Eric Bogatin: “The crosstalk problems can limit what PAM-4 can do”

Eric Bogatin: “The crosstalk problems can limit what PAM-4 can do” post thumbnail image

During DesignCon 2018, Sierra interviewed the Signal Integrity Evangelist Eric Bogatin. We could not resist to ask him about his first DesignCon. “You know, I looked back at all the DesignCons that I’ve been to, and I think it was in 1992, he said. It was back in the old days when it was the HP High-Speed Digital Symposium, so it’s been a long time.” Listen to this PCB expert’s thoughts on technology, crosstalk, PAM-4, educating the new class of PCB designers, and the signal integrity market.

0:24 What are your thoughts on technology?

This idea that technology keeps changing and evolving every year, I think that is one of the most important messages for engineers to grasp and get a feel for. I spent a little time teaching at the University of Colorado in Boulder and I see many students that feel that they have to stay in school more and they have to get a Masters. They have to get a PhD because they want to learn more. I keep telling them that learning never stops. The most important thing they need is the ability to pick up things on their own, because if you just stop learning when you get your degree, your knowledge has maybe a three-year time constant. At the end of the three or four years, you are not going to be current anymore. You have to be learning new every year.

I think that is one of the reasons why it is so important to come to conferences like DesignCon. I read an article for our Signal Integrity Journal and it was drinking from the fire hose again. Coming to a conference like DesignCon is like drinking from the fire hose because there is so much information, but it is part of keeping current in the new things that are happening.

1:36 How do we educate the new class of PCB designers?

The whole problem with training engineers for being able to jump into a new job and be effective immediately is a tough one in all aspects of engineering. Part of it is because in many university programs you get an undergraduate degree and you may have some good understanding of how to do derivations in the textbook but not how to apply it. Oftentimes, you don’t have a lot of that hands-on experience.

I think that we should expect universities to provide some basic foundation of engineering, like what is the difference between voltage and current and about impedance and about thinking about the behavior of circuits and being able to do simulations. Layout and hardware design is one of those areas that the more experience an engineer has with that as an undergraduate doing projects.

I see a huge movement in the maker community, of folks who do not have a double A degree, but are dabbling in building electronic components or are even building circuit boards, getting that hands-on experience. If they come into an undergraduate degree program, they are so much more effective at learning the material and using it immediately. In the maker community, there is the phrase that says “Learn by doing.” I do not think that is quite the right phrase. I do not think you learn by doing, I think you understand by doing. At University of Colorado, one of our themes is hands-on training. It is not to teach by doing hands-on, it is to understand by doing hands-on. You learn the principles. You learn the techniques and methods. Then, when you are building something or you are putting something together, that is when it all clicks. That is when you see how to use what you have learned. That is when you gain that understanding. I think that the hands-on opportunities are really to increase level of understanding.

Unfortunately, not very many universities have electronics programs with that opportunity for hands-on other than a few labs. It is very project based, project oriented, where you get that real opportunity. This whole idea of teaching circuit design, and how do you translate a napkin sketch into a physical board you hold in your hand, is a skill that very few universities teach. It happens that at University of Colorado in Boulder, just last year, we started a class that does exactly that. It is an undergraduate and graduate mixed class, specific, practical, printed circuit board manufacture and design. It is specifically to teach students how to go from a napkin sketch of the schematic to holding board assembled and working in your hand. We walk them through how to develop the schematic, how to source parts, how to do the connectivity of the board. Then, we show them that interconnects on a circuit board are not transparent. If all you are doing is connecting the dots and making the wires connected, many times it is not going to work, because of the electrical properties of the interconnects. That whole idea is a tough thing to teach, because it is about the practical application of the stuff you learn in the textbook to the physical design of interconnects.

I wish there was a magic way of instantly training engineers to do that. Ultimately, it comes from giving them some of the principles, giving them the opportunity to experiment, make mistakes, build a lot of boards and then to be able to hold something in their hands, to go through it a couple times to see what works and does not work. I hope that one of the ways that our students, and I talked to many engineers here at DesignCon, I hope that one of the ways that they gain that expertise is watching some of the videos that I have done, reading some of the textbooks that I have written, reading some of the papers I have written, because all of that is part of that process of moving up the learning curve and then using that information they have learned in real world application to really understand it.

