For theoretical physicists, there is a certain pleasure in seeing ideas and concepts applied and tested in experimental settings. For Steve Flammia, who is one of the principal scientists engaged in Amazon's quest to build a scalable and useful quantum computer, this interplay is something that he gets to enjoy every day. Flammia's introduction to quantum information came at the University of New Mexico, where Carl Caves was doing pioneering work in quantum metrology. For his postdoctoral appointments at Perimeter Institute and Caltech's Institute for Quantum Information (right before the creation of the NSF-supported IQIM), Flammia witnessed and contributed to the increasingly close relationship between the theories of quantum information, the experiments that tested their viability, and the complex engineering challenges that must be overcome to build a scalable quantum computer.
With a sense of adventure, Flammia accepted a faculty appointment at the University of Sydney, where he built a strong research group focused on applications of quantum information across a broad range of physics sub-disciplines, including condensed matter theory, topologically ordered phases, tensor networks, error correction, quantum optics, precision metrology, and classical statistical inference and machine learning. Thinking he had made a long-term research life in Australia, the Covid pandemic reoriented his spirit of new opportunity, and in 2020 he accepted an offer to join the Center for Quantum Computing at AWS, where Amazon is undertaking a massive and sustained effort to build quantum computers that will benefit both the company's core businesses, and the broader worlds of basic science and industry.
In the discussion below, Flammia reflects on Amazon's unique approach in this endeavor, and he emphasizes the numerous benefits and advantages that can be realized as a result of the Caltech and Amazon partnership. Between his own history as a quantum scholar at Caltech and his collaborations with John Preskill and other leading researchers, and the daily tasks and challenges in the lab, Flammia's perspective is uniquely suited to demonstrating that breakthrough achievements in quantum computing will need a vibrant and long-lived interplay between engineering and theoretical and experimental physics. As he emphasizes, that timescale will require patience and determination.
DAVID ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Monday, October 10th, 2022. It is great to be here with Dr. Steven Flammia. Steve, great to be with you. Thank you for joining me today.
STEVE FLAMMIA: It's a pleasure to be here.
ZIERLER: To start, would you please tell me your current title and institutional affiliation?
FLAMMIA: I am a Principal Research Scientist at the AWS Center for Quantum Computing. I also have a cross appointment at Caltech.
ZIERLER: When did you start at AWS?
FLAMMIA: A little over two years ago.
ZIERLER: Tell me about your work there. What do you do, and how does it fit into the overall effort at AWS in quantum information?
FLAMMIA: I work on—helping them build a quantum computer. [laughs] I spend a lot of my time lately thinking about how to calibrate a device. It is a very hard problem, to start with a pile of things that ought to be qubits and then turn them into things that behave like real physical qubits where you can do quantum gates on them. I have historically spent a lot of time thinking about things like characterization and validation, verification, of quantum devices as well as quantum error correction, but I've never been this close to experiment before. I'm spending a lot of time talking to experimentalists now. I've even run a few experiments myself, at this point. Overall, I'm trying to use my past life as a theorist and really now talk to experimentalists and make it happen.
ZIERLER: As a rough ratio, obviously it is all applied science if the goal is quantum computation, but how much of your day, how much of your agenda, is spent doing the kind of fundamental research you would be doing in an academic department, for example?
FLAMMIA: Lately, very little; probably 10%. I think that's a function more of the fact that the AWS effort is fledgling compared to older, more established industry efforts like IBM or Google. We have some catching up to do, and so it's kind of an all-hands-on-deck situation to get hardware off the ground. I think I have a bit of a longer leash than some of my colleagues because I'm a little more senior, and my boss, Fernando Brandão, tolerates me spending a little bit of time proving theorems or lemmas. But that is not what I spend most of my time on. Really I am focused mainly on helping experimentalists achieve some milestones with respect to our hardware.
The Caltech and Amazon Quantum Connection
ZIERLER: Between reporting to Fernando and having a joint appointment at Caltech, how much is Caltech in your world? How much do you feel the influence of—whether it's PMA or IQIM or John Preskill, how much of Caltech is baked into what is happening at AWS?
FLAMMIA: That is a great question. There is some component of that. I definitely feel it, but probably less than I would like, for now. The reason that I came to AWS, actually, was to have that connection with Caltech. I am really excited by the research that is happening at Caltech, and I want to be adjacent to that. Partly this is because of my personal circumstances; I have three young kids under age three. I have twins and a newborn who is about a month old. My bandwidth for doing things that are enriching relative to the goals, the immediate goals for AWS, is limited. I have gone to some seminars, I've gone to some workshops. Some of my colleagues have more interaction with the Caltech side of things. I hope longer-term, maybe a year from now, I see myself hopefully doing things like going to John Preskill's group meeting or seeing some more seminars on campus. Right now, I'm a little bit constrained. I'm the wrong person to give a great positive answer [laughs] but it's really my personal circumstances, not because there aren't great opportunities for that.
ZIERLER: You're not fully leveraging your courtesy appointment at this point.
FLAMMIA: Exactly, yes. That's not because I don't want to; it's just because of my constraints.
ZIERLER: So many questions about just the state of play in quantum computation at the industry level. I am so fascinated at how, with quantum information, there is now a level of industrial support for basic science that we haven't seen in this country really since the golden days of Bell Labs. It's really a remarkable thing. The first question is, when we look at AWS and Honeywell and IBM and Google and all of the startups, is it fair to say that they are all in a race for a shared goal, like a horse race? It's the same finish line, and it's about who gets there first? Or, are these very different companies both by size, by audience, by the things that they want to accomplish, and are they racing to build a different kind of quantum computer?
FLAMMIA: Definitely different companies are trying to build a different kind of quantum computer, if you define it to be the physical hardware that they are running. Companies like PsiQuantum are trying to build an optical quantum computer. You have companies like IonQ or Alpine Quantum Technologies trying to build ion traps. Others are trying to build more exotic things based on oscillators. I'd say most companies—IBM, Google, Rigetti—are trying to build some version of a transmon quantum device, or a superconducting quantum device. I think maybe the thrust of your question is, what are they trying to get from the end goal once they reach a quantum computer, not necessarily the physical hardware. Do I have that right?
FLAMMIA: I think that the goal is probably just to have a share of that market [laughs], a share of the people who are running algorithms on quantum computers. The way that they specifically do it is probably not super important to these players. Some of the players that survive in the long run I think will be running quantum devices and selling time on quantum computers to customers. I should pause for a second and just say that I don't have any special insights at the really big strategic level. I haven't talked to the CEO of Amazon, Andy Jassy or something, and picked his brain specifically about the long-term strategy. Probably even if I had, I may not be at liberty to say. [laughs] This is me putting together pieces from where I see the landscape and just using my brain and my intuition about these things to guess what the business strategy is. I don't have special insights. With that caveat, my feeling is people want to be using quantum computers as a cloud service, and you don't want to come in to a situation where your competitor can offer a cloud service that has a quantum computer on it, and you don't have one. That would be very bad.
ZIERLER: This is fundamental to AWS's business strategy?
FLAMMIA: My suspicion is yes. AWS has some huge market share in classical cloud computing, traditional cloud computing. If you are AWS, and let's say a few years from now some competitor comes out and says, "Here's my cloud. We can offer all the same scaling that you get on AWS cloud, plus we've also got this quantum computer," who are you going to go with? If you are believing that quantum computers can solve certain problems faster, and maybe we're in a future five years from now, ten years from now, where that does happen, you don't want to be caught in such a situation. You need to hedge, and you need to put some effort into this so that you can offer cloud quantum computing services to certain customers. That is I think the end goal for a lot of people. If you are a startup, your goal is to either actually provide a quantum computer or to get acquired by a big player, somebody who is putting in very serious money—Google, IBM, AWS, Microsoft. You want to become absorbed by that and get rich [laughs], I guess. Fundamentally, you want to be able to provide quantum computing services to customers via a cloud. I don't think anybody is expecting—I don't expect in my lifetime to have quantum laptops and quantum cell phones [laughs], unless there is a really radical innovation in technology. I think that is the end goal in the foreseeable-ish future.
ZIERLER: To the extent that AWS is pursuing a quantum computer scientifically in a different way than, for example, Microsoft or Google, can we reverse-engineer the way AWS is going about this, when we think about its end goal and its focus on cloud computing? In other words, is the process of building a quantum computer at Amazon—or AWS I should say—is the science being driven by what we hope quantum computing will do specifically for cloud computing? Whereas IBM might be thinking about general software, and Google might be thinking about search?
FLAMMIA: I see.
ZIERLER: Connecting the fundamental science to the different things it might be used for.
FLAMMIA: Certainly at AWS, one of our goals is to think very hard about fault-tolerant quantum computing. We are not the only ones thinking along these lines.
ZIERLER: Just to get our definitions in a row, what does fault-tolerant quantum computing mean?
FLAMMIA: Fault tolerance and error correction are intimately linked. Basically we want the analogous situation that we have on a conventional computer, where we basically have ideal computational operations happening on a quantum device. Error correction is the process whereby if an error occurs in the quantum device, we are able to correct it and continue our computation. Fault tolerance is where the hardware itself which is doing that error correction might actually spread some contagion. It might spread some of these errors. We want to be able to quarantine those errors, so that our error correction process, even in the process of faulty corrections, is robust to that noise that the hardware introduces. The goal, by applying techniques from error correction and fault tolerance, is to have ideal quantum devices, ones that function the way a theorist thinks about them completely abstractly, as just a perfectly functioning quantum computer. We want the error rates to be so low that we can just think about it as software running on a quantum computer, and that's it. That is the ultimate goal.
ZIERLER: The science, just the different way, when we look at Microsoft or Google, they're just pursuing a quantum computer in very different ways.
FLAMMIA: Right, so is it tailored specifically to cloud? I would say not so much. There are some people thinking about quantum networks and quantum communication, but if you just think about quantum computing, the end goal—yes, some people are thinking a lot about near-term applications. What do I mean by that? There is a popular phrase coined by John Preskill: NISQ—that stands for "Noisy Intermediate Scale Quantum"—devices. A NISQ device is one that runs on near-term quantum hardware. It is one where you don't presume the existence of a logical qubit; that is, an ideal quantum qubit which has very, very low error rates, lower than anything you would care about while you were running software. The question that is kind of the question for the very near term is, "Can we get anything useful out of a NISQ device?" I would still say it is not clear. There is a lot of effort that has been put into this question. I think there are a lot of promising avenues but nothing is a real slam dunk. A lot of people are pursuing applications on NISQ devices that might have commercial interest. That is a very interesting thing to do. Certainly my focus at AWS is on trying to get us to a fault tolerant quantum computer as soon as possible. I think the real interesting applications open up when we do have fault tolerance, and that's where I want to get as soon as possible.
