Sam Stanwyck: Quantum Error Correction and Research Partnerships

Transcript prepared with otter.ai and human copyediting.

Sarah Webb  00:00

This is Science in Parallel, and I’m your host. Sarah Webb. Next up in our series on quantum computing, a conversation with Sam Stanwyck, Director of Product for Quantum at NVIDIA. He’s responsible for the company’s product strategy, roadmaps and partnerships in quantum computing. Sam talks about how he moved from a physics Ph.D. into quantum computing, the challenges of quantum error correction,and NVIDIA’s work to build systems and software that efficiently link quantum processors with classical ones.

Sarah Webb  00:48

Sam, it is great to have you on the podcast.

Sam Stanwyck  00:50

It’s great to be here. Sarah, thanks for having me.

Sarah Webb  00:53

Tell me about your position in NVIDIA. How long have you been there? What do you do?

Sam Stanwyck  00:57

I’m responsible for our products in quantum computing. So our roadmaps, our strategy, our priorities, but we should build how we work together with our partners in quantum computing, and how we talk about it. I’ve been at NVIDIA doing quantum computing for about five years now. I joined in 2021 and at the time, I was actually the first quantum-computing-only hire at NVIDIA. Before that, I worked for a quantum computing startup, and my background is in physics. But I started as a product manager, and now I lead a team that builds a set of products in quantum.

Sarah Webb  01:33

So take me back a little bit to your scientific background. How did you get interested in science and computing? Sort of lay the foundation for what got you into quantum.

Sam Stanwyck  01:43

 I think the first lightbulb moment was high school physics. I liked math. I liked science. I don’t know if I was especially amazing at any of it. Physics just felt magical and something that made sense to me. And the thing I liked about it the most is the reasoning aspect: how you could start with a couple of very basic principles, and if you understood how to think, you could solve every problem. And so as a kid, I had a hard time, kind of memorizing the parts of a flower and things like that in biology and chemistry. And so to have this scientific discipline where if I understood it, there was a lot less memorization, just really appealed to me. And so I ended up kind of coming back to that in college and then doing a Ph.D. in physics. And in grad school, I didn’t really have a career of any sort in mind, let alone quantum computing. But during my Ph.D. was around when the first quantum companies started getting going, either companies like IBM and Google spinning up quantum efforts, or startups like Rigetti and IonQ and I was based in the Bay Area. And Rigetti started towards the end of my grad degree, and I actually started working for them a little bit as a contractor at the end. And it just seemed like an incredibly exciting application of one of the most difficult and interesting domains of physics. And so I jumped in, and now here we are.

Sarah Webb  03:17

So what do you find most interesting about quantum computing. What would you say is kind of the most exciting thing, maybe right now or coming in the near future?

Sam Stanwyck  03:29

So I’ll maybe give you two versions of the answer. Sure, in the big picture, the most exciting thing about quantum computing is this is a fundamentally different set of physics than we’ve ever been able to harness before. So basically, every piece of technology, from the transistor to an AI supercomputer, is based on electricity and magnetism, mostly, and quantum physics is very different. It’s very counterintuitive. It’s not something where you know kind of until you start learning about it and doing the math and calculating, you’d be able to wrap your head around. It’s very fragile and hard to kind of see and feel in our everyday life, even though it’s everywhere. And so the fact that we’ve over the last century come from discovering it to understanding the principles, to doing experiments on it, to now being able to leverage it for a different type of processor, and something that we think could accelerate supercomputers, accelerate AI help us solve new and interesting problems, is frankly amazing. That is the most interesting thing about quantum computing, perpetually.

Sam Stanwyck  04:41

I think the specific version of the answer is our progress in quantum error correction. And so to get briefly, a little more technical, quantum computers are made out of qubits, which can take many physical forms. All of them are very susceptible to noise. Noise and very hard to kind of control and do operations accurately on in the way that we take for granted with transistors and logic gates in computers. So the very best qubits make an error about once every 1000 times, which sounds like not a lot, but in transistors, it’s one in 10 to the 15 ish, so more than 10 orders of magnitude better, and nobody thinks we can bridge that gap because of the kind of fundamental fragility of qubits. So in order to solve important problems where we do need to do many, many, many operations of order 10 to the 15 and get accurate results, we have to correct the errors that the quantum bits are making in real time as they’re being made, faster than they’re being made.

Sam Stanwyck  05:45

So we can have what are called logical qubits, which are error free, or at least arbitrarily reduced in error, and that’s what we think we can use to solve important problems and relevant to NVIDIA, that’s one of the big places where GPUs and AI come in is being able to do this quantum error correction. And what’s just phenomenally exciting is we’re now seeing this happen. And so many groups have demonstrated logical qubits, the ability to create a qubit out of physical qubits that is more accurate than the physical qubits that went into it, and they’re demonstrating the ability to do error correction in real time. So have your quantum processor, which is computing and making errors, sending those errors to a classical processor, be it a GPU or CPU or something else, and applying corrections faster than they’re being made. And this has been done by many companies now across many different types of qubit. And so now we’re like, on the ladder that we want to climb where we have logical qubits, and it’s a matter of making them a lot better and scaling them up. And just like my answer about quantum computing, the fact that this works at all is amazing.

