# EPIC Online Technology Meeting on Roadmap 2021 for Co-packaged Optics

https://www.youtube.com/watch?v=mwhriE_Ghh0

[00:00] ladies and gentlemen it is 3 p.m 3 p.m
[00:03] here in the netherlands in the middle of europe
[00:05] and today we are worldwide with perhaps some thanks to anna
[00:08] the strongest participant listen agenda of all the online meetings we did since april last year
[00:15] and also with the biggest task to connect the drivers of the compact optics in the world
[00:19] with the supply chain of photonics that epic
[00:23] has become the next meeting on kopaki choptix
[00:28] is a dream come true we are asking the epic question to the leaders including intel
[00:35] facebook microsoft lenovo's broadcom adva with them and the rest of the epic supply chain
[00:43] will address the challenges of co-packaging
[00:48] the job of epic is to translate those challenges into business opportunities
[00:54] over the last two years big names like facebook and microsoft form the co-package optics collaboration in which they want
[01:02] to join forces
[01:04] with other hype scale data centers
[01:07] in their opinion co-packaging has the
[01:10] potential to address the challenge
[01:12] of data center traffic growth by
[01:14] reducing the power consumption
[01:16] of the switch optic electrical interface
[01:19] cost
[01:20] governs every road mapping decision we
[01:23] need to speed up the process
[01:25] while maximizing the yield on wednesday
[01:28] 3rd of march we are going to focus on
[01:30] three
[01:31] different areas of technologies one is
[01:33] equipment
[01:35] we want to address what alignment
[01:37] packaging and assembly equipment
[01:39] we need to address for co-package optics
[01:42] do we need die bonding semicolon-like
[01:45] equipment by the company evg
[01:47] or passive alignment base assembly from
[01:50] finetec
[01:50] or amicra or active alignment like
[01:53] ficontec
[01:54] xentech ims or tejema the second
[01:57] technology portfolio
[01:59] is silicon photonics and indium
[02:01] phosphide
[02:02] integrated circuits do we need to
[02:04] develop a standard silicon photonic transceiver
[02:07] or are we going to have just discrete components integrated in the co package
[02:13] in any case we need semiconductor process for integrating lasers detectors and modulators connected through glass or silicon bias
[02:23] we should bring to the table semiconductor companies like ase or atns
[02:30] what yield is expected at foundry level and what kind of individual testing is needed
[02:34] in epic we have infinera smart photonics ihp imac let's address this with them
[02:42] and with all the actors including those serving the testing industry like x4 quantified photonics or keysight
[02:51] plenty to discuss in this extraordinary meeting march the 3rd at 3 pm central european time
[02:55] do not forget to sign up to participate or simply watch live or later in youtube
[03:04] three at three that's easy to remember
[03:07] and it was easy to remember because about hundred and something people registered and the participant list is there for everyone to do business that is the purpose of this meeting
[03:15] i'm extremely happy today because we are i think only one or two members a way to reach this 700 number
[03:22] we are very close to reach 700 a welcome at s to the epic membership
[03:28] thank you very much for becoming one of our latest members especially being in the semi industry to to epic
[03:35] i also would like to remind everyone that i am part of a fantastic team of 15 people who dedicate their life to photonics and they work on this full time half of them with a phd position looking always to find collaboration with your members
[03:47] we organize events focus events to find technology business
[03:51] we provide access to your network so if you want to find a particular partner superior and customer contact us
[03:56] we'll scout it and make an introduction
[03:58] we help you raise capital
[03:59] we have the biggest website on finding a job in photonics jobs in photonics.com
[04:03] check it out and we also all of you know this because you're a big member because you're a member you have access
[04:08] to a long list of market reports from your from your development from the mates from fiverr reality they are all there for you to use and abuse.
[04:17] we also have here the list of the upcoming online technology meetings.
[04:21] today is a very special one called package optics.
[04:24] but also in two weeks we have one lidar applications we require mapping lidar for automotive and later for industry and even for agriculture so it's going to be a fantastic meeting.
[04:35] please sign up for it.
[04:35] i'm going to sign up for any of the upcoming online meetings that you find interesting.
[04:39] also in 2021 for the first time we are supporting the quantum industry and last week we had a fantastic meeting on quantum defense.
[04:46] next week quantum for medical industry.
[04:49] if you are interested in quantum technologies please do sign up for those.
[04:51] but today today is co package optics super excited.
[04:57] so first of all thank you so much for or to our media partners electro optics and nasa optics for making this possible for all the communication they have done proactively engaging all the industry on these meetings but
[05:08] also nobody ever can do anything great along anymore and this meeting is in cooperation with the kobo alliance and ethernet alliance thank you very much being all partners on this fantastic adventure
[05:19] most important this meeting wouldn't be possible without the support of our sponsors today thank you very much
[05:24] first of all knife force
[05:26] knife force all the way from sweden is the premier supplier of co2 laser splicing solution and fire processing equipment
[05:32] we go to switzerland exeter is just looking for a partner who provides micro optics for any kind of packaging assembly from telecom to any kind of sensing application
[05:41] accent residual panel for that if you're looking for a panel to do semiconductor lasers you have to go to finland
[05:46] semiconductor lesser manufacturing of modulite from the wafer growth all the way to packaging until reaching the the final part of the supply chain
[05:55] they provide even assistance to medical but also they provide fantastic semiconductor lasers for multi-wavelength applications in telecom
[06:01] but you are looking for a panel to provide birth systems and testing systems for the terabit per second communications
[06:08] multi-laying is a provider of testing
[06:10] equipment.
[06:11] If you're looking for a partner to provide equipment for diode bonding, direct diode bonding of any material, but especially the specialized semiconductor lasers, Bessie is your partner for equipment.
[06:22] Semicolon oriented packaging of semiconductor lasers.
[06:25] Packaging, no micro optics coming from Switzerland.
[06:29] Seuss Micro Optics is really at the forefront of what we call the European micro-optics revolution.
[06:35] They provide low-cost replication of any micro-optic wafer level optics supply for any market, but in telecom they don't fantastic optics for co-packaging optics.
[06:45] And finally ASM.
[06:47] Amigra, all the way from Germany, Amiga has really been the market leader on passive alignment for the packaging of transceivers for many years, and now they are the forefront of the equipment for the package optics.
[06:59] Thank you very much.
[07:01] Hello you for being our sponsors today.
[07:02] We love you.
[07:04] Thank you for being part of Epic, but I also love my colleague Anna Gonzalez, all the way from Barcelona.
[07:08] Anna, thank you so much for the.
[07:11] splendid work you have done today.
[07:13] what's in the menu?
[07:15] hello everyone and thank you jose for your kind introduction.
[07:18] so i am very happy to be here today to talk about package optics.
[07:22] with the key players in the industry as you can see in this agenda we will have intel facebook.com microsoft sequoia renovos to share their vision about the future and the current developments.
[07:33] optics and what is more important what is needed from the supply chain in photonics to get this job done.
[07:39] so we will talk about technology challenges and volumes requirements for development of new automation tools.
[07:47] what kind of micro optics will be required will compact this affects design and configuration of the chips.
[07:55] and to answer to answer all these questions we will have the entire supply chain in photonics attending the meeting so yes today we have a great supply chain with more than 100 companies.
[08:06] a and as you can see here today we have many companies that are users.
[08:11] or potential users of copaces so.
[08:14] all the companies that are vertical in.
[08:17] that are offering vertically integrated networking solutions.
[08:19] all the companies developing transceiver modules.
[08:22] even a company doing quantum computing.
[08:25] but also we have all the packaging ecosystem present.
[08:29] the company is offering packaging services automation equipment.
[08:32] we have companies doing testing equipment components.
[08:34] lasers and detectors also micro optics and different materials.
[08:40] and last but not least i would like to mention some of the european initiatives that are participating in this meeting.
[08:47] so i i would like to start with the passion project.
[08:50] which develops a new architecture for metro network.
[08:54] including transceivers in which the big cells are co-packaged on top of the silicon photonic web guides.
[08:59] and also the pylon lines which are working together to develop standard solutions for manufacturing of photonic devices.
[09:07] i want to start with pixab a pylon line which offers standard packaging and assembly.
[09:12] solutions for photonic integrated circuits.
[09:14] we focus on developing processes compatible with large ebony manufacturing.
[09:20] also we have fabulous the pylon line for large volume fabrication of free-form microptics.
[09:25] the new photon hub a photohub supports european companies that are not expert in photonics to develop new products based on photonic technologies and also helping them with financial business coaching and also technical training.
[09:41] and today i have an special announcement because a jpx pilot line which offers a large body manufacturing of indium phosphide of inner phosphite photonic integrated circuits is now hosting or they launched an open call for participation for companies that want to participate in the jbx demonstrator a program.
[10:03] so through this program jpx8 can help companies in supporting innovative businesses interesting in increasing the trl of their base.
[10:13] products and going from prototyping to commercial production.
[10:16] So yes if you want to know anything else about all these initiatives please feel free to contact us.
[10:22] And that's it back to cost thank you very much Anna.
[10:25] This slide only corresponds to the companies who register for the meeting today.
[10:26] So you're an epic member you're missing your logo there.
[10:28] Mostly you're right your logo should be there but you did not register for the meeting.
[10:31] This is not the result of any market study.
[10:33] This is just a visual representation of the participant list today.
[10:35] Thank you very much for all your support.
[10:37] These meetings of course also be live broadcasting YouTube.
[10:41] So hello YouTubers of the world.
[10:44] You want to get in touch with any of the participants today send me an email jose.com.com assac.com.
[10:49] I will be more than happy of making that introduction.
[10:54] And this is also valid for the people in the zoom room.
[10:58] You have the internal chat here use it to get in touch with each other over the next two hours.
[11:01] We want you to do business but of course after the meeting if you didn't get in touch with that person you wanted to do business with send me an email and make the introduction.
[11:14] anna i think it is time now to let the people from kobo and ethernet alliance to tell us how we are going to work together i'm very excited about this
[11:23] so jim theodoras thank you very very much for being with us this afternoon thank you it is 3 11 in the netherlands what time is it for you
[11:35] 3 11 in uh in sweden so we are worldwide we need to say all that to everyone so tell us how we're going to work together
[11:43] well i'm very uh honored to be representing the ethernet alliance today
[11:46] but i'm also honored to be co-sponsoring this event with epic i've been very vocal that the next stage in optical technology and communications is so difficult it's going to take the combined efforts globally of all the great minds and all the companies and all the associations tackling the problem from a to z to make the next stage happen
[12:09] and so um ethereum alliance is proud to be part of this effort and we're proud to work with others such as kobo and such as
[12:15] epic to work together to increase communication and and teamwork and partnership to see that this next level happens.
[12:24] so beneath the ethernet alliance um it began over a decade ago as just a way of companies to share marketing outreach and share booth space.
[12:33] uh the founding companies had plenty of money to do that themselves but we needed a neutral territory where we weren't competitors but we were partners working together on making the next thing happen.
[12:43] and that's how the ethernet alliance was formed but it's grown since then through a lot of strong leadership to be this very large organization that is all things ethernet.
[12:52] and if you look um at ethernet today it has grown into just all aspects of life in compass ethernet.
[13:00] and that's really what began as simply a way of connecting computers over a local area network or a lan.
[13:07] now is basically a fundamental way that machines communicate no matter what field or what industry so our goal is to promote
[13:16] the industry awareness accelerate industry adoption and also provide a neutral ground for multi-vendor interoperability.
[13:24] so we have a very large membership that continues to grow and um you know all the names at ethernet are basically members.
[13:31] we encourage all the uh additional kobo and epic members to take a look at membership as well.
[13:39] so one of the key things we do is interoperability testing.
[13:43] this has been key at all the levels uh best example i have is when we made the move to pm4.
[13:49] um you need to make sure that all these complex dsps can still talk to one another.
[13:54] and so we uh host these what we call plug fest where everyone can come in a neutral lab it's typically uh unh and in this neutral lab everyone plugs everything together.
[14:05] and everything's kept secret so if you do have problems then uh you know the world won't know until everything is fixed so.
[14:11] uh but we also do global outreach uh we have uh we participate and enable
[14:16] msas a lot of the msas use us for their marketing outreach so that's a win-win for both.
[14:22] uh we host tfs uh so we just hosted a very successful technology exploration exploration forum it was a week long.
[14:30] about what comes after foreigner gig very successful a lot of top-notch speakers i learned a lot and i thought maybe i couldn't and i learned a lot by participating.
[14:39] and then a lot of times well before things hit the ieee or the standards bodies they're incubated first in the ethernet alliance where consensus is built and then we go for that call for interest process at the various standards bodies.
[14:53] um and then obviously there's a lot of marketing outreach as well and trade shows.
[14:58] um we maintain what we call the ethan alliance roadmap it's about where ethernet is today but also where it's going.
[15:05] i know in the hustle and bustle of day-to-day activities sometimes you don't have the luxury of looking ahead.
[15:11] fortunately we're there for you to see where things are coming so you can plan well in advance of maybe what might be
[15:17] on your horizon your time horizon today.
[15:20] um the most recent value add that we added was a poe certification program.
[15:25] you know just a decade ago if someone told you you'll get all your power through ethernet you would probably have been called crazy and been committed but today ethernet is everywhere but also though they've gotten the devices got more complex so.
[15:37] if you remember the apple ecosystem like i am you find hey this perk doesn't work with that and that doesn't work with that but not everything works with everything and that's because it's gotten so complicated a handshake has to occur and this poe certifications way of knowing not only can you deliver the voltages and the currents and the power levels required but you can successfully hand shake with all the devices out there so that things don't catch fire and you won't be embarrassed when your product comes to market so that's a their most recent value add and i encourage everyone in the theory space to look at that.
[16:10] so please look at joining but if you're not quite ready to join you can still participate in all our efforts by just following us on facebook on twitter and
[16:20] On LinkedIn, we post everything, very active vlog, very active social media presence.
[16:25] So, um, please follow us there.
[16:28] It's going to take everyone working together to make the next step happen.
[16:32] And, um, and I, I'm excited about today's agenda.
[16:37] I can't wait to hear more about former speakers on the agenda today.
[16:39] Thank you.
[16:41] Thank you very much, Jim.
[16:44] We are super lucky to have you and to have the Ethernet Alliance sharing your vision for the industry for internet applications.
[16:48] And now let's hear about the Copacache group in Covo.
[16:52] A tiger from Senko, are you okay?
[16:58] We can see them.
[17:02] Thank you.
[17:02] Um, thank you, Anna, and thank you very much for Jim for introducing nice flight deck for the Ethernet Alliance.
[17:07] And again to Epic to have this very good opportunity to talk about, discuss about co-package optics that we really need.
[17:14] And let's make it happen.
[17:15] And, uh, today I'm representing Kobo, but I'm from
[17:20] uh cinco advanced components and the.
[17:23] let me switch over to the next slide so.
[17:26] as you know kobo has started our pass.
[17:28] back in 2015 to standardize uh onboard.
[17:32] optics module.
[17:33] and we did it a couple of years ago.
[17:36] and up our pass has been uh shifted to.
[17:39] kopacho optics and we.
[17:41] launched a new working group last year.
[17:43] so that we can.
[17:44] initiate the identify the challenges.
[17:48] and also talk about kopachi optics last.
[17:50] year and i i'm chairing that working.
[17:52] group.
[17:53] so let me describe this effort uh very.
[17:56] quickly.
[17:57] but the um this working group consists.
[17:59] with about like 50.
[18:01] membership companies in kobo and.
[18:04] surprisingly we have.
[18:05] more more and more people joining to.
[18:07] kobo now so um.
[18:08] if you are interested in uh please.
[18:11] contact melissa.
[18:12] which i put our contacts on the bottom.
[18:14] left.
[18:15] but the um we are having consist um.
[18:18] constant meetings and to talk about um.
