Achievable Big Stuff: IBM’s 5 in 5

Interview with IBM Research VP Jeff Wesler

EE Times covers the business of technology. Any given story might be a little more technology than business, or vice versa, but – yeah. Business is about the prospects for buying or selling. When it comes to semiconductors and other advanced electronics technologies, that usually means looking ahead to whatever products are being prepared for the next holiday gift-giving season, or the next upgrade or replacement of something that’s already out there. What I’m getting at here is that typically the time scale that most businesses look ahead is months. Sure there are exceptions —  automotive electronics, for example — but even in automotive, some OEMs want to eventually get down to months.

Anyway, the thing is: there are some technological challenges that take years to tackle, and can’t be tackled by any single organization alone. And there aren’t too many companies that can take the time to examine those kinds of challenges. IBM is one of them. The company’s IBM Research operation is renowned for being one of the few organizations on the planet with the wherewithal to think about technological roadmaps that sometimes look as far as 50 years ahead.

Every year for the last 14 years, IBM Research has designated five grand challenges that it believes technologists using the most advanced tools might be able to solve in as little as five years. IBM calls it Five in Five.

Jeff Welser is the vice president of IBM Research, Exploratory Science and University Partnerships. We gave him a ring to learn about this year’s Five in Five.

JEFF WELSER: Our whole theme this year for the Five in Five is about how we can supercharge the discovery of materials — new materials with new functionality — to enable a more sustainable future. So that was a theme we realized was particularly important given some of the challenges we’re seeing around climate change, COVID-19 pandemic, a lot of the sustainability issues we see in the world today.

So the five we focused in on was finding materials to help us capture and transform CO², harmful emissions, to mitigate climate change, for example. Or modeling the way nature actually produces fertilizers and nitrates so that we can actually do it in a way to grow food more sustainably while reducing carbon emissions and energy consumption.

Number three was, Can we rethink batteries themselves? Which are playing an increasing role in our renewable energy infrastructure, and also in electric cars. We think that not just for getting better performance in power, which we always do, but to make them more sustainable and less impactful on the environment.

The number four is really more generally for manufacturing of a lot of our high tech products, particularly chip infrastructure. Can we find ways to make those materials more sustainable?

And lastly, as we think about actually attacking the COVID-19 pandemic, we need to find new medications. Whether it’s vaccines or therapeutics or other medicines. Obviously that’s a very long process, ten years, billions of dollars usually. Could we find ways to actually combine AI and other technologies to look at reusing existing medications that maybe are already proven safe that could actually have an impact on COVID-19 as well.

BRIAN SANTO: I noticed that almost everything on the list did involve material science, often just basically dealing with elements. It’s chemistry, physics, not necessarily technology per se. I’m wondering, where does the technology come in? And I’m assuming the answer might be with creating the solutions for the problems.

JEFF WELSER: That’s exactly right. If you think about the way we in the world do materials discovery today, it still is very similar to Thomas Edison’s process, where you mix up a new material and you try it. And you mix up a material and you try it. And you keep doing that, obviously using the scientist’s expertise to figure out the best combinations to get the properties they’re looking for. Over time, of course, we’ve used high performance computers and simulation to help that process, to be able to simulate materials ahead of time. But it’s always been a challenge because in order to simulate materials at a scale, large enough molecules to really be something of interest requires a huge amount of computational power. So that even the largest supercomputers can really only simulate fairly small molecules or reactions if you want to do sort of an exact simulation.

We believe, though, with the advent of AI, artificial intelligence systems, as well as eventually quantum computing, combining that with HPC capability really does offer then an option for truly accelerating the discovery of materials.

To lay that out, if you think about what a materials scientist does today to go after new material, they know all the materials they’ve worked with in the past in that area, and they read papers of course about other people’s results and they then try to come up with some new idea about what they might try next. If an AI system could actually read through much larger corpuses of papers and patents and experimental results, and could understand what it was reading in terms of what the features are you’re looking for, it could then surface ideas and combinations that maybe the individual science wouldn’t be able to understand on their own or find on their own just because they can’t read that many papers. It still requires a science to come in and then decide which of these things make a difference, but that can make a huge speed-up in terms of how quickly they can find new materials to try.

In addition, if you think about actually trying to do the experiments and you see the results coming out, obviously we can try and model what those results mean to try to figure out what the next experiment should be. But again, AI systems are showing promise for being able to look at a series of experiments and results and hypothesize different experiments you might want to try or different combinations to go after based on the patterns they’re seeing from, again, a larger corpus of data than what an individual scientist could look at.

We even had some experiments going on where we have what we call our Robo RXN system, where the robotic machine itself can be run to mix the chemicals up, under the direction of both the scientist as well as the AI recipes that the AI system can produce on what combinations might be the right steps to actually produce a specific molecule. And actually run hundreds or thousands of experiments even remotely.

