Microsoft Research Podcast - Collaborators: Sustainable electronics with Jake Smith and Aniruddh Vashisth

Episode Date: July 11, 2024

Printed circuit boards are abundant—in the stuff we use and in landfills. Researcher Jake Smith and professor Aniruddh Vashisth discuss the development of vitrimer-based PCBs that perform comparably to traditional PCBs but have less environmental impact.Learn more:Recyclable vitrimer-based printed circuit boards for sustainable electronics | Nature Sustainability, April 2024Microsoft Climate Research InitiativeMicrosoft Research AI for ScienceStoring digital data in synthetic DNA with Dr. Karin Strauss | Microsoft Research Podcast, October 2018

From the computation point of view, we always thought that if somebody gave us like 100 different chemistries, we can do a bunch of simulations, tell you like 10 of these actually work. What we've been able to do specifically for Wittremers is that we're able to look at the problem from the other side and we're able to say that if you tell me a particular application, this particular chemistry would work best for you. In essence, what we were thinking of was that if aliens abducted all the chemists from the world, can we actually come up with a framework? If all of this work is successful in 10 years, maybe our materials design process looks completely different, where we've gone from this kind of brute force screening to an approach where you

start with the properties that you care about. they're defined by the application that you have in mind. And we use this need space to define the material that we would like, and we can use machine learning, artificial intelligence, in order to get us to the structure that we need to make in order to actually achieve this design space. You're listening to Collaborators, a Microsoft Research podcast showcasing the range of expertise that goes into transforming

mind-blowing ideas into world-changing technologies. I'm Dr. G in the booth today, IRL, with Dr. Jake Smith, a senior researcher at Microsoft Research and part of the Microsoft Climate Research Initiative, or MCRI. And with him is Dr. Anirudh Vashisht. He's an assistant professor of mechanical engineering at the University of Washington and director of the Vashisht Research Lab. Jake and Anirudh are working on a project that uses machine learning to help scientists design sustainable

polymers with a particularly exciting application in the field of the ubiquitous printed circuit board, or PCB. But before we get all sustainable, let's meet our collaborators. Jake, I'll start with you. You're a self-described chemist with relatively broad interests across applications, and you've done some pretty cool things in your career. Tell us about those interests and where they've led you and how they've contributed to the work you're doing now in MCRI or the Microsoft Climate Research Initiative. Yes, thank you very much for having me. So I started, like most chemists, poking things around in the lab and learning really fundamentally about how atoms interact with one another and how this affects what we do or what we see at our macroscopic level. And so after I left grad school doing this super basic research,

I wanted to do something more applied. And so I did a couple of postdocs, first looking at how we can more effectively modify proteins after we've synthesized them, so they might have a property that we care about. And then later doing similar work on small molecules in a more traditional drug design sense. But after I finished that, I wound up here at Microsoft.

We were very interested in one molecule in particular, one family of molecules, which is DNA. And we wanted to know, how do we make DNA at just gigantic scale so that we can take that DNA and we could store digital data in it? And because DNA has this nice property that it kind of lasts forever,

at least on our human scale, it makes a very nice archival storage medium. So we worked on this project for a while. And at some point, we determined we can kind of watch it blossom and find the next challenge to go work on. The challenge that we wound up at, I'll describe as the Microsoft Climate Research Initiative, the MCRI. We were a group of applied scientists from like natural scientist backgrounds within Microsoft. And we said, how can we make a difference for Microsoft? And the difference that we thought was Microsoft has climate goals. Microsoft wants to be carbon negative. It wants to be water positive and it wants to be zero waste. And in order to make this happen, we need novel materials, which really are a macroscopic view of, once again, atomic behavior. And we said, hey,

we understand atomic behavior. We're interested in this. Yeah, maybe this is something we could help on. And so here we are. We wound up with Anirudh and we'll go into that later, I'm sure. Yeah, yeah. So just quickly back to the DNA thing. Was that another collaboration? I had Karen Stra us on the practicalities of working with DNA once it's synthesized. And how would you do things like retrieve information from a big pool of DNA? Right. People could go back to that podcast because she does unpack that quite a bit. Well, Anirudh, you describe yourself as a trained mechanician who hangs out with chemists, hence your friendship with Jake here. But for your day job, you're a professor and you have your own lab that conducts interdisciplinary research at the intersection, as you say,

of mechanics and material science. So what made you want to move to that neighborhood and what goes on there? Yeah, well, again, thank you so much for having me here and super excited about this. Yeah, just a little bit of background about me. So I started off with my undergrad in civil and mechanics from IIT BHU, did a PhD in mechanics at Penn State and moved to Texas. Go back to the, what's the first one? It's Indian Institute of Technology in India. So that's where I did my undergrad there. And then straight away came to the US to do my PhD in mechanics at Penn State, and then ended up going to Texas, to Texas A&M University and postdoc in a chemical engineering

