 Yeah, welcome everyone. Beginning in 2018, the Purdue Engineering Distinguished Lecture Series really started inviting world-renowned faculty and professionals to Purdue Engineering to encourage thought-provoking discussions and conversations and ideas really with their faculty and students regarding the grand challenges and opportunities in their fields. Besides presenting a lecture to a broad audience of faculty, grad and undergraduate students, they also engage in an interactive panel with Purdue faculty, other stakeholders, and experts as well as students. Today's panel is about thermal and energy management challenges in large-scale sustainable computing systems. This is truly a grand challenge for engineering that arises at the intersection of exponential growth of the digital data of sustainability and new materials and electronics. Now without further ado, I'd like to introduce the moderator for today's panel, my colleague, Justin Weibel. Justin is a research associate professor in the School of Mechanical Engineering at Purdue University. He is the director of the Cooling Technologies Research Center, a graduated NSF IUCRC that addresses research and development needs of companies and organizations in the area of high-performance heat removal from compact spaces. Dr. Weibel's research group explores methodologies for prediction and control of heat transport to enhance the performance and efficiency of thermal management technologies and energy transfer processes. He has been a key contributor to the development of transformative cooling technologies supported by DARPA, NAMC, ONR, SRC, in addition to numerous sponsored research projects that transition these technologies to industry in projects with funding totaling over $10 million. Dr. Weibel will be the general chair of IEEE's I-Therm conference in 2021. He's associate editor of the IEEE Transactions and Components Packaging Manufacturing Technology. Dr. Weibel's academic record includes the supervision of 24 PhD students, 11 master students. He's authored more than 140 refereed journal and conference papers and was recently recognized as an outstanding teacher, engineering teacher by the College of Engineering here at Purdue and received the 2020 ASME electronic and photonic packaging division Young Engineer Award. Over to you, Justin. Thank you for the introduction, Arvind, and, you know, my panelists are equally and more accomplished than I am, so we'll get to introduce them here in a few moments, but I'm really pleased to moderate the panel today and I'll talk about, and I wanted to start and maybe I'll just tell everyone kind of what we'll go through is I'll read a brief abstract to just kind of set the stage for what the panelists will discuss today before I introduce them. And then we'll go ahead and have a couple of key topics to do that. So if I, if I kind of read a short kind of abstract that was prepared and wanted to explain what we'll talk about, you know, it is our digital economy and it's this rapid growth in the applications we know about and hear about cloud computing 5G, AI, autonomous vehicles. And all of these data driven technologies are generating huge amounts of data. And so there's a projections right by 2025 will have 175 zettabytes of data and no one knows what a zettabyte is because it's a ridiculous amount of right 10 to the power 21 bytes. And to put it in perspective, the total amount of data in the world that was kind of out there was only a single zettabyte in 2012. So this is really exploding. And all of this data is residing in these large centers that are the backbone of the internet enabling computation storage management of all of this information. And these data centers are extremely energy intensive so they use and consume about 1% of the worldwide electricity usage, maybe a little more. And there's projections for various growth models, whether that's going way up depending on how we intervene, or can we keep it right at a reasonable amount but regardless that's on the same carbon footprint is basically the aviation industry so it's a massive amount. And in this very big picture challenge right of a large data center energy usage, we can still drill down to like the transistors inside of it. And there, you've then you're dealing with heat flux densities equivalent to the surface of the heat flux of the sun. So, right, you've got this multi scale right problem where you have energy and thermal management from a few Matt nanometers to the size of football or soccer fields. So we need to engineer these computing systems to be right efficient reliable and sustainable. And so this is why we need to have right research at the intersection of this electronics packaging reliability techno economic policy. It's really imperative to these large systems so our panel today we've got industry and academic experts I personally feel that this is an area where there's been a great history of industry academic partnerships. And to discuss their perspectives on how engineers can rise to the challenge of this particular problem and talk us through that. So that's the background. I want to introduce our panelists I'll kind of just have them say hello short short bios here just so we don't take up too much time but the first Professor dredge Agonifers are distinguished lecturer he's a member of the National Academy of Engineering, a distinguished university professor of mechanical engineering at the University of Texas Arlington. He has not one but two centers he's the site director of an NSF IUC or see on energy efficient systems as well as the director of the electronics systems packaging center at UT Arlington. And the research is on data center cooling 3D packaging and the cooling and reliability of micro and power electronic systems. So welcome. Great to have you here. Of course, just absolutely. We also have one of our new faculty that we're really eager to start working with and collaborating with prevent Professor Rebecca chess. She's an assistant professor in mechanical engineering and holds a joint appoint appointment in environmental and ecological engineering at Purdue and her research is at the intersection of technology economics and policy to enable transition to decarbonized energy systems. So welcome Rebecca. Great to have you here. She's group that he is the director of the thermal mechanical solutions at Intel's data platforms group so he leads all formal and mechanical architecture execution and innovations development for Intel's full technology portfolio and that's this long list of different technologies and where they insert right there their devices. And of course