Integrated Photonics: It’s Always About High Performance

Full Transcript

https://www.youtube.com/watch?v=f2f2ZGWwxo4

[00:03] Okay, welcome everybody to the April.
[00:06] Okay, welcome everybody to the April 11th installment of the California Dreams Technical Series.
[00:08] Uh, today we are very glad to have Jonathan Clamin with us.
[00:11] Uh, Jonathan is a professor at the University of California, Santa Barbara.
[00:13] Um, where he's been since 2015.
[00:17] Before that, he heard he held um technical staff appointments at MIT Lincoln Lab.
[00:19] uh was also at uh several other universities.
[00:22] Um he's published over 250 journal and conference papers, lots of patents.
[00:25] He is not only a faculty member but an entrepreneur.
[00:27] Uh we were lucky to have him as one of the very early participants in um the formation of the California Dreams Hub.
[00:31] Um Jonathan is also the director of the nanofab lab at UCSB um which is really one of our flagship uh nanopab labs throughout the country and has contributed um so much important work as you heard in talks from Umeish.
[01:05] work as you heard in talks from Umeish and and others throughout this series.
[01:08] and and others throughout this series.
[01:11] Jonathan's also the recipient of awards such as the NASA early career faculty award, DARPA young faculty award, DARPA directors fellowship etc and a fellow of optica.
[01:13] such as the NASA early career faculty award, DARPA young faculty award, DARPA
[01:15] award, DARPA young faculty award, DARPA directors fellowship etc and a fellow of
[01:19] directors fellowship etc and a fellow of optica.
[01:22] So welcome Jonathan. Please take it away.
[01:24] Thanks Steve. Thanks a lot for the intro.
[01:28] Thanks everyone uh for joining.
[01:31] Um I I think as you know if you saw the announcement the title of the seminar is integrated photonics.
[01:33] saw the announcement the title of the seminar is integrated
[01:35] seminar is integrated photonics.
[01:37] Uh it's always about high performance.
[01:40] Um I'll share a few thoughts on that during the presentation.
[01:42] So here's here's a brief outline.
[01:44] I'll give a a little bit of background uh very briefly talked about UCSB and the nanofab and you have probably heard about our facility uh in other seminars and and likely will in future seminars.
[01:46] outline. I'll give a a little bit of background uh very briefly talked about
[01:48] background uh very briefly talked about UCSB and the nanofab and you have
[01:51] UCSB and the nanofab and you have probably heard about our facility uh in
[01:54] probably heard about our facility uh in other seminars and and likely will in
[01:56] other seminars and and likely will in future seminars.
[01:58] give a little bit of background on what photonic integrated circuits are.
[02:00] background on what photonic integrated
[02:03] circuits are. Uh and then highlight an application of Pix optical interconnects
[02:06] application of Pix optical interconnects for AI infrastructure.
[02:08] for AI infrastructure uh that has gained lots of interest.
[02:12] uh that has gained lots of interest especially recently over the years but.
[02:14] especially recently over the years but especially recently after Nvidia made.
[02:16] especially recently after Nvidia made some announcements at their uh flagship.
[02:18] some announcements at their uh flagship GTC uh conference.
[02:21] uh and then I'll focus on heterogeneous integration of 35 materials on silicon in support of silicon phatonics for optical interconnects but also for a few other things and uh touch on scaling and and commercialization opportunities.
[02:24] focus on heterogeneous integration of 35 materials on silicon in support of.
[02:27] materials on silicon in support of silicon phatonics for optical interconnects but also for a few other.
[02:28] silicon phatonics for optical interconnects but also for a few other.
[02:30] interconnects but also for a few other things and uh touch on scaling and and commercialization opportunities.
[02:33] things and uh touch on scaling and and commercialization opportunities.
[02:36] So ve commercialization opportunities.
[02:38] So ve very briefly um I'd be remiss to not mention the facility where we do a lot of this uh sort of fundamental device fabrication work the UCSB nanofab uh for those not familiar uh the nanofab um is an advanced nanop fabrication user facility we like to refer to it as a user facility because it is an open access facility users are not only internal UCSB faculty students researchers but also external uh we have users from other.
[02:41] mention the facility where we do a lot of this uh sort of fundamental device fabrication work the UCSB nanofab uh for.
[02:44] of this uh sort of fundamental device fabrication work the UCSB nanofab uh for those not familiar uh the nanofab um is.
[02:47] those not familiar uh the nanofab uh for those not familiar uh the nanofab um is an advanced nanop fabrication user facility.
[02:51] an advanced nanop fabrication user facility we like to refer to it as a.
[02:53] facility we like to refer to it as a user facility because it is an open.
[02:55] user facility because it is an open access facility users are not only.
[02:58] access facility users are not only internal UCSB faculty students.
[03:00] internal UCSB faculty students researchers but also external.
[03:03] researchers but also external uh we have users from other.
[03:07] uh we have users from other universities, government labs, and many universities, government labs, and many from industry.
[03:12] So, it's a big part of from industry.
[03:12] So, it's a big part of sort of the startup ecosystem in Santa Barbara and Southern California.
[03:18] Barbara and Southern California.
[03:18] Facility houses about $60 million worth of equipment and has been in operation on paper for more than uh 25 years in its current location probably 18 or 19 years.
[03:28] Um, and why uh it was developed?
[03:31] Well, it happened over time, but it's really been there to support innovation.
[03:35] We've got a user base that approaches 600 uh annual users.
[03:41] Uh, and about half of those actually are users from industry, from companies.
[03:46] And to our knowledge, this is the largest industrial base supported by any university uh nanofab facility maybe in the world.
[03:53] I say United States there.
[03:53] We operate uh by um uh charging user fees different rates for academics, industrial users and that's how we sort of fund our operation.
[04:03] But when um questions like what kind of
[04:08] um questions like what kind of capabilities do you have started coming.
[04:10] capabilities do you have started coming up in the context of the chips chips act.
[04:13] up in the context of the chips chips act uh we defined our facility as a greater.
[04:15] uh we defined our facility as a greater than or equal to 100 millimeter.
[04:16] than or equal to 100 millimeter facility.
[04:16] Um many of what we do is on small pieces, small wafers up to 100 millimeter or 4 inch.
[04:25] Uh but many of the tools also support up to 150 millimeter wafers.
[04:27] And we have maybe one or two tools that could even uh take a 200 millimeter wafer.
[04:32] I I won't go through the details on the type of equipment that's in the facility, but encourage you to visit our homepage where you could see sort of a detailed uh list of equipment.
[04:41] And if you're interested, there's lots of contact information on the website if you'd like to work with our staff or get access to the facility.
[04:50] Th this slide just gives an idea of some of the research areas leveraged for UCSB research and there's sort of a concentration.
[04:57] Um we do plenty of silicon work but lots of non uh silicon work especially with compound semiconductors and 35 materials.
[05:04] Um and the research spans electronics,
[05:09] the research spans electronics, photonics, MEMS, microfluidics, material.
[05:12] photonics, MEMS, microfluidics, material science, physics, chem, bio, uh and quantum.
[05:15] And you can see some nice images of things that have been built in our facility and sort of the scale of uh those structures, devices and uh and very small structures.
[05:27] Um last couple of things, if you just think about the impact of the facility, we like to show this slide to show where users come from.
[05:34] Uh so in terms of state impact, our hub is based in Southern California.
[05:39] So that's important to us.
[05:42] Uh this facility has been accessed by more than 200 California companies since 2006.
[05:51] Uh 260 companies across the country.
[05:54] Uh almost 200 of the companies that have accessed our facility are small businesses.
[05:59] Um active companies about 68 are local um in Galita and Santa Barbara area.
[06:06] um almost half of those were started and led by UCSB faculty or graduates and approaching a
[06:12] faculty or graduates and approaching a 100 academic institutions since 2006 are.
[06:15] 100 academic institutions since 2006 are really just showing uh impact across the.
[06:18] really just showing uh impact across the country and state but occasionally we.
[06:20] country and state but occasionally we have users from other uh countries that.
[06:23] have users from other uh countries that work with our staff to do fab.
[06:24] work with our staff to do fab development and um it has had an impact.
[06:28] development and um it has had an impact on the sort of start startup economy and.
[06:32] on the sort of start startup economy and ecosystem in our area.
[06:35] ecosystem in our area. These are a few examples of industrial users and their.
[06:38] examples of industrial users and their commercialization successes.
[06:40] commercialization successes. So, mostly recent things uh a highlight I like to.
[06:43] recent things uh a highlight I like to point to is uh Google AI.
[06:46] point to is uh Google AI. So, a lot of the Google quantum AI chips that we hear.
[06:49] the Google quantum AI chips that we hear about in the news were actually.
[06:51] about in the news were actually developed and built at the UCSB nanofab.
[06:53] developed and built at the UCSB nanofab.
[06:55] And it was only recently uh in the last year or two that Google built out their.
[06:59] year or two that Google built out their own facility to sort of scale up for.
[07:00] own facility to sort of scale up for production.
[07:02] production. And that's still local.
[07:05] They have a very strong presence uh in in our area.
[07:08] area. Number of other companies SLD laser just to give some examples is a.
[07:10] laser just to give some examples is a gallium nitride laser company that was.
[07:12] gallium nitride laser company that was acquired a few years ago by Kiosera.
[07:15] acquired a few years ago by Kiosera.
[07:16] Aluma I'll talk a little bit about later.
[07:18] That's a company I started about four four years ago that um that actually went public and is listed on NASDAQ now as of a couple of weeks ago um and has a lot of commercialization traction.
[07:26] Freedom Photonix is a photonic integrated circuits and laser company that was acquired by Luminar which is an automotive lighter company.
[07:36] Transform, you heard a bit about GAN power and RF electronics from Umesh a few weeks back.
[07:45] Uh transform was recently acquired by uh Renaissance.
[07:48] Uh and Crystal and Mirror Solutions was acquired by Thor Labs.
[07:49] They build these very low-noise semiconductorbased uh super mirrors that are used in lots of experiments and spectroscopy applications.
[07:58] Um so many of these companies developed processes at the UCSB nanofab and then either transferred those processes to industrial foundaries or to their own fabs uh that they ended up building as they started to commercialize their technologies.
[08:10] So a bit of background on
[08:13] Technologies.
[08:13] So a bit of background on uh photonic integrated circuits.
[08:15] I think uh photonic integrated circuits.
[08:15] I think many of you are aware of what these are.
[08:17] many of you are aware of what these are.
[08:21] Uh it's it's a primarily a single chip that integrates lots of optical fun.
[08:23] that integrates lots of optical fun functions like uh lasers or optical.
[08:25] functions like uh lasers or optical amplifiers.
[08:27] amplifiers uh optical modulators, detectors,
[08:29] uh optical modulators, detectors, passive elements, filters and things of that nature uh on a on a single chip.
[08:32] passive elements, filters and things of that nature uh on a on a single chip.
[08:35] Um so analogous to electronic IEC's uh but not quite as mature.
[08:39] so analogous to electronic IEC's uh but not quite as mature.
[08:41] Um there are lots of flavors, different platforms.
[08:44] lots of flavors, different platforms.
[08:45] Some of these are based on compound semiconductors.
[08:47] Some of these rely on silicon platonics uh as the base.
[08:50] So a little bit like in the electronics world where you might have SOI, FPSI, BICOS and things like that.
[08:52] little bit like in the electronics world where you might have SOI, FPSI, BICOS.
[08:56] and things like that.
[08:58] Not quite at that maturity level but maturing.
[09:01] Some PICSS have a few optical functions on a chip and there's a big benefit.
[09:05] have a few optical functions on a chip and there's a big benefit.
[09:07] Even the very simple thing um like a laser modulator is beneficial over having two separately packaged devices a laser and modulator.
[09:10] simple thing um like a laser modulator is beneficial over having two separately.
[09:13] is beneficial over having two separately packaged devices a laser and modulator.
[09:15] packaged devices a laser and modulator that then have to be separately packaged.
[09:17] that then have to be separately packaged and fiber coupled.
[09:19] Some picss have hundreds and thousands and tens of thousands of components.
[09:24] So not quite you know billions or trillions like we have in the transistor IC world but still quite complex compared to when people first built uh laser diodes on semiconductor chips.
[09:33] a few applications and some of these I'll touch on a little bit later but uh silic integrated photonics photonic integrated circuits were initially envisioned and developed really for telecom in the early days to integrate for example multiple wavelength lasers on a chip to support WDM applications or components uh for coherent receivers and coherent transceivers when coherent was being pursued for uh long haul and and there were substantial benefits to integrating multiple wavelength lasers or making tunable lasers uh on uh on chips for performance reasons for uh most in most cases for reducing cost, size, weight,
[10:17] cases for reducing cost, size, weight, and power.
