# LIGHTFINDER: WHAT HAPPENS WHEN A 50-POUND INSTRUMENT FITS ON A CHIP

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

[00:10] Everyone be prepar for something very special.
[00:12] Diana.
[00:13] Okay.
[00:13] So, it was nice.
[00:15] We created a little bit of suspense.
[00:17] I can't say we did that on purpose.
[00:19] Uh but now we have the right deck queued up.
[00:23] Uh so it's clear that the demand from the hyperscalers um has created an enormous amount of innovation in the photonix industry that has brought the entire industry many of whom are in this room from the fabs to the OSATs to the equipment companies to a whole new operating point in able in order to deliver photonic technologies at scale um for co-ackaged optics.
[00:48] But the optical test instrumentation has not kept up with the in incredible amount of infrastructure that has been developed for co-ackaged optics.
[01:00] And in order to be able to deliver co-ackaged optics at a true scale, we need to be able to test these photonic components.
[01:07] And what's clear is everyone in their lab has large instruments like optical spectrum analyzers from Yokugawa.
[01:12] Which you cannot do 10,000 simultaneous optical measurements at a single time without spending millions and millions of dollars on the instrumentation.
[01:21] And that's the pro one of the problems that Lightfinder is built to solve with our integrated photonic spectrometer on a chip technology.
[01:34] And this is part of a broader trend which is that instrumentation in general, scientific instrumentation is trapped in what I like to call the era of the instrument, which is we're all used to these really large devices that have tons of moving components.
[01:48] We have a lab.
[01:51] There are a lot of portable tools that exist today, but people who want to make high impact decisions don't use a lot of the portable low performance tools because you can't make a high impact decision with an instrument that you don't know works very well.
[02:04] And what
[02:08] hasn't been possible yet is how can you have a sensor that can achieve the performance of an instrument that you already trust but in a smaller form factor.
[02:18] So a lot of um sensing instrumentation needs to undergo the change that a lot of electronics and compute has gone in order to be truly miniaturized and high performance at the same time.
[02:30] At MIT we developed a technology called the digital forier transform spectrometer.
[02:35] This is unlike a dispersive spectrometer which is what most people are used to when they think of a spectrometer which has a grading that uh spatially separates light into its different wavelengths.
[02:48] Here on a silicon photonic chip, we've developed the most capable and scalable chipbased spectrometer um using a a standard silicon photonic platform.
[03:00] We have no fancy materials.
[03:03] Um and we're instead of dispersing the light um spatially, we're able to
[03:08] spectrally encode um the information in a forier transform approach in a series of MZIs um with optical switches that allows us to do spectroscopy in a different way um and allows us to fit it in a small miniaturized form factor.
[03:25] There are three key benefits to the technology we developed.
[03:27] The first is that we can exponentially scale the number of channels by adding another stage to our um architecture.
[03:34] So in most spectrometers you have to increase linearly the number of uh channel stages or channels um in order to get what you're looking for and that linearly increases the size and cost of your device.
[03:54] Um so with a five-stage architecture we can have a 1024 channel spectrometer which allows us to have a very very high resolution.
[04:03] The second benefit is we have a ultra high optical path length modulation and the third is that we benefit what's
[04:10] called from what's called felg's advantage or the multipplex advantage.
[04:13] which means we have a very high SNR.
[04:16] because we only have one photo detector in our entire architecture.
[04:18] our noise floor is significantly lower because if you have more photo detectors, you're stacking the noise of all the detectors on top of each other.
[04:30] Um, and we're able to get a very high SNR measurement.
[04:34] Our technology has won several awards since it start the research on it started back in 2018 at MIT.
[04:39] It's been ranked as one of the top 50 papers in physics by Nature Communications.
[04:44] It was featured on the MIT homepage.
[04:46] We were one of the winners of the MIT 100K competition which is a really big entrepreneurship competition.
[04:55] Have gotten tons of fellowships and awards from the US federal government and state agencies and we've actually so far been entirely funded by just grants which has been amazing to bring our technology to a TRL level 45 in 18 months.
