Illuminated: IEEE Photonics Podcast
Illuminated by IEEE Photonics is a podcast series that shines a light on hot topics in photonics, and the subject matter experts advancing technology forward.
Illuminated: IEEE Photonics Podcast
The Transformative Power of Optical Networking in AI
How do cutting-edge optical technologies elevate the performance of AI systems? Tune in to our latest episode of Illuminated, where we feature Laurent Schares from IBM Research. Laurent walks us through the pivotal role optical interconnects play in modern data centers and AI clusters, illustrating how these advancements are transforming high-performance environments. Discover the intricate relationship between advanced optical networking, system integration, and the groundbreaking hardware innovations that are driving the future of AI.
In the episode, the moderator and speaker we dive into the human side of tech innovation with insights on professional growth and mentorship. Laurent reveals his personal journey, underscoring the importance of adaptability, teamwork, and a supportive work environment. From peer reviews to mentoring the next generation of scientists, we unpack how contributing to the scientific community fosters both personal and professional development.
Host:
Akhil Kallepalli
Chancellor's Fellow and Leverhulme Early Career Fellow
University of Strathclyde, UK
Moderator:
Brandon Buscaino
Research Scientist
Ciena, USA
Expert:
Laurent Schares
Senior Scientist
IBM Research, USA
Have a topic you're interested in hearing about? Let us know!
Illuminated by IEEE. Photonics is a podcast series that shines light on the hot topics in photonics and the subject matter experts advancing technology forward.
Speaker 1:Hi everyone and welcome to today's episode of Illuminated. I'm Akhil and, as the past Associate Vice President for the Young Professionals, it is my pleasure to be your host today. As the past Associate Vice President for the Young Professionals, it is my pleasure to be your host today. I'm a biomedical physicist and engineer working at the University of Strathclyde as a Chancellor's Fellow and Leverhulme Early Career Fellow. However, in my role for the IEEE Photonics Society, I'm supporting and promoting initiatives very much like this podcast to raise the profile of valuable young professionals within various sectors.
Speaker 1:Now the Young Professionals Initiative is for graduate students, postdoctoral candidates and early career professionals Basically, anyone up to 15 years after their first degree. This affinity group within the IEEE Photonics Society is committed to helping one pursue a career in photonics. We're here to help. We're here to evaluate your career goals better understand technical pathways and subject matters, refine skills, grow your communication and your professional networks through mentorship and help basically in any way. We can Now on to our podcast. In this podcast, we're going to discuss optics and AI systems with our special guest, lohan Skars from IBM Research and moderator Brandon Buscaino from Ciena. In today's episode we will hear from Lohan as we discuss optics and AI systems, his journey, and also we go beyond academia with career and research advice from Lohan, and there is something for everyone today, so stay tuned. Now to your moderator and your host right after me.
Speaker 1:Brandon Buscaino received a PhD in electrical engineering from Stanford in 2020, where his research focused on electro-optic frequency comb generators and enabling efficient and resilient intradata center optical links using co-packaged optics. Brandon is now a research scientist at Sienna Corporation, where he helps develop the next generation of digital coherent optical modems operating at data rates of 1.6 terabytes per second. That's really, really good for my Netflix subscription. Brandon has co-authored over a dozen journal articles and conference papers, as well as several patents, and is an active technical reviewer. In 2021, brandon was awarded the Camino Outstanding Early Career Professional Prize.
Speaker 1:Now Brandon is also an active volunteer of the optics community. As the president of the Stanford Optical Society, he organized conferences, outreach events and industry seminars. After graduation, he's decided to continue his professional involvement, participating with the society and across groups and conference committees, including the Optical Fiber Communication Conference, as well as advocating for congressional funding for optics and photonics within the National Photonics Initiative. He serves in the IEEE Photonics Society Young Professionals Advisory Committee, as well as the Industry Engagement Committee, whose focus is to support members in the industry with educational, entrepreneurial and standard based resources. He could not have had a better moderator for today's episode. So, brandon, take it away.
Speaker 2:Hi Akhil, thanks again for that introduction. Yeah, I think we're going to have a really good discussion today. So, as many of us know, the recent surge in popularity of large language models such as ChatGPT has brought artificial intelligence to the forefront of our culture. It is now seemingly impossible to avoid AI in our personal and professional lives. What many do not know, however, is that research and development in AI has been ongoing for decades.
Speaker 2:While AI systems have become popularized due to their recent accessibility and impressive interfaces, artificial intelligence has powered a transformation in computing, finance, healthcare and many other fields. The performance and impact of these systems has been shepherded forward by hardware advances in semiconductor manufacturing, chip design and optical technologies. Today, we'll focus on that last item. Optical interconnects were already crucial for connecting servers inside and between high-performance data centers, and their proliferation into AI clusters is ongoing and seemingly inevitable. We'll explore these trends and more with our expert speaker and guest, laurent Skars.
Speaker 2:So Laurent Skars, from IBM Research, is an elected board member of the IEEE Photonics Society of Governors. He received his PhD in physics from ETH Zurich, switzerland, in 2004. After graduating, he moved to the US, starting as a postdoc and currently as a senior research scientist at the IBM TJ Watson Research Center in Yorktown Heights, new York. His current research focuses on advanced optical networking for AI supercomputers. He has led or contributed to numerous programs on optical technologies for computing networks which he has received an IBM Outstanding Technical Achievement and multiple Research Division Awards.