[10:20] But the other other benefit you see when you integrate these components on a on a single chip is an increase in performance.
[10:27] Things that you really can't do when you build a photonic system from discrete components.
[10:32] And I'll touch on some of those.
[10:35] Um so and datacom persists I'll come back to this later.
[10:38] So transceivers for shorter reach interconnects because data transfer uh is a little bit different now lots of data moving over uh shorter uh distances relative to say a decade or two ago.
[10:51] Picks are also used for other things like beam steering, building optical phase arrays that can shape and steer a beam that can be used for sensing, could be used for depth perception, uh is is used for example in uh navigation and docking.
[11:07] Um RF photonics, this is an example of a photonic integrated circuit we built to generate and distribute millimeter wave signals for RF applications.
[11:14] a very neat
[11:17] signals for RF applications.
[11:17] a very neat way to use optical true time delay.
[11:19] way to use optical true time delay instead of RFA shifters to be able to tune uh the uh carrier frequency for millimeter wave signals.
[11:22] instead of RFA shifters to be able to tune uh the uh carrier frequency for millimeter wave signals.
[11:27] tune uh the uh carrier frequency for millimeter wave signals.
[11:27] That was the first time that had ever been done for millimeter waves and we did what that with a fairly simple low silicon nitride pick platform.
[11:28] millimeter wave signals.
[11:28] That was the first time that had ever been done for millimeter waves and we did what that with a fairly simple low silicon nitride pick platform.
[11:31] first time that had ever been done for millimeter waves and we did what that with a fairly simple low silicon nitride pick platform.
[11:33] millimeter waves and we did what that with a fairly simple low silicon nitride pick platform.
[11:33] with a fairly simple low silicon nitride pick platform.
[11:35] with a fairly simple low silicon nitride pick platform.
[11:35] Lasercom using lasers for free space communications.
[11:38] pick platform.
[11:38] Lasercom using lasers for free space communications.
[11:42] free space communications.
[11:42] uh th this is prevalent now in space not only uh supported by space agencies such as NASA and the European Space Agency but also commercially companies that are deploying laser com uh com uh systems.
[11:45] prevalent now in space not only uh supported by space agencies such as NASA and the European Space Agency but also commercially companies that are deploying laser com uh com uh systems.
[11:47] supported by space agencies such as NASA and the European Space Agency but also commercially companies that are deploying laser com uh com uh systems.
[11:50] and the European Space Agency but also commercially companies that are deploying laser com uh com uh systems.
[11:51] commercially companies that are deploying laser com uh com uh systems.
[11:51] deploying laser com uh com uh systems.
[11:55] deploying laser com uh com uh systems.
[11:55] and I'll talk a little bit about sensing in the next slide pics are also being used in computing optical computing that is a real thing that again is not only in research but being pursued by companies and is also being leveraged for quantum photonics applications for doing things like uh generating entangled photon pairs.
[11:57] and I'll talk a little bit about sensing in the next slide pics are also being used in computing optical computing that is a real thing that again is not only in research but being pursued by companies and is also being leveraged for quantum photonics applications for doing things like uh generating entangled photon pairs.
[11:59] in the next slide pics are also being used in computing optical computing that is a real thing that again is not only in research but being pursued by companies and is also being leveraged for quantum photonics applications for doing things like uh generating entangled photon pairs.
[12:02] used in computing optical computing that is a real thing that again is not only in research but being pursued by companies and is also being leveraged for quantum photonics applications for doing things like uh generating entangled photon pairs.
[12:05] is a real thing that again is not only in research but being pursued by companies and is also being leveraged for quantum photonics applications for doing things like uh generating entangled photon pairs.
[12:06] in research but being pursued by companies and is also being leveraged for quantum photonics applications for doing things like uh generating entangled photon pairs.
[12:09] companies and is also being leveraged for quantum photonics applications for doing things like uh generating entangled photon pairs.
[12:11] for quantum photonics applications for doing things like uh generating entangled photon pairs.
[12:13] doing things like uh generating entangled photon pairs.
[12:16] entangled photon pairs.
[12:16] So for quantum photonics you need other functions that
[12:18] Photonics, you need other functions that maybe are a little different than the traditional functions that are used in um in uh uh dataccom and telecom.
[12:27] Uh and lastly using interconnects for array based systems.
[12:33] This is something we've been pursuing in California dreams with some of our hub partners.
[12:37] So, similar to the types of wideband transceivers that are built for say dataccom, but maybe slightly geared toward other applications like uh taking lots of information off of focal plane arrays or from RF phased arrays and transferring to uh to another uh place.
[12:55] Um you could use silicon photonics for example for doing that.
[13:00] And just coming back for a moment, what was the purpose of the title uh for photonix?
[13:06] What I'd like to say is that meeting performance specifications has always been a higher priority than achieving low cost.
[13:14] I think we did ourselves a disservice 10, 15, 20 years ago when we
[13:19] disservice 10, 15, 20 years ago when we said, "Hey, the reason we're pursuing silicon photonix is for low cost.
[13:24] Anyone that's tried to work with a silicon photonix foundry even at scale would be the first to tell you that silicon photonics is not cheap.
[13:33] Um it's it's very expensive to to run full silicon patonix wafers at foundaries that might come down o over time but we we we don't really do this for a low cost.
[13:44] We do it because performance requirements for a lot of these applications even what we would consider a volume application like data centers um are are extremely stringent.
[13:57] Um having lasers that are very stable over temperature and meet requirements.
[14:02] These are not very simple things to do.
[14:04] So uh so all I'd like to say here in summary is that we didn't go to silicon as opposed to compound semis for plotonics to get to low cost.
[14:15] It was really about having very stable high yield processes and being able to achieve the performance requirements for a lot of
[14:21] performance requirements for a lot of these applications.
[14:23] these applications.
[14:23] So a couple other examples on PICSS uh
[14:26] So a couple other examples on PICSS uh and after this slide I'm really just
[14:27] and after this slide I'm really just going to focus on silicon photonics and
[14:29] going to focus on silicon photonics and silicon systems and integrating 35s but
[14:32] silicon systems and integrating 35s but photonic integrated circuits also come
[14:34] photonic integrated circuits also come in 35 platforms indium phosphide gallium
[14:38] in 35 platforms indium phosphide gallium arssonide.
[14:38] The example on the top is an indium phosphide pick that we developed
[14:41] indium phosphide pick that we developed for atmospheric gas sensing.
[14:44] This is a pick that integrates two fairly stable
[14:47] pick that integrates two fairly stable and tunable
[14:50] and tunable lasers, phase modulators, uh detectors
[14:51] lasers, phase modulators, uh detectors and a number of passive elements
[14:54] and a number of passive elements um to detect in this case carbon dioxide
[14:57] um to detect in this case carbon dioxide in the atmosphere.
[15:00] It's a technology to be developed with NASA uh that is now uh
[15:01] be developed with NASA uh that is now uh going through steps to package for space
[15:05] going through steps to package for space qualification with an industrial uh
[15:07] qualification with an industrial uh partner and that could be extended to
[15:10] partner and that could be extended to other wavelengths for other gas species
[15:11] other wavelengths for other gas species like methane for example.
[15:13] This example on the lower left is also an indium
[15:16] on the lower left is also an indium phospide pick that we developed for FMCW
[15:22] phospide pick that we developed for FMCW lidar that has both a transmitter and lidar that has both a transmitter and receiver uh for that application.
[15:27] This receiver uh for that application.
[15:28] This one in the lower middle is a gallium arsonite optical phase array has phase modulators and optical gain on chip uh that could be used for a number of applications but this specific application was for industrial uh machining applications.
[15:39] And the last one, a widely tunable 1030 nanometer pick that we developed for topographical LAR that was also done in collaboration with uh NASA.
[15:53] I I mentioned earlier uh that data transfer is a little bit different than it was a couple of decades ago.
[15:59] Once upon a time, we really only cared about longhaul.
[16:03] Um and what's happening now is that optical interconnects are making their way into shorter and shorter reach applications data center interconnects inside data centers between data centers even few meter and people are considering optical for even centimeter reach for example to interconnect uh let's say GPUs with high bandwidth
[16:24] uh let's say GPUs with high bandwidth memory people are really thinking about memory.
[16:26] people are really thinking about doing that with optical even if it's doing that with optical even if it's only a few centimeters away because only a few centimeters away because there's a a bit a bottleneck there.
[16:34] Um there's a a bit a bottleneck there.
[16:36] Um uh and a lot of what's happening today uh and a lot of what's happening today for these short reach interconnects is for these short reach interconnects is in support of AI infrastructure.
[16:41] Indium in support of AI infrastructure.
[16:43] Indium phosphide is still widely used.
[16:45] You can integrate lots of transmitters and integrate lots of transmitters and receivers on indium phosphide chips but receivers on indium phosphide chips but silicon phatonics is really sort of silicon phatonics is really sort of coming uh to the forefront and uh I'll coming uh to the forefront and uh I'll just briefly mention the differences.
[16:55] just briefly mention the differences between these platforms.
[16:58] Um, it really between these platforms.
[17:01] Um, it really comes down to whether or not you can comes down to whether or not you can monolithically integrate a laser optical monolithically integrate a laser optical gain to overcome losses on the chip or gain to overcome losses on the chip or the actual laser that drives those the actual laser that drives those transceiver circuits.
[17:11] Uh, and substrate transceiver circuits.
[17:13] Uh, and substrate size.
[17:15] These are the substrate sizes in These are the substrate sizes in millimeters that are common for these millimeters that are common for these platforms.
[17:17] So you we we didn't go to platforms.
[17:19] So you we we didn't go to silicon platonics again coming back to silicon platonics again coming back to that point to scale necessarily just for that point to scale necessarily just for large volumes.
[17:22] It does have to do with
[17:25] Large volumes.
[17:25] It does have to do with performance.
[17:27] You can make much lower performance.
[17:27] You can make much lower loss waveguides for example in silicon platonics because of the type of tooling.
[17:31] You have access to on 200 and 300 millimeter wafers.
[17:34] You don't have access to that tooling.
[17:36] And it's not just uh lithography.
[17:39] It's how you do packaging and other things.
[17:43] You don't have access to those technologies for these other materials and other substrate sizes.
[17:44] So some of you might have watched the Nvidia GTC conference.
[17:48] I didn't watch the videos.
[17:50] I hope too soon, but I did read a lot of the articles and there was a huge buzz about silicon platonics again after some of the announcements that were made at that conference.
[17:59] Many of us in the field have known Nvidia's been working on silicon platonics and this supply chain for a few years now.
[18:01] Uh but it was a surprise to many people that they'd been doing this.
[18:03] I think it's the first time they really talked about this publicly.
[18:05] Um, and one point that I'll make, this is from a little bit earlier this year, uh, in in an article that says, um, uh, Nvidia CEO,
[18:27] article that says, um, uh, Nvidia CEO, uh, believes AI has to do a 100 times more computation now than when chat GPT was first released.
[18:36] And when I see an article like that, I take that to mean lots of power consumption.
[18:43] And one of the bottlenecks in the power consumption has become the interconnect.
[18:47] And what's happened in the industry, this is a slide from TSMC that is Nvidia's silicon ponics fab uh partner.
[18:54] Um is that years ago um uh sort of mega data centers really needed lower cost uh and maybe different pluggable optics transceivers.
[19:08] And this is all normalized to the amount of power uh that those pluggable transceivers consumes.
[19:15] And uh latency doesn't really mention cost here, but that's of course uh important especially if you're you know looking to plug in 32 of these pluggable modules uh into um a tray in a
[19:27] pluggable modules uh into um a tray in a rack.
[19:29] Uh and then there was a transition to co-packaged optics where silicon photonix transceivers started to get integrated nearby near say a switch ASIC.
[19:38] And doing that instead of using pluggables could reduce power and latency quite a bit.
[19:44] And now what's happening is for a lot of these AI applications you have things like GPUs and XPUs closely integrated on in some cases the same substrate package whether that be silicon or glass or something else and closely integrated with the silicon patonic transceivers to get the data on and off of those packages.
[20:05] And here just sort of shows the benefits to doing that when it comes to power and latency.
[20:13] And then this slide from Nvidia's presentation gave an idea of what that looks like.
[20:18] There's the co-acked optics uh inside um the supply chain that they've developed out.
[20:24] You can see how complicated uh the supply chain is.
