Issue link: https://iconnect007.uberflip.com/i/1369942
18 DESIGN007 MAGAZINE I MAY 2021 Hartley: ere's so much on Google that people don't know what to believe. Shaughnessy: I read an article that you wrote six years ago, Barry, and you explained that DDR specs can be downloaded from JEDEC. org. If you want the real specs, go to JEDEC. org and get them. Why don't people just do that? Hartley: DDR3 is the JEDEC-793B standard, so just download and read it and you'll know exactly what to do. Right, Barry? Olney: at's right. But there is a lot of data to go through. I should also point out that Andy and I were talking about datasheets always being wrong the last time we spoke, and it's the same with DDR specs. If you go to Micron, they have one spec, one timing analysis; JEDEC has another, and Xilinx has another. ey're all dif- ferent. ere's not one spec that's the same. Hartley: Right. at's a good point. Here's a comedic story about that: e first time I ever did DDR2 (I won't say whose device we were using as the controller), in the app note section for DDR2 for that device, they actually said, "We had such a hard time getting this thing to work that if you don't design it exactly as we did, we won't guarantee it will work." ey said, "We'll give you our Gerber files, and if you're a Cadence user, we give you the design file, and if you don't copy it exactly, we make no guarantees." And I thought, "Who are these people?" e lead engineer called me up and said, "Rick, did you read this? is is terrifying; are you going to do that?" Knowing he was the ner- vous type, I said, "We will take care of it, don't worry about it. Everything's good." I didn't do anything they said in the app note, and it all worked perfectly. Olney: Reference designs are always bad designs. Hartley: Almost always. I don't think I've ever seen a reference design that was really good. I really don't. Olney: I've seen DDR2 routed on four-layer boards in reference designs, and how they get it to work I'll never know. Matties: It sounds like it's not that complicated if you do your homework. Olney: If you follow a methodology, you can't get it wrong, really. Shaughnessy: So, designing DDR involves a few slightly different steps than a typical high- speed board? Olney: Each technology has different con- straints, so you want to keep the technologies apart. If you have a 5-volt or 3-volt technology next to a memory that's 1.5, then you're going to get more crosstalk because it's a higher volt- age; there's separation. You have technology rules, and you have the rules that separate the technologies. Hartley: I've told people for years that when- ever I would do a circuit board with multiple powers driving different technologies—let's say 1.5, 2.5, 3.3, and 5 volts—I make sure that I place the parts, as much as possible, so that they each have their own power distribution sections, for several reasons. One, you have fewer power planes because you can divide the plane up into multiple pieces; second, you're not routing 1.5-volt or 1.8-volt lines next to You have technology rules, and you have the rules that separate the technologies.