lab. And that's how I became like super familiar and fond of chemical engineers and chemists. And we moved to Seattle when I got the job at the University of Washington in 2021 with my wife and my daughter. And what we do in our lab is we make and break things now. And we try to see like, you know, when we are making and breaking these things, we try to see them from an experimental and assimilation point of view and try to gain some understanding of the mechanics of these different types of materials. Especially, we are very interested in polymers. I always joke with my students and my class that go about one day without touching a polymer. And I'm always surprised by the smiles or the smirks that I get. But in general, we have been super, super excited and interested about sustainable polymers, making sustainable composites. Particularly, we are very excited and interested in vitremer polymers. So let me just take a step back. I'll probably wear my professor hat straight away.

Yeah, let's do. Let's go. And I'll tell you just like taking a step back, what are the different types of polymers? So in general, you can think of polymers as thermosets or thermoplastics. So to Jake's point, let's just go to the molecular scale there. And you can think of polymers as a bunch of these pasta noodles, which can slide over each other, right? Or these bunch of pasta noodles, which are packed together. So thermosetet as the name suggests it's a set network the pasta noodles are kind of like set in their place thermoplastics is when these pasta noodles can slide over each other so you probably put too much sauce in there yeah so a good analogy there would be a lot of the adhesives that we use are thermosets because they set after

a while. Thermoplastic, we use plastics for 3D printing a lot. So those are thermoplastics. So they are solid. You can heat them up. You can make them flow, print something and they solidify. Multimers are very exciting because just like thermoplastics, they have this flowability associated with them. But more at a molecular scale, like if you think of a single pasta noodle, it can unclick and re-click back again. So it's made up of these small Lego blocks that can unclick and re-click back. Lego pasta. I like that. Exactly. So this unclicking and re-clicking can make them reprocessable, reusable, recyclable, gives them much longer life because you can heal them

and then uh vitrimers basically become the vampires of the polymer universe meaning they don't die well or they have like much longer life sleep every now and then to regenerate yeah um honorude sticking with you for a minute, before we get into the collaboration, let's do a quick level set on what we might call the secret life of circuit boards. For this, I'd like you to channel David Attenborough and narrate this PCB documentary. Where do we find printed circuit boards in their natural habitat? How many species are there? What do they do during the day? How long do they live? And what happens when they die? Okay, so do I have to speak like David?

Yes, I'd appreciate it if you try. No, just be your voice. Yeah, so PCBs, if you think about it, they are everywhere. PCBs are in these laptops that we have in front of us. Probably there are PCBs in these mics, automobiles, medical devices. So PCBs, they're just like everywhere. And depending upon like what is their end applications, they have a composite part of it to where you have like some sort of a stiff inclusion in a polymeric matrix, which is holding this part together and has a bunch of electronics on top of it. And depending on the end application, it might come in different flavors,

something that can sustain much higher temperatures, something which is flexible, things of that sort. And they live as long as we use the material for, like, you know, as long as we're using these laptops or as long as we end up using our cars. And unfortunately, there is a lot of e-waste which is created at the end. There's been a lot of effort in recycling and reusing these materials, but I'm confident we can do more. I think there's close to 50 million metric tons of e-waste, which is generated more than that actually every year. So a lot of scope for us to work there.

So right now, are they sort of uniform, the printed circuit board? I know we're going to talk about Vitramar-based ones, but I mean, other than that, are there already multiple materials used for these PCBs? Jake, you can even address that. Yeah, of course. So there are kind of graded ranks of circuit board materials that, as Anna said, might be for specialty applications where you need higher temperature tolerance than normal, or you need lower noise out of your circuit board. But kind of the bog standard circuit board, the green one that you think

about if you've ever seen a circuit board. This is like a anti-flammability coating on a material called FR4. So FR4, which is an industrial name for a class of polymers that are flame retardant, thus FR. And 4 gives you the general class. This is the circuit board material that we really targeted with this effort. Interesting. So, Jake, let's zoom out for a minute and talk about the big picture and why this is interesting to Microsoft Research. I keep hearing two phrases, sustainable electronics and a circular economy.

So talk about how the one feeds into the other and what an ultimate success story would look like here. Absolutely. So I'll start with the latter. When we set out to start the Microsoft Climate Research Initiative, we started with this vision of a circular economy that would do things that avoid what we, you know, can avoid using. But there are many cases where you can't avoid using something that is non-renewable.

