Bill Hargin of Z-zero: “A lot of people think of impedance and stackups as kind of a monolithic concept, but they’re pretty nuanced, in fact.”
Who owns the stackup design process? Is it the OEM’s design team, which is familiar with all the electrical requirements of the PCB, or the fabricator, who knows far more about the manufacturing process? We recently asked Bill Hargin, CEO of the stackup optimization software tool company Z-zero,to share his thoughts on stackup design. Bill built Z-zero around the need for software tools that can help designers facing stackup challenges. As Bill sees it, OEMs need to start exerting ownership of the stackup—and stop asking their fabricators to be responsible for stackup design.
Andy Shaughnessy: I’m here with my colleague Happy Holden, and Bill Hargin, the CEO and founder of Z-zero. How are you doing, Bill?
Bill Hargin: I gave myself the title Director of Everything.
Shaughnessy: And you had some fairly big news recently: Z-zero is collaborating with Siemens EDA to help optimize the creation of better stackups, which is the topic of today’s conversation. In our surveys, designers talk about how tough it is getting stackup right and all the trade-offs they have to make. What are your thoughts on why it’s so hard to get stackup right?
Hargin: I think the answer to that is simpler than what people might think, but it’s very detailed. But what engineers and designers have been doing for the last 25+ years is throwing their requirements over the fence to their fabs and trusting the fabs, who over time have acquired impedance calculation and simulation tools.We know most of the tools that are out there. But that part of the PCB design process has been delegated to the fabricators for the last 25+ years. And my view is that designers and engineers need to take more control of what’s happening in the stackup design process and own more of it. Bring it in-house. Siemens EDA calls it a shift-left strategy, and the concept is to take things that are known in the fabrication world and bring them to the left in the design process so that you’re making decisions that are going to affect the manufactured PCB earlier in the design process. And you’re bringing fabrication knowledge back to the OEM engineers and designers.
Shaughnessy: So it’s a matter of designers needing to know more about the fabrication process.
Hargin: Right. For instance, I went to a doctor this week, and this is a perfect analogy. I go to this particular specialist every six months or so, and he doesn’t remember what we’ve discussed in the past. He just had a patient leave his office. He got my file folder out, and I can see him thinking right in front of me, looking through my folder. I whip out my laptop and flip it open, and I’ve got graphs and Excel tables. I’ve mapped out my history of medications and the resulting symptoms in Excel, and I’m showing it to him. I have better data than he does.I used to hear that engineers are the toughest consumers in the world to sell to, because they analyze everything. Well, engineers need to do that with stackups and materials. Don’t just kick the can down the road and think your fabricators are going to do everything perfectly. And I’ve had engineers say to me, “Why do I give the same set of requirements to three different fabricators, and I even specify the materials, and I get three different stackups back?” In response, I’ve told them that they need to take more control of the process and leave less margin for the individual fabricators to figure things out separately.
Shaughnessy: In their defense, wouldn’t the fabricators say that three different fabricators are naturally going to be optimizing for different processes, chemistry, and sweet spots?
Hargin: I think that’s a myth, and I don’t fault you for saying that at all, Andy. I think that’s a common perception. But I think it’s a mythological perception that a guy in Shanghai manufacturing a board has a different process than one in Guangdong. I know there are differences in chemistry and how they do things, but in the end, you are specifying a board with impedance targets and tolerances, thickness targets and tolerances, and they all need to meet those specs. What I’m talking about is at the field solver level: they’re coming back with different results from different tools.Why is that? I think it comes down to operator differences. When you get a stackup back from PCB fabricator A and PCB fabricator B, it’s two different people probably running two different field solver tools. So right there you’ve added two degrees of freedom to the stackup design process. And different fabricators will send stackups back to you in different formats, too. One will send you a PDF, the other will send you an Excel spreadsheet, and if you have a third fabricator, they’re going to send you a JPEG.The cattle have gotten out of the barn, as it relates to stackups. Who owns the fact that the cattle are out of the barn? It’s the OEM design team; it’s their hardware, and it’s their jobs on the line. So if your cattle are out of the barn, what do you need to do if you’re a farmer? I come from farming people way back when in the Midwest. And the first thing you need is a barn. You can’t get them back in the barn if you don’t have a barn, and to me, the barn is the stackup software. The second thing you need is a fenced area, which is a process to get those roaming cattle back inside the fenced area leading to your barn.