I'd like to point out always he was recently recognized as well and will soon be awarded an outstanding mechanical engineering alumni award for Purdue because he got his PhD here with buzzer J Gore in 2004. So welcome issues. Thank you for joining. Professor Carol hand worker is the Reinhard Schumann Professor of materials engineering at Purdue. She also has a joint appointment I know and as well and her research applies thermodynamics kinetic theory experiments with phase transformations and interface motion to solve important industrial and scientific problems. And I know she's been a lot involved a lot in the adoption or facilitating the adoption of lead free solders and electronics and a whole host of lifecycle analysis reliability assessment of these systems. So thank you for joining Carol. I have Dr. Madhu Iyengar who's a senior staff data center engineer at Google. And so in that role he's involved in the design and delivery of Nick generation it and data center if infrastructure and I'll point out he's really a community leader in our field in our community on the industry side in thinking about energy efficient cooling technologies and road mapping activities he's the organizer and has helped been instrumental to writing the chapter on thermal management in the I triple genius integration roadmap as well. So very pleased to have him here. Thank you. Okay, so now I want to ask some questions and get the discussion going and stop talking myself. So I think we should start as kind of with some ice breakers because maybe many students and others in the audience are not really familiar with these topics so I start with you. If someone or a graduate student in the room wasn't familiar with what a data center is what it entails. Why is it consuming all of this energy and what are we doing for thermal management. Could you just give us a brief introduction to that this at the stage. So data center can be a large or small infrastructure their houses, electronic equipments, such as servers, storage devices and so on. And the purpose of it is really to transmit digital information. So we have it parts that may be a few hundred square feet, like the ones we built out to talk about that later. And we have companies like a Facebook, I don't think my do knows what that is. I'm just kidding. But as, for example, in the local area in Fort Worth 800,000 square feet, a data center, right. So, and more so it's, it's the size can be quite significant and you can certainly appreciate how important it is, especially during the covert times, being able to have access to information. So it's really an infrastructure that houses equipment dealing with digital data. And of course they, they are then consuming, right, large amounts of energy to do that right and and what is produced from that and the inefficiencies is a ton of heat right and how to manage that. Yes, I could talk about that if you want me to. No, I think, I think that's, you know, just kind of the premise. So when we talk about it. And so thank you. And, you know, I think we've got a couple of industry panelists on to so as she should do I kind of kick it to you to say, you know, I know this is an area where there's a lot of progress if you look at, you know, since the last decade. We all know that the installed capacity it's like 500% something else but we actually haven't used that much more energy for computation so what what have people done to solve this challenges, they're just looking backward a little bit maybe before we look ahead. You want me to go first just. Yeah, sure. Yeah, I think there's a huge boom. I think after the dot combust and data center growth in the 2000s. And I think there was an intense amount of scrutiny on energy consumption and efficiency. And a lot of good things came out of that. In terms of reducing the cooling consumption best practices. So I think the last 20 years has been a pretty awesome journey in improving the reducing the ratio of cooling energy use, divided by the power power used by the it. I think where I see is going is that's only part of the story but the other part of the story is devices are getting more and more powerful, which means they consume more power, they have higher power density. Part of the reason for that is, you know, you have the Moore's law where you keep adding devices and also reducing the length scale of the devices. But you also have something called the end of dinner scaling where until about 2006, as you shrunk the devices, the power consumption also reduced a great deal, but you don't see that happening to that much extent. So what it means is as you put more transistors on a chip, and you don't reduce the power linearly, you're going to have higher power densities. So relating that back to energy efficiency, the bar is now higher, you need to, you need to do the same highly efficient job with a much harder task of removing heat more, more effective. And would it be wrong to kind of say that, you know, in the last we were able to kind of identify there was a lot of low hanging fruit previously, right, we knew of these like horrible inefficiencies laying out data centers, you know, just, you know, mixing hot and cold those things have been resolved, and now moving forward, you know, especially with the end of this, you know, on the electrical device side, we could just scale it smaller reduce power. When that started to end, then it became more and more challenging and will be a lot harder going forward. Yeah, yeah, exactly so. Ashish, any thoughts on just that history. Sure. I think you guys said it correctly, but I think the, the history piece of it I think will continue with us for a little bit longer, in my opinion. You know, a lot of, when we established those huge data centers that Regie talked about or even smaller ones which may be very big cost investment for some of the companies which did that, they want to operate it for as long as they can. And at the same time, their data processing needs are continuously increasing. So they're trying to push the limits, right, I think even a small, you know, a decimal point improvement in efficiency can actually mean millions and millions of dollars of savings. So I do anticipate that the companies will just keep on pushing optimizing as they try to upgrade to the newer products which require more power. They don't want to really go about changing the entire infrastructure or building another 800 or 1000 or a million square foot data center. So how do I get the latest technology really service my customers in the existing infrastructure. But I guess in that so it's still going to be we can live off these