Effectively designing a Power Delivery System with Expert Lee Ritchey

Lee, I was just thinking about our conversation about the immense power that you and I both saw being exhibited at DesignCon just a month ago. If it's so challenging now, what does the future hold? Well, I'm thinking I might retire. You recorded your class. I'm out. Music. Hi, everyone. It's Judy Warner. Welcome back to this week's Ecosystem Podcast.

Today, I'm joined by high-speed expert, Lee Ritchie, who's going to talk about power delivery systems, how you can navigate some pitfalls in that area, and why they've become such a key topic of discussion these days. I also share some online classes he's done below in the show notes, so make sure you go check those out, and I hope you'll enjoy this conversation with Lee Ritchie of Speeding Edge. Hi, Lee. So good to see you again. I'm excited to talk to you about power delivery systems today.

Well, it's my pleasure to be a part of this little adventure. So, I was made aware that you did this online class about power delivery systems recently through Endeavor Media. So, I knew if you're teaching a class about it, it's something that our listeners would want to know about. So, why don't you start out by defining what a power delivery system is and what elements are involved in that from a design engineering perspective?

All right. Well, our delivery system's job is to make sure power is delivered to an integrated circuit inside a package so it can do whatever it's designed to do. Historically, we have considered the power delivery system to be the DC converter, the power ground layers on the board, and the capacitors that engineers place on the board. And that used to be enough.

As we have gone up the speed curve and the transistor count curve, that has had to be expanded to include the capacitors that are put on the IC package by the manufacturer of the component and the capacitors that's engineered into the IC itself. And none of the modern component we use will work unless all of those have been included. So PDS design now includes all of those together. And for us engineers who are designing the boards, two of those we don't have any control over.

One of those is, of course, the capacitors that the IC manufacturer put on the package. Uh-huh. And the other is what's on the IC. see, I forgot one element, and that's the package itself. Gotcha. We've got such high currents, the package itself has to be included in the design flow. I mean, of course, that is not in the hands of the board designer. Right. So how do we, well, first of all, a question, a personal question, our listeners might know, but I may not know.

I came from, you know, I have experience in the EDA space. And in that space, a lot of times they talk about the power delivery network or a PDN, you know, there's PDN software. What's the difference between a PDN and a PDS? Well, I've said a number of times, it's tomatoes, tomatoes. Some people, for some reason, want to use the word network. I consider that a bit misleading because mine says networks are a bunch of wires.

And, of course, a PDN does not look like that. If it had a bunch of wires, it wouldn't work. So the industry, for example, if you go to DesignCon, you'll see PDS as the common designator. Our delivery system, because that's in fact what we're engineering. Okay. Well, that clears that up for me anyways. So you mentioned all these elements that are involved and then, you know, there's the chip level stuff and then there's the stuff that a board designer needs to own and consider.

Why has it become so challenging and why did you decide to teach a whole course about this? Well, it most of all has to do with the level of integration we have in integrated circuits, where it is not uncommon to have an integrated circle with several hundred million transistors. And how that density is achieved is by making some devices so tiny that they can't tolerate a high power supply voltage. So we get hit twice.

We have to reduce the power supply rail voltage down and most of the time now it's under one volt. And at the same time, the currents that you have to engineer are in excess of 100 amps. And there was a paper I'd done at DesignCon this last January where that current was over 1,000 amps. And how do you get that much current into a part.

With copper layers and a thin circuit board inflated through holes and packages that typically have, if you're lucky, metal layers that are one mil thick. And that's basically what's driving pretty much all of the tools. Most of the tool vendors now have the ability to include all those elements in your model. And you kind of have to do that. Now, here's a loaded question. Are the models accurate? Depends. as Eric should want to say all the time.

When I teach my class, my mantra is an unvalidated health simulation is worth the papers written on, meaning that you have to somehow verify that the model you're including represents a real hardware because the tool itself does a good job. It's more often than not, it's incorrect models. Right. Most of us have had to resort to building prototypes and measuring them in order to get that. Historically, that's how we've all done it.

There exists some tools that if you are good at driving them, will give you a pretty decent model. The trouble is, there are so few people who can drive those tools. Yeah. Yeah. I mean, that's fundamentally a problem across the whole board, including PDF design. There's so few people with the skill right now that we're hurting as an industry. Which I, you know, which kind of speaks to my second point is why you decided to...