And there, what we really want to do is we want to recapture what we can't avoid using something that is non-renewable. And there, what we really want to do is we want to recapture what we can't avoid. And this project, you know, falls in the latter. And there's a lot of things that fall in the latter case. So, you know, we were looking at this at a very carbon dioxide-centric viewpoint, where CO2 is ultimately the thing that we're thinking about in the circle, although you could draw a circular economy diagram with a lot of things in the circle. But from the CO2 viewpoint, what led us to this project with Anirudh is we thought we need to capture CO2. But once you capture CO2, what do you do with it? You can pump some of it back into the ground, but this is economically non-productive activity. And so it's something we have to do. It's not something we want to do. And so what could we want to do with the CO2 that we've captured? And the thought was we do something economically viable with are great laboratories doing work on this. Oh, interesting.

And then we could look at our plastic design problem and say, hey, we have all this FR4 in the world. How could we replace the FR4, the explicit atoms that are in the FR4 with atoms that have come from CO2 that we pulled out of the air? And so this is the circular economy portion. We come down to the specific problem here. Andrew talked a lot about e-waste. Yeah.

I had great colleagues who also collaborated with us on this project, Biklinduin, Callie Frost, who have been doing work with our product teams here at Microsoft on what can we do to reduce the amount of e-waste that they put out towards Microsoft's climate goals. Right. And Microsoft, as a producer of consumer electronics and a consumer of industrial electronics, has a big e-waste problem itself that we need to actually take research steps in order to

ultimately address. And so what we thought was we have this end-of-life electronic. We can do things like desolder the components. We can recapture those ICs, which have a lot of embedded carbon in them, in the silicon that's actually there. We can take and we can etch out the copper that has been put over this to form the traces, and we can precipitate out that electrochemical way to recapture the copper. But at the end of the day, we're left with this big chunk of plastic plastic and it's got some glass inside of it too for completeness sake and the thought was you know how do we do this you can't recapture this with fr4 fr4 to go back to the spaghetti spaghetti is glued to itself it

doesn't come apart it rips apart if you try and take it apart and so we wanted to say you know what could we do and you know what could we do with Anirudh and his lab in order to get at this problem and to get us at a FR4 replacement that we could actually reach this complete circularity with? Interesting. Well, Jake, that's an absolutely perfect segue into how I met your mother, which is, you know, how you all started working together, who thought of who first and so on. I'm always interested to hear both sides of the meetup. So Anirudh, why don't you take the baton from Jake right there and talk about from your perspective, how you saw this coming together, who approached who, what happened? And then Jake can confirm or deny the story. Yeah, yeah. So it actually started off, I have a fantastic colleague and a very good

friend in CS department, Professor Vikram Iyer. And he actually introduced me to Biklan Nuang from Microsoft. And we got a coffee together. And we were talking about wit rumors, like the work that we do in our lab. And I had this one schematic, I forget if it was on my phone, or I was carrying around one paper in my pocket and I showed them, I was like, you know, if we can actually do a bunch of simulations, guide an ML model, we can create, for lack of a better word, like a chat GPT type of model where instead of telling like, this is the chemistry, tell me what the properties are, we can go from the other side, you can ask the model, hey, I want a bit of chemistry, which is the chemistry, tell me what the properties are. We can go from the other side.

You can ask the model, hey, I want a bitumen chemistry, which is recyclable, reprocessable, that I can make airplanes out of, or I can make glasses out of. Tell me what that chemistry would look like. And I think, you know, Biklin was excited about this idea, and she connected me with Jake. And I think I've been enjoying this collaboration for the last couple of years working on that. Was there a paper that started the talk

or was that just this napkin drawing? I think to give myself a little bit of credit there, I think there was a paper with the nice drawing on it. Yeah, there was a white paper, yeah. That's good. Well, Jake, what's your side of this story? This is awesome. We got the first half that I didn't know.

Filling in gaps. This is the Bitcoin-mediated half. I was sharing an office with Bitcoin, who apparently came up from this meeting, and I saw the mythical paper. I put this on my desk. And I'll plug another MCRI project that we were working on there, or at the time, where we were attempting to do reverse design or inverse design of metal-organic frameworks, which are these really interesting molecules that have the possibility to actually serve as carbon capture sorbents. Oh, wow. the possibility to actually serve as carbon capture sorbents. But the approach there was to use machine learning to help us sample this giant space of metal organic frameworks and find ones

that had the property that we cared about. I mean, you draw this diagram that's much like Andrew just described, where you've got this model that you train and out the other side comes what you want. And so this paper came down to my desk and I looked at it and I said, hey, that's what we're doing. It kind of, you know, went from there. We had a chat. We determined, hey, we're both interested in, you know, this general approach to getting to novel materials. And then, you know, we've already talked about the synergy between our interests and Microsoft's interests and the, you know, great work or the great particular applications that are possible with the type of polymer

work that Anirudh does. Yeah. So the University of Washington and Microsoft meet again. Well, Jake, let's do another zoom out question because I know there's more than just the Microsoft Climate Research Initiative. This project is a perfect example of another broader initiative within Microsoft, which has the potential to, quote, accelerate and enhance current research, and that's AI for Science. So talk about the vision behind AI for Science, and then if you have any success stories, maybe including this one, tell us how it's working out.