Speaker 2:Over his career he has worked across the stack from devices such as high-speed lasers, amplifiers and switches to optical interconnecting packaging and, more recently, into networking and system integration. He has more than 150 publications, 20 issued patents and is a senior member of the IEEE and Optica. Dr Skars has also been a longtime volunteer in the optical networking community. He's been an elected member of the IEEE Photonics Society Board of Governors since this year. He's currently the deputy editor-in-chief of the IEEE Optica Journal of Optical Communications and Networking and for the Optical Fiber Communication Conference. He has served as both a technical program and general chair, as well as a steering committee chair. He's been a frequent invited speaker, has been a journal guest editor on data center optics and has served on program committees of leading conferences. Outside of work he has been a long-time youth soccer coach and referee and he has widely volunteered to promote STEM education in schools.
Speaker 3:So welcome Laurent. We're happy to have you here. Thanks so much for having me. Thanks Brandon, thanks Arkeel, and also thanks to the Photonics Society for hosting this.
Speaker 2:So let's. Why don't we get started? So, laurent, to start off for the listeners, could you talk a little bit about what an AI system looks like, what are the basic building blocks and how are they all connected?
Speaker 3:Yeah, ai systems. Yeah, that's a good question. It's really a full stack play right. So I think it's not only the hardware, it's not only the software, but it starts, you know, at the very top level is the applications. You know chat, gpt, as you mentioned. You know it's more on the consumer side. Or you know we, ibm, we play in the enterprise space, so our platform is Watson X. In that context, you might have even seen that on TV in recent advertisements or so Now, under the applications, that's where it's getting interesting there's the AI models and large language models and that's really where the generative AI comes in.
Speaker 3:That really made the Chagik-Diesel and those applications take off in the last few years. But you know AI models, they rely on a full software stack. You know a data platform. There's a huge amount of data that needs to go in there. And then even under that, you know you have typically a very large number of servers that need to be, you know, connected to a network. Together you have storage where you store all your data and all that stuff, and then at the very lowest layer you have the hardware. It's typically lots of servers in large centers, especially for training, and what's slightly different in those servers than in cloud is that they are heavily GPU-based or accelerator-based. Then on the hardware side, of course, there's the connectivity, which is the focus of this podcast.
Speaker 3:Now maybe just one more thought on that front. You know AI systems. You talk about training on one side, but then also there's, you know, the inference. You know, when you go on ChatGPT or whatever model, you're putting one request there that makes essentially use of a pre-trained model already. That's before. Now, training systems tend to be very, very large at some point big data centers, high power consumption, but inference you want to do that at the edge, you want to do that on your handheld device, or something like this. So the requirements there are typically totally different Very low power, very small. In that sense, maybe just one more thing, since we are talking about communications here. What's also different than in cloud generally is that it's estimated that generally about 60% of all communications is for accelerator to accelerator traffic in these AI systems and if you take specifically training systems, that communication bandwidth goes up 90, 95%. So we're talking a totally different ballpark than what we've been used to in the past year.
Speaker 2:Wow, that's quite a lot of data and quite a lot focused on the training aspect. So what is your role in this ecosystem? What is your research focused on? How did you become involved in this type of research?
Speaker 3:Yeah, so that's also a good question. How did I get involved into networking for AI supercomputers? As somebody who was essentially trained in device physics, optics, device optics and so on, so it's been quite a journey. I've been working across the stack from PhD and early research years, more on photonic devices and interconnects, high-speed lasers, amplifiers, optical amplifiers and switches, and then all kinds of interconnects and packaging how you put that together and then more recently I moved more into networking and systems integration. Now, all this background essentially, I feel that helped give me a really broad perspective and it's kind of key for all the system designs. As I mentioned before, you know, we have to full stack across it and while nobody's an expert on everything, it really helps if you are able to see a little bit beyond your little expertise and you can bridge the different layers together. I think that's a broad, helpful thing. So, in terms of importance of optics here for these AI supercomputers, I think there's a big focus. As mentioned before, high bandwidth requirements, a big focus on data movement At data movement.
Speaker 3:We can essentially classify into two groups, the way I classify it. The first thing is to avoid data movement. If you can have models or applications such that you can keep them as local as possible, say within the server, within the GPU, within a rack, you don't necessarily need to move them all across a big data center. That makes your network design a lot easier. And you don't need need to move them all across a big data center so well. That makes your network design a lot easier. You don't need to shuffle bandwidth across when you don't need it.
Speaker 3:So that often requires you know a solid understanding of what workload is actually doing, what communication patterns you have in those workloads. One thing you need is often smart scheduling as well, so you know if this job is finished and I have another job in my pipeline, where do I place it best to minimize the communication requirements, stuff like this. And then, of course, on the model side, you know I think it's a relatively recent field and it's fair to say that those models will become more efficient over time. So to kind of minimize data movement if you don't need to do it. So I think that's the general thing on the avoiding data movement side, but then people take every bandwidth they get. So there's a heavy focus generally on building faster networks. Faster networks, vendor servers across your data centers and so on.
Speaker 2:That's amazing. So I guess that leads into what I was just about to ask is what are the main requirements for these clusters? Is it latency? You mentioned bandwidth. What is the overall cost of these systems, or maybe not in money, but in terms of the time and complexity that it takes to actually scale these systems? What's changed?
