MALAGA Keynote 8: James Watt, President Nokia Optical Networks

Thank you, and thanks to TIK and Jose for organizing this fantastic event. And special thanks to Jose for putting me on after you all had your coffee, not before. Nothing worse than being between the audience and their coffee. I'm going to talk about telecom today. I feel the need to narrow the scope since we've had such a wide-ranging set of wonderful conversations. Some days this is the stale old use of photonics, but it's still a very, very interesting part of the world for a bunch of reasons, and I want to get into them. If you look at optical networks these days, you can't get away from them. In fact, I would argue there's nothing you can do if we took all the optical networks away. We've got, I don't know, five billion kilometers of fiber. People suggest that's going back to the sun 20 times. And yes, we all have mobile phones, and there's some people who have probably not used a corded phone in maybe even their lifetime. But in the end, all that traffic comes back to optical networks. And so they're the foundation, if you will, of everything. And that's what we spend our time on. We spend our time on making sure those networks are there so that everything else that we talk about — whether it's IoT, whether it's AI, whether it's connected vehicles — can work. And you know, every time my guys have connected vehicles on a slide, I look at it and I say, 'Really? A fiber to every car? Are you serious?' No, they just mean there's going to be radio somewhere, and those radios have to get backed. And you know, 98.5% of the time that backhaul is fiber. So there's a lot of growth, a lot of pressure on these networks. But let's talk about where they came from first.

You know, light for communications is not new. I chose when people started using light — you know, 1084. One of my colleagues would say I'm a bit late with that number. He would say that smoke was somehow the absence of light, so we should start with smoke signals. But either way, it's been a while. The physics we needed — Chuck mentioned this earlier too — some of this stuff is not new, right? The physics is pretty old. Kepler, a little kind of important. Snell, you know, we need that in the middle. And yes, it's copper, but there's other waypoints along the way. First telegraph — I like this one. As you can probably tell from my accent, I'm Canadian. And one end of this was in Newfoundland. It was still part of the UK at the time, but it's Canadian now. The other end was in the UK, and the first letter was S. Why? Well, amateur historians in the crowd can answer that later. But you know, people got used to having reliable long-distance telecommunication.

Some inventions came along the way that helped. Semiconductor transistors — I'll get back to why they're so important to me, to my part of this, even if photonics — we worry about other things. Again, Bell Labs to the rescue. Shannon's theorem — Shannon-Hartley — yeah, I'm going to come back to that one. That's more of a problem for me today than a solution, but it's no less important than the others. Laser, key. Optical fiber, you know, '70, and where did that get us? Okay, first fiber optic installation, Chicago. These things were, you know, limited distance. As I explained, somebody had the 'how do you explain the ICS to children' comment. Everybody asked me to explain amplifiers, and I start with a flashlight analogy. You know, when you go outside at night, the flashlight only goes so far. But these were pretty important. Speed, yeah, by today's standards they weren't great, but they were huge, displacing copper. And that really started to move things. Then we got EDFAs, which really got things into gear, right? So 1987 — things, you know, again, you start charting that thing, you get another tick up in the curve, and that really got things moving. And then around the turn of the century, just before, René gave us pseudo-linear transmission, which kicked off the next set of progress, at least in my area. By the way, there's a lot of other things. You know, the comment about standing on the shoulders of giants — we're standing on the shoulders of villages, really. It takes a lot underneath.

And where have we gotten? So state-of-the-art — this was earlier in the year — 800 gigabits per second on a single wavelength, 6,600 kilometers. That's, you know, again, terrestrial, you can scale, but this is submarine cable — real world, real cable, real traffic. That's pretty much what you can do. And yeah, we can push it a bit further if you shorten the distance. We all know that phenomenon. But this is where you get with the cumulative separate efforts to date. And by the way, as one of my colleagues likes to point out, that telegraph I pointed out earlier — this is about 8 billion times the capacity. So if you want to know how far we've come in 125 years, the answer is 8 billion times. And yes, engineers can reduce everything to a number. Again, this is coherent transmission — a lot of interesting work on the modulation scheme, a lot of interesting photonic work to get it to work as well.

