Design Predictable PCBs through SIMBEOR® Electromagnetic Analysis

Zachariah Peterson
|  Created: July 24, 2022
Design Predictable PCBs through Simberian's SIMBEOR® Electromagnetic Signal Integrity Software

Making electromagnetic analysis accessible to anyone in the industry is what inspired Yuriy to create SIMBEOR® Electromagnetic Signal Simulator.

Yuriy Shlepnev is the founder and president of Simberian. In this episode, he will tell us about Simbeor simulation capabilities, and briefly educate us on rise times, signal integrity, and solving EMI. He will also show us the simulator in action and how it can be a lifesaver to PCB designers like you.

Tune in, enjoy and be sure to check the additional resources below.

Listen to Podcast:

Download this episode (right-click and save)

Watch the Video:

Show Highlights:

  • Yuriy talks about his background in computational electromagnetics and how he got started working on simulations for PCBs and magnetic analysis 
    • He realized that the byte rates and data rates were increasing, and everything was getting into the microwave domain
    • In 2006,  Simberian was founded as a project to make electromagnetic analysis accessible to anyone in the industry
    • SIMBEOR® was created to make electromagnetic analysis mainstream
  • Simulation capabilities accessible to lower data rates, Yuriy discusses rise times, signal integrity, and solving EMI simultaneously
  • SIMBEOR® 2022 includes three simulation modes, and one of them is the Fast SI which uses more approximate models for pins, pads, and vias. It is a full wave, but not 3D. It allows simulations to run faster for more mass, included
  • Simbeor SDK, a built-in software development kit in Altium Designer’s stackup manager. It uses the SFS solver for cross-sectional analysis, and that benefits Altium customers
  • Yuriy shows us the 3D field solver in action
  • Simulation of fiber weave effect on PCBs, is it possible? 
    • Coming soon in Simbeor SDK is the ability to build your own simulation tools
    • Monte Carlo analysis is a perfect way to convert numeric model variations into a probability distribution
  • What is the future of SIMBEOR® that designers can use to help them expedite important analyses for high-speed and RF designs?

Links and Resources:

Connect with Yuriy Shlepnev on LinkedIn
Visit Simberian website and learn more about SIMBEOR® Electromagnetic Signal Integrity Software
Read Yuriy’s Articles in SI Journal
Watch a related podcast episode: Simberian’s 3D Field Solver in Altium Designer

Full OnTrack Podcast Library
Altium Website


Get Your First Month of Altium Designer® for FREE

Transcript:

Yuriy Shlepnev: Finally, a couple of our customers approached us and said, "Well, why don't you kind of..." Not just explain what is possible to simulate and what is not, but you make it clear with some tools. And we built Simbeor SI compliance analyser with electrical.

Zach Peterson: Hello, everyone. And welcome to the Altium on track podcast. I am your host, Zach Peterson, and I will be talking today with Yuriy Schlepnev founder and president of Simberian. And father of the very well known Simbeor electromagnetic signal simulator. This is a very popular or well-known simulator if you are someone who frequents an event like design con. And I think it's going to be very interesting to talk to him about the newest version of Simbeor. As well as the place of these tools within the PCB design process. Yuriy, thank you so much for joining us today.

Yuriy Shlepnev: Zachariah. Thank you.

Zach Peterson: Yes, absolutely. And I would like anyone who is watching on the YouTube channel or watching on the LTM website to make sure that you check out the show notes. There will be some very useful links to where you can learn about this simulation program, as well as connect with Yuriy. Yuriy, what I usually like to do with folks who I find very interesting, including people like yourself who I've actually cited in my own work is to learn a bit more about your background and how you got into the electronics industry and how you got into working on simulation for PCBs.

Yuriy Shlepnev: Well, I have background in computational electromagnetics. My PhD was in this area, and application was microwave integrated circuits. But in around 2000 I switched to digital. So I realized that the byte rates, data rates are increasing and everything is getting into microwave domain. Now it's in millimeter wave domain even. So I decided to apply my knowledge to analysis of digital interconnects and joined in Ivida and became mentor. And in 2006, we started Simberian as a project to make electromagnetic analysis accessible to anyone in the industry. Back then it was really difficult to do. Now there are some easy to use tools, but Simbeor was a project to make electromagnetic analysis mainstream. We're still working on it, after so many years. During this process we developed a unique technology called decomposition electromagnetic analysis. And the major goal in all this to have analysis that weights those measurements. So the product is Simbeor electromagnetic signal integrity software.

Zach Peterson: So I think one of the challenges that I think a lot folks have with electromagnetic simulators is on the surface they seem very esoteric. And if someone has ever dived into something like COMSOL, they know that these can be very complicated. In fact my very first exposure to doing any kind of numerical simulation with electromagnetics, and in this case optics just based on my background, was in COMSOL. And I was pretty impressed by how complex it can get, especially because there was just one person in our research group and that's all they did was simulations on COMSOL, that was their entire thesis. So I think I'm a bit curious as to how do you see some of these programs having progressed and what have you done and what have other programs done to make these simulation capabilities more accessible to someone who might not be doing a PhD in engineering or physics or mathematics?

