[AH] My name is Alex Henderson and I’m the Needham optical networking and security analyst. It’s a pleasure to have Poet here to do some fireside chat. We’re going to be talking about the company’s fundamentals and outlook and what’s going on over the course of the next 40 minutes. … Welcome gentlemen.

Suresh as the Chairman and CEO maybe you can give us a brief introduction to Poet for people who are not familiar with the company. We’ve got a very broad audience. My guess is some of them know you but some of them probably don’t.

[SV 1:08], Good morning, good afternoon, good evening, … depending on where in the world you are. That definitely applies to me. Its great to be here to talk to you about Poet, what we’re doing, what we’re about. So fundamentally at a very high level Poet is a photonics hybrid integration company. There are two major buzzwords in the industry today, especially in Europe. One is hybrid integration and the other is silicon nitride waveguides. Basically we do the combination of those two. We’re a hybrid integration platform using a novel application of dielectric waveguides that assist in a seamless integration of electronics and photonics, and the assembly is done at wafer scale, testing is done at wafer scale. That provides the economies of scale as well as the size and form-factor and performance benefits to really any photonics sub-system that you want to design. Our primary approach to business development is to apply our technology to the kind of broadening field of data communications where we can provide subsystem level optical engines either with or without the electronics incorporated in it for a variety of different market applications. In data communications, we’re currently deploying through qualification our 100G/200G optical engine that we have designed and developed for either CWDM or LR4 applications. The major value proposition for Poet really is form-factor, cost, architecture and then economies of scale associated with the ways we do our integration. Fundamentally we have an innovation that allows our customers to innovate. And that’s what we’re really all about. You know we provide capabilities to the customers that they haven’t seen before, and they have an “A-ha” moment about how they might want to architect new products. We enable customers to innovate in this space. That’s kind of what Poet’s about.

[AH, 3:33] Let’s talk about what an optical interposer is. Its not a term that I think most people are familiar with. Sounds very technical for most people. What exactly is an optical interposer and how exactly is it different from the approach that is more commonly used today?

[SV: 3:56] That’s a great question and through a variety of different exercises I’ve been trying to simplify the message as to actually what it means and what it does. Let me see if I can play this out for the audience here. An interposer is anything that interposes between two mediums. So it basically is a means of communication. Hence the word interposer because we enable communications between individual components. Now why optical interposer? Because there is a term called interposer that is very widely used in the semiconductor industry. Its been in production since about 2015 and that’s the electrical interposer. And the electrical interposer is basically just called “the interposer”. Nobody says “electrical interposer”. We have to qualify specifically to say optical. So an electrical interposer interposes between electrical chips. It enables electrical components to communicate with each other when they are placed in close proximity on the interposer. Examples of electrical interposer, such as graphics and memory. AMD does it and Nvidia does it. There is processor and memory. Intel does it. And they call it something else but its fundamentally an interposer. So there are various companies that as a consequence of needing high speed communications have migrated to this concept of an interposer to enable co-packaging or close proximity placements of electronic components. What Poet’s been able to do is build upon this interposer concept and layer on, if you will, an optical medium that enables communications to occur optically. And when I say communications to occur optically, I mean guided optics through waveguides. What Poet’s been able to do is to create 1 to 2 layers of optical connectivity in the interposer. So that allows a degree of functionality that we haven’t really seen in the industry before. And what’s particularly unique is that the electrical and optical interconnectivities don’t interact with each other. So they can crisscross. There is a significant utilization of space, and our form factor benefits as a consequence of that. So that is what we call the optical interposer. Now how does it vary from..

[AH, 6:37] Just to be clear so it is both optical and electric interposer in a single use case.

[SV 6:44] That is correct. The optical because it has optical communication layers, interposer because it has electrical communication layers. That’s kind of the genesis of the term we’re using. Most, even at 400G today, most transceivers or data communication systems that are in vogue today are not necessarily guided optics. There’s a lot of free-space optics that is part of the transceiver manufacturing that tends to result in expensive assembly equipment and capability, active alignment, lots of micro-optics with lenses and so on. They tend to be fairly discretized in term of how these modules are put together. So Poet is fundamentally a guided optics basis. So we do waveguides and waveguide based communications. We don’t use thin film filters for muxes and demuxes we use arrayed waveguide gratings that are built into our waveguides. We use extremely high quality, very low loss dielectric waveguides which are currently in vogue. If you look at most of the publications coming out of Europe, people are talking about these kinds of waveguides. We’ve just been able to do it with an integration capability with electronics that nobody else has done so far.

That’s the way we’re different. What we can do with a single chip integration using the interposer, the form factor is about a factor of anywhere from 3 to 6 times smaller than what can absolutely conventionally be done with standard optics. In comparison to standard optics, there’s a world of difference in terms of what Poet does because its wafer scale assembly and capabilities.  Compared to silicon photonics, which is the other term that is broadly used, we are effectively also a form of silicon photonics, just like Rockley might say that they are a form of silicon photonics. We all do things differently using silicon as a medium. Poet’s silicon photonics uses or develops an optical interposer, which is an assembly platform, or a hybrid integration platform, that enables active components to be integrated on it. We have the flexibility of which device we choose. We can use a silicon photonics conventional modulator, or an EML laser, or a DML laser. We could use any form of photodetector or even Lithium Niobate modulators. We do have a very broad flexibility in terms of what that [?] platform can do, which is great because once we develop all of the basics, fundamentals, and capabilities which we’re doing today, the scalability and applicability of that platform into the future is just very seamless. We are able to use those same concepts and it doesn’t matter whether it’s a 200G per lambda Lithium Niobate modulator.. Yeah, bring it on. Because we have the platform with the CW lasers already incorporated and integrated in it. That’s the flexibility that our platform or our approach to silicon photonics has, which is broad hybrid integration capable, not necessarily a form fit for a specific application. We do have that platform mentality which allows us to morph and track and adapt to a variety of different applications with what we’re doing.