So we think there’s a lot of opportunity by combining these technologies then to actually accelerate materials searching. The ones we’ve chosen to highlight in Five in Five were ones that not only we think are important from a societal point of view, but ones that we’re actually doing some experiments ourselves as well internally or with some of our partners.

BRIAN SANTO: So it sounds like one of the themes behind a lot of the problems, the challenges that IBM has selected to try to address with this year’s 5 in 5 seems to be with discovery. I get how that works with trying to find a compound that binds CO² better. I get how that might work with trying to find new chemicals for creating batteries that might be more efficient. Same thing with photoresists. Again, trying to find a combination of chemicals that do what you want them to do.

Even with the anti-virals, the one that I’m not quite figuring out is the nitrogen fixing. I thought that was really fascinating. I was reading the background materials. The way the challenge was laid out was that ordinarily plants need nitrogen, and the way they normally get it is from nitrogen-fixing bacteria in the ground. And when we make artificial fertilizers, it’s a big, energy-intensive process to create an energy-fixing compound. This is something where it seems like you’re trying to replicate a bacterial action, which maybe might be less of a discovery than some of the others on your list. Am I thinking of this correctly? Or could I get you to maybe talk a little bit more about that process.

JEFF WELSER: It is of a different flavor in this case because we actually are trying to, in the first step, really understand how the bacteria even do it. There are these subset of plants that can actually very efficiently take advantage of making their own fertilizer in a sense. From the bacteria and the outputs that come in the ground. Whereas the process we use to make fertilizer is incredibly energy intensive and admits a lot of carbon.

Can we understand, first of all, how they’re doing that? And that’s of course a lot of us continue scientific analysis of it. Even understanding the chemical pathways, understanding in detail how that chemical pathway works. To do that kind of work does require potentially HPC kind of simulations. But again, they aren’t really good enough at the scale we need. That’s one when we get to having large enough quantum computers, that’s where we really think that kind of technology can shine to actually do really complex chemical equations. Which are in a sense a giant quantum-mechanical problem. That’s what the molecules are doing there in the end, is mixing their electrons back and forth in different quantum-mechanical states.

That’s a sort of longer-term approach we’re taking. In the shorter term, back to something a little more discovery-like, we can look at existing catalysts that we have today. Again, looking at large numbers of them to understand, would any of these shows the catalytic effects they show in the functions we use them for today, would that be applicable to actually helping us fix nitrogen at lower energies? Because catalysts in general tend to be good a making a particular chemical reaction happen more quickly or with less energy input. So looking thorough, again, how we use catalysts in a wide variety of areas today that touch on similar molecules or elements to the nitrogen-fixing process and then combining that with our, hopefully, increased knowledge of how bacteria themselves are doing it, could we then find a catalyst that might actually enable us to do this at much lower energy?

BRIAN SANTO: We’ve talked about AI. We’ve referred to high-performance computing — we’re about to enter the exascale era. We’ve talked about quantum computing, and we know IBM has done some amazing things there. Are the other technologies that are becoming sophisticated enough that could help with some of these? For example, I’m thinking I’ve read about using semiconductors as a substrate for doing a lot of chemical reactions in parallel on one substrate, for example. Any other technologies that are coming into play here that you think are going to develop in the next five years and help with some of these grand challenges?

JEFF WELSER: Well certainly the one you just brought up there on using semiconductors for microfluidics or even nanofluidics, enabling us to do very precise chemical reactions on a chip, and as you said, many reactions in parallel even. And I think what’s also nice about it is not only the ability to do it from the standpoint of doing many at a time or all on a smaller substrate there, but also the fact that you could have a system them be automated. So you could have a system where a normal computer is controlling the reactions you’re putting in, controlling when you release different chemicals. And this then allows a level of parallelization and throughput that you can’t get just with lots and lots of grad students in a laboratory. Although grad students do very good work, they are being limited by how much a human can actually do.

The Robo RXN that I referred to is sort of a macro-scale version of that. It’s not a semiconductor substrate, but it is basically a robotic chemical lab that we built in our Zurich facility that allows you to run a set of automated experiments 24 hours a day, seven days a week, lots of them in parallel, where you can then feed in… If you get a whole bunch of different potential reactions, you should be doing different combinations, you should be able to do those all very quickly, but it also limits your throughput. If you could have a system where you could actually let that go on its own, actually could make a difference in how fast then you can get that work done.