Yuriy Shlepnev: This is a great question. And even the original idea of Simbeor was to make electromagnetic analysis running somewhere in background without user involvement. But that was the goal. So the goal took many years to achieve. Simbeor can now run in automatic mode. And another problem where, well not a problem, but recently we noticed that more and more people tried to apply Simbeor to lower data rate. It was mostly developed for data rate above 10 gigabits and about 20, 30 years industry went on. So most of our customers was in this area and they didn't mind messing a little bit with electromagnetic settings to make it run properly and so on.

Yuriy Shlepnev: And with their experience, our experience so we gradually started to make it look completely automatic and run completely automatic. And about two years ago we decided also to address more data rate market below 10 gigabit per second. And start development of new tools for that market. After two years was the release of Simbeor 2022 that included all those. And I hope I will be able to show it to how it looks now. And what difference instead of kind of was automatic set up, especially.

Zach Peterson: Well this brings up a question because you said you wanted to make these simulation capabilities accessible to somewhat lower data rates. With digital signals the rise times can be such that a lower data rate can have similar signal integrity problems as a bit stream with higher data rates. Would you agree?

Yuriy Shlepnev: Rise time yes, you're absolutely right. Even below 10 gigabit per second, you might have rise time that brings you essentially a spectrum of your signal. And I wrote and article on this recently on how to define it, what it is. And so the rise time is a major part of this. And basically if you use components with faster rise time, you have to extend the simulation area in frequency domain and get quickly into microwave again simulation domain, and you have to simulate it properly. Basically not only signal integrity things, but also things related to EMI maybe solved simultaneously in the same way with the proper localization of every single structure on the board. So otherwise this energy that generated with faster rise time will be dissipated somewhere on your board.

Zach Peterson: Right, whether it's in the-

Yuriy Shlepnev: So creating more problems in EMC, EMI and so on. So yeah.

Zach Peterson: So I guess more of what I'm wondering is in terms of the mathematical approach that is taken in an electromagnetic field solver, working at maybe sub 10 gigabit per second, to above 10 gigabit per second. How does that compare and maybe how is it implemented in Simbeor? Is there an approximation taken below 10 gigabit per second that allows the solver to maybe operate faster, let's say?

Yuriy Shlepnev: Yes. In the last release of Simbeor 2022 we have basically three simulation mode. Both are the composition analysis, but one simulation we call it Fast SI show how it looks. Fast SI uses more approximate models for pins, pads, vias, most important thing. They are based on electromagnetics as well, but not full wave 3D. Full wave, but not 3D. And that allows to run it faster for more mass included. It includes crosstalk between traces and so on, but not crosstalk between vias for instance. So that's more kind of 3D electromagnetic problem. And then we position it as sufficient up to 10 gigabit per second, tested, validated, and so on with the frequency range a little bit closer to 20 gigahertz. But then if you need higher data rate or faster rise time, broader frequency bandwidths then... And Simbeor defines automatically what the bandwidth is.

Yuriy Shlepnev: And then you need to switch to 3D electromagnetic models that run all of magnitude or maybe couple of slower. But the benefit is always accuracy. So accuracy that that carry weights up to much higher frequencies. Simberian started many project with partners, with customers and results of some of those validation projects are reported on our website. It produces systematics and if some conditions are obviously not violated and so on. That's why you need full wave 3D decomposition analysis. So now we distinguish those areas. And the reason is we realize that our customers at lower data rates they basically not required. With electro magnetics they need lower cost licenses that we created. And so that makes life easier. But the simulation is, again is based on field solvers, cross sectional field solvers, and solvers for vehicles.

Zach Peterson: And so-

Yuriy Shlepnev: So it's not approximations formals or anything like that.

Zach Peterson: Sure.

Yuriy Shlepnev: Honest analysis.

Zach Peterson: And so you bring up a cross sectional field solver and I think that not every LTM user knows this, but there is actually a cross sectional field solver built into the layer stack manager in LTM designer for which Simbeor is responsible, is my understanding. Did you guys develop the model, the solver engine? How exactly is that implemented? Is it the same type of-

Yuriy Shlepnev: Yes.

Zach Peterson: Solver that's used in the newest version of Simbeor, or a previous version?