Speaker 3:Yeah, I think it's all of the above right. It's always money, but money, you know, is hard to talk about without having something that works right. So people don't want to overpay and you know the businesses who run those need to be profitable at the end of the day as well. So let's focus on really on the networking part here for the supercomputers. So I'd like to elaborate. I mentioned before that we need faster networks. That's probably something I would like to elaborate a little bit on here. So, compared to cloud or even HPC supercomputers of the past, compared to cloud or even HPC supercomputers of the past, often the interconnects have been 100 gig, 200 gigabit per second, ethernet-based in recent years, maybe InfiniBand for HPC as well, and that's typically the path on the server. You come from a CPU, you go to a network interface card, you go to the top of the rack switch and then you go to all your fabric in your data centers. Now, what's different in AI servers is that those typically have many GPUs on them. Often you know four GPUs per CPU or even more of that like this, and each of these GPUs has a lot of high-speed interfaces, multiple 200 gig, 400 gig or even more interfaces. So at the end of the day, if you say from a CPU-based system, you're coming out with 100 or 200 gig On the GPU side or the accelerator side of an AI server. Often you're coming out with 800 gig, 600 gig and people even moving into 3.2 terabit per second, which is easily an order of magnitude more in terms of bandwidth that you need to deal with at the network level to move around. But it also has a lot of implications for power and packaging inside the systems. How do you get this bandwidth into the system? And once you have them in the network, how do you get this bandwidth into the system? And once you have them in the network, how do you move it around?
Speaker 3:So the industry essentially on that front is looking into several concepts. One side, you merge the standard data center traffic with the GPU traffic. But there's also a concept recent years that people say, okay, we do a front-end network, which is a standard data center traffic, with the GPU traffic. But that's also a concept recent years that people say, okay, we do a front-end network, which is a standard data center network, storage and all that stuff. But then we also have a separate network for training just to deal with this high bandwidth communications between servers. That's often required. So I think that's on the training side.
Speaker 3:You asked about latency as well. So I think in training often we're limited, throughput limited. So that's a bandwidth play. You want as much bandwidth in the system as you want, but for inference, I think the rest is often you want to process as many requests in as short a time as you can. So there the latency aspect often becomes dominant over the bandwidth aspect. So Cherry is not one size fits all, but I think for network engineers it's a fantastic place and lots of room for innovation here all across the stack.
Speaker 2:So on that topic, could you give our readers some rough size for these networks? I mean, are they in one large data center, are they in one room of a data center, are they between data centers, and is that going to change in the future?
Speaker 3:It's all of the above. So I think the big models that are out there there are tens of billions, hundreds of billions of parameters that out, that there are tens of billions, hundreds of billions of parameters and people always speculate. You know, exponential growth is not going to continue forever, but people are talking about trillions of data centers and I saw recent press releases where people are talking about building single clusters of you know 100,000 GPUs. So that's just enormous in terms of systems. Now, are these already built? I think most are probably a lot smaller. You can say you know thousands of GPUs is not a surprise. But then also, I think there's probably a market for some leadership systems, but not everybody can support those with the very large ones. Right, if you go into enterprise or smaller customers, right, people might want to train their own systems. But you don't necessarily want to have 100,000 GPUs because you will never be able to recuperate that money. You want to be as efficient as you can.
Speaker 2:I see, yeah, 100,000 GPUs consume quite a lot of power. I guess that leads me to something that you've worked on in the past, which is could you talk a little bit more about what co-packaged optics is and how it could potentially help scale these data center networks?
Speaker 3:Yeah, so I think you mentioned power. Power is generally a big problem there. Power I see it in two things. One is the OPEX aspect. You know power costs money and if you need to operate all that pay, that you know. I think there's estimates of, you know, $1,000 per GPU per year or something like this. Now, 100,000 GPUs, that's a lot of money Then you need to, you need to recuperate that at some point. But the other side is also a thermal aspect. So all this power generates heat and in the past we've been able generally to air cool those servers. But there's a big trend right now into moving into more liquid cooling, especially as the power densities become higher. But there's a lot of room for innovation on the power delivery and cooling aspect on one side. But on the other side, wherever you can avoid power, it makes the cooling and the generation aspect a lot easier and it makes it, at the end of the day, cheaper as well.
Speaker 3:So one thing that people have been looking at at the connectivity side is if you look how data move from a GPU or CPU into a network, typically, you know we have a big printed circuit board, you have some sockets in the middle with GPUs or CPUs and you have pluggable transceivers at the edge of this box. That's where the optics comes in and go out with fiber, but to connect those pluggable transceivers and the modules you need some copper trace in between. You need some copper trace in between and also that copper trace, especially at high data rates, that becomes lossy, so to overcome those losses you need to drive it, and that consumes power at the end of the day. So I think that's one aspect where people are looking at well, how can I shorten this distance between the optics and where the data is actually generated on my GPUs or CPUs? Can I put those two actually together? Can I co-package my optical plugable transceivers right on the module where my data are being generated, on the module or right next to the model? There are different flavors on that front right. So I think that's essentially the paradigm. That's one of the paradigms. So you know lowering power.
Speaker 3:The other thing is also, once you come out from, you know a module you go to, you know a stack of electrical interconnects, you know different electrical pins and interfaces and all those take size. Once you go into, you know the copper tracers. Those take space up as well. So one play where Copac Chop takes potential benefits as well is how do I get more bandwidth per area or per linear dimensions right into my network, into my optics?
