Characterization and Compliance Testing for 400G/PAM4 Designs
Good. Afternoon and welcome to today's, electronic. Design webcast, our. Topic, today is. Characterization. And compliance, testing for. 400g. Pam-4 designs sponsored, by keysight, i'm, kristy martina with penton's design engineering, and sourcing group now. Let's meet today's speaker. Steve. Faecal is currently the 400g. Solution specialist, in the. Internet infrastructure, team within. The networking, and data. Center division, at keysight in this, role he, guides the requirements, definition for. Solutions, and solution, components, used for characterization, and, compliance. Testing of everything, within the data center like, ecosystem. Involving. 400g, class optical. And electrical links, Steve. Currently, represents, keysight, in the, CEI that's. Common, electrical, interface, project, working groups within, the optical, internet, working form where, he has made several technical, contributions. To. Confines testing, specifications. And methodology. In the. OIF, he, also serves, as the chair of the fiscal and link layer. Interoperability. Working, group he, also is a contributor, in I, Triple E 802. Three, BS in 802. Three CD projects, developing. Next-generation ethernet. Links, operating. In the range of 50 to 400, gigabits, per second, recently. He has made similar contributions. Within. The 32 G fibre channel standard, and working group. Robert, slay is a strategic, product, planner in, keysight. Technologies network. And data center organization. Is. Responsibilities. Include. Product, development, of pulse amplitude, modulation. Pam-4. Test. Solutions, based on key sites high-speed, electrical, and optical digital. Communications. Analyzer, robert's. Experience, at keysight, technologies. Includes. Five years in technical, support and over, 12 year 12 years in technical, marketing prior. To working at keysight. Robert, worked for 10 years. And west old telecommunications. In Vancouver. British Columbia Canada. Designing. Microwave, and optical. Telecommunication. Networks robert, earned his vsee, degree, from the University, of Victoria now. Let me turn things over to our presenters, gentlemen, the floor is yours. Thank. You Chris. Okay. So, what. We're going to do an agenda today we're going to talk. About what the state of the standards are many. Of them are very close to being released or have been released and then. We're, going to talk about really updates oh there's too much material to cover in just an hour on on, 400g. And Pam, for signaling, so, we're. We're going to focus on the updates Rob is going to cover the transmit. Side and I'm going to do the receive side and then we'll wrap, up with questions and answers. So. Let's start talking about the for, energy class standards, I use that word quite a bit 200 G class and don't, just think of Ethernet really, what this is is I, like to define it as all of these standards, I have. To do with data center networking that use a lane rate of 50 gigabits per second, so, all of them are very similar most, of them are using pam-4 although we do see some, other modulation, formats.
Involved. The, the, one to really kick the whole thing off was the. Ethernet. I Triple, E 802 dot three BS project, 400 gigabit per second, this. Is a single, mode fiber optical, links along. With the chip to chip and ship to module interfaces, the, technical, work on this standard is all complete, now it's been complete for a couple of months it's in, the process of a formal balloting, and will be published in late. Spring 2018. On. The electrical side o IFC e I this. Is a common electrical interface, there's a series, of documents. There that. Cover five, different reaches, everything from die to die on the same chip carrier to, back planes and everything in between and, these. Standards are in the same state that the. Technical work is complete, and they'll. Be published, either, in. You know very very late winter or early spring. 2018. So it's kind of in the same state as a, date, or 2.3 BS the. Remaining. Active project in Ethernet, is the CD, project, this is where we bring in the multimode, fiber so this doesn't have a 4G, speed but, it, has everything below it and against all 50 gigabit Lane rates it's multiple lanes and. It. Has its own ship to ship and ship to a module interfaces. As well as adding back planes and passive cables and this. Is entering what they call sponsor balance stage which is says. The draft is getting more firm so it's more. Difficult to make significant. Changes in it and. Then, a remaining project, in in, on, the Ethernet side is. The. Weather fight there's fibre channel project, fibre, channel 64, G we, just had a meeting here yesterday in Santa Rosa and I, would. Update the slide just based on yesterday, the third third. Draft it is complete, it's actually graphed for but it's the third complete draft is. Out now and I think this is gonna wrap up in about six months so, technical, work will be done about the middle of summer and then, it'll, go to, letter. Ballot and publication, and then, an Ethernet they, just are starting, a project to. Pick up the longest reaches this is really not inside, the data center it, would be a, big, metro and data center interconnect, and, they're trying to get into the 40 kilometer, we had that in energy and and the, slower rates but it's so far it's not addressed at, the higher rate so there's a project it's just exiting, the completion. Of the study group and, MB. Ready, to do the project start, what's. Coming up next is going to be the 800 G class. And these. Will be denoted with lane rates of 100 gigabits per second, or or slightly, greater and oh. I F is already actively, working on this on the next generation common electrical interfaces there. Are four reaches defined three of them have formal project starts, and one, of them the VSR, which is the chip to module already is in its second, draft stage so it has a complete draft out and then. Ethernet will be following up with something very similar to the. 802. Dot three BS but, doubling, the the Lane rate and this will be the 800.