Yuriy Shlepnev: That's a unique part about Simberian, all technology is built within the company. So including static or quasi static field solver for cross sections that are made so from pretty much any cross sectional PCB in package. And that technology was package in Simbeor SDK, software development kit, which was used by Altium to build a stackup manager. And back then we also moved into the Simbeor SDK all other solvers. Simbeor 3DML based on method of lines that I, my dissertation was on method of lines many years ago. So I keep developing it. And another solver is based on Simbeor's own 3D finite elements called Trefftz finite elements. It's wave based finite elements, completely unique technology. Also available in SDK. But so far Altium uses only the SFS solver for cross sectional analysis and that benefits Altium customers. And about two years ago we started thinking about why don't we also create some capabilities in Simbeor for fast checks of it's called electrical rule checking based on simulations on this cross sectional field solver, in particular in fast models for vias. And that what was major in Simbeor 2022.

Yuriy Shlepnev: And the reason we did it first, many problems that... We encountered problems that customer are trying to solve in legacy designs that are trying to move to higher frequencies where electromagnetic analysis basically doesn't make sense. You have to run simulation based checks first and fix those problems and then proceed with electromagnetic analysis. But they will jump in directly into electromagnetic analysis. But finally a couple of our customers approach us and said, well why don't you kind of not just explain what is possible to simulate and what is not, but you make it clear with some tools. And we built Simbeor SI compliance analyzer with electrical. So I hope I will show how it works just to see that electromagnetics, all the simulations are hidden and you don't have to mess with those.

Zach Peterson: Yeah, absolutely. And I think now would be a good time. And anybody who is listening to the audio stream of this, please head on over to the Altium resources website.

Yuriy Shlepnev: Yes.

Zach Peterson: There will be a page where we will link to the video and you can actually watch this demo in action. I think it's very valuable for anyone who has not seen how a 3D field solver works to be able to look at these results and at least understand what they're seeing even if they are not a power user of these types of applications. So why don't we take a look, I'm very interested to see this.

Yuriy Shlepnev: So just to make sure that I didn't prepared anything, I start from clean Simbeor and create new solution. This is regular way you do analysis. And defined bit rate 10 gigabit per second, 40 second rise times. This is important even for preliminary check simulation based checks. So next I import ODB plus plus file. A test created results at Altium. And for testing like this. Import. And there were some minor problems where Simbeor basically gets some dielectric parameters for some layers on this board. But this is a board. So we can look at it. Very simple in 3D view. Specifically created for signal integrity. Quick polygon feeling here allows us to see what every single. On a simple board it's really easy to see things here and kind of does make sense. And every signal integrity analysis must start with reference integrity analysis.

Yuriy Shlepnev: So basically we created the CRC mode and compliance conditions are three modes so far. And the first one is reference integrity. Lets just see if reference integrity. I need to select nets. Select nets, run. It tells me, the dialogue pops up and tells me what to do. Okay. With reference integrity we can immediately see problems in reference. And I need to extract reference conductors as well. So now we see that there are some problems with reference conductors. So here we can see bandwidths of each and one of the vias basically that more satisfy bandwidth for 10 gigabit per second. This one, it has two station vias, but they're too far. So this is the first type of analysis. Everything must pass. If it doesn't pass, we can proceed and see consequences of this. For instance, we need to relate impedance continuity, I click crown.

Yuriy Shlepnev: This is impedance of every single trace part most important vias as well. Right here. It's impedance deviation from 50 ohm goal with 5% deviation. And we can see that mostly all those traces missing on impedance is. And the field solver, Simbeor, you don't see it. In background it runs analysis for about 20 cross sections. And you can see vias have some stops and then boom, the impedance is too low, and for traces it's too high. So we can back drill those vias and quickly just trace with, by one mil for instance. And you can see that Simbeor immediately does it. And now impedance's are green, some vias are green, but this area didn't disappear.

Zach Peterson: So in this area where you have the red, it says I think 111 ohms is that correct?

Yuriy Shlepnev: Yes.

Zach Peterson: So this is because you're basically routing over a slot in your reference conductor.

Yuriy Shlepnev: Exactly.

Zach Peterson: Yeah.

Yuriy Shlepnev: Exactly. And there are strip lines, two narrows strip lines as well. 74, 75 ohm.

Zach Peterson: So what this is really showing you is this is just an extraction of the characteristic impedance in those particular trace sections. It's not input impedance looking along a specific direction along that interconnect.

Yuriy Shlepnev: No, that's specific impedance in every single cross section. For vehicles it's an input impedance of VA terminated on one side by the target impedance. So VA is not transmission lines. There are no characteristic impedance for VAs, but for transmission lines, every everywhere where metal can be turned into a piece of transmission lines and turned into piece of transmission line accounts along all the reference conductors around. So

Zach Peterson: That's what I'm getting at. So it does account for the presence of other conductors nearby.

Yuriy Shlepnev: Mm-hmm.

Zach Peterson: Perfect.

