PODCAST EPISODE

								 Judy Warner (00:00.834)
Hi, Ben. Thanks so much for joining me again. It's always great to have you on, and I'm excited to learn about your EDIcon presentation that you're doing next week. But why don't we start off by having you just introduce yourself to our audience.

Ben Dannan (00:15.336)
Hey, Trudy. Thanks for having me back on. I always love being on your podcast. So yeah, I'm a signal integrity and power integrity guru, I guess you could say. I love playing in this area. And so I get to work with folks like Steve Sandler and Heidi Barnes. I work at a big aerospace defense contractor. And I have my own LLC that just allows me to play and work with these folks, these amazing folks, and grow and learn and just.

do fun things in the field of signal integrity and power integrity. That's a good enough high level, right?

Judy Warner (00:48.598)
That's you in a nutshell. Yeah. We don't have enough time to talk about all the projects you're into, but, and you're an, an ex military guy too, and from the reserve, so why don't you give them a quick story about that.

Ben Dannan (00:59.156)
I am.

Ben Dannan (01:04.728)
Oh sure, yeah I was a combat system officer so for those Top Gun fans out there, if you like Maverick, I'm Goose. So I'm the backseater, I was the guy telling the pilots turn left, turn right, and the pilots just looked at me and said NAV, sometimes we're just called NAV. I just want to know this, trees get bigger, so stick goes down, trees get bigger, stick goes up.

Judy Warner (01:28.678)
Just the facts, just the important stuff. Well, let's talk about in regards to this power integrity expertise you've developed over the past several years. What are you going to be presenting at EDIcon? Because I don't even want to try to say the title of your talk. It's too long for me.

Ben Dannan (01:47.628)
Sure, sure. So it's on October 4th and I think it's at 1.30 p.m. The title is What is Enough? VDDQ package power integrity analysis with a DDR4 PHY. And the whole premise of this talk, it's a real-world story that I wanted to share with a colleague, another colleague and I actually worked on this, a colleague named Jim Kaczewski, he's a good friend and a great engineer.

And we wanted to present the story about an experience where we went through and we've been seeing this over and over and this is a topic that's actually pretty near and dear to my heart, which is vendor models aren't always right. And what do you do? And it's already a big challenge today if you're trying to do an end-to-end power integrity analysis and let alone with a full DDR

power integrity analysis where you have multiple DRAM, you have your ASIC, which has the DDR4-5, you have your package, your port effects, and your VRM model, and let alone these margins for the voltages are getting smaller and smaller, and so one more thing you have to worry about is just making sure your models are right. And so if your model is right, what do you do? Or isn't right, excuse me. If your model isn't right, what do you do? And so what if it's a die model? And so that's the whole premise of the story that we're gonna share.

It's going to be a lot of fun and really looking forward to it.

Judy Warner (03:15.778)
So first of all, so it was this project that made you sort of want to share this experience and this learning for people who are attending EDI Con, which by the way, if you haven't heard me say before, it's online and you can register and you can watch it live and get continuing ed credits or you can watch the recording later. So what should engineers who join your session expect to learn?

Why did you want to learn it?

Ben Dannan (03:46.804)
Well, I love power integrity first and foremost, so this is my wheelhouse. And I think one of the things that we want folks to walk away with is that they need to trust and verify the models they're getting from their vendors. Whether it's a capacitor or a dye model, it doesn't matter. If you look at the IBIS forum, right, the IBIS standard, they actually have a quality report. And most folks may or may not know that. And it's not...

It wasn't well known to me years ago, and it's been out for a while. And they actually are teaching folks to say, hey, if you're getting an IBIS model from a third party vendor, ask for the quality report. And so if you're getting a die model, why shouldn't you get some sort of quality report to hold these vendors accountable? I mean, these are just as complex, if not more so, than any sort of IBIS model. And that's part of the gap in the industry. I mean, you go and pull a capacitor model from Murata or AVX. There's no standard.

on how they made that measurement. And that's part of the problem, right? So being right matters, which is what one of a good friend of mine and a mentor says, Steve Sandler, and a friend of yours as well. And to be honest, having good models is where it starts.