Gigabit Project, will be starting with single mode fibers, and that, project really won't get started, until after. A BS is published that's, kind of where the digital, documents, are you can see they're wrapping up so, what's the general status of folks, working with these things and designing to them well, we're right to the stage now you know we turned, on the individual, components, in the ecosystem, and we're, actually starting to do some total systems where we put you know surveys, and host ships with. Cages, and. Connectors. And modules, on circuit. Boards and are turning them on and, you. Know when we do that as expected, with any, new. Standards, or some issues. Running into and it's showing it we're going to have to make some changes it's, a little bit tougher, with with. The 400g, generation, because with a shift to pay them for that was a fairly monumental. Change, and so, it's a little bit more difficult, than the evolutionary, changes we've had in past so. You know the normal things you expect with the parameter, tables and these documents are going to be updated, you, know with changes in tx/rx. As. Well as channel operating, margin parameters, and then. Several. New measurements, were defined because we're dealing with PM for modulation, now many, of those are on the translate. Side and we're, always going to talk about that but we're seeing some changes where they've had to go back and and and. Make. Some tweaks to the measurements, which means if you have these embedded in, your test instruments, be, sure you're always using the latest software version because, these measurements, do change and there, are in a state of flux right now and then, actually how we do the testing and some of the test setups are. Needing. To change because, of the difference. Of the modulation the fact that we're dealing with SEC links well cover, quite a bit of that in the RX section, and, with. That I'll just turn it over Rob all. Right thank you very much Steve, today. I'm going to talk to you a little bit about output. Or transmitter, characterization. Both, on the optical, side some key measurements there and on the electrical, side with. A focus, on some, of the parameters that are called out in 802. 3bs, and CD. We'll. Start with looking, at some of the key measurements, for optical, transmitters. And we'll, look back just for a moment, at nrz, or non-return-to-zero. Transmitters. And some of the measurements, that were made to, characterize, those outputs one. Of them being optical, modulation, amplitude ro ma that's the difference between the one power level and the zero level, extinction. Ratio is, the ratio of, the one and zero level so it takes into account the, DC. Offset there's. Another measurement, called transmitter, dispersion. Penalty, or TDP. That we'll talk about it's a BER, based measurement, and then, the very traditional I mask test which puts up a standards based essentially. A keep out region, for the transmitted, wave form on. The. Pam-4 side, of things we've got very similar measurements. There's one called the outer oma which, is the difference between the three level and the zero level so if you're looking at that and, for, signal, on the top right hand they're the. Bottom level is the. Zero and then it goes to one two and three, so the outer oma is looking at the bottom and top, rail the. Outer extinction. Ratio is the ratio of those, two levels and then, there's a new measurement called transmitter, dispersion, eye closure for, pam-4 that's, P dot Q and, the Q stands, for coronary. Notice. That there is no longer a traditional. Nrz math test and we'll explore, why that is in just a moment so. Here's, just a reference, for you going. And looking at the latest, draft which is 802 dot three BS draft, 3.5. From. Mid-october. And, here. You can see the summary, of our. OMA outer er and P dot Q. So. What is T dot Q well it essentially, tells you the performance, of your transmitter, your device relative. To an ideal, transmitter, and if, we were to go back and, look at the nrz, TDP. Measurement, this is literally a BER, based measurement, where we compare, the performance. Of your transmitter, with, that of an ideal or golden, transmitter, that's a perk and what. We did was, we looked, at how much extra, power how much additional power, was required, at the receiver you know to compensate, your. Device for its non-ideal. Performance, and the difference between. Essentially. That additional, attenuation. Or power. Was. The, TDP value. Because. It required quite a bit of test equipment and it, was a BER, based measurement, it was fairly expensive and quite time-consuming to.
Perform. That measurement, so standards, committee members decided, that they, wanted a different, approach, and develop the tdq measurement, where, we indirectly. Measure the symbol error ratio using an oscilloscope and that, means that it's a lot more cost-effective and it's. Fairly. Easy and fast. To, make that measurement compared. To the TDP measurement. So. This is a block diagram showing, the tdq. Measurement. On the, top left you can see your device your optical, transmitter, and it's got, to generate a FS, PRQ, pattern, that's, a short stress pattern repetitive, cautionary, so it's a pam-4 signal that's. 2 to the 16, minus 1 symbols, long and that's, passed through some. Variable. Reflector, to control, the return loss it's passed through a test fiber you. May need to filter that signal before it goes into the oscilloscope. In. Addition. You need to have a, clock. Recovery for, your optical, signal and this. Is because the coherence, if you were just to trigger the, BCA, or your oscilloscope using. A trigger. From your pattern generator or from your desk for example you, no longer have, coherence, between or, correlation, between the, trigger signal and the data signals so you really have to use clock. Recovery, specified. For Meg loop bandwidth with first-order meaning. No peaking in, the response and you also, need to have a very well controlled, reference, receiver with, specified, bandwidth, depending, on the bod rate that you're operating at and then. There's also an equalizer, all of these links are performing. On the order of 10 to the minus 4 so, they're fairly closed, eyes and they need to be opened up in order to be analyzed, so the tdq equalizer. Is a five tap T, spaced FF, II, we. And make a pair of histogram. Measurements, on that pam-4 signal and, post-process. The signal in, order to get the tdq, result so let's have a closer look at what's, happening, as part of that key, debt queue measurement. Again. We're leveraging the fact that the link itself the signal is operating. With a very high symbol, error ratio on the order of two times ten to minus four and so. The first step that we do is we statistically, determine. The FBR on your. Signal or on your duct using, an oscilloscope. Rather. Than attenuate, the signal to force errors like we did with GDP. We're. Now mathematically. Adding noise to create, errors and we, continue, to add noise, onto, the incoming signal and so, the target ser is observed, so, that's on your debt duct, the, next step and again this is all done in post-processing is, to, repeat that process for, a virtual, ideal, transmitter, so you don't physically, have to have this it's all done mathematically. But. We, continue, to add noise. To that signal with the same OMA as your. Gut until. We achieve the, SDR and. Then, the amount of noise that gets added to the ideal reference receiver, in order to reach that target ser, is going to be larger, than the. Amount of noise in your set and the. Difference, in those noise levels represents. Tdq. So. It becomes a fairly fast. And easy, measurement. To perform. One. Thing to note that that. PDQ, measurement has evolved, since the start of 802 3bs. Very. Early t deck versions, optimize, that PDQ, equalizer, by trying to minimize, the spread on the eye level so it minimize the ISI, and you want a maximum, i opening. In. The latest, draft, 3.5. And again as Steve said it's all the technical, changes, for bf have really been implemented, at this point so. What, needs to get optimized. Is the t deck queue number itself, and there's various ways. That. We're going to do too and we'll look. At those in more depth in a moment, another. Change was the, vs. Measurement, vs. Standard, initially. Defined a five tap T over two spaced FFP and now it's a piece based FF, II. The. Standard. Initially. Proposed, a. More. Traditional vessel constant. Bandwidth, which is 75%. Of the baud rate and now. That's been changed, to 50%, of the baud rate in. Addition. The initial. D. DQ definition, had two time slices, in the center, of the eye now those. Two times, those histograms, can be moved, they have the same relationship, to one another but you're allowed to move them slightly within, that, pam-4 signal and. So. This is what's in the latest draft there's still some discussions, for CD, about, some additional changes that if implemented, might, be rolled back into the, BS standard. But. Note that these changes were made to better represent what. A typical system is going to behave. Like and, so we're, really trying to emulate what the real receiver, is going to be doing when, making this PDQ measurement.
