Issue link: https://iconnect007.uberflip.com/i/1369942
MAY 2021 I DESIGN007 MAGAZINE 21 ticular, to extract sub-models from models because, quite oen, models won't work. It's pretty straightforward: You extract the topol- ogy, the transmission line into the simula- tion tool from the board layout, and then you attach the IBIS models to each end, but when they don't work, what do you do? You've got nothing. Now, I have soware that extracts the sub-models from a non-working model, and it creates a new model that I can use. e other issue is a lot of signals like PCI Express and Ethernet go off the board, so what then? You have a connector on the end of the board that all your signals go to. e trick there is not to use the connector model but to use the driver model in reverse and use the driver model so you can simulate the actual board. You can't control what happens off the board, you can only make the board perfect and then someone attaches a mile of cable to it. Hartley: at's the other thing—you basically have to simulate that cable at its impedance to determine how to design the board to drive that cable properly. Olney: Not really, because the impedance of the receiver, if you stick it in place of the connec- tor, you will have the right impedance. Hartley: e receiver will, yes, but what if it's a long cable? Long enough that you care about its impedance? Olney: at's what I'm saying: You can't control what happens off the board; you can only deal with what's on the board. Shaughnessy: What final advice would you give to someone who is working with DDR and is having a hell of a time with it? Any tips or tricks? Anything? Olney: Basically, you need to read a lot and educate yourself. With DDR3, 4, and 5, it is important to use the correct impedance of 40/80 ohms, and check the timing and channel isolation. Simulate the clocks/strobes and the two worst nets of each group as you don't have time to evaluate each individual signal. en compare the flight times of each signal to the clocks/strobes, which is a lot quicker. Hartley: I would say read a lot. If you have access to good simulation tools, use them. Understand at what point a transmission line becomes distributed vs. lumped and how you treat it. You must control its impedance. Usually with DDR, of course, you must control impedance; that's true of any high-speed design. If it's short enough you don't have to worry about it, but if it's a distributed link, of course you do. I treat all lines as if they are distributed, assuming they may route long enough to be distributed. If they're not distributed, then we simply don't terminate. Matties: is is great, guys. We appreciate so much your time, thoughts, and insights. ank you both. Olney: ank you. Great speaking with you guys, and Rick, nice to chat with you. Hartley: ank you, Barry. I read your column every month. I'm glad we had the chance to talk. DESIGN007 Understand at what point a transmission line becomes distributed vs. lumped and how you treat it. You must control its impedance.