Quantum Co-design with Andrew Houck
E54

Quantum Co-design with Andrew Houck

Sebastian Hassinger:

Welcome to The New Quantum Era. I am your host, Sebastian Hassinger, and this is the podcast that features conversations with researchers working at building technologies that draw directly on quantum physics and quantum information science. On this episode, we will hear another conversation I recorded at the American Physical Society Summit. Thanks again to the APS for providing space and to quantum circuits for their help. My guest is Andrew Hauke, a quantum physicist, a professor at Princeton University and director of the Co Design Center for Quantum Advantage or C2QA.

Sebastian Hassinger:

This is a national quantum initiative center funded by the Department of Energy and led by the Brookhaven National Lab with partners from academia and industry. The NQI centers have been really important forces for aggregating and focusing talent, resources and innovation in The U. S. And at the APS summit, a number of the center's leaders presented their progress and discussed their future plans. My previous conversation with Anna Grassolino was focused on her leadership of the SQMS center, so I was really happy when Andrew agreed to also give us a perspective from the CTQA.

Sebastian Hassinger:

So, take a listen, and I hope you enjoy. So thanks for joining us, Andrew. Thanks for having me. Our pleasure. And I mean, our, I guess, the royal sense.

Sebastian Hassinger:

I often do this with a a cohost, Kevin Rowney. He's not here this week. So but, yeah, thanks for joining me. So I I saw your talk as part of the DOE NQI center's sort of, you know, status update yesterday, and I thought it was really interesting. In particular, you know, you sort of you talked about the sort of trying to create a virtuous circle for increasing coherence time.

Sebastian Hassinger:

So that do you think that was sort of the central mission of the CTQA over the last five years?

Andrew Houck:

Right. So the central mission is to sort of build the components you need to bring us past the NISC era and into the era of fault tolerant quantum computing. And that means bringing together people who work in materials and in devices and in software and in error correction and architecture because the problem needs to be tackled at all of those levels, and you need to make sure your gains at all of those levels are compatible.

Sebastian Hassinger:

Right. Right. And so, I mean, there was the great story about tantalum. What was I mean, were you surprised that that was sort of it was almost like incredibly low hanging fruit. You described Kava, I think, was the researcher who said, yeah.

Sebastian Hassinger:

I I explained how all these oxides are super different. Don't make devices out of the hypnobium. Right? I mean, that was that was kind of the mean, was surprising how how you sort of ran into that at at the beginning of the century in a way.

Andrew Houck:

That's right. I mean, you know, some of these things are both low hanging fruit and or or at least are low hanging fruit in in retrospect, but also are not easy. Right? Other people have tried to make things negative tantalum. Other people have tried to follow the the work in our papers.

Andrew Houck:

And one of the challenges in coherence and one of the lessons is that there's a quasi infinite number of ways that you can mess up coherence. And if you're really only using one number, what's the lifetime of my qubit? What's the t two? Right. And that's it.

Andrew Houck:

If you make a qubit with a new material, you might have messed up four other things Right. But made three of them better. And if that's all you look at, you'll never know. Right. And so, you know, we we're using this playbook where we try and disentangle all of the different sources of loss.

Andrew Houck:

You can see which ones get better, which ones get worse. And from that, you can actually start to build a much more comprehensive story as to how this works.

Sebastian Hassinger:

And and what are the sort of what are the the the top sources of loss in in your mind after after sort of the last five years of sort of digging into Right.

Andrew Houck:

So so, you know, I should say all of this work is is in collaboration with with Natalie De Leon and and Bob Kava at An an early guest on the podcast. An early guest on the podcast. I know. I I heard her her her guest a while ago, and so this is this is a status update. Right?

Andrew Houck:

So so we we are working to improve many possible sources of loss. If you wanna make a superconducting qubit, you have losses that that often come from two level systems, sort of defects in some oxide, maybe some kind of impurity. We actually don't know, and this is a a big grand teller, what actually are the TLSs or or the many different kinds of TLSs. But some of them live at surfaces, some of them live at interfaces. Tantalum has been really good at bringing down the surface loss compared to other metals.

