# Live recording Optica Online Industry Meeting on Linear Drive Optics. April 30th 2024.

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

[00:08] [Music]
[00:23] we have been overwhelmed by your
[00:25] response many people have posed the
[00:27] Optica question direct to my team I they
[00:30] know what we do for them but what can
[00:33] active leaders in Optics and photonics
[00:35] do for us the answer is simple by
[00:38] joining Optica as a corporate member you
[00:41] encourage us to expand our trusted
[00:44] Global Network and support or new range
[00:46] of services get deep insights into
[00:49] what's next for your sector in photonics
[00:51] with free access to over 90 market
[00:54] reports plus exclusive access to the
[00:57] PowerPoint slides for the inperson
[01:00] meetings and corporate members get
[01:02] exclusive access to all Optica industry
[01:05] meetings in
[01:07] 2024 including if you're an investor in
[01:10] photonics you'll recognize where I am
[01:11] now hello from frankurt mes which will
[01:14] be the venue for the Eco conference
[01:18] 2024 the third PEC Summit in Silicon
[01:22] Valley the Optica laser Congress inaka
[01:27] Japan plus the Optica quantum industry
[01:30] Summit this time in Bristol UK and
[01:34] remember being a corporate member of
[01:36] optica also gives you discounted access
[01:38] to many partner events such as Peak
[01:41] Summit Europe driving Vision News and
[01:44] Peak International
[02:05] [Music]
[02:46] hello from Disneyland 2024 has been very
[02:50] hectic so far trips nonstop I decided to
[02:52] take a couple of days off and come to
[02:54] Disneyland with my daughter but I can't
[02:56] stop thinking about next Tuesday April
[02:59] 30th where the online industry meeting
[03:01] is on linear Drive Optics you may
[03:04] remember ofc 2024 where linear Drive
[03:08] Optics revolutionized the entire
[03:10] exhibition the Energy savings were a
[03:12] major need and the removal of dsps was a
[03:15] major cost savings that's why antios
[03:19] time was creating a really huge
[03:21] impression last year this year as well
[03:24] and atfc we discuss this directly with
[03:26] him here is the clip lpo is everywhere
[03:29] at this
[03:30] and it's actually surprising to me
[03:31] because I didn't know how many people
[03:33] are doing it like even companies i'
[03:34] never heard of before so um it is true
[03:37] that the lowest power pluggable module
[03:40] is an lpo module you get rid of the DSP
[03:43] you subsid it with a simple linear
[03:45] driver and a TIA so that's the good it's
[03:48] half the power of a full DSP there's
[03:50] another technology that also is getting
[03:52] mentioned a lot called LR linear receive
[03:55] rtime transmit and the power of that is
[03:58] somewhere halfway between the full lpo
[04:00] and the full DSP but still much better
[04:02] than the full DSP and one thing really
[04:05] good about pluggable modules is the
[04:08] system can adopt any kind of modu you
[04:10] can plug in the DSP module the lrm modu
[04:13] or the lpo modu so the system support
[04:16] all of this and depending on the
[04:18] customer and their qualification efforts
[04:21] or the the you know what they want to
[04:23] optimize for they can choose the
[04:25] appropriate uh technology and with first
[04:28] demonstrators here at ofc what does the
[04:31] community still need to do together yeah
[04:33] that that's the thing so we started this
[04:36] lpo MSA to come up with the the revised
[04:41] the relaxed tp2 specification that
[04:43] multiple venders can meet and there will
[04:45] be fully interoperable so today you know
[04:48] the the I spec is too Trent and uh it
[04:51] can be relaxed but that work has to be
[04:53] agreed on by all the vendors and that
[04:55] work just started but we hope to uh have
[04:58] results by eco it was ofc is the place
[05:01] where things happen and this is the year
[05:03] in which the First demonstrators on lpu
[05:04] are being seen and touched by everyone
[05:07] what is your Preferred Future in the
[05:09] coming o well to me the most surprising
[05:11] thing here was actually the first 1.60
[05:13] lpo module so there's not even system
[05:16] yet that can do 600 gig plug but with a
[05:19] tester you can you know generate the the
[05:21] waveforms I'm sorry the the signals and
[05:24] the bidder rate is actually pretty good
[05:26] for first attempt so I'm at this point
[05:28] I'm confident that there setion geek lpo
[05:31] will work as well that's why we brought
[05:33] Andy again we going to have a discussion
[05:35] with him and many others on the Optica
[05:37] Network about what is the future of
[05:40] linear Drive Optics and what will be
[05:42] your role please make sure your voice is
[05:44] heing this meeting register online in
[05:46] this link and see you on Tuesday April
[05:49] 30th all around the world on in Drive
[05:52] Optics the next hour we hour and a half
[05:55] we're going to get very strong
[05:57] discussions opinions and Technology
[06:00] competition and this is by Excel on
[06:03] behalf of optica this is Jose poo I am
[06:05] the CTO and I am really happy to be here
[06:08] today with all of you I would like to
[06:09] start by telling you that this is not a
[06:12] two three people job Optica is now
[06:14] 24,000 individual members 600 Corporate
[06:17] members and we are here to connect you
[06:20] our job is to connect companies in our
[06:23] industry and we provide as I said before
[06:26] key market reports and connection
[06:28] between our members we organize the
[06:30] events only to connect our members the
[06:31] result of today's meeting has to be for
[06:33] all of you to get in touch with
[06:35] potential Partners suppliers and
[06:36] customers I speak on behalf of the
[06:38] corporate engagement Council these are
[06:40] who tell us what to do and which
[06:41] direction to move and I would like to
[06:42] always thank them for their great great
[06:45] approach I would like to say that this
[06:46] is the third meeting of theine optica
[06:49] online industry series write down your
[06:51] calendars 21st of May Quantum sensing
[06:56] for many of you who are active in the
[06:57] field of photonics interative photonics
[06:59] in the room write down that date 21st of
[07:03] May but today today is all is all about
[07:05] linear Drive Optics and this meeting is
[07:08] being attended by about 750 people in
[07:12] YouTube and in Zoom right now 750 people
[07:15] any of you none of you paid any money
[07:18] this meeting is possible because of the
[07:20] support of our two sponsors today first
[07:22] I would like to acknowledge opty wave
[07:24] the design suite for photonic
[07:27] integration and Optical networks optic
[07:29] wave thank thank you very much for
[07:30] supporting this event and also I would
[07:32] like to thank aoto you're looking for
[07:34] materials for injection molding for
[07:37] packaging for nonthermal conductivity or
[07:40] for extra thermal connectivity between
[07:42] different materials aoto is your
[07:44] supplier thank you optic wave and aoto
[07:46] for supporting this event this event is
[07:48] our no our linear Drive Optics and this
[07:51] event for most of you has been who has
[07:52] been the first time you attend this
[07:54] event it has two groups of companies
[07:56] companies that can he with vision and
[07:58] with potential challenges for that
[08:00] Vision today those companies are Arista
[08:03] nuis light intelligence and axel link
[08:07] and companies that come to the meeting
[08:08] with solutions to enable that vision and
[08:11] today those companies are kyone
[08:13] photonics nlm photonics inol and light
[08:16] speed photonics also today we invited a
[08:19] market analyst and very dear friend of
[08:21] optica from five reality Mark lukovic
[08:24] these are the people who have been
[08:26] scheduled but this is just a meeting
[08:28] this is not a webinar so all of you at
[08:31] any time you are welcome to participate
[08:34] and to tell us what is your vision and
[08:35] how do you address other people's Vision
[08:38] these are the companies that are Optica
[08:41] corporate member and have registered for
[08:43] this meeting we have make sure that all
[08:45] of you play a role in the supply chain
[08:48] from Big technology to equipment
[08:50] manufacturers and users provider of
[08:52] fiber optics for components these are
[08:55] the Optica corporate members who have
[08:56] registered for this event and they
[08:58] player role in the future of linear
[09:00] Drive Optics if you are surprised that
[09:02] you registered to the meeting and your
[09:03] logo is not here this is because you are
[09:05] not a corporate member of optica please
[09:07] make sure you register as a corporate
[09:09] member ASAP we are building the global
[09:13] industrial Network and it is super
[09:15] important for us to have you to have you
[09:17] on board if you like what we are doing
[09:19] there is no better way to support us
[09:21] than to become an Optica corporate
[09:23] member I would like to also remind
[09:24] everyone that all of you have access to
[09:26] a long list of market reports so please
[09:28] make sure to go to
[09:31] opa.opa orgm marketreports at any time
[09:34] and download all the content there we
[09:37] have purchase market reports from
[09:38] companies like Yol de matis they are all
[09:40] there for you use them they for free
[09:42] because you're a corporate member and I
[09:44] also would like to remind everyone that
[09:45] this meeting is also live streaming
[09:47] YouTube so let me say hello to the
[09:48] YouTubers in the world hello YouTubers
[09:50] thank you very much for connecting to
[09:51] this meeting if you have any question
[09:53] ready in the chat and I will read it in
[09:54] the room and also please make sure that
[09:56] you register for all the upcoming
[09:58] upcoming events and with this I would
[10:00] like to also say that this also valid
[10:02] for the people with me here in the zoom
[10:04] room thank you very much for being here
[10:06] today we need your questions we need to
[10:09] make sure that you are interactive that
[10:11] you are active please write down at any
[10:13] time your question in the chat please
[10:14] say hello in the chat this is a meeting
[10:16] for connecting companies and there is no
[10:18] better way to start than to a person
[10:21] that for some time has been one of the
[10:24] drivers of the industry all the way from
[10:26] his time at some Microsystems all of the
[10:29] then many of you know what sparkk means
[10:32] today with us opening the meeting on
[10:34] linear Drive Optics we have no other
[10:37] than Andy bosin from Arista Andy Andrea
[10:42] thank you very much for opening the
[10:43] meeting the floor and the attention of
[10:45] everyone doing anything on lpo L is
[10:48] yours yes so a lot of progress has been
[10:52] made uh on linear plugable Optics over
[10:54] the last year and a lot of progress is
[10:57] being made uh with also the formation of
[10:59] the uh lpo MSA which was announced at
[11:03] ofc which has the goal to write a a
[11:07] common specification basically a relaxed
[11:10] specification that allows uh linear
[11:13] plugable Optics to interoperate from uh
[11:16] multiple vendors I should also note here
[11:18] that one of my big surprises at ofc was
[11:21] that literally every major Optics modal
[11:25] vendors had demonstrations and
[11:27] prototypes of linear plugable Optics and
[11:30] showed inability at the show floor for
[11:33] the E gig type Optics today so um uh
[11:39] what I wanted to highlight here is um
[11:41] you know the the history of course is
[11:43] the the power saving which is driving
[11:46] significant interest by end customers
[11:49] and uh the the results that we'll share
[11:52] with you here are speaking for
[11:54] themselves uh but on the power saving
[11:56] the power savings uh were even bigger
[11:59] than what we thought a year ago where we
[12:03] had identified that on the modal level
[12:06] uh removing the DSP would remove
[12:08] approximately half of the power what and
[12:11] that we thought at the time would result
[12:13] in a 25% reduction system power what we
[12:16] actually measured is a 40% reduction in
