Inside a collaboration between Microsoft and the Pacific Northwest National Laboratory
The Pacific Northwest National Laboratory (PNNL) in Richland, Washington, has turned to Microsoft for high-performance computing requirements. In what it calls ...
The Pacific Northwest National Laboratory (PNNL) in Richland, Washington, has turned to Microsoft for high-performance computing requirements. In what it calls a multi-year collaboration, the lab and the software giant will apply artificial intelligence to speed up research in clean energy. For details, the Federal Drive with Tom Temin spoke with PNNL Associate Director and Chief Digital Officer, Brian Abrahamson.
Interview Transcript
Tom Temin Well, give us the domain in which this collaboration is occurring. I mean, its energy, but what about energy?
Brian Abrahamson Yeah. So right now, we think there’s just tremendous potential in the use of AI for science. And the collaboration right now is really focused on advancing and accelerating scientific discovery in the field of chemistry and materials science. And those are two scientific fields that are really key to the discovery of new materials for batteries and energy storage and such.
Tom Temin Yeah, batteries are a big problem because it’s sort of a technology that moves along analog and we need step function, increase in battery capability to really realize these dreams of nirvana. Fair to say.
Brian Abrahamson That’s exactly right. You know, I think the discovery of new battery materials and the chemistry needed for those is part of our initial focus on this collaboration with Microsoft. And part of that is reducing dependency on elements like lithium, which can be difficult to obtain, especially in the quantities that the world needs it. When you think about the use of grid scale batteries to incorporate renewable energies into the power grid and other such use cases.
Tom Temin And there’s also the fact that batteries, as we know them now, eventually wear out in a way that’s not all that economic.
Brian Abrahamson Correct. Yes, they have a limited lifespan. And as we think about new battery chemistries that can improve both the density of what can be stored, the lifespan of those batteries, the accessibility of the materials used to produce the batteries, that’s all important.
Tom Temin All right. And now Microsoft is joining you. Let’s talk about the arrangement first. Do you have a CRADA with them. Do you have a contract. How is this structured.
Brian Abrahamson You know at the highest level what we’ve done is we’ve signed an MOU, a memorandum of understanding with Microsoft. It is for a multi-year collaboration where we are really starting with the exploration of using AI to accelerate discovery in chemistry, materials science. But we intend to move beyond that into other fields. And we have from a structure perspective, we then have a diverse set of projects underneath that that are advancing those goals. So as an example, we have certain engagements with the Department of Energy that we’re bringing this collaboration to bear to advance some of those outcomes. Microsoft has some internal investments that they’re making, and we as a national laboratory have some internal investments that we’re making to advance the goals. So there really is a portfolio of projects, if you will, underneath this broader multi-year MOU.
Tom Temin And I know what PNNL brings to this. You’ve got lots of scientists and people that know material science and know energy science and so forth. Microsoft writes code. So, what are they bringing to it? Brian Abrahamson You know, several things. When you think about the convergence of high-performance computing, advanced artificial intelligence models in the cloud, right. There is really a very beautiful convergence that’s happening there that Microsoft brings, you know, in case of chemistry, material science right there, delivering that through the Azure Quantum Elements platform, which is something that we’re working alongside of them to kind of prove out and to leverage, you know, the Azure cloud, some of the advanced AI models and some of the high performance compute necessary to run those AI models effectively, you know, all or part of the equation, and coupling that with some of the scientific expertise that irrespective of the compute right, that scientific expertise still remains essential to the development of these new materials and processes. And some of the AI models are specifically trained on science. And so, you know, you think about generic models, large language models, ChatGPT, etc. trained on lots of content on the internet. Right. But we’re bringing the better than this partnership. Also models that are trained specifically on scientific content in order to help advance.
Tom Temin We’re speaking with Brian Abrahamsen. He’s the chief digital officer and associate director of the Pacific Northwest National Laboratory. Give us an example. I mean, if you’re looking for a new storage, there’s anodes and diodes. And what’s the other end of the battery besides the anode and the cathode? Yeah, I was thinking of cathode. And is it a matter of like modeling molecules, for example, and behaviors that you can do with a supercomputer? Give us an example of how this might work.