[05:09] What's impressive
[05:12] about our architecture when you compare it to what exists today for a chipbased spectrometer is how broadband we're able to um detect.
[05:19] So we can detect right now a 420 nanometer bandwidth at a 0.05 nanometer resolution.
[05:28] The instrument that can do this today is a Yokugawa optical spectrum analyzer which is a 50 pound microwave size instrument.
[05:38] Most chipbased spectrometers would have tens of nanometers of resolution at this uh not chipbased but traditional spectrometers would have tens of nanometers of resolution at this bandwidth.
[05:48] And the chipbased spectrometers tend to have a much narrower bandwidth such as array waveguide gradings and they suffer from the same problem of linearly increasing their size and cost as you make them larger.
[06:01] Today we're able to cover the telecom band OCNL which is 1260 to 1680 nmters.
[06:08] But what we're actively working towards are the second and third generation of our technology where we
[06:14] can go into the longer near IR which enables a lot of really exciting sensing applications that we're working towards and then into the lower wavelength visible um near uh spectra.
[06:29] why this matters for the co-ackage optics industry.
[06:31] So when I was thinking about what to talk about today, I thought about how can we help this audience scale optical tests.
[06:36] Um and currently the AT companies can do 10,000 electronic tests simultaneously because the electronic test equipment is extremely small and cheap and high performance and they know that it works.
[06:54] And when we've talked to ATES, what's clear is that they want to be able to do 10,000 simultaneous optical tests if they really see CPO having a huge volume, which is one of the use cases that we're currently working towards.
[07:08] The second one is for channel monitoring.
[07:10] So, as WDM is becoming increasingly popular, and as we're
[07:15] starting to see an interest in going wide and slow as opposed to narrow and fast, you have to be able to do really high performance channel monitoring.
[07:25] on the WDM board, photo dodes only give you a power measurement, and most companies will have a wavelength meter, which they'll use to individually measure every wavelength of their channel.
[07:36] But you can imagine having several wavelength meters if you want to go significantly wider just increases the size of the chip and takes up a lot more space.
[07:47] And thirdly is in verifying the CPO modules.
[07:51] So a lot of it's interesting when you talk about how is CPO being tested now in what's being de what's being built in new product introductions.
[08:00] A lot of companies will go to the very last step and maybe test just the assembled module because in order to do the optical test, it would take significantly too much uh infrastructure or to do it on the wafer scale to be able to test every single die.
[08:15] Um but the problem with that is
[08:18] that you have to finish the entire assembly before you know whether your module works.
[08:26] Um, so we're interested in how we can help uh CPO test along the way because we can now enable a significantly cheaper and parallelizable uh optical test.
[08:36] So today we're working on releasing our first product which is simply our uh chipbased spectrometer which will act as a a chipbased optical spectrum analyzer um for the data communications industry.
[08:50] But what we're building towards is something much larger which is a platform technology that can do true spectroscopy on a chip.
[08:56] So between my PhD at Columbia University uh and my post-doal training at MIT, I've always been extremely excited about how we can take bulk optics um and miniaturize them by the power of silicon photonics.
[09:13] I was very inspired when I worked with Mika Lipson and I saw how the laser that we had miniaturaturized onto a chip h
[09:20] worked better than the NKT superc continuum laser that I was used to using in the lab and that really inspired me to continue in the field and to figure out how we can now bring all of spectroscopy onto a chip.
[09:31] And so seeing as I'm coming at the end of uh the conference and we're thinking about well what's beyond co-ackaged optics we're now going to have an incredible infrastructure that's being you know perfected in order to deliver on co-ackaged optics but of course beyond co-ackaged optics the industry will be looking for well what can we use all this infrastructure for and what we believe it is sensing.
[09:57] So we have had an incredible amount of interest for remote sensing across applications the semiconductor equipment industry, medical, more data communications and the energy industry.
[10:11] And I really believe that silicon photonics is what is going to be able to allow these industries to have true chemical
[10:20] compositional analysis embedded in their field operations which could potentially drive a volume that matches co-ackaged optics.