[12:19] the system power of a 51.2 t lpo switch
[12:23] compared to current 800 gig DSP Optics
[12:26] with a 700 DSP and uh projecting this
[12:31] forward to the one or 2.40 generation
[12:34] this would go up to a 1 kilowatt power
[12:36] reduction uh in the context of a one24 t
[12:40] switch now uh there maybe and this is
[12:42] with silicon photonics you know Apper
[12:45] apples additional power savings are
[12:47] possible with other modulators but they
[12:49] would apply of course to any kind of
[12:51] Optics um in terms of the link
[12:55] performance uh so what enables lpo is a
[12:58] high Quality Service implementation in
[13:01] the switch chip that essentially uh
[13:04] allows an endtoend channel uh between
[13:07] switch to switch going through both the
[13:09] electrical and the optical domains uh
[13:13] again a year ago we had a margin of
[13:16] error of 10 to Theus 4 to 10- 5 with the
[13:19] latest uh uh product measurements we
[13:22] actually between 10 to the 6 to 10 the 7
[13:25] Headroom compared to the uh the pref
[13:29] back uh error rate so this is an
[13:31] enormous amount of margin and there's no
[13:33] question that this is highly a highly
[13:36] stable um uh operating condition uh one
[13:39] of the other
[13:41] surprises was that the lpu Optics
[13:45] actually had
[13:46] Superior Oma performance compared to
[13:49] traditional retimed modules um the the
[13:52] theory why it is is that if you have a
[13:54] single quantization in the switch DSP
[13:59] you actually get fewer errors than if
[14:00] you have multiple um conizations of the
[14:03] signals both in the switch and as well
[14:06] as the uh module DSP uh but it is
[14:10] surpris it was surprising to ourselves
[14:11] that the symbol error distribution is
[14:13] actually Superior um going to one of the
[14:17] implementation challenges um the biggest
[14:21] challenge believe it or not is that the
[14:22] electrical challenge Channel loss
[14:25] between the short traces that are the
[14:28] ports that are right in front of the
[14:30] switch chip on the board versus the long
[14:32] traces that are the outermost ports
[14:34] ports is obviously quite different and
[14:38] you know the the total difference may be
[14:40] in the order of 13 DB um so that is is a
[14:45] very important thing to keep in mind
[14:47] that the linear plugable Optics has to
[14:49] be able to deal with uh transmit signals
[14:52] that have this range of generation and
[14:56] um you know it is what it is um
[14:59] this is with printed circuit traces I
[15:02] should highlight here if one uses flyer
[15:05] cables that uh difference can be reduced
[15:09] um but from a manufacturing standpoint
[15:12] uh PC boards are generally preferable um
[15:16] so this is measure data on a f2t switch
[15:20] that's a production product in um at
[15:22] Arista um this is a photograph of the
[15:28] latest um modules that had the best
[15:32] driver the linear driver chip on the
[15:33] market and um it turns out they actually
[15:37] are tp2 compliant at the optical
[15:39] interface and again uh this is was a
[15:42] surprise to us that that was even
[15:44] possible uh but it is conceivable that
[15:47] with the very best Driver Designs one
[15:48] can deach achieve tp2 compliance uh this
[15:53] is in fact more stringent than what is
[15:54] actually needed uh for the operation as
[15:57] I hotlined earlier um another
[16:00] interesting development is there is
[16:02] innovation on these driver chips uh
[16:04] there is the idea of adding um
[16:06] monitoring capability inside the driver
[16:09] to actually sample the ey and and allow
[16:12] tuning uh these chips are not in
[16:14] production yet but it I think it makes a
[16:16] lot of sense to make the driver somewhat
[16:18] more intelligent and um the other very
[16:22] important development over the last year
[16:24] was the introduction of the so-called LR
[16:28] Optics which is a linear receiver and
[16:30] return transmit it turns out the
[16:33] receiver part in an lpu world is
[16:35] actually the easy part the transmit part
[16:38] is harder right and so the L uses a half
[16:41] DSP for the transmit some people call
[16:43] this half rtim uh this n to say achieves
[16:47] full tp2 compliance with all the
[16:49] existing monitoring capabilities and
[16:51] this will reduce system qualification
[16:54] efforts um there's also significant
[16:57] power savings achieve by the lro
[16:59] compared to a full DSP measure data
[17:02] indicates 9 wat typical for an 800 gek
[17:04] dr8 L module uh the best number we had
[17:08] on an 800 gig lpo module was 7 and a
[17:11] half and this is silicon photonics to
[17:13] silicon photonics and the traditional
[17:16] DSP Optics depending on the generation
[17:18] of thep is between 13 and 15 watts so
[17:21] the lro provides significant power
[17:23] saving as well U and then looking
[17:26] forward to the 100 Gig to 200 200 gig
[17:29] Lane
[17:29] transition as you all know the 18 geek
[17:32] volume ramp has begun and it will ramp
[17:35] strongly this year into next year uh
[17:38] with a very large amount of volume over
[17:40] the next many years the E the six geek
[17:44] the tun geek PR Lane will start with the
[17:47] switch chips that support tun gek per
[17:49] Lane which may be available at the very
[17:51] end of 25 sud in volume in 2026 and
[17:55] Beyond so one key question people are
[17:57] asking is will lpo or lro work with tun
[18:01] geek perlane electrical and Optical and
[18:05] the uh analysis of for says this is
[18:08] definitely
[18:09] feasible uh basically the physics of the
[18:12] end channels don't really change the
[18:14] bandwidth of course is much higher uh
[18:16] what it will require is much shorter
[18:18] electrical traces on the motherboard or
[18:21] alternatively fly over cables uh but the
[18:24] same requirement is actually true for
[18:25] conventional DSP modules as well so we
[18:28] would expect first uh prototype uh
[18:31] demonstrations of linear plugable Optics
[18:34] at the system level uh I'm sorry were
[18:37] there were some early prototypes at this
[18:39] year's show but the first uh system
[18:41] level demonstrations of course have to
[18:43] wait until there's a platform available
[18:45] in uh either late 25 or in volum in
[18:48] 2026 um it also it may be required to
[18:52] use lro for 224 gig which certainly
[18:54] simplifies the channel um Power
[18:57] projections um this is a sort of a color
[19:01] chirt where lpo in blue is always going
[19:03] to be the lowest lro also significant
[19:06] power reduction compared to call it the
[19:08] best in breed future generation dsps and
[19:11] the current in Red 7 nanom type dsps and
[19:15] you know as you can see for uh the the
[19:18] the savings really increase as one goes
[19:21] from 800 gig to 600 gig to Future 3200
[19:25] gig modules where uh building a 3200 gig
[19:28] module with conventional current DSP
[19:31] technology is essentially not feasible
[19:33] whereas with an lro or lpo approach it
[19:36] fits in the form factor uh and the power
[19:39] envelope of a conventional osfp or qpd
[19:43] module and um just one line here on 3.2t
[19:47] Optics um there is some new renewed
[19:50] interest to increase the density of
[19:52] switch systems in uh 16 channels of Tek
[19:56] which would be a
[19:57] dr16 and this is the again the estimated
[20:00] power for these different type of
[20:02] systems where the lpo and the lro or
[20:05] clearly feasible whereas a conventional
[20:07] DSP uh is not um and uh again the
[20:11] attraction here is to get to one New
[20:13] Face Place sensity in the one 2.4t
[20:16] generation which matters for high
[20:18] density AI
[20:19] clusters so in um there's of course
[20:22] other power savings available which do
[20:25] not relate to the choice of DSP or lro
[20:28] or lpo which have to do with better
[20:31] lasers lower Optical coupling losses um
[20:35] better lower voltage modulators and so
[20:38] on uh all of this is of great interest
[20:41] and and so on and somewhat independent
[20:43] of the DSP discussion uh but um we hope
[20:47] of course that other improvements on the
[20:49] optical uh engine uh will come in as
[20:51] well um and we talked about the lpo that
[20:54] was announced uh at uh ofc uh there's
[20:59] weekly meetings and I I have to say
[21:01] there's very good progress that's being
[21:02] made in um characterizing you know the
[21:06] the spec uh challenge here and uh we
[21:09] hope to get the spec out certainly this
[21:11] year to allow enable a wide ecosystem of
[21:14] suppliers to build compatible Optics uh
[21:17] in
[21:18] summary the lpo delivers very very large
[21:21] uh powering system level uh and the L
[21:25] will as well uh not not quite the same
[21:27] amount but still sign ific enough that
[21:29] there's great interest in these power
[21:31] savings for large scale Ai and other
[21:34] applications uh the measured prefect
[21:36] data is outstanding um we we have lpo
[21:40] modules that support full tp2 compliance
[21:43] this does require a bit more power on
[21:44] the driver uh and is in fact more
[21:46] stringent than what is necessary tuna
[21:49] geek palanta we believe is totally
[21:50] feasible and lro may be required uh and
[21:54] we have of course the lp MSA to uh
[21:57] formalize these specification
[21:59] and that's all the sides I had thank you
[22:02] very
[22:04] much thank you very much Andy for a
[22:06] great presentation it is great to have
[22:08] you here it was great to be with you at
[22:10] ofc and could I ask you right now in
[22:13] your opinion from the whole industry
[22:15] what are the most important challenges
[22:17] that we need to
[22:18] cover yeah I mean from an end user
[22:22] perspective uh I think it's fair to say
[22:24] that no end user it doesn't matter it's
[22:26] a big cloud or small cloud wants to be
[22:29] in the Optics qualification business so
[22:31] the real challenge is that the industry
[22:33] has to deliver a fully tested a fully
[22:37] conforming solution that end users can
[22:40] deploy with high confidence so that
[22:42] there's no issues in the field if you
[22:44] know what I mean and clearly this is
[22:46] more challenging with lpo or lro than a
[22:50] conventional DSP however all the
[22:52] measurements we have taken so far on the
[22:54] next current generation of silicon with
[22:56] the advanced service indicate that
[22:58] there's plenty of margin available so in
[23:01] the end it comes down to in a
[23:03] multivendor situation with multiple
[23:05] system vendors multiple Optics vendors
[23:07] multiple Optics Technologies multiple
[23:09] Tia chips multiple driver chips right
[23:12] can we get to a point where there's a a
[23:16] high enough margin that people can
[23:18] deploy lpo and lro with confidence and
[23:21] we certainly believe the answer is yes
[23:23] but we're not there today there's a lot
[23:25] of people who want to get us get there
[23:28] to today and there the first question is
[23:30] coming from Ahmed at from optic W Ahmed
[23:34] please unmute your microphone switch on
[23:35] your screen and tell us squats on your
[23:37] mind perfect thanks um yeah thanks for
[23:40] the the presentation my question is
[23:41] actually about uh possibility to use
[23:44] probabilistic Ambit shaping uh with the
[23:46] modulation on these U lbos is it
[23:50] possible is it
[23:54] considered was that a question for me
[23:56] sorry yeah the question is about using
[23:58] the modulation like you are using you
[24:00] showed the B for uh like modulation
[24:03] scheme I'm just asking is it possible
[24:05] like to apply on top of that bistic
[24:07] amplitude shaping pulsing for for this
[24:11] uh modulation yeah I mean this is a
[24:13] theoretical question basically the the
[24:15] world Rel live in is defined by the
[24:17] existing I you know ethernet specs and
[24:20] existing service
[24:21] implementations and the existing tias
[24:24] and the existing drivers right so what
[24:27] we've demonstrated is it works well with
[24:30] all these existing components and some
[24:31] of these components in particular the
[24:33] drivers can be enhanced however you know
[24:36] assuming any making any other
[24:39] assumptions to changes in modulation
[24:41] formats or to further improve the cic