kind of percentage gains because it still means a lot right by economically there. Yeah, and but at the same time, like you started this panel, right, where we said in 2012 one zeta by 1025 175 zeta by right. Right, so I mean, as of now, only the half of our data that we have right now was just created in the last past two years. We have reached this inflection point where in the industries are getting reshaped by leveraging data. So while we see examples of digital transformation emerging, it's important to remember that only less than 2% of the world data has been analyzed right it's like, it's like the new oil right I think the term has been there for a while, that data is the new oil. And just like you dig in and you extract the oil, you have to that that data exists but people. They really started drilling it out and being able to use it in an efficient way. Yeah, yeah, but so then I guess that's where I you know you kind of say it'll continue and where I kind of ask and would be interested in all your opinions on is. So if it is this massive increase right all the data was in the last two years, we probably can't make efficiency gains at the same rate of the creation of data there's just no way to have exponential efficiency gains like there are increases in data. The use has to go up right significantly in the end and we can't kind of maintain this level I think that's kind of the problem we're facing. Exactly, I think, you know sub linear growth is a popular term right growth is is is going to happen. It's how how can you reduce the growth of energy with the growth of the business or the growth of the data data mining data consumption. So, trying to reduce that cost function or energy function or environmental function, subdue that as much as you can, while the actual growth of the internet and the cloud continuous. So we're talking about growth of like energy and economics and other things so now is my opportunity to pass it over to Rebecca I think you know your ex, that's your focus of, you know the lab and group you're establishing at Purdue is this techno economic assessment So can you just give us a little background about that and maybe where you see intersections with this community. Most of my work to date has focused on electricity energy storage. So, somewhat different than data storage. So you can't just say storage is a shorthand in this talk. But so thinking about how we take technologies that have already been adopted so for tumor electronics, everybody has a lithium ion battery in your cell phone. But how do you take those kinds of technologies and scale them up into systems that are low cost enough that we would want to use them in our cars or use them on our electricity grids to really incorporate more and more renewable energy and so when you start asking these different questions. What might matter a lot in designing a battery for consumer electronics might not be all that important. Once you get up to these bigger and bigger battery systems you know for if you're trying to store energy for eight 16 hours a day. You might have very different requirements than you might have. When you're thinking about these smaller scale systems and so how can you best manipulate the manufacturing process kind of change the design at like the fundamental materials level. In a way that will reduce the cost the overall system level to how can you kind of take advantage of those to ultimately build these decarbonized energy systems to provide the electricity that's reliable enough to power you know our cars to switch over transportation. But also providing all the other electricity that we use, whether it's in data centers or just in day to day life. And that's I liked the parallel between all of your work on battery systems and battery storage because you've thought about like the scaling of storage of energy from that perspective and this is on data. And I don't know if all the parallels exist but there are some and I think it is this is a techno economic driven thing in general and what we can do and if we need new technologies. You know, you know about trying to get technology transition right from right to industry it's hard and it's driven by right and more complex set of things than just right good thermal performance or cooling performance or other types of things. Good. Yeah, you want me a comment. Sure. I wanted to say that the projections and electricity usage that if you look at an hour show that in a slide later, what was projected. In 2010 has significantly dropped because of improved management. People like Google and Facebook hyper scale and hyper scale and and the cloud versus 10 years ago, they're now going to be 90% is going to be hyper scale versus the traditional being 90% in 2010. Right. So there is. So we are actually doing a really good job of of improved efficiency. So if you include best practices and and hyper scale, your efficiency rate really can go down quite a bit. You know, you improve quite a bit and thus reduce the electricity usage so so we're actually doing a pretty good job we but we got to continue to do that. Yeah, yeah, this the idea of just with scale you can do things at scale you might not be able to do in these small outfits but there's always the need for small to where there's like latency driven applications and need small but it as it migrates and you know this massive data explosion I think is is going to be stored on hyper scale systems. Okay Carol, it's, I want to involve you to because kind of the other angle and aspect of this is not the energy use just associated with okay the electricity being consumed but there's also these are not these are item you know products with a life cycle that get used and especially the turnover on electrical systems is quite fast and and the lifespan of them and there's an energy cost associated with that. I know you've worked a lot on life cycle assessment of the systems can you can you kind of tell us more about what you've done there and your thoughts. So we've been talking so far about the fact that you know with data centers. When doing a life cycle assessment, you have the original energy cost and all the other costs are like 11 different environmental and human health impacts associated with that anytime you just for the manufacturing and putting it into place. There are these costs that are associated with it that have to be divided by in where in is the number of years that the center is going to be operating. So it depends critically on what the reliability is, but the business models are associated with the replacement of the various parts of this. And it's much more