5:58 Are your videos available to the open media?

Over the years, I have taught lots of classes in signal integrity around the world. We recorded a lot of them. We posted all of them on our Signal Integrity Academy, so it is located at the website. We have got maybe 160, 170 hours worth of video training there. Maybe of that, there is about 40 hours of content that is available for anybody. You can go through the website You can browse through the six different courses that we have, each one with little pieces of video lessons in them. Maybe about a third or a quarter of them are available for anybody to view. The rest are available for subscription. We have a number of large companies that have a corporate subscription, so everybody in their company has access to it.

I would say one of the very thrilling things for me and very heart-warming things for me as I walk around DesignCon, I run into so many engineers that have watched our videos and say, “Oh it’s so good to see you in person!” I keep saying, “Aren’t you sick and tired of listening to me?” No, they love the videos! I know that there are a lot of engineers that have benefited from the videos.

7:11 Learning vs Understanding?

We found that it is not about learning. That is the beginning part, but it is the understanding is how you apply what you have learned. That comes from real world practice. That is where the hands-on comes in. They are trying some designs and thinking about the principles as you are doing that design, thinking about how they apply is so valuable. Then, you see what works and does not work. You begin to see how to use that learning that you have got in real world applications.

Eric Bogatin DesignCon 2018

“When you can marry what you learn in the classroom with what you learn in the real world, you become the most powerful engineer.”

I think that is the way the part of the revolution is happening these days, especially in the maker community, that combination of the learning part, the textbook… You go through the formal part of it. You understand the math. It is when you pull it all together that you understand the big picture. You understand how to use that formalism of electrical engineering into building real products. What works and does not. It is not that the textbook does not apply to the real world, it absolutely does. You have to learn how to apply it to the real world. When you can marry what you learn in the classroom and the textbook with what you learn in the real world, then you become the most powerful engineer and you are able to accomplish the most.

8:26 What are the benefits of attending DesignCon?

I keep saying this to everybody that I run into at DesignCon, that one of the real values of DesignCon rarely does you find in one environment, one place, all of these companies and also all of these training tutorials and boot camps. The article I wrote for Signal Integrity Journal is Drinking From A Fire Hose, because it is drinking from a fire hose. Rarely do you have that concentrated opportunity in a two or three-day period to come and have so much information available to you to take advantage of.

One of the themes at this year’s DesignCon, I have seen before, but it has not been as strong as here, is the number of training camps and boot camps and tutorials from some of the vendors even. They have all day tutorials on how to use scopes, how to use network analyzers, how to do probing of EMI, how to do a simulation for a high-speed serial link. These are real fundamental principles to understand and practical information to understand that many of these new engineers are not getting in school and they are available for free in a lot of these vendor-supplied workshops. Every workshop I went to… I was shocked. Every single workshop, the room was packed. There were at least 100 people in each of the vendor tutorials.

I think this has been a really valuable contribution for the DesignCon organizing committee to put in place the variety of training tutorials available for new engineers. It is an incredibly valuable opportunity and most of them are for free. All you need is an expo pass in order to get in them.

10:12 How will speed and data rates impact the signal integrity market?

A couple years ago, at one of the DesignCons, I saw that there was a lot more discussion about PAM-4 for higher speed. I wrote an article for, I think it was EDN, for Martin Rowe, looking into my crystal ball. I made a prediction there, which I am still looking around to see how well it is going to come true. Everybody talks about that next data rate.

I remember the days when it was before we were at one gigabit per second and everybody was saying, “Oh, how can you do one gigabit? You got to do optical for one gigabit. Cannot do it…” and then, of course, we got there. Then we said, “How are you going to do 2.5 gigabits? You cannot do it with copper. You got to do optical.” People were talking about optical backplanes and circuit boards. Then we were there and everybody said, “Oh, five gigabits. You cannot do five.” And then each step, then it was ten. Who can do ten gigabits per second? You got to do optical. Every generation, it is always that question of when is optical the right one?