Speaker 3:And co-packaging might help in that aspect as well, because now you don't need to fan out necessarily into electrical fields but you can put the optics right next to where the data are being generated. Now of course you know there's a lot of considerations that need to be solved there in that context, right, so I'd say generally co-packaged optics is about extending the bandwidth scaling roadmap for building my faster networks. I want to say for you know a given envelope, a fixed thermal envelope or power delivery envelope, and here at IBM we've done co-packaged optics projects together with partners funded, but there's also a bunch of large-scale industry projects that have come up in the last few years Now. Generally hardware development cycles are different than software development cycles. Often you need to incubate those technologies before they really become mature and usable at the large system scale.
Speaker 2:Right Incredibly difficult integration challenge as well. So I mean, on that front, are there issues with reliability and replaceability, when we talk about sort of upending this model, this pluggable transceiver model, and going to co-packaged optics, which you know higher data rates, highly integrated, right next to the switching chips.
Speaker 3:Yeah, so, absolutely so. I think one of the attractive parts of transceivers and switches being separate is essentially a disaggregated model. So now you can have companies who own the transceivers. If one doesn't work or whatever, you plug it out to plug another one in. I don't care about what's happening to my switch, that just can keep running.
Speaker 3:Now, if I co-package everything together in a co-package platforms Right? First question comes in there what if one transceiver fails? Who owns the problem? Actually, if I'm operating a data center myself, I want to be able to have this field replaceability, just keep it going. But on the other side, a transceiver manufacturer doesn't necessarily need to worry about how chips work, how co-packaging works and all that stuff. They focus on doing really well what they've been doing well and keep doing that. Now, if you put everything together, the transceiver manufacturer whoever needs to know much more about the packaging than they did in the past. We, as operators, may need to know much more about transceivers that we know in the past, right? So then, if something goes wrong you know, can I finger point to you, brandon, or something why did it go wrong? You know, am I owning the problem, are you?
Speaker 2:owning the problem.
Speaker 3:So I think people are looking at a lot of solutions around it. You know it's not a new thing to integrate things more. So you often look at redundancy at the system level. You know fail in place, do you have spare channels on one side, but then the other side of course you also got to work in improving your device reliability to the degree possible, right. So it's a whole you know kind of worms with potentially very big payoffs. But you know to develop this takes cycles.
Speaker 2:Yeah, it seems like a perennial problem between the consumer and the transceiver designers in terms of who wants to take responsibility for failures. Who wants to take responsibility for failures? Well, you know. So when I was working on some of this in grad school, you know co-packaged optics was coming about and there was a lot of talk about it. But you know, recently there have been some new types of technologies that have been entering this sort of power-saving space in, you know, intra-data center interconnects and especially for AI clusters, and they're called LPO linear pluggable optics and LRO linear retimed optics, and they've been proposed for again for power savings for inside the data center optical links. Just, could you give us an overview of how those are different from CPO and co-packaged optics and the traditional pluggable transceiver model?
Speaker 3:Yeah. So I think they're both very promising avenues. I would look at it probably from two points of view. One is the systems point of view. So it's kind of to keep up with connectivity requirements, make sure that within my given power and thermal envelope and cost envelope I can just keep scaling my cables and make them faster. So I would consider that for me as an operator. I would say that's cabling as a black box. It needs to work, not cost too much, within whatever constraints, and if that's the case I'm good, right. So just got to get fast over time. But then on the technology side, right. So I think it's what it is.
Speaker 3:It's somewhere in between, you know, co-packaged optics and the plugable transceivers and to some extent, to simplify it, really at the 50,000 miles level. Here it's what I mentioned. If we have my GPU or my CPU here, I go over the board trace to my edge of the transceiver, I need to drive this electrical line, I need to potentially retime it, and everything of this costs power essentially. And how can I minimize this power? Can I get by with only partial read timing, say only on the transmit side, only on the receive side, or things like this? So yeah, I think it's possible to some extent with a clearer system design. Now you need to have an end-to-end design. Really, you know what does your electrical channel and your transmit and your receive look, both on the transceiver side and on the module side, to really make sure that this channel is going to work at high speeds.
Speaker 3:If you get this working, does it help?
Speaker 3:I'm pretty sure in the short term we can get a little bit more continued bandwidth scaling for another generation or two, a little bit lower power than with fully repeated transceivers and so on, and there's less packaging involved there than, say, with a full CPO solution or a packaged optic solution.
Speaker 3:So in that sense, yes, but then again, if you zoom out in terms of how much power you're really saving, so I think it probably solves a thermal problem to some extent. But the total power needed at the system level typically is not moved that much by just pluggable transceivers, because the vast majority of the power is being consumed by your processors, by your memory, by your GPUs, accelerators, it's not by the cables. So even if you have cables, you have a single-digit percentage. If you improve that by a factor of 100%, you're not saving double-digit percentages of power. So I think it's something that needs to happen to keep the bandwidth scaling, if it's technologically feasible and if it's feasible it costs less, people won't consume it. But I think it's just one of those puzzle pieces that fit in into the whole system at the end of the day, I see, I see.
Speaker 2:So bandwidth is the most important and power scaling is important, but isn't the bottom line for a lot of these systems? So, okay, we talked about bandwidth and power. Are there other, perhaps non-technical challenges that are hindering growth or promoting growth? I guess in AI.