Yeah, absolutely. We are, and by we, I mean myself and my immediate colleagues, are certainly not the only ones interested in applying AI to scientific discovery at Microsoft. And it turned out a year or two after we started this collaboration, a bigger organization named AI for Science arose and we became part of it. who along with our kind of sister organization in research called Health Futures who work more on the biology side are interested in how AI can help us do science in a faster way but be maybe a smarter better use of resources way or the ultimate goal or the ultimate dream is see a way that we just can't think of doing right now a way that know, it just is fundamentally incompatible with the way that research has historically been done in, you know, small groups of grad students directed by a professor who are themselves, you know, the actual engine behind the work that happens. And so the AI for Science Vision, you know, it's got a couple of parts that really map very well

into this project. The first part is we want to be able to simulate bigger systems. We want to be able to run simulations for longer, and we want to be able to do simulations at higher accuracy. When we get into the details of the particulars of the VITRAMER project, you'll see that a ton of time trying to identify the appropriate simulation parameters in order to capture the behavior that we care about here. And so the first day after Science Vision says, we don't need Yuen to do that. We're going to have a drop-in solution or we're going to have a set of drop-in solutions that can take this work away from you and make it much easier for you to go straight to running the simulations that you care about. Yeah, there's a couple questions not on the list here, but you prompted them, no pun intended. Are these specialized models with the kinds of information? I mean, if I go to ChadGBT and ask it to do what you guys are doing, I'm not going to get the same return, am I? Absolutely.

Am I? Oh, no, no, no, no. I was saying you are absolutely correct. You can ask chat GPT and it will tell you all sorts of things that are very interesting. It can tell you probably a vitremer. It could give you an inerud spiel about the spaghetti, I'm sure, if you prompt it in the correct way. But what it can't tell you is, you know, hey, I have this particular vitremer composition, and I would like to know at what temperature it's going to melt when I heat it up. Okay, so I have one more question. You talk about the simulations.

Those take a lot of compute, am I right? You are absolutely right. Am I right? Yeah. So is that something that Microsoft brings to the party in terms of, I mean, does the University of Washington have the same access to that compute or what's the deal? I think especially on the scale, we were super happy and excited that we were collaborating with Microsoft. I think one of these simulations took like close to a couple of weeks and we ended up doing doing, I would say, close to more than 30,000 simulations. So that's a lot of compute time, if you think about it.

To put that in perspective, how long would it take a human to do those simulations? Oh, man, to try and actually go do all this in the lab. Right. First, you've got to make these 30,000 like starting materials. This in itself. Let's say you could buy those. Then to actually run the experiments, how long does it take? How much money?

That's like you're talking about like one PhD student there. It's like, you know, it takes like a couple of years just to synthesize something properly and then characterize it. Yeah. Yeah. No, I think the virtual world does have some pluses to it. This is a really good argument for AI for science, meaning the things that it can do,

artificial intelligence can do, at a scale that's much smaller than what it would take a human to do. Yeah, absolutely. And I'll plug the other big benefit now, which is, hey, we can run simulations. This is fantastic. But the other thing that I think all of us

really hope AI can do is it can help us determine which simulations to run. So we need less compute overall. We need less experiments if we have to go do the experiments. So it's a winnowing process. Exactly. Okay.

That's actually really interesting. And this is like the second or maybe even the largest vector for acceleration that we could see. Wow. Okay. That's actually really interesting. Man, this is like the second or maybe even the largest vector for acceleration that we could see. Wow. Cool. Well, every show I ask, what could possibly go wrong if you got everything right? And Anirudh, I want to call this the defense against the dark arts question for you.

You're using generative AI to propose what you call novel chemistries, which can sound really cool or really scary, depending on how you look at it. But you can't just take advice from a chat bot and apply it directly to aerospace. You have to kind of go through some processes before. So what role do people, particularly experts in other disciplines, play here? And what other things do you need to be mindful of to ensure the outputs you get from this research are valid? Yeah, yeah, that's a fantastic question. And I'll actually piggyback on what Jake just said here about Yvonne Zhang, who's like a fantastic graduate student that we have in our lab. He figured out how to run these simulations at the first point.

