A Brain Trust: Uniting a Community Around Building Neuromorphic Capability

Meet Katie Schuman

The SC26 Communications team spoke with Schuman about her origins, a chicken-and-egg problem 15 years in the making, and what she wants the HPC community to understand about where neuromorphic computing goes next.

Catherine “Katie” Schuman

Assistant Professor of Electrical Engineering and Computer Science

University of Tennessee, Knoxville

SC26: You grew up in the shadow of Oak Ridge, with a pamphlet about the Mouse House genetics program on your bookshelf. When did the lab stop being background scenery and start becoming a real destination for you? 

Katie Schuman: A lot of my family worked at Oak Ridge before it was Oak Ridge National Lab. They worked there back in the X-10 days, and my sister still works at the lab. She actually worked in the same building as the Mouse House, where my interest in it started when I was around middle school.

I don’t think I really considered it a path for me until I was in graduate school, actively pursuing a Ph.D. I realized then that I was really passionate about interdisciplinary science, and a national lab like Oak Ridge is a natural place to realize those sorts of collaborations. It’s strongly encouraged to pursue cross-disciplinary collaborations to figure out how to use your technology or science to help another domain, if you can. That made the lab a natural place for me to seek out, and it was convenient that I already knew a lot about it from having grown up just down the road.

SC26: In 2015, you accepted the Liane B. Russell Distinguished Early Career Fellowship at Oak Ridge, named for the geneticist whose work you’d grown up reading about. What did you think you would be working on then? 

Schuman: The fellowships at the lab are very important for young scientists to have an opportunity to start building their own research area. When I was applying, I was targeting building a neuromorphic computing capability at Oak Ridge, but through the lens of computer science, thinking about how to most effectively use neuromorphic computers for real-world applications and building the associated software and algorithms to make them more usable. That was my pitch for the fellowship.

One of the ways I program neuromorphic systems is by using evolutionary or genetic algorithms, so it was a very natural follow-on from my interest in the Mouse House and was very well connected to what Liane Russell herself did because she did a lot of genetics work. I was very fortunate to meet her before she passed away and to hear about her experience as a scientist at the lab. That fellowship really allowed me to kick off my neuromorphic research, build neuromorphic as a capability at Oak Ridge, interact with other fellows across the lab, and begin building interdisciplinary collaborations.

SC26: For a reader who keeps hearing about it but has never had it explained: what is neuromorphic computing? 

Schuman: I always say that neuromorphic is literally brain-shaped computing. That’s what the Latin term means. We’re taking inspiration from the brain, both in the system’s literal architecture and in how it operates. If you’re not really interested in the neuroscience aspect, it’s just a computer with certain properties that draw inspiration from biology. It’s massively parallel. It has co-located processing and memory. It’s asynchronous and event-driven, and it can operate at very low power across a variety of applications.

It can do neuroscience simulation, of course, because it’s meant to be a brain in hardware. It can also do AI applications. But it’s not just limited to those. It can work on a variety of other problem types that map well to those massively parallel, event-driven characteristics – things like certain types of simulations, graph problems, and optimization. So there’s this opportunity to use a new type of computer that’s extremely efficient, can accelerate certain types of computation, and can do a variety of different applications.

I always say that neuromorphic is literally brain-shaped computing. That’s what the Latin term means. We’re taking inspiration from the brain, both in the system’s literal architecture and in how it operates.

SC26: Your DOE Early Career project used Summit and Frontier to simulate neuromorphic hardware that does not yet exist. What is it like to use the world’s fastest computers to design a fundamentally different one? 

Schuman: I think it’s a fun use of the largest computers to say, “Maybe this isn’t the only way computing should be done. Maybe we should use it to figure out how to build the next supercomputer.”

I can simulate neuromorphic systems on my laptop, but at a relatively small scale. If we’re thinking about a future neuromorphic system that could be used in a supercomputer, you’d want to understand what billions-scale neurons in the system would look like, and that’s not something I can look at on my personal computer, or even local clusters. You really need HPC-scale systems to understand that. We’ve evaluated a variety of hardware characteristics to understand what the design should look like, how it changes across different problem domains, and how it changes as you scale up.

I do find it kind of ironic to be using these very large-scale systems that require a whole lot of power to design super-low-power computing for the future. But it’s kind of necessary. We have to.

I do find it kind of ironic to be using these very large-scale systems that require a whole lot of power to design super-low-power computing for the future. But it’s kind of necessary. We have to.

SC26: You have said the lack of available neuromorphic systems has significantly hindered the field. When did you first feel that barrier yourself, and what has it cost? 

Schuman: Basically, my entire career. I’ve been doing neuromorphic research for the last 15 years, and there just haven’t been neuromorphic hardware systems available to the broader computing community, or even to the broader neuromorphic community, in a way that lets people access and use them easily and effectively. There are very few hardware implementations and very few companies pursuing the development.

From the perspective of simulating these systems, our existing computers aren’t built the same way. That means it takes a whole lot longer to simulate something, and you need HPC-scale systems to understand how even a moderately sized neuromorphic system would behave. It’s not reasonable for me to use Frontier and Summit every day to understand what the compute will look like.

So it’s been a chicken-and-egg problem. Hardware companies are waiting for a killer application, but it’s really hard to understand how to build those applications without hardware to prototype on. You have to simulate what it will look like. Because our existing systems simply aren’t built the same way, it takes much longer to perform the same sort of computation, which could be orders of magnitude faster if we had the physical hardware.

