Issue link: https://iconnect007.uberflip.com/i/1441485
74 DESIGN007 MAGAZINE I JANUARY 2022 their need for simulation tools—and that is what Avishtech is all about. It's what leads to all these problems. Every laminator spends a lot of time and effort in testing and providing the right properties to the best of their knowledge and skills. But the question is how the data is used. Matties: One critical thing to consider prior to manufacturing is the computational side of what you're talking about. Amla: Yes. And that's where the frustration about data comes in, because people are doing model-based engineering. Big defense con- tractors are doing this work, so they're build- ing these finite element analyses and other models. When they ask for data, like the shear modulus of the laminate, they don't get that. Or the Z-axis modulus. ere is no good way to measure these properties and definitely not for every configuration or stackup. Trying to do that computation up front requires this data, which is not available, so you need a com- putational implement to predict the data. Similarly on the data transmission side, as Lee said, PAM2 is fine for 56 gigabits per chan- nel. But when you go to PAM4, then you're talk- ing about loss budgets as low 0.7 dB per inch. So, if you have a 4-mil trace, the copper eats it up. Even if the copper has zero roughness, it'll eat up all that 0.7 dB/inch budget, even if you're doing it on a zero-loss dielectric. At 112 Gbps PAM4, that's where the limit is, as of now. And that is with copper with roughness down to 0.06 microns RMS. Ritchey: One of the ways that this is being dealt with is with twinax for those long paths. I've seen that in servers; the long paths are not in the board. ey're in twinax and that's the big business model of Samtec. If you've ever seen their booths at DesignCon, they have a whole family of connectors that are aimed at that, because we haven't gotten the loss tangent down in laminate with glass cloth low enough. ere isn't any glass in a twinax; in fact it's foam PTFE, which has astonishingly low loss. We are successful at 56 gigabit per second in the kind of circuit boards with what we call NRZ, which is what all the internet is. e problem we're having is PAM4, which for the same clock frequency doubles the data rate, but the signals are smaller, so we don't have the margin we had before for loss or skew. at's the problem we're trying to solve right now. Matties: It sounds like the designer has a heck of a challenge regarding modeling and making all these decisions. Tarun, are you saying that some of the data points don't exist or aren't available? Amla: Especially if you're trying to do that for the defense contractors. A lot of this is hap- pening on the defense side. ey're trying to model this and trying to get more and more compact designs for whatever applications and they need this data. e same thing is hap- pening on the communication side where they want to have this information available. It's a multidisciplinary approach that is kind of tak- ing shape with model-based engineering, but they don't have all the data needed for mod-