[18:21] go back to optics and our scopes
[18:24] are to develop technical guidance and standards for copax optics implementations
[18:29] especially focusing on optical connectivity and remote laser sources
[18:33] and this is because we have quite quite a few membership in from the supply chains um that are strong in these areas
[18:41] and we'd like to focus on these areas and a couple of projects we initiated already
[18:46] one is white paper program to write some technical guidance identifying challenges and options to overcome
[18:55] and some some other discussions and open areas as well
[18:58] and recently we also decided to standardize the external laser source using embedded optics
[19:04] and this uh these projects are requiring more and more inputs from uh industries and industry leaders
[19:11] so hopefully we we can have a very good uh discussions within kobo and the again we hope this consortium is
[19:21] open to anyone so please don't hesitate
[19:24] to contact melissa on this
[19:27] and talking about optical connectivity
[19:29] and white paper
[19:31] we are targeting to release some kind of
[19:32] white paper probably around the summer
[19:34] time
[19:35] so keep your eyes on this
[19:38] but switching a little bit of gear to
[19:40] senko side um
[19:41] to describe some of the challenges of
[19:43] the optical connectivity side but the um
[19:45] as people here the speaker here will
[19:49] probably discuss today later today but
[19:51] the fibers
[19:52] per internet switch could be increased
[19:56] significantly and also there are some
[19:58] other areas to be discussed
[20:00] in terms of the optical connectivity
[20:01] described in the right hand side
[20:03] so i put some uh technical notes
[20:06] in in in a blog on my linkedin so if
[20:09] you're interested
[20:10] please visit and the you can probably
[20:13] discuss within this uh
[20:15] blog formats as well anyway that was
[20:18] fairly quick
[20:19] introduction of kobo and also some
[20:21] optical connectivity
[20:22] thank you very much thank you thank you
[20:24] very much tiger and thank you very much
[20:26] senko for all the support that you had
[20:28] done for epic and kobo and ethernet
[20:30] alliance yellow great
[20:31] the introduction is over it took 20
[20:33] minutes but today was worth it to have a
[20:35] longer introduction and now we're going
[20:37] to have intel facebook microsoft road
[20:39] con
[20:40] alva sequoia all of us but this meeting
[20:42] is not only about the speakers if you're
[20:43] looking for a webinar you are the wrong
[20:45] place this is for collaboration
[20:46] everyone in the room has to be active
[20:48] you have questions we ask
[20:50] and we took for cooperation and now we
[20:52] give the floor to one
[20:54] fantastic epic member intel thomas has
[20:57] been one star of the previous epic
[20:58] meetings and he follow up with all the
[21:00] introductions i made for him
[21:01] thomas legionberg thank you very much
[21:03] for taking the role of being the opening
[21:05] speaker of this fantastic event today
[21:07] the attention of everyone goes to you
[21:10] all right thank you jose and thank you
[21:11] to ethic for the opportunity to
[21:14] present here today um at six minutes you
[21:16] don't get an introduction i
[21:17] i assume we all understand that the
[21:19] challenges that that we're facing in
[21:21] terms of scalability of ethernet
[21:22] switches on i o power density signal
[21:25] integrity
[21:25] cost and so on and maybe margo ruby will
[21:28] also uh
[21:29] also give us more background so i'm
[21:31] going to jump right into
[21:32] to intel's activities and and our plans
[21:34] in in the area of copac
[21:37] photonics and i'd like to start with
[21:39] with a technology demonstration which we
[21:41] did
[21:42] almost a year ago now it was intended to
[21:43] be on the floor we are not seeing your
[21:45] your screen yet and we are
[21:47] so excited so yes
[21:52] yeah let me try again i thought i had
[21:53] done that but let me try again
[21:57] yes it's here you see it is not that
[22:00] clear yes
[22:01] okay good so i'd like to start with the
[22:03] with the technology demonstration we
[22:05] we we made last year and uh intended for
[22:07] 4c
[22:08] and it was built obviously as a learning
[22:10] and development platform
[22:12] for ourselves uh but also to demonstrate
[22:14] sort of more broadly that
[22:16] we've talked about copagus optics and
[22:18] ethernet switches for for a long time
[22:20] and
[22:21] we wanted to demonstrate that all the
[22:23] technology pieces are there we're ready
[22:24] to do it and
[22:25] and obviously there are tough
[22:28] engineering things to to deal with but
[22:30] the core technology is available and we
[22:32] can build it so we started building
[22:34] around uh
[22:35] one of the design points of 25t
[22:37] ultimately we had
[22:38] a silicon availability at the time due
[22:41] to timing at the 12.8 c switch
[22:43] so we had a fully functional switch
[22:45] system with a barefoot 1502 switch asic
[22:49] it was co-packaged with these 1.6
[22:52] terabit per second photonic engines um
[22:55] sort of you think of them as four times
[22:57] 400 gig dr4
[23:00] so eight sides of those to support the
[23:02] full 12.8 t bandwidth
[23:04] we populated in this configuration you
[23:06] see in the picture we had four
[23:08] of the photonic engines on the on the
[23:10] back here of the assembly
[23:11] uh copper heat pipes pulling out the the
[23:15] the heat out to to heat things out here
[23:17] and then uh
[23:18] these are socketed engines so the other
[23:20] four sides the other four sockets
[23:22] uh we've designed in a way that we could
[23:24] they could also be landing sites for
[23:26] flyover cables so we can bring cover
[23:28] ports to the front plate and and um
[23:30] that can support pluggables uh cover
[23:33] back planes whatever whatever you would
[23:34] do
[23:35] here so there's a flexibility and
[23:36] configurability at the
[23:38] assembly stage here what you don't see
[23:40] is the big cover plate that's that's the
[23:42] heatsink for the
[23:44] for the uh for the switch a ticket
[23:46] itself uh and then connectors on the
[23:48] front plate for for the optical ports
[23:50] and connectors for the uh for the cover
[23:51] ports here
[23:52] um so so this is a fully functional we
[23:55] hooked this up
[23:56] um loop back to a standard plugable df4
[24:00] in an off-the-shelf commercial ethernet
[24:02] switch and ran it for days
[24:04] uh error-free posttec we have
[24:07] a great df4 compliant operation you can
[24:09] see eye diagrams so
[24:11] a good operation at 112 gig pan-4
[24:14] transmitter here which is a new silicon
[24:17] photonic transmitter
[24:19] and and that was that was really like
[24:20] just to make that statement take
[24:21] technology is ready and and let's go do
[24:23] it
[24:24] uh products we're looking at scale
[24:27] deployment commercial intercept with the
[24:28] 51 t
[24:29] switch generation and with engines that
[24:31] are 3.2
[24:33] terabit per second then the comment you
[24:35] see here is that it's fully intended to
[24:36] be
[24:37] compliant and aligned with these
[24:39] emerging industry standards
[24:40] um in oif and coco and then elsewhere
[24:43] uh and and customer requirements
[24:47] um let's see
[24:50] on the next slide um at the at the heart
[24:52] of the the photonic engine with this
[24:54] transmitter chip so this is a 1.6
[24:57] terabyte per second
[24:58] it was a test chip eventually we'll have
[25:01] tx and rx on the same chip and we'll
[25:02] scale to 3.2
[25:04] terabit per second but but this was for
[25:06] the demonstration and it was first
[25:07] integrated ship here so we have
[25:10] at the center here you have laser raid 2
[25:12] by 16 laser ratio support 16 channels
[25:14] 100 and we have
[25:16] these are our on-ship fully integrated
[25:19] heterogeneously integrated lasers
[25:21] um we have two for redundancy even a
[25:24] very good laser once you put
[25:25] 500 of them in the package it will
[25:28] meaningfully contribute to
[25:29] to the failure rate of the switch so
[25:31] with redundancy we effectively take that
[25:33] out of the equation and and
[25:35] and don't see that in the in the total
[25:37] system fit
[25:39] we have an array of two by one switches
[25:41] to select which of the redundant laser
[25:43] the laser pair we're active and then
[25:45] we're we're
[25:46] modulators here 16 micro ring resonator
[25:48] modulators
[25:49] so we're switching from from the
[25:51] maxtender workhorse in our plugable
[25:53] modules to uh to these ring modulators
[25:56] for for size and for power efficiency of
[25:58] course
[25:59] and as you saw on the previous slide we
[26:00] have great performance here over a
[26:02] 100 gig pin 4 transmission so they're
[26:05] fully integrated with heaters control
[26:07] circuits and so on
[26:08] there's uh photo detectors on here so in
[26:11] this test chip the detectors are for
[26:13] monitoring only
[26:14] uh but they're capable of high speed
[26:15] operation so as we put the tx and rx on
[26:18] the same
[26:18] on the same dive we have we have that
[26:20] capability as well
[26:22] and then we route around the edge of the
[26:24] die here up to mode converters and
[26:25] v-grooves
[26:26] and the goal is to have high coupling
[26:28] efficiency in
[26:29] a high-volume manufacturing friendly and
[26:32] cost effective way with passive
[26:34] alignment
[26:34] and uh i didn't have room in the in the
[26:36] slide deck today
[26:38] if three slides qualifies a deck but but
[26:41] we have great results both in terms of
[26:42] the mode converter
[26:44] um performance and sensitivity to
[26:46] alignment and also the the v group
[26:47] capability with the
[26:48] passive alignment so we have a the great
[26:51] results
[26:52] to to see this to turn into a
[26:53] manufacturable process
[26:56] uh so where's this heading and i think
[26:58] that's the last slide here the thing
[26:59] that
[27:00] that we're building now on working on
[27:02] these 3.2 terabytes per second photonic
[27:04] engines
[27:04] and as i mentioned these are fully
[27:05] integrated tx rx die
[27:08] uh that we package with with the cdr
[27:10] driver t control circuits and so on so
[27:12] we're building fully testable and
[27:14] self-contained transceiver arrays if you
[27:16] will
[27:17] that evolved from from the photonics you
[27:19] saw on the previous slide
[27:20] so we're supporting both wbm uh
[27:22] interfaces and
[27:23] and therefore parallel interfaces with
[27:26] uh we have an unchip
[27:27] cwn laser array you can see that on the
[27:30] right hand side so the
[27:31] conceptual fork floorplan here so we
[27:33] laying them out with with the arrays of
[27:35] by wavelength and then having waveguides
[27:37] swizzle here so we could align wave
[27:39] wavelength to to uh to transceiver ports
[27:42] uh fully integrated mugs and then demox
[27:45] rekeeping the redundant lasers
[27:47] for for the purpose of mitigating a
[27:50] reliability risk
[27:51] um the receiver of course has has a
[27:54] provision for for diverse polarization
[27:56] diversity and receivers
[27:58] and then we made device improvement and
[28:00] i think one of the most notable is the
[28:01] reduced voltage so
[28:03] we're doing better on power dissertation
[28:05] of our modulator and there's more in the
[28:07] works there
[28:08] and then we're deploying some more
[28:09] advanced packaging that what you may
[28:10] have seen in our in some other
[28:12] publications so
[28:13] uh better signal integrity but also
[28:16] for of roughly four times the bandwidth
[28:18] density of the of the demo that we did
[28:20] and again the point is that this is all
[28:22] we're committed to an open
[28:24] standards-based ecosystem here um
[28:27] through oif through kobo and through the
[28:30] the the microsoft facebook driven uh
[28:32] optics collaboration um so i think that
[28:36] probably gets me to the six minutes and
[28:37] i'm happy to answer any questions uh
[28:39] here but that's a quick overview of of
[28:41] where we're heading with uh
[28:43] with the silicon photonics part of this
[28:45] project thank you very much thomas for
[28:47] this fantastic presentation i i we could
[28:49] have this presentation for
[28:50] 25 minutes for two hours and we'll still
[28:53] hear more and more
[28:54] room for cooperation we have a lot of
[28:56] questions in the chat but today everyone
[28:57] also gets the epic
[28:59] question we have the entire supply chain
[29:01] of photonic companies here in front of
[29:02] you
[29:03] what is highlight for us a couple of
[29:05] challenges what can they do for you what
[29:06] could you do for them
[29:09] so so i think we have i think we have a
[29:12] at the photonics level
[29:13] a compelling solution and and um
[29:16] i i think uh the uh the ability to put
[29:20] to put laser on there is gonna
[29:21] is gonna be helpful for us in terms of
[29:23] density and power and cost um
[29:26] i think there's there's a lot of
[29:27] challenges in terms of building up
[29:29] a healthy ecosystem certainly in
[29:31] packaging
[29:32] of building these photonic engines and i
[29:34] think that there's some
[29:36] connector fiber management uh and and
[29:40] um sort of uh maybe at the face of it
[29:43] and
[29:43] an offense to anyone who works on this
[29:45] but it may sound trivial but i think
[29:46] it's anything but trivial and similar to
[29:48] the laser once you pile
[29:50] many of something in there even even the
[29:53] small issues
[29:54] become become obstacles so i think uh
[29:56] compact resilient connectors and fiber
[29:58] assemblies and
[29:59] we don't want to eventually be limited
[30:01] by fiber density or connector density so
[30:03] so i think that there's work to be done
[30:05] there
[30:05] thank you so much thomas and be prepared
[30:07] because there's a lot of epic members
[30:09] experts in particular on that
[30:10] but we have some questions in the chat
[30:12] the first one is coming from lightweight
[30:14] stephen hardy thank you so much for
[30:16] being here what's on your mind
[30:19] so um i'm aware that intel has made a
[30:22] an investment in air labs which is
[30:24] working very hard on this co-packaged
[30:26] optics space did did you collaborate
[30:28] with with air on this particular
[30:30] demonstration
[30:31] and even if you didn't can you comment a
[30:33] little bit uh
[30:36] we did not work with it with ios on on
[30:39] on this demo and and uh
[30:41] i i think when you look at some of the
[30:42] technical detail of
[30:44] what ils is working on um
[30:47] i think it's not a that's it except fit
[30:50] or this way the focus is right now i'm
[30:52] not talking about sort of fundamental
[30:53] capability but
[30:54] it seems that focus is heading in in
[30:55] another direction um
[30:58] more on the on the the the co-packaging
[31:00] with the
[31:01] compute nodes or or similar the next
[31:04] question comes all the way from
[31:06] switzerland
[31:07] seuss micro biker optics girlfriend null
[31:09] what's on your mind
[31:11] so uh from the screenshot you showed
[31:13] from thank you
[31:14] jose from the screenshot you showed
[31:18] almost half of your real estate was mode
[31:20] size converters and big grooves
[31:22] is that uh compatible with your process
[31:25] and does this make sense in terms of
[31:27] real estate or
[31:29] or do you want to have a more compact
[31:30] solution in the future
[31:33] um i i think it's it makes it makes
[31:35] sense
[31:36] um and the
[31:39] the alternative it's a question that has
[31:42] come up
[31:43] but the alternative turns out that it's
[31:44] not that much more economical in terms
[31:46] of space
[31:47] and and density when you start putting
[31:49] in way more complex
[31:50] uh it's a matter if you want the
[31:52] complexity on the on the wafer and the
[31:53] die or do you want it in the assembly
[31:55] outside
[31:56] um so so this makes sense to us more
[31:59] compact
[32:00] is always better so so if there are
[32:02] ideas there
[32:03] we would love to hear about them um but
[32:05] it seems to be a trade-off more of where
[32:08] do you put your
[32:09] your space and your complexity rather
[32:10] than a a sort of a
[32:13] fundamentally changing the the total
[32:16] okay thank you the next question is from
[32:18] one of the most innovative companies who
[32:19] have an epic they are adding bto to
[32:21] silicon photonics lumiface gustavo
[32:23] villares what's on your mind
[32:25] thanks jose so my question is it's
[32:27] actually very simple if you could
[32:28] comment a little bit on the power
[32:29] consumption of the the two optical
[32:31] engines that you mentioned as
[32:32] it's i mean we know cpo it's a lot about
[32:35] uh power
[32:36] reducing the power consumption and can
[32:38] you give us a little bit of uh
[32:39] the idea of what you could achieve and
[32:41] what's your roadmap on that
[32:44] yeah so i think we we're starting to see
[32:46] some
[32:47] some uh i say targets or or
[32:50] specifications going into the standards
[32:52] and and i think we're we're in good
[32:53] shape to to to align with the
[32:56] with with those targets um i think
[32:59] eventually we'll
[33:00] we'll we'll do better but but i think
[33:02] i'd like to to generate more data and
[33:05] feel more comfortable before i start
[33:06] over committing on on or committing more
[33:08] than
[33:09] what has been put into the standards
[33:13] all the way from silicon valley chris
[33:15] fisner all the way from cti
[33:18] chris thank you very much for being here
[33:19] what's on your mind
[33:24] chris fisner yeah gotta get the mood
[33:27] on mu button first uh tomas nice
[33:30] presentation just follow on on the power
[33:32] consumption what are you most worried
[33:33] about is it on the laser side is it
[33:35] driver modulator there's a lot of talk
[33:38] about
[33:38] the direct drive simplifying the kind of
[33:41] the 30s cdr
[33:43] side of things yeah i mean
[33:48] it's kind of tried answered it's all of
[33:49] the above right that that to
[33:51] to to really to really um
[33:55] to really deliver on on on the promise
[33:57] of lower power you have to do well
[33:58] on all of them so i highlighted one one
[34:01] area we've made improvements on
[34:03] on drive voltage and on the modulator
[34:05] and that that contributes