Shaughnessy: That’s interesting. So, a lot of this confusion could be eliminated if OEMs took ownership of the stackup design process.
Hargin: Yes. Don’t rely on your fabricators to do all that. They don’t talk to each other; you’re the only one talking to each one of them as the OEM design team. They don’t collaborate; they compete with each other. And so, you’ve got to own the entire process. I get in these discussions every week, often every day: Which Dk numbers can we trust? One fabricator says it’s this, and another fabricator says it’s that, and the laminate vendor says it’s another thing. Which values can design teams trust? I‘d like to hear Happy’s opinion on this. But the common mythology is that those differences are due to individual fabricator processes, and that’s not true. It’s fundamentally illogical.
Happy Holden: It’s interesting that you talk about field solvers. In your expertise, is it the different way that people have put their field solvers together, or is it the independent variables like thickness and Dk that would result in fabricators coming back with a different stackup? Not so much whose field solver it is but that each one of them has their own numbers for thickness and Dk that may or may not be close to what it really is.
Hargin: Look at it in terms of degrees of freedom, which is usually used in modeling motion. You can have two degrees of freedom; you can have three degrees of freedom. Here we have at least four, if not five, degrees of freedom. We use HyperLynx field solver, but others may use different field solvers. There’s one set of Maxwell’s equations, but field solver A and field solver B might use slightly different meshing techniques. Now, do the field solvers usually agree? Yes, they do. But let’s say that the field solver is one degree of freedom.The second one is this: What parameters are being fed to the field solver? Are they the same, or are they slightly different? The third is the operator as a degree of freedom: Person A versus Person B at two different fabricators. They’re not sitting side by side, comparing their work. That doesn’t happen until the stackup gets back to the OEM. So, that’s a third degree of freedom. They could be using different Dks and many times they are—a fourth degree of freedom.Then the fifth degree: They could be using different percent copper values, and therefore different prepreg thicknesses. So you can get all that variation. That’s five degrees of freedom that we’ve identified that can lead to divergent stackup results, right?Let’s talk about Dk. This week, I was asked, “Which Dk do we use? Do we use the laminate vendor’s Dk? My fabricators use slightly different Dk values.” Some fabricators will send a stackup back to you and they won’t even tell you the Dk they used. They’ll just tell you trace widths and dielectric height. And sometimes they’ll include their calculated impedance values.Sometimes, if you give them a 50-ohm or 100-ohm target, they’ll just say, “Oh, yeah, it’s 100 ohms.” As if everything was exactly 100 ohms right on the button. And they won’t give you the Dk numbers. So you get all of this variation, and that’s why I say you need to fence those cattle in. I feel bad because I’m referring to people as cattle, but it’s just a metaphor.
Shaughnessy: Designers have been called much worse.
Hargin: You need to have a process by which you rein this whole thing in. If you take a big fabricator like TTM that’s worldwide, they’ll use different Dk numbers at different sites. I believe what people do is they take their test coupons from boards, and they backwardengineer the Dk from using the IPC impedance equations. Is that special sauce? No, I don’t think that’s the special sauce; I think it’s some guy using Excel. Is that due to their special processing? No, I think it’s a guy sitting in front of Excel, typing in equations, and pulling a number out of it, right?
Shaughnessy: Right. It’s a matter of communication, but that’s just part of it.
Holden: At HP, we were in the business of making test equipment, and one of the challenges with test equipment is it has to be about 10 times better than what it’s trying to measure. Being very accurate on these things was a hallmark of Hewlett-Packard excellence. Now, the second thing was that we made dielectric testing machines. The first generation went to 1.8 gigahertz, but you could put in a 30-mm disc of a core laminate, and it would give you back all the characteristics of that material. You could program the computer to do these measurements from 10 megahertz to 10 gigahertz, and from zero to 120° C, and from 30% to 75% relative humidity, and it would give you back a three-dimensional measurement.And the one thing about all FR-4s is that nothing is ever flat, because it changes with frequency, it changes with temperature, it changes with humidity. So, the first thing was that there’s no such thing as one Dk for your board. If you’re going to do it exactly, you have to run a range of what you think the actual environment is that it’s going to operate in.It was so important that I put a separate chapter in the Printed Circuit Handbook about characterizing materials because, the more expensive the material, the more stable and less variation you’re going to get. If you’re sticking with FR-4, it’s a very good mechanical platform, but electrically, it’s all over the map.