[10:29] If you can produce enough sensors to be able to really rein reinvision how sensing is done and bringing the lab out of the field which a lot of these industries rely very heavily on.
[10:43] So it was very nice to introduce you all to my company and would love to accept presents.
[10:52] Thank you very much Diana for being here.
[10:54] What what a couple of days you have given a fantastic presentation.
[10:59] Can you tell us what you can do for them and what they can do for you?
[11:01] Yes.
[11:01] So what we can do for you is um take measurements.
[11:05] So we have these TRL45 prototypes that we are uh actively demoing for customers and taking all kinds of new spectral measurements that we haven't seen before.
[11:17] So we would love to um learn about your needs and see if
[11:22] We can uh either you share samples with us or we can deliver a unit to you um.
[11:28] And get you in our queue for once we have our uh next batch of chips that we can start deploying them in more proof of concept studies what you can do for us.
[11:38] So I think um for the for packaging this is going to be something that we're be looking for partners as we try to scale.
[11:47] So if anyone is interested in work looking beyond co-acked optics into sensing we'd love to learn uh and work with you and yeah.
[11:57] I already heard that it's actually quite difficult to invest in this company they didn't take my money so you are interested in investing is actually wrong because they are over subscribed.
[12:08] Yeah, very impressive work um so I was wondering what are the engineering tradeoffs.
[12:12] So if I for example if I want more resolution or more bandwidth are they're trading off each other or.
[12:19] Yeah, great question.
[12:19] So, uh we
[12:23] depending on how we calibrate this device, we could give a narrower resolution um with a narrower bandwidth.
[12:34] Uh so we now have been pushing the limits of our bandwidth and resolution by adding another stage.
[12:38] So what I showed the results I showed are with a five stage device.
[12:43] If we had a six-stage device, um we can do an even higher resolution.
[12:49] Um the only thing is the more stages you add, you slightly sacrifice the speed of the measurement.
[12:53] Uh but at the same time, we can calibrate exactly the same architecture in a narrower bandwidth.
[13:02] So then you can have um higher resolution, narrower bandwidth and not sacrifice, you'd have the same number of channels.
[13:09] So you wouldn't sacrifice the speed.
[13:11] Can you do you think that that you can get to the fivepometer that the best Yukawa OSA uh are now capable?
[13:19] Yeah. So that's that's what we're trying to figure out is how we can test it.
[13:23] now when we do the testing, we use a tunable laser and so we need the resolution of that laser to be we probably will need another laser to be able to to test that.
[13:35] But that's something we're very interested in is how closely we can match Yokagawa.
[13:40] We have a lot of questions, but I need to ask you this question.
[13:42] Leon, you come from the Melanox group.
[13:43] When it comes to product optimization towards the scale up, there's very few people who know this better than you.
[13:48] What advices do you have for her?
[13:51] Uh I I I don't think she need my advice.
[13:54] I think that's that's a wonderful product and and I see a lot of application from that.
[13:57] Uh, and even if you have to increase a bit more stages, that's not uh that's that's not a big sacrifice.
[14:06] Uh, they're they're much smaller than Yukagawa, so there's a lot of a lot of room.
[14:11] Well, Diana, that's a compliment from Nvidia.
[14:13] Lars Diamond 365.
[14:16] Uh, great presentation, great product.
[14:18] Um, congratulations.
[14:18] Um, I saw your plan on going uh from uh Visual to N.
[14:25] would be uh the speed uh that one management does take when you have this set up?
[14:31] Um invisible or just across the whole bandwidth?
[14:35] The whole bandwidth.
[14:36] So with what we do right now, we are with thermoptic modulation.
[14:39] We do like millisecond level um speeds.
[14:42] We'll be upgrading to electrooptic modulation where I think we can go a lot faster.
[14:48] uh in terms of the lower wavelengths and longer wavelengths, we haven't designed the modulators to know yet, but those will tend to be uh not high-speed applications.
[15:02] So high speed is really in the telecom band where you you would need uh the spectral measurement to be taken as fast as possible.
