Full Transcript
https://www.youtube.com/watch?v=ge-ydSnJ8M0
[00:02] roads
[00:12] to move the speed of light.
[00:24] But every
[00:30] countip
[00:43] [music]
[00:54] Not human
[00:58] perons
[01:10] and build on time.
[01:10] Bring the light.
[01:23] Can change the market,
[01:26] Can change the market, but only
[01:29] not one by one in clean rooms,
[01:33] not one by one in clean rooms, but fast, precise, and thick.
[01:37] but fast, precise, and thick.
[01:37] Fiber rays must find their place.
[01:41] Fiber rays must find their place.
[01:42] Oh,
[01:42] lasers locked in tight.
[01:45] lasers locked in tight.
[01:45] No good before the build.
[01:50] No good before the build.
[01:50] Thermals held just right.
[01:53] Thermals held just right.
[01:53] Danger to come shouts the loudest, but
[01:56] Danger to come shouts the loudest, but it is not alone.
[01:58] it is not alone.
[01:58] Quantum sensing light, our defense need packaging of their own.
[02:01] light, our defense need packaging of their own.
[02:01] How do we package photons?
[02:04] their own.
[02:04] How do we package photons?
[02:06] photons?
[02:06] That's the volume AI demands.
[02:11] That's the volume AI demands.
[02:14] How do we scale the light with machines,
[02:15] not human hands?
[02:19] not human hands?
[02:21] Cost per gigabit, cost per port.
[02:25] Yield must rise.
[02:28] All the road map falls short.
[02:28] How do we package photons and build the future on time?
[02:34] Bring the light to the package.
[02:38] BRING THE PACKAGE TO the light.
[02:40] BRING THE PACKAGE TO the light.
[02:43] Hey, foundaries and equipment on the floor.
[02:46] System teams are waiting.
[02:49] They need more more alignment.
[02:51] Automated interfaces agreed.
[02:54] Wait for level testing at industrial speed.
[02:57] Not a demo, not a slide, not a promise.
[03:00] Still untried.
[03:02] PORTSMANSHIP SURVIVE.
[03:05] AI factories need light.
[03:07] AI factories need light.
[03:10] AI factories need light.
[03:14] How do we package photons?
[03:18] How do we package photons AT THE VOLUME AI DEMANDS?
[03:22] AT THE VOLUME AI DEMANDS? HOW DO WE SCALE THE light with machines,
[03:27] HOW DO WE SCALE THE light with machines, not human hands?
[03:30] not human hands?
[03:33] Cost per gigabit, cost per port, yield must rise
[03:36] cost per port, yield must rise
[03:40] or the road map falls short. How do we package photons and build the future on time?
[03:44] package photons and build the future on time?
[03:47] Bring the light to the package.
[03:49] package.
[03:52] BRING THE PACKAGE TO the line in just a few minutes.
[03:55] just a few minutes.
[03:58] Optica brings the industry to discuss the real test.
[03:59] to discuss the real test.
[04:02] Please stand by.
[04:05] Please stand by.
[04:10] I package the photons.
[04:14] Scale the light. Build it for volume.
[04:18] Build it for volume.
[04:18] Build it right.
[04:40] Welcome to a packed Optica online industry meeting.
[04:43] Today we're asking one direct question.
[04:46] How do we package photons at the volume AI now demands?
[04:53] AI infrastructure has a packaging problem.
[04:56] Hypers scale GPU clusters need optical IO on the package itself, not at the face plate.
[05:03] Nvidia's Quantum X and Spectrum X switches are early proof that the industry has accepted this direction.
[05:10] But shipping an announcement is not the same as populating an AI factory with
[05:19] Same as populating an AI factory with hundreds of thousands of optical ports.
[05:22] Hundreds of thousands of optical ports ready in time for the next GPU launch.
[05:25] Ready in time for the next GPU launch.
[05:28] Phutonix packaging has historically been slow, low volume, and handcrafted.
[05:32] Slow, low volume, and handcrafted.
[05:36] GPU deployment schedules are not.
[05:39] That gap is what we're here to close.
[05:42] Packaging is also the dominant cost in a photonic integrated circuit.
[05:45] Assembly, fiber attach, laser integration, and test account for the majority of a module cost.
[05:53] Module cost.
[05:55] Until packaging industrializes with wafer level processes, automated alignment and known good die testing, picks cannot cross from niche to volume market.
[06:06] And the same bottleneck is blocking multiple industries simultaneously.
[06:11] Datacom is the loudest voice, but quantum biomedical sensing, lidar, and defense face identical constraints.
[06:21] defense face identical constraints.
[06:21] Solving this once unlocks all of them.
[06:25] Solving this once unlocks all of them.
[06:27] In the next 90 minutes, we'll work through the specific problems.
[06:30] And to do that, [music] we've brought together two groups of experts.
[06:32] On the technology integrator side, Nokia, Nvidia, Jable, Photonix, and Light Matter.
[06:35] On the technology enabler side, Photon Bridge, Epiphany Design, Vera Leap, Fix Photonix Assembly, Fine and Cbrant.
[06:58] This meeting is free to attend made possible by our sponsors Fix Photonix Assembly, Ligantech, Vanguard [music] Automation, Nanoscribe, Fine, and Photon Bridge.
[07:14] So, let's get started.
[07:19] Hello everyone and welcome welcome to a
[07:21] Hello everyone and welcome welcome to a very special session.
[07:24] We have been very special session.
[07:24] We have been working very hard over the last months and we have seen record numbers on registrations because we are talking of perhaps the biggest challenge that the photonic industry is facing.
[07:34] Many of you were with us at OFC and you saw the huge boom in demand driven by AI and we saw the need to change the way that we have been dealing with photonic integrated circuits.
[07:49] It's all about advanced packaging.
[07:50] It's all about volume testing.
[07:50] is all about the scaling up and today we have a fantastic 90 minutes of your time to make sure that during the next 90 minutes you do business.
[08:00] Today as very well the interviewer said we are sponsored by Fix Lion Techch Vangar Automation Micronic Nanoscribe and Fine Techch.
[08:07] Thank you very much for making this meeting free for everyone in the room.
[08:12] everyone watching in the YouTube YouTube and everyone accessing this video on demand for as long as we leave it there and we have no intention of
[08:21] it there and we have no intention of ever taking it out.
[08:23] We also would like to remind that we do have two groups of companies Nokia Nvidia Matter and J
[08:27] Photonix came here with a shopping list.
[08:30] They came here because they need the rest of the companies in the room.
[08:34] And those rest of the companies in the room are everyone with us including
[08:37] Accelerant Fix, Fine Techch, Verif and Epiphany Design.
[08:44] Thank you very much for being with us.
[08:46] But everyone in the room is welcome to answer the needs of Nokia, Nvidia Light Matter and Jable Photonix.
[08:51] We do this because we love our corporate members.
[08:54] If you are not yet a corporate member, what are you waiting for?
[08:59] Optical.org/join /joinccorporate there you can find all the amazing membership benefits but the most amazing membership benefit is being friends with Dr. John Pew who put together this fantastic agenda.
[09:10] This is only part or agenda of online industry meetings.
[09:13] In a couple of weeks we have high volume so we have photonics for energy.
[09:18] Then we have 3D sensing.
[09:21] We're going to address topics like of
[09:22] going to address topics like of themology, applied imaging, aerospace,
[09:24] themology, applied imaging, aerospace, quantum safe networks, extended reality,
[09:26] quantum safe networks, extended reality, new space, quantum computing and CPO.
[09:29] new space, quantum computing and CPO.
[09:30] If you are interested in any of these topics, make sure you register as soon as possible in optica.org/industry.
[09:36] Please register as soon as possible.
[09:37] So we make sure that we answer and we draft the agenda of the meeting facing your needs.
[09:42] And when you want to meet us face to face in a couple of weeks only,
[09:47] second and third of June, we are all all of us going to Brussels for the Global Advanced Manufacturing Alliance, the GMA yearly event.
[09:59] We have a fantastic group of people there, including one of our keynotes today.
[10:02] Leon from Nvidia will be one of the opening keynotes at the GMA meeting.
[10:05] If you want to meet him in person, and I do, go to Brussels on the 2nd and 3rd of June.
[10:10] Also on Quantum 16th, 17th of June, we go to Glasgow together with Quantum 2.0 of Optica.
[10:15] The Quantum Industry Summit is looking to attract the CEOs of the quantum companies and see how we can make the supply chain resilient with Photonix
[10:23] supply chain resilient with Photonix technology.
[10:23] Do not miss this one.
[10:26] technology.
[10:26] Do not miss this one.
[10:26] Also, everyone is waiting for this one.
[10:28] The PE summit.
[10:30] What happens in the Uniper Dome stays at the Uniper Dome.
[10:32] will discuss with the key companies in AIdriven networks how we can bring photonix to satisfy their needs.
[10:38] Do not miss the PEX summit in October.
[10:40] And last but definitely not least in December we go to Paris to talk about perhaps the biggest upcoming market for photonics in Europe defense.
[10:44] We're going to address this with some of the tier ones with some of the primes of defense system industry in Europe and of course our flagship event.
[10:51] Where do we go every year?
[10:53] We go to Malaga 24 25 September the Global Photonics Economic Forum.
[10:55] We already published a list of keynotes.
[10:57] I think you are all amazed.
[11:00] No spoilers here.
[11:04] Globalonics economic forum.org.
[11:06] Go there and see the fantastic list of keynotes we have just published.
[11:08] But I would like to remind everyone that today is all about advanced packaging and volume production of photonic integrated circuits.
[11:11] So this meeting is also live
[11:24] circuits.
[11:24] So this meeting is also live streaming YouTube.
[11:26] So let me say hello to the YouTubers in the world.
[11:27] Hello YouTubers in the world.
[11:29] If you have any questions or comments, write them in the chat and I will read it in the Zoom room.
[11:34] And of course, this is also valid for the people here with me.
[11:37] 159 people here with me in the Zoom room.
[11:39] If you have any comments, questions, challenges, please write in the chat.
[11:45] We'll make sure that you voice the question out loud so the other 159 people in the room can hear you.
[11:51] Thank you very much all of you for this fantastic patience that you have for this introduction.
[11:57] Jump.
[11:57] what's going to happen in the room over the next 90 minutes.
[12:00] >> Uh thank you so much Jose and hello everyone.
[12:04] Uh I'm particularly excited today about hearing how the industry is moving beyond thinking deeply about individual devices and starting to solve that real manufacturing and integration challenge needed to bring Phutonix to scale.
[12:17] One of the crucial ways that's happening is by using that experience from the semiconductor industry and incorporating the experience of those who have been there and done it before.
[12:25] who have been there and done it before.
[12:27] The challenges are of course different uh but it's clear to me that that experience translates.
[12:32] So as Jose said this is very interactive meeting.
[12:33] Post your questions in the chat.
[12:35] We'll welcome you to unmute and ask your questions directly.
[12:39] So let's get straight into it uh with our first speaker and for the first time at an optica meeting with Nokia Armagan Isagi director of AI data center networking.
[12:49] Archan works at the intersection of AI infrastructure, high performance networking and next generation data center architectures.
[12:58] And today she'll share Nokia's perspective on how AI is reshaping networking requirements and where phonics and advanced interconnects will be critical to future scale.
[13:07] Armagan please share your slides.
[13:10] Uh the floor and the attention of everyone is yours.
[13:19] Thank you.
[13:22] Okay. Um, great. Um, just make sure everything is okay on the on the slide.
[13:25] Okay. Perfect. All
[13:25] Okay.
[13:25] Perfect.
[13:25] All fun.
[13:26] Fun.
[13:26] Uh, great.
[13:26] Thanks for having me.
[13:29] I'm Armagan Shaggy from Nokia and uh, today I'll focus on the key challenges of photonic packaging and integration and what we need to make photonic integrated circuits economically viable at AI scale.
[13:46] Well, uh there is a lot to talk about, but given the time constraint, I'll focus only on the most fundamental aspects of packaging and integration.
[13:56] Uh from a Nokia perspective, this this topic is is very timely as we run both uh silicon photonix line and um Indian phosphite fab and we build coherent as well as IMDD modules uh components for intra and interdcs with uh diverse capabilities including uh monolithic interior pick with integrated lasers uh multichip platform hybrid integration advanced packaging and manufacturing and any advancement.
[14:25] in in packaging integration techniques
[14:27] in in packaging integration techniques enabled by partners ecosystem will
[14:30] enabled by partners ecosystem will accelerate developments um and reduce
[14:33] accelerate developments um and reduce the time to market for us.
[14:36] Um so let me start with a the premise that the
[14:39] start with a the premise that the industry is acknowledging uh today that
[14:41] industry is acknowledging uh today that AI infrastructure uh is becoming
[14:44] AI infrastructure uh is becoming increasingly interconnect limited.
[14:47] Uh introducing different variants of
[14:48] introducing different variants of optical interconnects has been a way
[14:50] optical interconnects has been a way really to address this challenge.
[14:54] Um so the direction is is pretty clear but
[14:57] the direction is is pretty clear but there are several challenges that slow
[14:59] there are several challenges that slow down this this transition and we're
[15:02] down this this transition and we're still far from that scaling point.
[15:05] So over years uh we've made significant
[15:07] over years uh we've made significant progress in photonic integrated circuits
[15:09] progress in photonic integrated circuits design.
[15:12] Um but the packaging tax which is the cost and energy required to
[15:14] is the cost and energy required to couple the light off the chip into fiber
[15:16] couple the light off the chip into fiber still remains too high.
[15:19] And by some estimates, packaging accounts for more
[15:21] estimates, packaging accounts for more than 80% of the total module cost.
[15:24] Um, and it gets even harder in in in
[15:26] and it gets even harder in in in co-ackaged optics where you can have
[15:29] co-packaged optics where you can have multiple photonic suppliers in one.
[15:31] multiple photonic suppliers in one package.
[15:33] You have massive fiber counts and the huge uh yield throughput and cost pressure um at a scale that the industry has never designed to handle.
[15:43] So the real objective here is really to move futonic packaging away from the bespoke processes towards a foundry aligned high volume manufacturing model.
[15:53] and uh my focus today is obviously AI networking but with but the same packaging gaps show up in in sensing quantum biomedical photonics and so fixing them really lifts the whole field.
[16:05] Um the first challenge that I want to talk about is uh the fiber to to chip attach at scale.
[16:14] Um so today most fiber attach still relies on an active alignment as powering a device optimizing for pick signal and fixing it in place with epoxy.
[16:23] This requires submicron precision and it's done one device at a time.
[16:29] This is a slow um expensive and incompatible with AI
[16:31] expensive and incompatible with AI network volumes and is specifically
[16:33] network volumes and is specifically challenging for FA user or fiber array
[16:36] challenging for FA user or fiber array units. Uh the alignment and the UV cure
[16:39] units. Uh the alignment and the UV cure time seems to be dominant here. So there
[16:42] time seems to be dominant here. So there there are several techniques that are
[16:44] there are several techniques that are being developed and studied and each of
[16:47] being developed and studied and each of them has their own pros and cons and
[16:50] them has their own pros and cons and what's still missing is really the high
[16:52] what's still missing is really the high throughput uh passive alignment to guide
[16:55] throughput uh passive alignment to guide the optical fibers into place without
[16:57] the optical fibers into place without needing the device powered to find the
[16:59] needing the device powered to find the signal. Um and there is a deeper issue
[17:01] signal. Um and there is a deeper issue here that nothing is standardized.
[17:05] here that nothing is standardized. Today's flows remain uh tightly coupled
[17:08] Today's flows remain uh tightly coupled to foundry processes stack foundry
[17:11] to foundry processes stack foundry specific optical IO and then custom
[17:13] specific optical IO and then custom recipes. So every foundry or process
[17:16] recipes. So every foundry or process change uh triggers a new fiber attached
[17:19] change uh triggers a new fiber attached development and tooling qualification
[17:22] development and tooling qualification yield learning um also they don't
[17:24] yield learning um also they don't transfer between the different players.
[17:27] transfer between the different players. So until the industry converges on a
[17:29] So until the industry converges on a small number of reference flows, the
[17:31] small number of reference flows, the cost reduction curve that AI
[17:33] cost reduction curve that AI infrastructure demands um it really won't materialize.
[17:38] won't materialize.
[17:38] The second gap um emerges here as we move uh from uh pluggables to co-ackage in in near packaged optics uh in order to achieve uh the the required power efficiency and latency that photonix um demands is this is now placed extremely close to high power A6 and and GPUs.
[18:01] demands is this is now placed extremely close to high power A6 and and GPUs.
[18:04] Um this is a very challenging thermal environment as we all know.
[18:06] this is a very challenging thermal environment as we all know.
[18:06] So one common response is to use external lasers and feed light in by fiber.
[18:10] Uh that has clear benefits in thermal management and serviceability.
[18:14] that has clear benefits in thermal management and serviceability.
[18:16] But uh it's not the only viable architecture.
[18:19] it's not the only viable architecture.
[18:19] Um integrated laser as we know um has implemented uh in a number of Nokia's products offer higher reliability and enable use of lower power lasers or
[18:21] Um integrated laser as we know um has implemented uh in a number of Nokia's products offer higher reliability and enable use of lower power lasers or
[18:24] implemented uh in a number of Nokia's products offer higher reliability and enable use of lower power lasers or
[18:26] products offer higher reliability and enable use of lower power lasers or
[18:33] enable use of lower power lasers or fewer high power lasers uh with reduced fewer high power lasers uh with reduced uh uh blast radius.
[18:39] Um and this is uh uh blast radius.
[18:42] Um and this is important considering that the laser shortage that the industry is is facing.
[18:44] shortage that the industry is is facing.
[18:47] But uh there is a condition here that the laser should stay within their operating temperature window.
[18:49] the laser should stay within their operating temperature window.
[18:51] operating temperature window.
[18:54] So um like in my opinion what is missing here is that uh like we need a true system level code design thermal mechanical optical and electrical they need to be engineered together rather than in in in isolation.
[18:58] here is that uh like we need a true system level code design thermal mechanical optical and electrical they
[19:00] mechanical optical and electrical they need to be engineered together rather than in in in isolation.
[19:03] need to be engineered together rather than in in in isolation.
[19:05] than in in in isolation.
[19:07] So today these are often handled in in separate silos and the final design reflects a set of compromises.
[19:09] are often handled in in separate silos and the final design reflects a set of compromises.
[19:12] and the final design reflects a set of compromises.
[19:14] Uh and this is where the emerging substrate technologies such as glass interposers uh become very interesting.
[19:17] emerging substrate technologies such as glass interposers uh become very interesting.
[19:19] glass interposers uh become very interesting.
[19:22] Unlike silicon um glass is a strong thermal insulator acting as a barrier uh between the hot ASIC and temperature sensitive photonics.
[19:25] a strong thermal insulator acting as a barrier uh between the hot ASIC and temperature sensitive photonics.
[19:27] barrier uh between the hot ASIC and temperature sensitive photonics.
[19:30] U it's it's also electrically insulating which improves high frequency signal integrity
[19:32] it's also electrically insulating which improves high frequency signal integrity
[19:35] improves high frequency signal integrity and it's a coefficient of thermal.
[19:37] and it's a coefficient of thermal expansion can be tuned reducing.
[19:39] expansion can be tuned reducing mechanical stress and improving uh.
[19:41] mechanical stress and improving uh long-term reliability.
[19:43] So glass also supports advanced packaging features.
[19:46] supports advanced packaging features like ultra fine line RDL and better.
[19:48] like ultra fine line RDL and better optical waveguides and 3D integration.
[19:51] optical waveguides and 3D integration.
[19:53] So it really opens up uh design space that conventional approaches cannot.
[19:55] that conventional approaches cannot reach.
[19:58] reach.
[19:59] Uh there is a third gap as well which is the test and this is quite important in.
[20:02] the test and this is quite important in electronics.
[20:06] electronics. Uh the principle is very clear we don't package a faulty die.
[20:09] So no good die or KGD is is a foundational in photonics that principle is not fully.
[20:13] in photonics that principle is not fully and efficiently realized yet also we.
[20:16] and efficiently realized yet also we often don't know the full optical performance until late in the packaging.
[20:19] often don't know the full optical performance until late in the packaging.
[20:21] performance until late in the packaging.
[20:24] and wafer uh die level performance can be different from the package module and.
[20:27] and wafer uh die level performance can be different from the package module and.