Create this core so people could upskill. And yeah, at DesignCon, I don't know if you saw Steve Sandler's paper or his booth, but they were testing a 2,000 amp core power rail. And he had refrigerated probes. Oh, I believe that. Refrigerated. I believe that, yes. And he had water cooled last year at DesignCon, but this year it was refrigerated. And he said every semiconductor company has contacted him. So I guess we're pushing the power envelope for a variety of reasons.

So this sort of ties into conversations I've had with him. Thanks so much for listening to today's podcast. I hope you're enjoying our awesome expert guest. I wanted to let you know we've got a new website engineering platform at TheDoubleEcosystem.com. So go on over, join our community. I've got free downloads from Eric Bogutin, Sam Tech, Signal Integrity Resources, and so much more. And we've loaded it up all for you. And we've also launched a new career placement services.

Services so come on over join our community and make sure you do all the usual stuff on youtube and your podcast app thanks so much now let's get back to our podcast so in regards to the board designer. What are some common myths or misconceptions, specifically around the PDS, that maybe you're trying to address in your class? Well, certainly at the top of the list is application notes that almost never are correct. In fact, I can't say I've ever seen one that is correct.

Yeah. And of course, engineers who are beginners, that's what they have to work with. Yeah. Yeah. And when I teach my class, we go through and illustrate the things that are required, the elements that are required. And then I go and look at a typical application's note. And for example, this is historic. Way back when, a 0.1 microchip capacitor were all you needed. Things were so slow. And I still see application notes being printed with that on it.

And the frequencies that are involved in high-speed switching are so far above for those capacitors that are functional, they're just useless. Second is we know that if we have a wide parallel bus, we need some plane capacitors in the pinnacle circuit board to supply those frequencies. And to date, I have not seen an application note that suggests you should have plane capacitors built into the pinnacle circuit board.

And so if you're attempting sort of to illustrate how some of the smarter IC manufacturers have dealt with this is Intel has been doing this forever. The four-layer board does not have a plane capacitor, and so Intel has integrated onto the package and onto the die all the capacitors required for the part to do its job.

That's been going on for at least 30 years. Of course, the engineers at Intel know that, and they know that their customers are designing four-layer boards and don't know much about this, and so... They want their customers to succeed in what they've been doing. And lately, companies like Xilinx and Altera Broadcom are doing the same thing.

And I won't name who it is, but I saw an app note from one of those that at the bottom said, we have engineered all the capacitors on the dial and on the package you require for this part to work, other than the pastures for the switching noise on your switching power supply. And the last statement, this is on page one, because engineers don't know how to design power delivery systems. There it is, in writing, from IC manufacturers.

You can kind of guess they got some backlash on that and that last sentence is not on the current app note. Somebody got in trouble for that one. Yeah, we'll defend our customers right off. Yeah, we've done it for you because you're a dope. Yeah, pretty much that's the translation.

Unfortunately, they're correct. And that is happening quite a bit. And so for an awful lot of the parts, you don't have to have very much skill. Well, and I always like to say, and you and I have talked about this at different levels many times, I always say everybody's innocent because EEs are laying out boards and they were never taught it. You know, they're just thrown in the deep end of the ocean many times without all this insight, which is why I do this podcast in the first place.

Right. Right, is to help engineers navigate those gaps. apps. So anyways, I didn't want people to think we're saying they're dopes. They're not. So we're just here to help them. So I always like to, on the podcast, Lee, give our listener audience like one thing they could do today to make themselves better. Like, do you have one concrete tip? If someone listened today, I could go do this one thing to make my PDS better.

Well a lot of we had wide parallel buses like address and data buses if i could give you no other advice i'd say make sure you engineer a good interplane capacitor into your pin circle board stack up okay because there was a of two or three hundred emi problems i have trouble shot all except one have because been because that was not part of the board layout and it's required for for those high-speed signals.

Well, back to Steve Sandler, and I've said this many times, is that what he's told me, you know, as a power expert, is when he designed the power system for the International Space Station, like 200,000 pages had a stamp on them that says, does not include board effects. And now he said, now forget it. It's all board effects. Like, this is the difference of the times, right? Like you said, there's way more power now. Before, it used to not matter, and now it matters a lot.

So, a little bit, I do this in course to give perspective to people who have not done this before. It takes about 60 amps to start your car, and you know how big the cable is that goes through the battery. It's about as big around at least as your finger. and we want to do 120 amps in little films of copper called planes. And that's not simple.