Speaker 3:There always are? Right, there always are, you know, with any technology that's grown so fast, right? So I think you want to be able to scale up your workforce potentially very quickly with very skilled people. Now, those skills might not necessarily be existing right away. People need to be trained to some extent or learning on the job, right, which partially is that right. So I think what I hear is popular at the college, grad school level right now is really learning about the generative ai models, which is important, of course. But then you know if we're building whole systems, we need the rest of the stack as well. We do need networking skills. We need to. You know the lower level system design, chip design skills and eventually you know if you really want to go to the next generation on multi-year or five-year plus time horizon. You know it comes down to semiconductors and materials. You know how do you make those faster, how do you scale those as well. So I think skills is a big, big factor here as well.
Speaker 3:And maybe one additional thing I think that's been mentioned very much trust, trust generally in AI models. Obviously, technological changes come by fast, but there's often a societal aspect that gets discussed at the society level. Is society ready to adopt this at whatever level? I think two aspects we've heard generally the current generation of models. They are prone to hallucinations to some degree, getting better over time and I'm pretty confident those go away over time. That's one aspect. The other thing is then also copyright infringements. If you use chat, gpt or so, maybe it shouldn't do it, but less important than if you're a bank or hospital or somebody who might get sued by using data that are proprietary in that sense. So a need for known data sets, a need for governance all around to make sure that the technology is ready to be adopted from a legal standpoint as well, as, you know, a societal standpoint as well.
Speaker 2:So lots of challenges all across the board. Yeah, yeah, further evidence that AI is permeating every part of our lives, going forward probably. And you know, I just. I have one more question there's. You know, in recent years there's been some. You know, I just. I have one more question there's. You know, in recent years there's been some publications on optical circuit switches for hyperscale data centers and AI networks. Do you have any opinion or ideas related to this technology? Is it going to have a future impact on AI systems?
Speaker 3:Yeah, ocs, optical circuit switching. I've worked on this for a long time, so it's how much time do we have?
Speaker 2:yeah, maybe two minutes so it's just like.
Speaker 3:So I think it has been recent high leverage papers that have talked about significant potential in real systems at the hyperscale level in terms of power savings. But a resource utilization using OCS and potential in real systems at the hyperscale level, you know, in terms of power savings, better resource utilization using OCS, and there's also been a lot of academic research over a long time. There are various technologies that are, you know, being considered there. So I think to really, again, you know, if you want to adopt this technology at the system level, you need to have a technology at the system level. You need to have a really strong understanding of what your workloads are. What can you actually do with it. It's just not like a plug-and-play replacement for existing networking technologies. What can be accelerated with OCS to being smart about connectivity at a high level? With OCS to being smart about connectivity at a high level.
Speaker 3:So if you compare to, you know, optical circuit switches to electrical switching, which is really the elephant in the room here. So OCS at a very level is essentially just steering light. You come seeing light with one fiber and you go out to another fiber and essentially it's you. You know a better patch panel, automated patch panel that can switch faster. Um, what it does not do is like electrical uh switches. You know this switch at the packet level, at the flip level or even better, they have a lot of buffering, they have logics inside your switch and you know we don't have optical memories or buffers in that sense. So that's not something that ocCS can do right out of the bat here. The other thing is electrical switching is also a huge market. I think it's a lot of $10 billion per year or something like this and OCS market is orders of magnitude smaller with a bunch of companies there, but it's not like a thousand gorillas in there In terms of the technology.
Speaker 3:So there's a lot of technology that people are looking at. It can switch very fast, but typically there's a trade-off. If you want to switch at nanosecond or microsecond level, typically you know that's associated with a lot of insertion loss, maybe with polarization dependence and also with relatively small switch size. So there could be one application for that. The other extreme is that you really switch slow, maybe at the millisecond or even second level or something like this, and in that case you can build much higher rating switch matrices, also with low insertion loss. But obviously, if you want to switch at a packet level, that's not something you're considering in that case, because it's just way, way way too slow.
Speaker 3:So that's more like being considered for integration with, say, a software-defined networking stack where software can look at what's the resource utilization of my cluster. Are there some servers or some parts of the cluster that are not utilized perfectly well? Well, now I have my automated patch panel here. Can I just reconfigure that and really attribute resources, compute resources on demand in that context? So I think that's something that's very promising and I clearly see potential in that area. The faster ones you know people work on faster technologies and try to make all of this stuff faster as well. So I think it's a very exciting field to keep following and I'm positive that we see more innovation in that coming up.
Speaker 2:Well, there's a lot to chew on there. We could probably have another entire podcast, but I think this is a good time to bring Akhil back and chat a little bit more about some professional development topics. So, yeah, welcome back.
Speaker 1:Akhil, that's been absolutely uh fantastic. It was really good to listen. Um, obviously, because I do something slightly different to the put. Both of you were talking. I was sitting on google, sort of talking and searching for every single term should have just gone to chat gpt, it would have told me what was going on. We're gonna I'm gonna change the pace just a little and I'm uh gonna have a few questions when it comes to sort of professional development, career development and things like that. So, uh, this will not be like your phd viva. This will not be like your postdoc interviews. This is more like tell us why you're so amazing sort of segment of the podcast. Um, we've talked about a lot of things and I've noticed you've covered, you've had a lot of experience, laurent, in everything you've done. I'll start with the first question, to which I actually have a follow-up Could you give me and the audience of the podcast and background of where you've done your prior experience studies and sort of a snapshot view of your career?