[20:30] and this is because really the packaging itself uh fiber attached thermal mounting and and mechanical stress they
[20:37] mounting and and mechanical stress they all shift the optical behavior and
[20:40] all shift the optical behavior and that's a serious uh yield risk
[20:42] that's a serious uh yield risk specifically when photonic dies are
[20:44] specifically when photonic dies are co-ackaged with expensive AI6 uh that
[20:47] co-ackaged with expensive AI6 uh that makes the problem even worse so here u
[20:50] makes the problem even worse so here u there are two problems that need action
[20:53] there are two problems that need action So first of all like the optical wafer
[20:55] So first of all like the optical wafer level test requires submicron fiber
[20:58] level test requires submicron fiber alignment to couplers across hundreds of
[21:01] alignment to couplers across hundreds of devices per wafer. Again this is not
[21:04] devices per wafer. Again this is not efficient. This is slow expensive and
[21:06] efficient. This is slow expensive and and doesn't scale. So we need automated
[21:09] and doesn't scale. So we need automated optical probers and the automation
[21:12] optical probers and the automation really has to extend into the the test
[21:14] really has to extend into the the test as much as into the assembly. And the
[21:17] as much as into the assembly. And the second is that uh what's really missing
[21:18] second is that uh what's really missing is that
[21:20] is that the link between the foundry and the
[21:22] the link between the foundry and the OSAT of we need a unified interface
[21:25] OSAT of we need a unified interface standard for optics and handoff protocol
[21:28] standard for optics and handoff protocol that uh contractually guarantees yield
[21:31] that uh contractually guarantees yield uh reliability and package performance.
[21:35] uh reliability and package performance. Um so um just to summarize and what we
[21:40] Um so um just to summarize and what we need to close the gaps the gaps that I
[21:43] need to close the gaps the gaps that I just reviewed uh we need three things to
[21:46] just reviewed uh we need three things to happen. So first the industry need to
[21:48] happen. So first the industry need to converge on the reference manufacturing
[21:50] converge on the reference manufacturing flows particularly for fiber attach and
[21:52] flows particularly for fiber attach and laser integration so the tooling and uh
[21:56] laser integration so the tooling and uh yield learning can scale across
[21:58] yield learning can scale across different layers. Second, we need true
[22:01] different layers. Second, we need true system level co-designes of photonic
[22:03] system level co-designes of photonic packages with thermal, mechanical,
[22:05] packages with thermal, mechanical, optical, electrical considered from the
[22:08] optical, electrical considered from the start. And then the third is that we
[22:10] start. And then the third is that we need to establish the KGD or no good D
[22:14] need to establish the KGD or no good D for photonics. Um vapor level optical
[22:17] for photonics. Um vapor level optical probing test design and a standardized
[22:20] probing test design and a standardized methods that that match the wafer level
[22:22] methods that that match the wafer level performance to package performance are
[22:24] performance to package performance are are um essential. Um and uh yeah, thanks
[22:29] are um essential. Um and uh yeah, thanks um for your attention.
[22:32] um for your attention. >> Thank you Armagan. Really really
[22:33] >> Thank you Armagan. Really really interesting uh talk. So um a question
[22:36] interesting uh talk. So um a question from me. You you discussed active
[22:38] from me. You you discussed active alignment as being fundamentally too
[22:41] alignment as being fundamentally too slow and expensive for future AI
[22:44] slow and expensive for future AI demands. How realistic do you think is
[22:47] demands. How realistic do you think is fully passive high throughput alignment
[22:49] fully passive high throughput alignment in the near term? And what still in your
[22:52] in the near term? And what still in your view needs to be solved? And and I guess
[22:54] view needs to be solved? And and I guess to add to it, are Nokia trying to solve
[22:56] to add to it, are Nokia trying to solve that themselves or are you relying on
[22:58] that themselves or are you relying on innovation from this community?
[23:01] innovation from this community? >> Uh, of course we rely on the innovation
[23:03] >> Uh, of course we rely on the innovation from the community uh cuz like we don't
[23:05] from the community uh cuz like we don't work in that area. Uh but really um it
[23:08] work in that area. Uh but really um it might be at too good uh to be true at
[23:11] might be at too good uh to be true at this point uh having a fully automated
[23:13] this point uh having a fully automated like passive alignment but this is
[23:16] like passive alignment but this is really at the end what we need for this
[23:17] really at the end what we need for this scale because what we see today is that
[23:20] scale because what we see today is that uh that like fiber uh alignment uh
[23:22] uh that like fiber uh alignment uh procedure
[23:24] procedure um it it takes uh it's time consuming
[23:28] um it it takes uh it's time consuming and uh it's it does not scale. So that's
[23:31] and uh it's it does not scale. So that's very simple. So based on like what what
[23:34] very simple. So based on like what what we see today. So the time that it takes
[23:36] we see today. So the time that it takes for us to power on everything to align
[23:39] for us to power on everything to align and u um like apply the the the epoxy UV
[23:44] and u um like apply the the the epoxy UV cure it. So this whole process uh is not
[23:47] cure it. So this whole process uh is not uh uh like is very prohibitive for uh
[23:50] uh uh like is very prohibitive for uh larger scale uh manufacturing for AI
[23:53] larger scale uh manufacturing for AI demands. And uh this is like what I
[23:55] demands. And uh this is like what I wanted to ask the industry. I know there
[23:57] wanted to ask the industry. I know there has been some work around that but uh we
[24:00] has been some work around that but uh we really like want to rely on the
[24:01] really like want to rely on the ecosystem partners to work on it and and
[24:04] ecosystem partners to work on it and and uh to enable that. Uh that's that's
[24:06] uh to enable that. Uh that's that's going to um um in my opinion um um make
[24:11] going to um um in my opinion um um make a a very neat solution for uh today
[24:15] a a very neat solution for uh today demands that I'm sure um most of the the
[24:18] demands that I'm sure um most of the the companies are dealing with it today.
[24:22] companies are dealing with it today. Well, if you think you can enable that
[24:23] Well, if you think you can enable that in the audience, then please write in
[24:25] in the audience, then please write in the chat and we can connect you with
[24:26] the chat and we can connect you with Armagan. Uh there's a question from Mves
[24:30] Armagan. Uh there's a question from Mves Yusfi from Photon first. What's on your
[24:32] Yusfi from Photon first. What's on your mind?
[24:33] mind? >> Hi, thank you very much. So, I'm sorry I
[24:35] >> Hi, thank you very much. So, I'm sorry I forgot my anti-reflection coating so
[24:37] forgot my anti-reflection coating so it's still reflecting too much. Uh
[24:40] it's still reflecting too much. Uh question would be that this is a very
[24:42] question would be that this is a very nice idea. Thank you for the talk. But
[24:44] nice idea. Thank you for the talk. But you require a bunch of peripheral
[24:46] you require a bunch of peripheral technologies like the automation
[24:48] technologies like the automation equipment etc to come into play to be
[24:50] equipment etc to come into play to be mature in order for you to do this. Um
[24:53] mature in order for you to do this. Um do you have an internal timeline or do
[24:55] do you have an internal timeline or do you know when we could expect this to
[24:57] you know when we could expect this to happen? So well uh so the obvious answer
[25:00] happen? So well uh so the obvious answer is as as soon as possible but I think
[25:03] is as as soon as possible but I think just uh catching up uh with the vape of
[25:06] just uh catching up uh with the vape of the the co- package optics and near
[25:08] the the co- package optics and near package optic we're looking into a
[25:09] package optic we're looking into a timeline of 20 27 2028 really
[25:15] so I think yeah
[25:18] so I think yeah >> okay great well uh thank you Armagan
[25:21] >> okay great well uh thank you Armagan again that's a that's a great start to
[25:22] again that's a that's a great start to this conversation please uh stay in the
[25:24] this conversation please uh stay in the room and and contribute to the other
[25:26] room and and contribute to the other talks uh so Um, next I'm delighted to
[25:29] talks uh so Um, next I'm delighted to introduce Luron Gance from Nvidia, uh,
[25:32] introduce Luron Gance from Nvidia, uh, who works at the forefront of AI
[25:34] who works at the forefront of AI infrastructure and next generation
[25:36] infrastructure and next generation networking architectures. As Jose
[25:39] networking architectures. As Jose already said, Lyon is one of the keynote
[25:41] already said, Lyon is one of the keynote speakers for our upcoming, uh, global
[25:43] speakers for our upcoming, uh, global advanced manufacturing alliance meeting
[25:45] advanced manufacturing alliance meeting taking place in Brussels. And today,
[25:47] taking place in Brussels. And today, he'll share an NVIDIA perspective on the
[25:49] he'll share an NVIDIA perspective on the technologies and manufacturing
[25:51] technologies and manufacturing innovations required to scale AI data
[25:53] innovations required to scale AI data center infrastructure for the next
[25:55] center infrastructure for the next generation computing. Uh Lon, please
[25:59] generation computing. Uh Lon, please share your slides. The floor and the
[26:00] share your slides. The floor and the attention of everyone is yours. Thank
[26:02] attention of everyone is yours. Thank you. Sure.
[26:03] you. Sure. >> Thank you.
[26:16] >> Hey, can you see my slides?
[26:19] >> Hey, can you see my slides? >> Yes. Yes. Perfect.
[26:21] >> Yes. Yes. Perfect. >> Okay,
[26:22] >> Okay, great. So, thank you very much. Uh I'm
[26:25] great. So, thank you very much. Uh I'm Luron Gans from uh Nvidia research and
[26:28] Luron Gans from uh Nvidia research and today I'll give a short overview of how
[26:30] today I'll give a short overview of how AI data centers scaling is driving new
[26:33] AI data centers scaling is driving new requirement for networking futonics and
[26:35] requirement for networking futonics and advanced packaging. So in many ways
[26:38] advanced packaging. So in many ways futonics actually becomes synonymous to
[26:41] futonics actually becomes synonymous to advanced packaging. What it cannot scale
[26:43] advanced packaging. What it cannot scale without uh without advanced packaging.
[26:45] without uh without advanced packaging. And today AI scaling is not just about
[26:49] And today AI scaling is not just about uh it's not limited by the compute. we
[26:51] uh it's not limited by the compute. we actually need to move insane amount of
[26:54] actually need to move insane amount of data from one point to the other and
[26:56] data from one point to the other and that's driving uh a lot of innovation
[26:59] that's driving uh a lot of innovation and that's what that's one of the
[27:00] and that's what that's one of the bottlenecks in in uh in futonic scaling
[27:04] bottlenecks in in uh in futonic scaling or AI scaling now at all so sorry
[27:13] yeah
[27:15] yeah uh so let's when we talk about AI
[27:17] uh so let's when we talk about AI scaling uh let's uh define three
[27:19] scaling uh let's uh define three different domains so the first domain is
[27:21] different domains so the first domain is the scale up and this was already
[27:24] the scale up and this was already mentioned in the last talk but scale up
[27:26] mentioned in the last talk but scale up uh is about connecting a few GPUs uh uh
[27:29] uh is about connecting a few GPUs uh uh to to act as one big uh GPU to act as
[27:32] to to act as one big uh GPU to act as one super nod. It usually defines uh it
[27:36] one super nod. It usually defines uh it usually uses a very high bandwidth uh
[27:39] usually uses a very high bandwidth uh connections
[27:41] connections and uh at this point we are still using
[27:43] and uh at this point we are still using copper for uh for these kind of
[27:45] copper for uh for these kind of connections. Uh the other domain is the
[27:47] connections. Uh the other domain is the scale out where uh classical networking
[27:50] scale out where uh classical networking is is is being used. Uh this is where
[27:53] is is is being used. Uh this is where co- package optics actually uh uh
[27:55] co- package optics actually uh uh actually comes to play
[27:57] actually comes to play and uh in uh and and here we are
[28:00] and uh in uh and and here we are connecting a few Rexs or across an
[28:02] connecting a few Rexs or across an entire data center but there is an
[28:04] entire data center but there is an actually a third domain uh which is uh
[28:07] actually a third domain uh which is uh lately was introduced but I think it's
[28:08] lately was introduced but I think it's it's becoming more and more important
[28:10] it's becoming more and more important and uh it's becoming more and more
[28:13] and uh it's becoming more and more important and that is the scale across
[28:16] important and that is the scale across and in the scale across we are
[28:18] and in the scale across we are connecting a few data centers as I'm
[28:20] connecting a few data centers as I'm saying uh training does not stop at at
[28:21] saying uh training does not stop at at the at the wreck. We want to utilize
[28:24] the at the wreck. We want to utilize many data centers to uh to enable
[28:26] many data centers to uh to enable training.
[28:27] training. >> So we need to use the networking that we
[28:31] >> So we need to use the networking that we have today uh to connect several data
[28:33] have today uh to connect several data centers and that is uh and that brings a
[28:35] centers and that is uh and that brings a lot of uh new barriers as so uh for
[28:38] lot of uh new barriers as so uh for example latency how how much how much
[28:41] example latency how how much how much far is is far enough. how how far can we
[28:44] far is is far enough. how how far can we can we connect two data centers that
[28:46] can we connect two data centers that where latency does not kill us or or
[28:48] where latency does not kill us or or power does not does not limit or or kill
[28:50] power does not does not limit or or kill the or kill the link. So now that we
[28:53] the or kill the link. So now that we have defined all of this uh just for you
[28:57] have defined all of this uh just for you to just for you to show uh to see uh
[29:00] to just for you to show uh to see uh this is these are the scales that we are
[29:03] this is these are the scales that we are talking about uh scale out is moving in
[29:05] talking about uh scale out is moving in a stream scale and and and you can see
[29:07] a stream scale and and and you can see that the magnitude of these uh of these
[29:09] that the magnitude of these uh of these new data centers and this is the reach
[29:11] new data centers and this is the reach that we have to uh that we have to get.
[29:15] that we have to uh that we have to get. Uh one more thing is that uh optics
[29:18] Uh one more thing is that uh optics today uh as I as I mentioned actually
[29:20] today uh as I as I mentioned actually controls two of these domains which is
[29:22] controls two of these domains which is the scale out and the scale and the
[29:24] the scale out and the scale and the scale across. In the scale up we are
[29:26] scale across. In the scale up we are still using uh copper uh cables but they
[29:29] still using uh copper uh cables but they are very very limited in reach. If we
[29:31] are very very limited in reach. If we can make our uh optical links uh much
[29:34] can make our uh optical links uh much more efficient then at some point we can
[29:37] more efficient then at some point we can replace the scale up network uh from
[29:40] replace the scale up network uh from using copper to uh using optics and
[29:43] using copper to uh using optics and which will be kind of amazing if you
[29:45] which will be kind of amazing if you think about it because we'll have links
[29:47] think about it because we'll have links uh optic links from the scale of uh
[29:50] uh optic links from the scale of uh millimeters to the scale of kilometers
[29:52] millimeters to the scale of kilometers which is several orders of magnitude and
[29:54] which is several orders of magnitude and it's it's pretty amazing.
[29:57] it's it's pretty amazing. So now we we understood this let's talk
[30:00] So now we we understood this let's talk about the what consideration we need in
[30:03] about the what consideration we need in in AI networking and what makes uh a
[30:05] in AI networking and what makes uh a link efficient. Uh so let's start with
[30:08] link efficient. Uh so let's start with throughput because that is the most easy
[30:10] throughput because that is the most easy to understand. Uh you've all seen how uh
[30:13] to understand. Uh you've all seen how uh the data uh the data rate scales. Uh
[30:17] the data uh the data rate scales. Uh people have mentioned uh M's hyper m law
[30:21] people have mentioned uh M's hyper m law or people sometimes call it the Jensen
[30:23] or people sometimes call it the Jensen law every 18 months. It's it's it's uh
[30:26] law every 18 months. It's it's it's uh order of magnitude more. So throughput
[30:28] order of magnitude more. So throughput is is is kind of uh kind of obvious but
[30:32] is is is kind of uh kind of obvious but I want to raise awareness to something
[30:34] I want to raise awareness to something else that is that is that I found that
[30:36] else that is that is that I found that people le talk less about in our in our
[30:39] people le talk less about in our in our industry and I want to explain why it's
[30:40] industry and I want to explain why it's very very important and that is radics
[30:43] very very important and that is radics and what is radics is is actually the
[30:45] and what is radics is is actually the number of IO per as how many how many
[30:48] number of IO per as how many how many connections we have from each uh from
[30:50] connections we have from each uh from each switch and why this is critical
[30:53] each switch and why this is critical because the higher the radics the
[30:55] because the higher the radics the flatter the network. So I want to show
[30:57] flatter the network. So I want to show you a very simple example. Let's say we
[31:00] you a very simple example. Let's say we have a switch with with radics two which
[31:02] have a switch with with radics two which means it's only have uh two connections
[31:04] means it's only have uh two connections and we have another switch with four
[31:06] and we have another switch with four with radics 4 which means it has four
[31:09] with radics 4 which means it has four connections. So if I want to connect
[31:10] connections. So if I want to connect this node to this one I need to do one 2
[31:14] this node to this one I need to do one 2 3 four hops.
[31:16] 3 four hops. Uh and each hop is latency.
[31:19] Uh and each hop is latency. So uh in this kind of network uh I have
[31:22] So uh in this kind of network uh I have a lot of latency. But if I have a larger
[31:24] a lot of latency. But if I have a larger radics and a flatter network, then I
[31:27] radics and a flatter network, then I only need two hops in order to connect
[31:29] only need two hops in order to connect each node, which means less latency. But
[31:32] each node, which means less latency. But even more important, look at the number
[31:34] even more important, look at the number of switches I have here. I have only one
[31:36] of switches I have here. I have only one switch. Here I have three. So I saved
[31:38] switch. Here I have three. So I saved number of switches. I saved I saved
[31:40] number of switches. I saved I saved power. I saved cost. So Radics is very
[31:44] power. I saved cost. So Radics is very very important. And Radics directly uh
[31:46] very important. And Radics directly uh connects to the number of fibers we can
[31:49] connects to the number of fibers we can we can break out of the of the chip. So
[31:53] we can break out of the of the chip. So we need uh we need chips with uh with a
[31:56] we need uh we need chips with uh with a lot of uh a lot of IO's and that that is
[31:59] lot of uh a lot of IO's and that that is a big challenge in the optical community
[32:01] a big challenge in the optical community uh at this point. But hopefully now you
[32:04] uh at this point. But hopefully now you understand the the importance of that.
[32:07] understand the the importance of that. Uh other KPIs that are that are of
[32:09] Uh other KPIs that are that are of course important are reach uh as I
[32:12] course important are reach uh as I mentioned uh copper cables only can
[32:14] mentioned uh copper cables only can reach only 3 mters optics can reach from
[32:16] reach only 3 mters optics can reach from millimeters to kilometers. uh power
[32:19] millimeters to kilometers. uh power efficiency which I will mention uh in
[32:22] efficiency which I will mention uh in the in the next slide. Uh of course
[32:24] the in the next slide. Uh of course latency uh which we also addressed in
[32:26] latency uh which we also addressed in the in when I mentioned reics uh and
[32:29] the in when I mentioned reics uh and edge bandwidth density which packet
[32:31] edge bandwidth density which packet which is directly influenced by
[32:34] which is directly influenced by packaging because it defines the how
[32:35] packaging because it defines the how many how how many electrical uh how many
[32:38] many how how many electrical uh how many available electrical and optical beach
[32:40] available electrical and optical beach we have and they and they also limit the
[32:43] we have and they and they also limit the radics. Uh so let look let's look at the
[32:46] radics. Uh so let look let's look at the at the at the packaging. So today we are
[32:48] at the at the packaging. So today we are at the stage where we use coupon mcm and
[32:53] at the stage where we use coupon mcm and when you consider c coup is actually
[32:55] when you consider c coup is actually pretty amazing it's it's 3D stacking
[32:57] pretty amazing it's it's 3D stacking it's optics it's 3D stacking of optics
[33:00] it's optics it's 3D stacking of optics uh but today we are limited by the
[33:02] uh but today we are limited by the number of uh number of bumps so when so
[33:05] number of uh number of bumps so when so and and that makes the only way to scale
[33:09] and and that makes the only way to scale in this uh in this kind of architecture
[33:11] in this uh in this kind of architecture uh with regard to rate we can only
[33:13] uh with regard to rate we can only increase the rate so that's why we are
[33:16] increase the rate so that's why we are talking about 400g in in the next uh
[33:18] talking about 400g in in the next uh generation and from packaging point of
[33:20] generation and from packaging point of view uh that needs to we need to
[33:24] view uh that needs to we need to introduce new new kind of materials like
[33:25] introduce new new kind of materials like Tfln or any other new material that can
[33:28] Tfln or any other new material that can that can allow us to go to higher and
[33:30] that can allow us to go to higher and higher rates.