Andrew Houck:

Right. In our previous generation of devices, we brought down the surface losses, and that made the bulk losses in the substrate actually the dominant thing. So by moving to high resistivity silicon now, we've knocked that down and have some of the very best performing qubits you can imagine. And now the questions are what what is left? Well, the surface is still limit you, especially if you wanna make small junctions.

Andrew Houck:

Interfaces matter. The actual Josephson junction itself is still made out of aluminum. Right. That could be a problem both because of the aluminum oxide and because aluminum is a bit of a fragile metal. You can't sort of dunk it in anything.

Andrew Houck:

Mhmm. Tantalum got its name after the story of Tantalus in Greek mythology who who who essentially embodies frustration. And it was named Tantalum because because the the people who first started working with metal found it almost impossible to etch. And so it is it's it's incredibly chemically resistant. And that means that once you make something out of it, can sort of put it in various acids.

Andrew Houck:

You can do a very aggressive cleaning steps, that allows you to get rid of many of the residual hydrocarbons that can be causing loss. And so there's all of these kinds of things.

Sebastian Hassinger:

Whereas aluminum is much more

Andrew Houck:

Whereas aluminum, right, if we if we had our junctions on there and we did what we do to our tantalum, we would not have junctions on.

Sebastian Hassinger:

Interesting. Interesting. And that's I mean, so the the CTQA, the the acronym is center for sorry.

Andrew Houck:

There's two c's. Co design center for quantum advantage. Right?

Sebastian Hassinger:

The co design was the part that I wanna pick up on because that that even that description as far down in sort of the whole stack of quantum computing as that is, there's material there's basic material understanding, and then there's already design in the fact that you're bringing together materials in different to perform different functions. Is that is that sort of embody what you what you think of when you think of co design?

Andrew Houck:

So I actually think of two different things when when I think of co design. So there's sort of the more traditional way of thinking about co design, which is moving vertically through the stack. So we make these materials work. We need to make sure it whatever we do with materials can be implemented in a device, and that device can be implemented in some kind of cohesive architecture. Right?

Andrew Houck:

Sometimes you can imagine a a way to make a Transmon with a really good coherence time that doesn't fit into what industry is doing in their scaled systems. And so the nice thing about the materials gains is is they can often fit more readily into those systems. But then you you do error correction on top of that. So one of the big, wins in c two q a was some work out of Michel Debrais' group that showed that with these GKP states for error correction, you can get beyond breakeven. Was one of the first experiments to show error correction brow beyond breakeven.

Andrew Houck:

And in their first version of that, it was really hard because the Transmon materials were the dominant sources of error. Right. When they incorporated the materials gain, they were able to get beyond break even. And then you can take these sort of bosonic error correction modes up the stack to think about how you build scaled systems Right. Can they use to to more efficiently do sort of near term quantum algorithms.

Andrew Houck:

And by thinking of things all the way vertically through the stack, you can often get really important gains. Mhmm. One of my favorite ones from c two q a is this idea of using native bosonic modes. Mhmm. So some of the applications people might wanna study in quantum computing are studying, say, field theories.

Andrew Houck:

Right. Those have bosonic modes in them. And one of the costly steps is to turn those bosonic modes into a bunch of qubits. Right. But if you actually have a resonator sitting on your chip, if there's something about your your hardware, then you can natively use that Interesting.

Andrew Houck:

To encode the bosonic modes and save maybe a factor of a thousand in terms of the resource requirements. And and this is really a co design

Sebastian Hassinger:

Yeah.

Andrew Houck:

Between the hardware and software levels of the stack.

Sebastian Hassinger:

Would that require sort of custom system architecture to be able to combine bosonic, native bosonic operations with qubit operations?

Andrew Houck:

So so it would, though you could imagine a some some something in between general purpose and custom where you're you're taking advantage of that. But frankly, you know, we we have in our best processors, you know, a hundred, two hundred qubits. You know, we're we're at the point where we need to squeeze every last bit of computational power over out of every element. Right? You know, we we can afford to make custom circuits if they can do something particularly well.

Andrew Houck:

Interesting. There's another flavor of co design that I wanna talk about that I think is a little bit unique to c two q a and also, I think, embodies what we try to do with the Princeton Quantum Initiative, which which is another thing that I direct on campus. And that is what we call cross platform co design. So very often and and I think this is especially true for companies, but but also fields in general. There's a bit of a monoculture.

