Wild River ISI-56 Platform Accelerates SerDes Testing
I recently spoke with Al Neves, founder and CTO of Wild River Technology, about the release of their new ISI-56 loss modeling platform. Al explains why it was so critical that this tool meets the stringent requirements of the IEEE P370 specification (which he helped develop), and why he believes this is currently the best tool for SerDes testing and characterization.
Andy Shaughnessy: How’s it going, Al? Why don’t you tell us about this new ISI-56 modeling platform?
Al Neves: It’s going really well, Andy. We started making test fixtures for SerDes characterization for both compliant and characterization testing about 11 years ago. The focus was from 10 to 32G NRZ (non-return to zero). It’s either high or low, and that business went pretty well for us. Then PAM4, which is four levels and not two, came out, and in the last two or three years we’ve seen a big rush to 56G PAM4, and we have scores of existing customers asking us about 112G, even a few at 224G, PAM-6! Now, the issue is that there is a 9.0406 dB penalty versus NRZ because of the multiple levels. When you have multiple levels, your test fixtures need to be better by almost 10dB. When that happened, that business went through the roof. We couldn’t keep those products in stock. So, we focused on that and we slightly improved the ISI-32, which was for 32 G NRZ, and we’ve just released the ISI-56. It’s extremely good signal integrity; the product looks simple, but it isn’t.
Our company ran for a pretty good period without having any competitors. We have some competitors now, and as I was doing some analysis this morning, I see we’re exactly twice as good as our closest competitor, Keysight Technologies. Having said that, our product meets the highest signal integrity metrics for the new specification IEEE P370, and the Keysight test fixture does not. My comment is not to denigrate Keysight, they make very good equipment, just not so good test fixtures.
Shaughnessy: Why don’t you talk about the significance of IEEE P370?
Neves: One of the problems that plagued test fixtures and products was the lack of standardization and compliance agreement for signal integrity. There are two key indicators of good signal integrity: return loss and TDR impedance. Return loss is measured with a VNA; TDR impedance is measured with a TDR and then you do an impedance calculation. Having said that, there really was no standardization, and IEEE P370 changed that recently with a compliance specification. There are three tiers of signal integrity quality. Our test fixtures meet the highest tier, and they’ve met that tier for about nine years now, before the spec was even out. So, we’re pretty fanatical about our signal integrity, almost to a fault!
Having said that, the new requirements for these high-speed multiple-level PAM4 are pretty extreme. You’re trying to mimic a backplane and establish how long the SerDes can run a TX to RX at a certain bit error rate. If your signal integrity is not good, you’re not testing SerDes electrical length reach; you’re testing your test fixture signal integrity. Think about it in terms of the Hubble telescope. As long as the telescope was earthbound, you had to deal with the atmosphere and the atmosphere obscured the measurement. Poor signal integrity does the same thing. It disperses the signals you want and adds stuff that you don’t want, and you can’t accurately characterize SerDes if you have poor signal integrity.
I’ll give you an example of this, and this is published so I’m not violating any confidentiality agreement. We wrote a paper with Xilinx several years ago on the ISI-32, which we still sell. The signal integrity isn’t as good as the ISI-56, but we were able to run the 32G SerDes to 47dB of loss. They had never seen that before. So, their VP runs into the lab and says, “Gee, is that actually 47dB? We haven’t seen it before!” And I said, “Because the test fixtures you’ve been using are junk.” They laughed that nervous laughter. I said, “No, it really wasn’t a joke. Your test fixtures are junk. Now you have a good test fixture and now you’re seeing the true performance of your SerDes because you don’t have reflections and resonance and all the nasty stuff from junk test fixtures.” That was eye opening.
What’s happened is the IEEE P370 is a specification for 50 gigahertz. We’re working on 70 gigahertz right now. We almost have that done, and we’re going to have a plethora of products coming out in the 70-gigahertz space. So, we’re going to be achieving 70-gigahertz signal integrity performance. Most test fixtures have issues up around 26 gigahertz. If you’re interested, I have some graphs and we have some comparisons. It was a seminal paper on doing systematic SerDes characterization based upon design of experiments.