Yuriy Shlepnev: Yeah. And the third type of analysis is here, crosstalk. So let's just take a look at the crosstalk. And with the crosstalk micro strip structure, I need just a little bit proximity parameters to have larger distance where Simbeor finds crosstalk. And now we have, okay this is 23 millivolt, 25 and 12 millivolt on parts. Simbeor actually automatically finds the crosstalk areas. If you select just one net, it will find aggressors and tell you. It's a local estimation that uses cross section and then length of and rise time, obviously. It uses rise time. You change rise time, crosstalk you change.

Yuriy Shlepnev: But for the final analysis of crosstalk we can use those Fast SI's that I mentioned, where the models are built. And during all those runs Simbeor created some violation reports. I will not dive into it. It's a short kind of session. But now we created a linear network for further crosstalk analysis, as well as insertion loss, return loss analysis. And we just run and we'll look at insertion loss, return loss as well as crosstalk on different form in all nets simultaneously. So I selected them. And we can see it's just 22 cross sections. Simbeor runs analysis and runs analysis of linear network. And then shows pulse response. And you can see this is the net with those cutouts on the ground. You can see pulse response is really bad. This is crosstalk. This is crosstalk for impulse response, and you can see values of crosstalk fine crosstalk.

Yuriy Shlepnev: This is actually crosstalk that to the driver. And insertion loss, we can see how bad this net again there's six one. Return loss, you can turn it into any other compliance metric here. This is pulse on crosstalk. So what I wanted to show, this is kind of very consistent analysis in three months now. For instance, if I want to switch final, this is final analysis to 3D analysis to make it quick, I select just one net. And the way I don't think is exactly the same. I build model, and that model will have 3D electromagnetic models in addition to transmission line models. I click extract in the same way I upload eye diagram and there's parameters of this continuities. This is very far from any COMSOL mode you saw Zachariah. And Simbeor is running now analysis on my server, in my in office server, it support distributed computing, you can have as many servers as you have licenses.

Yuriy Shlepnev: And from the server results this is a fill structure, which is we can take a look at it's a multi like this. It's a Simbeor 3D geometry, very simple then. via like this was 10 gigahertz localization frequency. Both vias are parameterized and can be adjusted within Simbeor via analyzer tool. So everything just tune your design after you do the extraction. And well the simulation is going on and will be finished in the moment. So I should mention that the first two mode, ERC Fast SI included a new type of license called Simbeor verify. Which is just some $150, it's not low, per year license. And that sufficient for all needs up to 10 gigabit. And for ERC checking, even up to higher data rates. So we see now eye diagram of this link with electromagnetic models, 3D electromagnetic decomposition analysis, insertion loss, return loss. And what else? I selected eye diagram. That's...

Zach Peterson: Can you do impulse responses as well?

Yuriy Shlepnev: Yes, absolutely. So computed impulse responses in-

Zach Peterson: Oh, I see. You already did that.

Yuriy Shlepnev: Fast SI.

Zach Peterson: Yeah.

Yuriy Shlepnev: Yes.

Zach Peterson: Got it.

Yuriy Shlepnev: Yes. And the worst case with this link was.

Zach Peterson: c p>

Yuriy Shlepnev: Under the ground.

Zach Peterson: Yeah.

Yuriy Shlepnev: Yeah. But when you extracted it, you then can turn any kind of graph into other things you want. Take a look at phase delay at group delay,, crosstalks or crosstalk to insertion loss ratio. Or do integrated crosstalk analysis. Just select properness with crosstalk calculate and see that integrated crosstalk noise. Computed all types of compliance analysis, except channel operation margin. You can proceed with results in with. And analysis with IV's model, you can proceed with pipe beard, free tool, or with other tools like H spice and hypering for instance.

Zach Peterson: This is a lot packed into one program, which I find very cool. But I think the coolest part about this type of program where you've automated so much of the analysis is you don't have to take E and H field or E and D field results and export them into another program and then analyze them to get to where you've gotten.

Yuriy Shlepnev: Yeah.

Zach Peterson: Which is the important signal integrity metrics. You've just bypassed all of that, which is great.

Yuriy Shlepnev: Exactly. And this is more closer to production board, and you can see quickly right here the dielectric constant was, or dielectric was run. And the actual dielectric on the final board had different dielectric constant, and you can experiment quickly. And here it runs 690 cross sections in split second. And you saw this, this is real. And actually you can see variation of impedance here caused by different cutouts in the reference plane. You can see that all those compensation links, they have high impedance. And then there are variation along the traces, you can see how fine is the model. Almost 700 cross sections in a split second. And then you see complete picture of what's going on. So you can simulate all PCI express links or DDR links within a minute or two. And then progress with analysis, do more accurate analysis of crosstalk, and then do more analysis for the discontinuities if you have higher data rates.

Zach Peterson: One thing I just noticed, which should underscore the issue with two tight coupling in these links, is the length compensation sections are high impedance. Which I think is-

Yuriy Shlepnev: Exactly. Very reflective.

Zach Peterson: Yes.

Yuriy Shlepnev: Yeah. Very reflective.