Sarah Webb  06:55

So talk to me a little bit more about what’s involved in making that happen. Because it seems like, I mean, you’ve got qubits. You have your GPU, CPU, classical system. You’ve got information that’s being passed between the two what’s involved in making that error correction happen when you’re sending this across different types of processors?

Sam Stanwyck  07:15

It turns out to be as complex or more as the quantum computation itself. If you look at an estimation of the resources needed for computation with logical qubits, 99% ish of what’s happening is actually the error correction.

Sarah Webb  07:31

Oh, wow.

Sam Stanwyck  07:32

And is the actual quantum computation. And so I like to say that a quantum processor is mostly a quantum error correction machine doing a little bit of quantum computing, rather than mostly a quantum computer doing a little bit of error correction. So it’s incredibly important and computationally intensive. And what’s usually happening there are different ways to do quantum error correction, called different codes and different architectures. They do have some things in common. What’s usually happening is you have some set of physical qubits, let’s say 100, and you want this set of physical qubits to operate as a so called logical qubit. And you are running a set of operations on these physical qubits, which are called stabilizer circuits, where you are constantly generating what are called syndrome measurements. And syndrome measurements are telling you if an error has occurred and where it has occurred, but they’re not telling you that directly. You have just kind of some, some set of data, and what you need to get from that to what the error is, is what’s called a decoder. That’s what the classical processor does.

Sam Stanwyck  08:43

So your quantum processor is sending these syndrome measurements over some interconnect to a classical processor. The classical processor is looking at these syndrome measurements doing something called decoding. Where it takes those and decides, was there an error? Where is it? What’s the correction? And then it sends that back. And so all this time, the quantum processor is continuing to run. And so you can imagine, if you want your quantum processor to go very fast, and it’s making errors quickly, once you have a good amount of logical qubits, this task becomes very important to have a lot of compute with very low-latency transfer. You can’t be sending this over the internet, for example. And so, you know, just to briefly get into it, one of the things NVIDIA is doing is building the the architecture for this. So what should the interconnect be? What should the classical processor be? What should some of the software be? So if you want to use AI or parallel algorithms to do this, you can.

Sarah Webb  09:43

Cool. So what are you working on day to day in this area?

Sam Stanwyck  09:48

Yeah, we had a very exciting announcement of a new product in this space called NVQLink in October. And NVQLink is intended to solve the hardware and interconnect side of the problem and solve it in a way that works for everybody. Which is, you know, it’s difficult to build something that is great for everybody. There are lots of tradeoffs to be made, but NVQLink takes the best of NVIDIA and partner networking technology and allows companies that are building quantum processors or the systems that control quantum processors to use that to connect to GPU or GPU-CPU systems of different sizes, depending on their needs, and provides a software so that you can run error correction and other things that need low latency, real-time communication between the quantum computer and the classical computer. So that was our big announcement this year. There’s a software partner to NVQLink called CUDA-Q, which we’ve been developing for a few years. And CUDA-Q, for those in the audience who are familiar with other things, NVIDIA does, it’s the quantum equivalent of CUDA. It extends languages like Python and C++ so that you can use quantum processors in code that runs on CPU and GPU and QPU. And CUDA-Q is open source and is now connected to about a dozen quantum processors.

Sarah Webb  11:15

Where are we in this whole process of getting from the idea of quantum computing to the reality of, say, practical routine systems. How do you see this developing into something that people are using consistently, regularly, practically? How do you see that developing I mean, it feels like it’s just kind of on the edge of reality. So I’m sort of looking for your version of what the future looks like, or the near future looks like.

Sam Stanwyck  11:49

The near future looks like scaling and integration, in my opinion. So quantum computers exist. You can go run on a quantum computer built by a number of different companies now. Quantum computers are now deployed at national labs, like in Japan and in the US and in Europe. There are quantum computers that can be used together with supercomputers. And so already that is scientifically interesting to understand how to use a quantum computer. What it’s not yet is practical: So a quantum computer can’t solve any useful problem that probably even your laptop cannot solve, and you know certainly that a supercomputer cannot solve. So the challenge now is going from one logical qubit, maybe, or two logical qubits to 100 logical qubits, or 200 logical qubits, where you can think about a quantum computer solving or helping to solve useful and important problems.