[34:07] i think once we sold in the lower power
[34:08] 30s we're obviously going to see
[34:11] some significant improvement there sort
[34:12] of the whole the whole rationale behind
[34:15] power reduction uh power reduction here
[34:17] um
[34:18] i think the the the the laser i think is
[34:22] fundamental to to to our our value
[34:24] proposition i would say because
[34:26] we don't have the the the loss of of of
[34:29] uh optical power we get more of our
[34:31] photons into to actually put to good use
[34:33] in
[34:34] in the link right we don't have the
[34:35] coupling into to a fiber and then back
[34:37] into the to the optical interface
[34:40] um but i think it's it's really all of
[34:42] the above to get to
[34:43] to to really compelling numbers in terms
[34:45] of direct drive i think that's that's a
[34:47] that's an interesting concept
[34:49] i think it's it's much harder to
[34:52] standardize and get to that
[34:54] healthy uh healthy multi-source
[34:56] environment that
[34:57] that is a prerequisite for this to be
[34:59] broadly adopted so
[35:01] i i think that's that's a longer term
[35:03] discussion maybe
[35:04] thomas the next question is i want to
[35:07] address it differently
[35:08] we always discuss about the same topic
[35:10] active versus passive alignment and you
[35:12] are aware of how much i enjoy this topic
[35:15] the next question comes from a company
[35:16] this at the forefront of active
[35:18] alignment it comes from
[35:19] ficon tech greg fling thank you very
[35:22] much being with us
[35:23] all the way from next to bremen akin
[35:26] what's on your mind um thank you jose uh
[35:29] thank you thomas for the interesting
[35:31] talk
[35:31] i guess um i had a general question in
[35:33] terms of
[35:34] what uh are the greatest manufacturing
[35:37] challenges you face at the moment
[35:38] with with these modules moving forward
[35:41] to a degree you've answered that already
[35:42] so i can reduce my question maybe to
[35:44] answer
[35:45] one two and in terms of fiber passive
[35:48] alignment how many are you doing in
[35:49] parallel at the moment and how many
[35:51] would you like to do
[35:53] well so so i mean this this is
[35:56] this is pretty early stage from from a
[35:58] sort of product point of your
[35:59] manufacturing point of view right
[36:01] so so the designs and the the um
[36:04] design points we're looking at right now
[36:06] 16 arrays of 16 fibers
[36:08] um and i think that that's it's
[36:12] i think that's that seems to be like a
[36:14] good spot to be that it's that it's
[36:15] enough to be meaningful but it's not so
[36:17] many that it's that it's
[36:18] insurmountably hard at least to start
[36:20] with right so so that's where we are
[36:22] right now and uh
[36:23] but you're doing those individually no
[36:26] that goes in
[36:27] that goes in as an array okay okay so
[36:30] that's that's that's really
[36:31] the motivation because i think i think
[36:33] it comes up it's not religion that it
[36:35] has to be passive alignment and i think
[36:36] i think folks who are experts in active
[36:38] alignment have have good
[36:40] arguments that it's gotten very fast and
[36:42] very good um
[36:43] but i think that the so maybe the real
[36:46] differentiator is that we can put entire
[36:47] race down
[36:48] more effectively uh than than um
[36:51] than active but i think it's not
[36:52] religion so if there is a if there is a
[36:54] a better solution then um that's that's
[36:58] something we always want to hear about
[36:59] we have so many questions
[37:03] just is the process you have is that
[37:05] maybe
[37:06] putting in minimum footprint down on
[37:08] your real estate
[37:11] um i don't think i can say that with the
[37:14] great certainty that that
[37:15] that's the optimum point there but we're
[37:17] gonna revisit this topic
[37:19] later because i also have asm amigra in
[37:22] the room i also have ike's tech i also
[37:23] have fine tech i have all of them here
[37:25] for you thomas
[37:26] but before we go to next speaker two
[37:27] more questions quick ones the first one
[37:29] coming from france and my technologies
[37:31] or indium phosphide foundry here
[37:33] john lewis what's on your mind yeah
[37:36] hello everyone on my laser company so
[37:40] it's it's just are you curious to know
[37:44] but the temperature of operation you
[37:46] assume for the laser
[37:48] when you when doing the uh reliability
[37:51] estimate
[37:51] especially when getting to the point you
[37:53] uh you have a redundancy of the laser
[37:57] yeah so so the the uh
[38:00] we we have pluggable modules now that we
[38:03] rate up to 85 degrees
[38:04] uh case temperature so so that's that's
[38:07] one reference point most of what we what
[38:09] we
[38:10] sell and ship is obviously 70 degrees c
[38:12] on the case
[38:13] um but even even at
[38:17] uh when we look at actual realized field
[38:20] data
[38:21] we have something that looks like a
[38:22] laser fit of something like two fits so
[38:24] so and that's limited by device hours
[38:26] not by actually failures at this point
[38:29] so so the reality is that it's a very
[38:30] reliable laser
[38:32] even at high temperature the the issue
[38:35] is that once you put 500 of them even a
[38:36] very reliable laser it just adds up
[38:39] um and and the the
[38:42] if we can prove over time that we don't
[38:44] need the redundancy then maybe we'll
[38:46] apply the redundancy to yield instead
[38:48] and so so that there's no reason given
[38:51] how we make laces there's no reason to
[38:53] not put the redundancy and let's put it
[38:54] that way
[38:55] all right but it's not based on the
[38:57] calculation right
[38:59] and the final question about this
[39:00] fantastic q a is coming all the way from
[39:03] santa barbara a big friend of epic
[39:05] young professor bowers thank you very
[39:07] much for being with us today
[39:08] what's on your mind thanks that was a
[39:11] great presentation
[39:13] you said the two fit was not enough
[39:15] going forward so what what is the
[39:16] requirement for the laser do you think
[39:18] and similar to the modulated
[39:19] overall pick so what was the question
[39:22] jonah did you put up for a second
[39:24] you said i thought the laser was had a
[39:26] reliability of two fit
[39:28] yeah and but that was not good enough to
[39:31] run by itself
[39:33] so what is the requirement do you have
[39:34] to get below one or what do you
[39:37] well first first of all that the
[39:39] estimate is below one but your
[39:40] statistically limited
[39:42] by device hours what we can say with
[39:43] with a reasonable confidence level
[39:45] um and i think the d so at 500 lasers
[39:49] we're moving it up to more than a
[39:50] thousand fit
[39:52] for for the assembly just driven by by
[39:54] the laser and i think i would i would
[39:56] almost suggest ask one of the next
[39:57] speakers what is acceptable
[39:59] uh in in their networks from from from a
[40:01] fit rate
[40:02] um we we think that we're sort of right
[40:06] on the
[40:06] on the edge of what's what's good enough
[40:08] without redundancy
[40:10] but i think you're so familiar with our
[40:12] technology that you know that it's not
[40:14] it doesn't cost a lot more to put the
[40:16] extra laser in and it kind of just takes
[40:18] the whole question off the table here in
[40:19] first generation when we're all learning
[40:21] um that you go from from a thousand
[40:24] laser driven fit to two to four times
[40:27] four of the magnitude better and we
[40:28] don't even have to talk about it in that
[40:30] sense
[40:31] uh so long-winded answer to an easy
[40:34] question but
[40:35] how about how about the modulated photo
[40:37] detector what what
[40:39] fit level are they at
[40:42] um we we don't really see any
[40:46] uh so the modulator is new and there's a
[40:48] lot of work that needs to be to be
[40:50] to be done there um the but we don't see
[40:53] any on the detector we haven't seen any
[40:55] any
[40:56] any significant numbers in reliability
[40:58] testing or in in the field
[41:00] thomas you have gotten the record of
[41:02] more questions being asked after an
[41:05] opening presentation thank you so much
[41:07] and we continue now with the program but
[41:09] please stay with us because i'm gonna
[41:10] come back to you
[41:11] continue with the program we go from
[41:13] intel to microsoft
[41:15] mark feiler thank you very much for
[41:17] being with us today thank you for taking
[41:18] the role of following after
[41:20] after the amazing presentation from
[41:21] thomas let's see how we can
[41:23] help you help microsoft being even
[41:25] greater than what it is the floor and
[41:26] the attention of everyone
[41:28] goes to mark thanks
[41:31] and uh happy to happy to be um
[41:34] speaking today is everyone hearing me
[41:36] okay thumbs up
[41:38] okay great um yeah so uh
[41:42] i have a few slides i want to walk
[41:44] through um
[41:45] where uh i'm basically just going to
[41:47] present some
[41:49] high-level motivators and some of the
[41:53] requirements and then uh
[41:57] cap off by showing the application
[41:59] spaces that we're
[42:00] looking at applying co-package optics
[42:03] and
[42:03] i have one more slide actually on on
[42:05] standardization too so
[42:07] um i'm a principal engineer in azure
[42:10] hardware architecture and
[42:12] my focus over the last year has been uh
[42:15] increasingly on um co-packaging
[42:18] and um so uh first i i want to
[42:22] um talk about what what some of the
[42:24] motivators are
[42:25] and the initial um the initial thing
[42:29] that
[42:30] started pointing us to co-packaging uh a
[42:32] few years back
[42:34] was our observations about data center
[42:36] power and particularly
[42:38] the proportion of the power that was uh
[42:41] being
[42:41] dedicated to the network which doesn't
[42:45] generate revenue uh when compared to the
[42:47] overall
[42:48] data center power envelope so any power
[42:51] going to the network is taking away
[42:53] powers that we could be using
[42:55] for servers compute storage
[42:58] and other applications and those are our
[43:01] revenue generating devices
[43:03] and so we knew um that
[43:06] as time went on and we looked at scaling
[43:09] link bandwidths up
[43:10] um uh that the trajectory we were
[43:13] observing which you can see
[43:14] in the uh the chart on the right there
[43:17] um
[43:17] looked to be uh unsustainable so um
[43:21] power was kind of the first thing that
[43:23] that had us scratching our head and
[43:24] thinking okay this is
[43:26] you know we're gonna have to look at
[43:27] doing things differently um going
[43:29] forward
[43:30] we really see 400 gig um
[43:34] as being the last kind of generation
[43:36] where we can do things in the the
[43:38] way we've traditionally been doing them
[43:40] um and i'd say further
[43:41] compounding that you know looking at the
[43:43] transition to 100 gig and
[43:45] then in the future 200 gig electrical
[43:47] lanes really starts to push the limits
[43:49] of of
[43:50] copper interfaces where you know in
[43:52] particular in our current architecture
[43:54] we have a top of rack switch that's
[43:56] connected to our servers with uh
[43:58] direct attach copper cables and um
[44:01] as we're we're pushing those lane speeds
[44:03] up
[44:04] um that that starts to become
[44:06] fundamentally more challenging both in
[44:08] terms of uh
[44:09] reach but also just the the physical
[44:11] space that it requires
[44:13] um to connect all of those cables
[44:15] together in iraq
[44:17] um and then lastly um with with the
[44:20] increase of uh
[44:22] other applications in in the data center
[44:25] like ai and machine learning
[44:27] and then some some kind of more
[44:30] forward-looking topologies where we're
[44:32] looking at doing resource aggregation
[44:34] disaggregation uh such as storage or
[44:37] memory resources
[44:40] these applications typically have even a
[44:43] higher
[44:45] network dependency meaning they're more
[44:47] densely connected
[44:48] the nodes are more more highly meshed
[44:52] and the end points for these nodes
[44:56] are kind of order a magnitude higher
[44:58] than what we're used to for a typical
[45:00] server
[45:01] order magnitude or two even sometimes
[45:04] you know
[45:04] looking into the tens of uh terabits per
[45:08] per second per node um if we look far
[45:10] enough into
[45:11] into the future um so uh there again
[45:15] we're going to be power and space
[45:16] constrained
[45:18] and and the typical approach with with
[45:21] copper or plugables
[45:23] um to to scale these beyond just a
[45:25] chassis level or something like that
[45:27] is uh going to be limiting so we we
[45:30] really view
[45:31] optical and photonic innovations as
[45:34] being the key to
[45:35] enabling our our future data centers and
[45:38] those applications associated with it
[45:41] um on the left-hand side of this this
[45:45] chart um
[45:46] i want to talk a little bit more about
[45:48] the power
[45:49] but then i'll get to the right hand side
[45:51] which which shows that really the
[45:53] benefits of cpo
[45:54] extend far far beyond just just the
[45:57] power
[45:58] um so the left-hand side is just showing
[46:01] a plot
[46:01] of power efficiency basically of
[46:06] different
[46:07] form factors and technologies um where
[46:10] on the far left we're looking at
[46:12] pluggable transceivers and
[46:13] and the measure is in pico joules per
[46:15] bit um just think about that in terms of
[46:17] like
[46:18] uh the watts dissipated divided by the
[46:20] bit rate
[46:21] uh of the module and you'll get pico
[46:23] joules per bit so
[46:24] um uh traditional plugable optics
[46:28] in the 400 gig time frames
[46:31] um they're range anywhere uh from
[46:35] around um 18 to
[46:38] 32 picojoules per bit depending on the
[46:41] particular optic
[46:42] um and and the uh the node process for
[46:46] the the cmos there
[46:47] but um over time that that is expected
[46:52] to come down
[46:53] um but but uh kind of in these time
[46:55] frames we're trying to do an apple
[46:56] dabbles comparison as best we can
[46:59] um looking to onboard optics which uh
[47:01] tiger mentioned at the very beginning
[47:03] there
[47:04] um you can see a reduction in power uh
[47:06] because you're
[47:07] effectively moving the uh device closer
[47:10] to the asic it's
[47:11] mid board not all the way to the front
[47:13] face plate so the 30s required to drive
[47:15] those
[47:16] uh devices can operate at lower power
[47:19] levels
[47:19] but really the the big benefits come
[47:21] when you go to a full
[47:23] co-packaging paradigm where the photonic
[47:25] device and the asic are in the same
[47:28] uh the same physical substrate and the
[47:30] trace links are greatly reduced and you
[47:32] can look at things
[47:33] like um xsr certies which is a much
[47:36] lower power than the typical lr that's
[47:38] used with plugables
[47:40] and then projecting forward in the
[47:41] future you can see further optimizations
[47:44] um that that could get us close to the
[47:46] closer to the kind of five to ten pico
[47:48] joule per bit range
[47:49] um as we project out now the these
[47:52] power uh values here on the left are
[47:56] uh including the module and triplet side
[47:59] electrical interface
[48:00] uh dsp and and pick components and laser
[48:03] source but they
[48:04] they exclude the switch side um so they
[48:07] you know a little bit kind of undersell
[48:09] the the value proposition of
[48:11] co-packaging
[48:12] when you look at it at a system level
[48:14] you're going to see a larger overall
[48:16] system reduction in power i think
[48:18] um than what's just reflected by
[48:20] focusing on the module
[48:22] um if you look on the right side here my
[48:25] my my table is trying to uh highlight
[48:28] the fact that you know it's not really
[48:30] just just power but
[48:32] um with cpo going to cpo you know we
[48:35] we see uh multiple benefits um
[48:38] from optical reach you know you're going
[48:40] from dax and aocs that we typically
[48:42] build our
[48:43] our tour and tier one uh connections
[48:45] with
[48:46] um you're going you can cover anywhere
[48:49] from
[48:49] you know tens to hundreds of meters now
[48:52] with the same
[48:53] photonic chiplet at the cost level you
[48:56] know we expect the cost to come down
[48:58] pretty dramatically just because the
[48:59] the further degree of uh photonic
[49:02] integration but also
[49:03] that the volumes will will go up because
[49:05] the assumption here is that we'll be
[49:07] doing optics all the way to the server
[49:08] and
[49:09] obviously the server's uh server counts
[49:11] are much higher than what we have in the
[49:13] switches today
[49:14] um similarly bandwidth density we see is
[49:18] another
[49:18] um uh benefit of cpo particularly
[49:22] thinking about those high bandwidth
[49:23] endpoints for
[49:24] ai machine learning resource
[49:26] disaggregation
[49:28] um looking uh further at the
[49:31] operational aspects you know we have
[49:33] reliability
[49:34] um where again at a system level we
[49:37] expect
[49:38] uh the overall system of the cpo based
[49:41] switch or uh endpoint to be
[49:44] more reliable than what we have with
[49:46] plugables today
[49:48] essentially most of our failures in the
[49:50] data centers
[49:51] link failures are attributed to poor
[49:53] struck
[49:54] poor handling of the infrastructure and
[49:56] in the um
[49:58] the plugable modules themselves
[50:00] mechanical damage and things like that
[50:02] damaging the
[50:03] the interfaces and um really getting
[50:06] those out of the technician's hand
[50:08] should make for a more reliable and
[50:10] available infrastructure
[50:12] energy efficiency we've already touched
[50:13] on and then lastly you know latency
[50:16] we can see um
[50:20] tied to cpo in the sense that
[50:23] by by co-packaging and moving the
[50:26] chiplet closer to the asic and
[50:29] particularly looking at some of these
[50:31] [Music]
[50:32] future applications where we might be
[50:34] running a cxl
[50:35] or pcie type of interface at a lower
[50:38] baud rate and
[50:39] within our z um we can look at um
[50:43] uh you know bypassing factor or or
[50:46] having a very lightweight effect where
[50:47] the latency
[50:49] uh is critical particularly for some of
[50:51] these