[24:45] signs are of course limited to future
[24:47] implementations of these cices which are
[24:49] not available today so the the Ceres is
[24:52] what it is and the other thing I would
[24:54] like to note here is that all leading
[24:56] silicon vendors uh have shown at ofc
[25:00] that they can in fact Drive lpo Optics
[25:04] of course it doesn't mean that all the
[25:05] qualifications are done but from a
[25:08] Hightech the current generation of
[25:10] service technology I think there's no
[25:12] question that the implementations are
[25:14] good enough to have an entn Channel but
[25:17] there rest is very careful system
[25:19] implementation on both the motherboard
[25:21] the Optics modal the choice of tias the
[25:24] drivers and so on so it is not for free
[25:27] right you have to do a very very careful
[25:29] implementation but the current
[25:32] evidential experience you know suggests
[25:34] that it's really not an issue but a lot
[25:36] of testing still needs to be done okay
[25:39] thanks the second question is coming all
[25:41] the way from Switzerland from newel
[25:44] Winfred n from focus light Switzerland
[25:46] what's on your mind hello thank you um I
[25:50] I I was following the ramp session and
[25:52] other other events at ofc and for me it
[25:55] was not 100% clear if lpo has to be
[25:59] based on Silicon photonics or are there
[26:01] other Technologies modulations possible
[26:04] to do this or um because when we talk to
[26:07] our clients so not silicon photonics is
[26:10] not as widespread at one might see at
[26:13] ofc I get the impression at least from
[26:16] from our clients so so this is this is a
[26:19] question I have here in the in the
[26:21] audience and also for Andi so do you
[26:24] only see silicon photonics being an
[26:27] enabling technology for LP or other
[26:29] other technological options basically
[26:31] yeah it that's a very good question so
[26:34] linear Optics benefit from uh linear
[26:36] driver and the linear modulators so um
[26:41] Optics technologies that include vixel
[26:43] and emls that are not as linear uh have
[26:47] a challenge because the performance of
[26:49] the channel channel is not as good okay
[26:52] that doesn't mean you can do it but you
[26:54] would almost have to like pred distort
[26:57] or level shift you know the the single
[26:59] to address the nonlinearity of the
[27:02] modulation itself so it's it's it is
[27:05] easier with silicum phonics and with um
[27:10] Thum liate for example on the other hand
[27:13] the lro doesn't have this limitation so
[27:15] in an lro world it's just like today you
[27:17] can drive you know emls or wixel all day
[27:20] long so this is actually one of the
[27:22] advantages of lro is that it's less
[27:24] dependent on the exact Optical uh
[27:26] modulation performance
[27:28] thank you we have one more question we
[27:31] have many more questions but we have
[27:32] time so please keep asking then the next
[27:34] one is coming from wayam hang what's on
[27:36] your
[27:40] mind weam please unmute your microphone
[27:43] and tell us what's on your
[27:45] mind well we solve the it issue the next
[27:48] question is coming from
[27:50] act Abu Mas tell us what please switch
[27:53] on your camera unmute your microphone
[27:55] and thanks unfortunately don't have a
[27:58] camera I'm drawing from my desktop
[28:00] computer so my question is actually in
[28:03] line with the previous question so we
[28:05] see you know Trends in you know many
[28:08] Technologies like thin film lithium
[28:10] niobate you know to to improve the
[28:12] bandwidth rate of the modelator so the
[28:15] question is actually there's a
[28:16] limitation in Silicon photonic
[28:18] modulators right so you cannot go beyond
[28:20] certain and there's issues with law so
[28:23] people thinking now about then fil let I
[28:26] Obe it and this the the traditional
[28:28] question is that now is the packaging
[28:30] right the integration with the Silicon
[28:32] photonics is there like a a plan which
[28:36] which integration method will be
[28:38] considered is it like hybrid is it
[28:40] heterogeneous what's your uh view on
[28:43] that thanks yeah so if I can try to
[28:46] answer this um the obviously there's
[28:48] many great Technologies in development
[28:50] and people are doing that that they're
[28:52] working over you know very really hard
[28:55] to to get newer Technologies into
[28:57] production
[28:59] what we have observed recently is that
[29:01] end customers are the large end
[29:02] customers are getting quite risk averse
[29:06] in terms of their future purchasing
[29:08] plans basically they have to commit to
[29:10] millions and millions of Optics you know
[29:13] for like deployment next year and
[29:16] realistically they they want to use
[29:18] technologies that are in volume
[29:19] production mature stable guaranteed to
[29:22] work and there's lead times of course
[29:24] and so on so whatever they're going to
[29:26] deploy say by the middle of next year
[29:28] they have to decide today and
[29:30] unfortunately the newer Technologies
[29:32] including th libate are not at the point
[29:34] of volume production that they actually
[29:36] qualify for volume deployment in
[29:38] people's minds that doesn't mean that
[29:40] you know they won't get deployed in some
[29:42] places but in the volume case people
[29:45] will remain as far as we can tell with
[29:48] the existing technologies that are
[29:50] shipping in high volume because they
[29:52] don't see a
[29:54] choice okay so you still see s photonics
[29:57] right sorry I'm for I meaning silicon
[29:58] tonics is proven right um okay EML is
[30:02] proven vixel are proven people will ship
[30:04] I mean there's shortages it's there
[30:06] there's a supply demand imbalance right
[30:08] now in the market right so as people try
[30:10] to ramp up the ethernet geek Supply they
[30:14] they're naturally doing it in the way
[30:16] that they already have the technology in
[30:18] mass production there's no way around
[30:20] it okay yeah thanks s thank you very
[30:23] much Ahmed and now we go to the United
[30:25] Kingdom we are going to meet the CEO of
[30:27] one via suas you also have a question
[30:30] for Andy please speak your mind sure hi
[30:33] it's ask from one via can you hear me L
[30:35] and clear perfect uh quick question on
[30:39] um lpo based modal so what uh reaches
[30:42] are they uh planed to cover so
[30:44] LR uh Dr F
[30:48] FR yeah that's that's a very very good
[30:51] question so the the the impairment of
[30:54] the optical channel of course does
[30:55] increase with longer reaches um and that
[30:59] doesn't mean that it wouldn't work but
[31:00] the focus of lpo is the you know 500
[31:03] meter data center type environment and
[31:07] some people think the focus actually
[31:09] should be the 100 meter AI cluster where
[31:11] the power really matters right so the
[31:13] goal is not to displace the longer reach
[31:15] Optics in any way at all this is really
[31:18] focused on new applications that the the
[31:22] the new chips that support the new cice
[31:24] remember you need the latest cice for
[31:25] this even to work right and um inability
[31:29] with the the last generation of Optics
[31:31] is actually a lower priority so the the
[31:33] focus is really in the context what what
[31:35] the customer wants is lower power for AI
[31:38] clusters that works it's it's really
[31:40] simple right and um we we we have had
[31:44] sent in good results with lpo lro is
[31:47] even simpler to verify and they will go
[31:49] with one or the other in volume next
[31:51] year so much is clear for AI but the
[31:53] rest of the market just used DP
[31:58] so sorry suas let's move on because we
[32:00] have a few more questions Andy You're
[32:02] Gonna like the next question he coming
[32:03] all the way from Ireland it's coming all
[32:05] the way from Dublin City University Liam
[32:08] bar welcome to the meeting tell us
[32:10] what's on your mind Hey thank you
[32:13] apologies obviously I'm an academic here
[32:16] so not an industry person but I saw one
[32:18] of your slides you mentioned the move
[32:20] from Pam 4 vure to 16 quam I'm just
[32:22] wondering what time frame do you see the
[32:25] short reach networks transitioning from
[32:28] direct detection to coherent and do you
[32:30] think the the power consumption there is
[32:32] going to be an issue from the DSP side
[32:34] can we use dual po polarization where we
[32:37] can maybe get like 400 G or 800 G
[32:39] potentially per
[32:40] wavelength yeah yeah so this is the
[32:42] isold question of how far can you push
[32:45] PM 4 in reach as the B rates go up
[32:48] versus uh how how inexpensive or
[32:52] attractive does the six P become in in
[32:55] say a 16 gig or future world and and the
[32:59] current belief is that all the data
[33:00] center requirements within the data
[33:02] center and even within the campus uh
[33:04] setting can be met with Pam 4 which is
[33:07] the highest volume thing on the market
[33:09] and likely to be the lowest cost for the
[33:11] facil future where the 16 qu comes in
[33:14] and and I should mention here that you
[33:15] know the the business of coherent Optics
[33:18] is growing rapidly every year and it's
[33:21] it's a great business to be in it's a
[33:22] much higher selling price per per port
[33:25] and so on and so but um where it gets
[33:27] attractive is if you're running out of
[33:29] fibers so in a setting for example in a
[33:31] campus setting if you're not fiber rich
[33:33] between your buildings and you suddenly
[33:35] want to transfer you know hundreds and
[33:37] hundreds of PAB bits between these
[33:40] buildings you may not have a choice but
[33:42] to deploy coherent just because it's too
[33:45] much hassle to dig up the fibers and put
[33:48] in another you know 10 thousand of
[33:50] fibers so so there are use cases where
[33:53] if you're fiber poor so to speak or
[33:55] relatively speaking not fiber rich you
[33:58] you have no choice but the simplest
[34:00] technology for PM 4 that will always
[34:02] work is just paril right the basically
[34:06] dr8 so if you have enough fibus you can
[34:08] run that 10 kilometers all day long if
[34:10] you have to go F4 uh you're looking
[34:13] probably at 2 3 4 kilometers and there's
[34:15] some new polarization schemes where it
[34:17] may go to 5 six s but again the vast
[34:20] majority of data center traditional data
[34:23] center locations can be met with Pam 4
[34:25] so the the the real opportunity is
[34:28] beyond what pfor can do which is a large
[34:30] and growing
[34:31] market and the the next two questions
[34:34] I'm for from Silicon photonics foundies
[34:36] one on Tei liate and the other one
[34:38] actually second one is indium phosphate
[34:40] but the first one let's go to
[34:41] Switzerland to meet Amir gadini the CEO
[34:44] of lium Amir welcome to the meeting tell
[34:46] Andy Amir what's on your mind yeah thank
[34:49] you hello Andy can you hear me by way
[34:51] yes a quick question so I mean so you
[34:54] really dice about like different aspects
[34:56] of the bsp so actually wanted to ask you
[34:58] about the you what do you think about
[35:00] the difference between the differential
[35:01] drive like for gssg which is for silicon
[35:04] mostly and the GSG so let's say single
[35:07] Drive do we see any like a power let's
[35:10] say cost that you pay or maybe maybe
[35:12] some other thing you know bit between do
[35:14] think but there's a huge debate that you
[35:16] know maybe single drive is not going to
[35:18] make it because of some technical
[35:21] limitations yeah I I'm actually not the
[35:23] expert here but my observation would be
[35:25] that whatever driver chips people
[35:27] develop V oping for either the lpo or
[35:30] lro or or any kind of Optics are
[35:33] designed to handle any kind of
[35:35] modulation technology so naturally
[35:37] they're differential output and then it
[35:39] comes down to can you go to lower power
[35:41] if you do differential or you get a
[35:43] better signal and that's really within
[35:45] the optical engine so the End customer
[35:47] doesn't see that right all they care
[35:48] about is the performance that they get