Hargin: Yes, but I think what the fabricators are doing is simpler than what you just described. They’re not looking at temperature or humidity or fall results versus summer results, etc. They have a number, and it’s at one gigahertz. Your signal may be at 10 gigahertz, but the Dk they’ll use is most often at one gigahertz. And they just plug and chug, and here’s your stackup.
Holden: Everybody is working with a different set of data books that they created themselves.
Hargin: Correct. And they did it based on backing Dk out of an impedance equation. They do an impedance measurement and then they solve for Dk in the impedance equation, plug in the impedance that they saw, and try to pull their own home-grown Dk number out of it. To me that sounds primitive; it’s not the way I would do it.
Shaughnessy: Is it a matter of the OEMs just claiming ownership of the design, the stackup, everything? I keep going back to what you were saying about the OEMs punting to the fabricator who, as you say, doesn’t know all the stakeholders nor has all the info that the OEMs do.
Hargin: Let’s say I’m a fabricator front-end design guy, and I’m designing stackups. I might do five or more stackups in a day serving different customers. One of those stackups is the one I’m doing for you, the OEM. But you’re doing it the opposite way. You have one stackup that you’re doing, let’s say, with three different fabricators. In the end, who needs to own the divergent results? It’s the OEM that owns it. And, if they own it, they own the result.They need to take more ownership of how the process is done.If one fabricator is using a Dk of 3.75 and another one is using 3.6, and the laminate vendor has in their tables 3.65, well, you need to understand the frequency of the assumption. Okay, the fabricator is assuming one gigahertz, but what’s my signal frequency? And engineers on the design team need to take more ownership of the process. Which laminate numbers would I use? My way of looking at it is I’d measure them myself. Not just on impedance coupons, but if I’m using a material on a consistent basis, I would bring in some of that laminate and measure the raw laminate myself. Methodologies exist to do that. Our Z-field product does that. And you can remove the uncertainties. You could say, “Well, look, I’ve measured it myself from zero to 20 gigahertz. Here is the Dk and Df profile for this particular laminate construction.”It’s all about removing the uncertainty in my mind—unless you’re making some boards in Penang, Malaysia, and other boards in a drier climate, and humidity is a factor, like Happy was highlighting. Unless that’s happening, there shouldn’t be a strong reason that one location would have systematically different Dk numbers than another one.To be honest, philosophically, I don’t understand why design teams will spend dozens of hours laying out a circuit board and owning the CAD layout, then maybe dozens more hours using up to $100,000 worth of EDA tools, to design and simulate designs. Why don’t the same people that are putting that kind of investment into the board layout and SI and PI simulation process pay more attention to the spinal cord of their design—which is the stackup—and why do they delegate that to the fabricators? That’s a design-philosophy question.
Shaughnessy: You mention TTM. Julie Ellis at TTM recently said, “How can any designer possibly know all of this stuff that I’ve been working with for 30 years, and I’m still learning?”
Hargin: I actually talked to Julie a week or two ago, and we were talking about pressed prepreg thickness calculations and percent copper. One of the gaps in the design flow is that for a lot of stackups, the fabricators are calculating pressed prepreg dielectric thicknesses based on round numbers. For example, for the percent copper on a specific layer, they might say that, signal layers are 35% copper, or they might say 40% copper, or 50%, whatever they calculated.The fabricator then might use those percentages all day for signal layers, and they might
have a number like 90% for plane layers. But your design has a different percent copper on the various signal layers. Some CAD tools know that. Valor NPI knows that and those numbers could be used in the stackup design flow to refine prepreg dielectric thicknesses and therefore get more accurate impedance numbers.I find that people in this space who I deal with every week get confused about manufacturing process variation. Let’s say impedance can be plus or minus 10%, okay? And that’s a common standard; some people target plus or minus 5% and pay a bit more with a more expensive fabricator for tighter control. But that manufacturing variation includes copper thickness variation, Dk variation, percent resin variation, and a thickness variation; all these things that add up to impedance variation.But you don’t want to give away some of that plus or minus 5% or 10% on your nominal values. If you don’t center your nominal, and you’re off on the nominal, that’s a terrible design practice. And people get confused about that. They think, “Oh, it’s only a 1% difference. Oh, it’s only two, two and a half percent different.” But those percentages are eating into your tolerance-tolerance band. So, there’s the manufacturing tolerance and then there’s the engineering tolerance.