Speaker 3:That's a broad question, right. So I experience studies and sort of a snapshot view of your career? That's a broad question, right. So I did my studies, you know university studies, mostly in Switzerland, eth Zurich, undergrads. During that time, you know, there were exchange programs. I always was interested in broadening my horizon, so I did exchanges in France and Scotland, part of my studies over there, masters project in france, but then I went back to eth zurich for my, my phd.
Speaker 3:That was around the the time of the dot com areas, when I started. Now, fiber optics was really super, yeah, we need this huge connectivity. But then when I graduated, right, so I think there was like, uh, the job market dried up a little bit at least, uh, over what I was looking for. So, um, but then finally, you know, uh, through a uh colleague at ibm, uh, switzerland, I got in touch with, um, with the us. They say, well, we don't have jobs here, but are you interested in going to the us? There's a postdoc, exactly with what you're looking for. Say, well, yeah, worst case, you know, I get to travel. Uh, for free, going to the US, there's a postdoc, exactly with what you're looking for. Say, well, yeah, worst case, you know I get to travel for free over to the US, even if they don't take me. I go to New York and have a fun week there. But then you know, things get rolling and you know finally start with a postdoc here and then a permanent position later on.
Speaker 1:So that's how it started.
Speaker 3:Yeah, go ahead. No, no, go ahead. I'll let you finish. Yeah, so then, initially you know it's more on the physical layer, you know all the connectivity sides. But I was always interested. You know what can you actually do with the technology? So technology by itself, for me it's a means to solve, you know, broader problems, to have impact and in that sense you know the question for optics specifically is how can you make your system faster, how can you make them better? What's the application for all of this technology? So that's how it kept moving across the stack, broadening the horizon.
Speaker 1:That's fantastic. There's a diversity of geography, there's a diversity of experience, and the combination of all of that, effectively, is a culmination of everything that you're doing today. So, in all of that journey, what do you think was your biggest challenge, transitioning both geographically and your roles, and how do you think what worked for you to sort of overcome any of those challenges?
Speaker 3:Yeah, I don't think there's a single challenge in that sense. So I think what you learn you know I've been living in so many places working on very different areas Everything has essentially good aspects and bad aspects and essentially it comes down to you know, knowing yourself, what you like, what's really important for you, where do you feel comfortable? Well, you know, both challenged in a sense, and you know, academically and mentally and have an interesting job, but also you know, feel, feel at ease or have like a supportive group around you, both private and in and in. You know the workplace, what works for you individual, and I think that's a choice that everybody can only make by themselves, right. So, and I think I've found so something that's worked for me here, very happy, this, this environment fantastic.
Speaker 1:So I'm sure you've actually had and had conversations and met some very, very interesting people along the way any sort of perspectives on how it is to actually have a mentor? If you would like to mention somebody who's helped you during your career as a mentor and also, I'm sure by now you would have had a few roles where you were the mentor and somebody else is the mentee and this is a sort of a situation of being on both sides of the same table. Any sort of reflections on that if you'd like to share any?
Speaker 3:Yeah, okay, I think that's also a long discussion.
Speaker 2:Don't worry, we can.
Speaker 3:We can always bring you back for a part two and a part three yeah, no, I think generally, uh, it's important that that you open at some point, right? So, um, after you come into a new place you know, especially here, the ibm research center um, you, you come in as a nobody. Maybe you're an expert in just your specific part you did during your research, but there have been people here that have been working for 30, 40 years or even longer on technology and, at the end of the day, is, how can you learn from those? So don't think you're perfect on everything across the board. No, it's a team play, especially at the systems level. Here You've got to do your individual things and be really good at what you're doing, but then be open to what do others tell you, both on the technical level as well as on the way to get things done. So I think that's a lot of you know. You got to put in the effort, you got to put in the time to learn, to be humble about it, but then you know, get your hands dirty, get into trenches and work it out Right. So there's often no real shortcut to that. You know, at the end of the day, what's working best. Real shortcut to that. You know, at the end of the day.
Speaker 3:What's what's working best, I think, in this environments here is that you have a a bunch of experts who are really good in what they're doing, but also able to see across the horizon, to work with others together. You know, to make this a whole team play all across the board. So I think that's important in terms of um mentoring. And you know, when you come in as a, as a fresh phd grad, you don't necessarily know that that much. So that's important in terms of mentoring and you know, when you come in as a fresh PhD grad, you don't necessarily know that that much. So that's a little bit of a learning experience.
Speaker 3:You know Feedback. You know candid feedback. You know generally. You know try to be as positive as you can, but you know you still need to be open. Okay, there's some opportunity that you have. You could do that maybe better in that aspect. And then, as we go with newer people as well, same thing trying to pass on the knowledge that we have and say how do you get things done in a teamwork? We expect that we can trust everybody individually to do a really good job on their individual front, but then you still got to put all these pieces together. So mentoring, that's that's a very, that's a very broad perspective on mentoring yeah, so that's.