[35:51] um going back to my my earlier point on
[35:53] thin liit uh I think the current view is
[35:56] that the advantage of thin light are
[35:58] much bigger in the 224 Gig Generation
[36:01] where Sil bonics is kind of running out
[36:03] of bandwidth it still performs well but
[36:05] C liate performs better right and then
[36:07] also the power savings are more
[36:09] significant there so the I think that
[36:12] the window for C Li bit is in my mind is
[36:14] 224 gig
[36:17] P4 thank you very much Amir and I Amir
[36:20] loved your answer by the way and we
[36:21] continue another question comes from the
[36:23] way from Canada the Canadian photonics
[36:26] fabrication Center cpfc Daniel welcome
[36:29] to the meeting all the way from Canada
[36:31] tell us what's on your mind thanks thank
[36:34] you Andy for the excellent presentation
[36:36] um my question is going back to the
[36:38] coherent linear receive Optics uh so in
[36:41] traditional coherent Optics you have a
[36:43] local oscillator on the receive side and
[36:44] the DSP does a lot of work in matching
[36:46] that local OS to the signal when you
[36:48] pull the DSP off the the receive side
[36:51] what happens there what are the what are
[36:52] the problems with having to do that
[36:54] phase
[36:56] matching well uh you're talking coherent
[36:59] now right in there's one startup and I
[37:01] forgot out of Santa Barbara I forgot the
[37:03] name now that is trying to productize
[37:05] the DSP free coherent Optics and and you
[37:08] know if that works that's great so there
[37:10] are perhaps ways to simplify the whole
[37:13] operation of coherent in the context of
[37:15] a Data Center Reach uh and and this may
[37:18] take off that's again the beautiful
[37:20] thing with Optics is this always new
[37:22] innovation and great ideas how to make
[37:25] it even better um uh I I I don't know
[37:28] much about this particular startup in I
[37:30] can't even recall the name I apologize
[37:32] here um but it there was some ofc
[37:35] presentations from you know UC Santa
[37:37] Barbara on that topic over the the years
[37:40] and there certainly early demonstration
[37:43] now the negative is you're putting all
[37:45] the performance on the optical subsystem
[37:48] right so that has to be implemented in a
[37:50] very very good way to deliver the the
[37:52] bidder rate and uh the DSP of course has
[37:55] to be able to recover that
[37:59] I'm sorry the the main
[38:02] SCE thank you very much we got two more
[38:04] questions for you and this will be the
[38:06] the the final two questions because we
[38:07] need to move to the next speaker the
[38:09] first one is coming from the mill
[38:10] University the group of David PL David
[38:12] you're watching this hello great to see
[38:14] you great to see your colleagues here
[38:16] Bon Q what's on your
[38:18] mind hi Andy thanks for the great
[38:20] presentation I was wondering is when we
[38:23] talk about no DSP do are we considering
[38:25] pul shaping in that all
[38:28] shaping yeah well there's no in a linear
[38:31] Optics it's linear right there's no
[38:33] other shaping all the work is done by
[38:35] the DSP in the switch chip so the the
[38:39] the reason the performance is so good is
[38:41] that those dsps are digital LIF lots of
[38:44] tabs they actually better than the dsps
[38:48] generally in the Optics module itself
[38:50] now they have the overhead of course of
[38:51] the electrical Channel but again the
[38:54] with 20 plus you know tabs you can do
[38:57] amazing amount of recovery the the
[38:59] reason these dsps are as good as they
[39:01] are was they were designed to handle you
[39:03] know copper cables that are also kind of
[39:06] a very poor Channel but it turns out the
[39:08] same benefit URS to a combined you know
[39:10] electrical trace and an optical Trace as
[39:12] well so so the DSP is whatever it is and
[39:17] all the magic there is what you know
[39:19] Brom and Nvidia and marence on doing the
[39:22] dsps um and it's really that performance
[39:25] that enables the endtoend channel um um
[39:29] so does this answer your
[39:31] question it answered my question and now
[39:34] with the final question of the series we
[39:36] go to the next speaker who has a
[39:38] question for the first Speaker Mar
[39:40] lukovic CEO five reality and a very
[39:43] respected person in a photonic industry
[39:45] has a question about polymers Mark speak
[39:48] your mind you Sandy uh what is a
[39:51] realistic timing for organic uh you know
[39:53] polymers as a material what you
[39:56] say all I I have no idea because you
[39:59] know I'm actually not that involved in
[40:01] the most advanced Optics Technologies
[40:03] our our job at AR is to bring things to
[40:05] Market that we can ship next year and
[40:08] and you know we'll we'll test those
[40:09] modules when when they arrive but I I
[40:11] truly don't know when they happen I I do
[40:13] have one last comment though that um is
[40:17] um there is no emerging interest in
[40:20] 448 gig Pam 4 Believe It or Not uh which
[40:24] initially would require DSP of course to
[40:26] encode it but the belief would be that
[40:28] it's lower cost uh than 224 because it's
[40:32] half the number of channels now for 448
[40:34] you would need very high bandwidth and
[40:36] one of the few bandwidth things that can
[40:38] do this is of course the baate and the
[40:41] polymers so there may be another opening
[40:43] in the 448 generation but you know
[40:46] realistically this is we're talking I
[40:48] don't know 27 28 29 so this is several
[40:51] years in the future but the the world
[40:53] does not end at 224 gig I just want to
[40:55] make that clear and and where siliconic
[40:58] bandwidth most likely will add to 24 gig
[41:03] okay so the good thing of this meeting
[41:05] this is not a webinar this is a group of
[41:08] people who want to talk to each other
[41:09] Mark us for a realistic time for organic
[41:12] polymers to be available and mature Lou
[41:15] Johnson from nlm photonic but let me let
[41:18] me say this definitely right in the end
[41:20] what drives adoption of newer
[41:22] Technologies is that the existing
[41:24] Technologies no longer can do it right
[41:26] so it is true that in the 448 gig world
[41:29] it's game over for you know I I think
[41:32] it's game over for solic and then
[41:34] there's a question what else actually
[41:37] works see for those of you who are in
[41:39] YouTube right now you can go at any time
[41:42] search for Andy BOS right technology at
[41:44] right time and watch that video because
[41:46] it's fantastic but before that Andy
[41:48] there is somebody who wants to answer
[41:49] mark question H please tell us from NL
[41:52] photonics views what's in your mind when
[41:55] are we going to have available at
[41:57] industry ready organic polymers for
[41:59] silicon photonics yeah thank you Jose
[42:02] and
[42:03] Andy so as was mentioned this going to
[42:07] depend on customer on ultimately
[42:10] customer needs and I agree that there's
[42:12] opportunities at both the 224 and four
[42:16] and
[42:18] 448 B bandwidths but
[42:22] the realist realistic timeline with
[42:25] depending on Fab adoption
[42:27] probably
[42:29] sampling
[42:30] 2026
[42:32] is
[42:34] and broader availability 2027 2028 we
[42:38] are going to organize a meeting in two
[42:39] years for you Lois to show us to show us
[42:42] in 2026 majority of organic polymers it
[42:45] is great to have you here Louis you
[42:46] actually brought one slide to the
[42:48] meeting now it's the perfect time to
[42:49] share it with the world before we get
[42:51] the floor to
[42:53] mark all right
[43:04] thank thank you Jose for the
[43:06] all right thank thank you Jose for the
[43:06] introduction so as Jose mentioned Lewis
[43:08] Johnson CTO of analin fonics and our
[43:12] focus is on extending silicon photonics
[43:16] for the linear Drive era using hybrid
[43:19] organic modulation which can take
[43:22] advantage of low power linear drivers
[43:25] and particularly reducing the power use
[43:28] and providing good modulation depth on
[43:32] the on the transmit side with very high
[43:35] Extinction ratios low low and low
[43:38] insertion losses and this can provide
[43:41] you know along with Technologies like
[43:43] thin film lithium iate this can provide
[43:45] a further power savings beyond the
[43:48] linear driver alone we have some
[43:50] modeling data from work that we've been
[43:52] doing with noos semi recently showing a
[43:55] you know further you know 10 to 40%
[43:58] reduction being possible above and
[44:00] beyond the linear driver reduction and
[44:03] with a path to further Improvement as
[44:05] the organic materials
[44:09] are f are further optimized which is a
[44:12] advantage of Organics is that there is a
[44:14] sub substantial parameter space for
[44:17] further material
[44:19] Improvement and the advantage become of
[44:23] this technology becomes even larger at
[44:26] 200 G per lane or 400 G per
[44:31] Lane and a recent you know another
[44:35] recent development here and some of this
[44:36] was discussed in a some work
[44:39] co-published with the call Institute of
[44:42] Technology at eock is Organics are now
[44:46] able to survive up to 120 degre C
[44:50] long-term
[44:52] stability which has represented
[44:54] substantial advance for being able to
[44:56] bring the them closer to electronics for
[44:59] data center deployment thank you very
[45:01] much l congratulation on your work that
[45:03] you're doing with IM and with aim
[45:04] Technologies we want to try your staff
[45:07] we want to test your staff I want to see
[45:08] it in other Laboratories when can we
[45:10] access it and
[45:13] how yeah thank thank you Jose so
[45:17] we at this point for I mean just feel
[45:21] feel feel free to feel free to contact
[45:23] us and we are working with a couple
[45:26] translational resarch Fabs aims probably
[45:28] to furthest along but we are happy to
[45:30] partner with Optical with Optical module
[45:34] developers and other Fabs we have you
[45:36] know several other relationships
[45:37] underway there a couple of which Jose
[45:39] already already mentioned and I mean and
[45:42] even for for academics we can Prov we
[45:45] can provide samples and willing to work
[45:47] but our F our focus at this point is
[45:49] moving this technology from the small
[45:52] scale area where it's been at the Die
[45:54] Level towards wafer scale Fab scale
[45:58] deployment
[45:59] to provide efficient modulation for
[46:02] linear for commercial linear Drive
[46:04] Optics so contact them and you can use
[46:07] their polymer Lou congratulations on
[46:08] everything that you are achieving and
[46:10] now it is time all of you it is time to
[46:12] welcome our next speaker a person who
[46:14] has been very controversial and we all
[46:17] we all admire and respect mark lukovic
[46:20] thank you very much for being with us
[46:21] today the floor and the attention of
[46:23] everyone goes to you do you do you like
[46:27] like linear Drive
[46:30] Optics uh I can't say it's my favorite
[46:32] subject uh Jose but you probably know
[46:34] that uh uh so let's see I need to here
[46:37] we
[46:39] go
[46:41] um yeah I mean uh there there has been a
[46:44] precedent for linear uh 20 years ago
[46:47] people are still recovering
[46:49] from uh the uh the battles in dealing
[46:53] with a linear situation uh and um I mean
[46:57] this sort of thing was ignored in the
[47:00] whole discussion of the narrative
[47:03] unfortunately uh and uh I
[47:05] know recently Cisco has put put out
[47:09] something publicly where they claim that
[47:12] the ACL they were dealing with a half a
[47:14] decade ago made have may have had a
[47:16] problem with the linear interface uh I
[47:19] think that's uh it's a little bit
[47:21] historic revisionism uh but uh you know
[47:24] to the extent that you know they're
[47:26] willing to argue that it would maybe it
[47:27] would have been helpful at Cisco
[47:28] mentioned that a little bit earlier uh
[47:31] basically uh you know the fundamental
[47:33] the fundamental problem here
[47:35] unfortunately is that uh we needed to
[47:39] address these matters uh a a lot
[47:43] earlier uh you know when we were