Shaughnessy: I’m curious. You’re a small company; how did you guys get to this point where you’ve got a tool that’s better than the big guys? How are you all able to do that?
Neves: Actually, it’s based on a stubborn and uncompromising mindset: What’s your objective? Is your objective to build something that you think is marketable, or is it your objective to create the best signal integrity you possibly can and not care what the market wants at that time?
John Trudel is a good friend of mine and he once told me something very interesting. He’s a retired, brilliant guy who started Tektronix Labs and the whole TDR idea. John said, “Don’t listen to your customers.” I said, “No? Tell me why. You’re always supposed to listen to your customers.” John said, “Your customers will tell you what they’re struggling with today. You don’t want to design what they’re struggling with today. You want to guess at what they’re going to be struggling with three, five, or 10 years from now and build something for that market.”
So, once you start listening to your customers and you give them something, by the time you give it to them, it’s an obsolete product and it’s a different problem. You have to predict where the market’s going, and you maybe want to just do something that the market’s not ready to embrace it. When we first came out with these ISI platforms and our other test fixtures, the market wasn’t ready for it. They didn’t need it. Most people were doing 10G, maybe 20G, and we didn’t sell that many fixtures; We were ready to do 32G. We now have had requests this past year for 224 gigabits per second, over 100 gigahertz! We’re tied into the connector companies, using their products at beta level to design it in. We’re working with teams of people. For example, Samtec introduced a new 2.4-mm connector this last year. Now, the 2.4-mm connector has been in the market a long time, but Samtec made a better one. I looked at it and I said, “It’s better than what I’ve seen in the market, let’s transition to that solution.”
We used it and that’s the trick about how we’ve improved our signal integrity over other folks. They just grab what’s been available in the marketplace and what their purchasing people say is available for them to use. I’m not constrained by that. Then we used very extreme simulation methodology using HFSS and hundreds of hours of simulation time. There were some other proprietary tricks we used. The test fixtures look rather simple, but they are not. If we could all hit a 90-mile-an-hour fastball, baseball players wouldn’t make $15 million a year. It’s not easy doing that. Having said that, the test fixtures are of uncompromising quality and this ISI-56 is an example of that.
Shaughnessy: Right, it’s transparent.
Neves: Exactly, transparency versus opacity.
Shaughnessy: I remember Dr. Bruce Archambeault saying, “All of the connector companies will say, ‘These connectors are totally transparent.’” But he said, “There’s never been a connector ever made that was completely transparent.”
Neves: The issue I have with the connector companies is that you have two products: you have good products and bad products, but what you have for both of those is they generally don’t know how to use their own products (laughs). If you rely on most of the connector companies to help you do your electrical optimization into the PCB, you end up with a horrible result. There are exceptions to that. Molex is pretty good, and Samtec is very good, but most of the other connector companies relegate the customer optimization to junior engineers that don’t have the experience or time to get it right. I’ll deal with a connector company and they’ll throw some junior engineer who’s just learned HFSS, and I look at the result and I think, “You haven’t even begun to understand this problem. That poor young guy, or gal, is suffering under a lack of technical leadership.” Most of the connector companies are befuddled with that right now, and the encrypted HFSS model approach with poor stackup design, along with not setting up HFSS correctly is killing them.
I know that my statements are pretty strong, but when you think about this, we’re trying to achieve stellar signal integrity past 70 gigahertz. It’s a very difficult game. Everything has to be done perfectly, and you have to understand the problem. It’s not a compromising or easy space to be in.
Shaughnessy: It’s good to tell it like it is. So, who is the typical audience for the ISI-56? Who is typically going to say, “I’ve got to have this?” Is it for any specific segment?
Neves: Really, it’s anybody who’s developing SerDes and characterizing SerDes, both at the silicon level and the intellectual property level. Synopsys is a great customer, and Cisco, NVIDIA, etc. But it’s also OEMs, SerDes manufacturers, people who are buying SerDes to evaluate them, and the incoming quality control folks. If they’re buying a SerDes, they can use our test platforms to verify, not just whether it meets the performance, but also to say, “What’s the true performance of it?”