Zach Peterson: It's very clear here in this color coded view. And this is one of the reasons why I think folks need to get over the tight coupling in differential pairs, which is sometimes espoused in so many design guidelines and not very well defined. So it sounds to me like you could even use this on the front end when you're actually designing a channel, or maybe you're designing differential pairs.

Yuriy Shlepnev: Exactly.

Zach Peterson: You could build a quick test forward, put it in here and say, okay how close can I tolerate my two differential pairs to each other? If I need to apply a maximum of, I don't know, four length matching sections, let's say.

Yuriy Shlepnev: This is exactly what the intent was, the CRC. And that was driven by our customers, driven by demand that essentially you have to, how to do it in quickly even before the board is produce the final. And tune it in Simbeor, and then get back to way out. And you saw you can tune with, for instance, shape of the strip and tune vias and parts and so on. All those cutouts below the parts, sometimes they need it. So it can be used in different ways. And this Fast SI is the first, if you don't pass insertion loss with Fast SI for instance. And there is no need to do 3D electromagnetic analysis. Same with crosstalk in traces. At least, if you don't pass the compliance metrics, then you have to fix and then proceed. So one step at a time and ERC is kind front end.

Zach Peterson: So one thing I didn't see, or maybe you didn't show, was accounting for fiber weave. If you're using a somewhat open weave you can have skew accumulate. It could be totally random because of course you don't know if you're going to be routing one of those traces over periodic gaps versus only over the fiberglass. And it's difficult to account for in some of these programs unless you set up a material system that has a periodically varying dielectric constant in two dimensions.

Zach Peterson: So I would like to get your perspective on this. Because my feeling is that people say, well I think that fiber weave is going to cause a problem. Therefore, I'm just going to go with spread glass or find the tightest weave possible that I can use. And then they just do that. And then they say, okay forget about fiber weave. Or we're going to wait until we spin, and we'll see if there's a problem. What do you think about that approach? And is there any way that simulators can help address that?

Yuriy Shlepnev: Well, we work with some of our customers in= starting from, ISOL published a few papers on this. Simbeor has macro models to account the effects. But technically, almost no one uses this because of there are no data to get into the model. So parameters of the models are mostly unknown. And about a year ago was a customer, I cannot name right now, we start building probabilistic models for those. So knowing geometry of the fiber, if you exactly write that you don't know where your trace is. And then how it gets aligned with the fiber and so on. And that makes it really difficult and practical to build, to simulate in. And so we decided to build a probabilistic model that is based on analysis and 3D in SImbeor, where we get the geometry of the fabric.

Yuriy Shlepnev: It's a parameterized. Then parameters of glass dielectric and resin dielectric. And Simbeor builds a model of variation of the propagation delay from dependency from the position of the trace on this kind of 3D electromagnetic model. And then convert it into probability. And the goal was to do it for DDR links. Because of in DDR it's really to match the delay. And when DDR data links they aligned with the fiber, and they often align, it's the cheapest way to design the board no rotations. And like this. So the variation may be above the capabilities of a particular chip to compensated. So one way of compensate delays, do on the board additional links. But after you even do this then there is uncertainty how they aligned with the fiber. So this was a practical application, and it'll be included in Simbeor .

Yuriy Shlepnev: It's a part of SDK kit. So it's everything build around SDK. I hope we'll publish it because we're going submit it to Design Con. I don't know we will be accepted or not. But so far the project is more practical I would say. With ISOL a few years ago, we developed models that were with the measurements and predicting skewing differential press. But too many parameters are completely unknown for the model. You have to basically build a test vehicle to be able to extract parameters of the model, and then you know what the model is. And after this, it becomes useless.

Yuriy Shlepnev: But with this probabilistic approach, we are getting closer to practice, again driven by a customer. As most of the things, maybe all things in Simbeor, we are driven by customers. Including Altium. Simbeor SDK originally was developed for Altium to be able to do this stackup manager. Now many customers are using it, building their own in house tools, and you can use MATLAB C++ or Python to build your own tools with Simbeor SDK. And Simbeor SDK has kits like this and kit. So it's a little pre-announcement, but it's going to be done soon. We're working on it.

Zach Peterson: That's very cool. And I bring it up because you brought up probabilistic models. I actually talked with Heidi Barnes privately earlier this year about just such a model to be able to account for some sort of statistical or probabilistic variation, whichever approach you take, to impedance propagation delay caused by fiber weave effect above some frequency limit. And so the idea being that you can then get like a worst case that you could expect for a particular.

Yuriy Shlepnev: But what is the percentage of your links will fit into this worst case? You take 10% of your links for instance. What is the probability to have them within this bin? That's the major question. And it looks like it's really important to know. So is it just very small percentage of your links or you have to expect something really bad. And in reality, it looks like that the probability of having the worst case queue is very sizable. So it's increases actually with the number of cases, probability increases for worst case scenario. And then you basically decide what is a too high percent on simply worst case is maybe not... It's important to know that you may hit something really bad, but how often? What number of the links may hit it?