[47:44] contemplating pay4 we were contemplating
[47:47] jump on the 400 gigb ethernet before
[47:49] having to reverse on standards to 200
[47:52] gab ethernet you know contemplating uh
[47:55] strangle holds and the and and all the
[47:59] and all these sorts of things that have
[48:01] uh that have happened you know the power
[48:03] the cost but once the pan4 horses left
[48:07] the barn that is the wrong time to be to
[48:09] be addressing to be addressing these
[48:11] things and punishing suppliers you know
[48:14] like
[48:15] marll uh you know you know after the
[48:18] fact you know and uh you know kind of in
[48:20] plying that uh you know maybe there's no
[48:23] need for these dsps I am watching my
[48:25] time I was I promise
[48:27] um so um yeah you know very very quickly
[48:33] uh what What's what hasn't been
[48:35] addressed is just the age-old arguments
[48:38] that we've had with centralized versus
[48:42] uh uh you know distributed uh you know
[48:45] typ types of uh types of processing and
[48:48] it isn't you know it's there's tot it's
[48:50] totally ignored what would happen if we
[48:53] went to a centralized approach with
[48:55] these lpos uh that all what you're doing
[48:58] is you're just transferring the power
[49:01] and and all all that complexity to one
[49:03] place and I'm not sure that's what we
[49:06] really wanted to do uh you know I think
[49:09] uh and I think Andy reflected on this uh
[49:12] you know basically um you know this this
[49:14] solution of of half right retimed that's
[49:18] always that's been available it could
[49:20] have been talked about uh you know a a
[49:22] lot sooner it should have taken a year
[49:25] of Industry resources being used and and
[49:29] and that sort of thing you know the fact
[49:30] of the matter is yes store
[49:33] interoperability uh interoperability
[49:35] displays at ofc but how many how many of
[49:38] them were fudged and and I don't you
[49:40] know I don't mean to you know suggest
[49:43] anything onour but but you know I think
[49:45] I think a lot of suppliers felt they
[49:47] needed to to to to show this because
[49:49] that's where this convoluted rhetoric
[49:52] was going even more convoluted than
[49:54] co-packaged Optics uh and uh and really
[49:58] uh if you would have you know Google
[50:00] knew Google the that has the most uh you
[50:04] know influence on optical development of
[50:06] the
[50:08] hyperscalers they they knew uh in March
[50:12] in March of
[50:13] 2023 based on their own analysis based
[50:16] on their own experience uh and not on
[50:18] the demos of a few links they they knew
[50:21] you could not do 200g plus not to
[50:24] mention even 100g it's it's somewhat
[50:27] problematic this this sort of this sort
[50:29] of thing should not happen in in our
[50:31] industry it makes us look bad it's a
[50:33] hard enough industry and uh you know we
[50:36] have to be careful about these things
[50:38] and uh so you know Halo I know you know
[50:41] I I guess uh I'm hearing that Halo is
[50:44] not as good you know half ton is not as
[50:46] good as uh linear pluggable uh but but I
[50:50] don't know some of the feedback that The
[50:52] Oaf is getting you know it's it's kind
[50:54] of in the ballpark on Power and cost you
[50:56] know so I don't know I don't know what
[50:58] we're talking about and uh and yes it is
[51:01] insulting to technologist it's you know
[51:04] the idea that you could you're going to
[51:06] tell designers and technologists that
[51:09] there's no purpose for a DSP and you
[51:11] could just remove it you just remove it
[51:13] it is just one of the most in my 40 plus
[51:16] years of experience you know one of one
[51:18] of the most idiotic ideas I've ever
[51:21] heard uh I'm sorry to say that but you
[51:24] know I have to express my point of view
[51:26] uh and yes Nvidia could do it just like
[51:29] they could do co- packaged Optics thec
[51:31] portfolio we'll have to see where that
[51:33] goes uh uh and uh but but essentially
[51:37] essentially this is something that never
[51:39] should have happened it's a total
[51:40] embarrassment thank you so
[51:43] much thank you very much for a great
[51:45] presentation Mark I enjoyed it very much
[51:49] what do you think of Kobo no I'm kidding
[51:51] are there any questions are there any
[51:53] questions for Mark lukovic in the room
[51:56] puya the floor and the attention of
[51:58] everyone goes to you okay thank you
[52:01] everyone um so I would like to introduce
[52:05] our our next speaker uh actually it's a
[52:07] very exciting company that does great
[52:09] things uh without any further Ado I
[52:12] would like to invite Ron uh Schartz and
[52:15] puber from lightelligence the floor is
[52:17] all
[52:23] yours okay thanks thanks PUO um let me
[52:26] get started
[52:29] here all right am I sharing the right
[52:31] screen yes okay perfect so thanks P for
[52:36] the introduction uh my name is Ron sortz
[52:38] and truber director of engineering for
[52:40] lightelligence um lightelligence was
[52:42] founded in
[52:44] 2017 uh to develop and commercialize
[52:47] photonics Technologies for
[52:50] groundbreaking improvements in Computing
[52:53] Solutions and one of those is uh Drive
[52:57] optics for PCI and cxl
[53:00] applications so I'm going to share a
[53:02] little bit about that in uh in today's
[53:08] presentation and we go to next
[53:12] slide so here's the agenda we'll talk
[53:14] about how linear Drive Optics uh is
[53:17] superior for scaling we'll get into
[53:20] Optical
[53:21] cxl um as the emerging standard uh we'll
[53:25] talk about our case study how we proved
[53:27] it to ourselves and then we'll discuss a
[53:29] little bit about the um the active
[53:32] optical
[53:37] cable okay so first of all what we're
[53:40] hearing from all of our customers is
[53:43] that they do not want to pay the
[53:45] penalty uh for the re-timer they don't
[53:48] want to pay the power penalty and they
[53:50] don't want to pay the latency penalty
[53:52] and what you can see here is that the
[53:54] linear Drive is is excellent for scaling
[53:58] the signal loss in Copper is quite High
[54:01] even over short distances so what you
[54:04] generally find is that at 10 meters it's
[54:07] really it's really done and even at uh
[54:10] Gen 5 and Gen 6 speeds you know we're
[54:13] looking at maybe two to three meters
[54:15] even with active cables so to really get
[54:18] across the rack uh you're going to need
[54:21] Optics um the other thing we're seeing
[54:24] is that um PCI
[54:27] uh you know or the file latency right is
[54:31] is dramatic for ethernet it's it's got
[54:33] the built-in um FEC versus
[54:37] pci5 is uh is no effect and in PCI 6
[54:41] there's a lightweight effect so it's
[54:42] slightly higher latency but but overall
[54:46] um you know it's it's a consistent uh
[54:49] low latency versus the um dramatic high
[54:54] and variable latency of ether
[54:57] um furthermore the cable cross-section
[55:01] you can just look at here you know your
[55:03] your copper cables are very bulky versus
[55:06] your Optics is incredibly small this was
[55:09] an example using a single wavelength but
[55:12] once you add in
[55:13] multi-wavelength um you know those those
[55:17] uh cross-sectional diameters go way down
[55:20] um
[55:26] so um let's talk a little bit about
[55:28] Optical cxl so cxl is the emerging
[55:32] standard for memory disaggregation it
[55:34] adds cache coherency and memory
[55:38] functions to the ubiquitous PCI
[55:40] interface uh that's that's dominant the
[55:43] dominant load store interface that's
[55:45] used so cxl is a perfect um uh solution
[55:50] for uh memory disaggregation it's it's a
[55:53] perfect memory interface and and uh what
[55:57] cxl over Optics will do is allow you go
[56:01] across the rack so you can have your
[56:03] compute your GPU in in one rack you can
[56:06] have your memory in another rack they
[56:08] can be interconnected by series of
[56:11] switches um and it really enables this
[56:14] uh memory
[56:15] disaggregation and um our cxl over
[56:19] Optics uh provides low latency high
[56:22] bandwidth and a data center uh type of
[56:25] reach
[56:28] so um just a quick case study so we've
[56:32] uh demonstrated this at the flash memory
[56:35] Summit last year we won best of show
[56:37] award uh we're the first company ever to
[56:40] demonstrate um cxl over Optics uh using
[56:45] a PCI Gen 5 uh solution and what you can
[56:49] see here was we had a a super micro uh
[56:53] cxl enabled uh AMD Genoa CPU on one side
[56:59] uh with an Nvidia um A10 GPU in in the
[57:04] server
[57:06] and uh what we did was we compared
[57:09] storing the large language model on the
[57:13] the SSD memory you know on the same
[57:15] server versus moving that memory to a
[57:19] cxl memory expander that was on the the
[57:22] memory expander box which you can see on
[57:24] the right and so in this case we we put
[57:27] the large language model you know 10 m
[57:30] away and in fact the the decode
[57:33] throughput of the model was uh about two
[57:37] and a half times better you can see the
[57:39] results here and so we were running opt
[57:42] 66b which is a smallish uh model but it
[57:46] was about 120 gig so it fit on a single
[57:49] memory expander and uh we did the news
[57:53] Tech summarization and sure enough it
[57:55] was two and a half times faster than
[57:57] disc that just shows really the
[57:59] advantages of
[58:00] cxl uh you know compared to nvme um with
[58:04] all of those results
[58:07] here um and so what we're um you know
[58:11] showing here is our linear Drive uh cxl
[58:15] pcie Express AOC so uh today the PCI
[58:20] Express copper cable standard is copper
[58:24] link um and we've uh made some
[58:27] enhancements to that spec to be able to
[58:29] support uh linear Drive Optics um and so
[58:33] that's that spec uh you'll see copper
[58:36] link 1.0 being released uh shortly and
[58:39] being advertised um it'll support copper
[58:42] cables as well as active optical cables
[58:46] um and so be a perfect way to be have a
[58:48] compliant PCI
[58:51] cxl active optical cable um what's being
[58:55] shown here is cdfp style one so it
[58:58] supports uh cxl 2.0 PCI gen
[59:02] 5x16 uh we have a proprietary side Bend
[59:06] uh over Optics uh protocol so we can
[59:10] transmit all of the needed sideband
[59:12] signals across the op optical cable um
[59:17] so using a multimode Fiers type of
[59:20] solution uh you know we're getting up to
[59:22] 30 MERS easily we've actually had some
[59:25] tests uh with 50 um 5050 M uh as
[59:30] well um and even and even 100 so we can
[59:35] we can really uh push this as needed the
[59:38] end and latency you're really uh limited
[59:41] by the the time of flight it's you know
[59:43] it's a linear cable so it's one nond
[59:45] plus time of flight and so therefore
[59:48] most of these applications are in the 10
[59:50] meter range uh because you're talking
[59:53] about you know they still want the low
[59:54] latency from the from the GPU to memory
[59:58] um I mentioned power is a big concern
[01:00:01] the customers don't want to pay the
[01:00:02] retim or power penalty so it's uh you
[01:00:05] know we're less than 12 Watts uh per
[01:00:07] connector so what we're interconnecting
[01:00:09] here is CPUs gpus
[01:00:12] switches um you know and obviously all
[01:00:15] of the the various endpoints the memory
[01:00:18] uh type of
[01:00:19] endpoints um we're actively working with
[01:00:23] the PCI Sig and the optical working
[01:00:25] group to Define additional um optical
[01:00:29] cable based standards and those include
[01:00:33] uh qsfp DD and osfp XD and uh we'll be
[01:00:38] right there um to help Advance the the
[01:00:43] state-of-the-art uh in
[01:00:45] standards so that's the end of my slides
[01:00:48] and i' be happy to take any
[01:00:50] questions all right thank you so much
[01:00:52] Ron uh the questions are coming through
[01:00:54] but before I get to them I want to ask
[01:00:56] you the optical question what can you do
[01:00:58] for others and what can others do for
[01:01:00] you there are lots of folks in the room
[01:01:01] that that I'm sure are eager to
[01:01:04] collaborate with you so they they need
[01:01:05] some your Insight to know what they can