[20:28] complicated uh the supply chain is. And one thing I'll point out is that the
[20:30] one thing I'll point out is that the laser module that's feeding this in this
[20:33] laser module that's feeding this in this particular picture seems to be
[20:34] particular picture seems to be integrated in a separate pluggable
[20:36] integrated in a separate pluggable package. So it takes one of those
[20:38] package. So it takes one of those interfaces uh to package the laser
[20:41] interfaces uh to package the laser source that's feeding the silicon
[20:43] source that's feeding the silicon batonics inside uh inside that uh that
[20:46] batonics inside uh inside that uh that package. So right so this is the
[20:48] package. So right so this is the bottleneck uh you need wide bandwidth
[20:51] bottleneck uh you need wide bandwidth low latency and energy efficient uh
[20:54] low latency and energy efficient uh optical interconnect. So the
[20:56] optical interconnect. So the interconnect is a big fraction of the
[20:58] interconnect is a big fraction of the overall uh power budget.
[21:01] overall uh power budget. Um okay so um talking about uh
[21:05] Um okay so um talking about uh technology platforms again so silicon
[21:08] technology platforms again so silicon photonix has a lot of the components
[21:11] photonix has a lot of the components that we need active components like
[21:13] that we need active components like silicon based modulators germanmanium
[21:15] silicon based modulators germanmanium detectors and other things but to date
[21:19] detectors and other things but to date laser and optical gain functionalities
[21:22] laser and optical gain functionalities are not monolithic. Um, silicon
[21:24] are not monolithic. Um, silicon phatonics is already widely deployed for
[21:27] phatonics is already widely deployed for telecom and dataccom uh, but is in
[21:30] telecom and dataccom uh, but is in higher demand now especially because of
[21:32] higher demand now especially because of things like copacet optics that I just
[21:34] things like copacet optics that I just mentioned. But people are hoping to use
[21:36] mentioned. But people are hoping to use silicon platonics for other things uh,
[21:39] silicon platonics for other things uh, as well. And the rest of the talk is
[21:42] as well. And the rest of the talk is going to focus a bit on heterogeneous
[21:44] going to focus a bit on heterogeneous integration and especially how to
[21:46] integration and especially how to integrate lasers and uh, gain material.
[21:50] integrate lasers and uh, gain material. Um, this is how it's done today. uh
[21:53] Um, this is how it's done today. uh co-ackaging. In some cases, even in
[21:55] co-ackaging. In some cases, even in pluggable optics modules, uh companies
[21:59] pluggable optics modules, uh companies would integrate silicon phatonics chips
[22:01] would integrate silicon phatonics chips with laser chips sort of side by side.
[22:03] with laser chips sort of side by side. In some cases, there was even just a
[22:05] In some cases, there was even just a short fiber coupling the light from one
[22:07] short fiber coupling the light from one chip to the other. In some cases, the
[22:10] chip to the other. In some cases, the chips are sort of butt uh packaged and
[22:13] chips are sort of butt uh packaged and and light is butt coupled from one chip
[22:15] and light is butt coupled from one chip to another. In some cases, the lasers
[22:17] to another. In some cases, the lasers are flip chip integrated. Um and that
[22:20] are flip chip integrated. Um and that works reasonably well but is limited in
[22:22] works reasonably well but is limited in scalability and in most cases requires
[22:24] scalability and in most cases requires very precise alignment. So a little bit
[22:28] very precise alignment. So a little bit too much cost in the assembly and
[22:29] too much cost in the assembly and packaging which is really not what you
[22:32] packaging which is really not what you want. That's the beauty of silicon
[22:34] want. That's the beauty of silicon manufacturing. All of the cost is in the
[22:36] manufacturing. All of the cost is in the weight for FEB and uh packaging and
[22:38] weight for FEB and uh packaging and integration and even test is is very
[22:41] integration and even test is is very standardized. That's how you reduce cost
[22:43] standardized. That's how you reduce cost ultimately in the semiconductor world.
[22:45] ultimately in the semiconductor world. Another approach that is deployed
[22:47] Another approach that is deployed commercially. This is the approach that
[22:49] commercially. This is the approach that Intel uh commercialized bonding 35
[22:53] Intel uh commercialized bonding 35 materials directly onto uh SOI silicon
[22:56] materials directly onto uh SOI silicon platonics. And what they do typically is
[22:58] platonics. And what they do typically is bond coupons. Sort of thinned indium
[23:01] bond coupons. Sort of thinned indium phosphide wafers maybe 100 micron thick
[23:04] phosphide wafers maybe 100 micron thick couponons are bonded. The substrates
[23:06] couponons are bonded. The substrates removed and then the lasers are sort of
[23:08] removed and then the lasers are sort of fabricated on the silicon wafers and
[23:11] fabricated on the silicon wafers and light eancently couples between the
[23:14] light eancently couples between the silicon wavegu and the 35 material so
[23:16] silicon wavegu and the 35 material so that the optical mode can experience
[23:18] that the optical mode can experience gain. That's commercialized. I I believe
[23:20] gain. That's commercialized. I I believe the numbers Intel likes to share is that
[23:22] the numbers Intel likes to share is that they've shipped something like 8 million
[23:24] they've shipped something like 8 million transceivers and each of those
[23:26] transceivers and each of those transceivers had eight lasers. So 32
[23:28] transceivers had eight lasers. So 32 million lasers have been deployed and
[23:30] million lasers have been deployed and are in operation in the field. So this
[23:32] are in operation in the field. So this technology works very well. Another
[23:35] technology works very well. Another technology which I think is just really
[23:36] technology which I think is just really an incremental improvement over that is
[23:39] an incremental improvement over that is to leverage micro transfer printing
[23:41] to leverage micro transfer printing whereby instead of bonding sort of full
[23:44] whereby instead of bonding sort of full thickness coupons that then need most of
[23:47] thickness coupons that then need most of the substrate to be removed um you could
[23:50] the substrate to be removed um you could transfer thin film coupons you have sort
[23:52] transfer thin film coupons you have sort of a sacrificial etch layer that allows
[23:54] of a sacrificial etch layer that allows you to batch transfer many thin film
[23:57] you to batch transfer many thin film coupons from native substrates like
[23:59] coupons from native substrates like indium phosphate or galliumarodide onto
[24:02] indium phosphate or galliumarodide onto silicon. That's something we've worked
[24:03] silicon. That's something we've worked on quite a bit as well. Again, that
[24:06] on quite a bit as well. Again, that works, but I see it sort of just as an
[24:07] works, but I see it sort of just as an an uh an interim sort of incremental
[24:10] an uh an interim sort of incremental improvement over uh the coupons. What
[24:14] improvement over uh the coupons. What seems to be catching a lot of attention
[24:15] seems to be catching a lot of attention is doing direct growth of 35s on
[24:18] is doing direct growth of 35s on silicon. And that's what I'm going to
[24:20] silicon. And that's what I'm going to spend some time focusing on for most of
[24:22] spend some time focusing on for most of the remainder of the talk. The reason
[24:24] the remainder of the talk. The reason this is attractive is because you can
[24:27] this is attractive is because you can integrate many gain elements. So what we
[24:29] integrate many gain elements. So what we refer to as gain density per die can be
[24:32] refer to as gain density per die can be very large. This is potentially
[24:34] very large. This is potentially front-end compatible. So imagine a CMOS
[24:37] front-end compatible. So imagine a CMOS facility taking in an MOCVD tool or so
[24:40] facility taking in an MOCVD tool or so to do the deposition in house. They do
[24:42] to do the deposition in house. They do this already with germanmanium
[24:44] this already with germanmanium deposition selective germanmanium growth
[24:46] deposition selective germanmanium growth to integrate detectors for silicon
[24:48] to integrate detectors for silicon platonics and uh for doing silicon
[24:51] platonics and uh for doing silicon geranium by camos. So why not take
[24:53] geranium by camos. So why not take another CVD tool that could deposit
[24:55] another CVD tool that could deposit these materials and just do it right in
[24:57] these materials and just do it right in the the front end line? These other
[25:00] the the front end line? These other techniques require wafers to go out to
[25:03] techniques require wafers to go out to other facilities in many cases and
[25:05] other facilities in many cases and either then to go to a backend fab line
[25:07] either then to go to a backend fab line or come back into uh into the process.
[25:10] or come back into uh into the process. Um and cost could be potentially low if
[25:13] Um and cost could be potentially low if you can directly grow the 35s on
[25:15] you can directly grow the 35s on silicon. there is no 35 substrate uh and
[25:18] silicon. there is no 35 substrate uh and you're uh eliminating a lot of that sort
[25:20] you're uh eliminating a lot of that sort of assembly cost to put you know either
[25:23] of assembly cost to put you know either thin film coupons or uh chiplets onto
[25:26] thin film coupons or uh chiplets onto the substrate. If you only need one
[25:29] the substrate. If you only need one coupon or one chiplet that's totally
[25:32] coupon or one chiplet that's totally fine. Uh but if if you ideally want many
[25:35] fine. Uh but if if you ideally want many gain elements per die, you can see how
[25:37] gain elements per die, you can see how it gets very expensive to populate the
[25:39] it gets very expensive to populate the wafers in either of these cases.
[25:42] wafers in either of these cases. So let's talk a little bit about direct
[25:44] So let's talk a little bit about direct heterappoxy on silicon. This is not an
[25:47] heterappoxy on silicon. This is not an easy thing to do. People have been
[25:49] easy thing to do. People have been working on this on and off and even in
[25:51] working on this on and off and even in theory for decades. When we typically
[25:54] theory for decades. When we typically present the work that we've been doing
[25:56] present the work that we've been doing to people that haven't been following
[25:57] to people that haven't been following specifically what we're doing, usually
[26:00] specifically what we're doing, usually we just hear things like, "Oh, this
[26:02] we just hear things like, "Oh, this doesn't work. People have tried. You
[26:03] doesn't work. People have tried. You know, what are you doing that's any
[26:05] know, what are you doing that's any different?" Uh I would say the only
[26:07] different?" Uh I would say the only thing we've done that's different is
[26:08] thing we've done that's different is we've been working on this now for a
[26:10] we've been working on this now for a decade. literally 10 years persistently
[26:13] decade. literally 10 years persistently uh from the early days of Amphotonix
[26:16] uh from the early days of Amphotonix when amphotonix first uh started and uh
[26:20] when amphotonix first uh started and uh and we're pursuing this technology for
[26:23] and we're pursuing this technology for different applications for devices that
[26:25] different applications for devices that are a little easier um in terms of
[26:28] are a little easier um in terms of material quality requirements. I'm going
[26:30] material quality requirements. I'm going to mostly talk talk about lasers but
[26:33] to mostly talk talk about lasers but later talk a little bit about detectors.
[26:35] later talk a little bit about detectors. A very simple two terminal device that
[26:37] A very simple two terminal device that doesn't draw a lot of current. So it was
[26:39] doesn't draw a lot of current. So it was kind of inherently reliable. Whereas
[26:41] kind of inherently reliable. Whereas when people first pursued 35 on silicon
[26:44] when people first pursued 35 on silicon they were thinking about transistors and
[26:46] they were thinking about transistors and they were competing with silicon. I
[26:47] they were competing with silicon. I don't think that was the right
[26:49] don't think that was the right application to start with in pursuing uh
[26:52] application to start with in pursuing uh this technology. So lots of challenges
[26:54] this technology. So lots of challenges to overcome. These materials are very
[26:57] to overcome. These materials are very lattice
[26:58] lattice mismatched. There's a significant
[27:01] mismatched. There's a significant difference in the coefficient of thermal
[27:03] difference in the coefficient of thermal expansion of these materials. Some
[27:06] expansion of these materials. Some people refer to this as a polarity
[27:08] people refer to this as a polarity mismatch where you're growing a binary
[27:10] mismatch where you're growing a binary material on uh a you know monosilicon
[27:15] material on uh a you know monosilicon material that induces anti-phase
[27:17] material that induces anti-phase domains. Things don't quite match up.