Zach Peterson: Well, I was looking at it-

Yuriy Shlepnev: That's the key.

Zach Peterson: I was looking at it like this. If I know the worst case and everything that I design is better than the worst case, then do I really need to worry about the worst case anymore? I would say now a hundred percent of my links are better than the worst case I could tolerate. Would you agree with that interpretation?

Yuriy Shlepnev: The problem is you don't know that... Like in DDR. You have to create the ways or make them certain number for each particular data link. And then on top of this, you have uncertainty with the delay. So you can hit this worst case and how many of those links may possibly hit those worst case. And it depends on geometry of the fiber weave. And returning to the original question. Does it worse to switch to more uniform dielectric at high cost? Or is it okay to stay with this? And, and the question, I forgot to mention.

Yuriy Shlepnev: So obviously, what is the worst case, like additional delay difference between the links? What is the worst case? And what is the presentation of the links you expect within like 10% from this worst case. That's that's uncertainty and you don't know, is it going to be the worst case? So until you do the analysis and evaluation of this probability. And geometry of the fabric matters quite a lot on the probability distribution function. So essentially you end up with some probability distribution which is not by far. Absolutely it's completely different kind of distribution.

Zach Peterson: So the-

Yuriy Shlepnev: And we build it numerically within this kit. Just running a lot of numerical analysis, through numerical analysis, essentially. Again, a few things you have to know, like geometry of fabric is usually no parameter. And dielectric constants, it's the distribution of probability depends but not heavily. So if you get number for fabric from manufacturer of fabric, they provide some numbers. And then give the number for reason that produces composite the electric constant close to some observed on the board, measure it. Then you're in good shape to proceed with such analysis. So much small number of unknown parameters for such model. And again, it matters only where it is important. So for differential pairs you can mitigate skew in different way, with a pitch and so on. And with differential pairs, you have more flexibility and delay doesn't matter, so only skew. So again, you can evaluate skew worst case skew with a numerical model.

Zach Peterson: So that was just what I was about to get to in terms of the evaluation and building the final probability distribution that you would expect. Are you doing this through Monte Carlo and then just bending it into a histogram, or are you actually deriving a curve through a numerical solution, basically picking results.

Yuriy S cure for position of phase delay or impedance change with the position on the fiber. After this, you can either convert it directly in probability if the shape of the function is analytical or known. So sometimes purely for instance, and that's an observation published in many papers. So if it fits into the probability are distribution. But sometime it's not. So geometry of the fiber gives you some weird variation of phase delay or impedance with the position of the traces with the geometry of the differential pair. And in this case, yes you're right. Monte Carlo analysis is a perfect way to convert this variation into probability distribution. And then use it as a main tool for decision. Does that fabric fit my needs. If my link have this length, they possibly routed along the fiber. And what is possible outcome or the best case, worst case scenario.

Zach Peterson: Well, then I'm sure-

Yuriy Shlepnev: Something in between.

Zach Peterson: Then I'm sure you could go back into the input parameters, maybe swap it out for a newer weave, or maybe not newer but-

Yuriy Shlepnev: Exactly.

Zach Peterson: Swap it out for a tighter weave experiment and then compare again against your worst acceptable case.

Yuriy Shlepnev: Yes.

Zach Peterson: And then if everything lies above your worst case, now you can say, okay these six links or these 10 links or whatever I need to design are all going to operate within acceptable limits.

Yuriy Shlepnev: Exactly. It makes sense only in case if you want to shave down some dollars from the cost of your board and go with the cheap fabrics, cheaper materials, and still not have failures. And that' all in this analysis and probability models.

Zach Peterson: Well, when we're talking about someone who might be producing a million units of something, those couple dollars-

Yuriy Shlepnev: Yeah.

Zach Peterson: They add up.

Yuriy Shlepnev: That's exactly the case. Yes, they were involuted and they have real failures because of this. So just blindly switching and then chips are much smarter nowadays, but they still cannot handle the variation which is maybe giga second per inch easily variation due to the position on the board. Especially for along the fiber. So it started with cheap fabrics, or cheap materials. The whole fiber effect subject. I remember boards from Scott McMorrow or papers where they measured it. And some quantified papers from Oracle, many other. I have collection of those. And then dielectrics became more homogenous. And nowadays you can meet people that they'll tell you that fiber review is not a problem anymore. What are you talking about, and it's not important subject, but it is, it is important. In case if you want to use the same materials as people used 20 years ago. They're still much less expensive and they still have skew. I mean, fiber effects, they still have fiber effect those first papers from Oracle and Terra Speed.

Zach Peterson: Well, I think that-

Yuriy Shlepnev: So it's not went away. So homogenous materials, yes there are excellent homogenous materials.