SC26: That barrier seems to be what THOR is built to remove. Tell us about it. 

Schuman: There are neuromorphic systems; there are just relatively few of them, and it’s hard for the average person to get access. The whole point of the THOR effort, which is the Neuromorphic Commons project I’m part of, is to make some of these large-scale neuromorphic systems much more accessible to researchers and students than they are now, so we can broaden the community’s user base.

We’re really trying to provide a user facility much like these HPC systems, so folks can access it. To date, this hasn’t been something an individual research lab at a university could acquire. In fact, even at national labs, relatively few DOE national labs have acquired these sorts of neuromorphic systems. So, having something an undergraduate student can access and start playing with, I can only see that improving accessibility and usability will accelerate progress in the field. It also enables the community to come together to build large-scale software and algorithmic efforts, similar to how HPC has built the community around these systems.

The TENNLab Neuromorphic Computing Research Group, which Schuman co-leads, at its all-hands meeting in June 2026 at the University of Tennessee, Knoxville. The group spans faculty, graduate students, and undergraduate researchers.

Image Credit: TENNLab / University of Tennessee

SC26: The Nature review you co-authored had 23 authors across universities, national labs, and industry. What does it take to coordinate a collaboration like that in a field this young? 

Schuman: It’s really hard. Anybody who’s ever written a paper with collaborators knows it’s not easy, and to have 23 people with 23 different perspectives coming to that sort of article, it took a lot of meetings and a lot of discussion. This wasn’t a situation where a couple of people did the work, and everybody else tacked their name on it. Every one of us contributed to the content and the idea-shaping in that article.

One of my favorite things about the neuromorphic community is that we are intensely collaborative. We work together in these large teams all the time because no one person has all the knowledge across the entire compute stack to build their own neuromorphic system, software, and hardware by themselves. As the community grows, we really want to keep that as a core tenet of the culture. It’s often challenging and frustrating to coordinate at that scale, but worthwhile in the end. We’re meeting regularly with the lead institutions, but we’re also actively building up community leaders to make sure we aren’t leaving anyone out.

One of my favorite things about the neuromorphic community is that we are intensely collaborative. We work together in these large teams all the time because no one person has all the knowledge across the entire compute stack to build their own neuromorphic system, software, and hardware by themselves. As the community grows, we really want to keep that as a core tenet of the culture.

SC26: That sounds like a tight-knit group. 

Schuman: Very much so. I joke that it makes it really hard for us to review each other’s papers, because there are conflicts of interest everywhere. We all work together all the time. 

SC26: Your first doctoral student just defended. What does it mean for the next generation to have access to systems you could not reach at their stage? 

Schuman: He’s graduating before we’re really fully giving access to all these systems, but he certainly had access to a lot of hardware that I never had. He was snapped up pretty quickly for a job because of the skills he built up doing neuromorphic research.

It’s an exciting time to be a student. My students are going to be some of the first users as THOR pilot systems launch. That’s both frustrating and exciting. They’re going to be our guinea pigs for figuring out what works and what doesn’t, so they’ll encounter all the bugs first. But they’re excited about it because getting to be among the first people to leverage these systems will be great.

SC26: You describe this as a paradigm shift. What do you mean by that? 

Schuman: To effectively leverage a neuromorphic computer, you need a different way of formulating the problem you’re trying to solve. In the past, when we developed algorithms and software, we were thinking of a very specific target architecture. Initially, that was CPUs. Of late, GPUs have been a very strong part of the community, so you’ve already had to go through that transition of rethinking how you problem-solve from CPU to GPU.

There’s the same sort of change in thinking that has to be done when mapping to neuromorphic or quantum systems as well. With these future compute systems, you basically have to rethink the way you solve a problem on each of them because each is different. The same algorithm is not going to perform well on all of them. So that’s a huge challenge, and it does require this paradigm shift, and it’s hard. But there’s huge opportunity in making more efficient compute systems and doing so without the huge energy costs we often have to consider with today’s scaling.

SC26: You’ve referred to neuromorphic computing as ‘brain-shaped computing,’ essentially a brain in a computer hardware form factor. Let’s turn that relationship around. The brain inspires the architecture, but does the architecture give anything back to brain science? Is there an advantage in using neuromorphic compute resources to research brain-specific bio/medical applications?

One of the drivers of the development of neuromorphic computers is their use in neuroscience simulations, particularly to enable very large-scale simulations in near real time and efficiently.  The hope in the community is that neuromorphic systems will enable us to run computational neuroscience simulations on networks up to the human-brain scale.  As we continue to learn more from neuroscience, we can incorporate those findings into neuromorphic systems and use them to help understand which mechanisms in the brain contribute to the different neural activity we observe. 

SC26: Last question. What is the one thing you want the SC community to take away? 

Schuman: There’s huge opportunity for certain subsets of the SC community to make huge contributions in the field. We’re a very small, tight-knit community, but one of the big gaps we have is in programming languages, programming abstractions, and compilers. We just don’t have many researchers working in that space.

We’re at the point where we have hardware systems and resources, like THOR, that people can access. Now, we really need folks to come in and help us improve computing on these systems. The HPC community is used to thinking about parallel computing, which is what happens natively on neuromorphic systems. So, they’re already primed to consider what it would look like to use one of these novel compute platforms. I just want them to know that we’d like their help and would like to collaborate.

SC26: Where Computing Research Converges