[33:32] higher rates. Uh the other uh the other u
[33:35] Uh the other uh the other u consideration is that when we go to
[33:37] consideration is that when we go to higher rates we need very very strong
[33:39] higher rates we need very very strong services. These are the devices that
[33:42] services. These are the devices that aggregate the data and the more you
[33:44] aggregate the data and the more you aggregate the data uh the more latency
[33:46] aggregate the data uh the more latency you have and the more power consumption
[33:48] you have and the more power consumption you need. So you actually pay twice in
[33:51] you need. So you actually pay twice in this uh in this scheme. Uh hopefully you
[33:54] this uh in this scheme. Uh hopefully you want to get to a point where we can have
[33:56] want to get to a point where we can have something like coupon cost which is a
[33:58] something like coupon cost which is a which is an interpose and there then we
[34:00] which is an interpose and there then we increase the number of uh uh of bumps or
[34:03] increase the number of uh uh of bumps or increase the number of available
[34:04] increase the number of available channels that we can transmit the data
[34:06] channels that we can transmit the data and then we can go to a more moderate
[34:09] and then we can go to a more moderate rate or or at least the option of going
[34:11] rate or or at least the option of going to more moderate rate is is open up to
[34:13] to more moderate rate is is open up to us and that can save a lot of power and
[34:16] us and that can save a lot of power and again that that's uh we are shifting the
[34:20] again that that's uh we are shifting the uh we are shifting the demand from
[34:23] uh we are shifting the demand from faster modulator or faster optics to
[34:25] faster modulator or faster optics to more efficient optics and that's a bit
[34:27] more efficient optics and that's a bit different because when we have an
[34:29] different because when we have an efficient optics when when we have an
[34:31] efficient optics when when we have an efficient modulator uh one that is have
[34:34] efficient modulator uh one that is have a very low uh vpil or a very high
[34:37] a very low uh vpil or a very high pometer pervolt we can actually do the
[34:40] pometer pervolt we can actually do the thermal tuning uh with capacitive tuning
[34:42] thermal tuning uh with capacitive tuning and not resistive tuning we can save a
[34:44] and not resistive tuning we can save a lot of uh a lot of power that way so as
[34:47] lot of uh a lot of power that way so as I said the the shift can go from faster
[34:50] I said the the shift can go from faster modulators to more efficient modulators
[34:52] modulators to more efficient modulators And of course the maybe the the next
[34:55] And of course the maybe the the next stage is to go to a large scale plutonic
[34:57] stage is to go to a large scale plutonic interposer when we have uh an entire
[35:00] interposer when we have uh an entire system that that that sits on on the
[35:01] system that that that sits on on the pick and we have much more uh u much
[35:05] pick and we have much more uh u much more channels to connect to the the data
[35:07] more channels to connect to the the data and we can even go to lower and lower
[35:10] and we can even go to lower and lower data than than in the previous uh
[35:12] data than than in the previous uh generation.
[35:14] generation. So with that uh I want to summarize in
[35:17] So with that uh I want to summarize in in in one sentence that is packaging
[35:19] in in one sentence that is packaging increasingly defines the achievable
[35:21] increasingly defines the achievable communication architecture. So when you
[35:24] communication architecture. So when you enable us new ways we can think of new
[35:26] enable us new ways we can think of new architecture that can that can save more
[35:29] architecture that can that can save more power save latency or reduce the latency
[35:32] power save latency or reduce the latency and be much more uh efficient.
[35:35] and be much more uh efficient. And with that uh I'd like to thank you.
[35:38] And with that uh I'd like to thank you. >> Thank you so much Lon. Uh incredible
[35:40] >> Thank you so much Lon. Uh incredible progress as always from Nvidia. Uh I
[35:42] progress as always from Nvidia. Uh I want to go straight to uh Stefan Heinman
[35:45] want to go straight to uh Stefan Heinman from Fix. Uh what's on your mind?
[35:48] from Fix. Uh what's on your mind? >> Thanks Luren for this fantastic
[35:50] >> Thanks Luren for this fantastic presentation really summarizing it very
[35:51] presentation really summarizing it very well. Um do you have any kind of
[35:54] well. Um do you have any kind of guideline for density requirements of
[35:56] guideline for density requirements of the electrical and optical interconnects
[35:59] the electrical and optical interconnects in that scenario that you showed? Are we
[36:01] in that scenario that you showed? Are we down to the sub 10 micron into the
[36:04] down to the sub 10 micron into the micron level densities or what what
[36:06] micron level densities or what what where do you see this going?
[36:09] where do you see this going? Uh so we do have we do have requirements
[36:11] Uh so we do have we do have requirements which I think we published um in the
[36:14] which I think we published um in the past. I I I don't remember the numbers
[36:17] past. I I I don't remember the numbers uh by heart but uh I can I can refer you
[36:20] uh by heart but uh I can I can refer you to the to the publication where it's
[36:22] to the to the publication where it's specified.
[36:24] specified. >> Thank you.
[36:26] >> Thank you. >> And Lon uh Nvidia sits in I think it's
[36:30] >> And Lon uh Nvidia sits in I think it's fair to say a unique position across the
[36:32] fair to say a unique position across the ecosystem. Um where do you think
[36:34] ecosystem. Um where do you think stronger collaboration is needed between
[36:37] stronger collaboration is needed between foundaries, packaging providers, optics
[36:40] foundaries, packaging providers, optics companies and system integrators?
[36:43] companies and system integrators? >> Uh I I think uh
[36:46] >> Uh I I think uh I think all of them is is very very
[36:48] I think all of them is is very very important. I know and then this is this
[36:50] important. I know and then this is this is kind of an answer that that says
[36:52] is kind of an answer that that says nothing but but uh we really see that
[36:55] nothing but but uh we really see that every stage of the way every part is is
[36:59] every stage of the way every part is is connected to the other. So uh and and
[37:03] connected to the other. So uh and and it's something I always say that that in
[37:05] it's something I always say that that in the future we won't have like uh u
[37:08] the future we won't have like uh u analog designers that that are
[37:10] analog designers that that are responsible to the interface. We will
[37:12] responsible to the interface. We will have interface uh designers that are
[37:15] have interface uh designers that are that are in charge of the optics, the
[37:17] that are in charge of the optics, the the electrics, the packaging, everything
[37:19] the electrics, the packaging, everything is connected. And and one of I think one
[37:22] is connected. And and one of I think one of the most important thing is if if I'm
[37:25] of the most important thing is if if I'm I'm allowed to say is is that to enable
[37:28] I'm allowed to say is is that to enable uh software that can support this this
[37:31] uh software that can support this this multi- uh disciplinary u uh field.
[37:37] multi- uh disciplinary u uh field. >> Thank you. And uh we have a question um
[37:40] >> Thank you. And uh we have a question um in the chat from Taha II who wants to
[37:43] in the chat from Taha II who wants to know a little bit more about the
[37:44] know a little bit more about the photonic interposer. Taha uh could you
[37:48] photonic interposer. Taha uh could you um unmute and ask you a question.
[37:53] >> Yes, thank you. Thank you. So yeah my
[37:56] >> Yes, thank you. Thank you. So yeah my question is about uh to give if possible
[37:58] question is about uh to give if possible more details what about what you mean
[38:00] more details what about what you mean about photonic interpose. Is it glass
[38:03] about photonic interpose. Is it glass core substrate? Is it uh meant to uh
[38:06] core substrate? Is it uh meant to uh integrate some wave guide structure in
[38:09] integrate some wave guide structure in it or
[38:10] it or >> so? So so I I left it
[38:12] >> so? So so I I left it >> I left it vague on purpose because uh
[38:15] >> I left it vague on purpose because uh because any good solution that can that
[38:17] because any good solution that can that can route uh a lot of uh a lot of
[38:20] can route uh a lot of uh a lot of waveguide with with low loss is is a
[38:24] waveguide with with low loss is is a good solution. So I don't want to
[38:25] good solution. So I don't want to constrain you to to say this is oh this
[38:27] constrain you to to say this is oh this is glass or or silicon nitrite. I want
[38:30] is glass or or silicon nitrite. I want you to give the solutions to that. I
[38:31] you to give the solutions to that. I want you to find and if and if it's a
[38:33] want you to find and if and if it's a new material that that you discover that
[38:36] new material that that you discover that is also fine.
[38:38] is also fine. >> Thank you.
[38:41] >> Thank you. >> And lon um Amagan touched on this and I
[38:43] >> And lon um Amagan touched on this and I know it's going to come up again today
[38:45] know it's going to come up again today but many of the industry see still see
[38:48] but many of the industry see still see fiber to chip interface as one of the
[38:50] fiber to chip interface as one of the biggest unresolved bottlenecks. It seems
[38:53] biggest unresolved bottlenecks. It seems pretty fundamental. uh and how important
[38:56] pretty fundamental. uh and how important is reducing that FAU complexity such as
[38:59] is reducing that FAU complexity such as the number of active alignment steps and
[39:02] the number of active alignment steps and coupling loss uh to enable practical
[39:05] coupling loss uh to enable practical hyperskll deployment of co-ackaged
[39:07] hyperskll deployment of co-ackaged optics. So I I think that that that
[39:09] optics. So I I think that that that directly connects to what I said about
[39:11] directly connects to what I said about radics because this is this is one of
[39:14] radics because this is this is one of the main the main uh the main issues uh
[39:19] the main the main uh the main issues uh having uh low loss good connection that
[39:23] having uh low loss good connection that can also uh not limit us with with
[39:27] can also uh not limit us with with regard to the number of fibers that we
[39:29] regard to the number of fibers that we can take we can take out is uh is very
[39:31] can take we can take out is uh is very very important.
[39:33] very important. um that that will define the
[39:35] um that that will define the architecture of the of the network
[39:36] architecture of the of the network essentially.
[39:38] essentially. >> Thank you. And there's there's a further
[39:40] >> Thank you. And there's there's a further question from Stevie Cordet in the the
[39:43] question from Stevie Cordet in the the chat. Um Stevie, um could you ask your
[39:46] chat. Um Stevie, um could you ask your question?
[39:51] >> Um [clears throat] thank you for the
[39:52] >> Um [clears throat] thank you for the presentation. Yes. Um my question is
[39:54] presentation. Yes. Um my question is related to the latency. you mentioned
[39:56] related to the latency. you mentioned that this is a key point and I would
[39:58] that this is a key point and I would like to know uh what is the target
[40:00] like to know uh what is the target latency you're looking for in the case
[40:03] latency you're looking for in the case of the scale up scale out and scale
[40:05] of the scale up scale out and scale across that you show at the beginning
[40:08] across that you show at the beginning >> it it's really depends on the
[40:09] >> it it's really depends on the application so it's it's a different
[40:11] application so it's it's a different latency for training and it's a
[40:12] latency for training and it's a different latency for inference so for
[40:14] different latency for inference so for example you don't want the inference to
[40:16] example you don't want the inference to have a a large latency because then you
[40:19] have a a large latency because then you will be waiting a lot of a long time for
[40:21] will be waiting a lot of a long time for the the answer but in training maybe you
[40:23] the the answer but in training maybe you can you can uh you
[40:25] can you can uh you uh tolerate uh higher latency if you if
[40:29] uh tolerate uh higher latency if you if you do the training correctly if you can
[40:30] you do the training correctly if you can divide the the training or for different
[40:33] divide the the training or for different agents. So it's really depends on on
[40:35] agents. So it's really depends on on what is the application you are you are
[40:37] what is the application you are you are targeting.
[40:40] targeting. >> Thank you.
[40:42] >> Thank you. Okay, great. Well, thank you, uh, Lyon.
[40:45] Okay, great. Well, thank you, uh, Lyon. Um, and next I'm pleased to introduce
[40:48] Um, and next I'm pleased to introduce Sandeep Sane, senior director and head
[40:50] Sandeep Sane, senior director and head of packaging at light matter, a company
[40:53] of packaging at light matter, a company at the forefront of phutonics enabled AI
[40:55] at the forefront of phutonics enabled AI infrastructure and optical interconnect
[40:57] infrastructure and optical interconnect technology. Today Sandep will share
[41:00] technology. Today Sandep will share light matter's perspective on how
[41:02] light matter's perspective on how advanced packaging and phetonics can
[41:04] advanced packaging and phetonics can help address the growing bandwidth power
[41:06] help address the growing bandwidth power and scalability challenges facing next
[41:09] and scalability challenges facing next generation AI data centers and high
[41:11] generation AI data centers and high performance computing systems. Sandep
[41:14] performance computing systems. Sandep you've shared your slides. The floor and
[41:15] you've shared your slides. The floor and the attention of everyone is yours.
[41:17] the attention of everyone is yours. Thank you.
[41:24] Sandep are you on mute possibly?
[41:32] Ah,
[41:34] Ah, now I was on mute. Uh,
[41:38] now I was on mute. Uh, let me share.
[41:50] >> Everything coming through well, y'all?
[41:52] >> Everything coming through well, y'all? >> Perfect. Yes. Thank you.
[41:54] >> Perfect. Yes. Thank you. >> All right. So thank you for me inviting
[41:56] >> All right. So thank you for me inviting me here and uh what luron and armagan
[41:59] me here and uh what luron and armagan has said is really puts my presentation
[42:02] has said is really puts my presentation in the right perspective. You will sound
[42:04] in the right perspective. You will sound that we are all saying similar things.
[42:06] that we are all saying similar things. So which actually tells that for the
[42:09] So which actually tells that for the industry what are the big challenges and
[42:11] industry what are the big challenges and we are all cohesively saying similar uh
[42:14] we are all cohesively saying similar uh challenges that we are seeing to how we
[42:16] challenges that we are seeing to how we will what is needed to bring this
[42:18] will what is needed to bring this technology into high volume right. Uh so
[42:21] technology into high volume right. Uh so if you look at the we all know the pro
[42:23] if you look at the we all know the pro copper has some limitations and we all
[42:25] copper has some limitations and we all talked about it. So if you now look at
[42:27] talked about it. So if you now look at the promise of silicon photonics uh this
[42:29] the promise of silicon photonics uh this is the way we see the problem how we
[42:31] is the way we see the problem how we evolve silicon photonics technology
[42:34] evolve silicon photonics technology going from pluggables which is what
[42:36] going from pluggables which is what we're calling the current generation
[42:37] we're calling the current generation where you're getting energy efficiency
[42:39] where you're getting energy efficiency anywhere between 70 20 pjles per bit and
[42:43] anywhere between 70 20 pjles per bit and just for scaling we are calling that as
[42:45] just for scaling we are calling that as 1x bandwidth. as you move uh the silicon
[42:48] 1x bandwidth. as you move uh the silicon photonix devices closer to A6 uh say
[42:51] photonix devices closer to A6 uh say near package optics you gain efficiency
[42:54] near package optics you gain efficiency and also you gain bandwidth and further
[42:57] and also you gain bandwidth and further if you move to the generation 3 where
[42:59] if you move to the generation 3 where you now integrate these devices onto the
[43:02] you now integrate these devices onto the package itself the uh what we call 2D
[43:04] package itself the uh what we call 2D CPU you're further increasing the
[43:06] CPU you're further increasing the bandwidth from up to 10x but really
[43:10] bandwidth from up to 10x but really where the value really starts becoming
[43:12] where the value really starts becoming very clear is if I get into the 3D CPU
[43:15] very clear is if I get into the 3D CPU photonic space as well as interposes.
[43:18] photonic space as well as interposes. Somebody asked a question before uh
[43:20] Somebody asked a question before uh about optical interposes. We are
[43:22] about optical interposes. We are absolutely working on that concept and
[43:24] absolutely working on that concept and we already have some products out there
[43:26] we already have some products out there uh reference designs out there. And when
[43:28] uh reference designs out there. And when you go into the 3D space or interposer
[43:30] you go into the 3D space or interposer space, this is where you really now
[43:33] space, this is where you really now scale the energy down to 2 pjles or
[43:35] scale the energy down to 2 pjles or around that and then your overall
[43:38] around that and then your overall bandwidth against the pluggable space is
[43:41] bandwidth against the pluggable space is almost 100x. So from light matter point
[43:44] almost 100x. So from light matter point of view we are really looking at this
[43:45] of view we are really looking at this entire road map and solving the problems
[43:48] entire road map and solving the problems as is evolving and the demand of the
[43:51] as is evolving and the demand of the silicon photonics is evolving in the
[43:53] silicon photonics is evolving in the industry.
[43:55] industry. Uh just to tell you where we are today.
[43:57] Uh just to tell you where we are today. This is just where showing from a link
[43:59] This is just where showing from a link performance. We have already
[44:00] performance. We have already demonstrated some of these links. 56 gig
[44:04] demonstrated some of these links. 56 gig NRZ up to 16 lambdas per fiber, 112 4 up
[44:08] NRZ up to 16 lambdas per fiber, 112 4 up to 16 lambdas per fiber. And now we are
[44:11] to 16 lambdas per fiber. And now we are also working on 224 up to four lambdas
[44:13] also working on 224 up to four lambdas per fiber. Essentially these are all in
[44:16] per fiber. Essentially these are all in silicon uh in our labs today. uh we have
[44:19] silicon uh in our labs today. uh we have basically we are showing that we can
[44:21] basically we are showing that we can deliver 800 to 1T terabit per fiber that
[44:24] deliver 800 to 1T terabit per fiber that is available today and it can support
[44:27] is available today and it can support both buy up to 32t per optical engine
[44:31] both buy up to 32t per optical engine and you can support up to 1 kilometers
[44:33] and you can support up to 1 kilometers right and we are both standards or
[44:35] right and we are both standards or custom design as needed
[44:39] custom design as needed but beyond uh from a technology point of
[44:42] but beyond uh from a technology point of view uh we are all focused on while we
[44:44] view uh we are all focused on while we deliver what is needed for the market
[44:46] deliver what is needed for the market but we are heavily focused on what is
[44:49] but we are heavily focused on what is need what we can bring as a uh key
[44:51] need what we can bring as a uh key technology enablers with the 3D CPR
[44:54] technology enablers with the 3D CPR interposer programs and these are the
[44:56] interposer programs and these are the two activities that we are heavily
[44:58] two activities that we are heavily involved. Uh the first one on the left
[45:01] involved. Uh the first one on the left shows a 3D CPO what we call our L
[45:04] shows a 3D CPO what we call our L series. This is all based on chip on
[45:06] series. This is all based on chip on wafer assembly flows where your EIC die
[45:09] wafer assembly flows where your EIC die is attached on the pig die uh as shown
[45:12] is attached on the pig die uh as shown and is 3D integrated with detachable
[45:15] and is 3D integrated with detachable fibers at the perimeter. As Lauron said
[45:19] fibers at the perimeter. As Lauron said uh detach I mean fiber attach and giving
[45:22] uh detach I mean fiber attach and giving the radics that customer wants is a very
[45:24] the radics that customer wants is a very critical uh vector. Uh and we believe
[45:27] critical uh vector. Uh and we believe that detachability allows you to do at
[45:30] that detachability allows you to do at least solving a fiber attached problem.
[45:32] least solving a fiber attached problem. And the next question is how can you
[45:34] And the next question is how can you scale how many fibers can you really
[45:36] scale how many fibers can you really attach per edge which is a scaling
[45:39] attach per edge which is a scaling problem that we have to focus on. Uh but
[45:42] problem that we have to focus on. Uh but by doing so we can really deliver uh on
[45:44] by doing so we can really deliver uh on the CPU front on the 3D CPU front we can
[45:47] the CPU front on the 3D CPU front we can deliver a CPU engine uh up to 64 terabs
[45:50] deliver a CPU engine uh up to 64 terabs uh uh bandwidth. The one on the right is
[45:54] uh uh bandwidth. The one on the right is really the silicon in the optical
[45:56] really the silicon in the optical interposer what we talk about and we
[45:59] interposer what we talk about and we have this what we call as M series and
[46:02] have this what we call as M series and the again you can see all the chiplets
[46:04] the again you can see all the chiplets on top. These are all the EIC chiplets
[46:07] on top. These are all the EIC chiplets that are sitting on top of a photonic uh
[46:10] that are sitting on top of a photonic uh multi-retical interposer. These are
[46:12] multi-retical interposer. These are retically stitched optically and this is
[46:15] retically stitched optically and this is again built on a standard chip on wafer
[46:17] again built on a standard chip on wafer assembly flows that are fairly prevalent
[46:19] assembly flows that are fairly prevalent in the mainstream uh uh uh uh advanced
[46:24] in the mainstream uh uh uh uh advanced packaging technologies by doing so. This
[46:27] packaging technologies by doing so. This is a reference design that we have
[46:28] is a reference design that we have demonstrated that we can support up to
[46:30] demonstrated that we can support up to 114 terabits on a almost a 2,000 mm
[46:33] 114 terabits on a almost a 2,000 mm square die complexes. So this is what we
[46:36] square die complexes. So this is what we are really focusing on on bringing some
[46:37] are really focusing on on bringing some of these technologies into the market.
[46:40] of these technologies into the market. Now if you look at what are the key
[46:43] Now if you look at what are the key challenges uh for us to enable some of
[46:45] challenges uh for us to enable some of these part to high volume and it will
[46:47] these part to high volume and it will sound very similar to what Armagan has
[46:49] sound very similar to what Armagan has said which is very good to hear there's
[46:51] said which is very good to hear there's a same voice coming across right and
[46:54] a same voice coming across right and I'll go through four challenges that I
[46:56] I'll go through four challenges that I see from my point of view very similar
[46:58] see from my point of view very similar to what previously talked about number
[47:01] to what previously talked about number one delivering a known good optical
[47:03] one delivering a known good optical engine is a mustave right and there are
[47:06] engine is a mustave right and there are different packaging technologies here uh
[47:09] different packaging technologies here uh to showcase case. First one is a
[47:11] to showcase case. First one is a standard package where you have an XPU
[47:14] standard package where you have an XPU die complex sitting on a standard
[47:15] die complex sitting on a standard substrate connected to optical engine.