[01:01:08] how they can help yeah how they can help
[01:01:11] is really I think get involved with um
[01:01:14] with the standards organizations and
[01:01:16] really help uh really help us break sort
[01:01:18] of the um the technology wall where you
[01:01:22] know look folks are looking for optical
[01:01:25] solution they're looking for okay how do
[01:01:27] we get Beyond copper so there's so
[01:01:30] there's many companies now offering
[01:01:33] Solutions and I would just encourage you
[01:01:35] to you know to join um there's a short
[01:01:39] reach Optics uh group within ocp I
[01:01:42] encourage you to join the the cxl uh
[01:01:45] Consortium and as well as the PCI
[01:01:47] Express Optical working group uh get
[01:01:50] involved and um really help uh you know
[01:01:54] help uh the the industry move
[01:01:58] forward um okay uh I have a question
[01:02:01] let's go to to my friend Fernando Gomez
[01:02:04] Fernando uh why don't you put yourself
[01:02:06] on uh unmute yourself and tell us where
[01:02:09] you are and uh ask your question please
[01:02:11] and we'd like to see your face as well
[01:02:13] if have your camera on well you can see
[01:02:15] my face but uh if you want to hear my
[01:02:18] voice uh sure you know the question is
[01:02:20] fairly simple uh PCI Sig has uh you know
[01:02:25] for the long time not really uh
[01:02:28] committed to an optical interface um
[01:02:31] other than just a pure e to O and O to e
[01:02:34] but with the uh I guess the emergence of
[01:02:37] shter Link Optical uh links inside the
[01:02:41] data center and of course with cxl as a
[01:02:44] competing um in a protocol if you will
[01:02:48] will a PCI Sig eventually adopt uh
[01:02:52] perhaps after Gen 6 will they adopt an
[01:02:55] optical interface and what are the
[01:02:57] implications for vendors and you know
[01:03:00] part of it you already answered that
[01:03:02] you're involved in the standards WR and
[01:03:04] you know and that that is so uh you know
[01:03:06] the direction that uh appears to be
[01:03:10] taking shape at uh in any event there
[01:03:12] what is your what's your view on that uh
[01:03:14] know that's a great question so for the
[01:03:16] last six months uh they they launched an
[01:03:18] optical working group uh to specifically
[01:03:21] address that and there will be an ecn to
[01:03:26] the Gen 6 specification to Define
[01:03:30] Optical Reimers and further to follow
[01:03:33] that they will be defining connector
[01:03:35] standards for optical cables so this
[01:03:40] this is coming and I think the what the
[01:03:42] what the Sig um at least what we've
[01:03:45] heard from the Sig is they've agreed
[01:03:46] that yeah it's it's going to be needed
[01:03:49] probably you know starting in Gen 6 but
[01:03:52] then more prevalent for Gen 7 so the
[01:03:54] solutions that
[01:03:56] uh we Define in the ecn and in the Gen
[01:03:59] uh six and seven specs um you know we'll
[01:04:02] need to we need to sort of pave the way
[01:04:05] for Gen seven gen
[01:04:08] 8 okay and what about the things like
[01:04:12] Link Training on state machine uh status
[01:04:15] and things like that are those things
[01:04:17] that would also appear in Optical form
[01:04:21] or will those remain in the yeah so what
[01:04:24] um now that'll generally remain in the
[01:04:27] in the root complex endpoint or switch
[01:04:29] but what they are defining is this uh
[01:04:32] notion of an optical retim where in in
[01:04:34] the case of a retim there is the Link
[01:04:37] Training State machine in the retim
[01:04:39] itself so they're basically enabling um
[01:04:42] or adding an ecn to to say okay now that
[01:04:46] we have an optical interface uh what
[01:04:49] changes do we need to make to that
[01:04:51] training State machine so that'll be
[01:04:54] part of the ecn that that you see from
[01:04:56] PCI Express that'll be released this
[01:04:58] year I see well well that's an add on to
[01:05:02] already existing electrical retim
[01:05:04] anyways right so correct all right uh
[01:05:08] let Let's uh I will bring a couple of
[01:05:10] more question the first one from uh my
[01:05:13] friend deac Chama from Trump deac why
[01:05:16] don't you unmute yourself and uh ask ask
[01:05:19] your
[01:05:19] question uh yes thank you uh just just
[01:05:23] simple question I guess this is an AOC
[01:05:25] and I wanted to understand if there were
[01:05:27] any special requirements for vixel for
[01:05:30] these
[01:05:32] aoc's s any special requirements for
[01:05:35] what did you say for wixel
[01:05:39] for for the laser wixel um yeah so no
[01:05:42] this is um this is a is a multimode
[01:05:45] application so simple you know uh vixel
[01:05:48] driver Tia type of implementation so
[01:05:52] anywhere you know and the reason this
[01:05:54] was chosen is it's it's a highly
[01:05:56] reliable uh technology it's basically
[01:05:59] proven in the ethernet uh market so no
[01:06:02] there's no specific uh requirements on
[01:06:05] the
[01:06:07] laser okay thank you all right thank you
[01:06:11] so much Ron um actually the next
[01:06:13] question is from our next speaker uh
[01:06:16] from rohen from light speed bonics rohen
[01:06:18] why don't you ask the question from Bron
[01:06:20] and then I know you also have a one
[01:06:22] slide then we can start can you can
[01:06:24] share that one as well the Flor is yours
[01:06:26] ran yeah I was originally thinking that
[01:06:29] this is silicon photonic solution now
[01:06:31] that you clarified it trixel based what
[01:06:33] wavelength is this um
[01:06:38] they um yeah so uh it this is based
[01:06:42] based on you know like standard as I
[01:06:44] said vixel technology um and we do have
[01:06:50] um uh single mode Solutions as well so
[01:06:54] this particular uh
[01:06:56] implementation uh we believe it we you
[01:06:58] know we're using the standard um you
[01:07:01] know
[01:07:03] 850 type of uh
[01:07:06] implementation that that's interesting
[01:07:09] thank
[01:07:10] you all right why rohim why don't you
[01:07:13] share your one slide and and tell us
[01:07:16] what you do and and what you can do for
[01:07:19] others and what others can do for
[01:07:23] you um hello uh good morning good
[01:07:26] evening everyone I'm rohin from light
[01:07:29] speed
[01:07:29] photonics uh at light speed essentially
[01:07:32] we are building linear Drive Optics as a
[01:07:35] near packaged or onboard Optics solution
[01:07:39] unlike the expensive socket based
[01:07:41] onboard optic Solutions or product light
[01:07:43] connect is built as a surface mount
[01:07:45] component so it can be soldered near the
[01:07:47] ASC avoiding the long transmission lines
[01:07:50] which are basically the ones uh that you
[01:07:53] want to avoid for um lots of signal uh
[01:07:57] designed for both free space and
[01:07:59] pluggable uh fiber compatibility
[01:08:02] multimode fiber for legacy compatibility
[01:08:04] light connect essentially is as bare
[01:08:06] bonds as it gets like a physical layer
[01:08:09] interconnect uh it has a driver
[01:08:11] TI and uh you're looking at with a wixel
[01:08:15] array and a photo detector array
[01:08:17] combined with a custom micro Optics
[01:08:18] design all of it is packaged in a 7 mm x
[01:08:22] 4 mm LGA package uh for short range
[01:08:26] within the server chipto chip
[01:08:27] Communications we colate the beam to
[01:08:29] sufficiently large diameter to get easy
[01:08:31] alignment of free space connectivity up
[01:08:34] to 200 mm uh and you can see the solder
[01:08:37] components on the test for on the left
[01:08:39] here and we are also building an easy to
[01:08:42] integrate multi mode fiber version of
[01:08:44] this that is compatible with the
[01:08:45] standard mtpm connectors and as some of
[01:08:49] the speakers spoke about it is built on
[01:08:51] proven reliable components uh
[01:08:53] essentially vixel and the drivers uh and
[01:08:55] simplifying the highspeed design for our
[01:08:58] customers providing up to 68% reduction
[01:09:01] in power consumption compared to the
[01:09:02] regular pluga bles and it is
[01:09:04] integratable as a surface mount standard
[01:09:06] assembly process and provides a protocol
[01:09:09] agnostic F layer connectivity to run any
[01:09:12] standard protocol based on lvds like PCI
[01:09:14] cxl or Standard Optical ethernet
[01:09:17] protocols we geting up for commercial
[01:09:19] deployments as early as next year and we
[01:09:21] have a road map for 100g per Channel and
[01:09:23] 200g per Channel as a as a development
[01:09:26] with uh with the partners who are
[01:09:28] supporting us uh please feel free to
[01:09:31] reach out if you find our Tech
[01:09:32] interesting for your applications it's
[01:09:34] extremely compact form fact form factor
[01:09:37] you can put it near the package or the
[01:09:39] board yeah that's that's from light stre
[01:09:43] thank you so much Ryan uh and just a
[01:09:44] reminder for everyone if you want to get
[01:09:46] connected to anyone in this in this
[01:09:48] meeting just feel free to reach to any
[01:09:50] of us at Optica and would be happy to
[01:09:52] make the
[01:09:53] introduction um now with that I would
[01:09:56] like to actually uh introduce our next
[01:09:58] speaker which uh which has an
[01:10:00] interesting uh technology for us it's
[01:10:02] newbies communication and the uh the
[01:10:04] speaker is son Le son the floor is all
[01:10:07] yours uh please please share slides and
[01:10:10] we are all ears to hear
[01:10:16] yours hello can you hear me no than
[01:10:19] clear yes um thank you very much let me
[01:10:21] share my
[01:10:23] slide okay
[01:10:26] okay so basally in the presentation
[01:10:29] first my name is son I'm from New
[01:10:31] communication and in Z presentation I
[01:10:34] would like to talk about some important
[01:10:36] Prospect of linear dry uptic for 100 Gig
[01:10:39] and 200 gig and here I would like to
[01:10:41] broaden the discussion a little bit to
[01:10:44] cover a various Implement implementation
[01:10:47] scenario for linear dry Optics including
[01:10:50] CPO NPO or lpo so first I want to
[01:10:54] express our view uh we believe that
[01:10:57] linear dry optic is a very elegant
[01:11:00] technology uh to increase IO density
[01:11:03] Beyond Tabit per second per millimeter
[01:11:06] and to reduce the power consumption
[01:11:08] reduce the latency reduce the cost while
[01:11:11] increasing the implementation
[01:11:13] flexibility for highspeed optical
[01:11:16] interconnect for Data Center and for
[01:11:18] especially for machine learning and AI
[01:11:21] application so what you see here is new
[01:11:24] H6 00 uh 16 by 100 or 1.6 tab per second
[01:11:30] fup capacity uh linear Optical engine in
[01:11:34] very small form factor and uh our unique
[01:11:38] vertical fiber array attach approach
[01:11:41] allow for the optical engine to be
[01:11:43] placed in multiple rows around the a to
[01:11:46] active terabit per second per millimeter
[01:11:49] of iio density with very low latency and
[01:11:53] low power consumption below six bom bit
[01:11:56] including the laser and in general the
[01:11:59] optical engine uh it designed for
[01:12:02] machine learning and a application but
[01:12:05] we also have the different form factor
[01:12:07] which uh allows it to be placed in the
[01:12:10] plugable module as well so in the
[01:12:13] presentation I would like to present
[01:12:14] some measurement result about the
[01:12:17] performance of the optical engine in the
[01:12:20] linear dry application at 100 Gig and
[01:12:24] I'm going to discard Al is scalability
[01:12:26] to
[01:12:27] tun so you see here the basic schematic
[01:12:30] of the experimental setup we use here a
[01:12:33] commercially available 51.2 terabit per
[01:12:36] second switch uh with 100 Gig hour cus
[01:12:40] and we connect uh we we use a horse
[01:12:43] compant board to connect the switch to
[01:12:46] our xc600 evaluation Bard and you can
[01:12:50] see here the channel including the
[01:12:52] package channel the PCB on the horse
[01:12:55] connector and through the uh hor
[01:12:57] compliant board and also through our I
[01:13:01] cable and our socket and the channel on
[01:13:03] our evaluation board and then the