Zach Peterson: I think there's a perspective issue there because when someone says, oh fiber weave effect doesn't matter. I'm wondering if it's just because they're operating at low volume. I almost said low value, but that's not true. They're operating at low volume. They're not thinking about the fact that, hey this thing that we built that costs $5,000 to build this prototype, we're not trying to scale it up to a hundred thousand or a million units or 10 million units. And the people who do have to scale up to 10 million units they may not be able to afford a material that is more homogeneous. And so of course they have to compromise on cost somewhere. And it seems to me that the laminate material is an attractive place to start looking.

Yuriy Shlepnev: Absolutely. Companies that do not care about cost of the final product and so on that's they obviously solve problem with just switching to better materials. But look at the publications from Cisca.

Zach Peterson: I'm sorry, from where?

Yuriy Shlepnev: Design from Cisca systems. So they can and you can see the publishing papers on fiber review effect. A few papers at Design Con that I mentioned, this was distribution. And so on. A lot of measurements they're doing to figure out how to go away with was a cheaper dielectrics and they concern.

Zach Peterson: So I think this has been very interesting, and I'm happy to hear that there's a simulation program that is taking advantage or taking account of fiber weave in a more effective way. Because I've seen another software company, who I won't name, they just, I think they just assume Gaussian everywhere. And they try and fast track it. But you're saying incorrect.

Yuriy Shlepnev: No, it cannot be Gaussian. Gaussian was unlimited. And right here, you obviously have limits. Like this is one probability of delay was somewhere in the middle, for instance, and this is the highest deviation from this middle. And those highest deviation is defined by it's material. It's not go away. So obviously was duel direct distribution. There is a little bit of Gaussian distribution at each end, but overall it's not Gaussian by any means.

Zach Peterson: Well okay. So one thing that's done in random walks in finance, so I'm going to turn on my finance brain for a moment, is they will actually use a truncated distribution. So basically what they'll do is they'll have a distribution that accounts for all of the data observed throughout history. They'll trim it at the top and bottom ends, and then they'll set the extreme ends to zero. So they take a Gaussian shape and then they manipulate it so that it is in fact does have a worst case that's ever been observed and a best case that's ever been observed. Is that appropriate, or are you saying that the probability distribution for skew or any of these other signal integrity metrics, the shape is actually non Gaussian as well?

Yuriy Shlepnev: Absolutely non Gaussian.

Zach Peterson: Okay.

Yuriy Shlepnev: And I can explain in very simple terms. For instance, think about trace aligned with a fiber. And change the position of the trace across the fiber. You'll see the phase difference when it hits over the glass heels or over the resin values that's going to be maximum and minimal phase delay. And you cannot go then over the glass heels all lower than over resin values. And if you check this distribution of the face, most of the time it comes as. And what is probability of something happening if you have distribution of? It's an arcsine function. So functions that looks like up 1% like this. So you have high probability to have worst case scenario, in other words. And it's limited by nature, by properties of the material in real life. Of course it's going to, if you build a board and measure like Cisca systems guys did you'll see the top and bottom some Gaussian variation if you take a lot of samples.

Yuriy Shlepnev: So it's going to be not exactly limited on this end and on this end. But slightly different because of the fiber goes not always as close to surface and the model it goes up and down. So reason is a little bit variation thickness. So also thinks obviously Gaussian distributions. But overall when you take it into phase and then compute probability distribution for this shape, you'll see past arcsine distribution for the whole thing. So it's limited. It's not going to infinity anywhere. And artificial limiting is not even needed in this case. So it's sufficient to have it like this. So it's an interesting subject.

Zach Peterson: So let's say I build a million interconnects and I test them all. Does central limit theorem apply to all of those measurements?

Yuriy Shlepnev: I don't know frankly.

Zach Peterson: That's okay. My math brain says yes, but I don't know of a case where central limit theorem doesn't easily apply.

Yuriy Shlepnev: When you build it obviously, yes.

Zach Peterson: Yeah.

Yuriy Shlepnev: But in numerical experiment, probably no.

Zach Peterson: Okay. Well then there's some convergence to central.

Yuriy Shlepnev: Then you have to embed a numerical experiments some kind of randoms. That's another level.

Zach Peterson: Okay. Yeah so then it's Monte Carlo, not just for the weave, but for the individual interconnects itself. And then it becomes another level of Monte Carlo.

Yuriy Shlepnev: Yes. Distribution of also thicknesses and properties of the materials and so on. Not just position of the trace, position of the trace will not make it even close to Gaussian.

Zach Peterson: Yeah.

Yuriy Shlepnev: But all other variations, they all make it. But that's that second order here?

Zach Peterson: Yeah.

Yuriy Shlepnev: I would say. So the first order is how many samples and where of the worst case scenario probability of the worst case scenario.

