The winner of the DesignCon 2020 award was Istvan Novak, Principal Signal and Power Integrity Engineer at Samtec. Aside from his impressive career as an engineer, Novak has been a long time participant at DesignCon. In the following interview, Design News talks with Novak about past and future trends in Signal Integrity (SI) and Power Integrity (PI), as well as what the award means to him.
Design News: How has the signal integrity (SI) industry and technology changed in the last 20 years?
Istvan Novak: One change is the tremendous increase in data transfers, which has meant faster CPU clock speeds, data rates and cable signals. The bandwidth requirement is also exploding. Perhaps more importantly, the semiconductor industry has continued to change. Twenty years ago, Moore’s Law was in full force. Now it is significantly slowing down. I won’t call it dead yet. We still see improvement. However, the demand for bandwidth did not slow down. Instead, it actually continued to go up in terms of percentage rate. People have found other ways of satisfying increasing bandwidth. One example is by using different modulation techniques for to improve signal integrity such as pulse amplitude modulation (PAM) for stage 4. This technique is now excepted at the very high end of the data speed range. PAM-4 is a modulation technique whereby 4 distinct pulse amplitudes are used to convey the information.
Image Source: Keysight / PAM-4
Design News: Have power integrity (PI) challenges been addressed at the same pace as ones for signal integrity (SI)?
Istvan Novak: There have been major differences between SI and PI over the last 20 years. SI issues evolved from the electromagnetic compatibility (EMC) discipline sometime in late 80s to early 90s. In contrast, PI didn’t become an issue until 10 to 15 years later. What this tells us is that SI has had more time to mature. Today, there are a lot of SI standards and guidance available for high-speed signal designers. This is in very stark contrast to PI designs. Today, PI is probably in the same state or maturity level as SI was in the mid- to late-90s.
I have second thoughts as to whether PI will ever be as developed, as mature, as SI. The reason is that PI is definitely more complex than SI, which has come a long way. Back in the early 90s, high-speed signal integrity design was considered black magic. The problem was that people didn’t approach things systematically to understand what happens and why. Once the underlying physics were understood, the subject was no longer considered black magic. Technical works like Howard Johnson's very popular book on Signal Integrity helped a great deal.
The reason why we may not expect our understanding of PI to even catch up to where SI is today is that the PI problem definition is far more convoluted. In other words, it cannot be isolated in such a nice way as it was for SI.
Design News: Why is that the case? Why is PI more complex, more convoluted than SI?
Istvan Novak: For high-speed data signal integrity, we don’t typically worry about what happens to the signal when it goes into the silicon (chip). Most of the signal measurement and other checking for SI happens in the passive interconnects like cables, sockets and interfaces. And the passive interconnects can be treated in isolation from the rest of the system. They can be treated in segments.
It’s not that easy to break PI into smaller, easier to handle pieces in order to understand it. Consider a high-power and high-speed CPU. If you turn over a chip or the board, you’ll see a number of high-speed I/O pins with one dedicated return pin. You’ll also have power pins that are not very isolated or as numerous. Further, the ground plane is not separated between the high current and the high-speed sections. We like to keep the power planes intact. With high frequencies and high-speed signals, we don’t necessarily need to consider how the return signals spread over a large distance or plane because they will not.
This again is in stark contrast to power integrity (PI), where power related noise has a way bigger range of interaction from high frequencies all the way down to DC. This noise may interfere with the high-speed signal. My point is that we cannot easily, physically separate blocks of power. That is probably one of the reasons why standards or any general guidance are harder to come by for PI design.
Design News: Will improvements in models and simulation tools help us to better model and simulate more complex PI issues?
Istvan Novak: I believe it absolutely will help. But I’m not sure if underlying limitations will change because designs are driven by competitiveness. Would it be possible to design a system were the power can be compartmentalized and segmented very nicely? Yes, it is technically doable. It would be just much more complex. While it might be easier to design systems with segmented power elements, it would also require bigger, probably heavier and more numerous components. This is why is wouldn’t be competitive in the marketplace today.
This is why people try to push the limit on PI design, to allow all of these interactions I mentioned earlier to take place. Using detailed simulation and measurements, these detailed interactions can be taken into account. But it takes a lot of time to simulate and measure these interactions.
Design News: What activities are critical to ensure designs meet both SI and PI requirements?
Istvan Novak: Simulation and measurement validation are critical design activities. But equally important is understanding. These three activities are needed to ensure a good design and engineered approach. However, very often we may see people trusting tools blindly. They assume that the tools will give the correct answer to what they thought they asked from the tool. Such a believe can be extremely misleading. Even developers with experience and lots of support from the tool developers can be misled. It takes a lot of time to get a decent correlation to the measurement. But even if the correlations are good we must ask ourselves: “Do we understand why all that is happening? Can we explain to ourselves?”
It is true that, when I start my car and drive it, I don’t necessarily need to understand how the car is working on the inside. That is because the details have been worked out by the automotive designers. But the people who designed the car still need a very good understanding of what is happening and why.
Design News: In this age of reference designs and intellectual property, some would argue that designers don’t need to understand things like SI and PI for their designs to work.