[50:52] memory disaggregation type applications
[50:55] where where it's
[50:56] we have latency tolerances in the you
[50:58] know kind of tens of nanoseconds that
[51:00] we're sensitive to so
[51:02] um again big takeaway we see cpos being
[51:05] a distinct step forward
[51:06] in how we we build our data centers and
[51:09] the efficiency of those networks
[51:11] um and uh just kind of a big picture
[51:14] here
[51:15] and you know we've shown this slide
[51:17] around several times and i think it's
[51:18] useful but it definitely
[51:20] needs some work too but at a high level
[51:22] you can look at kind of three verticals
[51:24] of uh
[51:25] uh application spaces that that i've
[51:27] kind of referenced already but
[51:29] um the first vertical being the
[51:31] traditional networking so it's how we
[51:32] build our traditional front end networks
[51:34] today
[51:35] but going instead from a plugable
[51:36] paradigm to a cpo paradigm
[51:38] with the the end points are the rows
[51:40] here so
[51:41] uh are they the end point is the top row
[51:44] and the switches
[51:45] or whatever is doing you know some type
[51:47] of um
[51:48] connectivity uh in the middle uh between
[51:51] the end points is is in that bottom row
[51:54] um so moving over to
[51:55] some of the ai training or hpc
[51:58] applications
[51:59] um really it looks quite similar to what
[52:01] you have in the traditional networking
[52:03] the primary distinction there just being
[52:05] that the
[52:06] the ai and hpc application endpoints
[52:11] tend to have much larger bandwidth
[52:13] requirements
[52:14] and then lastly on the right-hand side
[52:16] the future systems that we call
[52:18] with disaggregation and that's looking
[52:21] more into things like uh just
[52:23] general resource disaggregation um
[52:26] storage
[52:27] uh things like memory desegregation um
[52:30] that that drives some different
[52:32] requirements and and um
[52:34] you know what we're really looking for
[52:36] is is to see how we can leverage
[52:38] the same cpo development across these
[52:40] applications
[52:41] uh there may be differences like i
[52:43] alluded to you know some being based on
[52:45] you know 100 gig camp 4
[52:47] uh for the traditional networking where
[52:49] the desag systems might be
[52:51] more cxl pcie based 32 gig and rz
[52:55] but the types of commonalities you can
[52:57] think of are you know
[52:59] 32 lane chiplet um would be
[53:02] uh useful in both applications um and
[53:05] you know they may
[53:06] just require some tweaks to the the
[53:09] front end
[53:10] electronics and things like that to keep
[53:12] them more power efficient but
[53:14] the big takeaway being is that some of
[53:16] the some of the developments
[53:18] and efforts being put into developing
[53:19] the the pick
[53:21] um for a networking cpo could also be
[53:23] leveraged
[53:24] in a disaggregation scenario so
[53:28] um the last thing i wanted to touch on
[53:30] is um
[53:31] uh standardization of cpo and and and
[53:34] you know i don't wanna
[53:35] undersell the complexity of what uh
[53:39] you know moving to a cpo based ecosystem
[53:42] entails because it's really uh you know
[53:45] kind of new territory and
[53:46] there are a lot of challenges to be
[53:48] solved um and
[53:49] you know it's our view um that that this
[53:52] needs to be
[53:54] done you know within a standardization
[53:55] framework um
[53:57] and that you know standardization is
[53:59] going to enable us as end users
[54:01] on large scale volume at lower cost
[54:04] it's going to give us vendor diversity
[54:06] uh within any
[54:07] single use case at least and and allow
[54:09] us to mix and match
[54:11] asics with uh chiplets um and
[54:14] uh hopefully it'll provide the reuse of
[54:16] cpo chiplets across
[54:17] yeah like i was just saying different
[54:18] different host asics so
[54:20] um you know the two primary efforts
[54:23] going on right now
[54:24] in standardization is oif uh they
[54:27] launched a co-packaging
[54:28] framework project uh back in the fall
[54:30] and then just at the
[54:31] meeting last week the q1 meeting um
[54:34] launched a
[54:35] 3.2 t cpo module project that's looking
[54:38] at really standardizing
[54:40] um the specifics in the module and ruby
[54:42] will get into this in her talk
[54:45] and then secondly um kobo
[54:48] as tiger mentioned earlier is really
[54:50] focused uh more on connectivity
[54:52] challenges
[54:52] and including external laser sources uh
[54:55] the two efforts are very complementary
[54:57] and there's a liaison relationship
[54:59] between the the standards organizations
[55:01] um to ensure that the um you know
[55:04] there's there's
[55:05] minimum uh redux uh uh redundant work
[55:08] going on so
[55:10] um yeah the ieee you know this the
[55:12] interfaces are based on ieee standards
[55:14] fr4 and dr4
[55:16] um open compute may play a role later on
[55:19] particularly as we look to going to
[55:20] wider interfaces or things like that
[55:23] where some some work there is being done
[55:25] in the
[55:26] packaging area um i have some url links
[55:29] down at the bottom for
[55:30] for uh announcements about some of these
[55:33] projects
[55:34] and um yeah that's that's what i wanted
[55:37] to say today and um you know i'll take
[55:38] questions and
[55:39] hopefully i've set up really well to to
[55:42] go into more about the specifics
[55:44] and um some of the uh some of the
[55:46] challenges that we see
[55:48] in in going to co-optics co-package
[55:51] optics sorry
[55:52] thank you very much mark super
[55:54] interesting presentation also very clear
[55:56] the link between the energy consumption
[55:58] and the different technologies for
[56:00] transceivers and how
[56:01] package reduce this energy for data
[56:04] centers
[56:05] so okay now the the the epic question
[56:08] uh what do you think the supply chain
[56:10] photonics
[56:11] that is present here can can do for one
[56:14] to to make a reality in your vision in
[56:17] compact
[56:19] yeah uh you know the biggest thing i
[56:21] would encourage them to do is
[56:22] uh participate you know and not only
[56:25] participate but contribute to some of
[56:27] these standardization efforts i mean
[56:29] we really need uh the voices of the
[56:32] experts there in the room
[56:33] and and pointing out you know what's
[56:36] feasible what's not giving us you know
[56:38] data driven um basis for the decisions
[56:42] that we're making
[56:43] i think we're off to a good start in
[56:45] those efforts in the oif and kobo
[56:49] but we really are looking for
[56:50] contributions there and i think that's
[56:52] what the
[56:52] supplier community can contribute most
[56:54] at this point is
[56:55] you know give give us a reality check
[56:57] help us to to be sure we're on track
[57:00] thank you very much mark we have a
[57:02] question in the chat
[57:04] a jour from max citrix would you like to
[57:06] make the question by yourself
[57:10] maybe also a good opportunity to explain
[57:12] a little bit what do you say
[57:13] accidentally say doing
[57:14] in what what could you do in compact
[57:17] optics
[57:19] well of course you want to sell optics
[57:21] no yeah
[57:22] you see it in my background so we are a
[57:24] manufacturer of micro-optics silicon
[57:26] and fuel silica and behind
[57:29] or beyond that of course also
[57:35] mems structures if you like that might
[57:39] help to align and assemble
[57:41] micro optics also for co-packaged optics
[57:44] then my question i think that's
[57:47] interesting for
[57:48] many people here is so thanks for the
[57:50] presentation mike anyway
[57:52] you have shown these sliders cpo
[57:54] benefits in terms of
[57:56] link energy versus technology level and
[57:59] of course
[58:00] um you don't have you haven't shown a
[58:02] timeline for the different technology
[58:04] levels
[58:05] right how do you see that what's your
[58:08] feeling for that yeah and and ruby
[58:11] has a a nice timeline um chart in her
[58:15] presentation coming up so i won't steal
[58:17] the thunder too much but um
[58:19] uh yeah that's a great question and and
[58:21] um you know we're looking at uh
[58:24] co-package um focus more on the
[58:27] traditional networking use case first
[58:29] and
[58:29] timelines for that is we're we're really
[58:31] hoping to have pilots
[58:33] um going into our data centers on a
[58:36] limited basis kind of in the mid 2023
[58:39] time frame and looking at being able to
[58:41] have something deployable by
[58:43] mid-2024 so you know it it's far enough
[58:46] out that there's still
[58:48] time to get it right you know but um it
[58:51] it's close enough that we don't want to
[58:53] take too much time
[58:55] with you know analysis paralysis and and
[58:58] really
[58:58] be trying to push it forward
[59:02] okay so thank you very much jorg thank
[59:04] you very much mark for this
[59:05] presentation so let's move to the next
[59:07] presentation let's say what the
[59:09] ruby from facebook has to say about
[59:11] copaca chapter oh very good
[59:13] ruby and we can see very well your
[59:16] presentation so the floor is
[59:18] yours perfect thanks anna okay hi
[59:21] everyone
[59:22] um great to hear all this great talk
[59:25] and uh see all this level of uh
[59:28] interesting for face
[59:29] for cpo um and i also see a lot of
[59:32] familiar faces for
[59:35] for front here and for anyone who
[59:37] doesn't know me i'm actually a technical
[59:39] sourcing manager focusing on next-gen
[59:41] optics and asic for facebook data center
[59:44] and in the following minutes let me walk
[59:46] you through facebook visions
[59:47] in co-package optics and really also
[59:50] code out some of the challenges
[59:52] and actions okay
[59:57] all right so the need to really involve
[01:00:00] our network infrastructure in terms of
[01:00:02] bank waste and performance comes from
[01:00:03] the traffic increase that we have
[01:00:05] experienced
[01:00:07] and it has forced us to consider a
[01:00:09] really more
[01:00:10] efficient network equipment in terms of
[01:00:12] power and performance
[01:00:15] to use and to reuse our existing network
[01:00:18] topology
[01:00:19] and code package as shown here offers a
[01:00:22] way to partially offset the power
[01:00:24] increase associated with the switch
[01:00:25] bandwidth scaling
[01:00:27] since by moving the optics package
[01:00:30] closer to the switch
[01:00:31] a lower power surface can be used to
[01:00:34] reduce the power consumed by the
[01:00:35] electrical interface between the switch
[01:00:36] and optics
[01:00:38] in this slides i'm showing a conceptual
[01:00:41] view
[01:00:42] of our switchman board inside the
[01:00:43] network switch
[01:00:45] on the left hand side you see the switch
[01:00:49] you see the you see the
[01:00:53] switch assembly which is aligned to our
[01:00:55] recent oif cpo project
[01:00:57] as mark just mentioned and the facebook
[01:01:00] microsoft jdf work
[01:01:02] and you can see that 16 of the 3.2
[01:01:05] terabit
[01:01:05] optics module are arranged in a close
[01:01:08] proximity to the 50
[01:01:10] 1.2 t-switch asic on the cpu substrate
[01:01:13] with four optical modules on each side
[01:01:18] the right-hand side is a close-up view
[01:01:20] of the optical optical module where you
[01:01:22] can see the layout of optical components
[01:01:25] heat spreader lga pads and also the
[01:01:28] fibers
[01:01:29] the optics that should support eight of
[01:01:32] the 400 gig
[01:01:33] interface including f4 and dr4
[01:01:36] and the also supports both external
[01:01:40] or integrated laser
[01:01:43] the optical module is also connected to
[01:01:46] the switch using a retime 100 gig ssr
[01:01:48] interface and we expect this device will
[01:01:51] serve
[01:01:51] one of the building block to enable our
[01:01:54] 51t
[01:01:55] generation with a lower power solution
[01:02:00] all right so let's talk about what we're
[01:02:03] seeing for the challenges
[01:02:06] when operating the data centers we face
[01:02:09] some unique challenges
[01:02:10] associated with the global scaling of
[01:02:12] our infrastructure and try to reuse the
[01:02:14] topology that we have
[01:02:16] i want to use this slide to outline some
[01:02:19] of the considerations
[01:02:22] first and foremost as we embark the cpo
[01:02:24] technology in the network
[01:02:26] cpu as expected has a limited in terms
[01:02:28] of fields of visibility
[01:02:30] that's why the previous discussion we
[01:02:33] talked about fit we talked about the
[01:02:34] power
[01:02:35] pikachu per bit and highly reliable
[01:02:39] lasers such as laser
[01:02:41] such higher reliable components such as
[01:02:43] laser
[01:02:44] which the level of network redundancy is
[01:02:46] definitely critical
[01:02:48] as you you might know the laser failure
[01:02:51] is known to be one of the primary
[01:02:53] contributor to optics failure
[01:02:55] so having reliable laser or
[01:02:58] enable the field service ability by
[01:03:00] using a plug of external laser
[01:03:02] can alleviate some of our operational
[01:03:04] concern
[01:03:07] another thing is the backward
[01:03:08] compatibility is required
[01:03:10] as is to ease the transition and
[01:03:13] minimize the disruption of network when
[01:03:14] we introduce or upgrade
[01:03:16] between generations so the challenge is
[01:03:18] actually not too
[01:03:19] lit not limited to cpo
[01:03:23] another thing is having the forward
[01:03:25] compatibility is also very highly
[01:03:27] desirable
[01:03:28] as you will lay the ground for future
[01:03:30] generation
[01:03:31] and ensure a sustainable technology role
[01:03:33] map
[01:03:36] another thing is having a compatibility
[01:03:38] with existing network topology
[01:03:40] is also critical like i said by
[01:03:44] moving the optic closer to switch with
[01:03:46] the shorter i o rich and less
[01:03:48] lesser this power uh we see cpo helps to
[01:03:52] preserve our e6c
[01:03:54] network radix and fit the power envelope
[01:03:56] when we scale up the switch bandwidth
[01:04:00] and last but not least as mark just
[01:04:03] mentioned
[01:04:04] we have to ensure the optics technology
[01:04:06] are interoperable with open and standard
[01:04:08] interface
[01:04:09] and really try to strive with the
[01:04:11] healthy ecosystem
[01:04:13] let me expand on this point in my next
[01:04:16] slide
[01:04:19] all right um facebook vision
[01:04:22] in our desire state is to build a
[01:04:25] co-package optics
[01:04:27] in a interoperable multi-vendor
[01:04:30] ecosystem
[01:04:31] and we like to have the optionality from
[01:04:34] the best breed of technology
[01:04:36] with this robust supply chain that's a
[01:04:39] big statement
[01:04:40] however i also want to acknowledge that
[01:04:42] uh that
[01:04:43] we do not want to block any innovation
[01:04:47] to optimize further power and density
[01:04:49] improvement
[01:04:50] even when it's not initially be subject
[01:04:53] to such standardization
[01:04:56] so to enable this end state
[01:05:00] we would like to continue to encourage
[01:05:01] your participation from the community
[01:05:03] such as epic in terms of standard
[01:05:06] standards and ongoing
[01:05:08] collaboration to really deliver the
[01:05:10] solution
[01:05:11] for increasing bandwidth density power
[01:05:14] savings
[01:05:15] system reliability enhancement and bring
[01:05:18] that
[01:05:18] to our data center
[01:05:22] like mark also mentioned earlier i would
[01:05:24] like to mention
[01:05:25] that we share the view that this
[01:05:27] technology can
[01:05:28] also be beneficial to application beyond
[01:05:31] the simple network switching
[01:05:32] such as in compute interconnects
[01:05:37] finally i would like to leave you with
[01:05:38] the romance shell at the bottom of the
[01:05:40] slide and give you some timeline
[01:05:42] associated with our switch generation
[01:05:45] looking back in 2017 facebook focused on
[01:05:48] the 100 gig
[01:05:49] e face plate face plate pluggable
[01:05:51] upticks for our 12 point t
[01:05:54] switch generation to preserve our 120a
[01:05:58] radix requirement
[01:06:00] in 2019 we announced the introduction of
[01:06:03] onboard optics to get familiarized with
[01:06:06] ourselves
[01:06:07] and with the advanced optics integration
[01:06:10] in terms of the field deployments
[01:06:14] this approach will carry over to our
[01:06:16] 2021
[01:06:17] when we introduced the 25.6 t switch
[01:06:20] with 200 gige
[01:06:23] in our 501t generation we see cpo will
[01:06:26] be coming into our data center
[01:06:28] and this will align to the 400 gig e
[01:06:30] pmds
[01:06:32] with 100k ssr interface to further
[01:06:34] enable the power reduction
[01:06:36] with a high level integration
[01:06:40] i think this one concludes my talk
[01:06:44] and this is not uh uh
[01:06:47] not a um simple task so i would like to
[01:06:50] acknowledge all the facebook team
[01:06:52] uh for this contribution and thanks to
[01:06:53] epic for having me on this talk
[01:06:56] thank you very much for the amazing
[01:06:57] support that facebook has given us in
[01:06:59] the last
[01:07:00] weeks and also we are aware that many of
[01:07:02] your colleagues are also watching in
[01:07:04] youtube many people wanted to be in the
[01:07:05] sunroom for facebook
[01:07:07] it's been it's been truly great ruby
[01:07:09] thank you for a fantastic presentation
[01:07:11] it is your first ever epic meeting and
[01:07:13] you've been the star
[01:07:14] but you get the epic question we have
[01:07:16] 699 companies in front of you
[01:07:19] what can they do for you and what could
[01:07:21] you do for them
[01:07:22] yeah so i think this is an excellent
[01:07:25] question and
[01:07:26] like you said jose like this is my first
[01:07:28] round
[01:07:29] um so i would like to see that you know
[01:07:32] don't
[01:07:32] i would like to encourage like don't
[01:07:34] stop asking your questions don't
[01:07:35] like you know stop asking the uh what's
[01:07:39] the key
[01:07:39] uh requirements that facebook or other
[01:07:42] data centers are looking at
[01:07:44] and i would like to also encourage to
[01:07:47] participate
[01:07:48] all those standards and trying to build
[01:07:51] that innovation and try to
[01:07:52] move this forward so i think uh
[01:07:56] that's why i'm never afraid to ask
[01:07:58] questions maybe sometimes i ask too much