So. One of the challenge is trying to find root. Cause for high. Feed at Q so let's say that you've got a P dot Q result, and it's higher than you'd expect where, you go to try and fix that problem well. You really want to start thinking like a receiver, so could. It be poor linearity, meaning. That the each, one of the levels is not equally, spaced and if, it's not then the, decision, decision. Threshold, that the receiver may not be optimally, set and that could certainly impact, your, system. Fer. Skew. If your skew isn't properly, aligned that could also, impact, fer, as can noise. So. Feed, a Q is fairly new but there's now starting, to become some more advanced, analysis, capability. Which can help you determine, what. Part, of your signal. Is dominating. The tdq results, so for example if you could make a partial PDQ, measurement that would tell you a lot about which, I was. Most. Problematic. Likewise. Being able to measure ser, for each eye and measure. Noise on each eye in order to determine. Perhaps. You've got a R in a relative, intensity, noise problem. Where, the noise on your top eye is dominating. The ser. Another. Key to accurate, repeatable. Tdq. Measurement. Capability, is to, ensure that you're using a very well designed equalizer, your t DQ equalizer, so. You want to have automatic, optimization of, both tap weights in the main cursor, in order to minimize that tdq value and you'd, also like especially. For speed, purposes, to be able to once. You determine, what your tap values are to be able to exceed your t, DQ AK equalizer, in order to speed the analysis, because it's a right, now it's an iterative approach and, it. Can be fairly time-consuming so. If you can shorten that test time especially in manufacturing, by, feeding your t DQ that will really, help your analysis, time. You. Want to make sure that you're using a reference receiver, that applies the correct frequency response. For, your oscilloscope. And so. There's techniques. Such, as the impulse response correction. Or IRC, that. Characterized. The non-ideal, behavior of, your receiver and, are. Able to mathematically. Correct, that and so what this means is that you're using essentially. An ideal, receiver, from. One instrument to the next in order to characterize your, gut so you get much better, repeatability. That way from one instrument to the next and then, you want to make sure that you start out with a very low noise receiver, this. Will impact, your TVXQ, results, as well. So. Keysight, has developed, as part of flexdca which, is a user interface for the dca, product. Line a pair. Of options, 9fp and 90 p that, provide a very large suite of measurement, capability, for analyzing. PDQ, one. Thing to note here is that what we've chosen to, do and this is by design was, build these measurements. Directly into, the user interface and that was done to ensure. That we, optimize. The measurement. Time so we reduce the measurement, time for you and you can see that we've got not just T at Q and outer oh ma and ER but we've have all those partial measurements, the partial Ser partial, T debt Q implemented, as well. From. A hardware standpoint, we've. Optimized, the solution, for both R and V and Manufacturing's, so again we've built that GDQ, measurement, capability, into the base firmware, the, hardware, itself samples, six times faster, this, new DC am or DC a mini.