The Potential Uses of Quantum Computers
ZIERLER: There is so much talk about not even knowing what quantum computers will be good for. At AWS, is that simply a rejected notion? That we know quantum computing will be good for cloud computing and that's where we're headed?
FLAMMIA: We do know that quantum computers will be good for some tasks, but I think what we don't really know is whether or not it will be commercially useful. We have a lot of evidence that there should be some commercial use cases. I don't think anybody has done something commercially that couldn't have been done on a conventional computer yet and then made money. You may have gotten some revenue, but you didn't make profit by doing it. I don't think that this has happened yet; maybe I'm wrong. That is what everybody dreams will happen. I think there is a genuine feeling throughout everybody in the community that, "If you build it, they will come."
ZIERLER: It's basic science. It's magnetic resonance, and now we have an MRI machine.
FLAMMIA: I wish I had at my fingertips the quote from John von Neumann where he talked about the use cases of a classical computer. He didn't call it a classical computer. But he said, "Look, what are we using these devices for?" This quote came from the 1950s or something. He was trying to pitch the notion of investing in computers. He said, "By the very definition, the applications we can't think of are going to be the ones that are most impactful." Maybe you even know this quote. It is because our limited brains are only thinking about the stuff that is right in front of us. There will be applications of quantum computers once we've built them. Our tiny brains are not really conceiving of what the most impactful applications will be. Those will arise. John von Neumann, as brilliant as he was, I don't think conceived of the internet, and that is one of the biggest applications of classical computing. Machine learning, maybe he had some intuitions for this. I think you could argue that he did. But I think all of the biggest applications are really beyond our conceptual reach at this time.
It may be that all of the things that we envision—there are some skeptical papers about, "Is quantum computing going to provide an exponential advantage for quantum chemistry?" Garnet Chan, here at Caltech, and many coauthors, are trying to argue, "Well, maybe not. Maybe quantum chemistry isn't the killer app." I know several people who are skeptical about this. So, okay, chemistry is out the door. That was a promising application for many years. Let's take the skeptic viewpoint. Maybe there are other skeptical viewpoints that you could have about various lines of inquiry where we sort of are advocating that this is where a quantum computer's strength will lie. Even if you take all that into account, I still think there's a great case to build this device, and it's exactly this von Neumann point of view. We cannot envision the applications where it will be most impactful. We just simply cannot. It is only by building this thing that we can harness its great power. We know from general theoretical arguments—we don't know; we have very strong [laughs] theoretical evidence—that this thing is going to be a tremendously powerful device. We won't actually be able to harness that power unless we build it.
ZIERLER: Why does industry need to be involved at this level then? It sounds like quantum information is still at a place where these fundamental questions about utility are very much up in the air. On that basis, why should this not be the domain of what academia does, which is just figuring out how nature works, fundamental research? Why do the Googles and the Amazons and the IBMS in the world feel so strongly that billions of dollars should be invested in the effort?
ZIERLER: I think it is FOMO, Fear of Missing Out. I think there's a bit of an arms race. Literally going back to this example of the cloud, if you are AWS and you have X market share in cloud computing, where X is some large number, you are really afraid of somebody else eroding that margin because they have some edge, even if that edge is pure marketing. If the edge is I tell my customers I have a quantum computer and the customers believe that a quantum computer is a valuable thing that will—let's actually move back even one layer and say you're a customer that uses AWS cloud or some other cloud service, and you find out that your competitors in your industry know all about quantum computers and are ready to pounce the second a commercially viable quantum application comes online. If you get caught flat-footed, you're in real trouble, so you want to hedge.
Even if you move back another layer from the industry players that are building quantum computers, there is a push from the people who are buying time on cloud, conventional, that they don't want to be caught flat-footed. They want to have some people who are knowledgeable enough about quantum that when a quantum device is viable, they can jump right in. Now, you see where the pressures are going. Now, there is pressure to be able to provide that to those customers, because if your competitor now has that and says, "Yeah, we're going to make this quantum device available so that you can train your workforce so that you're not flat-footed," even if that is just marketing, now that is going to start drawing market share. Now, you have this arms race mentality that happens in industry, where now I have to make sure that my margin doesn't erode towards my competitor, so now I have to invest in this.
I think certainly with our management—and like I said, it is not like I have had any contact with [laughs] the Andy Jassys or similar level people; I haven't—but my understanding of our philosophy and how we're pitching it up the chain is we're trying to be as honest as possible that this is a long-term bet, that we may get some revenue but we're not going to get profit from this. This is a long-term thing that we have to invest in. I worry that there are some players who are less honest about their pitch to their senior management, that "there is going to be profit soon, and there are going to be real applications tomorrow or today." I don't think that that is true [that real applications are arriving soon]. But I think that there is still a business case to be made, if you take a very long-term view. Big players like Amazon can afford to take that long-term view, and so I think they are investing with that long term in mind. I can't speak to other companies. I don't know what they tell VCs behind closed doors. I think there is a genuine fear that people are hyping our field too much and that there is going to be a shakeout, because people realize that those applications are not going to be profitable tomorrow, that the profit is going to come—my understanding is that VCs are investing on a ten-year time scale or something like this. They want to recoup their efforts across ten years. I think it's not a slam dunk that quantum computers are going to be profitable in ten years. If they decide that it's not profitable and they start to pull out, there is going to be some sort of contraction or some slowing of growth. I don't think that that is necessarily a problem. I think there are too many players that aren't serious, and a shakeout wouldn't necessarily be bad, although it might be bad for the individual people associated with those companies that don't survive. These are generally people who are highly educated; I feel like they will land on their feet. So, although that is maybe a personal tragedy for them for their companies to go under, I think that these are smart, talented people that should land on their feet. I hope that they do. But for quantum computing as a whole, it might not be so bad if we shrink a little bit and strengthen the fewer companies that are serious.
ZIERLER: Beyond the threshold of treating quantum computing as a shiny object for marketing purposes—that's only going to be viable for so long.
FLAMMIA: That's right, the rubber is going to meet the road.
ZIERLER: The customers are going to want to see, "Where does quantum computing benefit the cloud and my access to it?" and all the things that it does. This, by definition, will have to be speculative, but because it is what you are doing, because of all the money and effort and resources that AWS is pouring into this, when we do get to that level, what is the latest, most optimistic thinking about what scalable fault tolerant, error correcting quantum computers can do for the cloud?
FLAMMIA: There are two ways to do it. One is, where are the problems where we are most confident that quantum computers will solve them and classical computers won't be able to solve them? That's question one. Question two is, which ones of these problems are of interest to some potential customer? What would a potential customer want? Maybe it's some chemical company that wants a better reaction. I don't really know what these hypothetical customers might want from a quantum computer. I know that right now, for sure, there are people in academia who would love to use a quantum computer for research purposes. For sure, there are people who are paying to do this.
ZIERLER: For fundamental science?
FLAMMIA: For fundamental science, exactly.
ZIERLER: Everything from black holes to metrology.
FLAMMIA: Exactly. Anything like this. Or condensed matter physics. Google had this paper about preparing topologically ordered states on their quantum device. Great paper, great science. I think there's a ton of science like this, where academics want to be involved, and you can get them to pay money to run the experiments that they want to run, compiled onto a quantum device, on that device. Fantastic. Those would be paying customers today, if we had—IBM has a bunch of cloud devices accessible today. I think Fernando and Simone would kill me if I didn't pitch AWS to say that we have quantum devices as well that you can run, through third parties that run on the AWS cloud. You can go and run them today. I should have memorized the website so that I could pitch it! People will pay money to run their experiments on there, right now. But I don't think that we're really learning a lot that we couldn't learn otherwise, today. In many of these cases, not all, these things are accessible to classical simulation, and it's not a huge surprise what we get. There is a big market for people trying to do more fundamental science. Those people are definitely paying customers today. That's one of them. Then, longer term, where are quantum computers going to have an actual advantage over classical computers? This is a very hard question.
ZIERLER: Let's look at some of the basic limitations or challenges of classical cloud computing. What do we want? We want more power, we want more storage capacity, and we want more efficiency. Because these are carbons emissions nightmares, for example.
FLAMMIA: Oh, yeah. [laughs]
ZIERLER: On those bases, and any others that you want to throw into the hopper, what can a quantum computer do to make the cloud as we currently know it better?
FLAMMIA: In the Google quantum computational supremacy experiment that they did in 2019—it's a very artificial experiment. You mentioned carbon emissions, which I think is extremely interesting. They managed to sample from some specific probability distribution—that no one in particular ever really wanted to sample from, but that's a side issue. They managed to do it in a few minutes on their quantum device, much more efficiently than could be done on conventional computers. Since then, the advantage that they observed has eroded, as people attacked the problem using HPCs, high-performance computers, and a lot of tricks. But the energy consumption that has been involved to try to do those simulations on a conventional computer was enormous. So, there is a sense in which that advantage has persisted, even if the amount of time spent on a classical high-performance computer is now comparable to the time spent on the quantum computer. Quantum computations are fundamentally reversible. They're unitary dynamics, so they're reversible. In principle, they shouldn't take any energy. Of course with error correction, naturally to make a quantum measurement and then to delete the results of that measurement requires that you increase some entropy when you do this. Once you introduce error correction, it's not completely energetically free.
ZIERLER: What is the connection between reversibility and energy consumption?
FLAMMIA: If you have pure unitary dynamics, this conserves energy. You're not using any excess energy. Or any energy that moves into some subsystem moves out into some other system in a way which is conserved. But once you connect to some environment, which makes a measurement and then flushes out the data cache to make room for a new measurement, this is a lossy operation and it costs energy, fundamentally, to do this. Quantum computers have to do this as part of their error correction routine, but, in principle, a quantum computer is simply a pure, unitary operation. That is to say, it is a purely energy-conserving operation and therefore in principle would require no energy at all. You would just run it, and that would be it. So, it should be more energy-efficient, in principle.
ZIERLER: Orders of magnitude?
FLAMMIA: That's right. There is definitely a paper where they looked at the energetics of the Google experiments compared to these high-performance computing experiments that have subsequently simulated the Google experiment. It is definitely much more energy-efficient, even taking into account the fact that they used some huge dilution refrigerator, and that there's a huge amount of electronics that goes into actually operating their quantum device. It's a lot less than running a high-performance computer with thousands or—I'm not sure how many compute nodes they use in these experiments, but it's a lot.
ZIERLER: You mentioned how you are becoming more and more of an experimentalist.
FLAMMIA: [laughs] Yes.
The Quotidian Task of Building a Quantum Computer
ZIERLER: Without getting into any trade secrets, what does your office look like? Are people in clean environments? Are you working with these big devices? What does it look like for you?