Zach Peterson: So I'd like to know, quickly before we go, because we're getting short on time here. What do you see in the future of Simbeor and maybe other products that designers can use to help them expedite some of these important analyses that you need to perform for high speed designs and RF designs?

Yuriy Shlepnev: Well, the future is technically I think it's something like simulation driven route. Where everything in the interconnects become under control of that decides this is possible, this is okay, this is not okay. And sooner or later, right now industry is kind of simulators are separately, routers are separately and some companies are trying to merge them, but unsuccessfully due to the lack of quick simulation, fast simulation technologies that makes it possible.

Yuriy Shlepnev: But in reality we see this data rate is that only way to manage this is to do electromagnetic analysis obviously, and keep all the rules starting from the localization and so on. And quick estimate of is it successfully, two components are successfully connected or not. Just to make people not worry about what is what is impedance of and router should take care of it automatically. That's my kind of opinion on this. But we have a lot of plans at Simberian how we'll evolve the big plans making life easier. So essentially in this kind of direction. I didn't show you about this reports that Simbeor creates. It also has some tips on of how to fix the problem.

Zach Peterson: That's always helpful.

Yuriy Shlepnev: And most of the fixes you can do in Simbeor. We hope to do it automatically in the future. But so far I hope to do electrical rule checking not only with quasai static fuel solar, fast field solar, but also with 3D electromagnetic solars. So just for data rates 112 or 224 now people are running. So that's that's area of millimeter wave technology. So we kind of getting faster into this area. Well at Simberian we also invest a lot in simulation technology and our solar 3DTF based on Trefftz Finite Elements. For instance, last iteration was included infinite point boundary condition basically to handle cases unlocalized. So again, legacy design people created for 30, 50 gigabit per second, trying to 112 simulation frequency range is getting over a hundred gigahertz. Where in legacy design localization breaks out in those and how to handle the synergy that leak, you cannot simulate the whole board in 3D simulator.

Yuriy Shlepnev: You cannot restrict there, you have to create infinite planes at least. And lose the energy it'll begin somewhere, but this type of models is required, I think. And keep evolving also. Invest. So working at both end now, ERC and Fast SI and extending our 3D electromagnetic analysis capabilities where the major strength of Simbeor is actually this process called sink or swim, which starts from material unification, roughness model unification, and ends with analysis to measurement just kind of one flow. And that's in my DMI videos on the website where pretty much today I demonstrated it on simple test board. But usually I demonstrate everything from start to end where the end is analysis to measurement. So how to those measurements.

Yuriy Shlepnev: Our partner technology provides some test vehicles that are available for anyone who wants to test analysis to measurement and be able to repeat measurements, repeat simulations. And this is really important for people especially moving into this area. So it's easy to do in Simbeor this analysis. You saw this, even 3D electromagnetic analysis. But then when it gets to measurements. Imagine measure measuring it up to 20 gigahertz or 40 or 70 gigahertz. How many problems you're going to encounter.

Zach Peterson: Yeah, absolutely.

Yuriy Shlepnev: So that's area where it's still as difficult as it was 10 years ago or 20 years ago. Those are measurement, high quality measurements.

Zach Peterson: Well once-

Yuriy Shlepnev: And then have this correlation.

Zach Peterson: Yes. Yeah. And that's great that you have all of that built in to be able to do that measurement, to analysis correlation. Well, I would say once some of those simulation driven routing features are available, we would love to have you back to talk about how simulation technology is evolving for PCB designers.

Zach Peterson: Okay. Thank you very much Yuriy for joining us, this has been a lot of fun and very insightful. And I love talking signal integrity and of course related areas like power integrity, and RF design. To anyone who is listening or watching YouTube video, please make sure to subscribe to see all of our upcoming podcast episodes. Make sure to check out the show notes. We'll have links to connect with Yuriy on LinkedIn, as well as to learn more about Simberian and the newest version of Simbeor. Yuriy, thank you so much for joining us. And we hope to have you back again in the future.

Yuriy Shlepnev: Thank you Zachariah and thank you everyone.

Zach Peterson: And to everybody out there listening, of course we always say, don't stop learning, stay on track and we will see you next time.

About Author

About Author

Zachariah Peterson has an extensive technical background in academia and industry. He currently provides research, design, and marketing services to companies in the electronics industry. Prior to working in the PCB industry, he taught at Portland State University and conducted research on random laser theory, materials, and stability. His background in scientific research spans topics in nanoparticle lasers, electronic and optoelectronic semiconductor devices, environmental sensors, and stochastics. His work has been published in over a dozen peer-reviewed journals and conference proceedings, and he has written 1000+ technical blogs on PCB design for a number of companies. He is a member of IEEE Photonics Society, IEEE Electronics Packaging Society, American Physical Society, and the Printed Circuit Engineering Association (PCEA), and he previously served on the INCITS Quantum Computing Technical Advisory Committee.

Recent Articles

Back to Home