And I illustrate sometimes the last big design we did was a router, was a switch fabric chip that was 120 amps at 0.9 volts and the board had 29 different power supply rails on it. A couple of 128 gigabit links. I did all the SI engineering in about two days. It took a whole month to get the PDS right. Wow. So at what point in the design should design engineers be thinking about the PDS? Before you do the stack up of the pin circuit board. Okay.

So that's the critical piece. It has become more critical than all the signal integrity work you have to do. It's harder by a lot. That's wild. Why? Why? Why is that? because the primary things that are in most designs now are memory. Yep. And almost every connection is a differential signal. Uh. And differential signaling is, by comparison to routing a parallel bus, a slam dock. Uh. You route two wires the same length and you go have a beer and you're all done with the differential pair.

Unless you have skew. well you got that that's that has to do with choosing your board material doesn't it yeah not and then again again knowing really getting to know your board materials, is really important and not just blindly trusting those things as well dropping the other shoe you. And without question, PDF design just dominates everything we do. Okay. And if you were to look at the papers at DesignCon, it was dominated by PDF design.

Interesting. So you're saying stack-up is critical and knowing what's what with your circuit materials. Yeah, well, the material you use in a stack-up doesn't have much effect on the PDF result. Well, the pre-processing. Serial links, that is the effect. Oh, yeah. But copper, how much you have and where it is, is critical. Got it. Okay, I misunderstood you there. That's critical, yeah.

And that's why PDS design has to occur before anything else. And the other element of the stack-up is that plane capacitor we talked about, has to be engineered into the stack-up. Okay. All right, these are some really good tips. Plus, I'm going to send folks over to take your class. And if they're interested, they can go get the full treatment.

So before I let you go, Lee, I was just thinking about our conversation about the immense power that you and I both saw being exhibited at DesignCon just a month ago. If it's so challenging now, what does the future hold? Well, I'm thinking I might retire.

You recorded your class i'm out, good luck both my partner and i both decided that when they want us to do 128 gigabits per second we are we're finished it's like you guys are thinking we got you this far you're on your own now yes Yeah, he went to a design con, and we did this on purpose. We toured all the booths that were showing that sort of thing. Well, none of the signals were in pen circuit boards. Oh, okay. All the green X. Right, so it's all the flyovers and... Yeah, they're...

Because, yeah. So the question to be asked, and this is kind of off the subject, to be asked is why, and that is, at those rates, no matter how hard you work, you can't find a glass weave that's good enough to keep the skew under control. Right. Okay. So, Samtech and Molex and everybody else is just going to do really well then. Oh, yeah. In the future. Yeah. And the solution is laminate that does not have glass. And we are working with two developers on that right now.

I'm not. We'll see. When are we going to get to hear about that, Lee Ritchie? I had to sign an NDA. I can't tell you. I know you can't tell me, but when do you think any of that? Before this year is out. Okay. Well, that's helpful. Yeah, I don't want you telling secrets now. I just wanted to know when we might expect.

Okay, well, this is a perfect place to wrap up. We have test boards. How's that? Okay. Well, I want you to call me the minute that that's available, widely available, and you can teach our listeners and talk about that a bit. Okay. Okay, good deal. Well, Lee, it's always a delight to talk to you. You always have so much to say on all the leading and bleeding edge stuff. I'll definitely send people to your class. Where else would you like me to send them to?

Your website, your book? Where else would you like folks who are listening to go? We have five-day online courses. Okay. Over all the subjects, starting with fundamentals and getting your all ideas. The title is Signal Integrity and Getting to 56 Gigabits per Second. Okay. And what we are just talking about here is one element of that. That is an online course, and you can find out about it on our website. Okay, great. And I get to sit here and do it, which is kind of nice.

Yeah, right? The upside of remote work, right? What people can't tell is I'm looking out at the Pacific Ocean right now. Yeah. It's hard. Rough life. It's hard. It's rough. Somebody's got to do it, Lee.

It might as well be you. you've earned it you've ridden the ride through all these years so enjoy well Lee thank you again for coming today and teaching us about that and I will certainly put those links in the show notes okay thank you for our listeners make sure you go check that out there's some really great resources for you there I hope you've enjoyed this enlightening conversation with Lee Ritchie and we will see you next week until

then remember remember to always stay connected to the ecosystem. Music.