[47:18] substrate connected to optical engine. The dieto-d routing uh between the XPU
[47:21] The dieto-d routing uh between the XPU to the optical engine is on a standard
[47:23] to the optical engine is on a standard organic substrate. Fairly
[47:26] organic substrate. Fairly straightforward to integrate this but
[47:28] straightforward to integrate this but quickly we'll have because your
[47:30] quickly we'll have because your electrical bandwidth is going to be
[47:31] electrical bandwidth is going to be limited if I continue to be on a
[47:33] limited if I continue to be on a standard organic substrate. You will
[47:35] standard organic substrate. You will have to integrate optical engines on
[47:37] have to integrate optical engines on advanced packaging technologies. And
[47:39] advanced packaging technologies. And there are some of the technologies are
[47:41] there are some of the technologies are E-IB uh style technology that Intel has
[47:45] E-IB uh style technology that Intel has which is fairly easy to integrate I
[47:47] which is fairly easy to integrate I would say but then there are other
[47:49] would say but then there are other technologies like cos or cos l where now
[47:52] technologies like cos or cos l where now you need to dedicate the optical engine
[47:54] you need to dedicate the optical engine to the dia complex upfront and this is
[47:57] to the dia complex upfront and this is makes it very critical uh to ensure that
[48:00] makes it very critical uh to ensure that when we build the optical engine it's a
[48:02] when we build the optical engine it's a known good uh to ship it out to any end
[48:05] known good uh to ship it out to any end customer. uh think about it like almost
[48:07] customer. uh think about it like almost like an HBM. HBM people build it and
[48:10] like an HBM. HBM people build it and they ship it out. Uh optical engines
[48:12] they ship it out. Uh optical engines needs to be very similar approach.
[48:17] needs to be very similar approach. The second is how do you deliver
[48:18] The second is how do you deliver scalable fiber attached solutions and
[48:20] scalable fiber attached solutions and this is where we have gone through
[48:22] this is where we have gone through multiple attached schemes and what we
[48:24] multiple attached schemes and what we believe is when you look at the what is
[48:27] believe is when you look at the what is feasible uh if you look at the key
[48:28] feasible uh if you look at the key performance indicator for high volume
[48:30] performance indicator for high volume manufacturing uh those are the metrics
[48:32] manufacturing uh those are the metrics shown on the in the table. uh we believe
[48:35] shown on the in the table. uh we believe that shifting left to the wafer scale
[48:38] that shifting left to the wafer scale optical coupler attach is very critical.
[48:41] optical coupler attach is very critical. By doing so uh by attaching these
[48:44] By doing so uh by attaching these optical couplers and making this
[48:46] optical couplers and making this detachable uh as a part of the solution,
[48:50] detachable uh as a part of the solution, you can really test out the pig
[48:52] you can really test out the pig interfaces very cleanly before you and
[48:56] interfaces very cleanly before you and then before you commit an expensive EIC
[48:59] then before you commit an expensive EIC die on top as well as when you look at
[49:01] die on top as well as when you look at the EICPIC stack, you can test the whole
[49:04] the EICPIC stack, you can test the whole stack before you ship it out to any end
[49:06] stack before you ship it out to any end customer. If I do other traditional
[49:09] customer. If I do other traditional optical attach uh fiber attach methods
[49:11] optical attach uh fiber attach methods such as VG group, this is not possible.
[49:14] such as VG group, this is not possible. So moving to a wafer scale processes is
[49:16] So moving to a wafer scale processes is very critical and we have done some
[49:19] very critical and we have done some announcements if you go to OFC we have
[49:21] announcements if you go to OFC we have done very some of the big announcements
[49:23] done very some of the big announcements as as from light matter on how we are uh
[49:26] as as from light matter on how we are uh we have solving these challenges. Uh the
[49:29] we have solving these challenges. Uh the second piece is of course the cost of
[49:31] second piece is of course the cost of these FAUs. cost is a big deal here
[49:33] these FAUs. cost is a big deal here right one of the FAU is one of the main
[49:37] right one of the FAU is one of the main bomb costs and it is primarily because
[49:39] bomb costs and it is primarily because of the manufacturing it's manually done
[49:41] of the manufacturing it's manually done and it's driven by the PM fibers because
[49:43] and it's driven by the PM fibers because the requirement of the rotation of the
[49:45] the requirement of the rotation of the stress rods that really is very time
[49:48] stress rods that really is very time consuming so industry needs to think
[49:50] consuming so industry needs to think about how do you automate that how do
[49:52] about how do you automate that how do you bring the cost down for some of
[49:54] you bring the cost down for some of these connectors and make this
[49:57] these connectors and make this standardization is a must have
[50:00] standardization is a must have the third challenge is delivering to
[50:02] the third challenge is delivering to scale on HPM is really equipment and
[50:04] scale on HPM is really equipment and materials, right? Equipment you have to
[50:06] materials, right? Equipment you have to transition automated alignment in
[50:09] transition automated alignment in seconds, not not in minutes. We have to
[50:11] seconds, not not in minutes. We have to do be able to do parallel processing so
[50:13] do be able to do parallel processing so that you can handle alignment, dispense
[50:16] that you can handle alignment, dispense and u cure uh uh in parallel. Uh and
[50:20] and u cure uh uh in parallel. Uh and also we should be able to reconfigure
[50:21] also we should be able to reconfigure these equipments to support multiple
[50:23] these equipments to support multiple products. From a materials point of
[50:26] products. From a materials point of view, this is the biggest uh time adder
[50:28] view, this is the biggest uh time adder today. uh epoxy if we can have an
[50:31] today. uh epoxy if we can have an ideally epoxy just to showcase this
[50:33] ideally epoxy just to showcase this right we are showing how many epoxies
[50:35] right we are showing how many epoxies are used in a simple wafer level attach
[50:39] are used in a simple wafer level attach process and you say at least two or
[50:41] process and you say at least two or three epoxies are used if you can align
[50:43] three epoxies are used if you can align to one sing single epoxy that will be
[50:46] to one sing single epoxy that will be ideal it has to be reflow compatible uh
[50:50] ideal it has to be reflow compatible uh fast UV cure less than 5 seconds would
[50:52] fast UV cure less than 5 seconds would be ideal today this is the slowest step
[50:54] be ideal today this is the slowest step in the flow it takes sec tens of seconds
[50:58] in the flow it takes sec tens of seconds to
[50:59] to And of course lower TG would be helpful
[51:02] And of course lower TG would be helpful to conform to different warpage
[51:04] to conform to different warpage requirements of these devices.
[51:06] requirements of these devices. And the last one is really uh cost the
[51:09] And the last one is really uh cost the total backend cost and people are not
[51:12] total backend cost and people are not used to these numbers. uh if you come
[51:14] used to these numbers. uh if you come from a traditional uh uh traditional
[51:18] from a traditional uh uh traditional market semiconductor technology uh if
[51:21] market semiconductor technology uh if you look at the PAT package assembly
[51:23] you look at the PAT package assembly test cost in the of silicon photonics
[51:26] test cost in the of silicon photonics world uh just as a demonstration it's
[51:29] world uh just as a demonstration it's shown that package assembly test cost is
[51:32] shown that package assembly test cost is almost 60 to 70% of the total cost
[51:35] almost 60 to 70% of the total cost traditional market it is somewhere
[51:36] traditional market it is somewhere between 25 to 30%. So this is a big uh
[51:40] between 25 to 30%. So this is a big uh big thing for everybody to understand
[51:42] big thing for everybody to understand that oh wackage can cost so much and if
[51:45] that oh wackage can cost so much and if you look at reasons for it is fiber
[51:48] you look at reasons for it is fiber attach is one of the key things but
[51:50] attach is one of the key things but higher test times and lack of
[51:51] higher test times and lack of standardization similar to what Armagan
[51:54] standardization similar to what Armagan has talked about is really causing some
[51:56] has talked about is really causing some of these costs to be much higher. So
[51:59] of these costs to be much higher. So while we asked the test vendors and all
[52:03] while we asked the test vendors and all to figure uh to work with us on uh
[52:05] to figure uh to work with us on uh solving some of these problem but from a
[52:07] solving some of these problem but from a pack from an architecture product
[52:09] pack from an architecture product architecture point of view you really
[52:11] architecture point of view you really have to design products for DFTDFM which
[52:14] have to design products for DFTDFM which is a standard practice in the mainstream
[52:16] is a standard practice in the mainstream industry but in silicon photonics world
[52:19] industry but in silicon photonics world this has to come up front uh and really
[52:22] this has to come up front uh and really drive the equipment and material
[52:24] drive the equipment and material development to really push the
[52:25] development to really push the throughput uh higher through assembly
[52:27] throughput uh higher through assembly and
[52:28] and So with that uh thank you for my giving
[52:31] So with that uh thank you for my giving me time to give express my opinion and
[52:33] me time to give express my opinion and I'll be open to any questions.
[52:35] I'll be open to any questions. >> Thank you very much for coming and
[52:37] >> Thank you very much for coming and express your opinion. We you know how
[52:40] express your opinion. We you know how much we love light matter at Opticus. I
[52:41] much we love light matter at Opticus. I would like to start by saying that thank
[52:43] would like to start by saying that thank you. Thank you very much for everything
[52:45] you. Thank you very much for everything that you are doing for the organization.
[52:48] that you are doing for the organization. There are many questions in the room but
[52:49] There are many questions in the room but I would like to start if we could help
[52:51] I would like to start if we could help you. We have we have 187 people in the
[52:54] you. We have we have 187 people in the room who wants to help you. what would
[52:56] room who wants to help you. what would be the biggest headache, the biggest
[52:58] be the biggest headache, the biggest challenge that you would like the room
[53:00] challenge that you would like the room to help you with?
[53:01] to help you with? >> Uh if I were to pick means when I look
[53:03] >> Uh if I were to pick means when I look at the four challenges that I have
[53:05] at the four challenges that I have listed down, uh I would really pick
[53:08] listed down, uh I would really pick focus on making wafer level fiber attach
[53:12] focus on making wafer level fiber attach a primary way to solve this issue. And
[53:15] a primary way to solve this issue. And with that if you just take that problem
[53:17] with that if you just take that problem statement uh this has every aspect of it
[53:20] statement uh this has every aspect of it like fiber attach materials equipment uh
[53:24] like fiber attach materials equipment uh how do you bring that all together to
[53:27] how do you bring that all together to solve that problem. What we bring to the
[53:29] solve that problem. What we bring to the table is really the test chips and we
[53:32] table is really the test chips and we have our own expertise but we would
[53:34] have our own expertise but we would really love to collaborate to see how we
[53:36] really love to collaborate to see how we bring the entire ecosystem so that we
[53:38] bring the entire ecosystem so that we can solve this in a
[53:40] can solve this in a >> just one thing wafer level fiber
[53:41] >> just one thing wafer level fiber attached. Can you give me like a number
[53:43] attached. Can you give me like a number that would make you happy in terms of
[53:45] that would make you happy in terms of insertion losses fiber to chip?
[53:47] insertion losses fiber to chip? >> Oh, insertion losses it has to be less
[53:49] >> Oh, insertion losses it has to be less than 1.5 dB. I think if
[53:53] than 1.5 dB. I think if it's uh
[53:55] it's uh this is a classical issue right if you
[53:56] this is a classical issue right if you give the overall link
[53:58] give the overall link uh if you ask an photonix uh engineer
[54:02] uh if you ask an photonix uh engineer they will say give me less than 1 dB but
[54:04] they will say give me less than 1 dB but if you put realistic manufacturing hat
[54:07] if you put realistic manufacturing hat on your hat I think 1.5 dB is achievable
[54:09] on your hat I think 1.5 dB is achievable and that's what we need to demonstrate
[54:12] and that's what we need to demonstrate in high volume
[54:13] in high volume >> right now in the chat anybody who can
[54:15] >> right now in the chat anybody who can achieve less than 1.5 dB in session
[54:17] achieve less than 1.5 dB in session losses wafer wafer to fiber please write
[54:19] losses wafer wafer to fiber please write so in the chat we questions in the room.
[54:21] so in the chat we questions in the room. The first one is coming from Photon
[54:22] The first one is coming from Photon first in the Netherlands. What's on your
[54:25] first in the Netherlands. What's on your mind?
[54:26] mind? >> Hi, thank you. What a nice talk. Uh very
[54:28] >> Hi, thank you. What a nice talk. Uh very very interesting. Um I'm going to ask
[54:30] very interesting. Um I'm going to ask the same question I asked Nokia because
[54:32] the same question I asked Nokia because the key uh enabler for you to reach
[54:34] the key uh enabler for you to reach these targets is the automation
[54:36] these targets is the automation partners. They have to be able to help
[54:38] partners. They have to be able to help you to bring this to the field. What let
[54:42] you to bring this to the field. What let me ask your opinion again here. What do
[54:44] me ask your opinion again here. What do you think is the timeline for them to
[54:46] you think is the timeline for them to match your road map?
[54:48] match your road map? I think I would agree with
[54:51] I think I would agree with Almagan we need these equipments by 2027
[54:54] Almagan we need these equipments by 2027 28 time frame and it has these
[54:58] 28 time frame and it has these equipments are high precision and high
[55:00] equipments are high precision and high throughput that's what we need right
[55:03] throughput that's what we need right because if you do a wafer level and if I
[55:05] because if you do a wafer level and if I do a a lens attach or on a wafers level
[55:09] do a a lens attach or on a wafers level depending on the size of how many lenses
[55:11] depending on the size of how many lenses I'm attaching that's huge amount of time
[55:13] I'm attaching that's huge amount of time it takes per wafer so automating precise
[55:17] it takes per wafer so automating precise align Alignment, automation, parallel
[55:20] align Alignment, automation, parallel handling is absolutely must.
[55:26] Sandep you were very clear the 1.5 dB
[55:29] Sandep you were very clear the 1.5 dB you're thinking about fiber fiber to
[55:31] you're thinking about fiber fiber to wafer. Is there a a need for fiber
[55:34] wafer. Is there a a need for fiber bundles and also give us a list how many
[55:37] bundles and also give us a list how many fibers at the same time?
[55:39] fibers at the same time? >> Oh,
[55:41] >> Oh, so this goes to what Lon has said. It is
[55:43] so this goes to what Lon has said. It is all about radics, right? In general the
[55:45] all about radics, right? In general the road map we look at it today we look at
[55:49] road map we look at it today we look at if I look at a shoreline take a pick 33
[55:52] if I look at a shoreline take a pick 33 mm is a shoreline on the full reticle.
[55:56] mm is a shoreline on the full reticle. Um maybe today we are about 100 110
[55:58] Um maybe today we are about 100 110 fibers on that edge but the demand is to
[56:02] fibers on that edge but the demand is to double it every so generation. So in
[56:04] double it every so generation. So in general we'll have to double that to 200
[56:07] general we'll have to double that to 200 to 300 mill fibers on that edge. So you
[56:09] to 300 mill fibers on that edge. So you can imagine how much fiber per
[56:11] can imagine how much fiber per millimeter you'll need uh as you scale
[56:14] millimeter you'll need uh as you scale right and that technology is very
[56:16] right and that technology is very complex on how do you align how do you
[56:18] complex on how do you align how do you manage that whole uh uh uh in the real
[56:22] manage that whole uh uh uh in the real production environment. Sandy, please
[56:25] production environment. Sandy, please stay with us because we're going to have
[56:26] stay with us because we're going to have a couple of very interesting companies
[56:29] a couple of very interesting companies that want to ask you things, but I want
[56:31] that want to ask you things, but I want them to ask when they're giving the
[56:32] them to ask when they're giving the short presentation. But now, Sandep, if
[56:35] short presentation. But now, Sandep, if we are thinking about the scaling up,
[56:37] we are thinking about the scaling up, and I know you're working with all the
[56:38] and I know you're working with all the OSATs in the world, I would like to
[56:40] OSATs in the world, I would like to bring to the table one of the ones that
[56:42] bring to the table one of the ones that is making a big impact in the whole
[56:44] is making a big impact in the whole manufacturing of transceivers and CPO
[56:46] manufacturing of transceivers and CPO modules in the world. I'm talking of
[56:48] modules in the world. I'm talking of course about one of the largest
[56:49] course about one of the largest companies in Optica, Jable. Thank you
[56:51] companies in Optica, Jable. Thank you very much, Greg May, from being with us
[56:53] very much, Greg May, from being with us today. Tell us how you can help us, but
[56:55] today. Tell us how you can help us, but most important, how the the 185 people
[56:58] most important, how the the 185 people in the room can help you. The floor is
[56:59] in the room can help you. The floor is yours.
[57:01] yours. >> Hey. Hi. Can you hear me? All right.
[57:03] >> Hey. Hi. Can you hear me? All right. >> Loud and clear.
[57:04] >> Loud and clear. >> Okay, perfect. Just one moment, please.
[57:08] Okay. Just going to share my charts.
[57:16] Okay. Hopefully you can see my charts
[57:18] Okay. Hopefully you can see my charts now.
[57:19] now. >> Yes, we can.
[57:20] >> Yes, we can. >> Okay. So, thank you for the the leadup
[57:23] >> Okay. So, thank you for the the leadup and and actually the uh and I hope
[57:25] and and actually the uh and I hope you're seeing the right one. You're not
[57:26] you're seeing the right one. You're not seeing the preview. You're seeing the
[57:28] seeing the preview. You're seeing the proper
[57:29] proper >> We need to we need to do a display
[57:30] >> We need to we need to do a display settings uh change.
[57:33] settings uh change. >> Yes. Yeah.
[57:35] >> Yes. Yeah. Okay. Here we go. Um you know, I think I
[57:38] Okay. Here we go. Um you know, I think I I think my positioning in the
[57:39] I think my positioning in the presentations is is great because uh the
[57:43] presentations is is great because uh the previous uh speakers have teed up quite
[57:45] previous uh speakers have teed up quite a number of problems that are very
[57:46] a number of problems that are very familiar to us at Jable. Uh also Jable
[57:50] familiar to us at Jable. Uh also Jable is unique because not only are we a
[57:52] is unique because not only are we a manufacturer, a contract manufacturer.
[57:55] manufacturer, a contract manufacturer. So we are capable of making most of the
[57:57] So we are capable of making most of the products that have been described
[57:58] products that have been described earlier uh we have our own products as
[58:00] earlier uh we have our own products as well. So we are uh we share your pain.
[58:04] well. So we are uh we share your pain. We understand uh a lot of the problem
[58:06] We understand uh a lot of the problem statements that have been brought up so
[58:08] statements that have been brought up so far. Um so Jable Whoops, sorry. So Jable
[58:11] far. Um so Jable Whoops, sorry. So Jable just to flash up what we are uh is a $30
[58:15] just to flash up what we are uh is a $30 billion uh global behemoth uh
[58:18] billion uh global behemoth uh manufacturing all over the world. Uh we
[58:21] manufacturing all over the world. Uh we are in the uh intelligent infrastructure
[58:23] are in the uh intelligent infrastructure business units. So out of the 30 billion
[58:26] business units. So out of the 30 billion that we uh revenue last year uh
[58:28] that we uh revenue last year uh intelligent infrastructure was 3.7 uh
[58:31] intelligent infrastructure was 3.7 uh and thanks to AI as we were all seeing
[58:33] and thanks to AI as we were all seeing it's growing at uh grew at 34% last
[58:36] it's growing at uh grew at 34% last year. Um so it it is giving us numbers
[58:41] year. Um so it it is giving us numbers uh giving us quite substantial issues
[58:43] uh giving us quite substantial issues with scaling up. Um [clears throat] so
[58:45] with scaling up. Um [clears throat] so what Jable does is we don't we're not
[58:47] what Jable does is we don't we're not involved in designing the uh the pick uh
[58:51] involved in designing the uh the pick uh or fabricating wafers. We take the
[58:53] or fabricating wafers. We take the wafers in at the head end of our
[58:56] wafers in at the head end of our process. So we have wafer level
[58:58] process. So we have wafer level processing for silicon fatonic both uh
[59:00] processing for silicon fatonic both uh 12 in and 8 in. Um and from there we can
[59:04] 12 in and 8 in. Um and from there we can take in uh we can do the testing
[59:06] take in uh we can do the testing function. We can back grind we have
[59:08] function. We can back grind we have dicing. So be and that's uh we can move
[59:11] dicing. So be and that's uh we can move from there to flip chip uh the fiber or
[59:14] from there to flip chip uh the fiber or lens attach processes lit attach and
[59:16] lens attach processes lit attach and further functional testing. Um the nice
[59:20] further functional testing. Um the nice part of this is that can be all under
[59:21] part of this is that can be all under one roof. So there's no need to uh
[59:24] one roof. So there's no need to uh manufacture by FedEx or to ship product
[59:26] manufacture by FedEx or to ship product around the world to get it processed. Uh
[59:29] around the world to get it processed. Uh we can do that within one roof. And then
[59:31] we can do that within one roof. And then going forward from there in more of our
[59:33] going forward from there in more of our traditional business uh we can uh put
[59:36] traditional business uh we can uh put those uh optical engines or whatever
[59:38] those uh optical engines or whatever photonic parts onto circuit boards and
[59:41] photonic parts onto circuit boards and also do the electromechanical assembly.