[01:08:00] so please feel free to say that i
[01:08:01] shouldn't that i shouldn't ask this
[01:08:03] but you highlighted in your presentation
[01:08:05] and also was a little bit pointed out by
[01:08:07] john bowers in the beginning
[01:08:09] a reliability reliability and the
[01:08:12] the the dangers of the danger of having
[01:08:15] a non-working
[01:08:16] semiconductor laser in the previous
[01:08:18] meetings we organized a lot of
[01:08:20] events about wave level testing of
[01:08:22] indium phosphite
[01:08:24] do you do you still have a problem with
[01:08:26] the yield of semiconductor lasers
[01:08:28] is that a challenge or is it 100 working
[01:08:31] semiconductor lasers that go packaged
[01:08:34] yeah that's a great question i think you
[01:08:36] know without giving too much
[01:08:38] of the uh you know what we're thinking
[01:08:41] i uh uh
[01:08:45] it's it's gonna be really challenged
[01:08:48] like you know to continue on
[01:08:50] uh uh like i said that you know
[01:08:54] well it's a
[01:08:58] right now the semiconductor does have as
[01:09:00] especially for the laser it actually has
[01:09:03] its uh its needs and also has its
[01:09:06] applications
[01:09:07] but um yeah uh
[01:09:10] i the reason why i asked this question
[01:09:12] we do have great companies in the
[01:09:13] testing
[01:09:14] in the testing characterization of
[01:09:16] indium phosphide semiconductor lasers
[01:09:18] i already said in the beginning the
[01:09:19] introduction and this is a great rule
[01:09:21] for cooperation because as you have said
[01:09:23] everyone ruby is here open for
[01:09:26] collaboration summer
[01:09:27] cooperation this is an epic meeting and
[01:09:29] we have two questions for you in the
[01:09:30] chat the first one
[01:09:31] is coming from santa barbara ucsb john
[01:09:34] bowers thank you very much for this
[01:09:35] what's on your mind
[01:09:38] so you raised a good question about
[01:09:39] laser reliability but
[01:09:41] i wonder if we're already at the point
[01:09:42] where simplicity
[01:09:44] is more reliable right so if you had 500
[01:09:47] sources as thomas mentioned you know 500
[01:09:49] pluggables
[01:09:50] is not very reliable right just
[01:09:53] every time you plug and unplug them and
[01:09:55] the certies it's like you know if you
[01:09:57] didn't have integrated circuits right if
[01:09:58] you had
[01:09:59] a billion individual transistors that
[01:10:00] would be very very unreliable
[01:10:03] aren't we already at that point where
[01:10:05] co-package is more reliable than
[01:10:07] plugables
[01:10:09] um i think there's a big statement john
[01:10:11] so um
[01:10:12] i don't think that that's a
[01:10:16] there's still a lot of study that needs
[01:10:18] to be done uh
[01:10:19] and there are still quite a lot of uh
[01:10:23] uh data to be collected as thomas
[01:10:26] mentioned
[01:10:28] um i would say that um for the cpos that
[01:10:31] we do see the need for that to reduce
[01:10:34] the power however
[01:10:35] uh from the reliability or the you have
[01:10:38] to think that the cpo
[01:10:39] like we we see that pluggable is still
[01:10:41] gonna be in our generation
[01:10:44] um and is that in the point that to
[01:10:46] cutting over all the way to cpo
[01:10:48] uh that is a really a big question to
[01:10:52] answer
[01:10:52] but uh we still see a lot of tiers in
[01:10:55] our network we'll have the plugable
[01:10:58] and cpo will come in is it going to be a
[01:11:00] stagger quite
[01:11:01] step-by-step questions and we got to
[01:11:04] continue on with the incremental
[01:11:07] change in our networks so
[01:11:10] to answer your question uh it's really
[01:11:13] uh something
[01:11:13] uh there are more to study and there are
[01:11:15] more to uh
[01:11:17] to evaluate down the road ruby for me
[01:11:19] that sounded like a
[01:11:21] not yet that's what i really read
[01:11:23] between the lines
[01:11:24] john what is your answer to your own
[01:11:26] question is is cpo already more reliable
[01:11:29] than plugable
[01:11:31] i think so and i don't want to steal
[01:11:32] from uh
[01:11:34] gustavo's question but i think the
[01:11:36] advantage of integrating lasers like
[01:11:38] intel does is you can do wafer level
[01:11:40] burn-in and testing complete
[01:11:42] testing complete link right and so it's
[01:11:44] much
[01:11:45] i think higher yield than having to
[01:11:48] assemble things and test along the way
[01:11:50] with known good parts so
[01:11:51] just like you'd never make a server with
[01:11:54] with
[01:11:54] seven billion individual transistors you
[01:11:56] would never do to me even at
[01:11:58] 51 terabytes you wouldn't have 500
[01:12:01] 100 gig pluggables that's just un you
[01:12:03] know completely unfeasible
[01:12:05] or even 125 400 gig modules so
[01:12:08] i think co-package is the only way to go
[01:12:12] i i would actually think that this is
[01:12:15] still an intermediary step
[01:12:17] still internet but but we've been
[01:12:18] talking about so many intermediary steps
[01:12:20] but i still dream i still dream of
[01:12:22] further integration of photonics and
[01:12:23] electronics and cpo
[01:12:25] but that's a that's a futuristic dream
[01:12:27] and i think you shared that dream as
[01:12:28] well from the i see in your research
[01:12:30] but i would like to go to now lumiface
[01:12:32] gustavo i think we had a great
[01:12:34] introduction to your question from john
[01:12:35] powers but what's on your mind
[01:12:38] well i think my question goes i mean
[01:12:39] already quite a little bit a lot of
[01:12:41] things
[01:12:41] were discussed is more related to the
[01:12:43] fact that when we start to to co-package
[01:12:45] a lot of complex uh
[01:12:47] elements on top of the chip i do i do
[01:12:49] share the same vision
[01:12:50] as john off of uh yeah that simply be
[01:12:53] there no way than
[01:12:54] than than that but i i really had adding
[01:12:56] too much cost
[01:12:57] on the testing side especially if we
[01:12:59] start to think that now we have to do
[01:13:01] away for level burning and
[01:13:03] i just want to have a general opinion
[01:13:04] here if that that actually makes sense
[01:13:06] on the on the cost side because we
[01:13:08] still need to keep going to the cost per
[01:13:10] bit so
[01:13:12] that's a little bit what i wanted to
[01:13:14] start the discussion on that topic
[01:13:16] yeah so uh i do see to answer your
[01:13:19] question i do think that
[01:13:20] uh the wafer level burning is gonna be
[01:13:22] necessary and
[01:13:23] we need to find a way to really simplify
[01:13:26] the supply chain as well as the cost
[01:13:28] associated with that
[01:13:29] like you mentioned so um yeah this is
[01:13:32] something that i
[01:13:33] don't think it's going to be prohibitive
[01:13:36] in terms of cost but
[01:13:38] we would like to keep our eyes really
[01:13:40] close on this one and making sure as
[01:13:42] you know that data center is very we
[01:13:45] need to think about
[01:13:46] overall picture the total cost ownership
[01:13:50] so those are things that to be
[01:13:51] considered
[01:13:53] ruby i think we we hear from you we
[01:13:56] heard from intel and microsoft
[01:13:58] and we are all photonics people i think
[01:14:01] we need to hear
[01:14:02] for for this dream that that the young
[01:14:04] has put there
[01:14:05] we need to hear also from the
[01:14:06] semiconductor companies i think they
[01:14:08] also have
[01:14:08] uh uh to address this and i would like
[01:14:11] to see the view of a
[01:14:13] semiconductor pcb manufacturer leader we
[01:14:16] have atns
[01:14:17] in the room henrik thank you very much
[01:14:19] for being with us today
[01:14:24] atns is one of the market leaders in the
[01:14:26] pcb manufacturing
[01:14:28] and they have a wish and they have a
[01:14:29] dream to integrate photonics
[01:14:31] enlighten us tell us what's in your mind
[01:14:34] that's totally right i hope everyone can
[01:14:38] see my screen
[01:14:39] crystal clear okay perfect so
[01:14:42] uh first of all uh good afternoon good
[01:14:46] morning to all of you
[01:14:48] uh thank you for the possibility to
[01:14:50] present
[01:14:52] uh our approach on optical integration
[01:14:55] on pcb at the glance
[01:14:58] so my name is eric schlafer i am program
[01:15:01] manager in our research and
[01:15:03] the development department at 18 days in
[01:15:07] austria
[01:15:08] at 18s is coming from integration
[01:15:11] and printed circuit boards uh we are we
[01:15:14] have a
[01:15:15] revenue approximately one billion euros
[01:15:18] and we are dealing in six
[01:15:22] manufacturing plants in europe and in
[01:15:25] asia
[01:15:26] as well so
[01:15:29] uh what what is the dream or what is
[01:15:32] ongoing and
[01:15:34] uh in terms of optical integration uh
[01:15:37] 18s has on the left hand side you can
[01:15:41] see
[01:15:41] what the plan is to do for the technical
[01:15:44] integration
[01:15:45] yeah on the left hand side
[01:15:48] i just will put on my
[01:15:51] my laser pointer on the left hand side
[01:15:54] you see
[01:15:54] uh 18 days is stealing for a very long
[01:15:57] time
[01:15:58] uh to integrate electrical components in
[01:16:01] a stack up of a pcb
[01:16:02] yeah that might be we have we have
[01:16:05] launched
[01:16:06] uh with resistors and capacitors and now
[01:16:09] we're also able to
[01:16:11] integrate active electrical components
[01:16:14] uh to interconnect it somehow on one on
[01:16:18] both sides uh
[01:16:20] and for the short term yeah
[01:16:24] it means we would like to integrate
[01:16:27] electrical components within a carrier
[01:16:30] body
[01:16:31] yeah this is what we're already doing
[01:16:34] you see on the y-axis the ios per
[01:16:37] square millimeters and on the
[01:16:41] x-axis you see more or less the ios
[01:16:44] per millimeter per layer yeah so when it
[01:16:47] comes to
[01:16:48] a hdi pcb we are talking
[01:16:52] about a 10 uh interconnect
[01:16:55] or 10 how can i say this is more or less
[01:16:58] the pitch size here what
[01:17:00] what we are talking about yeah and
[01:17:03] uh the uh the goal what we are aiming
[01:17:06] is because we also are dealing with
[01:17:09] substrate like pcb
[01:17:10] yeah that we are using uh the
[01:17:13] integration concept
[01:17:15] yeah for a substrate like pcb
[01:17:18] where we can manufacture a line with uh
[01:17:22] below 10 and 5 microns
[01:17:25] yeah up to a flip chip pga
[01:17:28] and together what we are aiming in this
[01:17:31] in this
[01:17:32] uh project and we also we need
[01:17:34] additional partners
[01:17:36] along the supply chain this is what we
[01:17:39] are doing with
[01:17:39] fraunhofer for instance uh to integrate
[01:17:43] an electro optical glass interposer
[01:17:47] in the stack up of a pcb yeah this is
[01:17:49] more or less what we would like to
[01:17:51] achieve
[01:17:52] in the mid or long term plan
[01:17:55] so the advantages of this concept
[01:17:59] is more or less uh the modules with a
[01:18:02] high level of
[01:18:03] integrated components and this
[01:18:07] would be realized on pen level packaging
[01:18:10] this might be a big benefit in terms of
[01:18:14] costs here
[01:18:15] because we are talking of a
[01:18:16] manufacturing size of
[01:18:18] one port of 18 times 24
[01:18:21] or 21 times 24 inches
[01:18:25] second advantage is when we integrate
[01:18:29] a higher level of components
[01:18:32] electrical and optical it means we are
[01:18:35] able that we can shorten the circuit
[01:18:39] lengths
[01:18:40] of the signal path here and that means
[01:18:42] that
[01:18:43] that is a big impact in terms of losses
[01:18:47] and power consumption as well and
[01:18:51] the third advantage of such an idea is
[01:18:54] we can
[01:18:55] combine and link electrical components
[01:18:59] uh and photonics uh in the same
[01:19:03] stack up of a pcb or of an interposer
[01:19:08] and this concept should be targeted
[01:19:12] uh on application like miniaturized
[01:19:16] optical modules for plug cables
[01:19:19] or mid boards modules or it might be
[01:19:22] addressed also to
[01:19:25] to serve a higher level of integration
[01:19:29] for optical switches
[01:19:31] and what we are looking more or less in
[01:19:35] in
[01:19:36] this epic consortium is to find
[01:19:39] a strong partners here that we can
[01:19:43] uh on one hand side that we can check
[01:19:46] are we on the right page
[01:19:47] with this concept and on the other hand
[01:19:50] side that we also find
[01:19:52] potential partners along the supply
[01:19:55] chain yeah
[01:19:56] that we see that we are on the right
[01:19:58] page for the development
[01:20:00] and you're gonna find many of those
[01:20:01] partners in this consortium thank you so
[01:20:03] much for this
[01:20:04] we have a couple of questions for you
[01:20:05] the first one is coming from one of
[01:20:07] these companies it is like a one-stop
[01:20:09] shop for all photonic intrinsic circuits
[01:20:10] from
[01:20:11] design manufacturing all the way to
[01:20:13] testing photon path
[01:20:15] thank you very much douglas ring with us
[01:20:16] what's on your mind
[01:20:18] hi jose thank you thank you for the
[01:20:19] introduction no i'm just curious about
[01:20:21] the the waveguiding platform that you're
[01:20:23] using for the
[01:20:25] for your interposer which material do
[01:20:27] you do but both already
[01:20:29] insert here but it's i would like to if
[01:20:32] you can share more details on the
[01:20:34] guiding platform that you're using uh
[01:20:37] we are using it means that uh
[01:20:41] how can i say that the standard the
[01:20:42] carrier should be a standard ff4
[01:20:45] material
[01:20:46] what has been used for many many years
[01:20:48] yeah
[01:20:49] to to transfer to to transmit
[01:20:53] or to serve a platform for electricals
[01:20:56] but we're also looking to integrate
[01:20:58] glass
[01:20:59] it means the concept for the electrical
[01:21:02] optical glass
[01:21:05] electro optical interposer is based on
[01:21:07] glass material
[01:21:10] thank you very much er ruth hubert a
[01:21:14] ex ceo of multiphoton optics you have
[01:21:17] something to comment on this right ruth
[01:21:20] actually i was i was very thrilled that
[01:21:22] 80 and s is back
[01:21:24] back on the page and i was just you know
[01:21:27] like teasing
[01:21:28] them a bit with with the old old stuff
[01:21:30] we did
[01:21:31] i think 15 years ago or so uh where we
[01:21:34] had been already
[01:21:36] uh achieving world record with the new
[01:21:38] components
[01:21:39] and i just uh started to follow up with
[01:21:42] that
[01:21:43] i think yeah and i'm in excel more
[01:21:46] more than teasing more than this you
[01:21:48] need to you need to help them and i
[01:21:49] think there is nobody better to help
[01:21:51] than
[01:21:51] than you i would like to make them an
[01:21:53] introduction we have a father
[01:21:56] i always need you and we have another
[01:21:57] question from oliver
[01:21:59] kirsch oliver what's on your mind
[01:22:04] ah sorry for the conclusion i actually
[01:22:07] answered to the question from protonp
[01:22:08] from douglas okay so i don't have a
[01:22:11] question
[01:22:12] very good but it's great to have you in
[01:22:13] the room all the way from frank hoffer i
[01:22:15] said
[01:22:16] m in berlin it's mirrored but i know who
[01:22:18] you are thank you very much for being
[01:22:19] here
[01:22:19] and i think it is time to continue with
[01:22:21] this fantastic program we go to the next
[01:22:23] presentation
[01:22:24] and this is from a success story success
[01:22:26] story of the integrated photonic
[01:22:28] european ecosystem we go to one of the
[01:22:29] companies making a difference
[01:22:31] in the transceiver industry we go to
[01:22:33] sikoya svengo the ceo seco yeah it's an
[01:22:36] honor to have you here in this meeting
[01:22:37] the floor and the attention of everyone
[01:22:39] goes to my favorite transceiver silicon
[01:22:41] photonics company goes to you
[01:22:45] thank you very much jose it's always a
[01:22:47] pleasure
[01:22:48] attending those epic meetings here
[01:22:51] thanks for having me
[01:22:52] uh apic of course as we all know is the
[01:22:55] european photonic and
[01:22:56] industrial consortium but for us apex
[01:22:58] are actually
[01:23:00] electronic photonic integrated circuits
[01:23:02] so so no matter what you use it for apic
[01:23:04] is always great right so
[01:23:06] so making that statement all right um
[01:23:09] back to
[01:23:09] to the real topic so we we're going to
[01:23:12] talk about uh co-package optics mostly
[01:23:14] but let's just
[01:23:15] get started with just take a look at
[01:23:16] what we do today
[01:23:18] uh so yeah we are in the market for some
[01:23:20] time
[01:23:21] um mostly focusing on the transceivers
[01:23:23] and engines for
[01:23:24] wireless and and data centers um
[01:23:28] we had the privilege uh to really work
[01:23:30] with the foundries and develop the
[01:23:32] process that allows monolithic
[01:23:34] integration of electronics and and
[01:23:35] photonics so we can integrate both wave
[01:23:38] guides
[01:23:39] modulators pins we can also integrate
[01:23:41] the transistors
[01:23:43] and and we believe that's a very good
[01:23:44] platform really actually to
[01:23:46] to do co-package optic solutions um
[01:23:50] currently as co-package doesn't really
[01:23:52] exist in the market we're focusing on
[01:23:53] the transceiver level and we have a
[01:23:55] broad
[01:23:55] you know portfolio of products all the
[01:23:57] way from you know 25 gig to 100 gig 400
[01:24:00] gig
[01:24:01] working on 800 gig and and and beyond
[01:24:04] right um what i should say is though is
[01:24:07] that we are not in the foundry business
[01:24:08] so we are not a founder we don't own the
[01:24:10] phone we are not married to
[01:24:11] fundraise so we're you know very
[01:24:13] agnostic and you know transparent so we
[01:24:15] can you know use any any phone we're
[01:24:17] working with a
[01:24:18] large range of different foundries and
[01:24:20] and we're just selecting the phone based
[01:24:22] on the requirements that that we have
[01:24:24] all right um and i think that's a major
[01:24:27] transition a major change that recently
[01:24:29] happened that
[01:24:30] you know a lot of the technology
[01:24:31] developments are actually happening at