[15:09] Um some need measurements some other applications need measurements like once an hour.
[15:19] Aldo Font Tech.
[15:20] Yeah.
[15:20] So this is And from Font Tech.
[15:22] So I have just a curiosity question.
[15:24] So very very nice product. Impressive.
[15:26] question is you gave all the nice features compared to you took as example Yokohava.
[15:30] Uh but is there any weak point or the day that you launched this and they reach TRL9 they will go bankrupt?
[15:37] Like is there anything that I mean that you cannot do?
[15:41] Yeah. That or even worse than than the existing spectrometers nowadays?
[15:45] Sure. So something we're looking into is power handling.
[15:48] So how much power can we um look at in the chip?
[15:53] So obviously with a chipbased spectrometer, you know, that's something we're figuring out.
[15:59] And then what we also uh what else can't we do?
[16:01] We're trying to find a lot of the limits now. So power handling.
[16:09] The question was and though please no I have a curiosity question for you.
[16:11] Can we put this spectrometer in a f contact machine and when they are ready?
[16:19] Maurice has the hand up.
[16:21] I'm queuing up for the So I am queuing up for the first demonstration. So let's discuss afterwards.
[16:26] Okay.
[16:27] So when you have something which we can check in the lab sure is director of font you jump from.
[16:35] Yeah. Okay. So very interesting talk.
[16:36] This is a topic that I have looked into two years ago.
[16:38] So still I'm looking uh following the research trends in my tech radar.
[16:43] So I'm I'm Yum technology manager at Zeiss.
[16:45] Um so I noticed a recent work from from Cambridge.
[16:48] Um so this optical convolution u spectrometer.
[16:51] So it's looks from from the abstract from from the container of the paper published in nature photonics is everything looks great and um and they also demonstrated to use an IR spectroscopy to uh to show the promise of um of their struct uh spectrometer for healthcare um applications and um yeah and my question is how do you compare your work with um with uh with their approach and uh especially in terms of the T trl and also for high volume applications for example consumer products imagine you
[17:29] have a sensor for miniaturaturized sensor for uh healthcare um with your smartphones some application like that.
[17:36] yeah so um I mean it's hard hard to compare because you're comparing a company to a paper so we'll have to see where they decide to go with it.
[17:46] obviously the TRL level of we're at is higher uh but we'll have to see where they decide to go with it.
[17:51] I know in the paper they mentioned healthcare um so yeah we'll see.
[17:58] perfect answer loved it you wanted a packaging partner Jo from fix raised the hand what's on your mind.
[18:03] yeah but I actually have a technical question so we definitely interested working with you on the packaging but one of the and this it's related to packaging how do you get the light on the chip uh because if you compare your solution with a standard uh spectrometer getting light into your waveguides is is a challenge.
[18:21] so I think your your minimum minimum uh u yeah the minimum amount of light that you can measure is probably different than the standard technology.
[18:30] Yeah.
[18:33] So we use edge coupling.
[18:35] Uh we don't have any grading couplers in our design.
[18:38] Uh we do just a traditional fiber attach epoxy.
[18:41] The prototypes we have so far are packaged.
[18:43] They have their thermal control electronics all within a small box.
[18:49] Diana.
[18:49] Oh no, sorry to fromtech.
[18:53] Yeah, really cool product.
[18:55] Um, just a question.
[18:57] You mentioned millisecond um cycle time or measurement times um which would be really great.
[19:02] How do you interface with the machine?
[19:03] Is there a controller attached and do you provide something like an analog output which we would really love as a machine vendor to make very high-speed alignment tools or tuning tuning integrated systems using your your sensor?
[19:16] Yeah, sure.
[19:19] So I mean now it's attached to a PCB and it attaches to the computer with a USB.
[19:22] Uh so it's a very small electronic addition but yeah would love to talk more.
[19:27] I'll also line up for the testing.
[19:30] Congratulations and we're going to keep
[19:32] In touch.
[19:33] I want to see the progress from a passenger seat.
[19:35] Thank you so much.
[19:36] Thank you.
[19:38] [music]
[19:45] [music]