nuanced, it is much richer than just figuring out what are the energy costs in a given year. So I've done a lot of work with the electronics industry actually with Google, Microsoft, Cisco and Seagate, just in hyperscale data centers for hard drives. So those hard drives and SSDs are swapped out every two years. They still have three years of warranty in them, but it was brought up before. You want to be able to keep scaling the services and the equipment with with the demand. So you want to be able to go up a factor of two or a factor of four in storage. And so keeping the same physical footprint of the facility. So life cycle assessment is really sort of retrospective. What one really needs to do is think about this circular economy and Google is really great at this. They sell over a million hard drives a year in the open market that they retire. So, and other systems they they sell these sell these in a cascading way through the through the world economy. So these are other points just to remember. Yeah, and Carol, I said I'd warn you if your audio is cutting out so we did maybe you want to go sans video. Excellent. Yeah, so if maybe Carol you mentioned before you might, you're going to have to leave Paris and turn your video. Yeah, so is the audio better right now is better now yeah. Okay, so basically there are a lot of different components I don't know what you heard that was part of that. But there are many more components that are actually switched out and their life cycle costs and and economic costs associated with those. And it's really important that the more general public out there understand that the companies are actually being very aggressive with respect to making sure they're just not putting these in the trash, and that the the energy that's already embedded in all of these products. They, they actually aren't going to waste either. And that's another kind of critical perspective of this and one that's, I don't know I always talk about and think about is like this hidden energy use that I can't be ignored but I think industry does a great job of being aware of that. You know, from a research perspective what is, you know, I know there's a lot of faculty at Purdue working on life cycle assessment or however you know designed for life cycle. How does research help intersect and affect that process and inform what industry is doing how do we help and that. So one of the key ways that we can help is that if they're these technological gaps that we take into account. Economic assessment life cycle assessment logistics as quickly as possible and as early a stage as possible in the technology development. Because we don't want to develop a technology that nobody's going to use. So that really, it's been our approach that, you know, for low these ultra low low temperature starters that are being developed we're doing it with Intel and major companies, because we want to make sure that there are no show shoppers shows show stoppers along the way, and they see it as a critical technology. So we're getting involved with basically the whole range of stakeholders suppliers for the new technologies equipment manufacturers. We work with lamb industries, our lamb research that is developing a lot of tooling applied materials and with ASML. So, you know that that's how we try to to mitigate the risks of doing things that aren't relevant. We're working on, you know, advanced liquid cooling solutions and things I should be talking to you about whether or not we're using, you know, materials and fluids that have some horrible right long term costs or benefits and I should be talking to, you know, Rebecca about, you know, other kinds of economic assessments and how much they cost I think that's the point is that we've just got it. It's a big picture. This one is a big picture problem and it's, we won't get. There are lots of creative and good ideas but to get them matured early on we need to consider this even right and also operation environments are very important. So I know that director has worked a lot on corrosion effects. So in practice and I was talking with my colleagues at Juniper the other day where they, they talked about a case that they had heard of where the there's actually a gradient in corrosion from where the air inlet air comes in to the air that leaves the system. So it's, it's all of those things plus the harsh environments and the different locations where where things are going to be used. Well, I think, you know, we've kind of set the stage a little bit and had some conversation and I hope everyone in the audience now is thinking about questions they would have for the panelists so I'll just say and I should introduce Aditya, I came to die is PhD student, you know, working in this space and area so he he's kind of the moderator looking at Q&A so please do actually use the Q&A function if you have questions for the panelists will turn to him periodically throughout the event to kind of draw anything about what drops and questions if you have any so do type them there. I don't know if we have any now or if we can go in I have a couple of I have a couple of questions and I got from the Q&A that I can have for the panelists right now. Yeah, let's let's do a question from the panel. Okay. So there's one question in the Q&A about different levels of liquid cooling solutions for data centers and if they can talk about like some of the examples and their opinions about what the future of these and where they see the broader adoption of each of these technologies. I can, I can start on that one. If that's okay with you Justin. Anything yeah I'm not here to stop anyone from talking go ahead. Yeah, so I mean, liquid cooling right this is about liquid cooling in data centers I think liquid cooling is already been extensively used in data centers I mean first you have something maybe you call as free air cooling where you have data centers that's extremely cold temperatures we're just able to utilize air. Sometimes you try to pack more and more products or there's more system densification that happens at the data center. The power goes up. Sometimes what we're over time what we have seen is that more and more companies are moving towards some form of liquid cooling where either you take the heat away from the in the air and then you cool the air through liquid, and then companies are trying to bring that liquid closer and closer to the place where the heat is getting generated. So when it comes to like things like cold plate, they are actually now getting quite popular over the last few years. And at some point I think in the future we do anticipate that the