[01:24:33] the foundry level
[01:24:35] and and the companies like us we focus
[01:24:38] more on the
[01:24:38] on the solution level so bringing value
[01:24:41] to to the customer
[01:24:42] and understanding the the applications
[01:24:44] understanding the problems and
[01:24:46] and finding solutions for problems
[01:24:48] that's
[01:24:49] that's really one of the value
[01:24:51] proposition and that that we bring
[01:24:53] right all right um let's move on to the
[01:24:57] to the real topic
[01:24:58] um co-package optics um
[01:25:03] so basically what is it so we we
[01:25:06] we used words like embedded optical
[01:25:09] modules mid board optics or onboard
[01:25:11] transceivers
[01:25:12] for for many years those were
[01:25:15] miniaturized
[01:25:15] versions of transceivers that we use
[01:25:18] inside of the boxes
[01:25:19] or switches routers and
[01:25:22] co-package optics has been you know
[01:25:24] introduced as
[01:25:26] a word where the optics goes to the
[01:25:28] package level of the asics
[01:25:30] so that could be you know mostly
[01:25:32] focusing on you know
[01:25:33] switches at this point we could also be
[01:25:36] you know gpus or other things
[01:25:38] um so this is where this name co-package
[01:25:42] is coming from however i see
[01:25:43] a lot of people using co-package
[01:25:45] nowadays more in a more broader way
[01:25:48] so so everything that goes into the
[01:25:50] package
[01:25:51] i mean into the box is is being
[01:25:53] considered co-package so
[01:25:55] that's sometimes sometimes confusing so
[01:25:57] when you hear people talk about co
[01:25:58] packages just always ask
[01:26:00] is it really is that was with the
[01:26:02] package the
[01:26:03] the ac package level it really just
[01:26:06] meant
[01:26:07] or is it maybe maybe the box or
[01:26:08] something else what is the package
[01:26:10] actually
[01:26:11] um but in either case so we we started
[01:26:13] in the 90s with with transceivers and
[01:26:16] um pluggable transfers and and the main
[01:26:18] reason why we've done that is cost
[01:26:20] reduction so
[01:26:21] multi-sourcing standardization has
[01:26:24] helped a lot to bring the cost down with
[01:26:26] and and that that's where the whole
[01:26:28] industry was benefit from
[01:26:30] now if we go to the to the co package
[01:26:33] optics
[01:26:34] uh level uh really the main driver for
[01:26:37] that is as power reduction as mark and
[01:26:39] ruby and others have said already
[01:26:41] uh but we still want to benefit from
[01:26:44] from multi-sourcing and standardization
[01:26:46] right to keep the cost per year a bit
[01:26:47] down
[01:26:48] drive it further down and this is why we
[01:26:50] still continue
[01:26:51] to need to lean standards when it goes
[01:26:55] above that things become a little bit
[01:26:58] more you know
[01:26:59] weak when it comes to to standards and
[01:27:02] multi-sourcing the question always
[01:27:04] is multi-sourcing and standards at what
[01:27:06] level
[01:27:07] right so if we for example talk about
[01:27:08] co-package optics for
[01:27:10] network interface cause for the compute
[01:27:12] node the endnote as as mark described it
[01:27:15] you know um you can always dispose a
[01:27:18] network interface card if
[01:27:19] it breaks i don't know if you want to
[01:27:21] repair it
[01:27:22] uh the switch level you certainly want
[01:27:25] to repair rather than
[01:27:26] than scrap why that there's no question
[01:27:28] so
[01:27:29] um so at what level are we going to do
[01:27:33] the the centers and once we go to the
[01:27:34] disaggregation level or even to consumer
[01:27:36] type products things become even more
[01:27:39] unknown at this point but the but the
[01:27:41] ecosystem is building and what i'm
[01:27:42] trying to say here is
[01:27:44] that this entire co-package optics uh
[01:27:47] effort and initiative and all the
[01:27:48] alliances and partnerships that are
[01:27:50] building
[01:27:51] this is not meant just to build a 50-t
[01:27:53] switch this is a
[01:27:54] this is a paradigm shift this is an
[01:27:56] ecosystem that's building
[01:27:58] and and this is a whole new world of for
[01:28:01] photonics and it's it's very exciting
[01:28:04] each of these uh individual applications
[01:28:06] here have its own attributes when it
[01:28:09] comes to you know
[01:28:10] technical criterias like reliability
[01:28:12] cost
[01:28:13] bandwidth reach laser services the
[01:28:16] location of the of the laser is it
[01:28:18] is it really integrated into into the
[01:28:20] device or is it remote
[01:28:22] serviceability i touched on that already
[01:28:25] how do we do it with that and and
[01:28:28] there's a nice
[01:28:28] nice work going on in these different
[01:28:31] alliances but also in particular at oif
[01:28:33] where
[01:28:34] the company is mentioned here and most
[01:28:36] of them are
[01:28:37] you know have been given the talks
[01:28:39] already here today and they're working
[01:28:41] really hard on
[01:28:42] defining what it actually means at the
[01:28:45] at the technical level so what is it
[01:28:48] that
[01:28:48] um what is it that we do in in that
[01:28:52] regard
[01:28:53] see a couple of examples that we
[01:28:55] actually do so we have a 100 gig
[01:28:58] embedded optical module for for some
[01:29:00] time we're taking this to
[01:29:01] 300 gig next
[01:29:05] is the 800 gig which is which is
[01:29:07] sampling now
[01:29:09] so this is utilizing our 100 gig per
[01:29:11] lambda
[01:29:12] um silicon photonic chip that that we
[01:29:15] have developed
[01:29:16] so this integrates all the all the
[01:29:19] transmitter
[01:29:20] functions the receive functions the tias
[01:29:23] the drivers everything is monolithically
[01:29:25] integrated into
[01:29:27] into these two chips here in this case
[01:29:28] so each of the chip can do you can do
[01:29:30] 400 gig
[01:29:32] from here we're doing a 1.2 terabit
[01:29:36] and it's packaged in a way that you can
[01:29:38] you know um
[01:29:40] you can solder this down to the to the
[01:29:41] main board and the main board
[01:29:43] could be either you know any type of
[01:29:45] board it could also be the substrate of
[01:29:47] a
[01:29:47] of a you know at the substrate of an
[01:29:50] asic package basically
[01:29:52] and from here we take it then to to 3.2
[01:29:55] t um what we need to do to make that
[01:30:00] happen is to reduce power
[01:30:02] so as i think thomas mentioned that
[01:30:05] already power reduction is a is a
[01:30:07] critical point
[01:30:08] uh our initial 100 gig prolonger
[01:30:10] products we really focus on
[01:30:12] getting things work and make them
[01:30:14] interoperable
[01:30:16] optimizing performance and not so much
[01:30:19] driven by
[01:30:20] by reducing power so power reaction is
[01:30:23] something left we can do
[01:30:24] um there's some work a lot of work is
[01:30:27] going on on the modulator side but i
[01:30:28] also believe that
[01:30:30] um when it comes to power reduction
[01:30:32] there's a lot in it from the
[01:30:34] from the system perspective because the
[01:30:36] especially the the switch asics are
[01:30:38] currently not really optimized for for
[01:30:40] co-package optics
[01:30:42] all right so this is what we do um happy
[01:30:45] to take any questions
[01:30:46] and thank you for your attention thank
[01:30:49] you very much has been super interesting
[01:30:51] presentation
[01:30:52] and well you know what is the question
[01:30:54] that comes next now what can you do for
[01:30:57] them what can they do for you so
[01:30:59] i mean do you have any challenge for the
[01:31:02] supply chain in photonics that uh
[01:31:04] that is attending this meeting yeah i
[01:31:06] think we are one of the pioneers of
[01:31:08] silicon photonics so we have had
[01:31:09] all sorts of of issues and when it comes
[01:31:12] to supply chain and and partners so
[01:31:14] we've we've been really starting
[01:31:16] initially
[01:31:17] more working on supply chain than
[01:31:19] working on chip development so
[01:31:21] all the way from burn in from
[01:31:24] working with mems foundries for
[01:31:27] packaging purposes
[01:31:28] working with laser alignment uh
[01:31:31] fiber alignment type interfaces with
[01:31:34] wafer level testing so we have built a
[01:31:37] whole set of
[01:31:38] um of ecosystem that allows to do all of
[01:31:40] these things and then all of these
[01:31:42] individual tasks are now now fully fully
[01:31:45] automized and there's a supply chain
[01:31:47] from equipment
[01:31:49] from all the different equipment vendors
[01:31:50] available so i think we we contributed
[01:31:52] in that
[01:31:53] regard and we continue to to contribute
[01:31:56] um what we can offer is developing a
[01:31:58] solution so we have a set of
[01:32:00] technologies available
[01:32:01] and happy to you know customize that
[01:32:04] taylormade that you
[01:32:05] to really bring value to the customer
[01:32:07] and and to serve the customer needs this
[01:32:09] is what we
[01:32:10] can do uh when it comes to what what
[01:32:13] others can do for us well we
[01:32:15] happy to take your the challenge right
[01:32:17] so come to us with
[01:32:18] with challenges but give us problems
[01:32:20] give us give us applications
[01:32:22] uh because um you know you know at the
[01:32:25] cpo level
[01:32:26] a lot of things are standardized so you
[01:32:28] know there are these initiatives that
[01:32:29] have been mentioned a few times
[01:32:31] but there is i would say 90
[01:32:35] of what i see is not standardized 90 of
[01:32:38] of all the different opportunities that
[01:32:39] i'm seeing are not standards these are
[01:32:41] individual proprietary customized
[01:32:44] tailor-made uh type type products
[01:32:47] and and we love to do those things so so
[01:32:50] come to us and work with us and we
[01:32:53] will figure out together whether or not
[01:32:54] we come with it at a good conclusion or
[01:32:56] not
[01:32:58] we have one question from a jerk er
[01:33:00] exceed
[01:33:01] york would you like to make the question
[01:33:03] by yourself
[01:33:04] yes of course thanks anna thanks ben
[01:33:08] and a question for the modules you
[01:33:10] showed on the last page
[01:33:12] the different applications
[01:33:15] about what distance applications do we
[01:33:17] talk here
[01:33:18] oh those are all yeah i should have
[01:33:21] mentioned that
[01:33:22] all of those are currently dr you know
[01:33:24] do4
[01:33:25] fr4 type interfaces so so two kilometer
[01:33:28] up to two kilometer
[01:33:30] um but that's basically the baseline for
[01:33:33] all the designs that we do currently um
[01:33:36] we're working on 200 gig because at the
[01:33:39] at the 3.2 tera
[01:33:41] cpo it's you know
[01:33:44] there are some activities going on to
[01:33:46] really take this to 200 gig columnar
[01:33:49] uh which would reduce fiber count and
[01:33:51] and you know
[01:33:52] that would require you know um inverse
[01:33:55] gearbox you know so you would come in
[01:33:57] 100 gig electrical and then it would you
[01:33:59] know go to 200 gig optical
[01:34:01] uh so so there are some activities but
[01:34:03] but the baseline is really 100
[01:34:05] km therefore ffo thanks
[01:34:10] okay so thank you very much for the
[01:34:12] question thank you very much sven for
[01:34:13] this
[01:34:14] super nice talk let's move in the
[01:34:16] program so ned
[01:34:17] next talk is a from a yo
[01:34:20] ho hoja from ranovus
[01:34:23] so kojak if you want to share your
[01:34:26] screen
[01:34:27] sure yeah thanks so much
[01:34:32] excellent um good morning and good
[01:34:36] afternoon
[01:34:37] uh my name is rojas salimi i'm
[01:34:39] responsible for
[01:34:41] business development at ranovos uh
[01:34:44] so today i'm just going to talk about
[01:34:46] our cpo platform which is called odin
[01:34:51] um mark
[01:34:55] and you know ruby have done a great job
[01:34:58] covering
[01:34:59] the problem statement so i'm not going
[01:35:01] to talk about this chart
[01:35:03] so go over the odin and leave some more
[01:35:06] time maybe for
[01:35:07] questions now
[01:35:10] what is odin we designed this odin
[01:35:14] for as an scalable optical engine
[01:35:19] and uh focus on some of the key
[01:35:23] um innovation for this platform
[01:35:27] one of them is monolithic integration
[01:35:31] by that we mean that we can integrate
[01:35:35] some of the high speed electronics
[01:35:39] and control functions within the same
[01:35:42] die on that for example
[01:35:46] we are as we have said in the previous
[01:35:49] talks we are heavily invested on ring
[01:35:52] resonators
[01:35:54] and we believe that integration of
[01:35:56] driver
[01:35:58] with the ring resonator the control
[01:36:00] function and marks
[01:36:02] the marks within the same guy add a lot
[01:36:05] of value in terms of the control
[01:36:07] and hitting the performance cost
[01:36:10] target which are required by the data
[01:36:13] center
[01:36:14] so that's one key the other thing on the
[01:36:17] laser
[01:36:18] we have two options we build it in a way
[01:36:21] that we can do
[01:36:22] two for different application
[01:36:27] in our discussions with customers
[01:36:30] some of them prefer remote laser source
[01:36:33] because there are some applications that
[01:36:35] that app you know remote laser should
[01:36:38] make a lot more sense
[01:36:40] than laser attach and some of them
[01:36:42] especially on lower capacity
[01:36:44] laser attach is a key so that's another
[01:36:47] one that
[01:36:48] the platform supports
[01:36:51] the other thing we did was
[01:36:54] we didn't want to handle protocols in
[01:36:57] the device
[01:36:58] so we wanted to be protocol agnostic
[01:37:02] this way we can basically do direct
[01:37:05] drive
[01:37:06] or connect to any other edi externally
[01:37:10] and provide the functions that is
[01:37:12] required
[01:37:14] because of that we also get very low
[01:37:17] latency
[01:37:18] so our is in the few hundred picosecond
[01:37:21] range
[01:37:22] and to end from you know entering on the
[01:37:24] driver and leaving on the tia side
[01:37:27] so overall i think you know
[01:37:30] the agreeing on where to draw the line
[01:37:35] on what goes in the silicon photonics
[01:37:37] phonolytic integration
[01:37:38] versus the electronic die
[01:37:42] is really the critical decisions to hit
[01:37:45] the cost power targets for this
[01:37:48] um in terms of the soccerization
[01:37:51] we also see that as very important
[01:37:54] because you know
[01:37:54] they as previously speakers have
[01:37:56] discussed
[01:37:57] there's different requirements from data
[01:37:59] centers
[01:38:01] for example in some you need dr
[01:38:04] some you need fr so being able to
[01:38:08] socketize it and be able to remove it
[01:38:11] for the adding different features or
[01:38:15] looking at it as you know testing
[01:38:18] or basically uh
[01:38:22] manufacturability it's a very essential
[01:38:24] piece
[01:38:25] so this is you know very high level of
[01:38:27] the odin platform in terms of
[01:38:30] you know what we see as
[01:38:34] very important is really the
[01:38:37] standardization
[01:38:38] and having a tier one ecosystem to
[01:38:41] support it
[01:38:43] uh march last year we
[01:38:46] introduced our odin platform and we you
[01:38:49] know
[01:38:50] along with senco te and ibm
[01:38:54] and the list is much longer right now
[01:38:57] so for us the key is to have a strong
[01:39:00] ecosystem to be able to provide this
[01:39:04] end-to-end
[01:39:05] solution from you know the concept
[01:39:08] to the product so i think as mark and
[01:39:11] ruby mentioned
[01:39:12] oif is uh doing a fantastic job
[01:39:16] to standardize some of the key things
[01:39:20] for co-package optics as it applies to
[01:39:24] ethernet application
[01:39:26] for 51.2 we are a major driver there we
[01:39:30] see that
[01:39:31] very critical to standardize on some key
[01:39:34] functions
[01:39:35] to be able to hit basically price power
[01:39:38] and cost targets
[01:39:41] the the other thing is uh i think
[01:39:45] important is that we work directly with
[01:39:48] fabs
[01:39:49] and the fabs taking advantage of
[01:39:53] existing 300 millimeter high volume cmos
[01:39:56] manufacturing
[01:39:58] because that's you know there's a lot of
[01:40:00] investment has gone to that so if you
[01:40:02] can take advantage of that
[01:40:03] that it have too many boutique facts
[01:40:07] which are very hard you know to
[01:40:09] [Music]
[01:40:11] qualify a final product so
[01:40:14] by focusing on tier one fabs you know we
[01:40:17] can get to mature pdk
[01:40:18] faster we can have cad tool support just
[01:40:21] like the electronic industry
[01:40:24] and the qualified ip
[01:40:27] two pieces that again you know has been
[01:40:30] referred to
[01:40:31] several times in these talks are fiber
[01:40:33] and laser attacks which are very
[01:40:35] critical
[01:40:36] on co-package optics
[01:40:40] and flip chip compatibility is also very
[01:40:42] important because you know heterogeneous
[01:40:45] integration is going to be the key for
[01:40:47] growth of moore's law if you look at the
[01:40:49] next 10 years
[01:40:50] so these are the critical things on the
[01:40:53] fab
[01:40:54] on the laser technology what we are
[01:40:57] working on is basically
[01:41:00] being able to do high efficiency
[01:41:06] keep the high efficiency at higher
[01:41:07] temperature operation
[01:41:09] and qualify those lasers for particular
[01:41:12] fab
[01:41:13] that we are working for it and also
[01:41:16] working with the customers to
[01:41:18] standardize a multi-wavelength
[01:41:20] grid that we can look at beyond what we
[01:41:23] have today which is you know 400 gig
[01:41:25] fr4dr4
[01:41:27] so looking at that and also
[01:41:30] the disaggregation for compute that's
[01:41:32] another critical piece
[01:41:34] for that uh in terms of connectors i
[01:41:37] think you know
[01:41:38] uh we are partnering with senko as tiger
[01:41:42] mentioned
[01:41:42] it's very important to have these
[01:41:43] optical connectors that are low loss
[01:41:47] high density and obviously gives you a
[01:41:49] bit of room on your link margin
[01:41:52] and you know having the lga sockets that
[01:41:55] we can
[01:41:55] standardize on and that's really
[01:41:58] important because you know then