[27:19] domains. Things don't quite match up. It's another source of uh material
[27:22] It's another source of uh material defects. And there's lots of things that
[27:24] defects. And there's lots of things that we've done over time to sort of overcome
[27:27] we've done over time to sort of overcome and mitigate uh some of these issues. Um
[27:31] and mitigate uh some of these issues. Um and I'll and I'll touch on some of
[27:33] and I'll and I'll touch on some of those. Before I do, I just also touch on
[27:36] those. Before I do, I just also touch on quantum dot lasers. Why is there a
[27:40] quantum dot lasers. Why is there a strong renewed interest in quantum dot
[27:42] strong renewed interest in quantum dot lasers? Um well, quantum dots have also
[27:45] lasers? Um well, quantum dots have also been around for a very long time and uh
[27:48] been around for a very long time and uh not quite as commercialized as quantum
[27:50] not quite as commercialized as quantum well lasers. Um, but there are inherent
[27:53] well lasers. Um, but there are inherent benefits of quantum dots. Not just
[27:55] benefits of quantum dots. Not just generally if you need a good, reliable,
[27:57] generally if you need a good, reliable, and efficient laser source, but
[27:59] and efficient laser source, but especially if you're trying to integrate
[28:02] especially if you're trying to integrate lasers on other platforms, whether
[28:04] lasers on other platforms, whether you're butt coupling the lasers, wafer
[28:07] you're butt coupling the lasers, wafer bonding the lasers, or directly growing
[28:09] bonding the lasers, or directly growing the lasers on silicon, which is
[28:11] the lasers on silicon, which is primarily what we're doing. Um,
[28:14] primarily what we're doing. Um, generally speaking, quantum dot lasers
[28:15] generally speaking, quantum dot lasers lend themsel to lower threshold, higher
[28:18] lend themsel to lower threshold, higher differential gain, higher temperature
[28:20] differential gain, higher temperature stability. Carriers are much more highly
[28:23] stability. Carriers are much more highly confined in quantum dots than they are
[28:25] confined in quantum dots than they are in quantum wells. Line width enhancement
[28:28] in quantum wells. Line width enhancement factor is lower, which lends itself to
[28:31] factor is lower, which lends itself to lower noise, higher uh narrower line
[28:34] lower noise, higher uh narrower line with lasers. Because of that, these
[28:37] with lasers. Because of that, these lasers are less sensitive to optical
[28:39] lasers are less sensitive to optical feedback um more more so than say
[28:42] feedback um more more so than say quantum well counterparts and you can do
[28:44] quantum well counterparts and you can do things like modulation doping which uh
[28:47] things like modulation doping which uh also improve laser performance. Another
[28:50] also improve laser performance. Another very important benefit is that quantum
[28:53] very important benefit is that quantum dots are more tolerant to uh poor
[28:56] dots are more tolerant to uh poor material quality to dislocations.
[29:00] material quality to dislocations. There's a sort of cartoon picture here
[29:02] There's a sort of cartoon picture here showing what happens when uh maybe a
[29:05] showing what happens when uh maybe a threading dislocation makes its way
[29:06] threading dislocation makes its way through a few layers of quantum wells.
[29:09] through a few layers of quantum wells. Well, carrier migration lengths are
[29:11] Well, carrier migration lengths are fairly long in quantum wells. The
[29:13] fairly long in quantum wells. The confinement is not quite as strong as it
[29:15] confinement is not quite as strong as it is in quantum dots. And so carriers
[29:17] is in quantum dots. And so carriers migrate to those defects and uh
[29:20] migrate to those defects and uh recombine non-radiatively. So that's a
[29:22] recombine non-radiatively. So that's a big issue. the these uh quantum well
[29:24] big issue. the these uh quantum well lasers with lots of defects are
[29:26] lasers with lots of defects are inherently unreliable and um and lower
[29:30] inherently unreliable and um and lower efficiency. Instead, when you have,
[29:32] efficiency. Instead, when you have, let's say, a thread making its way to or
[29:35] let's say, a thread making its way to or through a few layers of dots, in some
[29:37] through a few layers of dots, in some cases, the dots actually trap that
[29:39] cases, the dots actually trap that defect. So, prevent it from going
[29:41] defect. So, prevent it from going through other layers. Um and the carrier
[29:44] through other layers. Um and the carrier migration lengths are a lot shorter. So,
[29:46] migration lengths are a lot shorter. So, carriers might be migrating over here,
[29:48] carriers might be migrating over here, but they get trapped in the dots. And so
[29:52] but they get trapped in the dots. And so um non-radiative recombination because
[29:54] um non-radiative recombination because of that material defect is a lot lower
[29:56] of that material defect is a lot lower in quantum dot lasers. And my colleague
[29:59] in quantum dot lasers. And my colleague John Bowers uh and his group has
[30:01] John Bowers uh and his group has published a lot of good papers showing
[30:03] published a lot of good papers showing that not only do these lasers work uh
[30:06] that not only do these lasers work uh but they can be very reliable. And so a
[30:08] but they can be very reliable. And so a number of papers out there we can look
[30:09] number of papers out there we can look at the reliability of these lasers as
[30:12] at the reliability of these lasers as the material quality has improved over
[30:14] the material quality has improved over time but also because of using uh
[30:17] time but also because of using uh quantum dots instead of um quantum wells
[30:20] quantum dots instead of um quantum wells and the reliability is approaching that
[30:23] and the reliability is approaching that of the reliability on native substrates
[30:25] of the reliability on native substrates like gallium
[30:26] like gallium arsenite. So how how do you do this
[30:29] arsenite. So how how do you do this there? There's a few different
[30:31] there? There's a few different approaches. In all cases you have to
[30:34] approaches. In all cases you have to start with CBD or MOCVD. you you can't
[30:37] start with CBD or MOCVD. you you can't really grow uh or nucleiate let's say
[30:40] really grow uh or nucleiate let's say gallium arsonite directly on silicon
[30:42] gallium arsonite directly on silicon without doing by let's say by MBE
[30:44] without doing by let's say by MBE without doing something else first and
[30:46] without doing something else first and in many cases what people have done
[30:49] in many cases what people have done traditionally is use miscut substrates
[30:51] traditionally is use miscut substrates very grossly miscut silicon substrates
[30:54] very grossly miscut silicon substrates four degrees but really like six degrees
[30:55] four degrees but really like six degrees and 10 degrees and in many cases what
[30:58] and 10 degrees and in many cases what they've done is uh this is one example
[31:00] they've done is uh this is one example where it was done on on on axis grown
[31:03] where it was done on on on axis grown germanmanium by CBD first and then once
[31:06] germanmanium by CBD first and then once Once you have a buffer and you can
[31:07] Once you have a buffer and you can transition to gallium arssonite from
[31:09] transition to gallium arssonite from germanmanium because the lattice
[31:10] germanmanium because the lattice constant is fairly similar between gas
[31:13] constant is fairly similar between gas and germanmanium you still have the
[31:14] and germanmanium you still have the polarity mismatch issue. Uh then you
[31:17] polarity mismatch issue. Uh then you could grow let's say a laser stack uh by
[31:19] could grow let's say a laser stack uh by MDE and there have been some nice
[31:21] MDE and there have been some nice demonstrations where uh that was done
[31:23] demonstrations where uh that was done and reasonable performance quantum dot
[31:26] and reasonable performance quantum dot lasers demonstrated on silicon. This is
[31:28] lasers demonstrated on silicon. This is a technique in the middle um that we've
[31:31] a technique in the middle um that we've approached over time uh growing the
[31:33] approached over time uh growing the initial buffer by MOCBD fairly thick
[31:36] initial buffer by MOCBD fairly thick buffer with a a sophisticated means to
[31:39] buffer with a a sophisticated means to filter and uh and sort of trap defects
[31:43] filter and uh and sort of trap defects and then switch over to MBE to grow
[31:45] and then switch over to MBE to grow quantum dots and and why the switch
[31:47] quantum dots and and why the switch because people tend to think that you
[31:49] because people tend to think that you need MBE to grow quantum dots and if we
[31:51] need MBE to grow quantum dots and if we want access to uh quantum dots uh for
[31:55] want access to uh quantum dots uh for this robustness to material defects then
[31:57] this robustness to material defects then we've got to do this sort hybrid
[31:58] we've got to do this sort hybrid approach, MOCBD plus uh MBE. Then
[32:02] approach, MOCBD plus uh MBE. Then there's the all MOCBD approach that
[32:04] there's the all MOCBD approach that we've been working on. One of the few uh
[32:06] we've been working on. One of the few uh if not only groups that's been doing
[32:08] if not only groups that's been doing this. If you can grow the whole stack by
[32:10] this. If you can grow the whole stack by MOCVD, why not do that? MOCVD is sort of
[32:13] MOCVD, why not do that? MOCVD is sort of the industry standard for production.
[32:16] the industry standard for production. You grow very high quality material by
[32:18] You grow very high quality material by MOCVD. So we've come up with techniques
[32:20] MOCVD. So we've come up with techniques to grow not only the buffer the initial
[32:22] to grow not only the buffer the initial nucleation on camos compatible silicon
[32:26] nucleation on camos compatible silicon but also to grow the quantum dots. You
[32:28] but also to grow the quantum dots. You can grow quantum dots by MOCVD. It just
[32:31] can grow quantum dots by MOCVD. It just hasn't been as common but I'll show you
[32:33] hasn't been as common but I'll show you some results later where our quantum dot
[32:35] some results later where our quantum dot lasers are as good or better than uh MBE
[32:39] lasers are as good or better than uh MBE uh quantum dot lasers. Um so so how do
[32:42] uh quantum dot lasers. Um so so how do we grow the initial buffer? There's a
[32:45] we grow the initial buffer? There's a number of ways to do this. One technique
[32:46] number of ways to do this. One technique that we really like is we take the CMOS
[32:49] that we really like is we take the CMOS compatible silicon and we sort of treat
[32:52] compatible silicon and we sort of treat the surface in various ways. In this uh
[32:55] the surface in various ways. In this uh particular case, we uh pattern um lines
[33:00] particular case, we uh pattern um lines of silicon dioxide so that we can use
[33:03] of silicon dioxide so that we can use very mature chemical etching that's used
[33:06] very mature chemical etching that's used in MEMS's processes to form these nanog
[33:08] in MEMS's processes to form these nanog groupoups. And we do that so we can
[33:10] groupoups. And we do that so we can expose the 111 interfaces. This is a way
[33:13] expose the 111 interfaces. This is a way to mitigate or trap those anti-phase
[33:16] to mitigate or trap those anti-phase domains that are inherent when you grow
[33:20] domains that are inherent when you grow um uh polarity mismatched material um on
[33:24] um uh polarity mismatched material um on silicon. And what happens is a lot of
[33:26] silicon. And what happens is a lot of those anti-phase domains either don't
[33:28] those anti-phase domains either don't form or form but get trapped really
[33:31] form or form but get trapped really close to this 35 silicon interface. And
[33:33] close to this 35 silicon interface. And then on top of that after doing the init
[33:35] then on top of that after doing the init initial nucleation and starting to
[33:38] initial nucleation and starting to planerize the growth we can get very
[33:40] planerize the growth we can get very nice high quality let's say gallium
[33:42] nice high quality let's say gallium arsonite films that also have smooth
[33:44] arsonite films that also have smooth surfaces and we can actually see whether
[33:46] surfaces and we can actually see whether or not there are any anti-phase domains
[33:48] or not there are any anti-phase domains leading to dislocations in this material
[33:51] leading to dislocations in this material even by doing an AFM scan and then to
[33:54] even by doing an AFM scan and then to improve the material quality further we
[33:56] improve the material quality further we do other things for example we take that
[33:59] do other things for example we take that initial galliumside buffer and we
[34:02] initial galliumside buffer and we subject to what we call thermal cycle
[34:05] subject to what we call thermal cycle analing. Increase the temperature um
[34:08] analing. Increase the temperature um cycle the temperature between 350 and
[34:10] cycle the temperature between 350 and 750 degrees in this case. That sort of
[34:12] 750 degrees in this case. That sort of forces some of those dislocations to
[34:15] forces some of those dislocations to move around. Uh in some cases they run
[34:18] move around. Uh in some cases they run into each other and annihilate uh and we
[34:20] into each other and annihilate uh and we can improve the material quality. Then
[34:23] can improve the material quality. Then uh we incorporate strained layer super
[34:25] uh we incorporate strained layer super lattises. Sometimes this is in gas gas
[34:27] lattises. Sometimes this is in gas gas or other materials because those defects
[34:30] or other materials because those defects also get sort of steered around because
[34:32] also get sort of steered around because of the strain fields around these uh
[34:35] of the strain fields around these uh super lattises and are forced to
[34:37] super lattises and are forced to annihilate and mitigate. And then what
[34:38] annihilate and mitigate. And then what you see here is what looks like very
[34:42] you see here is what looks like very high quality material on top of a uh
[34:45] high quality material on top of a uh nominally sort of defective uh buffer
[34:47] nominally sort of defective uh buffer that was grown on silicon. And the
[34:49] that was grown on silicon. And the material quality improves as we do all
[34:51] material quality improves as we do all this and the dislocation density comes
[34:53] this and the dislocation density comes down to a level that is acceptable for
[34:56] down to a level that is acceptable for making a fairly reliable device. And
[34:58] making a fairly reliable device. And there's an example where we actually
[34:59] there's an example where we actually grew quantum dots on top of this buffer
[35:02] grew quantum dots on top of this buffer for uh initial evaluation. I think seven
[35:05] for uh initial evaluation. I think seven layers of indiumide quantum dots and and
[35:09] layers of indiumide quantum dots and and this was all done by MOCBD. Here's an
[35:12] this was all done by MOCBD. Here's an example. uh very uniform uh quantum dots
[35:16] example. uh very uniform uh quantum dots grown by MLCVD on silicon and
[35:19] grown by MLCVD on silicon and photooluminescence emission at two
[35:22] photooluminescence emission at two different pump tow powers. And what
[35:24] different pump tow powers. And what you're seeing here is a comparison of
[35:25] you're seeing here is a comparison of the photooluminescence emission on this
[35:28] the photooluminescence emission on this buffer grown on silicon compared to the
[35:32] buffer grown on silicon compared to the same quantum dot stack grown on the
[35:33] same quantum dot stack grown on the native galliumide substrate. The PL
[35:36] native galliumide substrate. The PL intensity is about the same. The full
[35:38] intensity is about the same. The full width half max is about the same.