And how do we get there? What's moving things ahead? There's a few pieces — actually, there are really a lot of them, but I only have time to talk about a few or else Jose is going to be upset with me. One is we've taken networks apart and put them back together. And why did we do that? No, not because we were bored, although that's what some customers suggested when we first wanted to talk to them. Because we could get better networks out of it. We extracted the control plane, we've got things decoupled. We have still preserved the end-to-end nature of transmission, but we've got what I call good disaggregation. Focus people on the right problems. Focus people on the line. Focus people on the transponder. Yes, make it work end-to-end. And by the way, when I say make it work end-to-end, not only is it 800 gigabits per second across an ocean — by the way, it wasn't just across an ocean, as you saw, it had to go inland as well — but there's a few other properties we have to worry about when we want end-to-end. We need availability at a network level, that means there are protection mechanisms that need to be somewhere. 5G — somewhere. That's one of the traffic drivers. It's got some rather exacting requirements on network synchronization. Oh, okay, we need to do that end-to-end. So to do all that, yes, you need focus. You need focus on the pieces that matter. And the whole value of disaggregation is not that there was more room for people to write code to put it all back together — yes, there was — it's because we got the focus on the pieces we needed to get the right result. And I'm going to come back — there's a reason I point that out.

We got power out of the system too. Maybe not enough in time, because if I look at the power consumption going up in other parts of the ecosystem, it's going faster than I can take it out. But we took power out. We took power out of the transponders, we took power out everywhere. How did we take power out? Well, again, Essie Miller had this idea — not yesterday, by the way, 1969. You know what he thought we should do? He thought we should have photonic integrated circuits. And as somebody said earlier, maybe you don't always need them. And he's right, right? Premature integration is like premature optimization in the compiler-writing space — don't do it. But this is how we did it. We took things, again getting the separation of concerns, taking the pieces, getting the focus on the right thing. Okay, optical front end — yeah, let's integrate that. You see the pluggable there. Let's take that as a proxy for a 400-gig, or maybe tomorrow an 800-gigabit-per-second interface that goes 500, maybe tomorrow 1,000 kilometers — round up or down the numbers to match your favorite data sheet. The first 100-gig coherent interface, I can tell you, was a three-card set about that wide, that high, and that deep. Now that's 400 or 800. So if you want — when that was 2010, so 15 years. So you know, 125 years we got an 8 billion times improvement. 15 years we got a massive reduction in volume and took out 90-odd percent of the power. Still not enough, by the way, if you do the power-per-bit calculation, we still got work to do. But we've made progress, again leveraging science, engineering that's been going on. And going back to the comments earlier about how long it takes these things to materialize — yeah, they don't happen overnight, but they do add up, and they add up massively.

What else have we done? DSPs. The other thing that happened somewhere between 1977 and today is we moved a lot of the processing — and I know this is a heretical thing to say in this audience — but we moved a lot of the processing into the electrical domain. Why? Well, as Matt pointed out earlier, it was going down a learning curve like crazy. We took digital signal processing, we took the math. Remember 1998, pseudo-linear transmission — the digital signal processing, the continued growth that we could have for a set cost, for an amount of power, and applied more and more signal processing. Now we're in trouble. We got a bit of a problem. Another Bell Labs guy — it's always a Bell Labs guy — back to Shannon. We've hit the limit, right? We can fill a wavelength. We can get what we can manage into a wavelength. We figured that out. Yeah, there's a little bit here and there, and we can't always get it to go as far as we'd like. But we started to hit limits, and this is where the tale takes a turn. Now, there's still a lot you can do. Everybody knows that the goal is not just filling a wavelength, right? The goal is getting data, information, end-to-end. So now, everybody here's immediately, 'Oh yeah, but I know how to solve that,' which is good.

By the way, the other thing that happened — as we got great at putting data on networks, and by the way, not just the optical part of the networks, the IP part, all the others — these networks got really, really big. I have customers that have networks — thousands and thousands and thousands of network elements. And if you want to think about what that means in terms of effort to manage that network, you know, all the concerns we all of us have about the amount of labor, manual labor in a test sequence — it's the same problem. So if you think about all you would do to get that manual labor out of the test sequence, yeah, well, we have the same problem at the network level. The good news is there's some thoughts around automation and AI that can help, but it's a real problem to be solved if you want to keep these things running. And by the way, the only time civil society usually notices these networks is — and by the way, not just optical, data, mobile, whatever — is when they stop working, right? You don't see a headline in CNN that 'British telecoms network has been up for the last 3,000 days.' But you take Verizon, AT&T, Rogers out for what was it — 12, 14 hours — there'll be a lot of headlines, there'll be a lot of ink. This stuff, you can't go anywhere without it, so it's got to keep running.