[AH, 10:36] The technology that you are describing, is a subassembly component that goes into a layer of a multi-layer device that is probably manufactured by somebody else. Is that a fair way to characterize it?

[SV, 10:54] So the optical interposer is itself a silicon wafer. It is designed by Poet. It’s manufactured for us on a contract basis by Silterra, in Malaysia. Poet has consigned equipment that is proprietary to Poet in terms of the materials that it deposits and the materials that it etches and the materials that we measure. But at the end of the day what we get back [from Silterra] is an 8-inch silicon wafer that we then send for assembly. So our partner in China, Super Photonics, then gets this wafer. They assembly the components onto the wafer, test it, and once it’s singulated it gets shipped out. We have also established assembly capabilities in Singapore. As of this coming month we will be turning that on. And so we will have the ability to do research and development in Singapore for novel applications with our platform while Super Photonics focuses on manufacturing and go to market.

[AH, 12:06] Just to be clear on the process here: so you put it through. You get a wafer. You slice and dice the wafer. Then it goes to assembly. The parts are being manufactured in China. But then its shipped to a contract manufacturer who’s inserting it into modules that are going to somebody else’s transceivers?

[SV, 12:37] Our initial customers are transceiver module companies. So there is no CM [contract manufacturer]. We’re basically performing the task of a traditional CM. I mean if you look at most of the transceivers that are built today there will be a component like a laser which goes into a manufacturer who makes it into a TOSA or a Transmit Optical Sub-Assembly or a Receive Optical Sub-Assembly. Those then go to a CM. Then it goes to a transceiver module maker. What we do when we finish up assembly of our optical engine, it basically supplants the TOSA and the ROSA. It is a combined TOSA/ROSA into a single chip that then goes directly to the module maker. It’s a chip on-board assembly for a transceiver.

[AH, 13:14] So the end market here at the end of the day is probably the scaled out data center environments for the most part. Is that a fair statement?

[SV: 13:50] That’s where we are starting. It’s scaled out data centers and telecom to a certain extent, because a lot of interest in the 100G and 200G LR4 is for the telecom space. I mean it’s the client side which does [?] into the datacom. What we’re also seeing is a lot of interest in exoscale computing. We do have projects and programs associated with providing highly integrated laser solutions for AI, for example. And there is also a lot of interest with regards to exoscale computing as it relates to CPO or co-packaged optics. So we’re kind of seeing the conventional players interested in our platform for what it provides in the 100/200/400 and then we’re seeing the hyperscalers and the server compute guys interested in looking at what we’re doing for the co-packaged optics space which is quite exciting.

[AH, 15;00] So before we go down the curve of whose doing what with the product, can you talk about the dynamics around cost, the dynamics around performance, the amount of savings around the electrical envelope, the lossiness elements… the key parameters of why somebody chooses to use this technology versus alternative technologies.

[SV, 15:31] The biggest drivers to cost in conventional manufacturing is the time it takes to put some of these assemblies together and the testing that has to be done, and the various test insert points that have to be put in place. And then of course there’s no economies of scale. It’s a linear cost relationship. You put 10 people on 10 stations and then you start making these things one at a time. I mean the fundamental dynamics of cost are converting that into wafer scale capability and assembly. I mean we’ve spent the past two years now demystifying this concept, that photonics can actually be assembled at wafer scale just like electronics and that’s what we’re doing and that’s our vision. So the cost is anywhere from 30% to 50% lower, depending on which application you go after. Of course the higher end applications you see a much better cost benefit. LR4 you see a huge cost benefit because of how difficult standard LR4 modules are to make. CDWM-4 not as much because that price has come down. But even there we can demonstrate up to a 30% cost benefit because of what we do. That’s one vector. The other vector we measure ourselves on is performance. And you would say “Hey, 100G is 100G. Who cares, right?” (AH: “I wouldn’t say that!”) What we’ve been able to demonstrate with our 100G systems is eye margins over 40%, where most other companies 100G systems will have eye margins around 25%. I mean they would consider that to be very good. We are consistently over 35% to 40% depending on the temperature. So the RF capability of our interposer is so clean because we completely eliminate the wire bonds. So even at extremely low power levels, we can create wide open eyes.

[AH, 17:47] For people who don’t know what a wide open eye is that means the signal that’s going through the product is very clean and does not have any distortion in it.

[SV, 17:58] That’s correct. [With a wide open eye] you can’t confuse between a 1 and a 0 in a digital state. Not only that. Because of that, we’ve been able to operate these at up to 30% lower power because we can drop the currents down, we can drop the power down, and still get very good transmission. It’s cost. It’s performance. And then finally scale. I mean we do everything at wafer scale. Our interposer at the end of the day is a very advanced submount. But practically its like a submount. And that economy of scale is quite staggering compared to what others conventionally have been able to do. One thing that I would point out, is that because we do use the novel dielectric waveguides in our interposer, we are able to integrate the entire subsystem that includes transmit and receive. That has been a challenge for silicon photonics, silicon photonics in its conventional state. Its most successful implementations have been in parallel optics, like PSM-4 or DR4. You see very little, I think outside of Intel, you see very little CWDM-4 or FR4 coming out of the silicon photonics guys. And that’s because the demux of the receive chain on silicon photonics has always been challenged. Even Intel CWDM-4 does not use… I mean it uses external optics to do the demux function after 15 years of R&D. I think that’s one area where we do bring a significant advantage, it’s that our value proposition and our cost equation in the FR4 domain where you basically have wavelength division multiplexing, or different wavelengths of light being used to transmit information, we shine in those kinds of applications for sure.