Sam Stanwyck  12:51

I say helping to solve because the other challenge is tighter integration. Quantum computers don’t solve problems by themselves. This is not going to be a separate machine off solving a full problem; they accelerate very specific parts of a problem. And so if you’re wanting to use a quantum computer to help you design a drug or develop a more efficient process for creating batteries, you’ll use a quantum processor for a little bit of it, and a supercomputer or AI for the rest of it. And you’ll use it together as one system. So it’s a really exciting time. The fundamental pieces have been proven out, and now it’s just yeah about the scaling and integration needed to get to that future.

Sarah Webb  13:34

So I’m thinking about all the people who are working on quantum computing, whether that’s the national labs or academic researchers quantum computing startups, talk to me a little bit about how NVIDIA is working with some of those partners. How do you think about this process of integrating all of this toward going after these interesting science problems that are available to us?

Sam Stanwyck  13:59

NVIDIA is a platform company. We are building products intended for scientists and companies to build on top of and do their life’s work. We’re not kind of doing the whole enchilada ourselves, and so these partnerships are totally core to our strategy, and especially in a developing technology field like quantum computing. So with startups, we had a set of really exciting results in quantum error correction. I’ll name a few. Quantinuum, QuEra, Infleqtion are three different quantum computing startups, all of whom have used NVIDIA technology, our software and hardware, to accelerate their quantum error correction and do exciting demonstrations of logical qubits in the past year. We also work extremely closely with every national lab who’s working on quantum computing, and the national labs bring an amazing set of researchers and resources to this and are absolutely essential for driving all of this forward.

Sam Stanwyck  14:59

One kind of amazing result a few months ago, was with Lawrence Berkeley National Lab and NERSC where they used NERSC Perlmutter supercomputer to do a full wave microwave simulation of a quantum chip, a state of the art quantum chip with one micrometer resolution. And understandably that it’s not clear how big a deal that is. Designing quantum chips is incredibly hard and incredibly manual, and getting it right is a bottleneck now. And what we don’t want is design-fab-test cycles where we don’t fabricate exactly what we wanted, and we have to kind of go back to the drawing board. And so we want to simulate everything that we possibly can with as high resolution as we possibly can. And this is true for every computer, by the way. You know, this is what Synopsys and Cadence do, and this is key to NVIDIA’s business for our own GPUs. And we need to build the tools for quantum computers which haven’t existed yet. And NERSC and Lawrence Berkeley kind of took a big leap forward in developing that. They built a software package called ARTEMIS, accelerated by us on NVIDIA GPUs, scaled it up to a full supercomputer and showed that you could use it to dramatically accelerate the design process and reduce cycles. So that’s an amazing result.

Sam Stanwyck  16:16

The other set I’ll highlight is around NVQLink, and so there we’re partnering with just about every lab in the DOE so Oak Ridge, Berkeley, Fermi, Argonne, Brookhaven and many more to develop the specification for NVQLink. So all of those labs have authors on our papers about it, and start doing the first integrations and test systems. And I only see this continuing and expanding with the role of the DOE, especially the nqi centers, and with the convergence of quantum and AI in the recent Genesis mission.

Sarah Webb  16:52

What advice do you have, observations that you have, about where the field is heading that might be useful to people who are thinking about their career trajectories?

Sam Stanwyck  17:00

One general piece of advice is this is becoming a whole technology ecosystem, and we definitely need more people with quantum physics Ph.D.’s. That’s always going to be true. We also need chemists and biologists helping us look at applications. We also need electrical engineers and mechanical engineers helping us build cryogenic components and control systems and amplifiers and all of the pieces of technology that aren’t the qubits to make the quantum computer work. We need computer scientists thinking about and building the architecture for a new type of hybrid computer. We need product managers and developer relations and marketing people and business development people and creative people. So just about every type of technology job is now available in quantum computing, and so don’t feel like you need to go get a physics Ph.D. to contribute. I would be remiss if I didn’t say NVIDIA is hiring for just about everything, also in quantum computing right now. And so you can check our website, and if you think you’re a fit at all for any of the jobs, it does not hurt to apply. And again, in general, there are many, many open jobs in quantum computing now, both in the private sector and at the labs and in academia, with the NQI centers and with Genesis, I think everybody’s looking for people to help out, so don’t hesitate to dive in, I guess is my advice.

Sarah Webb  18:31

All right. Well, Sam, thank you so much. It’s been great hearing about your work and what’s going on at NVIDIA in quantum.

Sam Stanwyck  18:41

Thank you, Sarah, it’s great to be here

Sarah Webb  18:44

For more about Sam Stanwyck and NVIDIA’s work on quantum computing, check out our show notes at science in parallel.org. And to get all of our upcoming episodes, please subscribe to science in parallel wherever you listen to podcasts.

Sarah Webb  19:02

Science in Parallel, is produced by the Krell Institute and is a media project of the Department of Energy Computational Science Graduate Fellowship program. Any opinions expressed are those of the speaker and not those of their employers, the Krell Institute or the U.S. Department of Energy. Our music is by Steve O’Reilly. This episode was written and produced by Sarah Webb and edited by Susan Valot.