The other advantage of it is, we did some experiments with it already around some of the CO2 capture materials, and we were able to do it with our Yorktown lab coming up with some of the ideas, actually running it in Zurich and then the Zurich lab giving us the results, and the ones that look the most promising, shipping those to our lab in San Jose to actually do some more advanced characterization of the results that came out. So it really allows a level of collaboration then, which can also help accelerate the whole process.

BRIAN SANTO: All right. My next question was going to be about collaboration. IBM in the past has been pretty generous sharing the results of some of its research with other organizations to help solve the grand challenges. I’m wondering if I can get you to talk about some of the group efforts or the other organizations that IBM is working with in addressing some of this year’s Five challenges.

JEFF WELSER: I guess I’d highlight two. One is obviously on the vaccine front, how we can help with the COVID-19 pandemic. A lot of the stuff that we’ve been surfacing… we ourselves aren’t going to go and make the pharmaceutical drugs. We have been putting out into the web… We had a website starting with COVID-19 in the summertime. We put out various tools people could use. One was the ability to find potential therapeutics that could be reused for treating COVID-19. There have been a couple of papers just published in the last couple of months from groups that actually had identified a couple of potential candidates, where they see some potential use for an existing drug to help with COVID-19. A lot of that work comes from the ability for AI to read through large numbers of studies and find places where a doctor perhaps has given a medicine to a patient that has more than one illness, so he’s given to him for a specific illness but happens to notice that it’s actually helping with the other illness. And it’s a good way of starting to say, Hey, there’s something here that’s worth looking at.

In addition to that, we actually put out some tools as well for helping people to sift through the genomes of bacteria and viruses to help understand more how that might find targets for treating either as a vaccine or else treating for therapeutics COVID, or also helping understand how maybe a person’s own microbiome might be affecting how sick they get or how well they are able to fight off the virus. There’s a lot of study right now on this whole question or how our microbiome is linked in to our health.

So these are examples of tools we actually put out for really the community at large of researchers to use.

Lastly, alone those lines, we also had started this HPC consortium. We started originally talking to the national labs and OSTP in DC, but then rapidly pulled on Amazon and Microsoft and a lot of other folks came together. So we offered up over 400 pedaflops of basically free computing power on HPC systems around the world to people who submitted research projects to deal with COVID. Obviously this is all about accelerating that search.

The other one I’d highlight then is on the batteries. We’ve had a lot of work over the years with various consortium, including universities and national labs around just trying to find batteries that have greater and greater performance. The more recent stuff that we just did where we tried to think about sustainability, we did with some partners, Daimler, as well as Central Glass, who does electrolytes, and Citus, which is a battery manufacturing company, to show that we could build a prototype of a battery that had no cobalt or nickel in it, which are both heavy metals of concern, and in fact could be utilizing iodine-based electrodes instead, which is a much cheaper material to get and something you can actually get from sea water, so you don’t even have to mine it. It shows very good performance. So we’re going to continue to work with those partners to see how we can actually scale this up to be a scale that’s not just a laboratory bench experiment, but going beyond it.

In that case, what we’re looking for is, IBM’s never going to go make batteries. Is there another path where someone else can go take this technology and go make a battery that can actually have impact somewhere?

BRIAN SANTO: Anything that I haven’t asked you about involving the Five in Five that surprised you? That you found interesting or intriguing or just plain groovy?

JEFF WELSER: I would say nothing stands out particularly, other than just the complete flexibility of these systems once you start to build them and think about them on how you can apply them in so many different areas. Obviously we chose to highlight five here, but really we see this as part of what we call our future of computing work in general where we really think the ability of computers to advance and accelerate our ability to do science and scientific discovery broadly. We’re really just reaching a turning point. And it really does come from the fact that we have the addition of AI and eventually quantum to our HPC systems, which we already use effectively today.

That really ties in to something else that we’ve been pushing at the same time as the Five in Five. This is this whole notion of the urgency of science. We’re hoping to get more companies and organizations thinking about, and it really just reflects the fact that as we look at the world’s problems, we don’t have answers for all of them, but almost all of them can be helped if we really advance the science itself. That’s not just advancing science technologies, but the way we do scientific thinking and applying that. Whether it’s making policy decisions or decisions about how we should attack a problem. In our everyday lives as well as on the grand scale of discovering new material.

So I think this urgency or science is something that I hope transcends just the Five in Five we’re talking about now, or even in the computing work we’re talking about, and it’s really something that people should take to heart as they think about how we go after solving a lot of the problem in the world today.

BRIAN SANTO: Jeff, thank you so much for your time today. We appreciate it.

JEFF WELSER: Thank you. I’ve very much enjoyed it.

BRIAN SANTO: That was IBM Research vice president Jeff Welser. To learn more about IBM’s most recent Five in Five, we’ve got a link to the company’s web page that talks about the program. That’s on our own podcast web page.