Judy Warner (04:45.805)
Yeah.

Judy Warner (05:01.286)
Yeah, which yeah, I had hiding Steve on not long ago and we definitely hammered on that pretty good and of course I smiled right away just then because I recognized your quotes from Steve. He's, he's drilled into my brain too. So what kind of, without giving away your secret sauce or what you're going to show. Like, yeah. So what do you do? How do you know if you have a good?

model, like what are some solutions you're going to offer and some things you have found that work for you?

Ben Dannan (05:32.46)
Great question. And so some of the techniques we're going to show aren't, I wouldn't say there aren't, I would not say that they are new. You can find these in some great books by some other evangelists like Eric Bogatin and Larry Smith. In fact, we're actually referencing one of their books as part of this effort because it's right out of their book, but these things are easily forgotten and maybe you didn't think you could apply it in this manner. And so that's what we're showing in this talk. So in fact,

Judy Warner (05:52.532)
Okay.

Ben Dannan (06:02.964)
We initially got the die model from a third party vendor and in most cases people aren't going to Design their own DDR4 or 5 right? They're going to integrate a 5 from somebody else because they have their own IP on the ASIC to worry about But they have some requirement to have some sort of memory and so they're going to Integrate this DDR4 or 5 from a cadence or synopsis or somebody else. There's only a few vendors out there. You can buy these

Judy Warner (06:28.59)
Mm-hmm.

Ben Dannan (06:29.616)
And so in this instance, you have all this data coming in. And it's hard to vet it until you need to use it. And what we show is initially the DIMAL that we got from this vendor was not the right value, specifically the on-die capacitance or C-die. And we walked through the exercise, even after I approached the vendor many times, I said, hey, so you guys know that C-die is...

way too low. They're like, no, that's right. I'm like, are you sure that's the right value? Because there are other die models for other memory controllers and it's way higher, like orders of magnitude higher. And they're like, no, that's right. I'm like, okay, so can you like maybe do like a manual calculation to show me what that C die should be? And it's really not that hard to do this because all you have to do is, you know, look at the area of the caps that

you know what your DK is, because that's in the tech file, and you can calculate the on-dye capacitance because epsilon times A over D equals the capacitance, and multiply that by the total of capacitors on the grid, that should give you an approximate. And they're like, no, that's right. I said, okay, so they're gonna really make me dive into this improvement tool. And so that's part of the story that we're gonna share on October 4th and walk everybody through this process of how we went from A to B.

all the way to calculating what the minimum on-dye capacity needs to be. Using a technique right from Eric Boggerton and Larry Smith's book, to be honest. So it's not rocket science, but there's so much out there. The goal is to show folks a practical approach and a way to use these methods while showing them how to do complex modeling of this nature.

Judy Warner (08:13.623)
Right.

Judy Warner (08:21.726)
Well, you know, Eric, it'll be the first to tell you is as wise and guru E that he is, is every book he writes is the practical guide to right. So I'm sure he would love that you were citing that and making it practical. And also I should mention to the audience that your talks only like 25 minutes, right? So it's pretty easy to digest. Awesome. And then. I think it'd be really.

Ben Dannan (08:44.267)
Mm-hmm. Yeah.

Judy Warner (08:52.094)
I like the way you're kind of serving up as a real life case study that you guys went through and will be so relatable. It's almost like a technology story. Yeah. But you're seems really relatable than a bunch of theory.

Ben Dannan (09:05.308)
Yeah. I mean, we knew it was wrong from the get go. Just going about proving it, right? We're all working for big companies and some bigger than others and a big company like one of these, you know, where they're in the billion, it's hard to break down those barriers. And so if you're a smaller entity and you've bought some IP, how do you go about proving it? And that's the whole approach here that we're trying to show folks. So hopefully that's the takeaway. And they also walk away with, they need to hold their vendors accountable.

Judy Warner (09:21.063)
Yeah.