[35:40] width half max is about the same. there's just a slight wavelength uh
[35:42] there's just a slight wavelength uh shift. These quantum dots look very good
[35:44] shift. These quantum dots look very good and and the emission characteristics are
[35:47] and and the emission characteristics are good. Sort of proving to ourselves that
[35:49] good. Sort of proving to ourselves that hey maybe these quantum dots are very
[35:50] hey maybe these quantum dots are very robust to somewhat defective material.
[35:54] robust to somewhat defective material. Take a step back just to show the
[35:56] Take a step back just to show the quantum dot laser performance on native
[35:58] quantum dot laser performance on native gallium arsonide. Early on in this work,
[36:00] gallium arsonide. Early on in this work, we had to prove to ourselves and to the
[36:02] we had to prove to ourselves and to the community that we can indeed do this by
[36:04] community that we can indeed do this by MOCVD. You just have to use the right
[36:06] MOCVD. You just have to use the right metal organic sources and uh dial in
[36:09] metal organic sources and uh dial in your growth parameters. And I would say
[36:11] your growth parameters. And I would say these are you know uh competitive with
[36:14] these are you know uh competitive with record performing quantum dots grown by
[36:16] record performing quantum dots grown by MBE by other research groups by uh
[36:19] MBE by other research groups by uh companies. We might be the only company
[36:21] companies. We might be the only company uh university really doing this by um by
[36:24] uh university really doing this by um by MOCBD and and I'll skip some of the
[36:26] MOCBD and and I'll skip some of the details but you see you know very good
[36:28] details but you see you know very good high temperature operation uh and and
[36:30] high temperature operation uh and and attractive spectral characteristics. We
[36:32] attractive spectral characteristics. We also aged these lasers to show that they
[36:35] also aged these lasers to show that they are reliable. Um they didn't really age
[36:38] are reliable. Um they didn't really age at lower temperatures. So we had to
[36:39] at lower temperatures. So we had to increase the temperature to about 60
[36:41] increase the temperature to about 60 degrees. And then we start to see a
[36:43] degrees. And then we start to see a little bit of degradation in the
[36:44] little bit of degradation in the threshold current. So you extrapolate a
[36:46] threshold current. So you extrapolate a lifetime many years. The these would
[36:48] lifetime many years. The these would pass very stringent uh qualification
[36:51] pass very stringent uh qualification requirements for telecom dataccom and
[36:53] requirements for telecom dataccom and and other applications.
[36:55] and other applications. And then after we dialed in the quantum
[36:57] And then after we dialed in the quantum dot growth on native substrates said
[36:59] dot growth on native substrates said okay let's grow these on on our gallium
[37:02] okay let's grow these on on our gallium arssonite on silicon buffers the only
[37:04] arssonite on silicon buffers the only thing we did a little different and we
[37:05] thing we did a little different and we started to use indium gallium phosphide
[37:08] started to use indium gallium phosphide for the upper cladding instead of algas
[37:10] for the upper cladding instead of algas because we can grow this at lower
[37:11] because we can grow this at lower temperature and preserve the morphology
[37:13] temperature and preserve the morphology of these quantum dots but it turned out
[37:15] of these quantum dots but it turned out that the quality of that in gap wasn't
[37:17] that the quality of that in gap wasn't as good so the ptype conductivity wasn't
[37:20] as good so the ptype conductivity wasn't uh as good as it should have been so it
[37:22] uh as good as it should have been so it needs a little bit more optimization
[37:24] needs a little bit more optimization which is why you see a little bit of a
[37:25] which is why you see a little bit of a difference in performance in CW versus
[37:28] difference in performance in CW versus pulseed operation. We need to reduce the
[37:30] pulseed operation. We need to reduce the resistance of these lasers but
[37:33] resistance of these lasers but nevertheless working very well putting
[37:35] nevertheless working very well putting out tens of millows fairly low threshold
[37:38] out tens of millows fairly low threshold uh current and when you extract material
[37:40] uh current and when you extract material parameters um you know material gain and
[37:43] parameters um you know material gain and and spectral characteristics look very
[37:44] and spectral characteristics look very good for these lasers that were realized
[37:46] good for these lasers that were realized on uh silicon. So, so let me switch
[37:49] on uh silicon. So, so let me switch gears and talk a little bit about how to
[37:52] gears and talk a little bit about how to scale, how to process integrate.
[37:54] scale, how to process integrate. Everything I showed so far was, hey, we
[37:57] Everything I showed so far was, hey, we can grow these materials on planer
[37:58] can grow these materials on planer substrates on planer silicon and
[38:00] substrates on planer silicon and demonstrate things like lasers, but if
[38:03] demonstrate things like lasers, but if we're really trying to get these into
[38:04] we're really trying to get these into CMOS processes like CMOS silicon
[38:06] CMOS processes like CMOS silicon platonics, how do we do that? One way
[38:08] platonics, how do we do that? One way you do that is by doing selective area
[38:10] you do that is by doing selective area growth. So, the main motivation for
[38:13] growth. So, the main motivation for selective area growth in this case was
[38:15] selective area growth in this case was for process integration. If we're going
[38:17] for process integration. If we're going to take a silicon batonix wafer sort of
[38:20] to take a silicon batonix wafer sort of mid-process and uh form recesses expose
[38:24] mid-process and uh form recesses expose maybe the silicon substrate or the S so
[38:25] maybe the silicon substrate or the S so SOI layer where we'd like to integrate
[38:27] SOI layer where we'd like to integrate game we'd like to be able to grow
[38:30] game we'd like to be able to grow selectively deposit the materials just
[38:32] selectively deposit the materials just in that recess and not everywhere else
[38:34] in that recess and not everywhere else on like the high quality oxide that sort
[38:36] on like the high quality oxide that sort of encapsulated all the other uh
[38:39] of encapsulated all the other uh components. But there are other benefits
[38:41] components. But there are other benefits to selective area growth. For example,
[38:45] to selective area growth. For example, um lots of things change when you do
[38:47] um lots of things change when you do selective growth as opposed to sort of
[38:49] selective growth as opposed to sort of planer uh growth on a substrate. Thermal
[38:53] planer uh growth on a substrate. Thermal stress um is reduced significantly.
[38:56] stress um is reduced significantly. Instead of growing this material that's
[38:58] Instead of growing this material that's thermally mismatched with the substrate
[39:00] thermally mismatched with the substrate across the entire substrate, we're only
[39:02] across the entire substrate, we're only depositing these material in recesses
[39:05] depositing these material in recesses and and there's, you know, a fill
[39:07] and and there's, you know, a fill factor. or a certain density associated
[39:09] factor. or a certain density associated with that and that prevents film
[39:12] with that and that prevents film cracking when we grow the films on a
[39:15] cracking when we grow the films on a planer substrate. Sure, we get really
[39:18] planer substrate. Sure, we get really nice results um very good working lasers
[39:21] nice results um very good working lasers but occasionally the film cracks can
[39:24] but occasionally the film cracks can only grow maybe two or three microns and
[39:26] only grow maybe two or three microns and that's not enough for lasers. Uh so when
[39:29] that's not enough for lasers. Uh so when we do that on planer substrates when we
[39:31] we do that on planer substrates when we do selective growth in some cases we've
[39:34] do selective growth in some cases we've grown you know 10 15 micron films not
[39:37] grown you know 10 15 micron films not that that's needed but 10 15 microns
[39:40] that that's needed but 10 15 microns without seeing any film cracking versus
[39:42] without seeing any film cracking versus let's say two or three microns on a a
[39:45] let's say two or three microns on a a planer substrate. And and there's lots
[39:47] planer substrate. And and there's lots of other benefits to selective growth.
[39:49] of other benefits to selective growth. For example um aspect ratio trapping.
[39:52] For example um aspect ratio trapping. you can use the fact that you're growing
[39:54] you can use the fact that you're growing selectively to trap uh some of the
[39:56] selectively to trap uh some of the defects that generate when you're doing
[39:58] defects that generate when you're doing growth on planer substrates. So I'm
[40:02] growth on planer substrates. So I'm going to go a little bit quick through
[40:03] going to go a little bit quick through this section in the interest of time,
[40:04] this section in the interest of time, but we initially just had to dial in
[40:06] but we initially just had to dial in growth parameters to do selective
[40:09] growth parameters to do selective growth. Uh we were able to grow
[40:11] growth. Uh we were able to grow anti-phase boundary uh 35 material
[40:14] anti-phase boundary uh 35 material gallium arsenite and indium phosphide in
[40:17] gallium arsenite and indium phosphide in recesses right on a 001 surface. That
[40:20] recesses right on a 001 surface. That had never been done before. Usually
[40:22] had never been done before. Usually you've got to use a miscut substrate or
[40:24] you've got to use a miscut substrate or do something to the surface like those
[40:25] do something to the surface like those nano V-grooves that I showed before. But
[40:28] nano V-grooves that I showed before. But we were able to grow very high quality
[40:30] we were able to grow very high quality anti-phase boundary free gallium
[40:31] anti-phase boundary free gallium arsonite buffers with extremely smooth
[40:34] arsonite buffers with extremely smooth top surfaces. You see a little bit of
[40:36] top surfaces. You see a little bit of fastening near the edge because of how
[40:37] fastening near the edge because of how the growth sort of proceeds in in this
[40:40] the growth sort of proceeds in in this corner because the fab of these recesses
[40:42] corner because the fab of these recesses is not perfect. um but really high
[40:45] is not perfect. um but really high quality material and it took quite a bit
[40:47] quality material and it took quite a bit of work to sort of dial in the the fab
[40:50] of work to sort of dial in the the fab and growth parameters to get to that
[40:51] and growth parameters to get to that point. You could also put the V-grooves
[40:54] point. You could also put the V-grooves down in the recess and we did that and
[40:57] down in the recess and we did that and that opened up the window a little bit.