fluid is going to get even more closer to the silicon. And if you look if you just Google liquid cooling microchannel silicon microchannels you're going to see a lot of papers research work happening in that I do see a big hurdle till we get that point because it is. That's the place where the big companies which are making the product and the companies which are deploying those products they really have to come very close to make that kind of a reality happen. Another big shift that we are starting to see right now is something called as immersion cooling. I know this has been an area in the industry in the academia for quite a bit of time. I know Professor Dereji has been working on it for also for a long time. And I'm seeing an explosion of interest in that space. The sum of it is driven by edge computing and the anticipation of 5G coming. The amount of data that is going to get created is going to be so big so huge that it is not economical to move that data to a data center and get it processed over there for various reasons and if this time we can get into it or interest. So companies are trying to analyze the data either it's, you know, really utilize that oil that like talked about earlier, very close to the place where the data is getting generated. And that's leading to some computing happening at all kinds of locations, altitudes pollution, and you cannot really have this keep on designing those individual small data centers so there's something called immersion cooling where you have a dielectric fluid and you immerse the entire system in the dielectric fluid so you're not really utilizing even fans in that space. So that that is becoming very popular recently. And I expected to really explode in the coming years. I can inject to. Yeah, I was going to ask you to please please. Yeah. So I think just like she said that, you know, obviously you want free cooling. If you can just bring the outside air in certain environments you can do that. You know, from a thermodynamics as you know you, there's what we call a recommended zone so you have to control assuming that the biometric pressure is fixed you need two properties to fix the state. And so, let us say that says the relative humidity and drive up temperature so it has to be in a certain range so you might be able to do that and if you shift, it's okay. If it's out of there for a while. If that doesn't work but then what Cal said comes into play here if you start doing that. Then you're going to have some contamination so you bring the outside air and so you can have some contamination issue. It's a lot. We've done a lot of work in that area so I certainly refer you to that. So, follow up on that free cooling is directing indirect evaporative heat exchanger, where now you use the latent heat that's very significant so unless you're in the southern part of Florida certain areas where you know the Web of temperature does not allow you don't you don't have that delta T to work with you can do quite a bit of good stuff was Direct and indirect evaporative heat exchanger direct being better but indirect if you have to because it's a lot more flexible and plus you don't have to worry about contaminating things. So a lot of work in that area and certainly is being used extensively. People like IBM, I mean Google and so on and so forth, use that extensively for the big data center is very current start used to use spray and then cooling media and so on. Then you start talking about indirect liquid cooling I call it indirect right that's the IBM, because the, the, the coolant in this case water is not a direct contact with the, with the device, because it's not a dielectric. So there's a lot of work in that particular area but I mean, I remember being my IBM days you know thermal conduction module module I know you know that well they choose stuff. So, so been doing that for since the 80s, but there's a lot of work now in that particular area, you know, using people call it micro but not sure it's micro might be a mini channel. But there's a lot of work in that area we're we're right now working with a major company developed, I mean, looking at indirect liquid cooling. So, a lot of interest in that area certainly has limitations then you start looking at immersion cooling, the beauty about immersion cooling is, as you start, I think, my do alluded to this very well these are not scaling right since early 2000s right that the, the power stopped scaling as a square root of 1.4 so the power density. Stop creeping up so you had to fix the TD thermal design power, or, or the other way we're looking at is the frequency. And so in order to get performance you got to start putting a lot of course in there right so been doing that. But as you start moving now start looking at heterogeneous integration to gain leverage, right, because the frequency time between frequency start really increased four to five years right. It's still doubling transistors but taking more, more time, but you start doing this and start looking at heterogeneous integration, where you're stitching a bunch of components, either in two and a half D and 3D. That is really not something that a lot of the folks currently looking at at the data center level, but that's surely coming and immersion cooling is just a great way of doing it because it couldn't care less because it's dielectric fluid right. But again, it's the material people that win here to we've the PUE you can get 1.03 and stop no problem. It's the material compatibility issue that one has to work and so there's a lot of work that needs to be done in that particular area or else it's not going to sell. Yeah. So it seems like kind of what you're saying and she's as well as and I hear this question a lot is like, you know liquid cooling where is it. I guess it's always been there there's this suite of technologies there's going to be applications that need it right some applications that justify very high powers locally and for climatic conditions or other reasons and it's been an inward migration right do it freely with air if you can but different applications higher power densities more compact to the point where, you know, when there is a growing mining craze it made sense for the density of those systems to do immersion cooling and so these technologies are there and it's going to be the application that determines it. Maybe I'll had I see a question I want to get to soon but just because the term came up and and one of the trends and I'd like to just talk about it or I think make the students and audience familiar with it because we're going to hear about it a lot about the thermal management