It was like six months of like really long audio, how to make these simulations at the first point. It was like six months of really long audio. How to make sure that in the virtual world, we are synthesizing these polymers correctly and we are testing them correctly. So that human touch is essential. I feel like at every step of this research, not just like doing virtual characterization or virtual synthesis of these

these materials training the models but eventually when you train the models also and the model tells you that well these are like the 10 best polymers that would work out there you need people like jake who are like chemists you know they come in and they are like hey you know what like out of these 10 chemistries this one you can actually synthesize. It's a one-step reaction or things of that sort. So we have a chemist in our lab, also Dr. Agni Biswal, who's a postdoc. So we actually show him all these chemistries apart from Jake and Viklin. We show the chemistries to all the chemists and say like, okay, what do you think about

this? How do these look? Like, are they totally insane or can we actually make them? Yeah, we still need that like human evaluation step at the end at this point. Yeah, exactly. Ask a chemist. Well, and I would imagine it would be further than just this would be the best one or something like you better not do that one. Are there ever like crazy responses or replies from the model? No, it's fascinating. Models are very good. And particularly, we'll talk about models that generate small organic structures now, generating things that look reasonable. They follow all the

rules, but there's this next step beyond that. And you see this when you talk to people who've worked in MedChem for 30 years of their life well they'll look at a structure and they'll like get this gut feeling like you know a storm is coming in and their knee hurts and they really don't like that molecule um and if you push them a little bit you know sometimes they can figure out why they'll be like oh i worked on you know a molecule that looked like that 20 years ago and it you know turned out to have this toxicity and so i don't want to touch that again but oftentimes people can't even tell you. They just got this instinct that they've built up. And trying to capture that intuition is a really interesting next frontier for this sort of research. Wow. You know, you guys are just making my brain fry because it's like so many other

questions I want to ask, but we're actually getting there to some of them. And I'm hoping we'll address those questions with the other things I have. So Jake, I want to come, well, first of all, Anirudh, have you finished your defense against the dark arts? I think I can point out one more thing really quickly there. And as Jake said, like, we are learning a lot, particularly about these materials, like the bitumen materials. These are new chemistries, and we are still learning about the mechanical, thermal, rheological properties, how to handle these materials. So I think there's a lot that we don't know right now. So it's like a bunch of unknowns that are there. Well, and that's research, right? The unknown unknowns.

Jake, I want to come back to the vision of the Climate Research Initiative for a minute. One goal is to develop technologies that reduce the raw tonnage of e-waste, obviously. But if we're honest, advances in technology have almost encouraged us to throw stuff away. It's like before it even wears out. And I think we talked earlier about, you know, this will last as long as my car lasts or whatever, but I don't like my car in five years. I want a different one, right? So I wonder if you've given any thought to what things, in addition to the work on reusable and recyclable components, we might do to reverse engineer the larger throwaway culture. This is interesting. I feel like this gets into real questions about social psychology and our own behaviors with individual things. Why do I have this can of carbonated water here when I could have a glass of carbonated water? But I want to kind of completely sidestep that because—

Well, we know why, because it's convenient and you can take it in your car and not spill. Agreed. Yes. I also have this cop and it could not spill as well. Right, right. True. Recycle. Reusable. No, no, no. This is like a... It's an ingrained consumer behavior that I've developed. I'll slip into Jake's personal perspectives here, which is that it should not be on the individual consumer behavior changes to ultimately drive a shift towards reusable and recyclable things. And so one of the fundamental hypotheses that we had with the design of the projects we

put together with the MCRI was that if we put appropriate economic incentives in place, then we can naturally guide behavior at a much bigger scale than the individual consumer. And maybe we'll see that trickle down to the consumer, or maybe this means that the actual actors, the large-scale actors, then have the economic incentive to follow up themselves. So with the e-waste question in particular, we talked a lot about FR4 and, you know, it's the part of the circuit board that you're left over with at the end that there's just nothing to do with. Right.

And so you toss it in a landfill, you burn it, you do something like this. But you know, with a project like this where our goal was to take that material and now make it reusable, we can add this actual economic value to the waste there. Yeah. I realized even as I asked that question that I had the answer embedded in the question, because in part, how we design technologies drives how people use things.

Absolutely. And usually the drivers are convenience and economics. So if upstream of consumer consumption, how's that? Upstream of that, the design drives environmental health and so on. That's actually, that's up to you guys. So let's get out of this booth and get back to work. Well, Jake, to that point, talk about the economics. We talk about a circular economy, and I know that recycling is expensive. Can you talk a little bit about how that could be impacted by work that you guys do? Recycling absolutely is expensive relative to landfilling or a smaller alternative. One of the things that makes us target e-waste is that there

are things of value in e-waste that are like innately valuable. When you go recollect that copper or the gold that you've put into this, when you recollect the integrated circuits, they have value. And so a lot of the economic drive is already there to get you to the point where you have these circuit boards. And then the question was, how do we get that next bit of economic value so that you've taken steps this far, you have this pile of circuit boards, so you've already been incentivized to get to here, and it will be easy to make this, even if it's not a completely economically productive material versus synthesizing a circuit board from virgin plastic. But it's offset enough. We've taken enough of that penalty for reuse out that it can be justifiable to go to. Okay, so talk again off script a little bit, but talk a little bit about how Vitramers help take it to the last mile. Yeah, I think the inherent property of the polymer to kind of like unclick and reclick back again, the healability of the polymer, that's something that kind of drives this reusability and reprocessability of the material.