[59:44] also do the electromechanical assembly. Okay. So that's all within Jable's
[59:45] Okay. So that's all within Jable's remitt. Uh the photonics area is uh just
[59:49] remitt. Uh the photonics area is uh just over 10 years old uh but growing very
[59:51] over 10 years old uh but growing very rapidly.
[59:53] rapidly. Okay. Um most of the pe audience knows
[59:56] Okay. Um most of the pe audience knows this already. So some of the
[59:57] this already. So some of the applications we deal with uh are
[59:59] applications we deal with uh are transceivers. We uh we manufacture
[01:00:02] transceivers. We uh we manufacture products for other uh we have many
[01:00:05] products for other uh we have many customers that we make transceivers for
[01:00:07] customers that we make transceivers for but we also have our own product line uh
[01:00:09] but we also have our own product line uh from the uh acquiring the Intel product
[01:00:12] from the uh acquiring the Intel product line about uh three years ago. Uh so we
[01:00:15] line about uh three years ago. Uh so we have our own going out. Uh we're
[01:00:17] have our own going out. Uh we're involved with co-ackaged optics and
[01:00:19] involved with co-ackaged optics and products from that realm. Uh quantum and
[01:00:22] products from that realm. Uh quantum and uh LAR. So some of the applications that
[01:00:24] uh LAR. So some of the applications that were discussed earlier um those of us
[01:00:27] were discussed earlier um those of us who've been in the industry long enough
[01:00:28] who've been in the industry long enough remember the dotcom boom of the late 90s
[01:00:31] remember the dotcom boom of the late 90s and how things how crazy that got and
[01:00:34] and how things how crazy that got and the the kind of challenges it portrayed.
[01:00:37] the the kind of challenges it portrayed. Uh now we're in AAI. So basically we're
[01:00:40] Uh now we're in AAI. So basically we're back seeing it again uh but this time
[01:00:42] back seeing it again uh but this time with a lot more I would say a lot more
[01:00:45] with a lot more I would say a lot more vigor and a lot more investment a lot of
[01:00:47] vigor and a lot more investment a lot of real money in this uh so very exciting
[01:00:49] real money in this uh so very exciting business for us if we can keep up. So
[01:00:52] business for us if we can keep up. So the the hard part is actually keeping up
[01:00:54] the the hard part is actually keeping up to the demand that's going out there. Um
[01:00:57] to the demand that's going out there. Um I won't dwell too much on this. We we uh
[01:01:00] I won't dwell too much on this. We we uh recognize and acknowledge a lot of the
[01:01:02] recognize and acknowledge a lot of the uh packaging schemes that were presented
[01:01:04] uh packaging schemes that were presented earlier. This is a overly simply
[01:01:06] earlier. This is a overly simply simplified view of that. Um and fiber
[01:01:10] simplified view of that. Um and fiber attach we also recognize that. Uh so we
[01:01:14] attach we also recognize that. Uh so we we have been involved with fiber attach
[01:01:16] we have been involved with fiber attach uh for a very long time now relatively
[01:01:19] uh for a very long time now relatively speaking starting with single fiber and
[01:01:22] speaking starting with single fiber and coherent type of applications. Uh but
[01:01:24] coherent type of applications. Uh but now we're getting into uh into fiber
[01:01:27] now we're getting into uh into fiber array units and parallel fibers. Uh
[01:01:30] array units and parallel fibers. Uh there's a number of schemes for that as
[01:01:31] there's a number of schemes for that as we all know. Uh so there's edge coupling
[01:01:33] we all know. Uh so there's edge coupling but there's also uh V-groove. Um we are
[01:01:38] but there's also uh V-groove. Um we are seeing uh active and also passive. I
[01:01:41] seeing uh active and also passive. I think the uh with a passive there's a
[01:01:44] think the uh with a passive there's a simplification that comes in without
[01:01:46] simplification that comes in without having to power the unit, without having
[01:01:47] having to power the unit, without having to calibrate detectors, uh without
[01:01:50] to calibrate detectors, uh without having to do a lot of the overhead. Uh
[01:01:52] having to do a lot of the overhead. Uh but in when it comes to actually
[01:01:54] but in when it comes to actually manipulating the fiber array and
[01:01:56] manipulating the fiber array and bringing it into position, um and then
[01:01:59] bringing it into position, um and then curing, uh those are can be quite
[01:02:02] curing, uh those are can be quite lengthy even with passive alignment. So
[01:02:04] lengthy even with passive alignment. So we we do have a a big challenge here on
[01:02:07] we we do have a a big challenge here on and a big capacity bottleneck when it
[01:02:09] and a big capacity bottleneck when it comes to alignment. this for some of the
[01:02:11] comes to alignment. this for some of the volumes that we're seeing uh we're
[01:02:14] volumes that we're seeing uh we're getting many many many alignment
[01:02:16] getting many many many alignment machines required to keep up with it if
[01:02:18] machines required to keep up with it if we don't find a way to break the mold
[01:02:21] we don't find a way to break the mold and come up with a different way. Um so
[01:02:24] and come up with a different way. Um so as a Jel's core compy is volume
[01:02:27] as a Jel's core compy is volume manufacturing. So we really do
[01:02:29] manufacturing. So we really do understand what it takes to get
[01:02:31] understand what it takes to get something out the door in high volume
[01:02:32] something out the door in high volume consistently. uh in something with like
[01:02:36] consistently. uh in something with like photonic advanced packaging, uh yield is
[01:02:40] photonic advanced packaging, uh yield is a real potential issue because you're
[01:02:42] a real potential issue because you're dealing with so many different adhesives
[01:02:44] dealing with so many different adhesives and different manual steps. Adhesives
[01:02:46] and different manual steps. Adhesives are actually quite challenging to
[01:02:48] are actually quite challenging to manage. Uh everything from expiry date
[01:02:51] manage. Uh everything from expiry date to supply chain uh to how they're
[01:02:53] to supply chain uh to how they're handled, how they're stored. Um
[01:02:56] handled, how they're stored. Um automation is critically important. Um
[01:02:58] automation is critically important. Um but it in photonic packaging it runs
[01:03:01] but it in photonic packaging it runs crunchy because loads and unloading
[01:03:03] crunchy because loads and unloading especially are quite difficult to do uh
[01:03:06] especially are quite difficult to do uh automated in automated fashion. So there
[01:03:08] automated in automated fashion. So there can be if you're not careful uh have a
[01:03:10] can be if you're not careful uh have a lot of manual manipulation.
[01:03:13] lot of manual manipulation. um I don't think we spend enough time
[01:03:15] um I don't think we spend enough time really thinking about all the materials
[01:03:17] really thinking about all the materials that go into advanced packaging um to
[01:03:20] that go into advanced packaging um to make a robust product and it can take a
[01:03:24] make a robust product and it can take a number of iterations and experiments to
[01:03:25] number of iterations and experiments to actually figure out exactly what you
[01:03:27] actually figure out exactly what you want to do there to meet the robustness
[01:03:30] want to do there to meet the robustness as well as the manufactur manufacturing
[01:03:32] as well as the manufactur manufacturing volume objective. Um we see this with
[01:03:36] volume objective. Um we see this with work instructions the workforcees that
[01:03:38] work instructions the workforcees that we need. we're in big scale up. You
[01:03:40] we need. we're in big scale up. You don't have time to train a big
[01:03:41] don't have time to train a big workforce. You don't have time to deal
[01:03:44] workforce. You don't have time to deal with churn in the workforce. So, you
[01:03:46] with churn in the workforce. So, you have to have very clear instructions to
[01:03:47] have to have very clear instructions to your operations personnel. And then
[01:03:50] your operations personnel. And then finally, as as mentioned previously by
[01:03:52] finally, as as mentioned previously by some others, um production test is
[01:03:55] some others, um production test is really important and catching at this
[01:03:57] really important and catching at this earliest point of value ad. getting your
[01:04:00] earliest point of value ad. getting your production test to detect a build error
[01:04:02] production test to detect a build error or a defect early before you start
[01:04:05] or a defect early before you start adding in expensive parts like DSPs,
[01:04:08] adding in expensive parts like DSPs, lasers, and other parts like that. Um,
[01:04:11] lasers, and other parts like that. Um, so the bottlenecks kind of run in
[01:04:13] so the bottlenecks kind of run in parallel with the with the volume. So,
[01:04:16] parallel with the with the volume. So, um, the message here is to really you
[01:04:20] um, the message here is to really you don't have a lot of time, but the time
[01:04:22] don't have a lot of time, but the time that you have should really be spent in
[01:04:24] that you have should really be spent in high quality figuring out your prototype
[01:04:26] high quality figuring out your prototype and your pilot product. So spend the
[01:04:29] and your pilot product. So spend the time, invest the time and energy to get
[01:04:31] time, invest the time and energy to get that right and that will pay dividends
[01:04:33] that right and that will pay dividends in the future. If you don't do that,
[01:04:36] in the future. If you don't do that, you're going to proliferate uh problems
[01:04:38] you're going to proliferate uh problems into your mass production, end up with a
[01:04:40] into your mass production, end up with a lot of yield issues.
[01:04:43] lot of yield issues. Okay. So the approach that we do here
[01:04:45] Okay. So the approach that we do here with JBLE is that we take a DFM DFT
[01:04:49] with JBLE is that we take a DFM DFT philosophy. Um and what we're we've done
[01:04:53] philosophy. Um and what we're we've done I think which is fairly unique in the
[01:04:54] I think which is fairly unique in the industry is that here in Ottawa we have
[01:04:56] industry is that here in Ottawa we have an NPI environment with a lot of the
[01:04:58] an NPI environment with a lot of the same machines that you have would see in
[01:05:01] same machines that you have would see in a volume environment. Okay. So we have
[01:05:03] a volume environment. Okay. So we have our fiber attach we have die bonder we
[01:05:06] our fiber attach we have die bonder we have pick and place we have refflow in a
[01:05:09] have pick and place we have refflow in a captive NPI facility. So this gives you
[01:05:11] captive NPI facility. So this gives you the time to do your experimentation
[01:05:14] the time to do your experimentation um without interfering and uh competing
[01:05:17] um without interfering and uh competing with mass production in an environment
[01:05:19] with mass production in an environment that has where you're surrounded by
[01:05:21] that has where you're surrounded by subject matter experts. Um and what
[01:05:24] subject matter experts. Um and what we've been finding is that if we work
[01:05:26] we've been finding is that if we work very closely with the customers but also
[01:05:29] very closely with the customers but also with quality assurance labs so rapid uh
[01:05:32] with quality assurance labs so rapid uh turnaround of uh imaging and sectioning
[01:05:35] turnaround of uh imaging and sectioning in particular it's very important to to
[01:05:38] in particular it's very important to to drive and iterate to a a robust
[01:05:40] drive and iterate to a a robust solution. You know once you you have to
[01:05:43] solution. You know once you you have to figure out things like your temperature
[01:05:45] figure out things like your temperature profiles to get this integrity of your
[01:05:47] profiles to get this integrity of your solder correct make sure it's void free.
[01:05:50] solder correct make sure it's void free. Um we in Ottawa have a a fluxless reflow
[01:05:54] Um we in Ottawa have a a fluxless reflow system. Uh so instead of using chemical
[01:05:57] system. Uh so instead of using chemical flux we use a formic acid vapor that
[01:06:00] flux we use a formic acid vapor that reduces residue and allows uh uh keeps
[01:06:04] reduces residue and allows uh uh keeps voiding out of underfill and also
[01:06:06] voiding out of underfill and also produces uh clean photonic parts that
[01:06:09] produces uh clean photonic parts that are easy for that are uh less obstructed
[01:06:12] are easy for that are uh less obstructed with particles when it comes to line and
[01:06:14] with particles when it comes to line and attach. Um so we work closely with our
[01:06:17] attach. Um so we work closely with our customers. uh we apply I think you'd
[01:06:20] customers. uh we apply I think you'd recognize this similar to six sigma uh
[01:06:22] recognize this similar to six sigma uh we design we work collaboratively with
[01:06:24] we design we work collaboratively with the design uh we improve we measure the
[01:06:27] the design uh we improve we measure the outcome and we we circle back so as we
[01:06:30] outcome and we we circle back so as we we iterate to improve um and the whole
[01:06:33] we iterate to improve um and the whole goal is to get what we know very well
[01:06:35] goal is to get what we know very well from other industries is to get a lineup
[01:06:38] from other industries is to get a lineup with bill documentation bill of
[01:06:40] with bill documentation bill of materials
[01:06:42] materials uh work instructions everything uh test
[01:06:45] uh work instructions everything uh test methods that actually pinpoint where the
[01:06:47] methods that actually pinpoint where the source of the problem is get all that
[01:06:49] source of the problem is get all that out of the way and then phase it into
[01:06:51] out of the way and then phase it into the volume factory.
[01:06:54] the volume factory. I think if you don't get these if you
[01:06:56] I think if you don't get these if you don't plant the right seeds, you don't
[01:06:57] don't plant the right seeds, you don't get the right result at the end. So,
[01:06:59] get the right result at the end. So, we're uh we've really embraced that
[01:07:01] we're uh we've really embraced that concept.
[01:07:03] concept. And then finally, uh just a a bit about
[01:07:07] And then finally, uh just a a bit about what we do. So, here in Ottawa, we have
[01:07:08] what we do. So, here in Ottawa, we have a advanced packaging facility. Again, a
[01:07:11] a advanced packaging facility. Again, a captive NPI uh facility with a lot of
[01:07:13] captive NPI uh facility with a lot of the processes that you would recognize.
[01:07:15] the processes that you would recognize. I won't get into the detail there. And
[01:07:17] I won't get into the detail there. And then uh currently in Chable even though
[01:07:21] then uh currently in Chable even though the cor the corporation has probably
[01:07:23] the cor the corporation has probably hundreds of factories the one that we
[01:07:24] hundreds of factories the one that we deal with most is in Malaysia but we are
[01:07:28] deal with most is in Malaysia but we are currently uh bringing out factories in
[01:07:30] currently uh bringing out factories in India and the US as well.
[01:07:34] India and the US as well. >> Okay that's uh my presentation. Thank
[01:07:35] >> Okay that's uh my presentation. Thank you.
[01:07:36] you. >> Thank you very much Greg. Thank you for
[01:07:38] >> Thank you very much Greg. Thank you for being here. We keep receiving fantastic
[01:07:40] being here. We keep receiving fantastic amazing feedback from the from the all
[01:07:43] amazing feedback from the from the all the different big companies that
[01:07:44] the different big companies that collaborate with Jable. Um, is it
[01:07:47] collaborate with Jable. Um, is it limited to to big companies when it
[01:07:49] limited to to big companies when it comes to JOL because most of your
[01:07:50] comes to JOL because most of your customers at least the ones that I know
[01:07:52] customers at least the ones that I know they are typically large companies but I
[01:07:55] they are typically large companies but I think your presentation today said a bit
[01:07:56] think your presentation today said a bit otherwise.
[01:07:57] otherwise. >> No, absolutely. We especially in the NPI
[01:08:00] >> No, absolutely. We especially in the NPI facility we will work with just about
[01:08:02] facility we will work with just about anybody. Um, there is a hook there
[01:08:04] anybody. Um, there is a hook there though like the we're not like uh there
[01:08:07] though like the we're not like uh there are lots of research institutes that
[01:08:09] are lots of research institutes that will help you develop packaging but
[01:08:10] will help you develop packaging but they're disconnected uh from the vast
[01:08:12] they're disconnected uh from the vast production. uh you can get it through
[01:08:14] production. uh you can get it through the the research institute, but us we're
[01:08:17] the the research institute, but us we're our philosophy is to be a gateway to
[01:08:19] our philosophy is to be a gateway to mass production. So the little guys we
[01:08:22] mass production. So the little guys we work with are the ones that we think can
[01:08:23] work with are the ones that we think can get into the factory next that have uh
[01:08:26] get into the factory next that have uh really high potential for volume.
[01:08:29] really high potential for volume. >> We have a lot of questions for you. I'm
[01:08:30] >> We have a lot of questions for you. I'm going to address them quickly. The first
[01:08:32] going to address them quickly. The first one is coming all the way from the
[01:08:33] one is coming all the way from the masters and commanders of microtransfer
[01:08:36] masters and commanders of microtransfer printing acceler. What's on your mind?
[01:08:42] So uh the question I asked in the in the
[01:08:45] So uh the question I asked in the in the chat was um Jabel has an amazing
[01:08:48] chat was um Jabel has an amazing portfolio of capabilities you know all
[01:08:50] portfolio of capabilities you know all the way from you know design wafer you
[01:08:54] the way from you know design wafer you know fabrication packaging um and we
[01:08:57] know fabrication packaging um and we we're looking today at at these real
[01:08:59] we're looking today at at these real challenges in packaging and integration
[01:09:01] challenges in packaging and integration and the question I I posed was where
[01:09:04] and the question I I posed was where does Jabel intend to lead
[01:09:09] where do you want
[01:09:10] where do you want Interesting.
[01:09:11] Interesting. >> Yeah. The the one we look to for solving
[01:09:15] >> Yeah. The the one we look to for solving these problems we're all talking about.
[01:09:16] these problems we're all talking about. >> Well, have the skills.
[01:09:20] >> Well, have the skills. >> There's two prongs to that. We want to
[01:09:22] >> There's two prongs to that. We want to lead in the assembly methods and
[01:09:25] lead in the assembly methods and manufacturing methods. That's our our
[01:09:28] manufacturing methods. That's our our like we're not we're not going to be the
[01:09:30] like we're not we're not going to be the the people that are going to come up
[01:09:31] the people that are going to come up with the next or the the
[01:09:35] with the next or the the RDL or the you know the through silicon
[01:09:38] RDL or the you know the through silicon via and the you know the 3D
[01:09:41] via and the you know the 3D uh the best way to do 3D for example
[01:09:43] uh the best way to do 3D for example we're going to be the guys that we want
[01:09:45] we're going to be the guys that we want to lead on the best uh best practices
[01:09:48] to lead on the best uh best practices for adhesives you know kind of almost
[01:09:51] for adhesives you know kind of almost the boring stuff the adhesives the
[01:09:52] the boring stuff the adhesives the solders the you know the curing the
[01:09:55] solders the you know the curing the handling
[01:09:56] handling uh the you know the reliability behind
[01:09:58] uh the you know the reliability behind that, the quality behind that.
[01:10:02] that, the quality behind that. That makes sense.
[01:10:04] That makes sense. >> What what do you think about the
[01:10:05] >> What what do you think about the previous uh presenter discussion about
[01:10:08] previous uh presenter discussion about trying to find a single epoxy for all
[01:10:11] trying to find a single epoxy for all optical engines?
[01:10:12] optical engines? >> Well, that I think that would be the the
[01:10:14] >> Well, that I think that would be the the silver bullet, right? If we could find
[01:10:16] silver bullet, right? If we could find that, if there were such a thing, no,
[01:10:18] that, if there were such a thing, no, I'm I'm not the technical expert on
[01:10:21] I'm I'm not the technical expert on that, so maybe there is there is hope
[01:10:23] that, so maybe there is there is hope for that. But if you had a very fast
[01:10:25] for that. But if you had a very fast cure, like the ultimate would be a very
[01:10:27] cure, like the ultimate would be a very fast cure non-shrink
[01:10:29] fast cure non-shrink robust epoxy that robust against damp
[01:10:31] robust epoxy that robust against damp heat, shock, vibration, you would have
[01:10:34] heat, shock, vibration, you would have uh you'd have a a major advantage if you
[01:10:38] uh you'd have a a major advantage if you had something like that.
[01:10:40] had something like that. >> If anyone has a fast cure no shrink
[01:10:42] >> If anyone has a fast cure no shrink epoxy, please write so in the chat.
[01:10:44] epoxy, please write so in the chat. Thank you very much.
[01:10:45] Thank you very much. >> No shrink,
[01:10:46] >> No shrink, >> damp heat resistant, shock resistant.
[01:10:49] >> damp heat resistant, shock resistant. >> Thank you.
[01:10:50] >> Thank you. >> There is another question for you, Dan.