[40:59] that opened up the window a little bit. Not only is the material quality better,
[41:01] Not only is the material quality better, uh but we could grow high quality
[41:03] uh but we could grow high quality anti-phase boundary material in wider
[41:05] anti-phase boundary material in wider recesses. So instead of say 15 microns
[41:08] recesses. So instead of say 15 microns wide in the previous approach, 25
[41:10] wide in the previous approach, 25 microns, 100 microns wide. So if your
[41:13] microns, 100 microns wide. So if your device is large like a laser diode and
[41:15] device is large like a laser diode and you need to etch a ridge and put metal
[41:17] you need to etch a ridge and put metal contacts down, it'd be better to be able
[41:19] contacts down, it'd be better to be able to widen that recess a little bit beyond
[41:21] to widen that recess a little bit beyond say the 15 microns that would was
[41:23] say the 15 microns that would was demonstrated in the previous slide. And
[41:25] demonstrated in the previous slide. And so we grew uh buffers selectively not
[41:28] so we grew uh buffers selectively not only just galliumside, but we
[41:30] only just galliumside, but we incorporated things like thermal cyclic
[41:32] incorporated things like thermal cyclic annealing and strained layer super
[41:34] annealing and strained layer super latises to get the higher material
[41:35] latises to get the higher material quality on the flat bottom and on the
[41:38] quality on the flat bottom and on the recesses with V-grooves down there. And
[41:40] recesses with V-grooves down there. And you can see that the material quality
[41:42] you can see that the material quality gets better uh when V-grooves are down
[41:44] gets better uh when V-grooves are down in the recess. And just to see how good
[41:46] in the recess. And just to see how good it was and uh convince ourselves that we
[41:49] it was and uh convince ourselves that we could grow laser quality material, we
[41:51] could grow laser quality material, we just grew microisk lasers that don't
[41:53] just grew microisk lasers that don't require full cladding or doping on both
[41:57] require full cladding or doping on both of those platforms. And just to
[41:59] of those platforms. And just to summarize, in the interest of time,
[42:00] summarize, in the interest of time, these lays these are optically pumped
[42:02] these lays these are optically pumped lazing and the lazing threshold was a
[42:04] lazing and the lazing threshold was a lot lower when we had the bead grooves
[42:06] lot lower when we had the bead grooves down in the recess. So where do we go
[42:09] down in the recess. So where do we go from there? Um we've been working in
[42:12] from there? Um we've been working in collaboration with John Bow's group at
[42:14] collaboration with John Bow's group at UCSB with RFS Sunni AIM Photonix and and
[42:17] UCSB with RFS Sunni AIM Photonix and and some others uh to bring this into the
[42:20] some others uh to bring this into the silicon platonics platform. And this
[42:23] silicon platonics platform. And this particular example here we used MOCBD
[42:26] particular example here we used MOCBD growth to grow the initial buffer but
[42:28] growth to grow the initial buffer but then passed some of these samples to the
[42:29] then passed some of these samples to the MBE team to grow the quantum dots
[42:31] MBE team to grow the quantum dots because it's a little bit more mature to
[42:33] because it's a little bit more mature to grow the dots by MO by MBE. So hybrid
[42:36] grow the dots by MO by MBE. So hybrid MOCBD buffers and maybe even the lower
[42:39] MOCBD buffers and maybe even the lower cladding and then grow the dots in the
[42:41] cladding and then grow the dots in the upper cladding by MBE and then go and uh
[42:44] upper cladding by MBE and then go and uh fabricate these uh devices and you can
[42:46] fabricate these uh devices and you can see some really nice pictures of these
[42:48] see some really nice pictures of these devices. One of the issues is that when
[42:51] devices. One of the issues is that when you switch to MBE, you end up putting
[42:53] you switch to MBE, you end up putting polychrystallin material uh across the
[42:56] polychrystallin material uh across the entire wafer. So the wafer is going to
[42:58] entire wafer. So the wafer is going to need some cleanup. That's what I meant
[43:00] need some cleanup. That's what I meant when I said selective growth allows you
[43:02] when I said selective growth allows you to uh enable process integration. If we
[43:05] to uh enable process integration. If we could grow the whole thing by MOCVD just
[43:07] could grow the whole thing by MOCVD just on the recess, you wouldn't be
[43:08] on the recess, you wouldn't be contaminating the rest of the wafer. The
[43:10] contaminating the rest of the wafer. The other issue is that near the interface,
[43:13] other issue is that near the interface, near the edge of that recess, there's a
[43:15] near the edge of that recess, there's a lot of poly crystallin material. Uh that
[43:17] lot of poly crystallin material. Uh that actually has to be removed after the
[43:19] actually has to be removed after the growth. So in this case, there was sort
[43:20] growth. So in this case, there was sort of an air gap between the 35 waveguide
[43:23] of an air gap between the 35 waveguide and the silicon nitride waveguide. You
[43:25] and the silicon nitride waveguide. You could fill that with a slightly higher
[43:26] could fill that with a slightly higher index material instead of just leaving
[43:28] index material instead of just leaving air there um and demonstrate good
[43:30] air there um and demonstrate good working lasers. So some recent results
[43:34] working lasers. So some recent results um presented at the OFC conference and
[43:37] um presented at the OFC conference and reported in this paper that just came
[43:39] reported in this paper that just came out u very high quality uh MOCBD MBE
[43:44] out u very high quality uh MOCBD MBE hybrid quantum dots um that are
[43:47] hybrid quantum dots um that are operating at very high temperature and
[43:50] operating at very high temperature and also have very reasonable lifetimes uh
[43:53] also have very reasonable lifetimes uh as much as 6.2 two years for this batch
[43:55] as much as 6.2 two years for this batch of 13 lasers and a mean of about 2.3
[43:58] of 13 lasers and a mean of about 2.3 year uh lifetime that was at an elevated
[44:00] year uh lifetime that was at an elevated temperature. I don't remember exactly
[44:02] temperature. I don't remember exactly what temperature might have
[44:04] what temperature might have been at at best 60 degrees but could
[44:06] been at at best 60 degrees but could have been 30 degrees. Refer refer you to
[44:09] have been 30 degrees. Refer refer you to the paper. Um and a few more results
[44:12] the paper. Um and a few more results here just showing characterization of
[44:13] here just showing characterization of the coupling efficiency. So again when
[44:15] the coupling efficiency. So again when you've got this sort of air gap even if
[44:17] you've got this sort of air gap even if you fill it with BCB the light is
[44:19] you fill it with BCB the light is diverging significantly. So, we'd really
[44:21] diverging significantly. So, we'd really like to close that up. So, we're able to
[44:23] like to close that up. So, we're able to characterize the coupling efficiency
[44:26] characterize the coupling efficiency losses about 5.4 dB. That's not so bad.
[44:30] losses about 5.4 dB. That's not so bad. Um, and John's group was even able to
[44:32] Um, and John's group was even able to demonstrate some fairly sophisticated
[44:33] demonstrate some fairly sophisticated lasers with DBR mirrors in the silicon
[44:36] lasers with DBR mirrors in the silicon nitride ring resonators and DBRs on the
[44:39] nitride ring resonators and DBRs on the drop port. So, you know, sort of single
[44:40] drop port. So, you know, sort of single longitudinal mode lasers with reasonable
[44:43] longitudinal mode lasers with reasonable side mode suppression ratio. So, why do
[44:45] side mode suppression ratio. So, why do we want to do this all by MOCVD in the
[44:48] we want to do this all by MOCVD in the future? Well, look what happens when we
[44:50] future? Well, look what happens when we grow the structure selectively by MOCVB.
[44:52] grow the structure selectively by MOCVB. There is no gap. There's a little bit of
[44:55] There is no gap. There's a little bit of faceting, but you have high quality 35
[44:58] faceting, but you have high quality 35 material conformal buted up right
[45:01] material conformal buted up right against the the interface. The silicon
[45:04] against the the interface. The silicon nitrite is buried in the oxide here. The
[45:06] nitrite is buried in the oxide here. The the material has a little bit of
[45:08] the material has a little bit of faceting here. Um, so you could either
[45:10] faceting here. Um, so you could either remove that, but it's really only a
[45:12] remove that, but it's really only a micron or so of material. So that would
[45:14] micron or so of material. So that would be a small gap and just fill that with
[45:16] be a small gap and just fill that with oxide or just leave it there. H after
[45:19] oxide or just leave it there. H after all it is very high index material and
[45:21] all it is very high index material and if you look at simulations that faceting
[45:24] if you look at simulations that faceting doesn't do so much. Uh whether you
[45:26] doesn't do so much. Uh whether you remove the material and fill with SiO2
[45:29] remove the material and fill with SiO2 um and you only have let's say a micron
[45:31] um and you only have let's say a micron gap you can get very high coupling
[45:32] gap you can get very high coupling efficiency or just leave it there and
[45:35] efficiency or just leave it there and even with the faceting and sort of maybe
[45:36] even with the faceting and sort of maybe a little bit of redirection of the light
[45:38] a little bit of redirection of the light very high coupling efficiency can be
[45:40] very high coupling efficiency can be achieved. So that's uh sort of where
[45:43] achieved. So that's uh sort of where we're going in pursuing all MOCBD
[45:46] we're going in pursuing all MOCBD selectively grown quantum dot lasers in
[45:48] selectively grown quantum dot lasers in silicon platonics. Don't have too much
[45:50] silicon platonics. Don't have too much time. Um so let me just say uh something
[45:54] time. Um so let me just say uh something about scaling and commercialization
[45:56] about scaling and commercialization opportunities. Aluma is a company we
[45:59] opportunities. Aluma is a company we spun up about four years ago um that can
[46:04] spun up about four years ago um that can do uh MOCBD growth on very large
[46:06] do uh MOCBD growth on very large diameter wafers. So uh one of the
[46:09] diameter wafers. So uh one of the breakthroughs was to build indium
[46:12] breakthroughs was to build indium gallium marsenide based shortwave
[46:13] gallium marsenide based shortwave infrared detectors on large diameter
[46:16] infrared detectors on large diameter wafers up to 300 millimeter silicon. In
[46:19] wafers up to 300 millimeter silicon. In many applications 200 millimeter is just
[46:21] many applications 200 millimeter is just fine. That's a picture of the world's
[46:23] fine. That's a picture of the world's first indium phospite on silicon 300
[46:26] first indium phospite on silicon 300 millimeter wafer. You can see that it's
[46:28] millimeter wafer. You can see that it's very shiny reasonable quality material
[46:30] very shiny reasonable quality material that we demonstrated a few years ago.
[46:32] that we demonstrated a few years ago. And we've been leveraging this technique
[46:36] And we've been leveraging this technique uh and capability to build devices,
[46:39] uh and capability to build devices, photo detector arrays, large area
[46:41] photo detector arrays, large area detectors um that meet very stringent
[46:44] detectors um that meet very stringent performance requirements for a number of
[46:45] performance requirements for a number of applications. And uh this technology is
[46:49] applications. And uh this technology is broadly applicable even just as a sensor
[46:51] broadly applicable even just as a sensor in lots of things in mobile consumer uh
[46:54] in lots of things in mobile consumer uh automotive uh and if you expand and
[46:57] automotive uh and if you expand and think about quantum dot lasers and
[46:58] think about quantum dot lasers and quantum platonics and even 35 on silicon
[47:00] quantum platonics and even 35 on silicon electronics uh you can leverage this
[47:03] electronics uh you can leverage this technology for a wide variety of uh of
[47:06] technology for a wide variety of uh of applications. So that that's the
[47:08] applications. So that that's the breakthrough. Uh being able to deposit
[47:10] breakthrough. Uh being able to deposit these 35 materials not just on small
[47:12] these 35 materials not just on small substrates like we've always done even
[47:14] substrates like we've always done even if it were silicon uh at the university
[47:17] if it were silicon uh at the university but even being able to bring that to 200
[47:19] but even being able to bring that to 200 millime millimeter and 300 millimeter
[47:21] millime millimeter and 300 millimeter substrates. And why might you want to do
[47:24] substrates. And why might you want to do that? Well, it's really for scaling. And
[47:26] that? Well, it's really for scaling. And it's not just scaling for volume like
[47:29] it's not just scaling for volume like many more chips per wafer because you
[47:31] many more chips per wafer because you have a larger wafer, but it's about
[47:33] have a larger wafer, but it's about getting access to that highly automated
[47:35] getting access to that highly automated manufacturing environment that you have
[47:37] manufacturing environment that you have in the silicon world and being able to
[47:40] in the silicon world and being able to do monolithic integration with CMOS and
[47:43] do monolithic integration with CMOS and being able to do wafer scale integration
[47:45] being able to do wafer scale integration and packaging. A lot of the wafer scale
[47:47] and packaging. A lot of the wafer scale integration and packaging that we hear
[47:48] integration and packaging that we hear about in the news uh now only happens on
[47:52] about in the news uh now only happens on 200 millimeter and 300 millimeter
[47:54] 200 millimeter and 300 millimeter substrates.
[47:55] substrates. Um so Aluma is doing this has a facility
[47:58] Um so Aluma is doing this has a facility nearby UCSB. Um uh with some uh clean
[48:03] nearby UCSB. Um uh with some uh clean room uh environments houses a 300
[48:06] room uh environments houses a 300 millimeter production scale MOCD
[48:08] millimeter production scale MOCD capability but can grow on any substrate
[48:10] capability but can grow on any substrate size from 2 in up to 300 millimeter.