FLAMMIA: I don't ever go downstairs and work in the actual lab and touch the dilution refrigerator, but I have access to the actual device at the pulse level, where I send in specific pulses to—I have programmed a quantum computer in the sense that I have said, "Apply this pulse to this qubit for this amount of time, and now measure it," which is way beyond what most of my theory colleagues [laughs] would be comfortable doing. They would say, "Here's a quantum circuit. Do this Hadamard Gate on this qubit, and now make this measurement. Tell me the result." That's about the most that they can do. Basically, they could program a circuit, and that would be it. But a real quantum computer operates at a level lower than that where the Hadamard Gate or the gate that you program is fundamentally driven by some pulse. Maybe it's a voltage that you tune with some amplitude for some amount of time. This changes some Hamiltonian, which controls the device [laughs]. So, you take a qubit starting in some ground state of whatever, some system, and you rotate it for some amount of time, and then it ends in some final state. You have to then concatenate all of these pulses, you have to calibrate all of these pulses. This is really messy. I've done some of that. Really it's my colleagues that do most of this. I just have done enough of it that I can converse with my colleagues. They are doing all the hard work, I should say, of course. But I am becoming more conversant in it. These experiments look a lot different when you dig below the circuit layer of abstraction. You have to be comfortable with time-dependent Hamiltonians and things like this, to be able to really program a quantum computer at the lowest levels.
ZIERLER: Again, without getting into any sensitive details, between industrial espionage, China, the NSA being so concerned about cryptography and error correction, how much of your day or your overall research agenda would you consider sensitive, even of things of the national interest?
FLAMMIA: Honestly, very little. I think much of what goes on still in quantum computing, my impression is that it is still in the published literature.
ZIERLER: Does that point to just how fundamental things remain, how everybody is still just exploring?
FLAMMIA: Yes, I think so. I know that there are efforts where people are not publishing, in the U.S., at least, where much of their funding, I gather, comes from Department of Defense contracts. This is my understanding. These are people who are not really publishing. Some of the startups are not publishing very much. You can just guess how big their efforts are based on how many friends you have. You can say, "Well, I know at least ten people that work there, so they probably have at least 50 people, or maybe 100 people." You just do the math on what a salary is to retain these people, and you can tell that many tens of millions of dollars per year is their burn rate, for sure. For sure. And it has to be that, if they want to be competitive, frankly. Maybe it's $100 million a year or more burn rate. If they are publishing one paper every year [laughs] there is a lot that they are doing that isn't making it into the literature.
But, there is a huge amount of basic engineering that is going into just building up all the things to have a quantum computer. There is all the control hardware. A lot of this stuff, there is a lot in the literature that you can draw on. I think there are innovations that are happening at the engineering level that maybe aren't making it into the literature, at these individual companies, but the broad brushstrokes for sure are in the literature. There are a lot of people putting in this work to build these huge efforts. Much of that can be gleaned from what is in the literature. Much of what I do on a day-to-day basis is tweaks to things that are in the literature, to try to improve and streamline things, and just building infrastructure of tasks that are already laid out in the literature. If I want to make some measurement of T1 or T2, we know how to do this; it's in the literature. So let's do it well on this device. That involves writing a whole bunch of code, connecting that code that drives the actual hardware to run that device. That just takes time and effort. It's not like if I told people how it was done it would be a big spill of a bunch of secrets. Probably I wouldn't say it, but I don't think I would be giving anything big away if I did. That said, there maybe are a few things that are secrets. I think you need to have a secret sauce if you want to catch up to your competitors. I mentioned AWS started a bit late compared to some other competitors.
ZIERLER: AWS institutionally looks at Microsoft and Google as being further along, at least in terms of time scales?
FLAMMIA: I can't really speak for AWS as an institution, how they view it, but I can say personally that I see for sure Google or IBM as being further ahead. They have been doing this for more than ten years now, so I think it is fine and natural that they are ahead. Microsoft is a bit strange in that they are pursuing a very different approach, with a topological qubit. There, I think it's just hard to judge. They have taken a gamble, and their gamble is that their specific approach will pay off by first building a new type of physical qubit with a basic robustness, and that then they will be able to rapidly build on that once they have done it. I don't know that it is fair to say that we are ahead or behind them. But I think if you are building something that is closer to a more conventional approach, that is you have some basic physical qubit which you then build into a surface code, which is what something like Google or IBM is doing, then yeah, AWS has some catching up to do. But we also have some tricks up our sleeve, like for example being here at Caltech. Everybody wants to live here [laughs] in Southern California and be right here on campus where you can do things like hang out with John Preskill at his group meeting, which I don't do enough. I think we have a draw there, and we also have some technical tricks up our sleeve that we hope will catapult us into the lead eventually.
ZIERLER: Perhaps as a probe to how you understand how academic your job is, for example, if you want to publish or present at conferences, or even call up some of your IQIM buddies, like Dave Bacon, for example, can you do those things in a way that you would with no problem, no questions asked, if you were a professor of quantum information?
FLAMMIA: I am one of the few people that I think has very deep insights into this question. I was a professor for years before I moved to industry. I think a lot of people will maybe do a postdoc and then go to industry, but I don't know a lot of people who have gone the full route of like postdoc, started out as a junior professor, moved up the ranks to full professor, and then left. Some people have. My perspective is, I don't have as much freedom as I did as a professor, but I'm not sure I'd go back, either. I can't just call up specifically Dave Bacon and collaborate on a project. He is at Google now. I think you have to be careful with intellectual property. It is part of your mandate as someone working at a company that you can't just call up your colleague at another company and strike up a conversation. Those days are gone. I don't think it is out of the question to collaborate with somebody else at another company, but it would have to be—A, it would have to be approved, but it would have to be about something that was—it would be a little bit tricky to navigate that. That's enough of a disincentive that I don't just call up my colleagues and say, "Hey, I had this cool idea. Let's talk about it." Also, honestly I just have so much to do that I don't have the bandwidth to do that type of thing, like just call up a colleague. I do have collaborations that I continue with colleagues outside of AWS. I have projects that I have done and that I am currently doing with colleagues outside of AWS. For example, my colleague Ryan O'Donnell at Carnegie Mellon, he and I have been thinking about some very theoretical questions related to measuring and understanding properties of quantum systems. He is an academic. I have the freedom to pursue these types of things. There is no problem in doing this. But yes, you can't just call up your colleagues at other companies anymore.
ZIERLER: What about presenting at conferences or writing papers that would discuss your work at AWS at a level of detail that people really know what you're doing?
FLAMMIA: We are pretty committed at our company to publishing most of what we do. I think there is an approval process. The approval process I find kind of slow and frustrating personally, but that is because I came from this academic background, where I can just publish whatever I want all the time, so having any barrier to publishing feels burdensome to me. But it's really not that bad in the grand scheme of things. I know that my boss and my boss's boss, Simone Severini, have said they are quite committed to publishing as much as possible and keeping things as open as possible. So, I haven't experienced any scenario where they have said, "No, we're not going to publish this." Everything that I've said I want to publish has gone through the approvals process and it has been approved. So, I haven't felt that constraint. I have felt, like I said, this disincentive to calling up my colleagues at other companies and saying, "Hey, let's collaborate on a project." But nothing that I have worked on has been embargoed fundamentally. I've definitely been able to publish freely. You have to complete some speaker certification training before you can speak, because they don't want you going out and saying things like, "I speak on behalf of AWS." If you do that, somebody is going to be very upset with you.
ZIERLER: Yes, "My view neither confirms no denies—"
FLAMMIA: Yes, "My views are my own." Everybody puts this in their bios.
Quantum Metrology at the University of New Mexico
ZIERLER: Absolutely. Let's go back now and establish some history. When you were at University of New Mexico, first of all, were you attracted to work with Carl Caves because you were already in the quantum information mindset, or was that something new to you as a graduate student?
FLAMMIA: I learned about quantum computing as an undergraduate. I was an undergraduate at Penn State. There was nobody there doing quantum information at the time. I had heard about it from a friend of mine in high school, who had become an undergraduate at Caltech, in fact, and he had told me, "There's this new thing, quantum information. It's really cool." He was also a physics major. I screwed around a lot in high school. I was a really bad kid. I had gotten expelled [laughs] from my high school, and I really only found my way maybe halfway through undergraduate. I wanted to be a teacher originally, actually, and then I realized I got a lot of enjoyment out of doing the actual science classes and not so much the education classes. I started to put more serious effort into that during my junior and senior year. He told me, "Oh, yeah, there's this thing, quantum computing." I did a little reading about it and decided this was something I wanted to do.
I was like, "I'm going to go to graduate school," and when I was researching graduate schools, there were not very many programs where there was something specific in quantum information. Caltech had one, but due to the aforementioned screwing around, I don't think it was likely that I could get into Caltech. I think I did apply and didn't get in, actually. But Carl Caves and Ivan Deutsch at the University of New Mexico had a strong quantum information group. There were a lot of people, and I looked up their papers. Nielsen and Chuang had written their book at this time. Michael Nielsen had been Carl Cave's graduate student. There were other people like Chris Fuchs who had written papers that I had read about quantum information and quantum foundations that I found really fascinating. I didn't understand everything that was in them, but I was like, "This is a legitimate place, there is where it is happening, and I can go there and I can be part of something new." I also understood quite keenly that if I had to go toe-to-toe with all these brilliant people doing string theory or whatever, I would just fail miserably. I had to have an angle. There are so many smart people in physics, and I know that my strength isn't to outsmart the smart people. That's just not how I'm going to succeed. But if I can do something new, then maybe I can be successful and do something that I love. I think I got pretty lucky [laughs] that my instincts were good. So, I went to New Mexico to work with Carl Caves.
ZIERLER: What was he working on when you connected with him?
FLAMMIA: He was doing some quantum metrology. He had been doing some quantum information. I think he was interested in some things that are a bit deemphasized now in the literature, about trying to understand entanglement and its role in quantum computing. I think maybe 20-something years ago, there was a push to try to understand the nature of why quantum computers are better than classical computers by studying the one thing we believed they needed to have, which was entanglement, and superposition. I think subsequent work has shown that maybe that was a little misguided. I think some of the things he was doing at that time were to try to understand that connection. Things like NMR quantum computers had been proposed, where the state of the NMR quantum computer was very highly mixed, and there wasn't a lot of entanglement. It didn't seem plausible that this would ever be a route to a scalable quantum computer. I think that it is, in principle, possible, but it is not ever going to be practical, in my opinion, despite things like algorithmic cooling. I think Carl had spent some time trying to think about the nature of quantum computing. Carl is very driven by fundamental questions. As he said in his interview with you, he is only interested in fundamental questions, and I think he wanted to understand fundamentally the nature of the quantum speedup. I think that is what he was thinking about at the time.