Andrew Houck:

Right? Right. You you know, almost every superconducting person at some point went through Yale or or or at least went through somebody who went through Yale. You know? And that's wonderful.

Andrew Houck:

It's because, you know, they've trained really excellent scientists, and they they populate the companies. They populate most of the universities. But that means we all kind of think alike. We all learned the same lessons. And so when we tackle problems, we tackle them in the same way.

Andrew Houck:

Very often and, frankly, you know, in a in a modern era where everybody posts their papers in the archive, everybody sees everybody's result immediately, it's actually very hard to not think like everybody in your field, but the intersections with people outside of your field are harder to come by. And so some of the best ideas we have are taking ideas or approaches from one field and bringing them to another. So our work on tantalum, trying to figure out in a in a really robust way how to improve surface loss, followed what Natalie did in her postdoc where she tried to improve the coherence of diamond NV centers Right. As they got near surfaces for for purpose of quantum sensing. It's almost the same problem.

Andrew Houck:

Right. It's very different systems. Yeah. But the approach works really well. Yeah.

Andrew Houck:

You see this in other places. Right? So, Jeff Thompson and Judy Puri at Yale, and at Princeton, worked on erasure conversion in these neutral atom quantum computing systems. And that turned out to be an incredibly powerful approach to making the most out of their elements. And that is what inspired the dual rail qubit K.

Andrew Houck:

Which also implements erasure conversion in superconducting qubits, which is, of course, what Rob Schulkopf is working on at at Yale and at QCI. Yeah. Yeah. And so these ideas migrating from one physical system to another is what we call cross platform co design, because it really is helpful to get people who work on different systems talking and comparing approaches, problems, and solutions because often the idea that the problem has been solved in some other physical system. Right.

Sebastian Hassinger:

Yeah. Actually, in in conversation with Presco the other day, he suggested that neutral item platforms are almost a sandbox for experimenting with different approaches to error correction because you can reconfigure the qubits so plastically. Right? I mean, there's no connectivity issues to worry about. Just, you know, use the tweezers to bring the right

Andrew Houck:

ones Right. I mean, that that's certainly true. You know, you know, you if what you wanted to have is a thousand qubits that lasted a long time and had good enough gates that you could play with things in a reconfigurable way, that's that's definitely the playground you would want.

Sebastian Hassinger:

Right. That's so interesting. And when you when you think of sort of cross platform, when you think of the the defining line between one field and another, is that I mean, is that down at the granularity of, like, different qubit modalities? Or is it also different domains within physics? So maybe, you know, energy physics or astrophysics or other other domains as well.

Andrew Houck:

So so it it really goes at different levels of the stack, it goes in different ways. So, right, at the materials and device level, I think about it as qubit modalities. Right. Right. At the higher levels of the stack and applications, I think it's getting domain experts talking to quantum computer scientists and talking to quantum computing experts.

Andrew Houck:

And so, you know, one of the things we saw in c two QA was an enormous gain in efficiency for simulating nuclear physics problems that came from Nathan Wiebe talking very closely with a bunch of nuclear theorists Right. And learning about sort of how to use domain specific knowledge to develop sort of more interesting algorithms. Right. And so it's really about bringing people together. And, you know, frankly, this is one of the big challenges in science.

Andrew Houck:

Right? You know? Yeah. I I think of of quantum information as this big three d grid where you have sort of all the cubit modalities sort of horizontally next to each other, and then you have this vertical stack that sort of materials, devices, architecture, error correction applications. And then, of course, there's experiment and theory in all of those levels of the stack.

Andrew Houck:

Yeah. And it's you know, everybody sort of occupies a point or maybe an area in that. And when you build an institution like c two q a or like the Princeton Quantum Initiative, the goal is to try and populate that space. But you have to populate it with people who see the value in trying to bridge those connections because otherwise you have sort of, you know, a very sparse matrix with with almost no connectivity. And so the kinds of things we're trying to build are are really special collaborations.

Andrew Houck:

Yeah. And we're looking for people who see the value in talking to people outside of their their comfort zones and really trying to sort of enjoy science expansively.