coming up that you used or a J with heterogeneous integration. So, you lead kind of the thermal, you know, working group in IEEE for heterogeneous integration for those that don't know what that term is what it means and what it means for us moving forward could you just say a little bit more about that and explain. And Dirige said it very nicely right with the pure, I guess, sub device level performance, getting better but taking longer to get better. There's a huge exponential growth in innovation and stitching devices together, putting memory close to the ASIC logic chip, having other functionalities like photonics and maybe stacked memory. So, your cell phone is is not has a lot of heterogeneous integration just driven by all that functionality going into that phone and inside the phone there's heterogeneous packages. So, I think what happens then is purely from conduction you have conduction resistances, you have dissimilar devices with different requirements of temperature in close proximity. You have different devices heating each other. So I think it becomes a pretty rich space for thermal research, thermal innovation. So I think that is definitely an area that is a big growth area for innovation and research, which is why you're hearing about it a lot. And Carol, you of course work in this space as well and think about some of the materials and mechanical and reliability challenges about putting all of this stuff that used to be separate forcing it in the same package or there's thermal challenges but you know what other challenges do you foresee in that space that you're working on. So there's one particularly interesting one that I think most of the audience here hasn't heard about. So if you start having really large die. And we're working with somebody who has an 85 millimeter by 85 millimeter die, they're very thin. So now you have to take this package, and you have to assemble it on to a circuit board and you put solder on that. You think okay, you just put the solder on it melts the solder, and it makes the, you can make the interconnects very easily. So that is that those packages and some of the ones that my do was talking about. They're much more complex than they were before, and they warp, they work like crazy. So even before the solder melts, you have them warping like potato chips. And so with the conventional solders. So we're having issues just in minimizing these kind of manufacturing defects. And we have to be able to, to put all the, the heat dissipating systems on top of that we have to bolt them on top of it as well. And if these are warped, we can't do it at the end we have to actually do all of the processing with the applied load. So it's not just to make the make the boards make the components. We can ignore the manufacturing. There are a number of serious challenges with respect to that. And so one of them is that came up in a question was low temperature solders. Why would you use those well, if we don't heat have to heat the board and the component up as high doesn't work as much. So we will. So now we have a driver for low temperature solders. However, their maximum use temperature is not as high as some of the high temperature solders. So there are a number of problems like that that we're dealing with, not just in the, you know, in the, in the end, do we have a reliable solder drive but all that has to take and be taken into account in design and manufacturing, and we're not going to be able to to test our way into this, because one of the packages that we deal with right now, it costs $5,000 a package. So modeling simulation for the whole mechanical thermal electrical function is becoming more important. So those are just a few of those issues. Yeah, and I see, you know, I'll stick with one more. I'm looking at the Q&A one more question because I often see it at kind of the device level and then we can maybe talk about, you know, again some of the large scale energy challenges is, you know, there's, there's so much heat generation that we talk about either promising what's on the horizon for revolutionary ways to store process data. Not generating the heat in the first place. Is that a viable solution and what's there? Does anyone want to field that question? I think that's probably for computer scientists, but, but I agree with you. I think storing data, optimizing how you store data, redundant data and so that is something that needs to be looked at. And interestingly to either listening to Carol, there is so much opportunity for multi-disciplinary work. Really, that's, that is really what I see here. You can't really do, for example, thermal without understanding packaging. You can't really do thermal without understanding materials. In fact, Ravi Mahajan recently, Ashisha, he gave a, he received an award and gave a talk and his call to action was, you know, Tim materials that are order of magnitude better. But when you start talking about new materials, you got to start talking about their reliability. So those material people don't get away from us at all. Yeah. Excellent. And there's some actually to the panelists to it that you want to field any other questions by texts in the chat to if we don't get to them all and and J. So J Gore asked a question as well. And this is, you know, maybe thinking back about the energy uses. Is there any way to use all of this heat generation so is there, are there any viable or promising waste heat recovery, where some of this energy just doesn't go to waste. I, maybe I'll say a little bit about that but then turn it over to the panelists certainly it's very low grade so that's one of the difficulties for this right the temperatures we're talking about produced by electronic systems are certainly less than 80 Celsius type temperatures. And so, many of the conventional recovery mechanisms just are not viable and but Jay specifically said, asks about thermal electric generation to do some right powering of sensors or secondary circuits is that anything that is being considered or been considered. Maybe I can make a comment I think that's where our technical background actually reveals the drawbacks, as you said, when you calculate the car not efficiency. And, you know, real world devices maybe the refrigerant based ones they get to maybe 40% of car not. If you do the math and you do the ratios and Kelvin instead of Celsius, it ends up being you know sub 5C 5% conversion rates. I think. So that is one challenge the second challenges for energy conversion you end up having to put these devices in the heat flow path, and that really hurts the effectiveness or efficiency of thermal the thermal