[01:10:52] >> There is another question for you, Dan. stay with us because you are speaking
[01:10:53] stay with us because you are speaking next but Greg there is a question coming
[01:10:56] next but Greg there is a question coming all the way from the Netherlands John
[01:10:57] all the way from the Netherlands John from Epiphany what's on your mind
[01:11:00] from Epiphany what's on your mind >> yes thanks a lot so you said you're
[01:11:02] >> yes thanks a lot so you said you're already handling with small companies
[01:11:04] already handling with small companies that you where you believe right they
[01:11:06] that you where you believe right they actually can scale up and that's
[01:11:07] actually can scale up and that's important time to market that we don't
[01:11:09] important time to market that we don't need to have to redesign so how do we
[01:11:11] need to have to redesign so how do we design in order to seamlessly go to
[01:11:14] design in order to seamlessly go to manufacturing
[01:11:15] manufacturing >> early engagement
[01:11:17] >> early engagement >> and how do we engage
[01:11:19] >> and how do we engage >> well that's a good one that's It's
[01:11:22] >> well that's a good one that's It's always the challenge, right? Um I think
[01:11:25] always the challenge, right? Um I think you you you approach uh you approach I I
[01:11:28] you you you approach uh you approach I I think the gateway in Jable would be our
[01:11:30] think the gateway in Jable would be our salespeople just to uh get known to
[01:11:32] salespeople just to uh get known to them. They're the ones that do the
[01:11:33] them. They're the ones that do the conferences and and are always uh out
[01:11:36] conferences and and are always uh out with the customers. Get involved with
[01:11:38] with the customers. Get involved with them and and see if we can start an
[01:11:40] them and and see if we can start an early engagement like your concept and
[01:11:42] early engagement like your concept and then see uh see where we can help on
[01:11:45] then see uh see where we can help on DFM. You you you really have to nail DFM
[01:11:49] DFM. You you you really have to nail DFM and DFT early on. Um
[01:11:52] and DFT early on. Um >> yeah 100%. That's why I'm asking.
[01:11:54] >> yeah 100%. That's why I'm asking. >> No, not Yeah. Yeah. Yeah. No, it's it's
[01:11:55] >> No, not Yeah. Yeah. Yeah. No, it's it's I don't know. I don't have a very I wish
[01:11:57] I don't know. I don't have a very I wish I had a better answer for you, but I
[01:11:58] I had a better answer for you, but I think that's the best way. Like there is
[01:12:00] think that's the best way. Like there is a lot of activity in the marketplace. We
[01:12:02] a lot of activity in the marketplace. We do get approached by a lot of people.
[01:12:04] do get approached by a lot of people. >> One more question coming all the way
[01:12:05] >> One more question coming all the way from India. There's a new startup there
[01:12:07] from India. There's a new startup there called Qualon. They do silicon photonic
[01:12:09] called Qualon. They do silicon photonic gas sensors. What is in your mind?
[01:12:12] gas sensors. What is in your mind? Sandep.
[01:12:14] Sandep. >> Uh hi Greg. Uh can you hear me?
[01:12:16] >> Uh hi Greg. Uh can you hear me? >> Hi.
[01:12:16] >> Hi. >> Loud and clear.
[01:12:18] >> Loud and clear. >> Yep.
[01:12:18] >> Yep. >> Yeah. Perfect. Uh so uh we are basically
[01:12:21] >> Yeah. Perfect. Uh so uh we are basically designing pix for gas sensing
[01:12:23] designing pix for gas sensing applications. So we succeeded in uh
[01:12:25] applications. So we succeeded in uh designing passive and active photonic IC
[01:12:28] designing passive and active photonic IC but one of the challenge that we are
[01:12:29] but one of the challenge that we are facing from the last two years is how do
[01:12:31] facing from the last two years is how do we package the systems because uh
[01:12:34] we package the systems because uh everywhere we go we found the packaging
[01:12:36] everywhere we go we found the packaging for telecom or telecom but when it comes
[01:12:38] for telecom or telecom but when it comes to gas sensing because it requires
[01:12:41] to gas sensing because it requires inlets outlets and flow rate management
[01:12:44] inlets outlets and flow rate management you know pressure management all this
[01:12:46] you know pressure management all this becomes very much challenging in terms
[01:12:47] becomes very much challenging in terms of uh you know fixer sensing
[01:12:49] of uh you know fixer sensing application. Uh I think I just got a
[01:12:52] application. Uh I think I just got a reply from Photon first but yeah just
[01:12:54] reply from Photon first but yeah just want to hear from your end user. Thank
[01:12:56] want to hear from your end user. Thank you.
[01:12:57] you. >> Um yeah I I don't think that we would
[01:12:59] >> Um yeah I I don't think that we would provide a design service for that.
[01:13:01] provide a design service for that. That's quite a a unique application that
[01:13:03] That's quite a a unique application that would have some very interesting uh like
[01:13:07] would have some very interesting uh like considerations for flow for example
[01:13:09] considerations for flow for example that's not I would say a core competency
[01:13:11] that's not I would say a core competency of ours. we would be more able to help
[01:13:14] of ours. we would be more able to help you with uh making the mechanical part
[01:13:17] you with uh making the mechanical part like is you know molding material
[01:13:19] like is you know molding material choices things like that any kind of
[01:13:21] choices things like that any kind of machining uh any kind of uh assembly of
[01:13:24] machining uh any kind of uh assembly of that. So I think you you need to find a
[01:13:28] that. So I think you you need to find a third party that can do help you with a
[01:13:30] third party that can do help you with a with the unique part of your
[01:13:32] with the unique part of your application.
[01:13:33] application. >> Sandiv there are many people in the room
[01:13:35] >> Sandiv there are many people in the room but apparently one of them already reach
[01:13:37] but apparently one of them already reach out to you that they can really help you
[01:13:39] out to you that they can really help you on this. And now we go to the masters
[01:13:42] on this. And now we go to the masters and commanders of micro printing again.
[01:13:44] and commanders of micro printing again. Dan, please tell us shortly what you can
[01:13:45] Dan, please tell us shortly what you can do for others and others can do for you.
[01:13:49] do for others and others can do for you. >> Well, what I can do is introduce our
[01:13:51] >> Well, what I can do is introduce our CEO, Peter Smith, who will give give the
[01:13:54] CEO, Peter Smith, who will give give the presentation. Um,
[01:13:56] presentation. Um, >> I did not see Peter. Great to see you,
[01:13:58] >> I did not see Peter. Great to see you, Peter.
[01:13:58] Peter. >> Uh, so, you know, Accelan has been
[01:14:00] >> Uh, so, you know, Accelan has been around for I don't know 13 or 14 years.
[01:14:02] around for I don't know 13 or 14 years. >> I have one slide to share with us.
[01:14:05] >> I have one slide to share with us. >> Yeah.
[01:14:05] >> Yeah. >> So, I'll hand it over to Pet. Can you
[01:14:08] >> So, I'll hand it over to Pet. Can you see my Can you see my slides?
[01:14:09] see my Can you see my slides? >> Not yet.
[01:14:10] >> Not yet. >> Not yet.
[01:14:11] >> Not yet. >> Okay, just let me share here. Um, should
[01:14:14] >> Okay, just let me share here. Um, should be coming through loud and clear and and
[01:14:16] be coming through loud and clear and and up on the screen now.
[01:14:17] up on the screen now. >> I see it's it's loading. There it is.
[01:14:19] >> I see it's it's loading. There it is. >> Okay, great. Thank you very much. Um,
[01:14:21] >> Okay, great. Thank you very much. Um, Jose, John, it's great to be here. So,
[01:14:24] Jose, John, it's great to be here. So, I'll give you a very quick run through
[01:14:25] I'll give you a very quick run through on on um microtransfer printing
[01:14:28] on on um microtransfer printing technology and where Excel print is
[01:14:29] technology and where Excel print is today um in um providing manufacturing
[01:14:33] today um in um providing manufacturing solutions using MTP. So
[01:14:37] solutions using MTP. So micro transfer print is an incredibly
[01:14:38] micro transfer print is an incredibly versatile heterogeneous integration uh
[01:14:40] versatile heterogeneous integration uh mechanism um and it's being deployed
[01:14:44] mechanism um and it's being deployed right now addressing the major um
[01:14:47] right now addressing the major um concerns of key industries that we've
[01:14:49] concerns of key industries that we've all been talking about now primarily
[01:14:51] all been talking about now primarily high performance computing AI photonix
[01:14:53] high performance computing AI photonix communications when we say that it's
[01:14:56] communications when we say that it's it's a very diverse um integration
[01:14:59] it's a very diverse um integration mechanism what we mean by that is that
[01:15:01] mechanism what we mean by that is that we're not just talking about integration
[01:15:03] we're not just talking about integration of of devices like lasers and and um
[01:15:06] of of devices like lasers and and um photo detectors um but also other
[01:15:09] photo detectors um but also other materials um lithium niabate for example
[01:15:11] materials um lithium niabate for example and even um passive optics. So you can
[01:15:15] and even um passive optics. So you can have many many source wafers being
[01:15:16] have many many source wafers being transferred and integrated onto target
[01:15:18] transferred and integrated onto target wafers using the platform and that's
[01:15:20] wafers using the platform and that's incredibly powerful when you consider
[01:15:22] incredibly powerful when you consider all of the 3D integration uh challenges
[01:15:25] all of the 3D integration uh challenges that the industries are facing today.
[01:15:28] that the industries are facing today. So
[01:15:30] So what are the advantages? Um key aspect
[01:15:33] what are the advantages? Um key aspect of all of this is placement accuracy. So
[01:15:35] of all of this is placement accuracy. So very tight uh pitch but maintaining
[01:15:37] very tight uh pitch but maintaining precision uh while transferring
[01:15:40] precision uh while transferring thousands and thousands of of devices.
[01:15:42] thousands and thousands of of devices. So we're well submicron currently at
[01:15:45] So we're well submicron currently at around um 300 nanometers today but with
[01:15:48] around um 300 nanometers today but with research partners for example with
[01:15:50] research partners for example with Tindle and we're looking at sub 100
[01:15:52] Tindle and we're looking at sub 100 nanometer um and hoping to advance that
[01:15:55] nanometer um and hoping to advance that to a production stage later. Um so uh
[01:15:58] to a production stage later. Um so uh placement accuracy at volume is is a a
[01:16:01] placement accuracy at volume is is a a key differentiator here that provides
[01:16:02] key differentiator here that provides you with high density of course um but
[01:16:05] you with high density of course um but also precision delivers low parasitics
[01:16:07] also precision delivers low parasitics which um what you benefit there from of
[01:16:10] which um what you benefit there from of course is you maintain the performance
[01:16:11] course is you maintain the performance of the devices that are being printed
[01:16:13] of the devices that are being printed and the and the target application. Uh
[01:16:16] and the and the target application. Uh so um it's it's incredibly important
[01:16:18] so um it's it's incredibly important from a precision perspective to achieve
[01:16:20] from a precision perspective to achieve that and it also that placement accuracy
[01:16:24] that and it also that placement accuracy um has the additional benefit of um
[01:16:27] um has the additional benefit of um ensuring that you uh have low losses
[01:16:30] ensuring that you uh have low losses improved thermal uh profile performance
[01:16:33] improved thermal uh profile performance and that in turn producing higher
[01:16:35] and that in turn producing higher reliability.
[01:16:38] reliability. Another major uh benefit of the of the
[01:16:40] Another major uh benefit of the of the platform is in reduced material usage.
[01:16:44] platform is in reduced material usage. So when you're using MTP um with
[01:16:46] So when you're using MTP um with incredibly expensive wafers in in
[01:16:48] incredibly expensive wafers in in photonix for example, MTP has the
[01:16:51] photonix for example, MTP has the advantage that essentially you're you're
[01:16:52] advantage that essentially you're you're removing the device layer um which
[01:16:55] removing the device layer um which leaves you the um expensive uh source
[01:16:57] leaves you the um expensive uh source material wafer um for recycling and
[01:17:00] material wafer um for recycling and reuse um which of course is incredibly
[01:17:02] reuse um which of course is incredibly incredibly important for uh meeting the
[01:17:05] incredibly important for uh meeting the uh cost um metrics that everybody wants
[01:17:08] uh cost um metrics that everybody wants to achieve with um integration
[01:17:09] to achieve with um integration challenges. And critically, this is all
[01:17:13] challenges. And critically, this is all scalable for high yield and high margin.
[01:17:17] scalable for high yield and high margin. So you have a solution now which
[01:17:19] So you have a solution now which delivers on all of the performance
[01:17:21] delivers on all of the performance requirements but can meet the really
[01:17:23] requirements but can meet the really really stringent um margin and and um uh
[01:17:26] really stringent um margin and and um uh pricing u required uh by these
[01:17:29] pricing u required uh by these applications
[01:17:31] applications and it's supported by a growing network
[01:17:33] and it's supported by a growing network of uh of supply chain uh partners. Um
[01:17:36] of uh of supply chain uh partners. Um it's been discussed already about end to
[01:17:38] it's been discussed already about end to end how really important it is that the
[01:17:40] end how really important it is that the supply chain is connected for these
[01:17:42] supply chain is connected for these applications. So um micro transfer print
[01:17:45] applications. So um micro transfer print with acceler will deliver on that as
[01:17:46] with acceler will deliver on that as well. So just to give you an example um
[01:17:50] well. So just to give you an example um of um a current example of very high um
[01:17:54] of um a current example of very high um volume production here and we're proud
[01:17:56] volume production here and we're proud to say that we've been a technology
[01:17:57] to say that we've been a technology partner um with Seagate H for a long
[01:18:00] partner um with Seagate H for a long time in this area and Seate Seagate have
[01:18:03] time in this area and Seate Seagate have actually become um a leading integrator
[01:18:06] actually become um a leading integrator of lasers which is not what you might
[01:18:08] of lasers which is not what you might expect necessarily when we're all
[01:18:10] expect necessarily when we're all talking about high-speed comms. Um but
[01:18:13] talking about high-speed comms. Um but they have been an early adopter um in
[01:18:15] they have been an early adopter um in the use of lasers and in the dense
[01:18:18] the use of lasers and in the dense integration of lasers in a very very
[01:18:20] integration of lasers in a very very challenging application in hard disk
[01:18:22] challenging application in hard disk drive with all of the the yield and
[01:18:25] drive with all of the the yield and reliability challenges uh that hard
[01:18:27] reliability challenges uh that hard drive faces um for um hyperscalers. Uh
[01:18:31] drive faces um for um hyperscalers. Uh so um what you're looking at is is a
[01:18:34] so um what you're looking at is is a capability of printing millions and
[01:18:37] capability of printing millions and millions of lasers um with all of the um
[01:18:40] millions of lasers um with all of the um benefits that you uh you need in in a
[01:18:42] benefits that you uh you need in in a very very tough application.
[01:18:46] So I very very quick run through um love
[01:18:49] So I very very quick run through um love to hear from everybody um directly in
[01:18:51] to hear from everybody um directly in the questions on the chat. Uh also just
[01:18:54] the questions on the chat. Uh also just um reach out directly to us. Uh we have
[01:18:56] um reach out directly to us. Uh we have print centers operational in the US,
[01:18:58] print centers operational in the US, Europe and in Asia. um and able to take
[01:19:01] Europe and in Asia. um and able to take on um a lot of the uh the applications
[01:19:04] on um a lot of the uh the applications that are um driving
[01:19:07] that are um driving >> [clears throat]
[01:19:07] >> [clears throat] >> um the uh the business in in comms
[01:19:10] >> um the uh the business in in comms particularly at this point.
[01:19:14] particularly at this point. >> Great. Thank you so much uh Peter and
[01:19:16] >> Great. Thank you so much uh Peter and Dan. So uh the Seagate hammer example he
[01:19:20] Dan. So uh the Seagate hammer example he just showed was was a really powerful
[01:19:22] just showed was was a really powerful demonstration of industrial scale
[01:19:24] demonstration of industrial scale deployment. Uh how important was that
[01:19:26] deployment. Uh how important was that milestone for um Acceler proving that
[01:19:30] milestone for um Acceler proving that microtransfer printing could move beyond
[01:19:32] microtransfer printing could move beyond research into into high volume
[01:19:34] research into into high volume manufacturing?
[01:19:36] manufacturing? >> Yeah, it it's been a journey but
[01:19:39] >> Yeah, it it's been a journey but probably the best kept secret for quite
[01:19:40] probably the best kept secret for quite a while, right? Because it's not it's
[01:19:42] a while, right? Because it's not it's not simply that it's happened this year
[01:19:44] not simply that it's happened this year that um that Seagate is is um uh leading
[01:19:47] that um that Seagate is is um uh leading the charge in this type of integration.
[01:19:49] the charge in this type of integration. It's it's actually been happening for
[01:19:51] It's it's actually been happening for quite a while. Uh albeit um secretly. It
[01:19:55] quite a while. Uh albeit um secretly. It it is it's an important proof point, but
[01:19:57] it is it's an important proof point, but it's not unique in that many of our
[01:20:00] it's not unique in that many of our comm's customers now are beginning to
[01:20:03] comm's customers now are beginning to see those benefits. So it it has moved
[01:20:07] see those benefits. So it it has moved um you could say two years ago really
[01:20:09] um you could say two years ago really into a um high volume production
[01:20:12] into a um high volume production environment and now the telecom and
[01:20:15] environment and now the telecom and dataccom sector is actually benefiting
[01:20:17] dataccom sector is actually benefiting from the proof points that would were
[01:20:19] from the proof points that would were delivered actually by hard disk drive.
[01:20:21] delivered actually by hard disk drive. Even though it sounds like a different
[01:20:23] Even though it sounds like a different application, the challenges are exactly
[01:20:25] application, the challenges are exactly the same. Whether it's a, you know, with
[01:20:28] the same. Whether it's a, you know, with or without adhesive integration, the
[01:20:30] or without adhesive integration, the diversity of materials that have have to
[01:20:32] diversity of materials that have have to be handled, we passed way beyond uh
[01:20:35] be handled, we passed way beyond uh research now into high volume production
[01:20:38] research now into high volume production and it's all on the same tool platform.
[01:20:41] and it's all on the same tool platform. So the the tools that are um in use for
[01:20:44] So the the tools that are um in use for uh research and proof of concept are now
[01:20:47] uh research and proof of concept are now moving all the way through to the volume
[01:20:48] moving all the way through to the volume manufacturing scale.
[01:20:51] manufacturing scale. Thanks Peter. And there's a question in
[01:20:53] Thanks Peter. And there's a question in the chat. Mves Yusfi from Photon First.
[01:20:56] the chat. Mves Yusfi from Photon First. What's on your mind?
[01:20:57] What's on your mind? >> Hi. It seems to be the usual suspect
[01:21:00] >> Hi. It seems to be the usual suspect here. May I see Stephie again? What's
[01:21:02] here. May I see Stephie again? What's your strategy for non- good dyes?
[01:21:03] your strategy for non- good dyes? Because you're implanting dyes inside a
[01:21:06] Because you're implanting dyes inside a larger wafer. Um the non- good die
[01:21:08] larger wafer. Um the non- good die strategy is key to success in packaging.
[01:21:11] strategy is key to success in packaging. Do you do pre-esting? Do you do wafer
[01:21:13] Do you do pre-esting? Do you do wafer directly? Or do you do tested dyes from
[01:21:15] directly? Or do you do tested dyes from a gel pack? Just for curiosity. So
[01:21:18] a gel pack? Just for curiosity. So [laughter] it it's application um
[01:21:20] [laughter] it it's application um dependent um but we work with our um
[01:21:25] dependent um but we work with our um customers to produce the the source
[01:21:28] customers to produce the the source wafers
[01:21:30] wafers um with the um the expectation actually
[01:21:36] um with the um the expectation actually um that we're getting source wafers with
[01:21:38] um that we're getting source wafers with known good die uh available for
[01:21:41] known good die uh available for printing. Uh so in the early stages uh
[01:21:44] printing. Uh so in the early stages uh working with um our um source material
[01:21:48] working with um our um source material providers uh we'll be making sure
[01:21:50] providers uh we'll be making sure actually that when we're getting source
[01:21:52] actually that when we're getting source wafers uh we're in a position to
[01:21:55] wafers uh we're in a position to identify no gun dye and and print them
[01:21:57] identify no gun dye and and print them accordingly.
[01:22:00] >> Okay, farewell. Thank you so much Peter.
[01:22:02] >> Okay, farewell. Thank you so much Peter. Um, next uh up I'd like to introduce
[01:22:05] Um, next uh up I'd like to introduce Stefan Heinman, managing director USA
[01:22:08] Stefan Heinman, managing director USA and VP of sales at Fix Photonix
[01:22:10] and VP of sales at Fix Photonix Assembly. With more than 30 years
[01:22:12] Assembly. With more than 30 years experience across photonix, lasers and
[01:22:14] experience across photonix, lasers and advanced manufacturing, Stefan helps
[01:22:16] advanced manufacturing, Stefan helps lead Fix's work on scalable packaging
[01:22:19] lead Fix's work on scalable packaging and assembly. Stefan, uh, the floor is
[01:22:21] and assembly. Stefan, uh, the floor is yours. Thank you.
[01:22:23] yours. Thank you. >> Thank you, Yan, for the great
[01:22:25] >> Thank you, Yan, for the great introduction. Appreciate it.
[01:22:27] introduction. Appreciate it. >> You can hopefully see my screen.
[01:22:30] >> You can hopefully see my screen. >> Yes.