Sebastian Hassinger:

Science and and also, I mean, to me, that's the smartest approach for trying to answer the question of of, you know, what are the applications of these devices. If, you know, I keep thinking about Stan Ulam at at Princeton IS coming up with the Monte Carlo approach to, I think, neutron diffusion calculations or something in the late forties, early fifties. And then, like, twenty five years later, somebody's saying, you know, this would work on financial portfolios. You know? Like, the the scientific techniques to use these devices in the way that Feynman described to simulate many body systems is like I mean, it's it's scratching an immediate itch that you can you can satisfy with something we can build today or in two, three, four, five years, which may indirectly lead to sort of broader applications.

Sebastian Hassinger:

Those techniques are effective, and then it's a searcher for, like, what else can you do with these techniques.

Andrew Houck:

Yeah. That's exactly right. And and it's always you know, people are always asking me, like, what's gonna be the first application quantum computers? And it's a it's always a hard question because there's a there's only a very small handful of problems where you can prove in an end to end way that a well, okay. You can show in an end to end way that in in all situations, our best quantum algorithm is definitely better by a huge margin than our our best classical algorithm.

Andrew Houck:

No proofs in that space. But but but at least we we have a known algorithm that we can prove a runtime on on a quantum computer. There are also a whole bunch of heuristics where it's like, maybe a quantum computer will work on that. We don't really know if it'll be better, you know, but let's try it out. And and until we have the hardware, you can't try it out.

Andrew Houck:

And so, you know, I expect there to be surprises. Right. But I don't know what they are because they're surprises.

Sebastian Hassinger:

Surprises. Yeah. It's the c two q a for surprises. So so you are I mean, the DOE is in the process of sort of, you know, getting the centers to resubmit proposals. Are there are there any sort of course corrections or major new elements to what you're planning for the next five years?

Andrew Houck:

Definitely. I think it would be, you know, in a fast moving field like quantum information, you you can't imagine that that five years go by and you're like, we are we got it right exactly five years ago, 90 PIs, you know, spread across 25 institutions, and we're gonna stick exactly with what we have. We are we are planning to bring in some of the the really exciting developments, so we are bringing in a lot of the work on neutral atom quantum computing. Many people in our center were were part of those developments. Right?

Andrew Houck:

Jeff Thompson's work in neutral atom quantum computing. And that means we're we're sort of ramping down some of our efforts in quantum networking. We had some some Erbium based quantum networking work that was really exciting and and it made incredible progress, but it feels like the neutral atom works is more of a priority. There's been an enormous growth in an enormous effort that that arose in diamond growth Yes. At the Princeton Plasma Physics Lab thinking about bulk bulk diamond growth for for quantum sensing purposes.

Andrew Houck:

And that feels like it's important to bring in because the kinds of problems, again, that you you tackle in diamond Mhmm. Are the same as the kind of problems that you tackle in superconducting computing materials. And so we're going to have sort of one effort that's really focused on materials broadly written and how you can build up this field of material science for quantum information. And another that's thinking about quantum computing in the vertical stacked direction, especially on on problems around modular quantum computing. So most physical systems are gonna run into some kind of wall.

Andrew Houck:

Neutral atoms are gonna run out of laser power when they hit a couple thousand atoms. Superconductors are gonna run out of chip size, at least, and so you have to figure out how to go off chip. Ion traps, you know, there's only so many ions you can put in a trap before before the Coulomb forces sort of make things get get a bit awkward. And so we're getting to the era where everybody's quantum computer is getting past the point where it's it's comfortable Right. In a module.

Andrew Houck:

And so then the question becomes, what's the ideal module size? Mhmm. How many qubits do you want? What connectivity do you want? What kind of connectivity between modules do you need?

Andrew Houck:

What kind of fidelity? What kind of rates? How do you sort of take algorithms and error correction and compile them to those modular systems? And that's a very large co design challenge Right. Because, you know, you might have an idea of an algorithm you wanna do.

Andrew Houck:

You might have an idea of some error correction you wanna run. You might have some devices and some actual figures of merit for your modules and for your interconnects. And now how do you put all those together? And and, you know, what's the missing piece that

Sebastian Hassinger:

you

Andrew Houck:

could invent on the mod on the sort of device side that would make the architecture flow much better? Those those involve resource estimates, sort of computer scientists, computer architects, material people, device people, all working together. And and you just can't do that kind of thing outside of a center. Right.