I'll just like point out like, you know, particularly to the PCB case where we recently published a collaborative paper where we showed that we can actually make PCB boards using wet tremors. We can unassemble everything. We can take out the electronics and even the composite, the glass fiber and the polymer composite. We can actually separate that as well, which is, in my mind, like a pretty big success. And then we can actually put everything back together and remake a PCB board and, you know, keep on doing that. Okay, so you had talked to me before about ring around the rosy and the hands and the feet. Do that one just for our audience because it's good. Okay, so I'll talk a little bit about thermoset thermoplastic again, then I'll just give you a much broader perspective there.

So the FR4 PCBs that are made, they are usually made with thermosetting polymers. So if you think about thermosetting polymers, just think of kids playing Ring of Roses, right? Like their hands are fixed and their feet are fixed. Once the network is formed, there's no way you can actually destroy that network. The nice thing about Vitramers is that when you provide an external stimulus, like just think about these kids playing Ring of Roses again, their feet can move and their handshakes can change, but the number of handshakes remain the same. So the polymer is kind of like unclicking and re-clicking back again. And if you can cleverly use this mechanism, you can actually

recycle, reprocess the polymer itself. But what we showed, particularly for the PCB paper, was that you can actually separate all the other constituents that are associated with this composite. I love that. Well, sticking with you for a second, Anirudh, talking about mechanical reality, not just chemical reality, but mechanical reality, even the best composites wear out from wear and tear. Talk about the goal of this work on novel polymers from an engineering perspective. How do you think about designing for reality in this way? Yeah, yeah. That's a fantastic question. So we were really motivated by what type of mechanical or thermal loadings materials see in day-to-day life. I sit in my car, I drive it, it drives over the road, there is some fatigue loadings, there's dynamic loading,

and that dynamic loading actually leads to some mechanical flaws in the material, which damages it. And the thought was always that, can we restrict that flaw or can we go a step further? Can we actually reverse that damage in these composites? And that's where, you know, that unclicking, re-clicking behavior of Vitramar becomes like really powerful. So actually the first work that we did on these type of materials was that we took a Vitramar composite and we applied fatigue loading on it, cyclic loading on it, mechanical loading. And then we saw that when there was enough damage accumulated in the system, we healed the system. And then we did this again, and we were able to do it again and again until I was like, I've spent too much money on this test frame. But it was really exciting because for a particular loading case that we were looking at, traditional composites were able to sustain that for 10,000 cycles.

But for Vitramers, if we did periodic healing in the material, we were able to go up to a million cycles. So I think that's really powerful. Orders of magnitude. Yeah, exactly. Wow. Jake, I want to broaden the conversation right now beyond just you and Anirudh and talk about the larger teams you need to assemble to ensure success of projects like this. Do you have any stories you could share about how you go about building a team? You kind of alluded to it at the beginning. There's sort of a pickup basketball metaphor there. Hey, he's doing that, we're doing this. But you have some intentionality about people you bring in. So what strengths do each institution bring and how do you build a team? Yeah, absolutely. We've tried a bunch of these collaborations and we've definitely got some warnings about which ones work better than others. This has been a super productive one. I think it's because it

has that right mix of skills and the right mix of things that each side are bringing. So what we want from a Microsoft side for a successful collaboration is we want a collaborator who is really a domain expert in something that we don't necessarily understand, but who can tell us in great detail. These are the actual design criteria. These are where I run into trouble with my traditional research. This is the area that, you know, I'd like to do faster, but I don't necessarily know how. And this is the critical part, I think, you know, from the get-go. They need to themselves be an extremely, you know, capable subject matter expert. Otherwise, we're just kind of chatting.

We don't have anyone that really knows what the problem truly is, and you make no progress. Or worse, you spend a whole lot of resources to make progress. I'm doing air quotes. I love air quotes on a podcast. That is actually just completely tangential to what the field needs or what the actual device needs. So this is, you know, the fundamental ingredient. And then on top of that, we need to find a problem that's of joint interest where in particular, computation can help. You talked about the amount of computation that we have at our disposal as researchers at Microsoft,

which is a tremendous strength. And so we wanna be able to leverage that. And so for a collaboration like this, where running a large number of simulations was a fundamental ingredient to doing it, this was, you know, a really good fit that we could come in and we could enable them to have more data to train the models that we build together. Well, as researchers, are you each kind of always scanning the horizon for who else is doing things in your field that, or tangential to your field, but necessary?

