Issue link: https://iconnect007.uberflip.com/i/1245291
MAY 2020 I DESIGN007 MAGAZINE 19 a lot. I call it the plus or minus one mil rule; everyone is within a mil of each other. Get a compromise set of rules, I've done this multiple times, including stackups. Everyone has a different way they do stackups, but if a couple of companies can make it, the other companies can too. They haven't done it that way be- fore, or it's not the normal way they do it, but they can build it. Part of the problem is decoupling the sup- ply chain from the designer. How are they going to know what rules to use? They don't. Matties: When it comes to the designer, what should the expectation be? Korf: You send me a data package, and I build it as is. I can do my manufacturing, etch com- pensations, and that kind of editing, but I shouldn't have to run a netlist. Why should I? Your data should be good. Matties: The expectation is complete, accurate, and correct. Korf: Right. Matties: Is there an expectation that they have a basic understanding of process knowledge? Korf: That's a very good question. If I'm de- signing a board that hits the middle of the tech- nology of the fabricator I'm using—their sweet spot—it shouldn't matter how they build it. If I know I'm running at the edge of the tech- nology and there are some unique feature at- tributes, or I have to do something special in a process, I may want to know how they do it only because if something goes wrong on that board, I want to know what didn't work. Gen- erally, designers shouldn't have to worry about how it's done. Like I told my front-end people, you shouldn't have to worry about how they design. There are a lot of factors. Why did they do that? I don't know, and it doesn't matter— just do what they said. Matties: And then there are those "Oh, you did it just as I asked" moments where you think, "What was I thinking?" Korf: I worked through this with a large OEM. They were doing their first 5G boards many years ago. They were doing the anti-skew trace routing. You put a differential pair on, and you take one leg and snake it around to compen- sate for the skew. I told them that section is not a transmission line anymore. "Let's look at your TDR here, see those little bumps? That's where you snaked the trace. That's not a trans- mission line, and it's not 50 ohms anymore. This one says 90 ohms, but it's 40 here and 60 here." I've seen an impedance trace over a power plane, and suddenly, there was a gap in that power plane for some isolation reason. The trace runs over a plane, over a gap, and back to the plane. That gap, now the reference plane, is two layers down lower, and the im- pedance went up. You're going to ding me if the whole trace isn't 50 ohms. You didn't de- sign it to be 50 ohms; you designed it to be 100 ohms for two inches. Why am I paying for that design mistake? Matties: The designers' expectation is that you're the backstop to quality check to their work. Korf: It's their assumption versus their expec- tation. I don't know if there's a designer out there that designs a board intentionally wrong; Dana at Tiananmen Square in Beijing.