[01:42:00] we can qualify it and get the cost down
[01:42:04] rather than each one of them be a unique
[01:42:07] uh
[01:42:07] footprint um and the last item i think
[01:42:12] is on
[01:42:12] just like the electronic industry
[01:42:15] having these osats is critical to our
[01:42:18] growth
[01:42:19] because it allows us to manufacture the
[01:42:22] you know fiber attached laser attached
[01:42:25] and again not have too many boutique
[01:42:28] stuff that a supplier has to go through
[01:42:30] to enable it
[01:42:32] so in summary i think from our
[01:42:35] side to really build these optical
[01:42:38] engines which are scalable it's very
[01:42:41] important that we cooperate and
[01:42:43] we basically get some standards in place
[01:42:46] and have
[01:42:47] a strong ecosystem to support it
[01:42:50] and thank you very much for your time so
[01:42:53] i can pause here for any questions
[01:42:55] thank you very very much you know i
[01:42:56] can't hide this i'm a huge fan of
[01:42:58] quantum dot technology really really
[01:43:00] huge fan so when i see a presentation
[01:43:01] from ronaldo's
[01:43:02] and i see how far you have gone on the
[01:43:05] multi-waving laser by using the broad
[01:43:07] game material that quantum dot is
[01:43:09] i guess yes hats off to you thank you
[01:43:11] very much for what you are achieving
[01:43:13] but nobody can do anything great along
[01:43:14] anymore yet
[01:43:16] what can they do for you or can you do
[01:43:18] for them yeah
[01:43:19] so so i think as i said it's very
[01:43:22] important to work
[01:43:24] directly with the fabs and we need a bit
[01:43:27] of
[01:43:28] you know consolidations on the fabs you
[01:43:30] know as you know on the electronic
[01:43:32] industry
[01:43:33] you know there is like few tier ones and
[01:43:37] you know to because there's a lot of
[01:43:39] investments and they have to be
[01:43:40] qualified that to be ip suppliers
[01:43:43] they have to be cads so that's what we
[01:43:46] see as critical
[01:43:47] factor for success of you know cpo
[01:43:51] has the number of applications or
[01:43:54] you know amazingly is a big market
[01:43:57] and that's what we see as that you are
[01:43:59] really kind but you know what i see
[01:44:01] critical
[01:44:02] well the security guard is going to show
[01:44:03] me this light and you told me that you
[01:44:05] need
[01:44:05] high yield and low cost fiber attached
[01:44:08] technology i see this as critical
[01:44:10] because there are very different ways of
[01:44:12] doing that
[01:44:13] i would like to introduce you i know
[01:44:16] that you want to remain
[01:44:17] fabulous you want to remain i'm going to
[01:44:20] say you're going to remain switzerland
[01:44:21] in this but i have a lot of options i
[01:44:23] have a company
[01:44:24] in the room xm tech i believe exam tech
[01:44:27] tobias muller is in the room right
[01:44:29] action tech is our key company in active
[01:44:31] alignment machines
[01:44:32] tobias thank you very much for being
[01:44:34] with us today uh how could you
[01:44:37] add to this point and what could you do
[01:44:38] for a company like granovuz
[01:44:40] yeah thanks a lot so um i'm i must say
[01:44:44] um it's it's great that you are dealing
[01:44:46] just with these topics of
[01:44:48] um finding a way to really make it
[01:44:51] um deployable in a commercial way and
[01:44:54] this is also where we spend a lot of
[01:44:56] um thoughts and efforts on to put our
[01:44:59] business model
[01:44:59] in position um to really add up to that
[01:45:03] and say um for us today there still is a
[01:45:06] large uncertainty
[01:45:07] uncertainty which technologies and which
[01:45:10] um
[01:45:10] yeah product architectures will be
[01:45:14] really the answer to to solving this um
[01:45:16] scalability um
[01:45:18] issue which is which is obviously still
[01:45:20] there and
[01:45:21] um yeah let's say as a company we are
[01:45:24] automation and packaging um uh company
[01:45:28] um we we go for the approach to um use
[01:45:32] our very flexible platform
[01:45:34] um to first do a demo of um
[01:45:37] yeah new architectures bring in our
[01:45:39] expertise and also the expertise of our
[01:45:42] well our r d partners um to to uh
[01:45:46] go for the um yeah for the new um new
[01:45:49] approach
[01:45:49] uh do first a demo and then transfer it
[01:45:52] into a scalable production
[01:45:54] um system based on our yeah our open
[01:45:57] access
[01:45:58] assembly platform and um well in the
[01:46:00] past uh
[01:46:01] three years um uh or four years actually
[01:46:04] um this was was quite successful for us
[01:46:06] so i'm not sure jose if you've seen it
[01:46:09] yesterday we've been publishing um
[01:46:12] online that we are on place number 85 of
[01:46:15] the thousand
[01:46:16] strongest growing companies in europe um
[01:46:19] from financial times
[01:46:20] so we're pretty proud of that
[01:46:23] and um yeah was big big news for us we
[01:46:27] knew we were in the best thousands but
[01:46:28] that we are on place
[01:46:30] 85 is a huge thing actually except for
[01:46:33] the photonics industry you made a
[01:46:34] difference on the on the active
[01:46:36] alignment of the packaging and
[01:46:38] reconfigurability reconfigurability
[01:46:40] project when you see the different
[01:46:42] options that there are for the passive
[01:46:44] and active alignment of the fiber and
[01:46:46] your demands for high yield and low cost
[01:46:48] where do you stand in this debate i'm
[01:46:50] sorry to ask you this directly no
[01:46:52] yeah no that's quite fair jose thank you
[01:46:55] so
[01:46:55] so you see i think it's very important
[01:46:58] that uh
[01:46:59] we work directly with the fab see that
[01:47:03] that's i would say
[01:47:04] the most important thing because you
[01:47:05] know as you as i think
[01:47:07] the thomas at the beginning mentioned
[01:47:10] you know the v-groove is really tied to
[01:47:14] obviously the fabrication process so
[01:47:17] the fab you know we it's very important
[01:47:20] for that
[01:47:21] v-groove structure and then obviously
[01:47:25] you have the
[01:47:26] fiber ribbons and all the alignment
[01:47:29] that goes with it to make the product in
[01:47:32] a cost-effective way
[01:47:33] so that's the reason i say you know it's
[01:47:35] very important that we work as a
[01:47:36] ecosystem
[01:47:38] and direct fab engagement is one of the
[01:47:41] critical items
[01:47:42] and i'm glad to you know uh
[01:47:45] sync up after the this call and talk
[01:47:47] about that and maybe through an
[01:47:49] introduction to our fan
[01:47:50] there is lots of room for cooperation
[01:47:52] here introduction to the fab
[01:47:54] uh for any technology that can help with
[01:47:56] these needs from hoja they are very
[01:47:58] welcome for operations
[01:47:59] thank you very much for this and thank
[01:48:01] you for keeping quantum dot at the
[01:48:02] forefront of solving the
[01:48:04] multi-wavelength problem thank you so
[01:48:05] much for that
[01:48:06] from a personal point of view i want to
[01:48:08] continue with the program the next
[01:48:09] presentation is the one that we got most
[01:48:11] feedback about most questions about this
[01:48:13] you have youtube fans
[01:48:14] broadcom here bivek
[01:48:18] director frandi of broadcom thank you
[01:48:20] very much for being with us
[01:48:22] many people are asking about when is
[01:48:23] frozen gonna start the floor and the
[01:48:25] attention of everyone
[01:48:26] goes to uv back thank you jose
[01:48:30] thank you and let me know and you can
[01:48:33] see my screen
[01:48:33] i see it gorilla glass clear excellent
[01:48:37] excellent thank you
[01:48:42] all right uh i think uh
[01:48:45] everyone on the panel have talked about
[01:48:50] the need for co-package the need for
[01:48:52] advancement
[01:48:53] of the optical transceivers uh i viewed
[01:48:57] this
[01:48:58] and we at broadcom view this as the
[01:49:01] last mile effort that is needed to
[01:49:05] transition the optical transceivers for
[01:49:07] a true co-packaged
[01:49:09] solution and
[01:49:13] anything that we need to transition to
[01:49:15] co-package
[01:49:17] this requires semiconductor like
[01:49:19] scalability
[01:49:20] so most of the focus is going to be
[01:49:24] in terms of how the ecosystem can help
[01:49:28] us transition the optics
[01:49:31] from where we are today into a
[01:49:34] semiconductor like manufacturing
[01:49:37] ecosystem now this here we have an
[01:49:41] opportunity with the various use cases
[01:49:44] that are being projected for co-packaged
[01:49:47] optics
[01:49:48] i think what we really need is a step
[01:49:51] function
[01:49:51] improvement in the optical connectivity
[01:49:54] and anything that we are developing has
[01:49:57] to be thought about
[01:49:58] at scale the optical the interconnect
[01:50:01] connectivity has to be at at scale and
[01:50:05] we need the ecosystem to invest
[01:50:08] in these areas where we can create this
[01:50:11] scalability this could be in areas of
[01:50:15] fiber attach this could be in the
[01:50:17] foundry this could be in the packaging
[01:50:19] but we have to start thinking about
[01:50:21] scale
[01:50:22] and that's when we can reap the true
[01:50:24] benefits of
[01:50:26] co-packaged optics now with that
[01:50:30] as i mentioned co-package optics is not
[01:50:33] just
[01:50:33] about solving the problem of integrating
[01:50:37] optics to the switch
[01:50:38] but it is truly about how we can enable
[01:50:41] this
[01:50:41] uh technology for what we have
[01:50:44] in the next decade in terms of
[01:50:48] uh creating these uh this solution and
[01:50:51] it can be applicable to a variety of
[01:50:54] areas that includes you know
[01:50:57] taking the advantage of the xpu's the
[01:50:59] gpus the tpus
[01:51:01] or into the ai ml space
[01:51:05] or into the ethernet switching so the uh
[01:51:09] the opportunities are tremendous for
[01:51:11] this co-packaged optics and i think
[01:51:13] uh mark ruby all of them they
[01:51:16] they explained this in terms of the use
[01:51:18] case that's needed in the data centers
[01:51:22] so just moving forward in terms of what
[01:51:25] broadcom
[01:51:26] is doing as you all know we do have the
[01:51:29] leadership
[01:51:30] in terms of the core switch technology
[01:51:33] the
[01:51:34] service the dsp that comes with it
[01:51:37] and have created the sustained
[01:51:39] generational
[01:51:40] differentiation along with that
[01:51:44] what we are focusing on is in also
[01:51:47] developing the other pillars
[01:51:49] we see that each of these pillars are
[01:51:51] going to be very very
[01:51:52] key for the uh success of this overall
[01:51:55] co-package solution
[01:51:57] for all the applications where there is
[01:51:59] a need for
[01:52:01] optics scaling be it in the areas of
[01:52:04] mixed signal ic where we need to create
[01:52:06] power and performance uh you know in in
[01:52:10] technology agnostic be it be it in the
[01:52:12] by cmos technologies
[01:52:13] uh like the silicon germanium or in the
[01:52:16] cmos technologies it has to be power
[01:52:18] efficient it has to
[01:52:19] uh be interoperable etc
[01:52:23] and when it comes to the silicon
[01:52:25] photonics platform
[01:52:26] here we have to focus on creating a fab
[01:52:29] ecosystem where you are able to
[01:52:32] fabricate these highest density pics
[01:52:35] you know modulators the photodiodes all
[01:52:37] of this in silicon
[01:52:39] and create a low loss soi
[01:52:42] based waveguide platform so so
[01:52:45] and the very very important pillar is
[01:52:48] in terms of the advanced packaging and
[01:52:50] testing that is going to be needed
[01:52:52] because all of us see that uh you know
[01:52:56] for for optics to kind of get into that
[01:52:58] step function
[01:52:59] and being able to uh scale uh
[01:53:02] like semiconductor there has to be a lot
[01:53:05] of innovation and development that has
[01:53:07] to happen
[01:53:08] uh be it at wafer level test or the
[01:53:10] final tests
[01:53:11] be it in terms of the advanced packaging
[01:53:12] so that it can
[01:53:14] enable more heterogeneous packaging
[01:53:17] be it with technologies like tsv or 2.5
[01:53:20] d
[01:53:21] and 3d integration that that provides
[01:53:23] the density that is needed
[01:53:26] or even in terms of the uh the
[01:53:29] uh the light sources that are needed the
[01:53:32] lasers
[01:53:33] so broadcom already has a leadership in
[01:53:35] this with
[01:53:36] where we are uh manufacturing over 50
[01:53:41] million lasers per year from our
[01:53:43] internal
[01:53:43] laser fabs in addition to that we are
[01:53:46] also focusing
[01:53:47] on the high volume optical manufacturing
[01:53:50] as well as creating these
[01:53:52] very very high power multi wavelength
[01:53:54] sources that are going to be needed
[01:53:57] when we go down this path of remote
[01:53:59] laser source so
[01:54:01] if you see each of this pillar is
[01:54:03] important and
[01:54:05] to provide the industry leading
[01:54:07] economics that are going to be needed
[01:54:10] to create the volumes that are going to
[01:54:12] be needed
[01:54:13] and the power efficiencies that are
[01:54:15] needed at this sort of a scale
[01:54:19] and for this forum we are here on an
[01:54:23] epic journey through the
[01:54:24] next decade and and for us
[01:54:28] we have to benchmark ourselves against
[01:54:31] you know the very well established
[01:54:33] semiconductor industry so we have to
[01:54:35] really look up to how the
[01:54:37] semiconductor ics are manufactured today
[01:54:40] how the manufacturing ecosystem uh
[01:54:43] works in for the semiconductors and
[01:54:46] that is where we have the opportunity to
[01:54:49] innovate
[01:54:50] in transitioning the optics to
[01:54:53] semiconductor like
[01:54:55] manufacturing so with that
[01:54:59] where are the challenges that we see
[01:55:02] this is in terms of the density
[01:55:04] uh the volume and the economies of scale
[01:55:08] and and where where where i see uh the
[01:55:11] critical needs are is in terms of the
[01:55:14] innovation that's going to be needed in
[01:55:16] terms of the
[01:55:17] packaging where the osats are
[01:55:20] comfortable
[01:55:20] in handling optical devices through
[01:55:23] their manufacturing flows
[01:55:25] going through all the challenges of
[01:55:27] handling optics
[01:55:28] knowing how to handle optics how to you
[01:55:32] know
[01:55:32] get the light in and out etc coming up
[01:55:35] with
[01:55:36] very very important which is a very high
[01:55:38] density optical attached solution
[01:55:40] because if we are talking about high
[01:55:42] density multiple channels we need a very
[01:55:44] very very reliable solution there
[01:55:46] so coming up with the high density
[01:55:48] optical attach is going to be
[01:55:50] the key and uh coming up with automated
[01:55:53] optical opto electrical uh system
[01:55:55] testing this has to be at scale
[01:55:57] it cannot be you know testing single
[01:55:59] devices but it has to be
[01:56:01] where we are able to test multiple
[01:56:03] devices multiple channels at time
[01:56:06] very economically it cannot be an
[01:56:08] expensive solution
[01:56:10] so so all of this put together will help
[01:56:13] us
[01:56:14] establish a semiconductor like scale
[01:56:18] for the manufacturing and i think that
[01:56:21] is
[01:56:23] the biggest problem statement that we
[01:56:26] need to address
[01:56:27] and this is where most of the innovation
[01:56:29] uh needs to happen
[01:56:31] uh in the in the next few years along
[01:56:33] with the
[01:56:34] ecosystem partners that we have on this
[01:56:37] call
[01:56:39] yeah with that i come to the end of my
[01:56:42] presentation
[01:56:43] and open for any questions
[01:56:50] sorry i watch me thank you very much
[01:56:52] vivek super interesting
[01:56:54] presentation so then you in in your last
[01:56:58] slide you already said about what uh do
[01:57:00] you think are the challenges right
[01:57:02] for the future of copaca
[01:57:05] would you like to highlight maybe the
[01:57:07] one in which you
[01:57:09] are more interested in i mean the one
[01:57:12] that you think that is
[01:57:14] a that is the most difficult one then
[01:57:17] more challenging no
[01:57:19] i think i think i think i would like to
[01:57:23] invite all the osats the odms
[01:57:26] the equipment manufacturers to solve
[01:57:30] this problem of scalability i think what
[01:57:33] we have
[01:57:34] is a manufacturing problem and
[01:57:37] manufacturing at scale uh optics
[01:57:41] uh requires that level of a step
[01:57:43] function uh transition
[01:57:45] to manufacture at scale and that is
[01:57:47] going to be the most
[01:57:48] important problem for all of us to solve
[01:57:51] to
[01:57:52] really get into the economics of scale
[01:57:56] okay so you mean in the in the package
[01:57:58] uh
[01:57:59] the way of scalability you know the how
[01:58:02] to
[01:58:03] i mean probably your challenges for our
[01:58:06] companies doing automation tools right
[01:58:08] yes yes electronics packaging optical
[01:58:11] packaging as well as the test
[01:58:12] yes as well do you have any
[01:58:17] any thought about the volumes about the
[01:58:19] because we
[01:58:20] keep we have this automation companies
[01:58:23] asking
[01:58:24] uh what are going to be the volumes how
[01:58:26] fast
[01:58:27] they need to do the fiber alignment how
[01:58:29] fast they need to do
[01:58:31] their processes do you have any any idea
[01:58:34] of what are going to be these volumes
[01:58:37] um not specifically about the volumes
[01:58:40] but
[01:58:41] i think we saw the use cases that are
[01:58:44] coming for the co-packaged optics from
[01:58:48] compute and the
[01:58:52] the other networking areas i think all
[01:58:55] of this
[01:58:56] are an opportunity for tremendous volume
[01:58:59] and they speak to the volumes that are
[01:59:02] needed which are which are pretty much
[01:59:04] like
[01:59:05] any other semiconductor asics or chips
[01:59:09] in volume today so they need to be at
[01:59:11] that same
[01:59:12] scale okay we have a question from jan
[01:59:17] from bmvg consult hey john would you
[01:59:20] like to make the question by yourself
[01:59:22] uh yes please uh vivek the question that
[01:59:24] i had was whether you need the
