Issue link: https://iconnect007.uberflip.com/i/1295812
16 DESIGN007 MAGAZINE I OCTOBER 2020 parameters, and run. You can have simulation results within five minutes that give meaning- ful insight into what, if anything, you should change in your design. As an industry, we keep getting caught up in the idea that an analysis tool has to do every- thing with maximum accuracy to be useful. We think that if a tool can meet the needs of the most demanding experts, it will work for every- one, but that's not really true. The "expert" tools are often standalone applications that users string together to form their own analy- sis flows. The typical system designer doesn't have the time to learn yet another complex tool, let alone multiple tools and a way to string them together. We need focused toolsets that let designers perform basic analysis quickly and that don't have a huge learning curve. Five years ago, 25-gigabit serial links were the hardest thing going, and you needed the best people in your organization working on them. Today, we're trying to put them in mass production across scores of designs. How are all the teams trying to deploy this technology supposed to get their work done? The experts have all moved on to 112-gigabit links, and they're busy doing that. I think the real chal- lenge is, "How we deploy these advanced tech- nologies once they've been proven?" Shaughnessy: Do you see your typical users being degreed engineers? Westerhoff: Let me change that question a bit. I think the challenge is, "Are the people running SI analysis dedicated specialists?" The unfor- tunate truth is the majority of designs still don't get simulated in any significant way. HyperLynx has enabled a lot of people who aren't SI experts to run their own analysis. Our goal is to expand that pool of people because that's what the industry needs; the shortage of SI experts isn't going away anytime soon. Happy Holden: Todd is preaching the gospel, based on the time that I spent at Mentor ear- lier in my career. But this is everything from modeling to simulation to the trade-offs. It was tough 20 years ago, but it's much tougher now. Nolan Johnson: I spent my time in Mentor cham- pioning what we now call the digital twin—a brilliant design advisor that allows the user to make some high-level trade-offs between costs, productivity, thermal, SI, etc. early enough to keep you from running down the wrong alley. Whether or not you know anything about SI, Mother Nature is conspiring against you with very fast transistors. You can set the clock, but you can't change the fundamental characteris- tic of that transistor, which is going to turn on and off pretty fast. Westerhoff: The phenomenon of high-speed behavior on slow-speed signals is very real. One of the common rules of thumb is that you need to consider a net as "high-speed" whenever the delay of the net exceeds one-sixth of the driver's edge rate. Consider what that means for a driver with a 250 ps edge rate, which is pretty slow by today's standards. The longest a net could be and not be considered high-speed would be about 42 ps or roughly one-fourth inch. That means pretty much everything is high-speed unless you're dealing with old driver technology on an old process node. Johnson: And you can't escape it because, like death and taxes, Mother Nature is going to keep making things smaller and smaller, which favors SI, and we can't breed the SI gurus fast enough. We have to figure out how to simplify the first order to get you 80% there. Westerhoff: We've been talking about simu- lation and modeling for as long as EDA has existed, assuming that the design community We need focused toolsets that let designers perform basic analysis quickly and that don't have a huge learning curve.