Solution. Compared, to the traditional DC. A and. We've. Implemented. This impulse, response correction. Process that allows us to create an ideal, reference, receiver and, because. We've pulled out the, computer, we pulled out the display, and we've, reduced. The cost the manufacturing, cost we're, able to pass that cost savings, along to you. Very. Low, noise we have an amplified, receiver, so that also is really important, in providing. Accurate and repeatable tdq. Results. On. The real-time oscilloscope, side, we also have an OD, front-end. 33. Gigahertz, / 2 e that, can be connected, to our real-time series, the V Series X Q and V series, of solo scope and it's very powerful for, doing system level R in DB bug and troubleshooting. You. Can apply the ideal. Reference, receiver so, per the standard. 50%. Of the baud rate and we, have, built into the oscilloscope. Several. Nrz measurements, if. You're looking, for pam-4 measurements give. Us a call contact, your keysight rep, and we can give you an update on where those are as well but. As. You'll hear in a moment, using. Real time scopes, is a real. Benefit, when trying, to capture, rare. Events. So. Now we're going to transition to some key measurements on, the electrical, side of things and. Some. Of the measurements, that need to get performed, one is called level separation mismatch. Ratio, this is the linearity how equally-spaced. Your four levels are from one another another. Is called signal to noise and distortion, or, FM BR and. Even. Though it's, essentially. Technically, sound, for, BS there's still some debate over the specs for CD and if, there are changes to this measurement they may end up getting rolled back into the vs, standard. And another. Set. Of key measurements, is called output jitter these RJR msj for you for, uncorrelated, and even, an auditor, or ALJ, so. Having, a closer look at those you, may think that you recognize, a lot of these acronyms, so j4 sounds, a lot like J 2 J 5, J 9, EO. J has been around in some earlier standards, so. You might think that they're pretty straightforward to measure when. In fact these, new I Triple, E output jitter measurements, are made very, very differently, than more traditional or. Past, standards. In, the past J. 4 J 5, J 9, + EO j measurements. Were performed, on all edges, of an nrz pattern, for. 802. Dot 3 BS and CD these, measurements, now perform, on 12, specific, edges, of a PRBS, 13, q, so. The takeaway here is don't just pull out your scope, from press the EO j button on it because, you will not get the. An accurate, result you. Need to have very specific software, that's, going to target those 12 edges. Or. You have to do that manually, so. Why did. DF. Make. That change. Well there's recognition that, different, architectures. Are used to generate pam-4, signals, and if you have an architecture, such. As that shown in the, diagram, where you have two. NRZ patterns, that are generated. From two o'clock buffers, or two PLL's, it's, possible, to have different uncorrelated. Jitter, profiles, on your. MSB, and LSB and. As. A result of that if you were only to look at the clock, pattern, the JP stands, for jitter pattern, zero level two three level. The original, clock, pattern, that was defined, you. Could. Potentially. Miss some, jitter issues. And so, that's the reason why we're now, the standards committee has changed. To looking at twelve specific, edges is though to catch these. Problems. That might occur from that type of architecture. So. Standard. Specifies, which edges you shall make and initially. It was a very time-consuming. Measurement. To make because the standards specified, that each one percent histogram, had to have a million hits in it so, both real-time scopes, because, it had to acquire the entire pattern and sampling. Scopes because it you, know they'd sample, relatively, slowly, they. Took a long, long time to make this measurement on the sampling side it was, hours. So. Early, in January keysight. Presented, some advanced, methods to speed, up the test time while maintaining accuracy. And there's a reference that you can go to if you'd like to but. There were two main outcomes. One is the histogram size, change no longer you have to have one million samples in that histogram, and the other. Change was some. New acquisition, methods, were accepted, being, the edge model technique, that I'll tell you about. So. The edge model is where we can initially. Characterize, what that edge looks like so we create an edge model for each of the twelve edges and then. Using, that edge model when we acquire a sample, say sample one which is hi there we, know based on the edge model that through that one percent histogram.
That Is that. Transition. Actually was early and likewise. With sample, two if we measured low we knew we, would know where that sample. Would land in that one histogram, and it would be one. Percent histogram, and it would be late and so, this allows us to be extremely, efficient, with. Targeted. Sampling, and we can make those, output. Jitter measurements, very, very quickly now in, fact, that, new twelve edge jitter, method, that's been implemented in flex, pa has now reduced test time from hours, to, much less than a minute it's about 30 seconds. And. It can even be faster, than that depending on your your, hardware that you're using. So. We measure j4 ujr MSM even-odd jitter on that PRBS, 13q. And we. Will report not just the, parameters, that are required by the standard, but. We will also report, all of the individual. Results, so if there's something. Going on on a particular transition. You'll be able to see that that particular transition, is dominating, your result. In. Addition. On the electrical, side there's a number of custom, measurements, that need to be performed, such as steady state voltage s MBR that's that signal to noise and distortion ratio, measurement, I re, s MW i symmetry, mask width so. We have developed software to, facilitate this, this will automate, the. Oscilloscope, perform, the measurement compared. Against the result and generate a report for, both the sampling, scope that's the N 10 85, app and our, real-time scopes that's the n 88 36. App. 36a. Application. From. A hardware standpoint. On the electrical, side if you're using, a sampling scope you'd, be using the 86100d. With. An 86. 108 B plugin we. Also have other plugins, with remote heads and higher, bandwidth. Capabilities. On. The real-time. Scope you'd be using the V series because it has 63, gigahertz of bandwidth and that is required, in order to achieve the, 33, and 40 gigahertz. Vessel. Thompson, response required, by I Triple E and C II I. So. You might ask yourself well well. You've got a lot of capability, there which scope do I use to characterize my, pam-4 transmitter. And generally. Speaking, sampling. Scopes because, of the very low noise just architectural, II we can have much lower noise on a sampling, scope than, on a real-time scope, they, are going to be best for validating, and characterizing, pam-4. They're, going to give you the most margin, when you're making eye width, eye height measurements, for example. But. Real-time, oscilloscopes. They're very, important, when you're trying to troubleshoot pam-4, designs, they, can capture these one-time, glitches, and in. Fact because, of the very. High symbol. Error ratio on these, links they can actually do, some error capture, and. BER. Or. You. Know fer, performance. Measurements. For you and they. Can decode, that incoming. Data. Stream as well so. Real-time. Scopes have a tremendous. Amount of capability when, it comes to these pam-4 links, and they're a real key, tool to be using. So. With that I'm going to turn things over to Steve, and he's going to talk about import, or receiver characterization. Thanks, Rob. Ok. So I'm. Going to talk about two main topics here and then bonus, and quite a bit of depth and one of them is changes. In the input test set about alluded. To that in the introductory. Slides that. The. Big thing that we're seeing with pam-4, and. Lynx. Is that they don't run airfry they rely on real, hygiene for air correction to, improve, the link now be honest that wasn't something entirely, new with. For energy we had to add an energy with the back planes and with, one optical standard, the sr4 but, for many of you this is the first time you're dealing with links, that don't run air for, you by design, and there's, a couple of implications of that first when you're measuring BER all of, a sudden that number that you read in the for, the BRR measurement, or Cir measurement, is important. It has to be an accurate measurement it has to be a repeatable measurement, okay when the links ran on, air free basically. If you say anything other than zero in the display you, knew that you had a problem but now you you're, not going to see zero in the display so you're going to rely on the integrity of that measurement so, that's that's a big thing that is. A big change that we're seeing now and the, other one is in, the test setup because, these links don't run air free we got to think about that loopback link, in the bird that.