How does that work for recruiting, I would say? Yeah, that's a good question. I think, I mean, that's kind of like the job, right? For the machine learning work we did, we saw a lot of inspiration from biology where people have been designing biomolecules. The challenges are different for us. Like we are designing much larger chains,

but we saw some inspiration from there. So always like looking out for like who is doing what is super helpful and leads to like really nice collaborations as well. We've had like really fruitful collaborations with Professor Sid Kumar at UDelft and we always get his wisdom on some of these things as well.

But yeah, recruiting students also becomes like very interesting. They're not like people who can help us achieve our idea. Yeah. Jake, what's your take on it from the other seat? I mean, do you look actively at universities around the world and even in your backyard to like you, Doug. My perspective on like how collaborations come into be is they're really serendipitous. You know, we talked about how this one came into be and it was because we all happen to know Vikram

and Vikram happened to connect Vikram with Andrew and it kind of rolled from there. But you can have serendipitous, you know, meetings at a conference where you happen to, you know, sit next to someone in a talk and you both share the same perspective on, you know, meetings at a conference where you happen to, you know, sit next to someone in a talk and you both share the same perspective on, you know, how a research problem should be tackled. And something could come out of that.

Or in some cases you go actually shopping for a collaborator and, you know, you need to talk to 10 people to find the one that has that same research perspective as you. Second, Anirudh's observation that you get a very different perspective if you go find someone who they may have the same perspective on how research should be tackled, but they have a different perspective on what the ultimate output of that research would be. They can often point you in areas where your research could be helpful that you can't necessarily see because you lack the domain knowledge or you lack that particular angle on it. Which is another interesting thing in my mind is the role that papers, published papers play. That's a lot of Ps in a sentence, alliteration, that you would be reading or hearing about either in a lightning talk or a presentation at a conference.

Does that broaden your perspective as well? And how do you, like, do you call people up? I read your paper. I have cold email people. You know, this works sometimes. Sometimes this is just the introduction that you need. But the interesting thing in my mind is how much the computer science conferences and things like ChemArchive and Archive have really replaced,

for me, the traditional chemistry literature or the traditional publishing literature, where you can have a conversation with this person while they're still actively doing the work because they put their initial draft up there and it's still need for vision. And there's opportunities even earlier on in the research process than we've had in the past. Huh. And to your earlier point, I'm envisioning an Amazon shopping cart for research collaborators. Oh, he looks good. Into my cart. Anirudh, I always like to know where a project is on the spectrum from what I call lab to life. And I know there are different development stages when it comes to technology finding its way into production and then into broader use. So to use another analogy I like, pretend this is a relay race and research is the first leg. Who else has to run and who brings it across the line? Yeah, yeah. So I think the initial work that we have done, I think it's been super fruitful.

And to Jake's point, like converging to like a nice output took a bunch of chemists, mechanical engineers, simulation folks, machine learning scientists to get where we are. And as Jake mentioned, we've actually put some of our publications on archive and it's getting traction now so we've had some excitement from startups and companies which make polymers asking us oh can you actually can we get a slice of this this framework that you're developing for designing word trimmers which is very promising. So we have done very fundamental work, but now what's called the valley of death in research, like taking

it from lab to production scale, it's usually a very tightly knit collaboration between industry labs and sometimes national labs too. So we're excited that actually a couple of national labs have been interested in the work that we have been doing. So super optimistic about it. So would you say that the Vitramir-based printed circuit board is a couple of times. And in collaboration with his lab, we did actually make a prototype circuit board. We showed that it works as you expect. We showed that it can be disassembled, it can be put back together, and it still works as expected.

The break stuff makes stuff. I think to the spirit of the question, it's still individual kind of one-off experiments being run in a lab. And Anirudh is right. There's a long way to go from like a technology readiness level three where we're doing it ourselves on bench scale up to, you know, the seven, eight, nine, where it's actually commercially viable when someone has been able to reproduce this at scale. So that's when you bring investors in or labs that can make stuff in and scale. Yeah. Yeah. I think, I think once you're like close to seven, I think that's where you, you're pretty much ready for the big show. So where are you now? Two? I would say like two to three. Three, somewhere in that range. Yeah. Okay. The skills kind of differ depending on which agency you see you put it on. So Jake, before we close, I want to talk briefly about other applications of recyclable vitrimer-based polymers.

