye in the receiver. That's true in both AMI simulation and in the lab with actual hardware. Even though many receivers have a "full automatic" mode where they de- termine the best Rx settings based on the in- coming signal, transmitter equalization is still something that the user needs to program. And, as it turns out, determining optimal Tx settings, even with a fully automatic receiver, isn't that simple. Even for a moderately complex Tx, the number of combinations can be 200 or more. So aside from trying each possible combina- tion in sequence, how do you choose? There's a definite science to how you select the set of transmitter settings that will give you the best performance out of the receiver, but it requires a detailed understanding of the transmitter's equalization capabilities, the receiver's equal- ization capabilities and the characteristics of the serial channel. Evaluating the tradeoffs gets complicated quickly. So what a lot of designers do is just try as many combinations of Tx settings as possible, " Even for a moderately complex Tx, the number of combinations can be 200 or more. So aside from trying each possible combination in sequence, how do you choose? " sisoft: optimizing the state of the art