resistance goes up. Also, is a barrier. I do think there's a home for some of these devices for lower heat flux lower power applications and I think sensors and many other lower power highly distributed lower power applications might have a lot of use case for the advances in these devices. So just following up quickly, people in IBM Zurich and so on right there using a lot of the waste energy right for for heating and so on and so forth so depending where you are there's a lot of work in that area. There's some information on that as well. Yeah, that's sorry. Sorry, she should go ahead. Yeah, go ahead. Yeah, I think the way I'm seeing about energy recovery is that the companies are when they put out their data centers at large scale or small scale, their goal is not energy recovery. The goal is to meet their needs, which could be performance or it could be cost or something else. And energy recovery comes as an afterthought. However, that's changing. There are, and I think Europe is really leading in the forefront over here, for example, Amsterdam, which has the most number of data centers in Europe per capita. They are not going to allow any new data centers unless you have a plan of how you utilize the waste heat recovery and really you reuse it in a meaningful way. So I'm seeing that trend. I'm seeing countries which are putting limitations that if your PUE is higher than a certain number, then you know you have to pay more taxes or you have to buy back energy credits. There's a lot more movement in that direction. And hopefully, since this is such an important topic about your ecosystem about nature, about sustainability, that I hope that we do come up with a plan of how to utilize all of the heat that's getting generated. Yeah, and I think that's the point is that we can cool some very high power systems and do it, but it's also just as the raw energy consumption goes up, how are we thinking about just if there's any other ways to utilize it rather than, you know, reject it. And on this I think, you know, Rebecca, it's a good chance to kind of get your perspective here. I know you've thought a lot about, you know, the, whether it's in Europe or corporate policy and how it's evolving towards this sustainability conscious society and maybe is going to start driving some technology development in those areas. Can you kind of just talk about what's going on in that area. Sure. So I would say we're sort of in like the second to third generation of corporate sustainability goals sort of related to energy use and how do you companies sort of procure their electricity sources. Like 10 years ago when solar and wind were still relatively expensive compared to other technologies. There was a lot of effort for companies to buy the equivalent amount of energy that they were using from solar and wind and so basically take a year's worth of electricity usage and say we're going to purchase that solar or wind electricity. But there really wasn't a drive to actually use all of that renewable electricity themselves right so they've introduced more investment into the market to drive technology adoption in that sense but they haven't necessarily transitioned to 100% renewable or zero carbon electricity. And so more recently there's been a push to move from that kind of offset mentality to actually having 100% renewable or 100% zero carbon sources and so that means you have to adopt a lot more energy storage technology. And so to actually be able to use that renewable electricity later in the day or when there's no sun or wind. And so that's been like a shift as the sort of electricity technologies have increased and improved over time. So moving forward I think there's even a look towards how do you even become carbon negative and so I've seen some different companies who've pledged to ultimately have net reductions in CO2 and so investing in some kind of carbon removal technology. It's still pretty early in those stages but and I think a lot of technology companies in general are sort of on the forefront of that kind of thinking and so it's still pretty early more looking at investing in research and development of technologies that have been proven at like the lab scale but haven't necessarily been implemented in the market. Yeah and that you know that that's exciting because I think it is one if you just look at you know there's various statistics and other things out there but the IT industry I think as you mentioned and a few others in transportation is at the leading edge right are the ones who are trying to make that right conversion to right you know at these various stages you talked about. And given the fact that there are on the scale of transportation there is a substantial buying power especially if you're in these local markets like we're in Amsterdam or Amsterdam or other places where it might be outsized relative to other industries. Yeah. Yeah I wanted to make maybe an orthogonal point but related to Purdue education and the students in the out there. I think you've heard talk about life cycle analysis a broad multi part process. We just heard about you know TCO. I think it is important to know that as you try to become champions for new technology in industry you one of the skills that is useful is is to learn how to do the TCO. Almost like a business case end to end because if something was super easy to introduce it would already be used. So there is an element of effort that technology people need to do to make the ends meet and it may be very hard it may be tough ROI maybe 15 years but that is really a very important skill for even heat transfer people. If you want to make your energy recovery technology case you kind of have to figure out how does it make sense business wise with regulations without regulations and then you do have to sell it. Yeah yeah and that's a hard thing I can admit and I don't know if anyone else wants to confess here but it's it's easy to go to thermal management conferences there are other fields and it's often you kind of stop at you know heat flux and you stop at a temperature for technology development. You know we know what immersion cooling we know what even embedded silicon cooling can do but it can't stop there right that's not what's going to get insertion of the systems and that's a very difficult thing to translate across I hope that you know these panel discussions raise awareness of that if nothing else to try and speak outside your domain. Because the success does depend on being aware of actually getting this thing in and it's not a solo activity of course also when you run through the TCO or the case and you might fail