[48:13] size from 2 in up to 300 millimeter. Aluma is a proud user of the UCSB
[48:15] Aluma is a proud user of the UCSB nanofab that I mentioned uh earlier and
[48:18] nanofab that I mentioned uh earlier and has lots of partners for um scaling for
[48:21] has lots of partners for um scaling for especially for fab but even for scaling
[48:23] especially for fab but even for scaling the epi uh to beyond what the company
[48:26] the epi uh to beyond what the company can handle. That gives sort of a summary
[48:28] can handle. That gives sort of a summary of the technology or product uh uh
[48:31] of the technology or product uh uh offerings that include detector arrays,
[48:33] offerings that include detector arrays, large area detectors, now quantum dot
[48:36] large area detectors, now quantum dot lasers, quantum platonics and even
[48:38] lasers, quantum platonics and even nanocale semiconductors for 35s. Just a
[48:41] nanocale semiconductors for 35s. Just a few pictures showing some of the
[48:43] few pictures showing some of the detector arrays demonstrated on these
[48:45] detector arrays demonstrated on these large diameter substrate platforms that
[48:48] large diameter substrate platforms that can be delivered for evaluation or can
[48:51] can be delivered for evaluation or can be assembled into focal plane arrays,
[48:53] be assembled into focal plane arrays, large area detectors for consumer
[48:55] large area detectors for consumer applications that are either pins uh
[48:58] applications that are either pins uh APDs or even geiger mode APDs. In the
[49:00] APDs or even geiger mode APDs. In the silicon world, people refer to those as
[49:02] silicon world, people refer to those as spads. And aluma is also partnering with
[49:05] spads. And aluma is also partnering with infotonics has been for the last few
[49:07] infotonics has been for the last few years but publicizing more because what
[49:09] years but publicizing more because what aluma does is can grow those structures
[49:11] aluma does is can grow those structures on the 300 millimeter wafers. Um and and
[49:14] on the 300 millimeter wafers. Um and and aluma did the growth on uh some of those
[49:16] aluma did the growth on uh some of those MOCVD MBE uh hybrid quantum dot lasers
[49:19] MOCVD MBE uh hybrid quantum dot lasers that I showed earlier. So we're we're
[49:21] that I showed earlier. So we're we're now able to do this on 300 millimeter
[49:23] now able to do this on 300 millimeter scale not just on small wafers like
[49:26] scale not just on small wafers like we've done in the past for um university
[49:29] we've done in the past for um university level demonstrations. In addition to
[49:31] level demonstrations. In addition to gain, you can also think about in
[49:33] gain, you can also think about in integrating other components. Aluma has
[49:35] integrating other components. Aluma has some uh customers in contracts to add 35
[49:39] some uh customers in contracts to add 35 materials for quantum photonics. For
[49:42] materials for quantum photonics. For example, another functionality a
[49:44] example, another functionality a different type of modulator material
[49:46] different type of modulator material either al gas or indium gallium
[49:47] either al gas or indium gallium phosphide that could be used for mod
[49:50] phosphide that could be used for mod modulation or for um entangled photon
[49:55] modulation or for um entangled photon pair generation in quantum. And what's
[49:57] pair generation in quantum. And what's unique about Aluma's capability is that
[49:59] unique about Aluma's capability is that it could take that to 300 millimeter,
[50:02] it could take that to 300 millimeter, add that layer to an S so SOI silicon
[50:04] add that layer to an S so SOI silicon platonics platform and get access to
[50:07] platonics platform and get access to very low loss uh waveguides which is
[50:09] very low loss uh waveguides which is what you really want for quantum. Um and
[50:11] what you really want for quantum. Um and then a program that kicked off a few
[50:13] then a program that kicked off a few months ago using this for what you know
[50:16] months ago using this for what you know sort of was always the holy grail of 351
[50:18] sort of was always the holy grail of 351 silicon uh nano sheet transistors in
[50:21] silicon uh nano sheet transistors in this case and there's a collaboration
[50:22] this case and there's a collaboration with some other hub partners UCSB and
[50:25] with some other hub partners UCSB and Teladine a little bit different the way
[50:27] Teladine a little bit different the way than the way this was done before using
[50:30] than the way this was done before using selective area growth and very
[50:32] selective area growth and very aggressive forms of aspect ratio
[50:34] aggressive forms of aspect ratio trapping uh that we hope to publicize
[50:37] trapping uh that we hope to publicize exactly what we're doing there to to
[50:39] exactly what we're doing there to to realize very high quality defect-free
[50:41] realize very high quality defect-free films on silicon for things like HBTs
[50:46] films on silicon for things like HBTs that you heard about from Mark Rodwell
[50:48] that you heard about from Mark Rodwell and others in in other seminars. So, let
[50:51] and others in in other seminars. So, let me wrap up uh and say that um MOCBD MBE
[50:56] me wrap up uh and say that um MOCBD MBE hybrid approaches are are very good for
[50:59] hybrid approaches are are very good for interim demonstrations and for low
[51:01] interim demonstrations and for low volumes. Uh please uh take a look at
[51:04] volumes. Uh please uh take a look at that paper that we published with John
[51:05] that paper that we published with John Bower's team that's published in JLT
[51:08] Bower's team that's published in JLT now. very excited about those results
[51:10] now. very excited about those results and uh for those of us that working on
[51:12] and uh for those of us that working on the project we know that there's a lot
[51:14] the project we know that there's a lot of room for improvement so we expect to
[51:16] of room for improvement so we expect to have a lot lot of publications in the
[51:18] have a lot lot of publications in the com coming months showing what we're
[51:20] com coming months showing what we're doing because that was really just you
[51:21] doing because that was really just you know initial uh demonstration I think
[51:24] know initial uh demonstration I think ideally you'd like to do this all by
[51:26] ideally you'd like to do this all by MOCVD uh and and that's what we hope to
[51:29] MOCVD uh and and that's what we hope to do in the future just grow everything
[51:31] do in the future just grow everything selectively by MOCVD and that recess it
[51:34] selectively by MOCVD and that recess it simplifies the process integration and
[51:36] simplifies the process integration and can make very high quality
[51:38] can make very high quality uh lasers and if if I look back I can
[51:41] uh lasers and if if I look back I can remember you know just talking about
[51:43] remember you know just talking about sort of that transition from MBE to MOCD
[51:46] sort of that transition from MBE to MOCD for commercialization and scaling I
[51:48] for commercialization and scaling I remember a time probably almost 20 years
[51:50] remember a time probably almost 20 years ago when vixels were being
[51:51] ago when vixels were being commercialized mostly for communications
[51:54] commercialized mostly for communications applications now vixels have exploded
[51:56] applications now vixels have exploded for sensing applications people were
[51:58] for sensing applications people were trying to figure out how to grow those
[52:00] trying to figure out how to grow those DVR mirrors by MOCVD uh and and saying
[52:03] DVR mirrors by MOCVD uh and and saying oh this is really hard and a lot of
[52:04] oh this is really hard and a lot of people were saying you can't grow vixels
[52:06] people were saying you can't grow vixels by MBE Those of you in the field know
[52:09] by MBE Those of you in the field know that all commercial vixels are grown by
[52:11] that all commercial vixels are grown by MOCVD now. Um and and some of that is
[52:15] MOCVD now. Um and and some of that is because of the volume scaling that was
[52:17] because of the volume scaling that was required for consumer markets to get
[52:20] required for consumer markets to get vixels in iPhones uh for example. And
[52:22] vixels in iPhones uh for example. And there's lots of other examples where
[52:24] there's lots of other examples where some of these technologies transition
[52:26] some of these technologies transition from MOCVD so sorry from MBE to MLCVD.
[52:29] from MOCVD so sorry from MBE to MLCVD. If you can grow it by MOCVD, it'll
[52:32] If you can grow it by MOCVD, it'll eventually go there if if scaling and
[52:34] eventually go there if if scaling and commercialization is required. People
[52:37] commercialization is required. People have told me and and said that you can't
[52:40] have told me and and said that you can't grow quantum dots by MOCVD. It happened
[52:41] grow quantum dots by MOCVD. It happened in conference session. Uh one of the
[52:45] in conference session. Uh one of the first talks said somebody asked the
[52:47] first talks said somebody asked the speaker that talked about MBE quantum
[52:49] speaker that talked about MBE quantum dots. Hey, what can you grow quantum
[52:51] dots. Hey, what can you grow quantum dots by uh MOCVD? And the speaker said
[52:54] dots by uh MOCVD? And the speaker said no, you can't do that. And then I gave a
[52:57] no, you can't do that. And then I gave a presentation just after that showing
[52:59] presentation just after that showing very high quality MOCVD
[53:01] very high quality MOCVD uh quantum dots and people in the room
[53:03] uh quantum dots and people in the room were kind of confused. Um so my
[53:05] were kind of confused. Um so my prediction is that as the demand for
[53:07] prediction is that as the demand for quantum dot lasers and integration with
[53:09] quantum dot lasers and integration with silicon uh grows all commercial quantum
[53:12] silicon uh grows all commercial quantum dot lasers will eventually be grown by
[53:14] dot lasers will eventually be grown by MOCVD. Last thing maybe I'll just leave
[53:16] MOCVD. Last thing maybe I'll just leave up if you're interested in these topics.
[53:19] up if you're interested in these topics. Uh we've got a really nice session
[53:20] Uh we've got a really nice session happening at the Optica Advanced
[53:22] happening at the Optica Advanced Photonics Congress. um silicon phatonics
[53:25] Photonics Congress. um silicon phatonics manufacturing customer requirements um
[53:28] manufacturing customer requirements um and fab read readiness Nvidia is
[53:31] and fab read readiness Nvidia is speaking there to jazz uh ST micro
[53:34] speaking there to jazz uh ST micro electronics some other really large
[53:36] electronics some other really large volume silicon platonics fab so please
[53:38] volume silicon platonics fab so please consider attending this meeting to hear
[53:39] consider attending this meeting to hear about what's happening in the industry
[53:42] about what's happening in the industry thanks very much for your time I'm sorry
[53:43] thanks very much for your time I'm sorry if I went a few minutes over but happy
[53:45] if I went a few minutes over but happy to uh take any questions if we've got
[53:47] to uh take any questions if we've got time
[53:52] thank you Jonathan timing was good We do
[53:53] thank you Jonathan timing was good We do have time for
[54:00] questions. I will have one if somebody
[54:02] questions. I will have one if somebody else doesn't, but I I'd like to get
[54:04] else doesn't, but I I'd like to get somebody else
[54:09] in. Okay, I'm going to start with a
[54:12] in. Okay, I'm going to start with a possibly self-s serving question.
[54:16] possibly self-s serving question. Um the UCS
[54:19] Um the UCS uh beef nanofab has had so much success
[54:22] uh beef nanofab has had so much success in tech transition. You gave a lot of
[54:25] in tech transition. You gave a lot of examples of that, but I wonder if you
[54:27] examples of that, but I wonder if you could take a step back and talk a little
[54:29] could take a step back and talk a little bit about why you think that's true and
[54:32] bit about why you think that's true and then what you think the California
[54:34] then what you think the California Dreams Hub could do um to to enhance the
[54:39] Dreams Hub could do um to to enhance the model that you've been so successful
[54:42] model that you've been so successful with so far.
[54:44] with so far. Uh, great question, Steve. Thank you.
[54:46] Uh, great question, Steve. Thank you. Um, I think one of the reasons,
[54:51] Um, I think one of the reasons, um, sorry,
[54:53] um, sorry, I think one of the reasons we've had a
[54:55] I think one of the reasons we've had a lot of successes, labto fab
[54:58] lot of successes, labto fab demonstrations and some of those went to
[55:00] demonstrations and some of those went to industrial fabs for commercialization.