[01:59:27] testing the opto electrical testing on
[01:59:29] wafer level on package level so why
[01:59:31] where do you think that
[01:59:33] is more needed uh
[01:59:36] i would i would say both i would say
[01:59:38] both
[01:59:40] okay thanks
[01:59:44] okay thank you very much and we have
[01:59:46] another question from john
[01:59:48] from you is ucsb
[01:59:51] vivek um your last slide showed 3d uh
[01:59:55] packaging is that preferable or the only
[01:59:58] approach in your mind
[01:59:59] as opposed to integrating electronics
[02:00:00] photonics on the same wafer
[02:00:04] um i think
[02:00:07] packaging today has evolved in the
[02:00:10] electronics industry
[02:00:12] where we are able to take advantage of
[02:00:14] the 3d heterogeneous packaging
[02:00:17] so staying off monolithic gives us the
[02:00:21] opportunity for the electronics and
[02:00:23] photonics to scale differently
[02:00:26] they don't need to scale together so
[02:00:28] there is a
[02:00:29] inherent uh benefit that we get
[02:00:32] but as we said what provides the best
[02:00:35] density is going to be the winner here
[02:00:40] related to that at some point i don't
[02:00:42] know where you know
[02:00:43] a hundred terabits perhaps are you close
[02:00:45] to using an entire reticle
[02:00:47] in which case you don't have room for
[02:00:48] any optics and you have to do 3d
[02:00:50] how far away are you from that reticle
[02:00:52] limit
[02:00:54] uh yeah you mean specific to the uh
[02:00:58] optics or the electronics like at a
[02:01:01] at 51 terabits how much of a reticle
[02:01:03] does that use up do you have room to put
[02:01:06] photonics monolithically in there or
[02:01:07] does it have to be 3d
[02:01:10] yes uh i cannot specifically answer
[02:01:12] towards the
[02:01:13] future roadmap of where we are but
[02:01:16] i think from an industry why we are
[02:01:19] hitting the reticle
[02:01:20] limits of for for the
[02:01:24] generation of the asics that are being
[02:01:27] manufactured today
[02:01:28] so there is a need to grow vertically
[02:01:32] or to go 3d
[02:01:36] thanks okay thank you very much for this
[02:01:39] discussion and answer for your
[02:01:40] presentation vivek well um for the
[02:01:42] companies attending the meeting you saw
[02:01:44] a broadcom has many challenges
[02:01:47] uh please let us know if you want to
[02:01:49] contact
[02:01:50] with them especially the the the company
[02:01:52] is doing packaging in the room okay
[02:01:54] so let's move to our last presentation
[02:01:57] to our last talk for today
[02:01:59] so last but not the least benjamin from
[02:02:02] adba
[02:02:03] so benjamin if you want to share your
[02:02:05] screen
[02:02:09] yes
[02:02:13] right so thank you very much so i would
[02:02:16] like to talk a slightly different
[02:02:18] application here
[02:02:19] [Music]
[02:02:21] sorry no no perfect
[02:02:25] okay so i would like to talk about a
[02:02:27] slightly different application here than
[02:02:28] switching and that is
[02:02:29] inter data center interconnect because i
[02:02:33] have the view that
[02:02:34] the technical challenges at least that
[02:02:35] we see there are very much similar to
[02:02:37] what we see in the switch chain albeit
[02:02:40] to a
[02:02:40] much less extreme extent so
[02:02:44] a few question words about advance
[02:02:47] so for those of you who don't know advan
[02:02:49] provides systems for
[02:02:50] all kinds of optical communication
[02:02:52] networks generally we
[02:02:54] divide that into two categories on the
[02:02:56] left hand side we have the edge networks
[02:02:58] that's access and mobile front hole and
[02:03:00] backhoe and so on
[02:03:01] and in the center and on the right hand
[02:03:03] side we have the core and metro networks
[02:03:05] these are a bit more complex so they are
[02:03:07] mostly based on dwdm and coherent
[02:03:09] transmission
[02:03:10] and that also includes the data center
[02:03:11] interconnects on which i want to focus
[02:03:13] here
[02:03:14] so if we zoom in on a data center a
[02:03:16] typical network looks like this
[02:03:18] on the bottom we have the servers which
[02:03:20] are which are served by a
[02:03:22] top of racks or tier zero switch and
[02:03:24] then over multiple tiers with
[02:03:26] increased complexity and increased
[02:03:28] transmission reach and so on
[02:03:31] we reach the edge of the data center
[02:03:32] where we have these coherent
[02:03:33] transmission modules
[02:03:34] and there we connect to long-haul
[02:03:36] networks or metro networks
[02:03:38] and if you look from a very high level
[02:03:39] then you see the challenges that are
[02:03:41] faced here are very much the same on all
[02:03:43] levels
[02:03:44] so we need of course to increase the
[02:03:45] data rate per channel that
[02:03:47] demands that we increase the interface
[02:03:50] speed between
[02:03:50] silicon chips and optics we want to
[02:03:53] decrease the
[02:03:54] power consumption the hope here with co
[02:03:56] package optics is that due to decreased
[02:03:58] transmission distances electrical
[02:04:00] transmission decision
[02:04:01] where we can get that down we also want
[02:04:04] to increase face plate density
[02:04:05] just to achieve higher bandwidth per
[02:04:07] rack unit so that way
[02:04:09] we have to fit more ports on one rack
[02:04:11] unit or
[02:04:12] and or we have to increase data rate
[02:04:15] and of course with everything we do we
[02:04:17] want to decrease costs as well
[02:04:20] so this is all very nice to have and
[02:04:21] then the question is when do we see it
[02:04:24] so certainly we don't have it today so
[02:04:26] not at the current generation 25 terabit
[02:04:29] for the next generation at least we
[02:04:31] already have some standardization effort
[02:04:32] that we already mentioned of
[02:04:34] cobo and so on but
[02:04:37] if that's a mass deployment will still
[02:04:39] remain to be seen
[02:04:41] so do we will we see it then at the 100
[02:04:43] terabit generation
[02:04:45] um not sure um could be let's find out
[02:04:48] but what i want to say here is that we
[02:04:51] already see steps in this direction at a
[02:04:53] different application level and that is
[02:04:54] the
[02:04:55] intel dci level if you for example look
[02:04:58] at the oif there's another
[02:05:00] standard it's called the ictroser it's
[02:05:02] short for integrated coherent transmit
[02:05:04] receive optical sub assembly
[02:05:06] as i said it's aimed at the coherent
[02:05:08] market the current generation is 400 gig
[02:05:10] that's 64 gigabot
[02:05:12] 16 quant transmission dual polarization
[02:05:14] and this particular package is solar
[02:05:16] reflow compatible so if you look at the
[02:05:18] back side you would see a bga
[02:05:20] so this can be directly soldered onto a
[02:05:23] host pcb
[02:05:24] this includes almost everything so dual
[02:05:27] pole iq modulator coherent receiver
[02:05:29] driver tia and it has a control unit so
[02:05:32] it can be
[02:05:33] addressed digitally from the outside
[02:05:35] world
[02:05:36] and is a very compact you can see it on
[02:05:38] the right side in the picture compared
[02:05:40] to your one euro coin
[02:05:42] it's only 22 by 50 millimeters and 3.5
[02:05:44] millimeter
[02:05:45] high it's a non-hematite package because
[02:05:47] this particular instance it's aimed at
[02:05:49] the silicon photonics market
[02:05:51] um it does not include a laser however
[02:05:53] there's a second type of this
[02:05:55] package defined and that is a standard
[02:05:58] gold box more aimed at the inner
[02:06:00] phosphate guys
[02:06:01] and this one does include a laser as i
[02:06:03] said it's standardized in the oif the
[02:06:06] implementation agreement is also termed
[02:06:08] ic choser so if you are interested you
[02:06:10] can look up more information in there
[02:06:13] and so but how does it compare to the
[02:06:15] co-package optics initiative
[02:06:17] so even more so than in the coherent
[02:06:19] space and you need a
[02:06:21] more compact form factor so in the oif
[02:06:23] they are talking about the
[02:06:25] 51 terabit generation
[02:06:28] and there they want to define or
[02:06:31] assemble a sub board which contains
[02:06:34] eight
[02:06:34] of the 400 gig transmission modules and
[02:06:38] to address
[02:06:38] 51 terabits you need 16 of those so you
[02:06:41] really need to fit a lot of
[02:06:42] modules next to the switch asic then of
[02:06:45] course these modules
[02:06:46] each should be exhaustive meaning it
[02:06:48] should include everything
[02:06:49] this also includes then dsp and of
[02:06:51] course rf electronics modulator receiver
[02:06:53] again a control unit laser is a big
[02:06:56] question we already heard a lot of talk
[02:06:58] about that today
[02:06:59] so the talk here also in rf is do we go
[02:07:02] for an external multi-wavelength laser
[02:07:04] source
[02:07:04] with that amount of scaling i think it
[02:07:06] makes sense right you reserve a rack
[02:07:08] unit just for your laser source you can
[02:07:09] build in sun redundancy
[02:07:11] and then just fiber couple
[02:07:14] your lasers to the silicon engine and
[02:07:17] also you need a lot of
[02:07:18] rf bandwidth so the natural choices here
[02:07:20] are either bga or lga
[02:07:22] bga offers slightly higher bandwidth but
[02:07:25] you would have a problem replacing a
[02:07:27] broken module
[02:07:28] so the go-to solution would be lga but
[02:07:31] here
[02:07:31] this eats somewhat into footprint
[02:07:33] because you need a clamping mechanism to
[02:07:34] hold the molecule down
[02:07:36] both in common is that you have a very
[02:07:38] easy way of cooling
[02:07:40] assuming you flip chip everything on a
[02:07:41] carrier ceramic or organic
[02:07:43] and solder that onto your hose board
[02:07:46] and then you just place the heatsink on
[02:07:48] top and that's
[02:07:50] very low thermal barrier so it can't get
[02:07:52] any better than that with
[02:07:53] this kind of assembly and last but not
[02:07:56] least i already mentioned low cost
[02:07:58] and number always cycles is 80 of a
[02:08:00] module because it's associated with
[02:08:01] packaging
[02:08:03] plus minus so and most of that is then
[02:08:07] associated with the optical packaging so
[02:08:09] taking the laser out here is
[02:08:10] again a step in the right direction i
[02:08:12] think that
[02:08:13] leaves only the fiber chip alignment and
[02:08:15] here you probably also want to reduce as
[02:08:17] much as possible the complexity
[02:08:19] and maybe go for a single alignment step
[02:08:22] of multiple fibers
[02:08:24] so these representations on the right
[02:08:26] hand side are just examples so
[02:08:28] for the exact implementation you can get
[02:08:30] more creative
[02:08:31] so you can think about 3d integration we
[02:08:34] just saw that or heard that
[02:08:35] you can think about monolithic
[02:08:36] integration nowadays as possible like
[02:08:38] sequoia is doing
[02:08:39] putting the rf electronics on the
[02:08:41] silicon die
[02:08:42] or moving the control electronics and
[02:08:44] the dsp or phi
[02:08:47] so there aren't the solutions here in
[02:08:48] any combination
[02:08:51] so one more word about the laser so this
[02:08:53] is not so
[02:08:54] much a topic then for the switch guys
[02:08:56] but very much for the intel data center
[02:08:59] interconnects
[02:09:00] i think generally there are two
[02:09:01] approaches you can integrate it somehow
[02:09:03] or you have an external like the
[02:09:04] modulation laser source
[02:09:06] for the integrated way um i think the
[02:09:08] most elegant solution would be
[02:09:10] monolithic integration
[02:09:11] like the group of john bowers has shown
[02:09:13] and we see at inter
[02:09:15] so here you place directly some indium
[02:09:17] phosphide
[02:09:18] dye on top of your waveguide but this
[02:09:20] would
[02:09:21] require that your foundry would allow
[02:09:23] first of all indium phosphide in your
[02:09:25] in their clean room and then have the
[02:09:27] processes in place to process it
[02:09:29] accordingly then you can shift the
[02:09:31] burden more towards the packaging guys
[02:09:32] and
[02:09:33] just put the laser dye on top or
[02:09:37] surface couple or edge couplet and then
[02:09:39] there's an immediate step where you have
[02:09:40] an edge cavity
[02:09:42] basically you have a hole in your
[02:09:43] silicon chip and put a reflective soa or
[02:09:46] whatever in there and build your laser
[02:09:48] from that but that again would
[02:09:49] demand some active optical alignment is
[02:09:51] fairly complex
[02:09:53] with the external laser source this
[02:09:55] makes more sense if you have a lot of
[02:09:57] modules to be addressed so
[02:09:58] again here you need to really scale your
[02:10:01] your network to to make this viable
[02:10:05] all right and that's it um thank you for
[02:10:07] your attention i'm happy to take any
[02:10:09] questions
[02:10:10] thank you very much for this fantastic
[02:10:12] presentation and also i would like to
[02:10:13] tell you
[02:10:14] vivek and everyone that actually
[02:10:17] tomorrow we are releasing our market
[02:10:19] report on photonics for 5g
[02:10:21] and i think you are mentioned almost in
[02:10:23] every other slide so thank you very much
[02:10:25] for
[02:10:25] for this a advice a key member of epic
[02:10:28] you know the drill you know the question
[02:10:30] what can you do for them
[02:10:32] what can they do for you
[02:10:35] so i think most of it was already
[02:10:37] mentioned in principle i think a lot of
[02:10:39] all the things are already there one
[02:10:41] thing i think is slightly missing is the
[02:10:43] monolithic integration of layers i
[02:10:45] already mentioned for 1550 nanometer
[02:10:48] but apart from that i think everything's
[02:10:50] there you can always improve slightly in
[02:10:52] many things but you just need to get
[02:10:55] people to talk with each other so
[02:10:56] it's very good what you are doing here
[02:10:58] right where we can talk with each other
[02:11:00] that's really the most important
[02:11:01] takeaway of this meeting that we can all
[02:11:03] talk for each other and i think
[02:11:05] right now we have to say oh just a few
[02:11:07] more words before we send everybody away
[02:11:09] the first one is that for me that was a
[02:11:11] really really truly epic meeting i
[02:11:13] i rarely had the chance to have in the
[02:11:14] room intel microsoft facebook it was
[02:11:17] everybody commenting and trying to help
[02:11:19] each other
[02:11:19] and what's the most important for me
[02:11:21] there are four takeaways costs really
[02:11:22] cost driving the debate of active versus
[02:11:24] passive alignment
[02:11:25] it's an open debate we saw two different
[02:11:28] approaches
[02:11:29] from two different machines that are
[02:11:30] present in the room as well i don't want
[02:11:32] to cover more than that but i think you
[02:11:33] all know what i'm talking about
[02:11:35] we also talk a lot about reliability and
[02:11:37] test i love the comments from john
[02:11:38] bowers on this
[02:11:39] still is a limiting factor i love also
[02:11:41] the comments from ruby from facebook
[02:11:42] about this
[02:11:43] if we talk about the yield or the
[02:11:45] multi-wavelength laser required to
[02:11:47] what is what level of testing is
[02:11:49] required a package and a wafer level
[02:11:51] still is an open debate we do have the
[02:11:53] different meetings to address that one
[02:11:55] and also for me also the semiconductor
[02:11:58] scale manufacturing and also a change of
[02:12:00] mindset
[02:12:00] trying to bring more semiconductor
[02:12:02] thought to this actually on this topic i
[02:12:04] would like to say that today
[02:12:06] my ceo carlos lee talked to salah nasri
[02:12:09] the managing director of international
[02:12:10] semiconductor executive summits
[02:12:12] and they see great potential for a
[02:12:13] collaboration a joint action between
[02:12:15] epic
[02:12:16] and semi to bring more and more
[02:12:19] some broadcasts and sd microelectronics
[02:12:21] to this debate because i think they
[02:12:22] should drive it
[02:12:23] also as a honestly in the beginning i
[02:12:26] would like to tell everyone
[02:12:27] that we have now an open call an open
[02:12:30] call
[02:12:30] for demonstrators on indian foster
[02:12:33] photonic internet
[02:12:34] circuits 50 50 percent funding
[02:12:37] in kind contribution fund is a huge huge
[02:12:40] opportunity for any company
[02:12:42] that needs to have indium fossil
[02:12:44] photonic integrity circuit
[02:12:46] chips very fast and around very reliable
[02:12:48] from jpegs thank you very much for this
[02:12:50] they actually asked me to say this and
[02:12:52] of course
[02:12:52] i'm more than happy to really bring this
[02:12:54] huge opportunity not just for the epic
[02:12:56] members but for any company
[02:12:58] in our ecosystem and for all of you for
[02:13:00] roger i have to say that even though you
[02:13:02] think
[02:13:02] that the meeting is over the meeting is
[02:13:04] not over the meeting starts
[02:13:06] now it is all about the follow-up so
[02:13:08] anna and me we did our best we do our
[02:13:09] best to make sure that we find some
[02:13:11] business opportunities between all of
[02:13:12] you
[02:13:13] but now what we really want we really
[02:13:14] want is that you
[02:13:16] get in touch with each other and do
[02:13:18] business if you want to get in touch
[02:13:19] with any of the participants today send
[02:13:20] me an email hosted.com
[02:13:23] and i will make the introduction
[02:13:25] everyone that was in the room today has
[02:13:27] the epic mindset which is
[02:13:29] we need to work together and let's do
[02:13:31] that here on behalf of a fantastic thing
[02:13:33] on medical expenditure photonics per
[02:13:35] laser expert optics and quantum expert
[02:13:37] an innovation manager who's totally
[02:13:38] crazy with a great guy antonio raspa
[02:13:40] marketing and event managers my ceo all
[02:13:43] of you know carlos lee
[02:13:45] we love you we continue willing to
[02:13:48] travel again so we're at mask
[02:13:49] and wash your hands and let's see each
[02:13:52] other very soon
[02:13:53] let's see each other very soon the virus
[02:13:54] is almost over we need to travel and see
[02:13:56] each other until
[02:13:57] next time take care bye