Sebastian Hassinger:

Yeah. I think you mentioned sort of the potential for very complex connectivity sort of topographies, right, where there might be nearest neighbor and then one far neighbor as a as an example, sort of and I can imagine extremely complex graphs that are sort of designed to solve a particular algorithmic problem. Is that sort of what

Andrew Houck:

you're That's the kind of thing. Right? So so, you know, for instance, in neutral atom quantum computing, you have a high degree of reconfigurability within a module. But, you know, nobody has yet demonstrated module to module links between neutral atom systems. In superconductors, it's very natural to make nearest neighbor coupling, but some of the the error correcting papers on LDPC code show if you can get a little bit of longer range coupling, then that's a lot better.

Andrew Houck:

But also, how do you couple from one chip to another chip? Right? You you probably don't wanna go from fridge to fridge, but from chip to chip, you know, do you just use a wire? Do you use a cavity? Do you do you how how do you build those kind of couplers?

Andrew Houck:

So there there's an enormous number of questions on the hardware side. And then what does that mean for error correction? Right? You know, what what does that mean for algorithms, and how do you sort of mesh this chopped up grid of small processors that are connected probably with worse links than you have just from neighboring qubits? How do you how do you do that sort of chopping up in division in a way that that is that is efficient?

Sebastian Hassinger:

I'm glad you're thinking that far ahead because it'll be great when we get the, you know, those those sort of optimal modules as you put it of of function, but you're right. I mean, as when I start looking at the way the architectures are are evolving, it does feel like they're gonna top out at some point before they're really achieving the kind of scale we want from and it's very hard to imagine a million superconducting cubits in one fridge on one chip. Well, I mean,

Andrew Houck:

one one fridge may be one chip, I think I think, is harder. But but if you imagine it on one chip, right, because, you know, the reason you have to think about it now is that, you know, you don't wanna then build up your single module to be a universal quantum computer that can solve any of this problem. Or maybe you do, but that might back you into a corner where going beyond that module size isn't actually the right path. Right. And so you sort of need to think, like, okay.

Andrew Houck:

I'm gonna make some decisions that are going to make the best quantum computer I can make on a single chip. But I'm gonna recognize that once I go past that, maybe what I want on one chip isn't a general purpose NISC or a quantum computer. It's a machine that's optimized to do x. Right. And that thing can be connected to other similar modules, and all of the interesting computation comes in the connectivity layer and not in the module itself.

Sebastian Hassinger:

I see.

Andrew Houck:

And so you have to sort of think about those architectures and those problems you wanna solve in order to design the modules and to design the physical systems. And so you can't just say, well, when we get to that problem, we'll we'll figure it out because, you know, you might have spent many years of effort going not quite the right way you wanna go. Right.

Sebastian Hassinger:

And you mentioned the Princeton Quantum Institute. How do you manage sort of the the collaboration? What's what's the advantage of of having those two sort of entities? I mean, one's obviously, you know, Princeton specific and the other is is multi institutional and also private and public. How do you manage that that collaboration between the two?

Andrew Houck:

So so, I mean, there there are sort of very different entities. Right? So so as one of the co directors of the Princeton Quantum Initiative, our goal is to build up quantum information science on campus. Princeton doesn't think just in terms of what are we gonna do for the next five years. Right.

Andrew Houck:

We are we are making one of the largest investments we have ever made in any field by building up a building for quantum information science, by bringing new people to campus. You don't make that kind of investment for five years. That's a fifty, hundred year investment that this is something that we think will be important. As we build that up, we are trying to bring people to campus who will collaborate and will work together and who see the value in coming together in a in a new discipline that brings together people with different different disciplinary trainings. Mhmm.

Andrew Houck:

CTQA is more aimed at tackling the kinds of challenge that is that are needed immediately to bring us forward as a nation. Right? It's it's part of the National Quantum Initiative. It's a national effort to make sure that we keep our lead in quantum information science. And it's built on an idea that this three legged stool of collaboration can solve problems that you can't solve just at a university or just at a company or just at a national lab.

Andrew Houck:

Right? Universities are really good at innovating new ideas and at trying them very quickly.