Brian Abrahamson So traditionally there’s a lot of trial and error, kind of an Edisonian approach to the discovery of some of these new materials. And to the extent that we can transition some of that upfront work to more computational simulation, leveraging artificial intelligence to help narrow down the playing field of kind of the universe of elements and structures that can be explored. And, you know, in this case, where we went from 30 something million potential elements and structures to be evaluated, the artificial intelligence helped narrow that playing field significantly to several dozen that we could then use some of our human expertise to then evaluate and then start to synthesize candidate materials in laboratory. So, it literally is just an acceleration of what otherwise is a laborious process.
Tom Temin Yes. Because if you had, say, 30 million potential structures, and I’m presuming you mean molecular structures at this level of material research is beyond how far does this piece of steel bend? Then there’s no way that humans could test 30 million of them. So possibly nothing would happen if you couldn’t narrow it down.
Brian Abrahamson Correct. You’re talking decades and decades. And so, this is where that simulation — and simulation and leveraging computation is nothing new in this — the addition of some of these more advanced artificial intelligence models that can help to improve some of that upfront simulation and computational work.
Tom Temin And is it also a vendor that has the computational capability in its own cloud versus the traditional way of national labs and other federal elements building bespoke supercomputers?
Brian Abrahamson Yeah, you know, we think of that as it’s not an either or. It’s and we think the role of high-performance computing in its traditional capacities, we think of some of the leadership class computing facilities of the day. We play an incredibly important role in scientific discovery. In addition to that, right? We think some of the, you know, the hyperscalers people like Microsoft with, you know, significant horsepower, also have a lot to offer and bring in some of those models and compute to there. So really, it’s an end for us.
Tom Temin And for those at the lab that absolutely sure, they have a structure they want to work on and develop. You might have unlimited demand, but not even this resource is not unlimited. So how you parcel out who gets to do what?
Brian Abrahamson You know, that’s a big part of what this multiyear collaboration is about, is one, it’s testing out and creating proof points of progress with some of these new AI models, you know, running in the cloud. But secondly, it’s also about improving the accessibility of those models in those computing environments to the scientific community and finding ways to do that. You know, that’s what’s one thing that I think a lot of scientists talk with is sometimes with a lot of the computing that’s available, there are queues. You’re waiting. There’s a lot of demand for those resources. And for extent, we can use the cloud to help create additional capacity for scientific research. We think there’s a significant benefit to that.
Tom Temin And will Microsoft have people on hand, for example, to say, well, here’s how we can run this model, or this is the AI model that probably would give you the best answer or something like that.
Brian Abrahamson Absolutely. Microsoft is bringing some deep expertise in developing some of the AI and machine learning models that have been trained on scientific literature and content, and then surfacing those through the Azure cloud. And then we’re applying those to some of our domain specific challenges, right, in chemistry, material science that we have as a national laboratory in terms of our goals and our mission around, you know, a clean energy future.
Tom Temin And is Microsoft getting money for this? Are they getting the intellectual property? I mean, there’s an exchange of funds. Fair to say?
Brian Abrahamson Yeah, I think for both organizations. And, you know, clearly Microsoft has commercial interests in licensing technology to help do this work. You know, this collaboration is really about applying some of those technologies to our mission and iterating on that together as we move it forward. And to the extent we can accelerate kind of our mission outcomes and help create a tool and a capability that can be used by the broader scientific community, we think there’s a win there.
Tom Temin So ultimately this could feed your technology transfer process.
Brian Abrahamson Absolutely. And, you know, to the extent the discoveries are made that could be scaled up by industry and whatnot, that’s a big part of what the national laboratories do. You know, we don’t always bring things to industry scale, right. We do that technology transfer to let others do that.
Tom Temin And by the way, is one area getting back to the research itself, batteries that are power dense but not weight dense, because from what I’ve seen, electric cars and these pickup trucks coming them, some of them would collapse your garage floor and all they do is use their battery up to move their batteries around, basically what they’re doing.
Brian Abrahamson Absolutely. When you think about the density of materials, I mean, there’s a lot of weight involved. And to the extent that there’s opportunities to discover materials that have different characteristics that can improve upon that, right. You know, I think the future will tell us. But, you know, all of those things are certainly in play.
Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.
Tom Temin
Tom Temin is host of the Federal Drive and has been providing insight on federal technology and management issues for more than 30 years.
Follow @tteminWFED