Shaughnessy: It sounds like they’re making decisions based on these numbers that may or may not be accurate. I mean, every little tradeoff you do can affect something down the line. And like you said, you don’t know what the numbers are for each for a prototype shop, much less for overseas. So it sounds like a lot of it is kind of a crapshoot.
Hargin: I would say that an informal description of the design team’s job is to reduce uncertainty in their designs. If you have firstpass prototype success, and production is all done on schedule, you’ve reduced uncertainty down to an optimal level. I’m not really saying it’s a crapshoot; what I’m saying is, it’s our job as designers, engineers and design teams to reduce uncertainty wherever we can in our design flow. Just because I didn’t get bit by something on my last design doesn’t mean that I’m not going to get bit by the same issue on my next design.Glass-weave skew is an example of that. Just because it worked yesterday doesn’t mean it’s going to work tomorrow. That’s how my son thinks when he’s driving to and from college. He says, “I don’t get tickets.” Well, until he got a speeding ticket for going 20 over the speed limit. Anyway, I think we need to reduce uncertainty. That’s why I use the metaphor of getting the cattle back into your barn and keeping track of more things than you were keeping track of yesterday. Because when speeds increase, the margin of error decreases.Your margins decrease as speeds increase. You’ve got to be improving your design methodology; you can’t do what you were doing two years ago, you’ve got to be improving. That’s what engineers and designers are getting paid to do. If you’re starting to work on PCI Express 5, you can’t use the same techniques you were using with PCI Express 3.
Shaughnessy: Are standards helpful in doing the stackup?
Hargin: Yes and no. The IPC-4562 standard for copper foils is helpful in terms of copper thickness; it gives you a definition of copper thicknesses: how thick is half-ounce copper, and how thick is one-ounce copper. But it handles copper roughness, which is much more important these days, in a nebulous way. The definitions in 4562 about copper roughness are not universal, and they’re not current.Glass is defined in IPC standards, but each manufacturer can create their own wrinkle in how they implement that standard. Some companies like Nan Ya, where both Happy and I used to work, have their own glass manufacturing. But other laminate vendors source their glass from multiple sources. That could be a source of variation.
Holden: Does the Z-zero impedance analyzer replace the HyperLynx analyzer? How exactly is this used by Siemens EDA?
Hargin: We both use the HyperLynx 2D field solver. I’ve been using that from the get-go in an OEM relationship with Siemens EDA. The HyperLynx 2D field solver is also used by the HyperLynx SI and PI simulator, and it’s also used in Xpedition.The Z-zero software, Z-planner Enterprise, can send data to and from the stackup editor that Xpedition and HyperLynx use. In our environment, there’s a lot more detail. For example, in the materials library, we have about 150 materials, and probably another 10 materials, I would say, by the end of this year. We have a lot more granularity, as it relates to stackup and materials, and we send that data to and from Xpedition or HyperLynx.The other thing we do that’s a little different is that our tool realizes that a given production part number may have multiple versions of a stackup. Fabricator A, B, and C, version one, version two, etc. A lot of people think of impedance and stackups as kind of a monolithic concept, but they’re pretty nuanced, in fact.
Shaughnessy: Bill, is there anything we haven’t covered that you’d like to mention?
Hargin: Just that stackup design has been a passion of mine for a long time. The seeds of Z-zero were planted in my mind back in my HyperLynx days. And I thought that a tool should exist that does what my Z-planner Enterprise product does. I thought that if speeds kept increasing, engineers would need to have a tool that handles the granular details of stackups. And that’s the journey that I’ve been on with Z-zero.We spoke with you a few months ago about Siemens EDA distributing my Z-planner Enterprise product. As Max Clark and I mentioned at the time, we realized we were both pursuing the same goal: To take manufacturing knowledge and move it to the left in the design process with an EDA tool that would handle all the uncertainties in stackup design.
Shaughnessy: Well, congratulations, Bill. Thanks for speaking with us today.
Hargin: Thank you, Andy, and Happy, it’s always great talking with you.
Holden: Thanks, Bill