ZIERLER: Was it just him and Ivan, or was there a larger center like an IQI, at New Mexico when you joined?
FLAMMIA: When I joined, there were some connections with some people at Los Alamos, but New Mexico is a big state; you have to drive over an hour to get to Los Alamos. And, there were some people at the Santa Fe Institute who had joint appointments like Chris Moore. You did get exposed to some of those people. I should say, driving in New Mexico is very different from driving in Los Angeles. It takes you an hour to drive anywhere in Los Angeles, and it's not a very enjoyable hour. But it's pretty enjoyable to drive to Santa Fe. You get to see really beautiful scenery, and there's no traffic, so you're not constantly pressing the brake, and like, "Ugh," cursing that somebody just cut you off or something. So it's not unenjoyable to drive up and collaborate with those people. So, yes, there were some—like Howard Barnum was at Los Alamos. You did see these people, but not constantly. I feel like at the time, it was definitely for sure punching well above its weight. There weren't many centers that had anybody doing quantum information. Ivan Deutsch does quantum information and also quantum optics, and so it was a good place to build a foundation for quantum information science. I think it still is.
ZIERLER: Tell me about developing your thesis research. How did that go?
FLAMMIA: I think it was kind of hodgepodge, to be honest. I didn't really know what I wanted to work on. It took me a very long time to find myself, as you can guess from my back story. I worked on a few different projects. I remember Joe Renes had been thinking about the SIC-POVM problem, and I got interested in this.
ZIERLER: What is that? What's the problem?
FLAMMIA: [laughs] Oh, man. It's a problem about trying to find a certain set of pure quantum states which have equal pairwise overlaps. So, in a d-dimensional quantum state space, can I find d2 quantum pure states whose pairwise overlaps are all equal? That's a precise statement. It's a very attractive thing to work on if you don't know what else to work on [laughs], because it's so well defined. Part of the challenge, I think, when you're a beginning PhD student is you don't always know what you want to work on, and you don't know how to choose problems, so well-defined problems are attractive problems. This was very well-defined, and I felt like I understood the problem. I felt like I might be able to solve it. And I still haven't solved it, 20 years later, but that's okay.
I tried to work on that a little bit. I tried to work on some other entanglement-type questions that Carl had been asking which, as I said, subsequently the thrust of research in our field has really shifted away from these questions. I don't think those questions are impactful long term. We learned something by studying that at the time, but ultimately I feel like it was a bit of a dead end. I did spend some time working on metrology, and this was maybe largely due to the influence of Sergio Boixo, who is now at Google. He got quite interested in Carl's earlier work on generalized uncertainty relations, and he was reading those papers. He and I had a lot of discussions about how to understand quantum metrology. There was an influential paper that came out around that time by Lloyd, Giovannetti, and Maccone, about quantum metrology. It might even be called "Quantum Metrology." They use these ideas due to Carl Caves and Sam Braunstein in a pretty fundamental way. I was influenced greatly by Sergio, who thought there was a lot to do in this space. Towards the end, I started to feel like that was something I wanted to do, was quantum metrology. In the end, my thesis was kind of a hodgepodge [laughs] of just papers that I glued together, with somehow entanglement being key. Even though it wasn't key in the long run, that was the unifying picture that I presented at the time.
ZIERLER: Did Caltech loom large when you were a graduate student and you were already thinking about quantum information?
FLAMMIA: I think so. I graduated in December 2007, and already by that time, they were such heavyweights. Really, really influential things were happening at Caltech. I can remember some of them. Development of really key ideas. Sergey Bravyi and Kitaev coming up with magic states was in the time period where I was a graduate student, and that happened at Caltech. Guifré Vidal coming up with simulation algorithms for one-dimensional quantum systems. Frank Verstraete had some really important work that he did at Caltech at that time as well on that topic. Barbara Terhal wrote influential papers about understanding complexity of quantum systems, and things like this. She was a postdoc at Caltech. There's just too many things to count, to be honest. Even leaving things out, I would feel bad.
ZIERLER: Did that plant a seed after you graduated?
FLAMMIA: It was really clear that Caltech was the place to go. It was totally the place to go, I think especially if you had more of a physics bent. I think if you were more computer science-y, you might argue some other places could be just as good or better. Umesh Vazirani at Berkeley had done so many influential things as well, so you might want to go work with Umesh if you were more of a computer scientist. But I think if you had a physics background like me, it was clear that Caltech was the place to go. Before I graduated, Perimeter was pretty new and very attractive. Michael Nielsen had just announced he was moving there. IQC had a really big effort as well, so I think Waterloo loomed really large as well. I think those were really, in North America at least, the two really big hot spots, if I remember correctly.
ZIERLER: Why Perimeter first? What was attractive to you about that?
FLAMMIA: Because you got a three-year postdoc offer, and it felt like you had a longer leash, because you were just a researcher. You weren't hired on somebody's grant to work with a professor. It was just, "Here's some money. We're going to pay you for three years. Do some research." Which is scary when you're a beginning postdoc, because unless you have a very clear idea of what you want to do, you're really getting thrown in on the deep end. It's kind of trial by fire. In the end, actually, I never applied to do a postdoc at Caltech when I graduated, even though it was the place to go, because I felt it was too intimidating.
FLAMMIA: Yeah. I had—I still have it, to be honest—imposter syndrome. Everybody says this, right? Magnus Carlsen, the chess world champion, said in some interview recently that he has imposter syndrome! So, I had it, and I was like, "There's no way I am as good as these people. I'm not even going to bother applying." John actually invited me out to give a talk when I was still a graduate student, which I look back in retrospect as being his way of—I know now enough about the process to realize that what had happened, almost surely, is that Carl had run into John at a conference, and John had asked Carl, "Who are the promising students these days?" Probably Carl had mentioned me, so John invited me to give a talk, which I did.
ZIERLER: On what? What was the paper?
FLAMMIA: It was about quantum metrology, and it was about using k-body couplings to enhance quantum metrology in a way that is, in a sense, beyond the Heisenberg limit, although it turns out you should probably redefine the Heisenberg limit. The sexy way to sell it is to say it's beyond the Heisenberg limit, but the Heisenberg limit, if you define it to be a scaling with one over the number of particles, we showed that you could achieve a scaling which is like one over the number of particles squared. This was observed subsequently in experiments, in fact, so we know that such a scaling is possible. It doesn't violate the theorems about Heisenberg scaling; it just violates certain definitions of what it means to scale. It cheats in a way, but it cheats in a good way, because it was validated by an experiment. That's the good way to cheat. So, John invited me to give this talk, and I was too scared to apply, because I didn't think I was worthy. Actually, also at the time, it just seemed very hard to ever make a career out of quantum information.
ZIERLER: This is also the middle of the financial crisis.
FLAMMIA: The financial crisis hadn't yet hit. I was applying for postdocs January 2007. It was hitting in the end of that year, I think, or maybe the next year.
ZIERLER: 2008, yes.
FLAMMIA: But I was just really happy with my life in New Mexico, and I didn't really want to leave. I told myself that unless I got—I think I only applied to Perimeter? No, maybe I also applied to IQC. I only applied to go there, at Perimeter. I said, "If I don't get it, I'm just going to leave the field." [laughs] Because I was like, "This is a good position. It's got stability for three years. I can do whatever I want. I'm just going to go there, and if I don't get it, I'm going to leave the field."
From Perimeter Institute to Caltech
ZIERLER: To the extent that entanglement was the connecting thread of everything you worked on for your thesis, was that the game plan at Perimeter, to refine and expand on entanglement research?
FLAMMIA: No, the game plan was, "I still don't know what I want to do with my life, and this gives me three years to figure it out."
FLAMMIA: Michael Nielsen was just moving there, and I knew Michael personally.
ZIERLER: You knew him through Carl?
FLAMMIA: I knew him because Carl had introduced us, and I had traveled to Australia—this was my first time moving to Australia. Actually, I moved there and spent a month there, and I worked a little bit with Michael. We never published together, but I talked a little bit with him, and I was like, "This is a guy I would like to work with" as well. It was attractive to try to go there. And Gottesman was there. Gottesman is brilliant, of course. I was like, "Maybe I can make something happen if I go there. If I wind up not there, then probably I'm not destined to be in this field, so I'll just do something that makes me happy, and I'm not going to try to worry about it too much." Because it seemed very stressful. It is still stressful, I think. The academic landscape has changed dramatically, but it is pretty stressful if you stake your hopes on an academic position. Pedigree matters. As much as we wish that it didn't, it matters. So I was like, "Unless I get something that gives me some pedigree, I'm just going to tap out, I'm going to find some job that makes me very happy, and I'm going to live here in New Mexico, where my life is great. And that's it." But then I got the postdoc, so I moved to Canada. [laughs]
ZIERLER: What were some of the big ideas at Perimeter at the time in quantum information? What were people excited about?
FLAMMIA: There was a lot going on. This was I guess 2008, 2009, 2010. I started thinking about some different things at this time. I started doing my own thing, to be honest. [laughs] I can remember some interesting things that were happening at this time. I think the color code people were... Like I remember meeting Héctor Bombín. He was my colleague for a little while. He was talking about the color code. There were some interesting things happening then. I think there were some interesting things about topological order and topological quantum error correction, at that time. I think some of the theory questions about Hamiltonian complexity seemed compelling and interesting at that time. Some of these things seemed rather hard to approach from my background, and I wound up writing some papers on very different topics, because I didn't know how to contribute to some of those thrusts at the time, so I just did different things.
ZIERLER: What was the research culture like at Perimeter? Would people collaborate, or was it mostly single-author papers?
FLAMMIA: It was less collaborative than I would have wished. I found it challenging, to be honest. Daniel Gottesman, for example, is kind of an intimidating fellow. If he reads this, he is going to be surprised to hear that, maybe. He's really smart, and when you talk to him, he is very quick to jump to the logical conclusion of what you're about to say before you have even reached it. I found that really hard. Like I said, with imposter syndrome, I didn't really want to go talk to him. It made me feel dumb, to talk to him. It's not his fault. It's not my fault, either, in any sense, but I felt bad feeling – I would talk to him and he knew what I was going to say before I could articulate it, so that made me feel bad about myself, and bad about my abilities to do science. So my not very healthy way to deal with that was, "I'm just going to avoid talking to Gottesman." [laughs]
I sort of turned my focus inward. I talked to some people that I knew personally, like Nick Menicucci and I had some collaborations where we worked on a very dark-horse candidate for building scalable quantum computing, which is continuous variable cluster states. Nick has pushed this very hard, actually, and done really great work, together with experimentalists like Olivier Pfister at the University of Virginia, and Akira Furasawa in Japan, and now others as well. In Scandinavia, I forget exactly where, they have built large-scale experiments building 2D cluster states on thousands of nodes now. It's really interesting stuff. This was something that I contributed to a bit with Nick and others at the time. I just picked some problems where nobody else was really working on it and I felt like I could have an impact but I didn't have to feel like I was competing with people where I felt like everybody was smarter than me. [laughs]
ZIERLER: I'm thinking of Lee Smolin and people like that. Were people who were in astrophysics, cosmology, starting to make connections about the potential value of quantum information in their fields?