[01:22:31] >> Yes. Right. Um so yeah what are we fix
[01:22:34] Right. Um so yeah what are we fix assembly is um a startup company or is
[01:22:37] assembly is um a startup company or is is an established packaging foundry uh
[01:22:40] is an established packaging foundry uh for uh integrated photonics circuits and
[01:22:42] for uh integrated photonics circuits and also mems and we are based in the
[01:22:44] also mems and we are based in the Netherlands and also have United States
[01:22:48] Netherlands and also have United States facility. So what do we do? Uh we
[01:22:50] facility. So what do we do? Uh we support all P platforms whether it's
[01:22:52] support all P platforms whether it's silicon photonics in phosphide
[01:22:54] silicon photonics in phosphide lithiumate you name it. Um that cover
[01:22:57] lithiumate you name it. Um that cover the wavelength range from UV to Cban.
[01:23:00] the wavelength range from UV to Cban. While we play a lot in the dataccom
[01:23:01] While we play a lot in the dataccom industry, we also very heavily engaged
[01:23:04] industry, we also very heavily engaged in quantum um and sensing applications.
[01:23:08] in quantum um and sensing applications. Um we take it from prototype to volume.
[01:23:10] Um we take it from prototype to volume. Um we actually in you know we heard from
[01:23:12] Um we actually in you know we heard from Jable a very great presentation. We do
[01:23:15] Jable a very great presentation. We do digital twins and not only design the
[01:23:18] digital twins and not only design the product but we also do digital twin
[01:23:20] product but we also do digital twin assembly. We bring bring it all the way
[01:23:22] assembly. We bring bring it all the way from designing your product DFM DFT all
[01:23:25] from designing your product DFM DFT all the way to uh the digital assembly line.
[01:23:28] the way to uh the digital assembly line. And we have worldleading expertise uh in
[01:23:31] And we have worldleading expertise uh in integrating a lot of different material
[01:23:32] integrating a lot of different material systems together. Active passive
[01:23:35] systems together. Active passive integration just in in one dimension or
[01:23:37] integration just in in one dimension or in two and a half uh dimensions. Um so
[01:23:42] in two and a half uh dimensions. Um so um and when we talk about advanced
[01:23:44] um and when we talk about advanced volume packaging, what do we do in terms
[01:23:46] volume packaging, what do we do in terms of scaling product? Uh as you can we do
[01:23:50] of scaling product? Uh as you can we do we address all these pain points or in a
[01:23:54] we address all these pain points or in a different to a different level that you
[01:23:55] different to a different level that you heard before from the end users. um fast
[01:23:59] heard before from the end users. um fast reliable wafer level processing of
[01:24:02] reliable wafer level processing of tested components of known good dice and
[01:24:04] tested components of known good dice and the core expertise is summarized here um
[01:24:07] the core expertise is summarized here um two two different streams of the core
[01:24:09] two two different streams of the core expertise one is wafer level assembly
[01:24:11] expertise one is wafer level assembly processes that are um a pop of active
[01:24:15] processes that are um a pop of active and passive uh whe whether it's an optic
[01:24:18] and passive uh whe whether it's an optic or whether it's a vixel or a gain medium
[01:24:21] or whether it's a vixel or a gain medium or a photo diet or or another material
[01:24:24] or a photo diet or or another material system we put them on we put dye on
[01:24:27] system we put them on we put dye on wafer in a fully automated way with an
[01:24:30] wafer in a fully automated way with an accuracy of 300 nanometers and below.
[01:24:33] accuracy of 300 nanometers and below. And then that's the process capabilities
[01:24:35] And then that's the process capabilities that we offer today uh on machines such
[01:24:37] that we offer today uh on machines such as the Vikrron Nano. We have uh several
[01:24:40] as the Vikrron Nano. We have uh several of those in our floor and we're actually
[01:24:43] of those in our floor and we're actually running off processes that have been
[01:24:45] running off processes that have been developed um at IMC and other research
[01:24:48] developed um at IMC and other research institutes and we make them ready for
[01:24:50] institutes and we make them ready for taking it in at the tables um and the
[01:24:53] taking it in at the tables um and the Ozets big of the world. We're also
[01:24:56] Ozets big of the world. We're also addressing other issues that you know
[01:24:57] addressing other issues that you know what other what are other means for cost
[01:25:00] what other what are other means for cost reduction in manufacturing and that's
[01:25:02] reduction in manufacturing and that's shown in the lower one where we take uh
[01:25:04] shown in the lower one where we take uh where we go from individual hemetic gold
[01:25:06] where we go from individual hemetic gold box manufacturing to um flowline lead
[01:25:10] box manufacturing to um flowline lead frame assembly with semiatic packaging
[01:25:13] frame assembly with semiatic packaging around.
[01:25:15] around. Um and last but not least um just wanted
[01:25:19] Um and last but not least um just wanted to round up our presentation with one of
[01:25:21] to round up our presentation with one of our sample projects in the datcom
[01:25:23] our sample projects in the datcom industry uh where uh we we do attach
[01:25:26] industry uh where uh we we do attach this fiber attaches in a fully automated
[01:25:29] this fiber attaches in a fully automated way for the different configurations
[01:25:31] way for the different configurations that are out there whether it's a
[01:25:32] that are out there whether it's a vgroove or an edge attach and you see
[01:25:35] vgroove or an edge attach and you see typically process capabilities of these
[01:25:37] typically process capabilities of these different of these different uh two main
[01:25:40] different of these different uh two main streams uh vgroove is a very established
[01:25:43] streams uh vgroove is a very established process that is fully
[01:25:44] process that is fully passive gives you less than 1 dB
[01:25:46] passive gives you less than 1 dB insertion loss per facet and co posted
[01:25:49] insertion loss per facet and co posted the question hey if you have less than
[01:25:51] the question hey if you have less than one and a half dB please raise your hand
[01:25:53] one and a half dB please raise your hand here we are we have we have that we run
[01:25:55] here we are we have we have that we run that in in semi high volume production
[01:25:57] that in in semi high volume production environment with very good process
[01:25:59] environment with very good process capabilities whether it's a V-groove um
[01:26:04] capabilities whether it's a V-groove um the challenge that we have for vgroove
[01:26:05] the challenge that we have for vgroove is that they need special low index
[01:26:07] is that they need special low index clues and we heard a lot about epoxies
[01:26:09] clues and we heard a lot about epoxies if you go to edge attach uh insertion
[01:26:12] if you go to edge attach uh insertion losses are a little higher process cap
[01:26:14] losses are a little higher process cap capabilities are yet still a little
[01:26:15] capabilities are yet still a little lower and that is part of you know good
[01:26:17] lower and that is part of you know good die strategy. Uh what do we see um in
[01:26:20] die strategy. Uh what do we see um in the future for us um to address and this
[01:26:23] the future for us um to address and this is nothing new. We heard that from other
[01:26:24] is nothing new. We heard that from other speakers as well losses uh is always
[01:26:28] speakers as well losses uh is always good better known good dice designed for
[01:26:31] good better known good dice designed for test and faster curing uh but also
[01:26:34] test and faster curing uh but also reflow compatible epoxies. So we're
[01:26:36] reflow compatible epoxies. So we're going we're heavily um investing in um
[01:26:39] going we're heavily um investing in um epoxy-free interfaces uh not only for
[01:26:42] epoxy-free interfaces uh not only for that purpose but also since we're
[01:26:44] that purpose but also since we're pushing into the visible in the UV in
[01:26:46] pushing into the visible in the UV in many other applications we need epoxy
[01:26:49] many other applications we need epoxy for interfaces [gasps] to the point of
[01:26:51] for interfaces [gasps] to the point of FAU supply chain the attach process
[01:26:53] FAU supply chain the attach process itself is probably not the bottleneck
[01:26:54] itself is probably not the bottleneck the cost driver it really is the single
[01:26:58] the cost driver it really is the single mode PIM PM mode fiber area that also
[01:27:01] mode PIM PM mode fiber area that also was mentioned by light mirror in his
[01:27:02] was mentioned by light mirror in his talk in that talk of arranging
[01:27:06] talk in that talk of arranging PM fibers in an array in an fully
[01:27:08] PM fibers in an array in an fully automated way to minimize cost.
[01:27:11] automated way to minimize cost. Thank you very much.
[01:27:13] Thank you very much. >> Thank you Stefan. Great work as always
[01:27:14] >> Thank you Stefan. Great work as always from Fix. Um I see numbers less than 1.5
[01:27:18] from Fix. Um I see numbers less than 1.5 dB there which is uh fantastic which
[01:27:20] dB there which is uh fantastic which meets the spec. I see fairly large error
[01:27:23] meets the spec. I see fairly large error bars there. Can you comment on you know
[01:27:26] bars there. Can you comment on you know how repeatable the process is?
[01:27:29] how repeatable the process is? Yeah. And u absolutely and that's well
[01:27:32] Yeah. And u absolutely and that's well there's always outliers and this is you
[01:27:34] there's always outliers and this is you know the summary of when we start with
[01:27:36] know the summary of when we start with processes all the way through the
[01:27:38] processes all the way through the different stages of the of the process
[01:27:39] different stages of the of the process development proc uh stage development
[01:27:42] development proc uh stage development process. Outliers are there in the
[01:27:44] process. Outliers are there in the beginning and the boxes basically the
[01:27:46] beginning and the boxes basically the color boxes are where we end and so uh
[01:27:49] color boxes are where we end and so uh you see the progress as we go through
[01:27:51] you see the progress as we go through the different phases of our development
[01:27:53] the different phases of our development processes with our partners.
[01:27:56] processes with our partners. >> Thanks Stefan. Uh in the interest of
[01:27:58] >> Thanks Stefan. Uh in the interest of time, I'm going to move on and I'd like
[01:28:00] time, I'm going to move on and I'd like to introduce uh John John Anderson,
[01:28:02] to introduce uh John John Anderson, chief commercial officer of Photon
[01:28:04] chief commercial officer of Photon Bridge. Uh John, please share a slide
[01:28:07] Bridge. Uh John, please share a slide and tell us how your work is addressing
[01:28:08] and tell us how your work is addressing one of the industry's key challenges,
[01:28:10] one of the industry's key challenges, how to efficiently integrate laser
[01:28:13] how to efficiently integrate laser sources into photonic platforms.
[01:28:16] sources into photonic platforms. >> Well, hello everyone. Uh good afternoon.
[01:28:18] >> Well, hello everyone. Uh good afternoon. Uh thanks Jose and John for the for the
[01:28:21] Uh thanks Jose and John for the for the invitation to come share what we're
[01:28:23] invitation to come share what we're doing at Fodon Bridge today and uh
[01:28:25] doing at Fodon Bridge today and uh really enjoying the opportunity to
[01:28:26] really enjoying the opportunity to follow the the great slides that
[01:28:28] follow the the great slides that preceded this my short presentation and
[01:28:30] preceded this my short presentation and helped set the stage. So packaging
[01:28:33] helped set the stage. So packaging lasers and how heterogeneous integration
[01:28:35] lasers and how heterogeneous integration unlocks DWDM power for CPO. I actually
[01:28:38] unlocks DWDM power for CPO. I actually think listening to the presentations
[01:28:40] think listening to the presentations that preceded maybe there's a different
[01:28:41] that preceded maybe there's a different subtitle. I really enjoyed um the the
[01:28:44] subtitle. I really enjoyed um the the presentation from like matter and the
[01:28:46] presentation from like matter and the comment about we need to move from
[01:28:49] comment about we need to move from minute taking minutes to align photonix
[01:28:52] minute taking minutes to align photonix interfaces over different materials to
[01:28:54] interfaces over different materials to seconds. So perhaps an alternative
[01:28:56] seconds. So perhaps an alternative subtitle reflecting on this is how to
[01:28:58] subtitle reflecting on this is how to solve that problem. How to move move
[01:29:00] solve that problem. How to move move phonic alignment from minutes to seconds
[01:29:02] phonic alignment from minutes to seconds in packaging. That's really [snorts]
[01:29:04] in packaging. That's really [snorts] what we do and where the photon bridge
[01:29:05] what we do and where the photon bridge shine platform shines. Um let me find my
[01:29:09] shine platform shines. Um let me find my mouse so I can go to the next slide.
[01:29:12] mouse so I can go to the next slide. Perhaps don't need to spend too much
[01:29:13] Perhaps don't need to spend too much time on this slide. I I think the Nvidia
[01:29:15] time on this slide. I I think the Nvidia presentation probably addressed a lot of
[01:29:17] presentation probably addressed a lot of this already, which is great. But in in
[01:29:20] this already, which is great. But in in line with what we hear across the
[01:29:21] line with what we hear across the industry, data centers in AI are going
[01:29:23] industry, data centers in AI are going to move to in principle by 2028 scale up
[01:29:28] to move to in principle by 2028 scale up optics and it's going to be slow, wide,
[01:29:30] optics and it's going to be slow, wide, multicolored, lots of lasers. Um the big
[01:29:34] multicolored, lots of lasers. Um the big challenge for the industry from a
[01:29:35] challenge for the industry from a packaging perspective as we see it is
[01:29:38] packaging perspective as we see it is how do you package first eight lasers
[01:29:40] how do you package first eight lasers then 16 lasers then 32 lasers because
[01:29:42] then 16 lasers then 32 lasers because that's the way it needs to go to meet
[01:29:44] that's the way it needs to go to meet the ever expanding bandwidth
[01:29:46] the ever expanding bandwidth requirements in a way that's actually
[01:29:48] requirements in a way that's actually manufacturable at volume and packageable
[01:29:50] manufacturable at volume and packageable at the right optical output power and
[01:29:52] at the right optical output power and the optical output power is really
[01:29:54] the optical output power is really important because that drives the
[01:29:55] important because that drives the performance of the system how many CPU
[01:29:57] performance of the system how many CPU engines can you power and what's the
[01:29:59] engines can you power and what's the link budget and the speed and bandwidth
[01:30:01] link budget and the speed and bandwidth of that CPO engine. And this is where we
[01:30:04] of that CPO engine. And this is where we believe and hope Photon Bridge will have
[01:30:06] believe and hope Photon Bridge will have a role. Photon Bridge is all about
[01:30:08] a role. Photon Bridge is all about bridging between materials. We have a
[01:30:10] bridging between materials. We have a platform that's built for laser
[01:30:12] platform that's built for laser packaging. Hopefully, you can see the
[01:30:14] packaging. Hopefully, you can see the video that's playing on the right hand
[01:30:15] video that's playing on the right hand side. You can see a placement of a laser
[01:30:17] side. You can see a placement of a laser or a modulator or whatever active device
[01:30:19] or a modulator or whatever active device you like onto a piece of silicon. And
[01:30:22] you like onto a piece of silicon. And it's all about singlestep optical and
[01:30:24] it's all about singlestep optical and electrical bonding in one go. We do in
[01:30:27] electrical bonding in one go. We do in seconds full passive alignment to less
[01:30:29] seconds full passive alignment to less than 200 nanometers based on a flip chip
[01:30:32] than 200 nanometers based on a flip chip placement accuracy of two micron. That
[01:30:34] placement accuracy of two micron. That means you can do it at speed two 200
[01:30:36] means you can do it at speed two 200 nanometer or less optical alignment at
[01:30:39] nanometer or less optical alignment at two micron placement accuracy at speed
[01:30:42] two micron placement accuracy at speed fully passively. Our platform includes
[01:30:44] fully passively. Our platform includes thick film wave guides which means you
[01:30:46] thick film wave guides which means you can deliver and transport high optical
[01:30:48] can deliver and transport high optical powers and you can be on wavelength.
[01:30:51] powers and you can be on wavelength. Basically thick film for for us means
[01:30:54] Basically thick film for for us means low process sensitivity which makes the
[01:30:56] low process sensitivity which makes the design and the performance of the MX
[01:30:57] design and the performance of the MX just better and all this adds up to a
[01:31:01] just better and all this adds up to a manufacturable laser attach platform
[01:31:04] manufacturable laser attach platform highly scalable standard flip chip tools
[01:31:06] highly scalable standard flip chip tools ey line type lithography so no exotic
[01:31:10] ey line type lithography so no exotic processes required um so what I thought
[01:31:12] processes required um so what I thought I'd do next is share a little bit of an
[01:31:14] I'd do next is share a little bit of an insight into what we're doing with this
[01:31:15] insight into what we're doing with this platform this is the photonbridge laser
[01:31:17] platform this is the photonbridge laser source designed to scale picture of our
[01:31:19] source designed to scale picture of our toer actually that we've We've announced
[01:31:21] toer actually that we've We've announced we are packaging with fix. So great to
[01:31:23] we are packaging with fix. So great to also see fix on this presentation. It's
[01:31:26] also see fix on this presentation. It's got on silicon wavelength multipplexing.
[01:31:28] got on silicon wavelength multipplexing. It's building on our high on silicon
[01:31:30] It's building on our high on silicon power. It's scaling colors by addition
[01:31:32] power. It's scaling colors by addition amplification which means you can move
[01:31:34] amplification which means you can move through eight 16 and then 32 colors. And
[01:31:36] through eight 16 and then 32 colors. And perhaps really important today we bond
[01:31:39] perhaps really important today we bond fully manufactured um burntin tested
[01:31:42] fully manufactured um burntin tested Indian phosphide dye. Our process means
[01:31:45] Indian phosphide dye. Our process means that we have a path to multiple
[01:31:47] that we have a path to multiple qualified sources. So that offers offers
[01:31:50] qualified sources. So that offers offers supply chain redundancy and resilience
[01:31:52] supply chain redundancy and resilience that we think is unique to the platform.
[01:31:54] that we think is unique to the platform. We're developing it today. Stay tuned.
[01:31:56] We're developing it today. Stay tuned. Target performance is greater than 30 ms
[01:31:58] Target performance is greater than 30 ms per wavelength per fiber and targeting a
[01:32:00] per wavelength per fiber and targeting a 16% wall plug efficiency, which is what
[01:32:02] 16% wall plug efficiency, which is what we believe the market needs today. Um,
[01:32:06] we believe the market needs today. Um, if you're working on CPU integration or
[01:32:08] if you're working on CPU integration or advanced packaging, we'd love to talk to
[01:32:09] advanced packaging, we'd love to talk to you. Um, thank you again for taking the
[01:32:12] you. Um, thank you again for taking the time to listen. I really love to hear
[01:32:13] time to listen. I really love to hear questions but if you'd like to reach out
[01:32:16] questions but if you'd like to reach out these are my coordinates let's say
[01:32:19] these are my coordinates let's say >> thank you very much and indeed there are
[01:32:20] >> thank you very much and indeed there are many people in the room working on this
[01:32:22] many people in the room working on this but one of them is really working on
[01:32:24] but one of them is really working on this Leon Gans from Nvidia when you see
[01:32:26] this Leon Gans from Nvidia when you see that they can do 0.2 two micrometer
[01:32:29] that they can do 0.2 two micrometer passive alignment. Is this magic? What's
[01:32:32] passive alignment. Is this magic? What's on your mind?
[01:32:33] on your mind? >> Uh so what I would like to know is do
[01:32:35] >> Uh so what I would like to know is do you see a limit to the number of lasers
[01:32:37] you see a limit to the number of lasers you can you can package? So where where
[01:32:40] you can you can package? So where where would you draw the line 36 72? Can you
[01:32:42] would you draw the line 36 72? Can you reach uh more than that?
[01:32:45] reach uh more than that? >> I I I I think from the bonding
[01:32:47] >> I I I I think from the bonding performance no there's no limit. I think
[01:32:50] performance no there's no limit. I think the challenges really then become
[01:32:52] the challenges really then become thermal management and packaging those
[01:32:54] thermal management and packaging those and managing them all together.
[01:32:57] and managing them all together. So yeah, I I I don't we don't know what
[01:33:00] So yeah, I I I don't we don't know what the limit is today. We feel very
[01:33:02] the limit is today. We feel very comfortable and we're designing eight
[01:33:04] comfortable and we're designing eight colors today and we see a clear path
[01:33:06] colors today and we see a clear path towards 16 and 32. Do we think it scales
[01:33:08] towards 16 and 32. Do we think it scales beyond that? Potentially, but we'd have
[01:33:10] beyond that? Potentially, but we'd have to dig deeper to to find that boundary.
[01:33:12] to dig deeper to to find that boundary. >> Okay. Thank you. Thank you very much.
[01:33:14] >> Okay. Thank you. Thank you very much. Very interesting.
[01:33:17] >> Very well. Thank you. Thank you so much,
[01:33:19] >> Very well. Thank you. Thank you so much, John. Uh let's let's move on to uh
[01:33:22] John. Uh let's let's move on to uh Travis Scott, senior product manager adv
[01:33:25] Travis Scott, senior product manager adv advanced packaging systems at fine. Uh
[01:33:28] advanced packaging systems at fine. Uh Travis, please share your slides and uh
[01:33:32] Travis, please share your slides and uh tell us how fine are enabling reliable
[01:33:34] tell us how fine are enabling reliable scalable hybrid integration and
[01:33:36] scalable hybrid integration and packaging process that can move fetonics
[01:33:38] packaging process that can move fetonics from R&D environments towards repeatable
[01:33:40] from R&D environments towards repeatable high volume manufacturing.