[01:13:06] optical channel here is optically loop
[01:13:09] back okay so you see here two uh
[01:13:13] measurement set of pron the blue circle
[01:13:17] is a prefect P versus the channel loss
[01:13:20] from 10 to uh 22 DB uh when we switch
[01:13:26] off analog Equalization within the
[01:13:28] optical engine and the Red Square here
[01:13:33] uh curve here is when we switch on
[01:13:35] analog Equalization within the optical
[01:13:38] engine and you can see here that by
[01:13:41] switching on the analog Equalization uh
[01:13:43] we have a much higher better performance
[01:13:47] uh and this is our first generation of
[01:13:50] analog Equalization and for our
[01:13:53] production revision we have implemented
[01:13:56] a much uh better stronger analog
[01:13:58] Equalization with more boosting and we
[01:14:01] could achieve a much better performance
[01:14:03] more than other magnitude performance
[01:14:06] Improvement especially as a high Channel
[01:14:09] loss okay so just to remind you so for
[01:14:13] the uh popular UK Cas of lpo NPO or CPO
[01:14:18] we have typical performance expectation
[01:14:21] as one minus8 prefect PR to cover 20 DB
[01:14:25] loss 15 and 10 DB loss so now based on
[01:14:29] that typical requirement uh you can see
[01:14:32] that when we don't have analog
[01:14:34] Equalization within the optical engine
[01:14:37] uh we can only support the CPO but lpo
[01:14:41] and NPO applications seem to be
[01:14:43] difficult with our first generation of
[01:14:46] analog Equalization uh we can reach 17
[01:14:50] DB Plus at the br1 E min8 so we can
[01:14:53] support CPO NPO and most of lpo for our
[01:14:58] production version with what we call
[01:15:00] Advan analog Equalization we can reach
[01:15:03] 22 DB of loss at the one Min 8br so we
[01:15:06] can support all of the application and
[01:15:10] the cost of Advance analog Equalization
[01:15:12] actually quite small only 0.8 B per bit
[01:15:16] or 0.6 wat per 800 gig and it is
[01:15:20] significantly lower than a typical 7
[01:15:23] watt power consumption of 800 gig
[01:15:25] Optical re timer so the message here is
[01:15:28] that for 100 Gig strong Equalization
[01:15:31] analog Equalization enable on the
[01:15:34] application SC from CPO NPO to lpo and
[01:15:39] that it at 100 Gig so how does it scale
[01:15:42] to 200 gig so let's start with 100 Gig
[01:15:46] case we have here again the prefect
[01:15:48] versus the channel loss and for tun our
[01:15:51] analysis shows that uh we're going to
[01:15:54] have the uh 2 to3 DP SNR penalty and
[01:15:59] with a good implementation uh with uh in
[01:16:02] our case with the different design for
[01:16:04] driver Tia we believe that we could AC
[01:16:06] 2db penalty and then you end up with Z
[01:16:09] Cur here for 200 g so now for tun G now
[01:16:13] within the discussion within oif the buy
[01:16:15] is quite high for the lp application now
[01:16:19] we need to reduce uh the thres of the pr
[01:16:23] properly to one in minus 6 is the our
[01:16:25] assumption then we as the typical
[01:16:28] implementation for the channel loss can
[01:16:31] go up to 27 BB even higher but NPO CPO
[01:16:35] the channel loss could be reduced to 20
[01:16:37] to 50 DB so now 27 DB is very high so it
[01:16:42] means that it seem to be difficult to
[01:16:44] achieve the desirable performance at the
[01:16:46] channel loss but for CPO and NPO for 20
[01:16:50] DB loss uh it possible so now for if we
[01:16:55] want to scale the t g uh line direct DY
[01:16:58] for help your application we need to
[01:17:01] reduce the channel loss and it could be
[01:17:03] done through using either fly over cable
[01:17:06] or VC and with that we believe that the
[01:17:10] channel loss can be reduced to around 18
[01:17:12] DB and it possible so and we believe
[01:17:16] that it is a way to go so the message
[01:17:19] here is that strong
[01:17:21] Equalization uh in needed at tun gate
[01:17:24] and it going to be enable line direct
[01:17:27] dve at two for NP CP for lpo properly
[01:17:32] fly over C obiously going to be needed
[01:17:36] so here's my summary for the case of 100
[01:17:39] gate we believe that with uh analog
[01:17:42] Equalization uh with a welld designed
[01:17:46] analog Equalization we could support all
[01:17:48] application on the UK from CPO NP to lpo
[01:17:53] at T not get it's going to be challenged
[01:17:56] for CPO and NPO with advant Equalization
[01:17:58] is possible for lpo we need to reduce
[01:18:01] the channel loss uh using either VC or
[01:18:06] uh fly cable but the bottom line is
[01:18:09] linear direct dve up to scale to 200 g
[01:18:12] with Advan
[01:18:13] Equalization and with that I would like
[01:18:15] to thank all of you for your attention
[01:18:18] thank you very much for a great
[01:18:19] presentation song um my regards to all
[01:18:22] your colleagues you're doing a fantastic
[01:18:24] job at NIS communication we are really
[01:18:26] amazed what can you do for others what
[01:18:28] can others do for
[01:18:32] you what are your key challenges at this
[01:18:35] moment that companies can help you
[01:18:37] with oh so basically I think I agree
[01:18:39] with Andy that the big challenge for lpo
[01:18:42] Is to have the uh to have the test
[01:18:46] platform so the customer can test and uh
[01:18:49] and be confident about the the product
[01:18:52] and especially in op durability testing
[01:18:56] so having the uh the the procedure to be
[01:19:00] compli at uh tb2 is important and with
[01:19:04] that we also working on some uh Advan
[01:19:07] feature within the driver to any B tp2
[01:19:12] compli thank you very much yeah so we
[01:19:15] have a few questions for you in the room
[01:19:17] the first one is coming from Mark
[01:19:18] lucovich Mark what's in your mind yes hi
[01:19:22] um I wanted to ask you uh you know your
[01:19:24] most Valu valuable solution is the
[01:19:27] vertical line card which is going to
[01:19:29] enable U pluga bles indefinitely and and
[01:19:33] remove your discussion uh uh remove the
[01:19:35] discussion about CPO and N uh uh when
[01:19:39] are you guys going to start talking
[01:19:40] about that Solution that's your most
[01:19:42] valuable asset of VC actually we
[01:19:47] um yeah I'm not sure I'm the right
[01:19:49] person to answer the question but uh for
[01:19:51] sure VC is very important especially at
[01:19:53] tun the gig and and also we're working
[01:19:56] with uh a few partner to push it forward
[01:20:01] in the S of Licensing and stand up um uh
[01:20:06] designing uh plugable for that uh but
[01:20:11] yeah so it it uh it will come as the
[01:20:14] right time the owner can say so uh
[01:20:16] that's perfect at the right time is the
[01:20:17] perfect answer for Market the next
[01:20:19] question song is coming all the way from
[01:20:21] the UK from one via designer of
[01:20:23] electronics tell us what's on your
[01:20:28] mind hello uh just a quick question um
[01:20:32] so when you say unlock e EQ do you mean
[01:20:35] uh just a standard ctle uh with
[01:20:40] this uh yes thank you for the question
[01:20:43] so there's a few we have a few is a c
[01:20:47] plus picking and of course we have some
[01:20:49] special design in our so uh yes
[01:20:54] unfortunately I'm not an ICC designer so
[01:20:56] that's all I can
[01:20:58] say suas is an IC designer so please
[01:21:01] make sure you follow up afterwards the
[01:21:03] next question is coming all the way from
[01:21:04] CMC
[01:21:06] microsystem ammed all the way from
[01:21:08] Canada tell us what's in your mind uh
[01:21:10] yes it's actually about the because I
[01:21:13] think uh they mentioned the test like
[01:21:16] the customers to test their modules
[01:21:18] right like have you done any yield
[01:21:20] analysis right because when you are at
[01:21:22] the production like volume production
[01:21:25] right you you're supposed to have like
[01:21:27] yield analysis so customers know you
[01:21:29] know what's your yield what's the cost
[01:21:31] all that information is very important
[01:21:33] right rather than you know let the
[01:21:35] customer test right so what is what are
[01:21:39] the yield look like for this
[01:21:43] technology yes so so basically we have a
[01:21:47] good year for the Silicon photonic IC
[01:21:50] for analog IC the yield is very very
[01:21:53] high is on Sy and geranium the years is
[01:21:56] very very high and for the packaging and
[01:21:59] the fiber uh fiber cing so we're working
[01:22:03] to improve the yield but uh the yield
[01:22:05] going to be as good at the the yeari of
[01:22:08] cnic and photonic I so overall we are
[01:22:10] quite confident to deliver the product
[01:22:13] in the mass
[01:22:14] production yeah but sorry like sorry if
[01:22:17] I want to understand that so so you have
[01:22:20] the yield like you have a number and
[01:22:22] then you have the accumulative yield
[01:22:24] right right you have to be
[01:22:26] at like 99% right because you have many
[01:22:29] stages right and each stage will
[01:22:32] actually lower yield right so uh that is
[01:22:35] true but we have the few key component
[01:22:37] here we have the photon I see ACC in
[01:22:39] there is many component integrated in
[01:22:43] the single IC for the analog IC I can
[01:22:47] tell you that the year close to 100% of
[01:22:51] yes for the photonic I see then we are
[01:22:53] improving the VI and it it getting there
[01:22:57] and for our uh for our F attach and
[01:23:02] packaging and then y keep improving uh
[01:23:06] so we are improving the year every day
[01:23:08] okay okay thanks thank you very much
[01:23:10] soon there are three companies who want
[01:23:12] to introduce their technology to you and
[01:23:13] to the rest of the speaker we're going
[01:23:14] to start with one company who may help
[01:23:16] you improve the deal we're going to
[01:23:17] Keystone photonics Philip Philip thank
[01:23:20] you very much for being with us please
[01:23:21] share your screen and tell us what you
[01:23:23] can do for others and others can do for
[01:23:25] you yes um thank you very much for
[01:23:29] introducing me here so we have heard a
[01:23:31] couple of times that um that lpo puts a
[01:23:35] lot of um um heavy weight on on the
[01:23:39] optic the in the DP world yeah well
[01:23:42] things are rather soft but if you if you
[01:23:44] leave away the um DSP then then things
[01:23:46] get heavy in Optics and and we heard
[01:23:49] that a couple of times and he was
[01:23:50] reading that he doesn't want to be the
[01:23:52] company doing the optical qualification
[01:23:54] we heard Ed here NIS is moving the yield
[01:23:56] up but but I think um what it always has
[01:23:59] to be done if you move yield up or
[01:24:01] you're looking at yield tasting and that
[01:24:03] was that is what what Keystone photonics
[01:24:05] does we provide products for optical
[01:24:08] tasting so how we do this we have
[01:24:11] industry proven products for tested
[01:24:13] measurements we have the optical prop
[01:24:15] for that we take fiber arrays or other
[01:24:18] fonic components and we 3D print a micro
[01:24:21] optic structure on it we have extremely
[01:24:23] high position alignment we have very
[01:24:26] good beam shaping because it's created
[01:24:28] at the place where you need it with high
[01:24:31] resolution lithography we use machines
[01:24:33] from bangard Automation and with that we
[01:24:35] achieve super low variation now how that
[01:24:38] can be used it's used for wer level
[01:24:40] testing mostly and modu level testing so
[01:24:43] we're capable of creating PR for for uh
[01:24:47] proving through grating couplers or uh
[01:24:50] we can create probes that go into edges
[01:24:52] or Wafers and that allow to qualify your
[01:24:55] components at wafer level we can do the
[01:24:58] same thing also on module or assembly
[01:25:00] level for testing Optical lenses we can
[01:25:03] tune the mode field of 3D printed fiber
[01:25:06] arrays and mirror lens fiber arrays to
[01:25:09] anything between two and about 40
[01:25:11] micrometer and that really allows you to
[01:25:13] test the optical components in a
[01:25:15] manufacturing settings so we're in
[01:25:17] manufacturing in many silicon photonic
[01:25:20] faps now we collaborate with most of the
[01:25:22] photonic um pasting company so we're