FLAMMIA: Yeah, they were. Thank you for reminding me of what happened at this time. The Hayden-Preskill paper, "Black Holes as Mirrors," that paper came out during this time, and that was definitely a splash. That was a fantastic paper. It was something where I felt like I could actually understand what was happening as well. It was clear that it was a toy model, from the perspective of high-energy physics, and I knew very little or next to nothing about high-energy physics, but I felt like I could understand this toy model. I definitely remember hearing high-energy people talking about it in the Black Hole Bistro, the little coffee shop at the top floor of Perimeter Institute. I could hear high-energy people talking about it and asking basic questions about quantum information and not knowing what the quantum information people were talking about. But I knew, and I thought, "But everybody knows that." I remember very distinctively, one example was Lenny Susskind was sitting behind me or next to me or something, and he's like, "What is this twirling? What are these people talking about, twirling?" I was like, "I know that. Everybody knows what twirling is!" I think I may have even answered that question. I mean, Lenny certainly doesn't remember this, I'm sure. He doesn't even know who I am. But I remember it, because Lenny's famous, right? That was pretty cool to hear real connections building right here before my eyes.
Yeah, that definitely happened at that time, and I remember that pretty strongly. I remember I tried to talk to some people about this. It was very hard to talk to the string theorists. It was just hard to understand their language. I wanted to learn more about string theory at that time, but it was just hard to make connections. I feel like I understood a little bit more what was happening from the people who were doing loop quantum gravity. You mentioned Lee Smolin. Some of the postdocs there, I talked to about science. Also, we played foosball together after lunch. [laughs] Then I understood a little bit more about the style and approach that they were taking. It made a little bit more sense from my background. So, I was able to learn a little bit about that stuff, but we never really collaborated.
ZIERLER: I know at Perimeter, it is a world of theory you're living in.
ZIERLER: But because in parallel, at IQI, we're starting to think about experimentation, condensed matter—the "M" is on the horizon. It's about to become IQIM. Were those similar things percolating at Perimeter? Were people starting to think about quantum information in terms of hardware and experimentation?
FLAMMIA: At Perimeter, maybe not. But just up the road in Waterloo, I think they were. They had experimentalists there. You could go and talk to them, and I did. I talked to them a little bit. There were plenty of people doing experiments at Waterloo, at IQC, the Institute for Quantum Computing there. But I don't know that this percolated so much to Perimeter. I think Perimeter prided itself a bit on—
ZIERLER: Pure theory, all the time?
FLAMMIA: Well, yeah, I mean pure theory is, in a sense, pointless, if it doesn't connect to experiments. Definitely there were experimentalists that would come and give seminars there, and high-energy theorists would make predictions about things like, "Oh, I predict to see this particle at this energy." It's not like they were explicitly trying to be devoid of connection to experiment; I think that they craved connection to experiment. But at the same time, I don't think that within the theory group for quantum information, this was a point of focus. Not at that time.
ZIERLER: Did you get over the imposter syndrome enough to put out feelers to Caltech at a certain point, or how did that happen?
FLAMMIA: I actually sort of had a replay of my same reasoning. I only applied to Caltech for a postdoc, with the same reasoning. I did not get over my imposter syndrome. I'd still say I'm not over it, to be honest, although it's definitely better. By the time I finished my postdoc at Perimeter, by the time it was ending, the financial crisis had happened, and there were no jobs at all, in North America, at least. I think my colleague Toby Cubitt said this to me—"The dark ages of quantum information happened after the financial crisis." There were many smart people who left the field at this time, or who took jobs that were clearly beneath their stature within the field. Really brilliant people. Maarten Van den Nest is a name I remember at the time, and he had made a lot of really interesting contributions. He left the field, because he couldn't get a permanent job. Dave Bacon left the field because—he may have been able to get a permanent job, but he decided to leave the field. Aram Harrow I think had a permanent job and for various reasons decided to take a research faculty position around this time. These are names that are really brilliant. I mean, Aram Harrow is a luminary. Dave Bacon is brilliant, right? It just seemed [laughs] really like it wasn't very promising to pursue an academic track.
This was prior to there being big industry efforts. There were some industry efforts, but I didn't find them very compelling at the time. You had these papers being written in industry, people talking about tuning up a one-qubit gate or something. What am I going to do with one qubit, or two qubits? It's totally boring, from my perspective, when you think about the possibilities of a scalable quantum computer. So, I didn't see industry as something that would appeal to me at the time. I thought I would kick the can down the road a little bit. I used the same reasoning. I said, "I'm just going to apply to this place, and if I don't get it, I'm just going to leave. I'm going to find a job somewhere, a job that makes me happy in a place that I want to live, and then I'll leave the field, and that will be that." I applied, and lo and behold, I got it. [laughs] So I moved to Caltech, and that's how I got to Caltech. [laughs]
ZIERLER: Tell me about your initial impressions when you finally arrived. What was happening here?
FLAMMIA: Oh, god! The day I arrived, my first day that I came in, I think it was a Wednesday, and everybody said, "Oh, the Preskill group meeting is on Wednesdays. It's catered. And you're in luck because Kitaev is talking." I thought, "Oh, awesome. Great." So my first day, I got to attend the group meeting and sit through basically a three-hour-long meeting where Alexei Kitaev talked about classifying phases of matter for two of those hours [laughs]. It was really exciting. It was exhausting. It was interesting. It was fantastic. It was overwhelming, but in a good way.
ZIERLER: Now, the collaborative environment that you were hoping to have more of at Perimeter, did you have that at Caltech?
FLAMMIA: I did. I got more of that. I didn't write a lot of papers while I was at Caltech. I think I spent some time trying to—there's this syndrome where I really felt like I had to take a really big swing while I was here. Like I said, I mentioned a bunch of names like Terhal, Verstraete, and people like this, or Bravyi. I was like, "Man." I looked at some of their contributions like, "I need something that's at that level if I'm going to prove my worth here." I don't know that I really [laughs] did that while I was here. I tried to prove some things about topological order, stability of topological order, while I was here, that never really panned out. I tried to prove some things with Spiros Michalakis and Norbert Schuch. Spiros, in the end, did prove some very interesting results along these lines, with Sergey Bravyi and Matt Hastings, and then some follow-up work with coauthors. But I spent some time trying to prove some things like this, and never really getting anywhere.
But I did have some collaboration. It was exciting. I had a lot of great discussions with people. Like my office mate, Stephen Jordan, and I, we came up with a quantum algorithm that had an exponential speedup, which unfortunately, after we had written a draft, it turned out that other people had already come up with this quantum algorithm. But it was a collaboration, nonetheless. That sort of thing happens in research, of course, so that's okay, but it was exciting to collaborate on different things. I had never done research on quantum algorithms, so to have that interaction with Stephen Jordan was fantastic. I really learned a lot from that, even if we didn't ultimately publish a paper on that. Yes, I loved having a little bit more interaction, and a little bit more collaboration.
I did have collaboration also at Perimeter; I don't want to be too down on it. Also, these things are also your personal mindset. I think it's clear; everybody has mental health issues. "Mental health issue" is kind of a loaded term, right? It's a term that carries certain baggage, so I don't mean to imply some of that baggage. But some of my introversion was due to my own choices, not necessarily due to the institutions or environments that I was in. If I'm choosing to only apply to one place and go there, probably I'm not primed to really be the best person to go out and do a bunch of collaboration! But I think it had gotten a little better by the time I had gotten here to Caltech, so I was a little more primed to have some better interactions.
ZIERLER: When you talk about a big swing, hitting for the fences, what was the frontier at that point? What were the big things to aim for?
FLAMMIA: Certainly any algorithm that had an exponential speedup was a big swing. That would be something big, at that time. There was a lot of emphasis on trying to understand phases of matter and trying to understand different schemes for computation involving surface code-like architectures. People tried to study things like string-net models, which had been around for several years, but trying to do computation with topologically ordered phases in an interesting way. That was something that was really interesting at that time. It's still really interesting. Trying to understand the stability of topological phases of matter, which like I said, Bravyi, Hastings, and Michalakis had contributed to—this sort of thing was really important because we wanted to know that the surface code was viable. From a computation perspective, we wanted to know that the surface code was a viable path to quantum computation. I think these sort of proofs are more fundamental and interesting than that, but I think trying to study surface code quantum error correction, and all the abstract theoretical questions that you're led to from that concrete practical question, those were all very interesting topics. Very broad abstractions like these string-net models; what can you do with that? Can you do fault-tolerant quantum computation with these things? Those were questions that I think were really interesting at that time, and I think many of those questions are still interesting. Some questions about Hamiltonian complexity were interesting at that time.
Some of the things that I was working on at that time I think weren't broadly viewed as being interesting by the theory community, but I think are more important nowadays. Just prior to that, I had worked on something—when I was a postdoc at Perimeter, I had originally done this work with David Gross and Yi-Kai Liu and others on trying to study tomography or estimating quantum states using something called compressed sensing. I was trying to find ways to learn quantum states in a more scalable, more efficient fashion. There had been some work on this by people like Scott Aaronson, who had done very abstract questions from classical learning theory to try to learn quantum states efficiently, but they don't give efficient estimators that are consistent estimators in the sense of consistency of an estimator in statistics. So, they redefined the problem, which makes it solvable, but not necessarily useful to the way that experimentalists wanted to use these tasks. Our approach with compressed sensing was trying to find ways that we could solve these problems that would be useful to experimentalists. I worked on some things like this. I worked on a paper with Yi-Kai Liu on trying to find a Monte Carlo method to estimate fidelity efficiently. I think some of this work has had impact. I'd like to think it has impact. [laughs] So, I think there was a theme at that time to try to find ways to just understand, "Is our quantum device performing as we think it is?" Those questions were a little more trying to bridge this gap between what the experimentalists were doing and what the more hard-core theorists were doing. That's kind of the niche that I found a lot of my subsequent work ending up in, actually, from that point forward.
ZIERLER: Now, the "M" that joined with IQI, did that register with you? Did you feel the experimentalists, more condensed matter—was that happening around you? Did you plug into that at all?
