video_OCP2026_Lumilens_NPO_ScaleUp與ScaleOut_20261015

All right. >> [clears throat] >> Good afternoon. I'm glad there's still so many people still stay. Um It's very difficult to follow. We have a so many elite speaker in front of me. Um And I also notice I'm the only person represent a startup company. There are gazillions of startup company out there and uh appreciate Frank Erica Jimmy. I so invite us to give a talk over here. Most of the technical part is actually already covered. And um we just want to share some of them our position. All right, so we are very focused um from day one MPO is the one of the product in our mind. All right, so eventually MPO CPO and I still remember 2 years ago we are only 2 years old. Honestly. And uh we launched the company couple months ago. But 2 years ago when we started at the very beginning many industry friends ask me so why you guys jump in to build another transceiver company? Uh we cannot answer. Um but now is the I think finally it's time, right? So people talk about MPO. So the the market um really grow super fast, right? We have a at the number over here. Uh some of them are I quoted from uh Light Counting maybe it's already old, few months old. But overall the market for the photonic interconnect market grow like a crazy for last few years. So it's a good it's a good for startup company, right? So the pie keep growing and uh every company the AI photonic company we have a chance to grow, right? We have a chance to innovate, to come out with a good solution. So, down to the a lot of people showed a very rise of all the network with a scale out, scale up, and all these interconnect. But for us, it's a very simple, right? So, scale out is all pluggable. Is it dominated by pluggable? It's been there very successful, keep growing. And uh but we see only last few months, let's say from this year, the new opportunity start to open up. And uh I saw all these different buzzword. We can call it MPO. I heard people call it pluggable CPO, the pluggable MPO. Um I also heard there's new buzzword, right? So, maybe the OIF define they call the MPO HDI or something. But the same meaning. So, it's the it's the MPO somehow the adding to the market for scale out. It's not competing with pluggable, but allowed a new market emerge addressed by the high density MPO. The scale up part from um last couple years, everybody know, hey, the copper may run out of a gas eventually, right? If copper can cover, keep using copper copper. Many, many different kind of optical solution emerged. So, MPO, CPO, it's definitely focused, right? To address the the the use case. So, for the the scale up for the uh with the MPO, so what is top level benefit? So, many speaker already talk, right? So, it's it's a density, is the power, is the cost. It's really on the top level. Can we apply the pluggable? Right, so this is there's a density issue. We cannot even fit, right? Even though you have all these different kind of benefit. So, the long-term goal for the whole industry is moving to directly move to CPO. But the CPO also have their challenge, right? How do we quickly ramp up? How do we solve all those yield issue? I control the cost, right? There's a many, many other issue. But from today to the really huge volume deployment, we need something is deployable today. So, that deployable solution for the scale up uh is MPO. So, power-wise because MPO rise is all based on the linear optics, we eliminated the DSP. Inherently, the power is a lower. So, people also previously show the the comparison with the tables. This is the one we can drive the uh the the picojoule per gigabit, right? So, to lower gigabit uh per per bit to the below 10 level. And the cost is all associated with uh when we have a higher density. This is the old days when we talk about we double the speed, let's say 800 gig to 1.6 T. And uh we could not just double the price, right? So, the cost for the end user is um is when we use a simple example, right? 6.4 T. The next MPO around the corner. That's a four times bandwidth compared to 1.6 T pluggable. So, with that, obviously, right? So, the price going to be much lower. Not times four time. Right? And then I want to also uh touch base for the for the scale out. Whether there's a room for MPO to play. So you can see this is oversimplifying model. Right? Typically from Nick. Uh the traditional models go to a top rack switch. And nowadays a top rack can be like a remote tour like in the middle of a row or the end of row. And some of network they collapse right with a uh that top rack switch with leaf and then you go to spine a super spine. So overall the pluggable still have its play. So the traditional pluggable even nowadays for the 51.2T switch um it's already proof working right started deployment with uh 64 by 1.6T. And then the XPO um we heard a lot of uh good promotion and uh it's it's a gathering momentum to extend the pluggable. Um really it's good enough to cover with the traditional DSP enabled or full retime or half retime TRO LRO all these different buzzword. But at this moment is whether from the pluggable we can directly go to CPO. Right? Industry is waiting. Industry actually is driving. So you can see that Broadcom and others like have been pushing for the CPO. But there's a limitation. This will take time to ramp to really achieve the goal. So the new emerging use case we call the MPO but as I said right so many different terminology college. Um it's it's it's more like Verizon is from Tomahawk 6 um perspective. They call it ICA, but it's just Right? So, it's it's MPO or pluggable MPO, whatever we call it. But it's it's kind of like a bridge the gap to allow the switch a build a common platform with a connector to support imagine a card or daughter card. And then we added a MPO or the optical engine on that card. So, that is the the use case. So, overall optical connectivity for the scale out we actually need more option. It's definitely not one solution cover all. So, we as we said, right? So, we need the traditional pluggable. We can see the IXP has its play uh higher density, especially when the network is moving to 200T switch. And MPO all these new defined MPO also has uh its advantage. Has its market play. And eventually when the CPU is getting mature CPU also uh will play a major role for the scale up. Yeah, I think this page is a little bit repeated, right? So, >> [snorts] >> at least two or three speaker before me already talked um for the integrated IRS versus ELS. Every company um have their own position, right? So, imagine for me they're the leader for the ELS P. And uh it's definitely can see the advantage. But overall it's related to especially for startup company. For us, we