Shaughnessy: Right. SerDes seems to be a big thing. Barry Olney writes about SerDes in his column sometimes and he’ll get thousands of reads right away. Everybody’s hungry for information about SerDes.
Neves: That’s the fundamental of the physical level. That’s where the actual 1s and 0s, the actual data is occurring—at its physical level. That’s fundamental to the performance metric of the system.
Shaughnessy: So, for the ISI-56, is there a learning curve for this? Can a signal integrity engineer just pretty much jump on it and hit the ground running?
Neves: We have measured S-parameters that are ready to be simulated. I run a one-hour training class. What I like about the product and our test fixtures is that we throw out a lot of energy on the front end in an integrated solution. We have a Keysight ADS simulation kit that we give our customers. We have measured S-parameters so they can run that in MATLAB. We have a comprehensive user’s manual. We have graphs. Recently we came up with COM metrics for all the channels and we provided that as a report. We have COM metrics for 32G NRZ, 56G PAM4. COM is a figure of merit for those data rates, so that’s really helpful. We also have good application support. The WRT product training that we do really helps our customers. We don’t really get a lot of application issues after the training, so there’s no real learning curve.
Shaughnessy: That’s great. Is there anything you want to add? Anything we haven’t covered?
Neves: When you’re deciding on a solution, there are a host of things that are really important. The homogeneity of the platform is important. The materials that you select when you build your test fixture—that’s called weave—there’s been a lot of discussion on that. We talked about launch integrity, which is measured by TDR impedance. The return loss is measured by a VNA and what that return loss looks like. Again, there are two other solutions on the market right now. One is a USB solution that I would consider poor to very poor signal integrity. It’s magnetically introduced ISI, and I don’t really know how the industry is using that platform effectively because the signal integrity is poor. It’s convenient, it’s USB-controlled, and you can control the channel loss with a USB computer, but it doesn’t relate to how a backplane works with dielectric and skin effect losses.
The Keysight platform was released fairly recently, and our test platform is twice as good if you compare the S-parameters that they published versus the ones that we published. Their platform is probably good out to about 10, 12 gigabits per second, certainly not 56G PAM4 as they claim. You just have to look at the power spectral density and the Nyquist. I don’t want to get too technical with that, but these platforms either meet the requirements or they don’t. It’s not really that complicated.
Shaughnessy: That’s great. So, what’s next? What are you guys working on for 2021?
Neves: We’re working on the complete family of 70-gigahertz test fixtures. Then there’s a company working on 1.35-mm connectors up to 90 gigahertz, and we’ll be working on that Q4 2021. Then we’re going to work on some test fixtures that torture the SerDes in some very novel ways, introducing mode issues and resonance. We’re coming up with a whole series of products to do that in Q2 of 2021. We’re getting calls right now, as I mentioned earlier, from our customer base going up to 224 G. The next thing that’s happening in the industry is that mil-aero is coming on strong. They’re trying to do PCIe gen 4 and the military’s problem is pretty intense because they have to build Class 2- and 3-compliant systems. Their signal integrity is not as demanding as the commercial, but let me give you an example. With some of the stuff that my company makes, if you put it in a tank and a tank round went off, some of the compression mounted connectors might not work. Their problem is pretty challenging, actually. We’re getting a lot of business right now from the mil-aero folks because their signal integrity requirements have gotten pretty extreme in the last year or two. They’re going up to 32 G, and they’ve been at 10 G with issues, so this is a big step up for them.
Shaughnessy: That’s a good segment to be in. The checks don’t bounce from them.
Neves: It’s the circle of life, right? My company’s doing well, so I pay too much in taxes, but that money goes to the federal government. The federal government then pays mil-aerospace companies and then the mil-aerospace companies give me that money back again. It’s the circle of life (laughs)!
Shaughnessy: That’s a great point to end on! Thanks for talking with me, Al.
Neves: It’s always a pleasure talking with you, Andy. I hope we can get together next year.