Might Not be that's, writing the air detector, loopback data, that. Link itself might not be air free and that creates some problems that's what I'm going to talk about quite a bit and then, I want to talk just touch a little bit on the importance, of agressor testing, you know that's been the standards for a long time you need to do it most people haven't done it in the past but. Because of the reduced signal-to-noise, ratio. That we have with PM for links it's really, important not only for the input measurements, with a with. A bird but the output measurements, we see noticeable I closure in not. Only electrical, but even optical, links with with. Adding, the aggressors so it's important that those be present so now, let's let's, jump into talking, about when the links, don't run air free by design, and what, the implications, are that when we're doing infant testing so. The bit error rate in the, bit error rate ratio. Tester, continues. To be the principal tool but continues, to be the principal tool and you, spend just a couple of minutes. Kind of walking through how that setup, works and many of you are familiar with it but it's really important we understand, is to see. The context, of this so what this issue is okay. So we're testing bytes under test and we're looking at a receiver. Or input, and we. Want to determine its ability, to correctly. Interpret data, that has no, impairments, in it so the impairments, are within, the allowed range of the compliance, of the transmitter, in the channel and as, long as those impairments are within that. That. Put, meet that criteria, the, input should be able to detect them and again, you know in the past it would be air free now, it's at a suitable air level it won't be air free so, now we're going to put the bird in the into. The test setup so the, pattern generator, you know birth to have two and half they have a pattern generator and air detector, so, the pattern generator, its job is to create a data pattern, that.
Is Stressed and what stressed is that means that the impairments. Are calibrated. We we, have various, mechanisms in, the bird to, put. Noise and, jitter, stresses. Into, the signal and quite, often meaning, an external, channel or ISI, in. A test setup so that could be a circuit board trait and then, that stress signal, goes into, the device that we're testing the important we're testing, now. We you. Know whether or not that receiver, makes. Correct decision, is only. Known within that device you gotta, get that signal out some how to measure it okay so the way we do that is we loop back because. We. Need to get that signal back out to measure the the, BER so we, take the signal out and we route it back to a transmitter, in the device a separate, transmitter, and we, read transmit, that signal through, a clean, channel. Or very clean link, into. The air detector, of the bird that's the other half of the birds so. What the bird of the air detector, does is it has a replica of the test pattern, and it, aligns, it in time where the incoming loopback, signal, and it, compares bit by bit or symbol by symbol depending, if it's n or p.m. for with. The replica. Of the test pattern counts, the total number of bits or symbols and the total, number that were incorrect, and does, the ratio to determine bit, error ratio so. That's the way the bird works the, key is that part. That says clean loopback link, you, go and pass these links ran air free and now by, design they're not air free so, what happens if we have errors in that well any air where my arrow is drawn, at the in. That signal caused by my. Any corruption mechanism. Will, be counted, by you know the air, detector, so, the air could actually be at the employee, air detector itself so, those errors, are added, to the. Real errors that the DUT experienced. And once, we have that we can found results, because we're going to overstate. The. Air raid, because. We're going to have additional errors. Produced. By the input of the air detector, okay. So. Now I notice, I move the air I should, set that clean link the clean link now with. P.m. for really, goes beyond just the cable we tend to think about just use a short cable to, get the clean link but. It has a lot to do with the retransmits. And then the device under test itself, and transmitters. Themselves. And pam-4 are quite, often a contributor. When you look at these link, margins, air.
Margins A big. Contribution is, from, the nonlinearities, and noise and distortions. And transmitter, itself so. You. Know that's beyond the control of, to be able to assure, that that transmitter, is clean so. You know the question is will, the loopback link, be, clean, enough that B air free enough to make, an accurate measurement. So. Let's look at some test scenarios, and in. All of these in the walk through a series of cases and I'm going to be testing a module, an optical module and I'll be testing an electrical, input so it's the transmit employed it receives an electrical, signal that drives, the the, total so would be driving a pixel or you know go through a laser driver modulator, or whatever depending on the module configuration and, in. A test setup we'd have a compliance, test board and, when. We start with a 100, G K so this is a legacy 100, G and these. Links are designed to it run air free so, we have the bird pattern generator just as you saw in a previous diagram. And because, the link is designed, to run air free okay that's what we're testing at that. TX. Input again we, use quite often we will use the same module we'll just take an optical short optical, patch cord and wrap. It around from, the optical transmitter, out put to the receive input, and then, we take that signal and feed, that back into the birds in the air detector, of the birth and, this. All works because of, the values and like now as I, just said earlier that you know aggressors. Are important, and for the sake of simplicity I don't show them but to do this test properly. We would have to have aggressors in it okay. Now. Let's look at the numbers, in this so if, we're looking at the standard, in 100g, case the chip 2 module the, the, BER requirement. Is essentially, air free it's 1 in 15, is not practicable, to test because that takes days due, to measure so it's. Tested, to one eat well that's what the standard says we test at 2012, and we're, looking for air free operation essentially is air free so, now, if I take and I same, standard and I look at the optical, loopback or. The optical path plus, the chip 2 modulo the entire link again, is, 1 minus, 12 that's what the bird criteria. Is so. Essentially that's air free so our loopback path in this case is air free and if it's a good module, then.