Sebastian Hassinger:

Right.

Andrew Houck:

Most of the new ideas in quantum information science still come from universities Right. And find their way into companies. Companies can build things at scale that you just can't build at the size of a single PI in a in a university, especially when your graduate students, you know, keep leaving and you have to keep training new people. Yes. And national labs have this enormous investment in infrastructure and talent, especially around very specific problems.

Andrew Houck:

So in c two q a, we leverage investments in material science. There is the beam lines Right. The light sources, the the center for functional nanomaterials that have incredible expertise at material science and and these billion dollar tools Right. For doing material science that had not really been applied to problems in quantum information and an enormous amount of expertise in codesign and in advanced scientific commuting

Sebastian Hassinger:

Right.

Andrew Houck:

And trying to apply that expertise and that infrastructure to problems in quantum information is incredibly important. And so CTQA, I feel like, is focused on bringing together academic talent spread around the country, industry, and leveraging national lab investment. Right. Whereas Princeton is trying to build something that is much longer lived and really builds a completely internal talent pipeline of people who wanna collaborate up and down the stack and horizontally around the stack and between experiment and theory.

Sebastian Hassinger:

Excellent. And and Princeton's investment at that scale in quantum information, is that because of a belief in sort of the practical applications of quantum computing, or is there a belief that quantum information is is sort of the next foundational advancement for quantum mechanics and and physical sciences generally, or both?

Andrew Houck:

Yes. Oh, yeah. Right. It it's both. Right?

Andrew Houck:

Mean, I was asking that question.

Sebastian Hassinger:

It's clearly both. You

Andrew Houck:

you wanna make an investment in something that you think is going to have an impact on the world. Right? That's that's one of Princeton's mottos is Princeton in the service of humanity. We we wanna have an impact. We also care about deep foundational questions on the nature of of the universe, and and quantum information brings together both of these things.

Andrew Houck:

Right? You know, that's why I was attracted to this field twenty five years ago. You know, I I I switched between physics and electrical engineering as my undergraduate major enough times that I think everybody was incredibly annoyed with me. You know, but but but I but but I had this sort of this this yearning to to study deep and interesting questions about the universe, but also do useful things. And and this this allows both of those those to go together.

Sebastian Hassinger:

It is the hundredth year anniversary of of quantum mechanics. Would you hazard a guess to what the next hundred years will bring in terms of the science? Well, mean, I I

Andrew Houck:

you know, hundred year your guesses are are are are always better when somebody isn't recording.

Sebastian Hassinger:

It is the the record will be long gone.

Andrew Houck:

Yeah. That's what they always say when they trap you into saying something dumb. No. You know, I I think one of the things that's really exciting about quantum information is we are building systems that are becoming macroscopically large. Right?

Andrew Houck:

A a Right. A million cubic quantum computer is basically a cat. I mean, it's, like, less fuzzy and cute, but it it's it's essentially Schrodinger's cat. And as you build these systems up, you can start to really ask questions about, do we actually understand quantum mechanics? Do we do we understand quantum mechanics as it turns into these macroscopically large objects?

Andrew Houck:

Is there is there something else lurking there? Can we really understand this sort

Sebastian Hassinger:

of interface between classical and quantum? That's really interesting. Excellent. Well, that was a fantastic conversation. A great overview of c two q a.

Sebastian Hassinger:

I'm looking forward to what you guys do in the next five years.

Andrew Houck:

I'm looking forward to it too.

Sebastian Hassinger:

Impressive. Thank you very much.

Andrew Houck:

Yeah. Thank you so much for having me.

Sebastian Hassinger:

Thank you for listening to another episode of the podcast, a production of the New Quantum Era hosted by me, Sebastian Hassinger, with theme music by OCH. You can find past episodes on www.newquantumera.com or on blue sky at newquantumera.com. If you enjoy the podcast, please subscribe and tell your quantum curious friends to give it a listen.

Creators and Guests

Sebastian Hassinger
Host
Sebastian Hassinger
Business development #QuantumComputing @AWScloud Opinions mine, he/him.
Andrew Houck
Guest
Andrew Houck
Princeton professor, director of the Co-design Center for Quantum Advantage, and co-director of the Princeton Quantum Initiative