Quantum computers — I was very relieved to hear earlier somebody's comment on how soon we would or wouldn't have how many qubits. The only problem is that in the meantime people aren't waiting, and we have a little bit of harvest-now, decrypt-later activity going on. So all this optical stuff we put out — and I think everybody in the room knows how easy it is to get some light off of fiber if you need to — we kind of made a big, attractive attack surface. Also, as I said, nobody knows this when it works, but the regulators are noticing the importance, and we're starting to have some help — at least my customers are starting to have some help. And that means: what happens with regulations? Privacy, safety, digital safety. And as I said earlier, sustainability — a lot of pressure on that, a lot of pressure on power. The good news is — and this is though where footprint and handprint go hand in hand — yes, we need to reduce the footprint. We still got a lot of work to do there, but as my CEO likes to say, 'There is no green without digital.' I tend to say, 'There is no green without optical.' It's the place where we can find the lowest power per bit, and that's the contribution we can make — not just taking that power per bit down, but expanding things into the layer. By the way, I say that in telecom, but you know, that goes for a lot of other applications as well.

So I didn't know, but I predicted that Jose or somebody would ask me, 'Where's all this going?' And by the way, how are we going to solve all those problems? So I looked out a little bit too far — I said 10 years, not five years. And I'm not a very patient guy. So where do I want this to be? We need to massively increase our capacity from where we are today. Why? All the drivers we live every day — all the data, all the small things, where people talk about life sciences, all the data people are working so hard to create has to go somewhere, right? So we have to get there — petabytes, exabytes. You know, everybody's going to have their own number, and it probably won't be mine. But we need — you know, we can do a petabyte per second. We need to get in the exabyte range. And how are we going to do that? Right, I said Shannon was in the way — you saw the stop sign there. Well, three things. One: people. To be honest, my biggest concern with getting to exabytes is whether we're going to have the people to do it. Now, the last two days on that front have been very reassuring in some ways, but I'm very concerned that we need to continue to attract the people — not just the researchers, but also the engineers — to pull this off the way we have over the last 15 years, over the last 125 years. It's not a given. Now, I have Bell Labs to rely on, and a team there who works to recruit the best and brightest researchers. But I've got to find — my teams have to find the engineers. And I can tell you — and it was great to see we had David here, you know, and academics here — because we need to find ways to have the right engineering talent to pull this off. That's number one.

Number two: we need technology, and that's where you all come in. Yeah, I have my list — I'll give it to you in a second. I know everybody's going to ask. But what I've heard in the last two days is going to offset my worry about people, because I'm more than ever convinced that there are people out there — some of you I've met, some of you I haven't yet, maybe some who weren't here — who are working on the things we'll need. And yeah, we may need new science, but we need a lot of other things, and that's where you come in. And as I say, you know, that's the answer to how we're going to get from petabytes to exabytes. It's a whole-of-industry effort, if you will. It will be a long — it's like your Visa bill at the end of the month. How did it get there? A little bit here, a little bit there, and it all adds up in the end. And that's what it takes. If you look at what it took to get 100-gig coherent single carrier working — yeah, we can point out some important pieces: some math here, some physics there, some engineering here, some cleverness there. But it was a lot. Even if you just think — we had a DSP. Think about what it took to get to the point where we could get all the gates we needed on that DSP. We built on that. So there's a lot. Now, I said I have my list, so — is it a shopping list? Yeah, sort of. Is it a list to inspire questions and head scratching? Absolutely. So what do we need? What could get us from petabytes to exabytes?