Then. We're not going to be adding any additional errors. You, know as, long as we have clean set up in a short cable running between that compliance, test board. And the error detector input, then, this, method, is going to work fine and this is the method you've probably been using for years okay. So what changes in a fam form okay so, let's look at that it all gets into the number so it's the same same test setup and initially we're going to try it with the optical loopback, okay. But remember that these don't run here free so, the test criteria for, the chip two module, on the host side is only. One a minus eight now quite often they're a bit better than that but they're not air free and so. So, we're going to have some errors in that and that's what we're trying to measure again that DX input and we. Do our optical, loopback well. We take the same standard, we look at our 802 that three BS and we look at all the optical, links optical. Plus chip two module link the, BER requirement. Is 2.4. E to the minus fourth so, it's worse, than. The signal we're trying to measure we could potentially have more. Errors, in the loopback path than, we actually have in the in. The signal that we're trying to measure and I was working with a customer just a few weeks ago that was trying to do this and initially, they were trying this technique with the optical look back and wondering, why it wasn't working and and this, is this is the answer right here so. So. You might think well ok you know that's because, the Rossa isn't that great we know the receivers, have are challenged, and for pam-4 so. Let's use an optical, or an instrument grade OTE converter instant. Grado de converters, they have a lot lower noise floor it, might, be higher bandwidth, it might have two little filters, a lot of things that should clean that signal up can, we still make that measurement well, remember. I was showing when I showed that arrow getting bigger on where the loopback path is it includes. The transmitter, so. Yes. This probably, does have a bit better BER, performance. But. There are transmitter, distortions, are can be linearity, to be signal noise ratio problems. Since, the Vic so you're going to have some skew in the eyes you might have too much skew so, all of those things could, contribute to causing, some errors at the transmitter, within the module, of where, the optical, signal is being transmitted. So, this technique, you know any technique, using an optical, loopback. Is, suspect, in in the module that's that's a problem and if, you already see where this is going. Can say that the existing, you know modules, if we don't have some special, stuff build in the module, testing. This particular, lecture little input is going, to be very very difficult to do or impossible, to, to. Make this measurement okay. So. Now let's look at another approach, and some of the electronics. That we've seen and this customer, that I was working with in Asia they, had this very same problem, and, they had they did have this capability in, their you know after they tried the optical, they, try the electrical, so they have the ability in their their, interface, electronics. That drive the tow serosa with. The those, side their. Interface electronics, have a test, mode that. Allows a host side loopback so it takes the TX input, signal and rather, than driving the. Pixel. Driver with it they, wrap that signal around into, the received, data output, and they. Retransmit. That signal back okay, and so that's what they were trying to do and. They. Weren't getting very good results okay. And we, were able to work with that customer, and and clean, some things up and we actually did get results that were passable, but, you, know it's. Very iffy, and here's the things though. The problem is even with this host side loopback, remember. That that chip to module, a link. Is only specified, to work to 1e minus, 5. And, you know quite often we see something better than that but we see numbers in 20 minus 7 20, minus 8 is is. On the high end of what they run and and. A big a big artifact. In that is is a connector, so so. You know this technique might work but. You, have some caveats that have to be in place the. Transmitter. Has to be really good you have to have a very linear transmitter, very low noise very. Little skew in your pixel the. Obviously. The channel, has to be very short it's always the case when you're using a an air, detector, but. The, other, thing that gets, to be a problem with these modules is reflections. Reflections, going, through the module connector to the module. Compliance, board will. Cause a problem and in their testing, they actually. Didn't have this bundled into the module yet it was still on and. Valuation board neighborhood will access, this signal with a cable so, they were eliminating. That that connector, and using cleaner connector, and we, were able to squeak, by and get it to work but I suspect in the final module you, might have a problem because the reflections, caused by.
That By, that connector, and then, of course you have to carefully equalize, the. Oh bird input now you have a powerful equalizer, in there and it, allows the manual tuning and, with some modes we can get in and we, can see what's going on to do the tuning and by, doing all that in this particular case wherever, we get acceptable, results, but. The big question is how can you assure this you know how do how, can you assure that, your loopback, have. Really, is not introducing, significant, errors and distorting, your measurement. The. Proper way to do it in the best way if you have the hardware to do it is to build an air chapter in the device okay. So you. Just illuminate the loopback path altogether so, in this case you would have an error checker, monitoring. The the, TX receiver. An input and most. Of the silicon. That does this actually, has two air chapters as one in each direction so. It can monitor, the optical signal as well so. Now that you, know again, are required, measurement. Is when you minus five but, we're not using the loopback path, we're using the management interface in the, module, so, the bird is being used as a pattern generator, but the air detector is not being used in the bird right now so. That's so we get the the air cow from the, air counter, built in the module itself and. We. Do that to the management interface now one of the problems is that you lose a lot of the automation. The bird has in making this measurement and. You. Know if you're using our bird with our software, with our ma t70 software, we've, alleviated, this problem because we're seeing a it's, starting with the host chips we're seeing a lot of people going air checkers inside and so, we put some tools and I'll cover that in another, slide as to what, the advantages, of those are so. Let's talk about the pros and cons of using internal, air checkers even if you have them some people won't like to use them and there, are there are some trade-offs in it now certainly because, of this loopback issue, it's great it eliminates all that problem, and if you have one we really, you use, the internal air checker that's the best. Way to go but, the challenges. Are you know the advantages. Are you're not going to overstate the burr because you don't have a loopback path and the, other advantage is that you you're not just doing a component test think, about it if the air checker is built inside we, don't need that module compliance, board anymore right so we can actually do this one system so if you're a system integrator you're, building a switch card and you, have your host chip and you want to test your circuit board layout make sure there's are any reflections, on the circuit board they're too bad and go, through the connector you can test the entire link, now and measure. That burr you, know with the assumption, that your host chip can generate. The the correct test pattern which, most of them can but, what are the downsides. Well the biggest out side is that you're limited on the selection of test patterns that you can use you, have to design the air checker into, Silicon when, you when you design the chip and you. Know many, of them have allow, more than one pattern. To be used which is useful but, maybe you want to test, with a specific pattern because you're looking for a pattern. Sensitivity. Or something like that you. Might not have the ability to do that with the internal air checker and. It. Requires, a special test mode to go into the air checker so that's not the mission mode and just, because you're having to reconfigure, the data path with, inside the chip itself you, might be missing some things in the data path that, a loopback signal, would show, you know we see problems, with errors, being introduced elsewhere in the in the, data path and then, and, then you lose the ability all, these automated measurements, such as a jitter, tolerance sweep, as I said if. You're relying on the internal error checker you, don't have that capability to, subvert but, now with our new, bird control software we.