Speaker 1:That's very interesting because it quite directly connects onto aspects of what do you look for in a person? So say, somebody at a career stage is at an early career stage is listening to this and they would like to work with you. Or somebody in a person. So say, somebody at a career stage is at an early career stage is listening to this and they would like to work with you or somebody in your sort of career position, and they're trying to look around, network and try and meet more people to find out the answer to a very simple question what are you looking for in terms of a person at my career stage being able to work with you and work in this role? So, based on that, could I ask you a question about what were you looking for? If you were looking for somebody, let's say, a few years behind you in your career, a younger self, or maybe somebody out there who wants a similar career path, what sort of qualities and attributes do you think are good to cultivate at an early career stage?
Speaker 3:Yeah, I think that's also a broad question. I think there's no size fits all there and there I'm really on the front. I want to do these interviews myself. I want to have them evaluated by other people, not necessarily by AI. So I would not necessarily trust that my AI gives me a perfect first back, because often you know, of course you've got to have the qualifications, you've got to be, you know, in a technical, quick, moving field, as always, you've got to be, you know, top notch technically qualified and able to do that. But then, on top of that, you've got to be open. You've got to be open to learn. I think all this technology is changing so fast. If you want to keep doing what you keep doing things move, the market moves, the technology moves You've got to be able to learn all the time.
Speaker 3:Something we're looking for is motivation. I know a lot of other people are motivated. What do they really want to do? Motivated, but not in the sense of okay, here's my way or the highway and I don't want to do that, right. But you know, are you able to fit in here? Are you able to learn? Are you humble about this, right? So it's kind of a play between motivation and also drive. You know, are you willing to take the initiative by yourself or do you need a lot of hand-holding, right? So I think of course you need some hand-holding when you come in there, but you know, is the person willing to go this extra step and say, okay, yeah, this is something maybe I can come out of. I'm trying to do this. You know, take an initiative, take a lead on something like this.
Speaker 1:So it's a mix of, you know, qualification and you know, being humble about yourself, being willing to learn, being willing to play together and putting in the effort. Yeah, you're always looking for somebody who's ready to take the initiative, who's proactive, open to learn. These are the sort of things that we always hear, but it's always good to sort of sometimes listen to good advice twice, so it's always good to sort of hear all of that again. I've got two more questions and I'll pull brandon into this as well. Um, you've both volunteered for the society. You've both volunteered externally as well. What do you think that has added value in terms of your career, and how would you say sell the idea to somebody? Because we all know that volunteering is interesting. It helps you meet some very interesting people, you gain skills. There's a plethora of things you can achieve by sharing simply your time. What have your experiences been and how would you basically convince somebody that volunteering is a good idea?
Speaker 3:Yeah, also a broad set of questions. I can talk about that for an hour, right. I think on one side, you know it's an altruistic aspect to it and there's a more, you know, self-driven development aspect to it. A more self-driven development aspect to it, the altruistic effect. Science generally relies on peer reviews. Somebody, I'm putting time in to review your paper, you're putting time in to review my paper, whether it's blind or not blind, this type of thing. You need feedback to really drive the quality up there and things get better over time. So I think that's where the altruistic aspect comes in At the end of the day, especially junior people, not only junior people, but they benefit a lot from having experts look over your papers and say point out, what can I do better?
Speaker 3:In that sense, and I think that's passing on the knowledge, whether it's paper review, whether it's driving conference, whether it's mentoring, I think that's passing on the knowledge in what it's paper review, what it's you know, driving conference, what it's mentoring. I think that's generally a you know training, the next generation. Moving on, you know what works, what doesn't work Right. So on the other side, for myself as well, I think I do get a lot of things out as well. So, on one side, uh, certainly, um, working with top-notch people is that you have access to um, to the leading edge of knowledge, right? So if I um say I need to know something about uh lpo or so and brandon knows more about this than I do, right, so we can't do a talk uh company secrets, that like this. But I could call up Brendan and say, hey, give me a lowdown in 15 minutes instead of me having three days of doing to read it. I'm still not at the same level, just to give an example here. Right? So I think that's maybe a more selfish approach.
Speaker 3:But then, beyond that as well, it's often, i'd'd say, driving research agendas. You know when, when, when you meet with new people and you see what are the leading edge people working on, right, so what, what's really leading edge? What people are working on, where are the gaps in generally industry roadmaps, right? So if we want to say, push in OCS or co-packaged optics or whatever forward, right, so, okay, we might have this technology piece and this technology piece and this technology piece, but how you put it together, right? So? And I think having a broad network in that sense or, you know, knowing people is really helpful in that sense of kind of driving a full agenda that you know may benefit the industry as a whole or maybe you know a virtual company or something like this in terms of getting things done. So I think there are multiple aspects to this that are that are all important, right?
Speaker 2:yeah, I mean I think yeah, yeah, I think that you covered, um, yeah, quite a lot of the, the altruism and and uh, the networking parts.
Speaker 2:I think for me, um, my involvement with the, with the, uh, the optics community and these professional societies, really started in grad school.
Speaker 2:I I received a lot of support from uh professional societies like I truly photonic society, and that really was, um, uh, really helped me build a lot of community that I didn't didn't have in grad school and that was, you know, I think, instrumental in me deciding to go into optics and optical communications, and I think that that was something that you know, when I look back on a lot of these professional development activities that I'm involved in today, that's sort of the core driver is that I was given a lot of opportunities when I, you know, didn't have that community or was just starting out in this field, and it really was.
Speaker 2:It's really an opportunity to give back. So that's what I really am passionate about, and also, there are a lot of good things that come with giving back. Like, as Lauren said, you get to meet a lot of interesting people, you get exposed to new ideas and you really get to be a part of this community, the scientific community that's pushing progress forward. So it's very fulfilling on a personal level and probably also on a professional level too.