In light of their importance to the Climate Research Initiative and AI for Science, so what other industries have polymer components that have nowhere to go after they die but the landfill? And will this research transfer across to those industries? An excellent question. but the landfill and will this research transfer across to those industries? An excellent question. So my personal view on this is that there's a couple of classes of polymers. There's these very high value application uses of polymers where we're talking about the printed circuit boards. We're talking about aerospace composite.

We're talking about the panels on your car. We're talking about things like wind turbines where there's a long life cycle. You have this device that's going to be in use for five years, 50 years. And at the end of that, the polymer itself is still probably pretty good. You could still use it and regenerate it. And so Anirudh's lab has done great work showing that you can take things like the side panel of a plane and actually disassemble this thing, heal it keep it in use longer and use it at the end of its lifetime. There's this other class of polymers, which I think are the ones that most people think about, your Coke bottle. And VITAMIR seemed like a much harder sell there. I think this

is more the domain of biodegradable polymers in the long run to really tackle the issues there. But I'm very excited in this high-value polymer, this long-lifetime polymer, this permanent install polymer, however you want to think about it, for work like this to have an impact. Yeah. From your lab's perspective, Anirudh,

where do you see other applications with great promise? Yeah. So as Jake said, places where we need high-performance polymers is where we can go. So PCBs is one, aerospace and automotive industry is one, and maybe medical industry is another one where we can actually, if you can make prosthetics out of vitrimers, prosthetics actually lose a little bit of their stiffness, you know, as you use them. And that's because of localized damage. It's the fatigue cycle, right? So what if you can actually heal your prosthetics and reuse them? So yeah, I feel like, you know, there's so many different applications, so many different routes that we can go down. Yeah. Well, I like to end our collaborators shows with a little vision casting. And I feel like this whole podcast is that. I should also say, you know, back in the 50s, there was the big push to make plastics. Your word is Vitramers.

So let's do a little vision casting for Vitramer-based polymers. Assuming your research is wildly successful and becomes a truly game-changing technology? What does the future look like? I mean, specified future, not general future. And how has your work disrupted this field and made the world a better place? I'll let you each have the last word. Who'd like to go first? Sure, I can go first. I'll try to make sure that I break it up into computation and experiments, so that once I go back, my lab does not pause on me. Yeah, so I think from the computation point of view, we always thought that if somebody gave us like 100 different chemistries, we can actually bottle it down to like, we can do a bunch of

simulations, tell you like 10 of these actually work. What we've been able to do specifically for Vitramers is that we're able to look at the problem from the other side. And we're able to say that if you tell me a particular application, this particular chemistry would work best for you. In essence, what we were thinking of is that if aliens abducted all the chemists from the world, can we actually come up with a framework? So I think it'll be difficult to get there because as I said earlier that, you know, you need that human touch. But I think we are happy that we are getting there. And I think what remains to be seen

now is like, you know, now that we have this type of framework, like what are the next challenges? Like we're going from the lab to the large scale, like what challenges are associated there? And I think similarly for the experimental side of things also, we know a lot, we have developed frameworks, but there's a lot of work that still needs to be done

in understanding and translating these technologies to real life applications. I like that you're kind of hedging your bets there saying, I'm not going to paint a picture of the perfect world because my lab is going to be responsible for delivering it. Jake, assuming you haven't been abducted by aliens, what's your take on this? I view kind of the goal of this work and the ideal impact of this work as an acceleration of getting us to these polymers being deployed and all these other applications that we've talked about. And we can go broader than this. I think that there's a lot of work, both within the MCRI, within Microsoft and outside of

Microsoft and the bigger field, focused on acceleration towards a specific goal. And if all of this work is successful in 10 years, maybe our materials design process looks completely different where we've gone from this kind of brute force screening that Anirudh has talked about to an approach where you start with the properties that you care about. They're defined by the application that you have in mind. You want to make your Vitamir PCB. It needs to have a specific temperature where it becomes gummy. It needs to have a specific resistance to burning. It needs to be able to effectively serve as the dielectric for your bigger circuit. And we

use this like need space to define the material that we would like. And we can use machine learning, artificial intelligence in order to get us to the structure that we need to make in order to actually achieve this design space. And so this was, you know, our big bet within AFRA Science. This is the big bet of this project. And with this project, you know, we take one step towards showing that you can do this in one case and the future casting would be we can do this in every materials design case that you can think about. Hmm. You know, I'm thinking of Lane's track analogy again, but you know, you've got mechanical engineering, you've got chemistry, and you've got artificial intelligence, and each of those sciences is advancing, and they're using each other to sort of help advance in various ways.

So this is an exciting, exciting project and collaboration. Jake, Anirudh, thanks for joining us today on Collaborators. This has been really fun for me. So thanks for coming in and sharing your stories today. Thank you so much. Of course. Thank you.