follow the customer guidance. And we follow the customer guidance. We offer the solution based on the co-design co-defined solution. And we strongly believe for this generation MPO if we can fit in the integrate laser with MPO this is a better way to go. Right? So, there there are many other benefit, of course, right? Advantage, disadvantage. But the major major advantage is how do we lower the cost? So, we have the integrate laser and we eliminate the PM fiber, the connector, the patch cord. And we also eliminate the laser power loss. Just think about it, right? So, when we have the It's a good news if I have a 400 mW laser, right? Uh I heard or even higher um power output power laser. But during this process we deliver to the modulator. They have a major loss. Um with the integrate obviously we can eliminate those possibility. Um also the small foot foot print. So, smaller size, higher density. So, all those are direct tangible benefit uh at least for my company's position, right? So, we support if we can use the integrate laser solution where we can, we do it. But some customers, hey, they specifically request they want to use a ELS due to maybe serviceability, due to some other reason. Uh we can support both. >> I think one thing most of the speaker didn't touch base on is um when we talk about MPO or eventually CPU, the most important part is actually the manufacturing. Who actually can control the manufacturing, right? Can improve the manufacturing, can can actually innovate a better way, right? To improve the yield. This is the way to win for MPO, right? Build sample, right? So we can do a live demo, we can show a very good data. Um but the important part is who actually can quickly ramp up. So this is all the way down from the chip level, like the PIC, the EIC, right? So how do we control it? How do we co-design to the advanced packaging level? What kind of new technology uh we can bring to the table, right? How do we further improve like a fiber attachment? How do we kind of like when we mentioned the the integrated laser? Is that a traditional free space laser coupling, like go through the dual lens, go through the isolator, couple into the side hole? Or there's a different way, right? Can we further reduce the laser number, right? Can we still take advantage of the high upper laser available in the market, and uh >> [clears throat] >> so we can we can come out what we call the the laser optical subassembly. It's kind of like it's a separate building block within the module, even though still integrate, but we can separately testing, assembly, right, to further reduce and then we do the burn-in and further reduce um uh the module build loss or improve the yield. And overall, as as a optical engine, how can we separately testing, right? And then we can do the fiber attachment, we can do other integration. So, these are the one and I think one of the speaker mentioned also um it's critical for us to co-design, co-development with the system house. Instead we set a full industry, full industry, where I used to set a demarcation point between the optical transceiver. So, this is what you want, this is what you define, I can deliver with lower cost, lower power, meet your requirement, and then I wash my hand, hey, I'm done. It's your issue, the system issue. We are different. From day one, we need to think about how do we integrate with the laser, right? So, with a laser perspective, I'm sorry, the system, the system perspective, right? How do we think about how do we integrate the signal integrity, the power integrity, right? The optimized thermal design. So, those are the things it's will decide which company going to win with the MPO or down the road with CPO. So, overall, we are considering is um um from the system perspective, the GPU rack level, the compute tree level, and for different kind of switch. So, most of the MPO, even a dress the scale up use case. Most of them actually is not sit together with uh with the the GPU. It's actually sit together uh closer to the scale up switch. So, it support high density cross rack scale up switch connectivity use the optical way. So, I guess um this is uh the last page we want to also mention a little bit um what's uh why so unique for Luminance. So, from day one we're considering the market is already very crowded for the optical connectivity, right? So, there's a so many different optical transceiver company over there that are doing fantastic job, right? So, they offered the industry with a good volume, good pricing, everything is good. But, what are we are considering from day one is um we build a team is more like 1/3 of the domain expert is a system domain expert, right? So, we have a lot of people from Cisco, from Juniper, from Arista, Palo Alto Networks, you name it. So, very strong on software. And um so, we come out this unique one platform to address the pluggable, the MPO, and the CPO. And the overall beyond the strong system expertise we also have a roughly maybe 1/3 of the people working on advanced packaging. They're not optical people. They're not system people, but they have been involved deeply in last 10 years. It's only a few company you can name it in the industry, right? Working on the advanced packaging. Of course, we have a a major optical domain expert from all these well-known company, right? The old days Finisar, JDSU, you name it, Avago, and then Intel. So, overall um our difference is a good combination of all the domain experts. And the other important part is um from day one, we are very focusing on manufacturing, testing, process. Right? How do we introduce automation? How do we introduce all this data injection? Every single step for the manufacturing, we collect data. We analyze data, right? So, we can we can easily right? So, we call the manufacturing execution system. So, we know we can [clears throat] how do we where is the gating item? How do we further improve the whole end-to-end process? And the last one over here I want to mention, from day one, we have this model. We co-design, co-define, work closely with our lead customer. But going forward, this might be a new model. So, NPO, CPO, we are more like the whole industry is moving away from the traditional standard I build pluggable module. I follow the standard. Let's talk about who has a bigger manufacturing capacity, who has a better cost, who has a better performance. NPO, CPO to some extent going forward will be more and the more costume based. I will design specifically to meet a one large customer requirement. And from day one we need to work closely with the end user. So that's one, the other major benefit at least Lumentum spring to the table. That's uh all my talk. Thank you.