Fiber. Hollow core is getting a lot of attention recently. It has some nice properties that could help a lot. People get to figure out how to manufacture it, but that could help. My friends in the radio business love to laugh at me because I spend all my time — I make these launches, we have our photonic service engine — and they say, 'Yeah, but you know, you're only using one mode. Nobody — that's old-fashioned. You need to be multi-mode.' Okay, maybe they got a point. So can we do some — you know, coupled modes, not coupled modes — how do we — what's the processing to take that? There's another area. More bands would help, right? We've got two that we use in telecom. If I have a third one or a fourth one, that would give me a lot of capacity. What do I need there? Well, and here another theme begins: materials. Right? How did we get EDFAs? How did we get Raman? Right? What's behind that? Again, the science — been there for a while. How did we get those technologies in volume? So yeah, so materials work. Maybe we could have the E-band, maybe the S-band, maybe one that people haven't talked about. Right? Things we talk about for a long time — sometimes they go to fruition, sometimes they don't, and something else pops up. So that could help with capacity too. Modulation — yeah, I mean, we've got some fundamental physical limits. But now we're talking about how we cram as much as we can into a fiber, and we're talking about how we do that cost-effectively. Right? So yeah, maybe we could find some different ways — bigger channels, different modulation schemes that we haven't thought of. Again, you still got Shannon's limit, but yeah, maybe we can get better ways of doing things there.

By the way, just to get where we need to — I got mentioned, I think, earlier, but you know, this is a material science problem, right? We get ourselves distracted with the DSP, we get ourselves distracted with network architectures, disaggregation — but at the core, this is a material science problem. So yeah, the modulators — yeah, we probably do need new materials. Now, do I know exactly which speed needs which new material, or even how far I can push a material? No. Because every time I think we're out of gas, we find a way to push it a little bit. It's like inside the data center — every generation people say, 'Ah, InfiniBand is dead, that won't work, we need something else,' and we push it further. So yeah, materials. And the list goes on, but as I said, the real list is the set of people in this room, the set of people who aren't here who are working on things. Because it will take a number of things. We've got people working hard on it, but we know we don't do everything right. It is — and you know, one of the great things about a number of the talks these last two days was the use of the word 'ecosystem.' It is an ecosystem, and it's not always a finely focused one. It can be a broad ecosystem, but it is an ecosystem. We all leverage different pieces, we all contribute different pieces, some of us have different priorities. You know, quite frankly, some days I'd give a lot for 0.1 dB. Others, not so much. And that's good — we sift those things apart, make sure we get it.

But in telecom, where have we gone? We've gone from bright lights going about 500 kilometers to relay the fall of Troy, to the letter S at a very, very slow bit rate across an ocean, to 800 gigabits across an ocean on a single wavelength. So now you can think of capacity as well as bandwidth. And it's going to keep going. When I was looking at the first slide, I was thinking of the Billy Joel song where it starts with a whole bunch of historical events. That's what it felt like to me. And you know, we didn't start at Burning Man. None of us were here when this started, but we've kept it going, and it's going to go on and on and on. And I know why — everyone here, everybody working on this — it will go on. Thank you.

Sure. And look, it depends — you know, manufacturing is a hard thing. And the first question always is: what do you want, how much of it do you need, when? And you know, there's different applications inside the data center, right? Different span length than outside the data center. My usual joke, by the way, about hollow-core fiber in small audiences is: yeah, I wonder if we can figure out how to use that undersea? Because if you think about what happens when the cable breaks — how do we do it? So there's a whole bunch of practical aspects people are working on. This is something that, you know, we've had people in the labs working with teams that have been working on this for a number of years. And I think they have some good ideas. But, you know, manufacturing this stuff — and the fiber manufacturers in the room can chime in to tell me how wrong I am — is still a hard problem. But if they solve it, the properties are very cool.

You know, I think there can be. Now, the question is: why and where? And if you take a look at the challenges that got pointed out a few hours ago — the amount of investment to get through a learning curve on something of substantial complexity is just beyond a small company, right? And it's not just the monetary investment, it's the people investment, it's the diversity of skills that you need. Now, on the other hand, on focused areas, right, it absolutely is the case that much smaller organizations can do it. Now, the question is, as I said, where and why. And you know, this is true for any company. I think that the presentation on startups said it well: you need a mission, and you need it to be something — yes, there's a time component — but you need it to be something you can accomplish. And you know, if I look today at the set of fabs out there in the, let's say, the non-bulk-silicon space — so I take the whole broader space — there's a pretty big dynamic range today. So I think there's room for a dynamic range. The low end — I don't comment, I don't know. James, thank you very much for a great presentation.

Thank you.