Have This mode, called the duck. Control interface, and it allows you to access those internal, air, counters. In your duck and through. The management interface if it's a module and then. Integrate, that and essentially, it substitutes your, air counter, for, our air detector so all of measurements, such as jitter tolerance that, are built into the bird are accessible, to you with. It with this control and it also is handy, because it allows you to control other device registers, once, we have this interface, in place so it's it's it's just nice for troubleshoot. In general. So. Now how can we validate, the loop and make sure that the loop really is error-free you know because I said if you're, going to rely on on the loopback, method, you, really have to validate you have a an air-free. Load. Or sufficiently. Air for you might not be air free it might be Yan said you're measuring 1u minus 5 and you're, measuring 180 minus 11 or something you, know that that's close enough that you're several orders of magnitude that, you're not going to influence. That we're measurement that much but. But. Basically you have you know I went through all these criteria you have to have a clean transmitter, you have to have low low, reflections, you have to have a good. Low. Loss channel, and so, we really need to validate so, here we are that it's my blue arrow that I put on that we need to validate at length because, we can try to use loopback and we need to validate that, really is there free so. The ideal method, to do this is, once, again requires, hardware, in your chip and that's, to put a pattern generator, instead of an error checker or a pattern generator is easier it takes takes, much, fewer gates to. Build one you can build something, on based on PRBS or q PRBS with, very very little gate, count doesn't eat up very much of your real. Estate but you could put that in in your device and if you have the ability to put a test pattern into. The transmitter. That is providing. The loopback path then. All we have to do is we need to know what that pattern is we, program, our air detector. To. Use that as a reference pattern and then we're actually validating. The real, loop-the-loop. Back link, that, you're going to be used using, for your bird measurements, so if this, measures air for your sufficiently, air free then, you know you're good to go, okay. You, might not have that method you might you might have an internal error checker but, you know again you want to test with a different pattern or something so um. You, you're using this mode with the link. Side loopback and you're, comparing, the results now. Between with the internal, error checker with, measurements, in the, air. Detector, it with inside the bird itself and this happens to be the moment I mentioned I was working with this customer in Asia a few, weeks ago this is exactly, what they were doing and and they were reporting that their internal error checker was, reading a much much a lower, air level then, the Bert was and what's going on well we did some troubleshooting and all the problems were errors introduced, in their their, loop backlink so that's the that's. The alternate but in both of these cases you, see you do need to, do interviews methods, you do need to have some additional hardware built. Into your debt so if you're going to be building modules or or host, ships host, ships is required by the standards, but modules. It's not but. You really need to think about this too if you want to measure the BER and you know individual. Parts of the link you, really need to put these additional, air. Checkers and pattern generators, and into your device when you design silicon. Ok. Now I'm going to spend just a couple minutes real quickly talking about crosstalk, aggressors as I, said we need them for pam-4. Because, the reduced signal noise ratio we, we lose 9.6 DB when we go from energy to to, pam-4 and that bites us in a lot of ways and now, all of a sudden it's really important that we have those we, need them for both optical, and electrical testing, I've actually, seen, it I what I wanted to put it in the slide apologize, for that the, the material, I saw was copyright it was valuable your copyright, permission to, duplicate.
