video_OCP2026_Huawei_NPO標準化全景_20261015

Um what I want to do today is to uh discuss the what IPEC is and what we think we see the future of NPO and um and then conclude and hopefully I can make that a little bit of entertaining entertaining at the same time. Um so who's IPEC? IPEC is the International Photonic Electronic Committee. Right? It's open. Anyone that's interested can join. Um the fee to join is very low. Um we want it's um incorporated in Geneva and we want to make a unified standard and an ecosystem that includes everyone and every we want to make it equal for every uh participant to uh to be able to present standards. Um and we had a very very good consensus track record into IPEC so very little fight so it's very it's very good so far and uh we want to also we have a stream that looks at advanced research and we're looking at innovation. We have around 60 uh member at this point. So if we look at NPO, well here we have two Huawei system. One is uh um uh superclusters on maybe I can put the laser pointer. Here we have the the supercluster here. Right? And here we are using full optical interconnect and Huawei's been using full optical interconnect. We have Vixel versions and we have silicon photonics versions and we've been using that for I'd say around 10 years, right? Before the pandemic so it's around 10 years. And we also have uh intra copper and uh inter inter inter rack optics for our um accelerator solutions and enterprise systems, right? And as we move along and we're increasing the rate of of our of the optics. I know we'll move from 25 gig up to 200 gig to 400 gig, what happens and we've seen that in long haul as well, right? The optics moves closer and closer and closer into the system, right? And that's been a truth so far. And the question is what are we going to do for 448 gig, right? And there is very little left for for copper and that's why we need optics. And for people that think about OCI and all we're making making more parallel it history has been very brutal for those uh those system. We've been going up in rate forever, right? And I've started my career discussing with customers that 10 gig is going to take over 2.5 gigs, right? And now we are here and people say 50 gig is better and we're no we're going to go we're going to go more wider, but I think you know the question is how are we going to do 448 and that's where the future is. That's where we need to look at. Um like we need to do OCI and I mean we need to do the the the wide stuff and that's that's good. There is a window of opportunity for those companies, right? But what I'm saying is in the future we have to look at that. And so the optical interconnect the density the power and all you manage that that's very that's very attractive for the optics, right? And we have to look at that and we need to look at the long-term operation, reliability, stability and all of those challenges in the optics and that's what we need to to look at. Okay. Um so the optical evolution from pluggable to uh to CPO, I think we've seen that, right? The pluggable is right now it's mostly fully retired. It's at the edge of the system. It's at front panel is far from the from the channel from the from the A6. Um and well, we also have linear optics in our free time, but no, mostly what's sold today is fully re-timed. Um then you get closer in PO, that gives you some some benefits in manufacturability and supply chain availability, and then you have CPO. And then when I ask the AI like why is it exactly that um where did that whole workshop almost exclusively on NPO? This is what I got. A kung fu match. Right? So. Well, I hope I didn't I didn't push it too far. >> [laughter] >> So, anyway, so if you look at the story, right? So, supply chain round one, right? Supply chain maturity, and PO is winning. Uh serviceability, uh well, we can repair it we can repair the system if we have yield issues. Um manufacturing yield, also cost efficiency because we can pull uh different components from different suppliers. Um and we have standardization and ecosystem, and that makes the you know, if you have a bigger ecosystem, um that increase also competition and innovation, and in the end we get uh lower cost, right? But ultimately, ultimate energy efficiency, ultimate integration is CPO. And I will refer the blue guy to Dan here, he will train him, and he will win more rounds. Okay? This is what's going to happen. Well, he's just not trained enough, but eventually >> [laughter] >> but eventually eventually, you know, Dan will do his job, and the blue guy will win, right? Okay. So, practical balance between the performance and ecosystem readiness, I think this is where the sweet spot is right now is on NPO. Right? It's a question of time. You know, we Subu said timing, timing, right? Timing is important. I think the timing right now is for mostly NPO. Right? So, it's best balance for high performance and engineering feasibility. Uh why do we standardize? We standardize to have technical optimization, right? So, we have consensus. We try to optimize the system. We give it a bit there. We We gain it a bit here, right? And we get the right split of um for example, system integrity, right? So, we can have a split on how many uh block array to get from from this part of system or this part of the system, and we end up with the optimal solution. Um we also uh help market expansion for the ecosystem by making very clear what the solutions are, what part of the solutions are, and everybody can carve a very a niche and optimize there. And ultimately, cost reductions because um if you have an open ecosystem, uh you open the competition, and you find ways to reduce costs, right? And uh hopefully, you know, we can increase also margin at the same time. What do we standardize uh for an NPO solution? You need to look at standardizing the optical connector, the system requirement as a whole, the optical interface, right? So, that's a niche that IEEE has been uh doing for a long time, and we keep using it, right? We thought We heard in this conference that Ethernet is what's going to carry the data for long from uh from from time we we um uh standardized the uh ELSFP, the laser, right? Matt talked about that, the electrical connector, the the CMOS, the low-speed electrical control interface, all the controls, I squared C, I 3 C, right? Right now we're doing this standardization movement. Uh high-speed interface, uh what the serializer looks like at 448 by 56 by 54, right? And the mechanical dimension because we want those to be replaceable, right? So, that's the thing we're looking at at uh standardizing. So, IPEC had a few demos uh at the CIOE. We cannot show them here, but you know, like uh next year if you go to CIOE, IPEC usually has a big booth there. Right? We showed electronic connectors. We had connectors from TE. Um We had connectors from Amphenol, from MHE, and uh I think this is also from MHE. So, we had a few uh connector demonstration, all for NPO, all within the booth, and allowing people to actually uh plug and feel get the look and feel of those solutions. And uh we also had system uh demos uh with optical connector modules in full switch. And we had system from MaxLinear, from Huawei, from um from uh HiSilicon, and uh we also had some um uh Inspur server Mate One, which was participating in the demo. At the end, well, I think as we said, I think NPO it the timing is now. It It's got a good balance between maturity of the technologies and integration that it offers uh and availability of parts. And the IPEC has started a project um and 7.2T uh NPO uh to contribute to the industry, and everybody is welcome to join the project. So, thank you very much. >> [applause] >> So, we have 5 minutes and I think the speakers are booting up, but we could take a few questions for any of the speaker. If you had any questions.