Istvan Novak: The world is going toward simplifying the process in which, for instance, a big vendor marketing a new chip may say; “You don’t need to do SI. Here is a reference design - just copy it and you will be happy.” And that is a working solution, if the reference design is done correctly. If so, it saves a lot of time as people don’t have to recreate all of the groundwork to understand the SI and PI details. However, anyone who relies on this solution for multiple generations of system design should agree that it carries a big risk.
First of all, the users who rely too heavily on reference designs may become depended or at the mercy of the silicon vendor. If anything goes wrong such as a slight change in design or requirements or even an honest mistake, then not understanding SI and PI design might cause big problems. Companies that lack or don’t want to invest in design resources and rely heavily on reference designs are creating a very strong dependence on others.
Again, I’m not against it, I’m just pointing it out that we need to understand that a dependency can be hard to break.
Design News: You’ve been an active participant at DesignCon for many years. How has DesignCon changed in the last 20 years? Where might it or should it go in the future?
Istvan Novak: I remember that the very first predecessor to DesignCon was a marketing roadshow from HP. In the early 90s, HP trade show busses with their equipment would travel around. As you probably remember, HP broke up at the end of the millennium into a computer part (HP) and an instrumentation part (Agilent). Later still, the instrumentation part became known as Keysight.
I still remember one such roadshow which was turning into a conference when they began to invited consultants and people who used their equipment to make speeches, give lectures and present technical papers. And that is how we started.
Back then, the main high-speed data transmission topics were cross-talk, simultaneous switching noise, reflections, matching. Today, SI topics are focused on glass weaves and challenges between passive interconnect and the silicon.
Design News: What do you mean by glass weaves?
Istvan Novak: A printed circuit board (PCB) laminate has a glass fabric embedded in it. The glass fabric is a woven structure like the fabric in our clothes. A PCB is a 3D structure which is emerged in epoxy resin and different epoxy materials will behave differently electrically. This problem first started to become notices in the early 2000s. It is still one of the major topic still ongoing it these days.
Design News: So, what does the future hold?
Istvan Novak: In SI, people would still like faster and faster interconnects and more and more bandwidth. Both can definitely be done in a number of different ways. But each solution has its own limits.
Another trend is the replacement of copper (Cu) PCB connections and board traces with optical elements. But the question of when optics will replace Cu connections has been going on since the mid-90s. The question seems to be asked every three year and the answer has been that the Cu signaling connects have continued to improve – pushing out the need for optical. The improvements have gone hand in hand with the improvement of the silicon. Better silicon can handle more degradation on the passive channel, so we don’t need to switch over to the optics yet.
There is another reason why the switchover to optics is not happening. If I just look at the transmission medium, then maybe optical connections might be a better choice. The problem is that making the translation from the electrical domain to the optical domain takes a lot of power. Today’s big data communication servers dissipate kilowatts of power. Switching over to an optical interconnect would require a significant amount of power per channel on hundreds or maybe thousands of connections. We cannot cool the servers any further and that is why the search for a better interconnect will continue.
Another trend that will continue is the replacement of portions of the printed circuit boards with cables, whether it is Cu or optical cables. And in parallel to this trend will be the gradual introduction to Cu cables and optical interconnects inside of our system boxes.
Design News: What do you mean by Cu interconnects in the chassis?
Istvan Novak: In the traditional set up, the Cu cables are outside of the box, e.g., a PC or server. Inside the box is the PCB which is used among other things to carry the signals. But the PCB are getting to the end of their capabilities unless some significant technological changes happen. This may result in more cables inside the chasses be it Cu or optical cables.
Design News: Let return to any future trends in SI and PI.
Istvan Novak: Additionally, the sophistication of silicon signal processing will be increased. Twenty-five years ago, there were the telephone modems for Internet dial up that ran at 56 Kbs. The digital signal processors at the time had to run at a much higher clock rate compared to the modems, which wasn’t a problem as the CPU clock speeds were already very high.
Now I see the need for more and more processing capabilities, probably inside the silicon.
One the PI challenge, I want to be optimistic that progress will be made but I’m not sure. I hope at least some baby steps will happen with PI.
Finally, AI seems to be coming. It would definitely change the we are getting to the point where the human brain may not be able to handle the big data designs that are being created. AI might be able to help.
Design News: You have been selected as the Engineer-of-the- Year for DesignCon 2020. Why is that an important award?
Istvan Novak:. What I’ve noticed from my many years participating in DesignCon and its predecessor it that, even though this is a worldwide conference, DesignCon has become like a family. It’s a place were very good experts mostly from the industries but also from academia have come together. I am very much pleased that this community chose me as engineer of the year. But it’s more than that. As my friend Albert-Laszlo Barabasi shows in his book (“The Formula”), once you have excellence in any given space, the selection of a winner is almost arbitrary. So, I offer my congratulations to all of the finalist. They are all excellent and great and, in this regard, any one of them could have easily won the title. This is how I feel.
John Blyler is a Design News senior editor, covering the electronics and advanced manufacturing spaces. With a BS in Engineering Physics and an MS in Electrical Engineering, he has years of hardware-software-network systems experience as an editor and engineer within the advanced manufacturing, IoT and semiconductor industries. John has co-authored books related to system engineering and electronics for IEEE, Wiley, and Elsevier.