John Lynch: So, I'm here at ECOC 2024 in Copenhagen with Julie. Julie, could I ask you, would you mind introducing yourself and what you do? 

Julie Eng: Sure. Hi, I'm Julie Eng, and I'm the CTO of Coherent. 

John Lynch: Great, thank you very much. We met coincidentally this morning at breakfast. So can you tell me, how did the day go for you? How was the first day at ECOC for you? 

Julie Eng: It's been busy. It's been busy already so far. Let's see. After I saw you, then I went to the plenary talks. One of the interesting things was Ed Lee from Nvidia gave one of the plenary talks, and he was talking about CPO in the scale-up domain. In addition to silicon photonics, he actually mentioned VCSELs and also DWDM, so that was super interesting. And then we, interestingly, are having a quantum key distribution demo, so I went to see that. And then I gave a talk. 

John Lynch: And that leads into my next question because you did a really interesting presentation on co-packaged optics. For someone new to the technology like myself, can you tell us what are co-packaged optics and how do they differ from the pluggable optics that we've been using for years? 

Julie Eng: Yeah, great question. So, the pluggable optics, as you know, you have a switch or server, and the pluggable transceiver plugs in through the front of that switch or server or router. What the co-packaged optics is, it's really like a linear pluggable optics version, so with no DSPs inside, exactly the same transceiver. You just make it smaller, and you move it inside the box that's next to the switch or XPU chip. 

And now you say, "Well, why would I do that?" Because I love the fact that pluggables plug in through the front, they give me flexibility, I can replace them if I need to. The main reason you would do that is then you don't have to drive that electrical trace to the front of the box. And driving that electrical trace to the front of the box consumes power. So you can get lower power by putting it closer to the switch or XPU chip. And the second thing is the front of the box, you usually have to plug in the whole transceiver—optics and everything. And if you look at moving it inside, now the only thing that has to come out through the front is the fiber. And so you can get higher bandwidth density. 

So in applications where having higher bandwidth density and lower power are more important than having the flexibility of the multi-vendor ecosystem that pluggable transceivers offer you, that's where you'll see people using the co-packaged optics.

John Lynch: My next question is a little bit selfish. So we've been your distributor now for 25 years and we've been selling mostly pluggable optics. So my question is, what do you think the future is for pluggables now in the new CPO world? 

Julie Eng: Oh, that's a great question. I think CPO is a new technology and a lot of times when new technologies come along, people think they turn on immediately and all old technologies turn off immediately. But I showed a market model today in my talk, and what we show is we believe the pluggable transceivers continue to grow through the end of the decade, and they remain the biggest part of the TAM, $25 billion by 2030. CPO starts from zero, it grows faster off a much smaller base, and at least our market model is $5 billion for CPO by 2030. 

So you still have a huge, huge usage of pluggable transceivers. And the reason you have that is what's so valuable to our customers is being able to plug everything in through the front. If you have any kind of failures or anything, you move it out, you put it in. All the modules look the same, so you have very big security of supply by a multi-vendor ecosystem. And then the most important thing, I think, is because you plug them in at the end, you can actually make architectural decisions after you've already developed the switch or router. Even once it's on the data center floor, you could decide, "I want it all to go 30 meters and be VCSEL-based transceivers." But maybe it's going to go 500 meters and be silicon photonics-based transceivers, or 10 kilometers and be Indium Phosphide-based transceivers. And our customers can make all those decisions at the end. So in the portion of the network which is where you value that flexibility and multi-vendor ecosystem more than that last little iota of power or bandwidth density, you're still going to use pluggable transceivers. 

John Lynch: AI is the buzzword of the day. So how is CPO designed to address the needs of AI and ML workloads like bandwidth, latency, and scale? 

Julie Eng: Right. Well, I think what you see is the whole data center is growing with the AI and ML workloads. And then inside the AI portion of the network, we usually separate it into "scale up" and the "scale up" is networking multiple GPUs or XPUs together to act like a single compute core. And then the "scale out," which is the networking together of all those compute cores to form essentially a supercomputer that is the AI factory. So that scale out portion is all optical already and all pluggable transceivers. What's new is that scale-up portion is on copper, but because of the push of AI, those processors are getting to be higher and higher speeds and more and more lanes, it's getting harder and harder to actually serve that with copper. So we think one of the main usages of CPO is actually to replace copper in that scale-up domain. So that's great for all of us because that's a TAM expansion. 

John Lynch: From an engineering perspective, you're the Director of Engineering for one of the biggest optics companies out there. So what are the major technical challenges that you encounter in your job? Like how does Coherent continue to increase bandwidth while trying to maintain package size? 

Julie Eng: Yeah, great question. So I would say the biggest challenge right now is actually that the speed with which we're changing data rates is getting faster and faster and faster. So if you think about it, I made a chart once that was the speed of the data rate as a function of time, and I could fit it to an exponential. I mean, it's just growing exponentially. So what it means is we have less time to innovate between every data rate. So that's a big challenge. For us, it's an opportunity because we're a tech—we're a high-tech company with all of the technologies in-house. So normally when the technology splits faster, that benefits the technology leader. So for us as a company, that's fine, but it's still a challenge. 

And then we also have to ramp them in manufacturing faster than ever before. Like I looked at how long did it take from when we shipped the first 10,000 units as a whole industry, till we had to ship 10 million units. And at 10 Gig, it was like 10 years. And now it's like three years. So all of that coming faster, faster, faster is a challenge. But it's a good kind of challenge to have if you're a high-tech company like us. 

John Lynch: And last question, Julie. Outside of CPO, which we've been talking about, are there any other exciting trends you see in the industry?

Julie Eng: Yeah, sure, sure. I'm super excited about the optical circuit switch. That's a new product, a box-level product, right? And I think we have a very differentiated offering there. We're seeing expanding customer interest and expanding use cases within customers. It's very new. We estimated that to be a $2 billion TAM by 2030, created from scratch, from zero. So that's good. 

Outside the data center, I'm interested in the DCI because that's also pushing the DCI—the fact that you have so much data that has to go, you know, sometimes the data centers are power constrained, so you're putting the data between multiple data centers. So DCI, and even the transport layer is starting to get to be taxed by the AI system. So I'm interested in all of those areas. 

I'm also interested in the data rates going up because the way to keep a robust roadmap for pluggable transceivers is, you know, you have two choices. You have the transceiver. You either raise up the data rate of all the components and/or you increase the number of lanes. So we're at eight lanes of 200G now, so the next step would be 400G. So like at OFC 2025, we showed our 400G EML. So we're at the bleeding edge of the new components to support the future data rates. So I'm very excited about that too. And I'm also excited about CPO, even though you said besides CPO.

John Lynch: Listen, Julie, that's super. I really appreciate the time you gave us. I know you're super busy and getting time with you is not easy. So thank you very much for taking the time to have the conversation and enjoy the rest of the show.

Julie Eng: Okay, thank you very much. Great to be here with you.