FLAMMIA: Yeah, it did start to happen when I was there. I had some great conversations with people like Jeff Kimble, at that time. Something else that happened around that time—boson sampling, I think, came out around this time, and I had some conversations with people like Jeff Kimball, who were like, "Here's a serious theorist who is proposing some experiments that we might be able to do, that are testing something interesting about quantum computation." So I had some conversations with Jeff Kimball at that time. It didn't wind up going anywhere at that time, and subsequently it turned out that random circuit sampling turned out to be a better approach, I think, than direct boson sampling, for various reasons. But this started to build some connections.
Other things that happened at that time—I think Oskar Painter was getting involved, for example. Now, of course, I talk to Oskar semi-regularly, because he is also involved with AWS. I didn't have any interactions with Oskar at that time. But I think there's a ramp-up period with these things, and I think I didn't have enough time to get to the top side of that ramp. I started to have some interactions, but it hadn't built up quite enough yet. We were still in the ramping-up phase. Some other people in the group who were naturally a little closer to experiment, I think they saw that ramp-up a little better, like Alexey Gorshkov, Liang Jiang, I think they experienced the upside of the "M" more immediately than I was able to. I definitely started to feel it, but couldn't take as much advantage of it as I would have liked to.
ZIERLER: Coming in as your second postdoc, was John Preskill a mentor, a collaborator? How would you interact with John?
FLAMMIA: I would interact with John mainly through our group meetings, and he would come to lunch to talk to the postdocs. My feeling with John is he knows that you've got to talk to young people, because the young people are the only ones who have enough time to actually fully immerse themselves in the literature. When you're a senior person, you do enough management that you start to lose a little bit of fine-grain touch with the details of what is happening in the literature. But the young people, postdocs and senior graduate students, they're deep into the details. So he would come to lunch and ask us what we were thinking about. Probe us. I would talk to him a lot at lunch. I would talk to him at the group meetings. I would occasionally go to his office and talk to him about some of the things that I was thinking about. But I could never interest him enough [laughs] to become a coauthor on any of my papers. He would always ask me some question, and then he would have that "w" shaped quasi-smile that he's kind of famous for, and he would ask some question like, "Well, why don't you sample it in this way? Wouldn't that smooth over the fluctuations better?" You'd say, "Well, I thought it would do this," and he'd go, "Mmhmm." And you'd go, "Oh man, he thinks my result is garbage! Oh, I better go back and think harder about what he just said." So, I don't know, that was mainly my interactions with John.
FLAMMIA: Maybe he perceives it differently; I don't know.
ZIERLER: What was the job market like?
FLAMMIA: Oh, it was terrible.
ZIERLER: Were you thinking about a third postdoc at that point, or what were your options?
FLAMMIA: No, my options were, "I'm really done at this point, so—"
ZIERLER: The pattern, right? It's like, "Perimeter or bust"; that works. It's "Caltech or bust"; that worked. So what's the next one?
FLAMMIA: At that time, Dave Bacon said, "I've got this grant." At this time, like I said, there are famously talented people who are drastically underemployed. The job market in North America is terrible, but it's starting to recover.
ZIERLER: Where is industry at this point circa 2012, 2013?
FLAMMIA: Industry is starting to make some investments. Google at this point had invested in a D-Wave quantum device, which D-Wave—say what you will about D-Wave; at the hardware level, they are trying to build devices that do something. I think a lot of people take issue with the way that they market the abilities of their devices. I am certainly sympathetic to that. But Google invested in that, thinking, "Well, let's see what they're capable of doing." They put some very smart and capable people on that, to try to see, "What can you do with this actual device?" IBM, I think, was doing some interesting work at this time as well. People like Jay Gambetta and Jerry Chow—lots of other people were starting to really build hardware at IBM. Sergey Bravyi was at IBM. Barbara Terhal was at IBM. Graeme Smith, John Smolin, others, were at IBM, and Charlie Bennett, of course. Who deserves a Nobel Prize, by the way.
ZIERLER: Yes, he does.
ZIERLER: After this one, maybe that paves the way for him.
FLAMMIA: Yeah, I hope so. I hope so. Yeah, so industry was doing really interesting work in theory, both at a pure theory level with people like Charlie and Sergey and others—Graeme Smith—as well as stuff that is quite closely connected to experiment. Jay Gambetta did these really landmark papers on randomized benchmarking, building on this work of Joe Emerson and others, to develop this idea of estimating incoherent noise rates in quantum devices. Jay Gambetta really pushed this at IBM, did experiments with it, and I think he was central, and he was in industry at that time. So, there were industry players getting involved. Microsoft was also of course involved, but Microsoft is sort of the exception that proves the rule. I don't really like that expression, but they were involved, but had this very different approach, so it's kind of hard to compare them.
So, there were industry efforts, but like I said, unless you were excited about smaller-scale experiments, I didn't find it super compelling to want to work on one-qubit randomized benchmarking at that time. It didn't seem interesting, because it was only one qubit. I said, "Call me up when you have 100 qubits. I don't even know what to do with a quantum device that has less than 100 qubits." So I was not really interested in industry. There were some jobs; they didn't seem super compelling. In retrospect, this was too short-sighted. So, industry was there, but it didn't seem like the most appealing career path. It seemed like you would be stuck working on some rather detailed things about device physics. Which of course I'm doing now! But it's in service of a greater—now I've seen the light.
ZIERLER: You have your 100 qubits now!
FLAMMIA: That's right, yes. So, that was the landscape. There were also other jobs, like national lab type jobs, as well. Backing up, famously underemployed people like Dave Bacon and Aram Harrow were working at the University of Washington on soft money as research professors, paying their own salaries out of grants that may or may not get renewed, on the whims of the federal government. This seemed very tenuous. They somehow talked me into moving there, actually, because my impression at the time was, "Wow, I'm going to go to Seattle. Never been to Seattle; great. I'm going to work with Dave Bacon and Aram Harrow, and overnight, there's going to be like three theorists at this place, and it's going to be actually world class, to have three theorists in one location." Which it was, circa 2011 or 2012 or so, to have three theorists in one location. So, I decided to move there, just because I actually thought, "Well, maybe, just maybe—I mean, surely they're going to put Dave Bacon and Aram Harrow on the tenure track. They'd be insane not to do that." But I don't think I appreciated just how much institutions take soft-money people for granted, even if those soft-money people are Dave Bacon and Aram Harrow. I think there are institutional headwinds to trying to do something like that. That's not to say anything negative about the University of Washington; I just think it's just a fact that you underappreciate people that are in those positions, and it's only afterwards, when they say, "Oh, yeah, now I'm going to this other place," that institutions realize what they actually had. Aram is now at MIT, for example. I think when Aram gets a job offer at MIT, Washington says, "Oh gosh, maybe we should have put that guy into a tenured position."
I think I was pretty naïve when I took that, thinking that maybe, just maybe, there was a hope that I could be an academic, someday, so I went there. It didn't really pan out. I think I spent most of that year thinking about what I would actually do when it didn't pan out [laughs] and looking for a landing. So it wasn't my most productive year, but I did enjoy talking with Dave and Aram, and the people in the CS Department there as well, and some of the students that have stayed, that graduated from there, and have gone on to careers, like Elizabeth Crosson and people like this. I enjoyed talking with them.
Faculty Life in Sydney
ZIERLER: What was the next move for you then?
FLAMMIA: Sydney! There weren't that many job opportunities in North America, still, although it was starting to open up.
ZIERLER: You're a bachelor at this point? You're unfettered and can go wherever you want?
FLAMMIA: Yeah, because when you move every few years, it's hard to have a stable relationship, isn't it? [laughs] It's the disadvantage of being in academia. I thought, "Let me do something totally different and exciting." I had been to Australia before, like I said, with Michael Nielsen, visiting in Queensland. I had been to this absolutely fantastic workshop that they have every year in Coogee Beach in Sydney that was organized by Stephen Bartlett and Andrew Doherty and others, where you're steps from the beach. The beaches in Sydney, I have to say, on record, are way better than the beaches in Southern California. There are smaller crowds. The water is warmer. They're just better. So, they would invite me down there; I think they were grooming me. [laughs]
When you're from North America, your impression of Australia is kind of like, "Well, there's just as bunch of poisonous animals down there. I don't really want to go down there!" [laughs] But once you've been there a few times, you realize, "Wow, it's kind of nice down here. I could see myself living here." I started to habituate to this viewpoint. Eventually, I was like, "Yeah, let's do it. Let's make this move." They had a job opening down there. Even that, it was still tenuous at this time, recovering from the financial crisis, so I wasn't initially even on a hard money position. I was on a soft money position with a promise, a written promise, that when the budget turned over in a subsequent year, I would be shoehorned in to a hard money position. [laughs] So, even this was tenuous, but I figured that was good enough for me, and it was exciting enough to just pick up and move to another place, that it was worth trying to do it, so I did it.
ZIERLER: Culturally, were there different approaches in quantum information that you got exposed to in Australia that you might not otherwise have had?
FLAMMIA: Yeah, there's a lot of quantum optics in Australia. I would say there's a lot of really good people doing optics in Australia, and then there's also people doing more solid state-based approaches, people like Michelle Simmons, Andrew Dzurak, Andrea Morello, at the University of New South Wales. There are other approaches. I mean, there's lots of other approaches. I am remiss for not mentioning other colleagues by name, but what I was most familiar with at the time was people, like Andrew White, who were doing optical approaches to quantum computing. When I moved down there, that was what I was most familiar with, and there was a bunch of that sort of stuff going on in Australia. I saw that happening, and that was pretty interesting. What I saw most exciting was to collaborate with people like Stephen Bartlett and Andrew Doherty, who were my friends at that time. I've written a bunch of papers with those guys. Also people like Gerard Milburn as well, although I don't think we've written any papers together, sadly. I think that's what I mainly wanted to do when I first moved down there.
ZIERLER: Could you have made Australia a long-term proposition?
FLAMMIA: I did. I mean, I was there until 2020.
ZIERLER: Full professor, as long as you wanted?
FLAMMIA: Yeah, I did. Like I said, I had this weird thing where they managed to convince their department that I was the guy to hire, but there wasn't money on the budget, so they said, "We're going to pay his salary first here, and then we're going to hire him," some weird shell game situation. [laughs] But it was effectively a hard-money position from the start. I was quite happy there. I started at basically the equivalent of assistant professor and was promoted through the ranks to full professor, and I could have stayed there indefinitely on a continuing position. It was the equivalent of a tenured position. I was happy there. I was very happy there. But yeah, a bunch of things changed, so I wound up moving back.
ZIERLER: What was some of your key research on the faculty?
FLAMMIA: I'd say a couple things that I'm most proud of while I was there are, honestly, supervising some really brilliant students while I was there, or co-supervising.