[01:25:25] ready um to to support
[01:25:27] you thank you thank you very much second
[01:25:31] question who wants to also help you here
[01:25:33] song and the rest of the speakers is the
[01:25:35] company that is is actually sponsoring
[01:25:38] the event today thank you very much a
[01:25:41] aoto for being with us today nakar the
[01:25:44] Flor orientation of everyone is
[01:25:47] yours okay thank you so much can I hear
[01:25:50] my voice crystal clear okay thank you
[01:25:54] uh just I will introduce my company
[01:25:56] profile uh and our
[01:25:59] materials so can you see my slide yes
[01:26:02] but please go presentation mode I
[01:26:09] okay contrl
[01:26:13] L
[01:26:14] sorry yes it's good
[01:26:17] enough yes okay so uh let me introduce
[01:26:20] our company profile thank you very much
[01:26:22] for opportunity so my name is
[01:26:25] from aoto aot is known as a food
[01:26:29] industry company and also we supp
[01:26:32] electronic materials such as called a
[01:26:36] build up fil called APF APF is
[01:26:40] initiation materials used
[01:26:45] between now is one Theo standard
[01:26:48] materials for Semiconductor
[01:26:50] manufacturers so just now uh we have S
[01:26:54] line up for Co package ofes so one is
[01:26:58] polymer based Optical webite mat called
[01:27:01] K series just we suppli sever materials
[01:27:06] based on photography Technologies and
[01:27:09] just we
[01:27:15] can material with a single mode Optical
[01:27:19] transmission and also another is
[01:27:22] adhesive material called set such apply
[01:27:26] for adhesion for fiber and ADH fiber
[01:27:32] optical fiber so just please if you
[01:27:35] interested to our materal please please
[01:27:38] contact
[01:27:44] this thank you so much thank you very
[01:27:47] much aoto being one of the sponsors
[01:27:50] today and the third company s that
[01:27:53] wanted also to speak after you is a
[01:27:55] company means a lot to Optica we're
[01:27:57] going to one of the leading suppliers of
[01:27:59] design suits in the Optica membership
[01:28:01] optic wave thank you very much for being
[01:28:04] with us today tell us tell us what's on
[01:28:05] your mind and tell us how others can
[01:28:07] help you and you can help
[01:28:09] others sure please go presentation mode
[01:28:12] thanks Jose hello everyone so yeah I'm
[01:28:16] just I'm Ahmed I'm VB at optic Optical
[01:28:19] system at optiwave optiwave is a a
[01:28:23] software company
[01:28:24] um we are established like about 30
[01:28:26] years ago uh so we are a long time in
[01:28:29] the business of uh designing software
[01:28:31] packages used for Designing photonic
[01:28:34] components and systems uh we have as you
[01:28:37] see here eight packages of software op
[01:28:39] the system of the spies which is and opt
[01:28:41] fddt these three software can be used
[01:28:44] for Designing the transceivers basically
[01:28:47] and um uh we have uh in an optic system
[01:28:51] software over 600 components with and
[01:28:53] visualizer which can be used for
[01:28:55] simulating the different devices uh and
[01:28:58] parts in in the transceiver uh we have
[01:29:02] uh uh capability to operate our software
[01:29:04] with external software like bython or
[01:29:06] mlab or cab and also we have a a huge
[01:29:11] library of examples uh which can enable
[01:29:13] the users to kind of start uh his
[01:29:17] simulations for the different uh uh
[01:29:20] system levels or device levels uh if you
[01:29:23] are interested in downloading our
[01:29:25] software you can go to optiweb website
[01:29:27] and you you may download 30 days of free
[01:29:30] trial of any of these software packages
[01:29:33] we'll be very happy to to work with with
[01:29:36] you on setting up your design and and
[01:29:38] maybe uh if you need any any help and in
[01:29:41] simulating your your transceivers thank
[01:29:44] you thank you very much Ahmed for all
[01:29:46] the great support that optic wave has
[01:29:48] been given to Optica it is time to give
[01:29:51] the floor to our closing keynote speaker
[01:29:53] I go to a company that many of you
[01:29:55] visited during ofc because they had an
[01:29:57] lpo demo for everybody to touch and they
[01:29:59] actually working with oif on a
[01:30:01] standardizing the interconnection I'm
[01:30:03] going to give the floor to Tiger from
[01:30:06] Axel link the floor and the attention of
[01:30:08] everyone goes to you you are all closer
[01:30:11] enjoy right thank you um Jose and
[01:30:14] everybody um um so aink is manufacturing
[01:30:17] and also development uh LP on a modules
[01:30:20] right now so I'm going to share a little
[01:30:22] bit of uh results from our very
[01:30:26] preliminary uh test and experiments and
[01:30:28] simulation uh to validate some of the
[01:30:31] points that andd brought up but before
[01:30:34] jumping into Data uh here are some
[01:30:36] summary of the lpo benefits uh lower
[01:30:39] power lower latency module cost and
[01:30:41] reliability and today I'm also going to
[01:30:43] touch based on the heat generate from
[01:30:46] the module and also overall switch
[01:30:48] system that will also impact to power
[01:30:51] and
[01:30:52] reliability so this is the uh demo
[01:30:56] summary from ofc that Jose just
[01:30:58] introduced uh we had the uh 40 uh 64x
[01:31:03] 800 gig switch from vron having our 800
[01:31:06] gig dr8 and 2x F4 modules uh
[01:31:10] demonstrating the atp2 I and also be uh
[01:31:14] just to clarify these modules are based
[01:31:16] on silicon photonics and we were
[01:31:19] comparing with lpo lro and DSP based
[01:31:22] modules and all of these working pretty
[01:31:24] good uh and also we had the line coming
[01:31:28] from oif boost uh that's linked from the
[01:31:32] their n NPO uh signal and we are
[01:31:35] receiving pretty good way as well and
[01:31:39] furthermore we did uh continue our
[01:31:41] experiment by uh doing some simulation
[01:31:45] and also experiment fitting up all the
[01:31:47] 64 ports uh with DSP and lpo modules and
[01:31:53] the uh to manipulate the results we uh
[01:31:57] fix the fan speeds and this particular
[01:32:00] uh data is showing the speed at 75% and
[01:32:03] the xaxis is showing this uh switch Port
[01:32:06] counts uh and also total switch is y
[01:32:09] axis uh just because we fixed the fan
[01:32:12] speed uh at 75% we only saw 20% part
[01:32:16] difference however this is going to
[01:32:19] relate to the next slide of uh just to
[01:32:22] CL for this is a dr4 Optics there are
[01:32:25] eight Optics um left hand side is
[01:32:28] another uh chart Matrix of the xaxis we
[01:32:32] have a switch overall power consumption
[01:32:34] at 64 channels uh with the DSP and lpo
[01:32:38] running and the y- axis we have the
[01:32:41] switch temperature and we plotted uh the
[01:32:45] data from when we measured the overall
[01:32:48] par consumption on temperature at the
[01:32:50] fan speed of 50% 75% and 100% and of
[01:32:54] course um the difference uh is
[01:32:57] relatively similar to the previous
[01:32:59] result 20% however you can see the
[01:33:02] significant uh lower temperature uh with
[01:33:05] the lpo and the of course this is a
[01:33:08] temperature pile Matrix it cannot be
[01:33:10] linear but if you uh have the linear
[01:33:12] line like this on the left hand side uh
[01:33:15] we can expect to uh reduce the fan speed
[01:33:19] uh probably that will contribute to good
[01:33:23] uh amount of the power reduction and uh
[01:33:25] I'm pretty much supportive of what Andy
[01:33:27] said 40% of the system power consumption
[01:33:30] reduction coming from not only the power
[01:33:32] consumption at the lpo but the uh
[01:33:35] cooling system is going to be more
[01:33:37] efficient and relating to this
[01:33:39] temperature topic uh make gave a
[01:33:41] presentation at ofc uh on the
[01:33:44] reliability and presented that the lpo
[01:33:47] compared to tsp based module has a three
[01:33:50] times better
[01:33:51] reliability uh coming from the uh heat
[01:33:55] uh so uh we're hoping to further invest
[01:33:59] uh experiment and uh develop our lpo
[01:34:01] module to elaborate these features and
[01:34:06] and lastly uh this is the average module
[01:34:09] Park
[01:34:10] consumption uh of lpo lro DSP these
[01:34:13] numbers are matching uh what Andy
[01:34:17] presented um I didn't have too much
[01:34:19] about lro but we working on it uh but uh
[01:34:23] like Andy mentioned LP MSA having uh
[01:34:26] industry leaders uh working on the
[01:34:29] specifications uh probably coming around
[01:34:31] the Q3 this year so hoping to have more
[01:34:34] data uh complying to MSA specification
[01:34:37] later on thank you very much thank you
[01:34:40] very much for a great presentation and
[01:34:42] thank you for the collaboration that you
[01:34:44] are doing right now for entire benefit
[01:34:46] of the industry tiger what can you do
[01:34:49] for others and what others can do for
[01:34:50] you sure um like uh I think as I
[01:34:54] mentioned but we we we' like to have
[01:34:55] more Partners to verify modules but they
[01:34:58] also to further uh have the efficient
[01:35:02] modules we probably would like to hear
[01:35:05] what are other um technology enablers to
[01:35:08] lower the power consumption so we like
[01:35:11] to hear from components manufacturer to
[01:35:14] uh further enhance our uh modules I
[01:35:18] would like to remind everyone that this
[01:35:20] was an online production the meeting
[01:35:23] will be
[01:35:24] on YouTube for everyone to follow I'm
[01:35:26] going to give the floor to a closing
[01:35:28] question before we close the meeting and
[01:35:30] that has the honor to be winred n from
[01:35:32] focus light winred ask away for Tiger
[01:35:35] from aeling hello tiger good to see you
[01:35:39] again um so it's very interesting when
[01:35:43] I'm looking at your presentation I like
[01:35:44] to know so you provide now an lpo model
[01:35:47] so is that more a pull from industry
[01:35:50] from companies uh like companies to depl
[01:35:53] employer products or or do you see also
[01:35:57] from a production point of view it makes
[01:35:58] more sense to go into the lpo direction
[01:36:00] so you like to push it a little bit more
[01:36:03] so is it is the way pull push is it more
[01:36:05] on the push side or more on the pull
[01:36:07] side how do you see that th this is my
[01:36:09] personal opinion not accident any
[01:36:12] represent anybody but the um hearing
[01:36:14] from the very uh broader view of power
[01:36:18] consumption to app AI cluster and
[01:36:21] machine learning things uh do we have
[01:36:23] enough power to support in general like
[01:36:27] uh uh like a data center uh especially
[01:36:30] Japan is having data centers hypers
[01:36:33] skaters or big ones are in one one area
[01:36:36] and if we are adding these uh uh new
[01:36:40] features to the data center I don't
[01:36:42] think power generation is enough in in
[01:36:45] the area so it's I think personally P
[01:36:48] technology to enable or the to have our
[01:36:52] life better I think thank you thank you
[01:36:55] and thank you everyone it is 5:30 p.m.
[01:36:58] in the Netherlands 11:30 a.m. in
[01:37:00] Washington DC at 8:30 a.m. in California
[01:37:04] this meeting was Global and produced by
[01:37:06] Optica I would like to thank all of you
[01:37:07] for the great presentations remember
[01:37:09] everyone who has been registered will
[01:37:10] receive an email asking them to join
[01:37:13] Optica as a corporate member please do
[01:37:15] so we need your support if we like what
[01:37:17] we are doing if you like what we are
[01:37:18] doing here the only way for us to keep
[01:37:20] our jobs is to become corporate member
[01:37:22] of optic we will be sending you our CVS
[01:37:26] to work for a corporate member thank you
[01:37:27] very much all of you for a great event
[01:37:29] and thank you Andy beos for being with
[01:37:32] us today and setting up the scene myself
[01:37:35] I'm a Believer on lpo you can quote me
[01:37:37] on that they could get each other wash
[01:37:40] your hands no it's not pandemic anymore
[01:37:42] enjoy the relations enjoy and see you
[01:37:45] very soon all the best byebye
[01:37:50] [Music]