[55:02] industrial fabs for commercialization. Some went to some of the, you know, hub
[55:05] Some went to some of the, you know, hub partners that we have like Kelladine and
[55:07] partners that we have like Kelladine and Northrup and HRL and others. Part of
[55:10] Northrup and HRL and others. Part of that is because we've always had a very
[55:12] that is because we've always had a very strong collaboration between
[55:15] strong collaboration between materials research groups and device
[55:18] materials research groups and device research groups at UCSB and also some
[55:20] research groups at UCSB and also some very good circuit designers. Um
[55:24] very good circuit designers. Um uh and uh I think that has something to
[55:28] uh and uh I think that has something to do with it because when you get a nice
[55:29] do with it because when you get a nice mix of people that are not just
[55:31] mix of people that are not just developing the materials for the science
[55:34] developing the materials for the science aspects but really trying to develop
[55:36] aspects but really trying to develop materials for a device and you've got
[55:38] materials for a device and you've got circuit
[55:40] circuit experts thinking about the
[55:42] experts thinking about the application, there's sort of a a natural
[55:46] application, there's sort of a a natural transition to go uh commercialize
[55:48] transition to go uh commercialize things. And um so I think that's a major
[55:51] things. And um so I think that's a major factor. And one thing that I'm really
[55:53] factor. And one thing that I'm really excited about um which we've been, you
[55:56] excited about um which we've been, you know, building out uh in California
[55:59] know, building out uh in California Dreams is the fact that we've now got an
[56:01] Dreams is the fact that we've now got an even bigger mix of not only compound
[56:03] even bigger mix of not only compound semiconductor materials, device and
[56:05] semiconductor materials, device and circuit experts, but also silicon
[56:07] circuit experts, but also silicon experts. And we've got Moses and Moses
[56:09] experts. And we've got Moses and Moses 2.0 in the mix and various of our
[56:14] 2.0 in the mix and various of our partners, not just us, but right Sandia
[56:16] partners, not just us, but right Sandia and others are trying to bring 35s into
[56:19] and others are trying to bring 35s into Moses. So I think that's really going to
[56:21] Moses. So I think that's really going to open things up. The fact that we you've
[56:24] open things up. The fact that we you've got sort of the entire ecosystem,
[56:26] got sort of the entire ecosystem, customers at the far end, Moses that
[56:28] customers at the far end, Moses that could be a storefront and figure out how
[56:30] could be a storefront and figure out how to broker all the requests. um
[56:33] to broker all the requests. um everything in between all the way down
[56:35] everything in between all the way down to hey there's a customer that wants to
[56:37] to hey there's a customer that wants to work with our nanofabs through the hub
[56:40] work with our nanofabs through the hub to just develop a new etch process
[56:42] to just develop a new etch process module for something. Um we've got the
[56:44] module for something. Um we've got the whole gamut now as a hub. Uh and there's
[56:48] whole gamut now as a hub. Uh and there's lots of examples that that are we we can
[56:50] lots of examples that that are we we can discuss not just historically that are
[56:52] discuss not just historically that are but that things are happening right now
[56:54] but that things are happening right now within our hub. So I I think we are in a
[56:58] within our hub. So I I think we are in a really good position to sort of you know
[56:59] really good position to sort of you know build that model out and and expand um
[57:02] build that model out and and expand um out into other parts of the ecosystem
[57:05] out into other parts of the ecosystem packaging and prototyping that we
[57:07] packaging and prototyping that we haven't traditionally done.
[57:10] haven't traditionally done. Thank you. That was very helpful. I'll
[57:12] Thank you. That was very helpful. I'll I'll also add that um you talked about
[57:14] I'll also add that um you talked about the collaborations in your lab and um
[57:16] the collaborations in your lab and um that's very important and that extends
[57:18] that's very important and that extends to the staff supporting your lab and the
[57:20] to the staff supporting your lab and the and the culture surrounding the lab
[57:23] and the culture surrounding the lab between the faculty and
[57:25] between the faculty and um students and staff. Um a great team
[57:30] um students and staff. Um a great team to work with and it's clear what an
[57:31] to work with and it's clear what an impact that has and that it's not just a
[57:33] impact that has and that it's not just a collection of equipment. Yes, thanks for
[57:35] collection of equipment. Yes, thanks for pointing that out Steve. Sure. I see a
[57:37] pointing that out Steve. Sure. I see a hand up from Priya.
[57:39] hand up from Priya. Yes. Hello. Um hi Professor Clampin um
[57:42] Yes. Hello. Um hi Professor Clampin um agree with everything Steve said and uh
[57:44] agree with everything Steve said and uh as HP we have a close relationship with
[57:47] as HP we have a close relationship with the UCSB clean room as well. Um my
[57:50] the UCSB clean room as well. Um my question was the result that you showed
[57:52] question was the result that you showed with aluma and with like uh growing
[57:54] with aluma and with like uh growing indium phosphide and silicon and other
[57:56] indium phosphide and silicon and other 35s is very impressive and you know like
[57:59] 35s is very impressive and you know like being able to scale to that large um of
[58:02] being able to scale to that large um of a wafer is best of both worlds. So my
[58:05] a wafer is best of both worlds. So my question is do you have any comments on
[58:07] question is do you have any comments on your thoughts on um singulation uh are
[58:10] your thoughts on um singulation uh are you able to use the same singulation
[58:12] you able to use the same singulation methods uh as the silicon camos world is
[58:16] methods uh as the silicon camos world is use uses or is it more the 35 approach
[58:19] use uses or is it more the 35 approach or if you had any comments on that in
[58:21] or if you had any comments on that in terms of getting from that integrated
[58:23] terms of getting from that integrated wafer to the diaphragm.
[58:26] wafer to the diaphragm. Um I I think I understood the question.
[58:30] Um I I think I understood the question. Uh are are you are you talking about
[58:33] Uh are are you are you talking about singulation like to expose waveguide
[58:35] singulation like to expose waveguide facets or um
[58:38] facets or um so once you have like the Indian
[58:40] so once you have like the Indian phosphate as a 300 mm I'm just wondering
[58:43] phosphate as a 300 mm I'm just wondering you get how you get from the wafer form
[58:45] you get how you get from the wafer form to die oh I mean just standard dicing um
[58:49] to die oh I mean just standard dicing um and I can give a little bit of
[58:51] and I can give a little bit of background some of the demonstrations we
[58:53] background some of the demonstrations we did materials growth actually happened
[58:56] did materials growth actually happened on the 300 millimeter wafers because
[58:58] on the 300 millimeter wafers because some of those wafers go through part of
[59:00] some of those wafers go through part of the FAB at uh at amphotonics. Recesses
[59:03] the FAB at uh at amphotonics. Recesses are formed to expose and partially etch
[59:05] are formed to expose and partially etch into the silicon and then the wafers
[59:07] into the silicon and then the wafers make it to the partners to do the
[59:09] make it to the partners to do the initial nucleation and buffer growth. A
[59:11] initial nucleation and buffer growth. A lot of that growth happens on the 300
[59:12] lot of that growth happens on the 300 millimeter scale. Um even the quantum
[59:14] millimeter scale. Um even the quantum dot growth with IQE as a partner, but
[59:17] dot growth with IQE as a partner, but sometimes what we do is uh cut the wafer
[59:20] sometimes what we do is uh cut the wafer into die um maybe to do the final growth
[59:24] into die um maybe to do the final growth of the dots at UCSB instead of at IQE.
[59:27] of the dots at UCSB instead of at IQE. And then most of the FAB demonstrations
[59:30] And then most of the FAB demonstrations for lasers they you know die coupons
[59:33] for lasers they you know die coupons were were were were diced in order to do
[59:36] were were were were diced in order to do the 35 fab at the UCSB clean room. A and
[59:39] the 35 fab at the UCSB clean room. A and so all the lazing results that you see
[59:40] so all the lazing results that you see were um carried out that way. In
[59:43] were um carried out that way. In parallel AIM has been developing out all
[59:46] parallel AIM has been developing out all the 35 process at 300 millimeter scale
[59:49] the 35 process at 300 millimeter scale and uh it's getting close to being
[59:52] and uh it's getting close to being ready. Um most of the steps are are
[59:54] ready. Um most of the steps are are there. There's a few things to uh tweak,
[59:57] there. There's a few things to uh tweak, but a lot of what's happening now is
[59:58] but a lot of what's happening now is just moving away from doing stuff on
[01:00:00] just moving away from doing stuff on small pieces and doing it all at the 300
[01:00:03] small pieces and doing it all at the 300 millimeter scale. So imagine Aluma doing
[01:00:06] millimeter scale. So imagine Aluma doing the the buffer growth and maybe the
[01:00:08] the the buffer growth and maybe the lower cladding and then IQE doing the
[01:00:09] lower cladding and then IQE doing the quantum dot growth and then all the fab
[01:00:12] quantum dot growth and then all the fab happening at in platonics. And our hope
[01:00:14] happening at in platonics. And our hope is one day Aluma does the entire uh
[01:00:16] is one day Aluma does the entire uh laser growth. But at the end those
[01:00:18] laser growth. But at the end those wafers go back into the fab and it's
[01:00:21] wafers go back into the fab and it's almost like they're any silicon
[01:00:22] almost like they're any silicon platonics wafer. the the you know the
[01:00:24] platonics wafer. the the you know the gain uh structures are integrated.
[01:00:27] gain uh structures are integrated. They're completely buried in the oxide
[01:00:28] They're completely buried in the oxide just like a geranium detector or
[01:00:30] just like a geranium detector or something else. They sit below a few of
[01:00:33] something else. They sit below a few of the metal uh interconnect layers. Um and
[01:00:37] the metal uh interconnect layers. Um and the wafers just get uh you know diced in
[01:00:39] the wafers just get uh you know diced in most cases at Sunni's tap facility up in
[01:00:41] most cases at Sunni's tap facility up in Rochester.
[01:00:43] Rochester. Okay. Well, thank you so much.
[01:00:45] Okay. Well, thank you so much. Appreciate the insights. Thank you.
[01:00:48] Appreciate the insights. Thank you. Thank you for the good question, Priya.
[01:00:50] Thank you for the good question, Priya. Um we're past the end of the hour. Uh if
[01:00:52] Um we're past the end of the hour. Uh if there's one more quick question, I'll
[01:00:54] there's one more quick question, I'll take it. But I don't see any hands
[01:00:57] take it. But I don't see any hands up. We have positive comments but not
[01:01:00] up. We have positive comments but not questions in the chat window. So thank
[01:01:02] questions in the chat window. So thank you Jonathan again for such a great
[01:01:04] you Jonathan again for such a great seminar. Um and as we finish up here, I
[01:01:07] seminar. Um and as we finish up here, I will just remind people that we have uh
[01:01:09] will just remind people that we have uh more great seminars coming up. Uh next
[01:01:12] more great seminars coming up. Uh next week uh Hamemeda Agassi from UC Irvine.
[01:01:16] week uh Hamemeda Agassi from UC Irvine. Um the week after that we have Steve
[01:01:18] Um the week after that we have Steve Zamuk from PDF solutions along with
[01:01:21] Zamuk from PDF solutions along with Jonathan's colleague from UCSB, Brian
[01:01:24] Jonathan's colleague from UCSB, Brian Bibbo speaking of the excellent uh team
[01:01:26] Bibbo speaking of the excellent uh team that supports the nanofab lab there at
[01:01:28] that supports the nanofab lab there at UCSB. Jonathan Hacker at Teladine. And
[01:01:32] UCSB. Jonathan Hacker at Teladine. And then uh we are very lucky to have Subu
[01:01:35] then uh we are very lucky to have Subu Ayair who recently turned from leading
[01:01:37] Ayair who recently turned from leading the NAPMP program at the Department of
[01:01:40] the NAPMP program at the Department of Commerce from UCLA. And interestingly
[01:01:43] Commerce from UCLA. And interestingly and coincidentally, we have two in a row
[01:01:45] and coincidentally, we have two in a row from UCLA. The week after that, Chiwi
[01:01:48] from UCLA. The week after that, Chiwi Wong from also from UCLA. So, uh please
[01:01:52] Wong from also from UCLA. So, uh please visit
[01:01:54] visit our sites and um sign up for the series.
[01:01:58] our sites and um sign up for the series. Make sure you get the notifications and
[01:02:01] Make sure you get the notifications and uh check out the YouTube videos for any
[01:02:03] uh check out the YouTube videos for any seminars you miss. Thank you everybody.
[01:02:07] seminars you miss. Thank you everybody. Thanks again Jonathan.
[01:02:10] Thanks again Jonathan. Thanks Steve. Thanks everyone.

https://www.youtube.com/watch?v=f2f2ZGWwxo4

Summary

TL;DR — Integrated photonics, particularly silicon photonics, is crucial for high-performance applications like AI infrastructure, driven by the need for advanced optical interconnects. The development of these technologies relies heavily on advanced nanofabrication facilities and heterogeneous integration of materials, with a focus on enabling scalable and cost-effective manufacturing.

Key points

• The UCSB nanofab facility plays a vital role in supporting innovation and commercialization, serving a large industrial user base and contributing to the local startup ecosystem.

• Photonic integrated circuits (PICs) offer significant advantages over discrete optical components, enabling higher performance, reduced size, weight, and power consumption for applications ranging from telecom to AI and sensing.

• Silicon photonics is becoming increasingly important for AI infrastructure due to its potential for high-bandwidth, low-latency, and energy-efficient optical interconnects, especially in co-packaged and chiplet-based architectures.

• Heterogeneous integration, particularly the direct growth of III-V materials on silicon, is a key focus for enabling monolithic lasers and gain functionalities in silicon photonics, with quantum dot lasers showing promise due to their robustness and performance.

• Scaling manufacturing to large-diameter wafers (200-300mm) using techniques like MOCVD is essential for commercialization, enabling integration into automated silicon manufacturing environments and reducing costs.

Takeaway — The advancement of integrated photonics, especially silicon photonics, is critical for future high-performance computing and AI, with ongoing research focusing on heterogeneous integration and scalable manufacturing to meet demanding performance requirements.