you might fail for a couple of reasons markets not big enough or something but when you do that effort and then you watch full that assumption may change in five years or 10 years and then you can come back so I just want to do it's not directly related to the panel but I think it's an important item to be cognizant. Yeah, yeah, no I think it is. I think it is actually directly related or an important aspect of this to think about. Okay, I timing wise you know we've got a few minutes I don't know if you have anything that you know that hasn't been brought up that we should ask. Yeah, there is one that Dr. Gupta kind of briefly alluded to with respect to PUE. So there's been a lot of focus on you know industry getting down the PUE and it's been going down steadily. And the goal is to get it as close to one right so the question is, what needs to be done to create a drastic reduction in the it energy consumption that's coming to keep it close to that value of one and also theoretically like what can we do to keep it as low as possible and what's the theoretical minimum. Yeah, I mean, one comment I would say is PUE is a ratio. So it's a ratio of the total energy or power of the data center to the power of IT. So it does, it does define how efficiently we are reducing the parasitic losses, but changing the denominator is also very important. So ultimately the total electricity bill is what matters, or also matters very important. Yeah, I wanted to add there that we have really been doing a lot of good work and efficiency. And, as I said, when you talk about the improved management and the hyperscale, big change right to the hyperscale systems right versus the traditional significant change almost flip. That is definitely saving on energy. But then you also got to talk about at the at the component level, when you start, you know, putting things together, like this 3D stacking and so on. I think I'm not sure that the, that the data center community, certainly people like Ashish and them who make Intel who make this device are aware of it but I'm not sure that the data center community is ready to see how, how do I cool the heterogeneous integrated system where I have, you know, a microprocessor and memory stacked on top of each other and several of them, or maybe might be using your either to inhabit the Intel. Yes. So, I'm not sure we prepared for that. So the other thing I also want to say from an education point of view, I really encourage universities that talk about heat transfer to have a packaging class that also includes materials, right, and so on and about the ROIs and business cases that my dude is talking about, but I think it's needed. Otherwise, you get left behind you might start saying I'm doing microchannel right, and then you find out oh this daggle microchannel is a plug in nightmare right. You keep saying you're focused on, we all know that the heat transfer coefficient inversely proportional hydraulic diameter but so I'm, it's not about just getting that to zero right and go to infinity, there's a lot of other things like pressure drop and so on. So, how do you understand the system, both from a packaging I talked about this to unhappy the stacking and so on. Are we anticipating that, and then the education, how are we integrating these things as we start teaching, because you can't just teach thermal management course without understanding that. So, yeah, maybe Justin you might say a word or two on that. Yeah, maybe I'll say a word or two on that only because there's a lot of recent activity going on at Purdue in this area I should just, you know, because the students online as well. Harold knows this and is involved but Purdue is part of of a national effort to kind of help improve awareness of packaging, even at the undergraduate level in the curriculum because it, you know electronics are such a ubiquitous system and the packaging of electronics touches so many things that there's people from all fields who don't have an awareness of basically the assert insertion of them from a thermal materials, radiation hardening perspective and various other challenges and so with, you know, Purdue in collaboration with the University of Arizona State Georgia Tech SUNY Bingham 10 has right excellent resources in this area we're actually co developing and co teaching across those universities, a junior level, kind of awareness course for students who are interested in packaging I don't know Carol do you want to say any more on on that but it that it's completely true right we need this, we need education on that whole area, and even the electrical circuit design. One thing that we are trying to do is insert problems, even in the first year of engineering where they're learning how to handle data and use graphing programs and use statistics, we're inserting some of those. Some of the topics related to packaging and even heterogeneous integration radiation hardening into the first semester freshman year, because we don't want to wait until they're juniors to say oh and by the way, look at all of this these issues with respect to So we're working very hard to just introduce this in the core courses and introduction to materials, electrical engineering one on one, first in circuits. It's really a holistic approach to get the students to be aware of these as early as possible, and so that they can build on it that it's not. Okay, you have to take, you have to take thermodynamics and heat mass transfer and all these things and then once you are anointed, then you can take the advanced heat transfer course, or some packaging course. No, they've been exposed to it along the way. It's awesome. Yeah, yeah, that is the goal is to just be very excited about this you can tell. Yeah, it's a really fun thing and that is the goal so I see Arvin's picture, which I guess means you're going to tell us that we're out of time. Fascinating out of time but I can see no lack of energy here. I mean, we could, we could keep going but but I know many of us want to hear derages talk as well so we should. Yeah, I'd like to add, you know, to all the participants here, please note that and including by the way our guests are panelists, a station mother please feel free to join us for derages talk he's giving the distinguished lecture. 15 minutes from now at 430pm and a real thanks to all of the panelists and that the effort supporting the Q&A at the back end of this really exciting panel, truly a grand challenge moving forward for the next few decades so really appreciate the great discussion. And again, it might all of you to join derages talk here. They might have mature at 430pm. Thank you.