Ben Dannan (09:34.392)
Ask for some sort of quality report, right? Verification, don't just take a model and willy-nilly just throw it in your stimulation, yeah.

Judy Warner (09:39.456)
Mm-hmm.

So what I want to know is did you go back to that vendor and say, No, it's not and I can prove it.

Ben Dannan (09:47.424)
Oh, so we yeah, so we proved it and you know what they did. They came back and said, oh yeah. That that's the wrong value we gave you after we picked at it and we had schedule slip in our program and so at that point we're like OK, so what's the right value? So they gave us a new dye model and we had calculated that the dye model on dye capacitance needs to be around 6.2 nano ferrets. It was initially 3.0.

to 9 nanofarad, so almost half what we calculated the minimum needed to be. And they came back and said, yeah, the actual value is like 6.49. So really close to what our calculated minimum was, like really close. So we dropped that into our model and we were failing their spec miserably, right? There's an AC voltage ripple spec that we were assessing at the die bumps on our package.

Judy Warner (10:29.815)
Hmm.

Ben Dannan (10:43.976)
And we were failing it by over 2x, their limit, with the initial dye model of 3.29 nanofarads. As soon as we dropped it in, the new dye model, we were passing just barely, but we were passing. So it just shows you that what we calculated and what they gave us, these methods are very practical and applicable to any sort of application like this.

Judy Warner (10:43.982)
Mm-hmm.

Judy Warner (11:08.79)
Well, you know, I was listening to an executive do a talk recently and, and she was an executive like at Flex or something like Bigwig. And she was talking about some challenges they went through with COVID and having a global team and all that. But the one thing I took away that I thought she goes, well, let's not just sit around and admire the problem.

You know, and I'm like, guy, don't we do that though? Like, yeah, it's those darn models, right? But I really love that you guys, you know, sadly, you know, your schedule slipped, but that you actually took the time to do it and bring it back and like you said, hold the vendors accountable and you actually didn't admire the problem or just fix it and go on. I think there was probably a good learning in it for you and the vendor. So very cool, love it.

Ben Dannan (12:00.36)
Yeah, yeah, there's some sanity check that we're going to show folks on how to do some basic sanity checking, right? That's part of this. I mean, we even showed a comparison to a Micron SDRAM, a Bi-16 SDRAM, and we showed, by the way, this is a great example. Every Micron SDRAM, they provide a quality report as part of their IBIS model and for their

Judy Warner (12:22.574)
Hmm.

Ben Dannan (12:26.304)
the documentation from Micron so great. So why shouldn't you have the same level from any other vendor? And so we looked at the dye model on dye capacitance and it's right there. It's more than double the initial value that we got from our IP vendor. And we showed that to this IP vendor and they're like, yeah, no, it's still right. And so just showing folks how to go through this method and apply this to get the right answer, I think that's probably the fulfilling part.

Judy Warner (12:42.135)
Hmm.

Ben Dannan (12:54.92)
teach and help folks understand what to do if you're not getting the right answer from the vendor and how to prove it. These aren't easy to model, right? There's a lot of complexity in the model like this. You have 72 DQ nets, your strobe, your CAC, all running to toggle all those simultaneous switching nets to create that noise, to get that overall response. And so it takes a lot of work to prove this to a vendor.

like one of these big guys.

Judy Warner (13:26.358)
Well, it's great that you did it and I'm sure there'll be some really good takeaways. So Ben, thanks so much for coming on and teasing up your class and I'll definitely be there. And I'll call you later if I have questions because I probably will. But sounds really great. Well, thanks again for coming on, appreciate it.

Ben Dannan (13:47.572)
Yeah, thanks Judy for having me. Always a pleasure.

Judy Warner (13:50.582)
So for our listeners and our audience, make sure you go, I'll put the links to EDIcon that you can go register there. It's free and you can watch it online and get your continuing ed credits or watch it, just register and they'll send you the recording later, but make sure you choose Ben's session. Thanks for being with us. Ben, we'll see you next time. And for our audience, we'll see you next week. Until then, remember to always stay connected to the ecosystem.