ZIERLER: How international was the crowd? Was it primarily Australians, or all over the world, Asia?
FLAMMIA: You have a lot of students from the rest of Asia. You have postdocs internationally. Many of your graduate students—the culture there is—in the U.S. at least, people often leave home at a younger age, I think, than in Australia, at least in the cities there. It is really expensive in Sydney, and a lot of people live still at home while they're undergraduates, and until they get a PhD stipend, they don't have money to leave home, so they're like, "Well, okay, I'm going to stay here, and I'm going to do my undergraduate and graduate school here. Then I'm going to go abroad." So we would get students from the Sydney metro area who would come to the University of Sydney, which is like a premier, definitely top I would say three institutions in Australia. The Sydney metro area is maybe six million people, something like this, so you're getting some of the top students in a pool of that size. That's almost as big as a country like Switzerland or Austria. So you get some really good students that come through and they decide to stay for their PhDs. You get to draw on that pool. That's mainly what we would get—some international postdocs, students from Asia who would come on PhD stipends, and the local crew, [laughs] students from the Sydney metro area who decided to stay for their PhD. These are very good students, of course.
ZIERLER: Just to foreshadow to your decision to join AWS, to go back to how unimpressed you were with single qubits, 2012 or 2013, were you tracking some of the industrial developments in real time that made this a decision a long time coming, or was this an out-of-the-blue opportunity? Let me add to that, 2020—you can't help but notice—where does the pandemic fit into all of this?
FLAMMIA: The pandemic features large in my decision, actually. I definitely took a lot of notice in 2014 when Google wrote a paper where they did error correction of bit flip errors on nine qubits. This was a really big splash. If I recall correctly, if you read the abstract quickly, you cannot tell that they didn't do full quantum error correction. But that's okay. I think the point is, it was pretty eye-opening that they were able to do this, even though it wasn't maybe as big as I remember my first impression. Nine qubits was a lot, at that time. You're like, "Wow, they actually fabricated a 9-qubit device and did something non-trivial"—like really non-trivial—"with this device. That's really impressive." It was pretty clear they were going to get some momentum from this, and it was also clear at that time that other players were serious about investing in this in the long term. So I think you definitely started to pay attention to industry at that point.
I think experiments got more and more interesting. It became more and more indispensable to read some of the experiments that were being done, increasingly so. That happened year on year, like, "I need to read about these experiments that are happening, because they're really doing interesting things, and fidelities are getting better and better." It's like, "Okay, I don't just have a 90% fidelity CNOT gate; I have a 99% CNOT fidelity." It's starting to look like, "Wow, okay, this is very impressive." Also, the goalposts are getting closer on the theory end, because you get more and more techniques where the thresholds get higher and higher, we're going to meet in the middle, and it's gonna happen. That was happening progressively while I was at Sydney.
ZIERLER: Did that plant a seed as early as 2014 that some of the really interesting things that were happening were happening in industry?
FLAMMIA: I don't know if it planted a seed that early. I was pretty happy there. Life is good in Sydney. I think it's hard to explain to somebody who has been in the U.S. their whole life, but—I mean, politics is dysfunctional everywhere, is my suspicion, but having some distance from the dysfunction of U.S. politics was strangely relaxing. I didn't have to worry about that.
ZIERLER: The Trump administration was a long way away.
FLAMMIA: That's right. I was in Sydney when Brexit was happening, and I remember ribbing my friend Ben Brown—"Hahaha!" And he said, "Yeah, but you guys have Trump." I was like, "That'll never happen!" [laughs] Oh, man. So [laughs], I think it really looked like I had the good life. You just get on a bus and you're at the beach. Awesome. I've got access to great students, great colleagues. I think it's pretty hard to get students of the caliber that we had at Sydney at most institutions in North America, so I was blessed to have students that I either supervised as an advisor, or as a co-supervisor, or did projects with them, people like Jacob Bridgeman, Chris Chubb, Dom Williamson, David Tuckett, Robin Harper. I'm forgetting names here. There were just so many great students that came there that I worked with. I apologize that I'm forgetting names. This was a real pleasure. Unless you get John's job, unless you're the Feynman Professor at Caltech, what students are you going to get that are better than them? It's really hard, actually. It's hard to get great students. I thought I really had a gravy train going, at that time. It didn't feel like it was the right time to move.
A Pandemic Move to AWS
ZIERLER: What happened, then, in 2020? Did AWS reach out to you?
FLAMMIA: Yes, AWS did. Fernando called me in December of 2019.
ZIERLER: You had known Fernando from your Caltech days?
FLAMMIA: I think Fernando joined after I had left Caltech, but I had known Fernando. We had talked a bunch at conferences. He knew me from my work, and I knew him from his work, for sure. I think he saw me as someone who would be a valuable contributor, as someone who knew a lot about this interface between measuring properties of quantum systems and sort of passing it up the stack towards the error correction layer. A lot of the work that I had done at Sydney at that time—the other stuff that I had done as well, if I can talk a little longer about Sydney—some of the stuff I had done, also with Ben Brown and others, was on studying biased noise, which originally had been done by John Preskill, actually. He did one of the first studies on trying to understand the role of noise bias, and whether or not you can use the bias of noise, whether or not phase flips are more likely than bit flips in a quantum device. Can you use this to raise the error correction threshold or reduce the logical failure rate in a quantum device? We had done a series of papers showing that this was a promising avenue, actually, and that maybe we should pursue this. I had also done some collaborative work with Shruti Puri and others at Yale. I had done a sabbatical. I was visiting Yale. My now-wife was doing her medical residency at Yale at the time, so I was living in New Haven—I forget exactly which years, but during the Trump administration, actually, so I got some of that stress back in the end. I think Fernando had seen this. He knew some of that work. He had known some of my work on characterizing noise in quantum devices, and he said, "This is a guy who might be able to contribute." He had contacted me in December 2019.
I had some personal circumstances. My wife and I had moved back to Australia. We had decided to transfer her medical license so that she could practice in Australia. The Australian bureaucracy was dragging their feet on approving her credentials, so she had been unemployed for months as a doctor. At some point if your medical license lapses, you're not a doctor anymore, so she decided to move back. Yale took her back and said, "We will hire you as a faculty member here at Yale, and you can practice medicine under your license. You can continue to practice." We had made the decision that we would live apart for a little while until we could work it out, and then she would move back. So we were committed, actually, to living the dream, in Sydney—"Let's do it." The pandemic, in January or so—when we had made this decision to live apart for as long as it would take, we had thought maybe six months. We made this decision in January of 2020 or so. [laughs] I had told Fernando, "Yeah, I'm kind of interested. Tell me more. I guess I'll interview for the position just to see what's up. Tell me more. I'm not going to say ‘no' right now. I want to hear you out." Because maybe I can see on the horizon that this might become an issue if her license never gets approved. I might need to move back, so let me hedge. So, I interviewed. It was a very interesting offer. Great.
ZIERLER: Compelling? Giving up a full professorship in a city that you love?
FLAMMIA: It's hard to give up a full professorship. That's right. Also, you've got it made. You become good at it by then. I know how to write a paper. I know how to pick research projects that are going to be attractive to my colleagues. I know how to write papers. I know how to supervise students. I know how to get funding. I know how to do the boring things associated with being a professor, like sit on a committee. I can do it. It feels good to be successful at something. It feels good to be competent at something. It doesn't feel good [laughs] to not be good at something and to struggle. But that's how you grow. It's tough to make that choice. I saw it as risky to move to industry and to start from scratch, so I saw that as a big risk, but it was attractive.
Also, like you said, you start to see this tide building, all these brilliant experiments coming out of industry, all the money going into industry. It seems harder and harder to have an impact in a purely academic setting when unless you're hiring at a huge rate in an academic setting, how are you going to keep up with that? It seems challenging. Then the craziest part of all this is coming on a personal note. My wife—like the very last day of February, 2020, she flew back to New Haven, and something like three days later—so the headlines were starting to build in the news about this whole COVID thing is a thing—maybe three days after she got there, she told me, "I'm pregnant," [laughs] which we subsequently found out was twins.
FLAMMIA: Yeah, it got even crazier. I think within two weeks from that point, they were talking about closing the borders. So, I needed to make moves really, really fast. We had bought a house and everything down there. I basically said, "Yeah, Fernando, I'm totally taking this job." [laughs] "Sydney, I quit." Basically. They said, "Yeah, how are you going to teach your course?" I said, "I'm going to teach it over Zoom from New Haven, and I'm going to teach it over Zoom between the hours of like 12:30 a.m. and 2:00 a.m."—or something—"from New Haven." And that's what I did. I got on literally the last possible flight out of Australia. I basically threw the keys to some property manager person and said, "Please try to mail me my stuff." Our dog—we had this German Shepherd, beloved family pet—flew on a different plane to me to a different airport. I didn't even know if they were just going to euthanize the dog on arrival because I couldn't even get there. I flew to Hawaii, which was the only way I could get out of Australia. I thought they might quarantine the island. It was totally crazy. I managed to get on a connection, after maybe five different rebookings, to land in Dallas. Then I eventually drove to New Haven. [laughs] It was totally insane. It was all happily ever after in the end, but I took this job under somewhat stressful circumstances. But, I should say, it was an exciting circumstance. I genuinely wanted to take the job. But my hand was absolutely forced in a pretty crazy pandemic-inspired way.
ZIERLER: Because you had to make this decision so quickly, in an alternative universe, if you had more time to consider, do you think if Fernando was expressing interest that you would have seen, "What about Microsoft? What about Google?" Would you have done sort of a broader tour of industrial approaches to quantum computing? Or, was AWS and Fernando just so compelling, because in Caltech, that it just made it all the more easy, right then and there?
FLAMMIA: I've got to say—[laughs]—Fernando isn't sitting behind me right now, I should say. He's not sitting behind me, but I have to say, to some extent, it was the latter.
ZIERLER: It spoke to you?
FLAMMIA: It did. I feel like if I had knocked on other doors, I feel like people would have been receptive. I'd like to think so. What was most compelling to me about AWS was getting in on the ground floor. They were starting something new. If I had joined IBM or Google or Microsoft, I was concerned about being a cog in the wheel. It's like, "Okay, Steve, you're here. Now you're going to work on x, and you're going to be part of this."
ZIERLER: Which is less academic, in some regards.