[01:33:47] I see your slides, but I don't hear you.
[01:33:49] I see your slides, but I don't hear you. Travis, I believe you might be on mute.
[01:33:56] >> Hey, can you hear me now?
[01:33:58] >> Hey, can you hear me now? >> I can.
[01:34:00] >> I can. >> Rookie mistake. Sorry about that. Um,
[01:34:03] >> Rookie mistake. Sorry about that. Um, good morning. Thank you Optica for
[01:34:04] good morning. Thank you Optica for giving us the space and for curating a
[01:34:06] giving us the space and for curating a really great uh presentation today. Um,
[01:34:09] really great uh presentation today. Um, yeah, so at Fine, we manufacture die
[01:34:11] yeah, so at Fine, we manufacture die bonding and die assembly equipment. Um
[01:34:14] bonding and die assembly equipment. Um this is currently qualified at 0.3
[01:34:17] this is currently qualified at 0.3 micron at 3 sigma. We're trying to push
[01:34:19] micron at 3 sigma. We're trying to push this down to below 200 nmters as well.
[01:34:23] this down to below 200 nmters as well. Um we offer a prototype to production
[01:34:26] Um we offer a prototype to production platform strategy. So you could start
[01:34:28] platform strategy. So you could start with a manual tabletop bond for your R&D
[01:34:31] with a manual tabletop bond for your R&D and then you could go all the way up to
[01:34:33] and then you could go all the way up to automated production.
[01:34:35] automated production. Um the platform itself is a very
[01:34:38] Um the platform itself is a very accurate pick and place machine at its
[01:34:40] accurate pick and place machine at its at its core functionality. And then on
[01:34:42] at its core functionality. And then on top of this we uh we can put on a
[01:34:45] top of this we uh we can put on a various different array of uh modules.
[01:34:49] various different array of uh modules. This covers all types of different
[01:34:51] This covers all types of different bonding technologies from thermal
[01:34:53] bonding technologies from thermal ultrasonic ultraviolet and adhesive
[01:34:56] ultrasonic ultraviolet and adhesive curing um your tactic bonding for laser
[01:34:59] curing um your tactic bonding for laser assembly. Um and then on top of this
[01:35:01] assembly. Um and then on top of this fine techch has 30 years of experience
[01:35:03] fine techch has 30 years of experience in providing um solutions as a service.
[01:35:06] in providing um solutions as a service. So, we don't just sell machines. We
[01:35:08] So, we don't just sell machines. We don't just sell tooling. We don't just
[01:35:09] don't just sell tooling. We don't just sell modules. We help you put these
[01:35:12] sell modules. We help you put these things together and to bring your
[01:35:13] things together and to bring your products and prototypes to life.
[01:35:16] products and prototypes to life. Um, I'd like to get a head start on Jose
[01:35:19] Um, I'd like to get a head start on Jose and John's question of what we can do
[01:35:21] and John's question of what we can do and what you can do for us. Um, so what
[01:35:23] and what you can do for us. Um, so what we can do for you is provide an
[01:35:25] we can do for you is provide an extremely highly accurate and modular
[01:35:27] extremely highly accurate and modular bonding system for your R&D and
[01:35:29] bonding system for your R&D and prototype and production needs. Um but
[01:35:32] prototype and production needs. Um but what we would very much like from the
[01:35:34] what we would very much like from the industry is we're busy building up our
[01:35:36] industry is we're busy building up our next um uh platform or our next system
[01:35:40] next um uh platform or our next system to cover the next 5 to 10 years of the
[01:35:42] to cover the next 5 to 10 years of the industry requirements. And what we're
[01:35:45] industry requirements. And what we're looking for is partners from the
[01:35:46] looking for is partners from the industry to tell us exactly what you
[01:35:48] industry to tell us exactly what you need from a die bonding and die uh
[01:35:50] need from a die bonding and die uh packaging or like advanced packaging
[01:35:53] packaging or like advanced packaging system. Um so today I've already made
[01:35:56] system. Um so today I've already made quite some notes. Uh it's been a very
[01:35:58] quite some notes. Uh it's been a very interesting discussion. Thank you very
[01:35:59] interesting discussion. Thank you very much for the input of all the speakers
[01:36:00] much for the input of all the speakers so far. Um, but if anybody out there
[01:36:03] so far. Um, but if anybody out there really wants a fine tech system to do
[01:36:06] really wants a fine tech system to do what they need that it currently doesn't
[01:36:08] what they need that it currently doesn't cover today, please reach out and get in
[01:36:09] cover today, please reach out and get in touch and I'd be happy to talk to you.
[01:36:11] touch and I'd be happy to talk to you. Thank you very much.
[01:36:13] Thank you very much. >> The the key thing, the key challenge I
[01:36:15] >> The the key thing, the key challenge I find that keep solving and making making
[01:36:17] find that keep solving and making making things fantastic for industry is when it
[01:36:20] things fantastic for industry is when it comes to to to direct diode bonding.
[01:36:22] comes to to to direct diode bonding. There was a very clear request from
[01:36:24] There was a very clear request from Jable on the UV curing epoxy. Is there
[01:36:27] Jable on the UV curing epoxy. Is there anything that you're working on on this?
[01:36:31] anything that you're working on on this? >> Um, so we offer like a a wide range of
[01:36:34] >> Um, so we offer like a a wide range of dispenser technologies and then we have
[01:36:36] dispenser technologies and then we have our own uh like proprietary ultraviolet
[01:36:39] our own uh like proprietary ultraviolet curing strategy. Um, so we also know
[01:36:42] curing strategy. Um, so we also know that this is very critical when it comes
[01:36:43] that this is very critical when it comes to active alignment. Uh, so we're also
[01:36:46] to active alignment. Uh, so we're also kind of dipping our toes into the world
[01:36:47] kind of dipping our toes into the world of active alignment. um and trying to
[01:36:50] of active alignment. um and trying to understand you know epoxy drift during
[01:36:52] understand you know epoxy drift during curing epoxy uh behavior before and
[01:36:56] curing epoxy uh behavior before and after curing. Um so at the moment we
[01:36:58] after curing. Um so at the moment we don't have any kind of off-the-shelf
[01:37:00] don't have any kind of off-the-shelf solution for this. Um but yeah we've
[01:37:02] solution for this. Um but yeah we've always faced the same issue as the the
[01:37:04] always faced the same issue as the the industry as well with like epoxy drift
[01:37:06] industry as well with like epoxy drift and cure drift and this kind of thing.
[01:37:09] and cure drift and this kind of thing. >> Travis congratulations what you're
[01:37:10] >> Travis congratulations what you're doing. I know I know market is booming.
[01:37:12] doing. I know I know market is booming. I know there are many orders. I know you
[01:37:14] I know there are many orders. I know you are doing fantastic job with this 0.3
[01:37:17] are doing fantastic job with this 0.3 micrometer accuracy with pass with the
[01:37:19] micrometer accuracy with pass with the passive alignment. Allow me to bring
[01:37:21] passive alignment. Allow me to bring another company that is really doing a
[01:37:24] another company that is really doing a huge huge impact on the semiconductor
[01:37:26] huge huge impact on the semiconductor industry but is pushing the photonics
[01:37:28] industry but is pushing the photonics industry as well. I would like to bring
[01:37:29] industry as well. I would like to bring to the table for the first time at an
[01:37:31] to the table for the first time at an optical online industry meeting
[01:37:33] optical online industry meeting Katherine Fisher head of business
[01:37:34] Katherine Fisher head of business development from Verer Lee. Welcome
[01:37:37] development from Verer Lee. Welcome Verer Lee for the first time at an
[01:37:39] Verer Lee for the first time at an optica online industry meeting. The
[01:37:40] optica online industry meeting. The floor is yours.
[01:37:43] floor is yours. So, thank you very much and I hope you
[01:37:46] So, thank you very much and I hope you can see my slides.
[01:37:48] can see my slides. >> Gorilla Glass Clear.
[01:37:50] >> Gorilla Glass Clear. >> Excellent.
[01:37:52] >> Excellent. So, thank you very much for having the
[01:37:54] So, thank you very much for having the opportunity to introduce um the company
[01:37:57] opportunity to introduce um the company and myself. Um let me start. I'm as you
[01:38:01] and myself. Um let me start. I'm as you said the new kid on the blog at Optica
[01:38:06] said the new kid on the blog at Optica uh but and not entirely new to the
[01:38:09] uh but and not entirely new to the field. So I started out in microoptics
[01:38:12] field. So I started out in microoptics with Reinhardt focal got pulled into
[01:38:15] with Reinhardt focal got pulled into semiconductor manufacturing processes
[01:38:17] semiconductor manufacturing processes and equipment at Suss Microte and now
[01:38:21] and equipment at Suss Microte and now with leap I'm trying to connect the dots
[01:38:24] with leap I'm trying to connect the dots um maybe shaving a little time off the
[01:38:28] um maybe shaving a little time off the journey from lab to FEP. So let's see if
[01:38:32] journey from lab to FEP. So let's see if we can get close to that uh within the
[01:38:35] we can get close to that uh within the community.
[01:38:37] community. So, LEAB is a special machine
[01:38:39] So, LEAB is a special machine manufacturer located in the northern
[01:38:41] manufacturer located in the northern part of Bavaria with more than 20 years
[01:38:44] part of Bavaria with more than 20 years of experience in building semiconductor
[01:38:47] of experience in building semiconductor equipment and a track record of hundreds
[01:38:50] equipment and a track record of hundreds of tools in all top manufacturing FAPS
[01:38:52] of tools in all top manufacturing FAPS worldwide. On the left you can see a
[01:38:57] worldwide. On the left you can see a typical example of a tool we are
[01:39:00] typical example of a tool we are building for semiconductor industry
[01:39:02] building for semiconductor industry handling wafers from one cassette to
[01:39:04] handling wafers from one cassette to another and in between the process
[01:39:08] another and in between the process module on the right you see inside the
[01:39:12] module on the right you see inside the tool from above and actually what you
[01:39:14] tool from above and actually what you can see is nothing. There's a big blue
[01:39:19] can see is nothing. There's a big blue empty space and so you might ask
[01:39:22] empty space and so you might ask yourself why it's that empty. Um the
[01:39:25] yourself why it's that empty. Um the free space is for your process module
[01:39:27] free space is for your process module giving you the possibility to go to the
[01:39:31] giving you the possibility to go to the market with a full automated platform.
[01:39:34] market with a full automated platform. We integrate your process module and you
[01:39:39] We integrate your process module and you can concentrate on your process and by
[01:39:42] can concentrate on your process and by that we can enable you to transfer your
[01:39:45] that we can enable you to transfer your process in light speeded from lab to
[01:39:46] process in light speeded from lab to FAP. So that's um especially when it
[01:39:51] FAP. So that's um especially when it comes down to wafer level based
[01:39:54] comes down to wafer level based processes in pix manufacturing.
[01:39:57] processes in pix manufacturing. So quick and short and I'm
[01:40:01] So quick and short and I'm happy to get any inquests of who has any
[01:40:06] happy to get any inquests of who has any interest in kind of getting his perfect
[01:40:09] interest in kind of getting his perfect technology into an automated platform
[01:40:12] technology into an automated platform which I heard a few times now in all the
[01:40:15] which I heard a few times now in all the presentations that this is missing the
[01:40:18] presentations that this is missing the automation part and we would be happy to
[01:40:21] automation part and we would be happy to help here.
[01:40:22] help here. >> Kine welcome to optical. We typically
[01:40:25] >> Kine welcome to optical. We typically focus on the peak itself and having a
[01:40:26] focus on the peak itself and having a company that really can match this with
[01:40:29] company that really can match this with handling transport and automation to
[01:40:31] handling transport and automation to calculate the overall yield is great.
[01:40:33] calculate the overall yield is great. You are going to play a fantastic role.
[01:40:35] You are going to play a fantastic role. We're going to make a lot of
[01:40:36] We're going to make a lot of introductions but allow me to bring to
[01:40:38] introductions but allow me to bring to the table our closing speaker today and
[01:40:41] the table our closing speaker today and we actually save something very
[01:40:43] we actually save something very interesting because we make a living out
[01:40:45] interesting because we make a living out of collaboration. Many of you saw in the
[01:40:47] of collaboration. Many of you saw in the news that Epiphany really is
[01:40:48] news that Epiphany really is collaborating with many of the global
[01:40:51] collaborating with many of the global foundaries globally and recently had an
[01:40:54] foundaries globally and recently had an announcement collaborating with genoton
[01:40:56] announcement collaborating with genoton global foundaries offering already
[01:40:58] global foundaries offering already access to organic polymers and we've
[01:40:59] access to organic polymers and we've been definitely waiting for this making
[01:41:02] been definitely waiting for this making success of the collaboration we love
[01:41:03] success of the collaboration we love this John Epin CEO of Epiphany design
[01:41:06] this John Epin CEO of Epiphany design the floor and the attention of everyone
[01:41:08] the floor and the attention of everyone is yours send us away the way that we
[01:41:10] is yours send us away the way that we deserve
[01:41:11] deserve [laughter]
[01:41:12] [laughter] >> thank you very much co uh very rarely I
[01:41:15] >> thank you very much co uh very rarely I That's such a nice welcome but yeah I
[01:41:17] That's such a nice welcome but yeah I think in photonics indeed collaboration
[01:41:20] think in photonics indeed collaboration is key and that is the collaboration
[01:41:23] is key and that is the collaboration between everything and that is very
[01:41:25] between everything and that is very important point for us because we want
[01:41:27] important point for us because we want to have
[01:41:30] to have first of all we want to change the slide
[01:41:32] first of all we want to change the slide because it's your vision and our
[01:41:34] because it's your vision and our expertise. So we want to bring uh your
[01:41:36] expertise. So we want to bring uh your expertise and make it available
[01:41:39] expertise and make it available and um the first one is uh what do we
[01:41:43] and um the first one is uh what do we do? We do offer design solutions and
[01:41:46] do? We do offer design solutions and supply chain support. So we bridge the
[01:41:48] supply chain support. So we bridge the whole supply chain um where packaging is
[01:41:52] whole supply chain um where packaging is the first line actually we heard that uh
[01:41:54] the first line actually we heard that uh to shape to to shape off the cost but
[01:41:57] to shape to to shape off the cost but the first point where you can save that
[01:41:58] the first point where you can save that is in the design make the right design
[01:42:00] is in the design make the right design choices uh and collaborate with the
[01:42:03] choices uh and collaborate with the right partners in order uh to be yeah to
[01:42:06] right partners in order uh to be yeah to to have a
[01:42:08] to have a small uh yeah design to market to chips.
[01:42:13] small uh yeah design to market to chips. Yeah. And we also do that by making
[01:42:15] Yeah. And we also do that by making design kits for our customers to make
[01:42:17] design kits for our customers to make their
[01:42:19] their technology available and that's also
[01:42:21] technology available and that's also what we are planning uh and will do with
[01:42:23] what we are planning uh and will do with NLM to make their process available. We
[01:42:26] NLM to make their process available. We will also help them uh developing the
[01:42:28] will also help them uh developing the PDK for that in order to be accessible
[01:42:32] PDK for that in order to be accessible and this all we combine also internally
[01:42:34] and this all we combine also internally with the hybrid laser design module for
[01:42:36] with the hybrid laser design module for example where we have pre-ested
[01:42:38] example where we have pre-ested validated uh hybrid lasers that can be
[01:42:41] validated uh hybrid lasers that can be packaged and where package is already
[01:42:43] packaged and where package is already validated so you don't have to
[01:42:44] validated so you don't have to experiment in the processes there
[01:42:51] and what we call it always it's ABCD
[01:42:54] and what we call it always it's ABCD it's assembly based chip design
[01:42:57] it's assembly based chip design because you really want to fit there in
[01:43:00] because you really want to fit there in standard
[01:43:02] standard um chip design um
[01:43:05] um chip design um what what people offer that's why I
[01:43:07] what what people offer that's why I asked for Jel how how can I access you
[01:43:10] asked for Jel how how can I access you how can I make uh chips for you in order
[01:43:12] how can I make uh chips for you in order to have a smooth transition for our
[01:43:14] to have a smooth transition for our customer
[01:43:15] customer what is very important of course is the
[01:43:17] what is very important of course is the design verification yeah because what
[01:43:20] design verification yeah because what brings for example and we heard that in
[01:43:22] brings for example and we heard that in many of these talks bringing EIC's and
[01:43:25] many of these talks bringing EIC's and PIC's together is very important and
[01:43:27] PIC's together is very important and that will be done via uh design and
[01:43:30] that will be done via uh design and verification. Why verification? Because
[01:43:32] verification. Why verification? Because you want to be really sure that you're
[01:43:34] you want to be really sure that you're doing the right thing and not lose time
[01:43:36] doing the right thing and not lose time and money. Yeah. So typically we start
[01:43:39] and money. Yeah. So typically we start with the schematic and the simulations
[01:43:42] with the schematic and the simulations u and mask layout and then do the
[01:43:45] u and mask layout and then do the verification fee go all the way back for
[01:43:48] verification fee go all the way back for checking it with DC. Yeah. Also do post
[01:43:51] checking it with DC. Yeah. Also do post layout simulation. Is it really what I
[01:43:54] layout simulation. Is it really what I want to get? And also lately uh yeah
[01:43:57] want to get? And also lately uh yeah layer versus layout versus schematic to
[01:43:59] layer versus layout versus schematic to really see is it what you will get what
[01:44:01] really see is it what you will get what you can package that such that you can
[01:44:03] you can package that such that you can go to the mark with your designs and
[01:44:05] go to the mark with your designs and then
[01:44:07] then let me see that yeah also verification
[01:44:11] let me see that yeah also verification in design is key for advanced packaging.
[01:44:14] in design is key for advanced packaging. Yeah, in order to have what you need and
[01:44:16] Yeah, in order to have what you need and cross uh crossplatform design modules
[01:44:20] cross uh crossplatform design modules such as hybrid lasers or uh designing
[01:44:22] such as hybrid lasers or uh designing for for organics and polymers is is key
[01:44:26] for for organics and polymers is is key to make uh technology available and then
[01:44:29] to make uh technology available and then thanks a lot
[01:44:31] thanks a lot and happy to receive any question.
[01:44:33] and happy to receive any question. >> Thank you. Thank you. Thank you. Well,
[01:44:34] >> Thank you. Thank you. Thank you. Well, thank you John for for being with us and
[01:44:36] thank you John for for being with us and for what you're doing for the industry.
[01:44:38] for what you're doing for the industry. But one key question and I think many of
[01:44:40] But one key question and I think many of us have this question in mind whenever
[01:44:42] us have this question in mind whenever we hear the current ecosystem on PDKs is
[01:44:45] we hear the current ecosystem on PDKs is there a way is there a way or is there
[01:44:46] there a way is there a way or is there going to be a way in the future that I
[01:44:48] going to be a way in the future that I can actually easily transfer my design
[01:44:50] can actually easily transfer my design from one big foundry let's call it
[01:44:52] from one big foundry let's call it global to another big foundry let's call
[01:44:54] global to another big foundry let's call it tower is that going to happen
[01:44:57] it tower is that going to happen >> yes actually that is what we're doing
[01:44:59] >> yes actually that is what we're doing for our customers we go one level higher
[01:45:03] for our customers we go one level higher uh and design agnostic of pdk such that
[01:45:06] uh and design agnostic of pdk such that um indeed the technology transfer from
[01:45:09] um indeed the technology transfer from one fab to another easy.
[01:45:12] one fab to another easy. >> I would like now to thank everyone but
[01:45:14] >> I would like now to thank everyone but what I'm going to do is going to cut the
[01:45:16] what I'm going to do is going to cut the live stream in YouTube. So for that I
[01:45:18] live stream in YouTube. So for that I have a closing video for those of you
[01:45:20] have a closing video for those of you who want to stay for for a drink after
[01:45:22] who want to stay for for a drink after the event. Stay with us and we'll have a
[01:45:24] the event. Stay with us and we'll have a nice virtual drink. But now allow me to
[01:45:27] nice virtual drink. But now allow me to close the live stream.
[01:45:41] This is opt. [music]
[01:45:52] This is lighting redefined.
[01:46:05] All right. So, allow me to ask I to
[01:46:07] All right. So, allow me to ask I to kindly close the live stream. This
[01:46:09] kindly close the live stream. This becomes now a private event. I can undo
[01:46:11] becomes now a private event. I can undo my tie and breathe again. And I would
[01:46:13] my tie and breathe again. And I would like to thank everyone. Oh my god, so
[01:46:15] like to thank everyone. Oh my god, so many people joined today. For those of
[01:46:16] many people joined today. For those of you who had to leave because you had to
[01:46:18] you who had to leave because you had to pick up the children, because you had to
[01:46:19] pick up the children, because you had to start working,