Speaker 3:Yeah, maybe one additional thought on there is that I wouldn't take it for granted. I think to some extent it's a privilege to be able to serve the broader community in that sense. Right Just to give it back to you know, to pass on knowledge to others, to other generation, I think it's an honor, it's a privilege to some extent as well.
Speaker 1:And it's very interesting because the idea of actually giving back to a community that's actually supported somebody at the very start of their career sort of makes a very, very positive feedback loop work, where, effectively, somebody gets a lot of support at the early stages of their career, turns around to become eventually the leaderships and the mentors of that particular organization, wherever that may be, is always a very good feedback loop, because you've been through the process, you've done everything that somebody after you wants to do, so I really appreciate that that's, that's the cycle that we are perpetuating in that way. One final question, and after this I will basically summarize everything that we've talked about uh, in what was it? Now we're on podcast number four, five. I don't know which uh future version. We're on um.
Speaker 1:We're here, we're sitting. We're talking about um technical aspects of the research and the output that the both of you specialize in. We're talking about careers, trajectories, developments, all of these aspects. If you had to distill everything down, this might be yours or this might be somebody else's in your career, a mentor or somebody who's helped you out. If you have to give one piece of advice to somebody who's a young professional who wants to sort of chart a similar course in academia, industry, big organizations, wherever they want to go. What would that one piece of advice be?
Speaker 1:It's for me or for Brendan.
Speaker 3:Brendan, you want to go. What would that one piece of advice be?
Speaker 2:It's for me or for Brendan? Brendan, you want to go first? No, no, wow. One piece of advice? Well, I think if I had to distill it to one thing, I'd say be passionate, show your passion, show your enthusiasm, show your excitement. I think that Brian Laurent could talk about this. When you see someone at work who is passionate about what they do, it inspires others, and I think that it's really only going to lead to good things in your professional career. So, okay, I've got many other pieces that I could probably give, but that's, I think, the main one. Right Be passionate and be engaged.
Speaker 3:Yeah, I think that's the first and foremost one. I think, in addition to that, you know, be prepared to get your hands dirty right. So I think, both at the technical level, you know, go to Trent, just do the work, really sweat the little details if you need to. I think that's super important. But then also later on as you move a little bit up the chain. So I think what you don't want to do at the later level is micromanage a technical deal that's a recipe for disaster in that sense. But that being said, I think there's a big difference between micromanagement and deep technical dives.
Speaker 3:I think you've got to trust your people. You've got to make sure that they all work perfectly well, but also got to make sure to hold them accountable. Here's what to say If you have a problem, well, we'll find solutions around it. We're easygoing about this, but you've got to go and explain to me in really accruciating details what you're doing at the end of the day, so that we have confidence that the system is going to work right. So I think so it's this passionate thing combined with, you know, putting in the time and the effort, that and the mentality essentially to, you know, to pull the things off the ground.
Speaker 1:That is extremely, extremely good advice. I'll basically add on to that to anybody who's listening. One of my closest colleagues and mentors in my career had once said, when I asked him a similar question, his response was sometimes the people that you work with are almost more important than the actual work that you do. So make sure that you're surrounding yourself with good people and make sure, whether you are at a level where you're building a team or you're a part of a team, make sure that you perpetuate that positive culture around you so that you enjoy going to work every single day. And don't we want to do that? Don't we all want to enjoy our times when we go to work? So we've talked about a lot.
Speaker 3:We've talked about a lot. We've talked about, maybe, maybe on that front, just one last episode. I remember at the very beginning, when I was here just telling a story right off the top of my head here, uh, I went skiing one one sunday and I think I was like, yeah, it's a hard technical problem. And I was like, uh, yeah, gotta figure it out. So maybe a little bit tense on the ski lift, but then, talking to you know, coincidence, sitting to a very, uh, senior people person next to me, you gotta talk on the chairlift on the way up on that side. So, so where do you work? So you're working at um, at uh, at ibm research, that's what I said. Right, oh, great, so you don't need to go to work, you're actually getting paid for having fun and I think that a lot of that summarizes it right.
Speaker 3:So you know always have the little things that you've got to work through, right, but I think having this supportive and broad environment, I think that goes a long, long way.
Speaker 1:Yeah, I've now got this image in my mind of you on a ski slope trying to do the math on the ski going. That's it. I've solved the problem. Now I can go down the slope. We've talked about a lot today. We've talked about networks and data centers. We've talked about complexities of larger systems, growth of data centers, co-packaged optics, career challenges, mentorships I mean the list is is so long that we've decided we're now on episode six. So thank you very much, brandon and Lohan. This has been an absolute fantastic chat. I really enjoyed the conversation. I'm sure everybody listening has as well. I'd like to thank you very much. If you have any final thoughts, now's your opportunity. Otherwise I will sign off for today.
Speaker 3:Be passionate. Thanks so much for your opportunity, Otherwise I will sign off for today. Be passionate. Thanks so much for the opportunity. Really appreciate it being on here. It's been a pleasure, excellent.
Speaker 1:Thank you very much, and we'll remember what Brandon said let's be passionate. Thank you very much everyone for listening. It has been an. I'm talking to both Lohan and Brandon today, and join us again next time for the next episode. Thank you for today and bye, bye.