FLAMMIA: Yeah. Well, I want impact. Really, in my world, I'd like to think that my work matters and that I'm having an impact. I think it is harder to do that if I am not calling the shots. I really got a sense, by getting in on the ground floor, I would have not just impact in terms of my scientific expertise, but strategic vision. When you're a more senior person, one of the things that you develop through years of experience is strategic vision, and an understanding not just of, "Well, I'm going to write this paper, and this is the next paper in this series of papers about this topic," but you say, "Why am I doing this? How am I going to get to my ultimate goal? How do I do this, and how do I plan for this?" And it's even bigger than that. The phrase "logistics wins wars"—quantum computing is a marathon, and you have to have supply lines to feed the huge numbers of people that do that. How do you do that? How do you take the talent that you have under you, and put them on problems, and drive them to solving those problems and then integrate that?
When you're a full professor, you have a huge amount of integrated expertise as somebody who can solve concrete technical problems but also manage those things, and then also build a program which has a five or more year vision, and then integrate those threads. This was a skill set that I had developed, which I felt would be lost if I had just become a cog in a wheel somewhere. What I was promised was, "You know what? You're going to be here. You're going to be on campus. You're going to be on Caltech. You're going to have these connections to all the brilliant science that's happening at Caltech. And, you're going to have the ability to get in on the ground floor with this planning and vision for this totally new approach. It's going to be funded to the tune where you're going to be able to compete, longer-term, with all the big players. Go!" That's really compelling. That's really hard to say "no" to. In the end, I probably would have made the same decision. Who knows? It's counterfactual. But it was really compelling. I think it was very hard to say "no." In a no-pandemic world, if I had said, "No, I'm going to stay a professor," that would have been the lazy, safe, risk-averse choice. I would like to think that I would have made the risky choice and come here.
ZIERLER: COVID helped you along, though.
FLAMMIA: [laughs] COVID helped me! Oh, god. I don't know if COVID helped me, but it forced my hand, and I think for the better, for that specific thing, yes.
ZIERLER: You talked about the ground floor at AWS. What was it? What was that ground floor? What was the beginnings of the project?
FLAMMIA: You want to build a quantum computer, and you don't even have a building yet. You start from nothing. You have to build up all of this. You have to build—everything. You have to hire all the talent. You have to build a lab, which means a huge amount of control electronics, dilution refrigerators, everything. You have to write the software that controls the hardware. You have to write the software interface that lets somebody like me write something like "CNOT from qubit 1 to qubit 2, enter." [laughs] All that stuff. It's huge. It's an absolutely enormous amount of stuff that you have to build up. Then you have to polish it. Getting in on the ground floor is building all that stuff, but also developing the direction that it moves. "Am I going to take this approach? Am I going to build a bog-standard surface code, or am I going to try to tune it in some way?" Like I had mentioned this work that I had done on biased noise—I had written this paper. The lead author is now a graduate student at Harvard. He was an undergraduate at Sydney at the time, and I had helped supervise his project, along with Ben Brown and Stephen Bartlett, David Tuckett. Juan Pablo Bonilla Ataides, a brilliant guy. We had this code. Other people had looked at this model before but nobody had really studied it in the context of error correction. The XZZX code, we were calling it. I had this vision, like we can actually—we had done studies of this code and showed, if you just tune things in a really simple way, you can get achieve fault tolerance sooner. So, I really thought, "I am going to influence the way in which we actually develop these quantum computers." Everybody was like "surface code, surface code"—the bog-standard surface code that had been proposed years ago by Bravyi and Kitaev, building on Kitaev's idea for the toric code, here at Caltech. We had a very, very tiny incremental change on this, but it had a big impact on quantitative numbers, and I thought, "These are the kinds of things that I can contribute to the vision." I wanted to do this sort of thing. That's what I mean by ground floor.
Measuring Progress Toward a Quantum Computer Breakthrough
ZIERLER: Steve, for the last part of our talk, we'll measure, in this compressed time frame—it's only two, two and a half years after this whole thing started for you—what at AWS are the big obvious areas of progress that you can see so far, and what might feel as brand-new as when you were in building mode in early 2020?
FLAMMIA: I'd say the amount of infrastructure that we've built is huge, by which I mean, like I said, all of these things like building labs, setting up dilution refrigerators, hiring brilliant people, writing software. All that stuff, we've done. It's like building roads. Now, we have networks of highways, we've cleared whole forests, and things like this. That stuff, there has been tremendous progress on. It's not very sexy, though. The stuff that is still relatively nascent is things like building a small number of qubits and doing demonstration experiments for things like quantum error correction. We don't have any published work on things like this at the moment, and I think it's fair to say this stuff is relatively nascent. I think if it weren't nascent, we would have already published it. If it were already state of the art, we probably would have published it. I don't think I'm giving any secrets away by saying that. So, it's relatively nascent. But it's early days. We have only been in it for two years. Our competitors have been in it for many years longer than that. But we're learning so fast. There are some advantages to starting late, because you don't have to make the same mistakes. It's sort of good to make mistakes; we're making different mistakes, I think, and we're learning faster than our competitors did. So, we'll catch up. We will catch up. And that's exciting.
ZIERLER: In terms of your sense of time scale and keeping your own attention fresh, these are things that could occupy you for as long as you want them to?
FLAMMIA: Yes, I think so. If I look at the recent experiments that have come out from other industry leaders, they're really interesting now. They're scientifically interesting things that they are doing. I think we're a laboratory on the cusp of the revolution where the types of hardware that we're building are going to be independently scientifically interesting, and it's at that point that we've crossed the threshold, from my perspective. We're not necessarily commercially viable. Okay, fine. We do have to worry about that, as employees of companies. But, my perspective as a scientist and as somebody who is interested in fundamental questions and who wants to have impact scientifically, we're crossing that threshold right now. That's really, really exciting. And we're going to do that. I see that in the papers that are being published today. I'm really excited that I'm alive at a time that I can contribute to that.
ZIERLER: How do you measure that excitement relative to recruitment? The best graduate students and postdocs who might otherwise consider faculty positions at your Caltechs, your MITs, Harvards, are they thinking about AWS as a place where they can do their best research in quantum information?
FLAMMIA: Some people are. Yes, definitely. I think the people who are extremely mathematical, maybe not, because I think it's a bit hard to monetize some of that. I do think that some companies try to have a bit of a budget set aside where they can have a few people thinking on blue-sky things. Microsoft does this, for sure. It's fine for them to bankroll some really brilliant minds—Jeongwan Haah, Matt Hastings, people like this. Those people also do practical computations as well. Not every company invests in that way. I think right now, our portfolio is less focusing on that, although we do have some people thinking about more mathematical things as well. I think people of a more mathematical bent might seek an industry job, but there are very few of those around, so I think they tend to want to think, "I'm going to go on an academic track."
People who are really interested in quantum computing specifically, I think these jobs are just very compelling right now. The salaries are better. You get to talk to a lot of really brilliant colleagues. I think before, the industry efforts were relatively small. What's your payoff if you take an industry job ten years ago? You might have three other colleagues, and you can't freely talk about your work with all the other colleagues. What's the benefit? Now you've siloed yourself. But now, I've got so many colleagues that I can talk to, all the time. I can name all of my brilliant colleagues at AWS that I talk to on a day-to-day basis. There are so many of them that even if I were siloed there, which I'm not, it would be a very rich intellectual environment. Of course, couple that with being here at Caltech as well; it's great to be at AWS. I think even though Fernando isn't behind me with a gun to my head saying it, I have to say, there's a really strong pitch to want to work at a place like, in particular, AWS, but also other industry places as well, to be fair to them.
ZIERLER: Last question, looking to the future. In the way that you're always going to want your research to have that impact, you want that intellectual engagement, you want the best of both worlds of industry and fundamental research—
FLAMMIA: [laughs] Yeah, I do!
ZIERLER: —what are the benchmarks—one year, five years, ten years, twenty years, however you want to define those benchmarks—where you're hitting all of these areas that are going to keep you happy and satisfied at AWS? What does that look like?
FLAMMIA: Oh, man. I would love it if we can publish some papers soonish on our hardware. I think that is probably the next milestone that I would like to see us meet. The bar is so high now for industry-led hardware papers. There are also academic efforts that are producing really spectacular efforts in things like quantum error correction. There was a paper by Andreas Wallraff within the last year where he did quantum error correction in a three-by-three surface code. That's a fantastic result. If you're putting out results that aren't at that level, I think as an industry PR person, which I'm not, but I'm imagining what they might be thinking, you don't want to put out work that looks subpar to state of the art. But those labs, like Andreas Wallraff's lab, has been working on this stuff for many, many years, and is huge and well-funded. So you don't necessarily expect to put that out tomorrow, when you've started. I do look forward to putting out something that is publishable from our hardware soon. I can't speak to where we are exactly at right now—I'm not authorized to do that—but I'm looking forward to doing that. That's a big milestone that I'm really looking forward to. Then, once we're doing that, it's going to be a huge win. I can't wait to celebrate that day. Then probably the next milestone is to do something—that first thing isn't necessarily going to be the most scientifically interesting thing, but it will be interesting from an engineering standpoint and from just this milestone of, "We're doing serious hardware."
The next thing is to do something scientifically interesting. One of the experiments I mentioned was Google's paper on preparing a topologically ordered state on a quantum computer. Something like that I see as being scientifically interesting. I think that is the next milestone that I look forward to us meeting at AWS. People are already meeting it, so that is the state of the art, in my opinion. I think we will get there soon. Beyond that, the real next milestone is fault-tolerant quantum computing, to really demonstrate something where you actually have an error-corrected logical qubit, using the principles of fault tolerance, where your entire error budget is actually decreased relative to your physical hardware. It's at that point that things really get interesting. I'm not sure when that's going to happen, but I don't see any fundamental reason why we can't achieve that, and everybody is working towards that goal now. I think people are really aligning on that.
John really coined this term "quantum supremacy." The real quantum supremacy is not random circuit sampling. It's not boson sampling. I personally feel like those things are a bit distracting. They don't solve any interesting problem. Our theory community is somewhat fetishizing them, because we can answer those questions. The real quantum supremacy is building an error-corrected logical qubit. That's the supremacy. I think a lot of industry efforts are aligning in this way, and that's what I really look forward to. I'm not exactly sure when we'll get there. I'm super excited to achieve that.
ZIERLER: It's so significant that it's worth remaining engaged even if this breakthrough happens right around your retirement party?
FLAMMIA: Oh, yeah, totally. I will die with a smile on my face if it happens right before I die. I will be thrilled.
ZIERLER: Steve, this has been a great conversation. I want to thank you so much for doing this.
FLAMMIA: Thanks a lot, David.
- The Caltech and Amazon Quantum Connection
- The Potential Uses of Quantum Computers
- The Quotidian Task of Building a Quantum Computer
- Quantum Metrology at the University of New Mexico
- From Perimeter Institute to Caltech
- A Pandemic Move to AWS
- Measuring Progress Toward a Quantum Computer Breakthrough