It But I actually saw, an example with a real optical, module, where, we were making that with the measurements, that Rob talked about they're measuring the optical output, measurements, from. The pam-4 signal and, when, they put a received, optical, signal back into the module, in the, in, the on the, received, path that, actually, close the I measurable. Amount a good, amount of light closure so it's very important that you do it ok, so what are the requirements for investors, when we look at the standards, they, say that the. They, basically. Emulate. The ideal transmitter, so they're going to have the minimum arrive, time and they're going to have the maximum amplitude and, they, have to be at the same data rate and they. Say but they can't be phased along they have to be asynchronous, to. The clock, that we're testing and why, is that asynchronous necessary, this is really a key point a lot of people missed this because, if they're using a bird they, have the you know clock output, and they want to use that as the clock to drive the the. Vesser, well here's the problem okay, so let's let's look at I'm going to show this with energy, just for see it's easier to see but it's the same effect with pam-4 so. Here we have an eye that's in the victim Channel that we're looking at and we, put an aggressor, and aggressor. If it's static in phase you. Know it's good a couple energy at some point in time is going to cause possibly. Cause some eye closure based, on how much coupling we have and, you. Know we're in in, time does, that closure occur, it's. Only a problem if it occurs where the receiver you know what the slicer is looking, at the. The, input to evaluate, the input state so if it occurs in other places in the eye like near the transition, as I show here it's, not a problem at all now, you might say well my real device, all, my multiple lanes in my back channel they're all coming off a common clock so they are going to be synchronous, so, what's the problem well the problem, is your, test environment even. Though it's synchronous clock may, have different delays because the cable lengths are different than the trace links on your circuit board so you know in a test environment you might have a situation I show on the left but. In your test environment the, delay is such that it's happening right at the slicer, and it's causing a problem so, by putting asynchronous. Clock requirement. In that. What that does is, that put, some face lift because you have a few parts per million different, in any two different clocks so. Every time that goes, through, when. We get that the next stressor edge is going to be a slightly different point in time and you can see it's, going to slide through the entire, the. Entire unit, interval and eventually. It's going to land up with where the slicer is and if we've got a problem, you're, going to catch it so that's the reason the standards so you need a synchronicity. Now, that, might be a problem you know that's expensive I got to buy another clock source you buy another pattern generator well, you might not necessarily need, to do that if you have a multi-channel. Bird and your. Bird allows, you on the second channel to have a separate, separate, jitter. Profile. Which some do and some don't our. New berta for Pam Ford does have, this ability then. You can actually get away with a common, clock and all you need to do is dial the sinusoidal, jitter in that with, an amplitude greater than half a UI to, guarantee that you are you. Know passing. That transition, as address or edge to the entire unit interval and you, know program the lowest frequency, you can for, your SJ, frequency, and, and. That, will suffice for the asynchronous, because again what we're trying to do is just whip that edge through well. Now one last thing on speak, on this is use, pam-4 as, the aggressors, some, people use in RZ you can use energy but. You're greatly over stressing the part so if you if, you test, within RZ and you passed fine but, if you fail with energy, you might want to you, know find a good p.m. for aggressor. Waveform and use it because there's much lower high frequency energy as you can imagine since. The distribution, of transitions, are they're not all 0 3 3 0 transitions. In a pam-4 signal. Ok, so uh, so. It's lasting and I don't just, wrap up showing our tools for this our birth, that is really optimized, for 400 GS was designed from the ground up, to do 400, G claps it. Does a native, pam-4 as well as nrz the, pam-4 is not created, by taking, two energy, signals and in analog. Combining, them in some birds do but, we actually do a native generation, so that's a gives, you you know you don't have any any calibration issues, and EQ issues.
To, Deal with that we, allow your direct programmability, of linearity so you can do linearity, test with this and full. Range of the emphasis, and all the capabilities. You expect in a full feature in RZ birds are present. In in, this bird as well and on, top of that I show a four channel arbitrary waveform generator it's, a real good source, for aggressors, and then, in some of these standards, you need to inject random. Interference at the far end of the channel and so, how do you do that where you need a, and. You, have sufficient, crest. Factor with, an aw g based noise source to do that it's, a more than adequate for all the standards, to meet compliance so is a very good way to to. Do the testing. Okay. And now reps wrap up to our presentation, so let's go through the summary and, let. Roth start out with assurance, thanks. Big stuff ad so just to summarize, as. You've. Seen with pam-4 there's a number of new measurement, challenges that, have. Been introduced, and we are continuing, to discover. New, challenges. As we turn on lengths, so be. Prepared for, additional. Changes down the road, key. Optical, transmitter, measurements. Include tdq, outer, oma and outer ER in order to perform. These measurements, very quickly and accurately and, repeatedly, I will. Make sure that you're using a solution that has been, optimized. Per, the latest standard. And that. It has the, ability to create an ideal, reference, receiver using. Impulse response, correction, and that you're starting out with a very low noise receiver. Electrically. Key measurements, are output jitter that's the j 4 UJ r ms + EO j as well as s ND r and i symmetry, mass quit and. Make. Sure that when you are making those output jitter measurements, that you're focused, on those 12 edges and we. Do have new capability. There as well, okay. And. On receive side. Quite. A bit with the. Look. Back path now you know your traditional method you've been using a burke for years and years may not work with these links that do not run air free by design. So you need to take some special caution if you do have to use move back to assure that the new back path is, the. Errors of control of in it and then. Internal. Error second capability, you know that's that's a big way to eliminate, the use of the loop back and with, our m87 t birth control software we. Make that really easy that, you don't, lose your jitter tolerance measurements. Just because you're using an internal error checker and the. Use of crosstalk, agressors is absolutely, imperative not only in in the, input testing with births but the output measurements, with the with, the Scopes we, really have to have those in place and. For. Any more information you, know we could give you a whole list of things to look for on our web but the easiest way to find is just remember one thing keysight.com, /, line / pam-4 and, that, will give you the list of all of our tools and solutions for. Protesting. In this application, space. Okay. Now on, to some questions. Here. Is a first one if, the place I can, get the PRBS. 13q. Test pattern. Understand. Question is, it somewhere, we can get the PRBS, 13q. Test pattern so that's built into our building, with a bird the bittern generators, and if, you're asking where you can actually get, what. The pattern definition, is it would be in an appropriate, standard socket that would be defined in 802 dots three BS. Okay. And we have time for one more question. When, Wilbert, m84. T support, FEC, encoding, and decoding oh. That's. A good one um that's. That's a difficult thing to do with versus particularly multi channels, because of the as you probably are aware there's. A very, very high gate count to to, implement that but we're working on that we're, going to start out with a with. A. Single. Channel implementation. You know for the signal, lane variance of a CD. And then you'll see that moving to multiple, channels in the past but it's not, give anything, short-term, just just just you, know hold, tight and look.
In The in. The next year sometime and you'll see so Mike is either. Okay. Thank you now we will reply, to any, other funny unanswered. Questions, by email within the next day or two and that. Concludes, today's presentation. On behalf of electronic, design, I'd like to thank esight response and today's event and of course all of you for joining have, a great rest of the day.
2018-10-17 18:13