Now, you cannot do 25 with NRZ, you have to do PAM-4. Then we got to 25 NRZ. Now it is, you cannot do 56 with NRZ, you got to do PAM-4. I think there is a potential problem in transition to PAM-4 and why it is not necessarily the best solution to get the highest data rate. Here is the fundamental problem with PAM-4. PAM-4, you have taken the voltage range and now you have split it up into four bits, so you have three different eyes. That means that the voltage level for each eye is a third. That means the power level for the aggressor bit is three times the lowest level. That means automatically, you get ten times worse signal-to-noise ratio for crosstalk. That means that the kind of crosstalk level you can tolerate… If you are at -35 DB and you need 20 DB below that as the absolute worst case… That is where your noise is going to be. Then on top of that, we are on a 10 DB margin, you are talking about -60, -70 DB isolation between this aggressor channel and the poor victim channel over here. Keeping that level of isolation on the board… Maybe you can do that, you pull the traces apart, but doing that in the package and the BGA escape region under the package, I think is a really serious problem.

When you are doing one channel at PAM-4 – I think you can make one channel at PAM-4 work, maybe two or three – but when you are doing 50 channels PAM-4 and all those signals are coming out under the BGA, I think the crosstalk problems potentially can limit what PAM-4 can do. As you start implementing these large systems with PAM-4, I think this problem is going to raise its head. Maybe we might be rethinking NRZ or switching to optical or cabled backplanes, like what Samtec and Molex and FCI are doing.

13:13 What is the limit to copper interconnects on circuit boards?

I have talked to many folks and thought a lot about what is the limit to copper interconnects on circuit boards? It is not so much… Everybody says, “You got to use better material.” There are some materials out there that are so good low-loss, that they are almost transparent. It is not the dielectric materials that are limiting performance. It is the copper connectivity. There, it is all about the line width. Because the conductor loss is fundamentally how high a data rate you can go, and how far distance, and it is party also dependent on what that equalization technology is.

Everybody is thinking that -35 DB may be the highest practical amount you can compensate for equalization. It is really easy to put in the numbers for how far can you go at -35 DB attenuation for given line width before you run out of signal at the other end. That is something on the order of if you the 30 inches or so, and depending on the line width, that is something on the order of about 50, 60 gigabits per second. I think it is possible to go, pushing the limits, to 50, 56 gigabits per second NRZ in copper, but it means using a little wider lines.

That means that you’re going to have to think about re-architecting a circuit board, so you build it like a highway system, the interconnects, the street system. You have wide traces that are going to have low loss for long distances. Then you have the narrower traces that go board to board in the fabric for the short distances. You partition your layers with a wide conductor with thick dielectric for the controlled impedance, and you have a few of those long-haul super highways. Then you have multiple layers that have the narrower line, higher innerconnectancy for the short distances. You engineer the stack-up for attenuation control and for how far each run is going to be on that layer. That is how you get the most out of the circuit board.

I think that approach will get us to the 56-gigabit backplane kind of range. Beyond that, is where I think cable interconnects come in. That is the approach Samtec has been pushing that with their flyover technology and they use literally cable because the circumference of the conductor of the cross section is wider than what you need on the final lines on the board. I think there is a long future for circuit board technology and transitioning looking ahead at cable technology.

15:52 How can the capacitors function very well at 50 gigahertz?

Well, I do not think there are too many components that operate well at 50 gigahertz, but I do not think we need very many. I think there is a lot of work with some of the capacitor vendors here at DesignCon showing off capacitors and the mounting features that will do DC blocking into the 50 gigabits per second kind of range. The connectors are always a challenge, but I am so impressed with what connector com- like Samtec and Molex can do. They do phenomenal jobs with their connectors. It is always the vias that are the weak link in engineering transparent vias. With good design tools and watching out for the stub, we can engineer controlled impedance and transparent vias.

Then it is the packages. They are also part of the weak link, but I think it is possible to engineer transparent or acceptable packages in the 50 gigabits per second kind of regime. I think there are challenges, but it requires just good design engineering to accomplish those.

Watch our DesignCon 2018 videos:

Tags: , , , ,

Leave a Reply

Your email address will not be published.