Hello, and welcome to the Physics World Weekly podcast. I'm Hamish Johnston. This week, we're celebrating the winners of the Physics World breakthrough of the year award for 2024, and we have 2 podcasts for your listening pleasure. This episode is supported by the journal Reports on Progress in Physics, which offers unparalleled visibility for your groundbreaking research. This year's award is all about error correction in quantum computing, and we are honoring 2 independent teams.
I've spoken to the lead researchers of both groups, and we're presenting those conversations in 2 different episodes. In this podcast, I'm in conversation with Mikhail Lukin and Dolev Blufstein at Harvard University, who, along with their colleagues, have implemented quantum error correction on an array of trapped atomic cubits.
In a second podcast, I chat with Google's Hartmut Nevin, who leads a team that has made a major breakthrough in implementing quantum error correction in a processor that uses superconducting qubits. In principle, quantum computers can solve some problems that cannot be computed on conventional processors.
However, the quantum processors available today are very susceptible to disruption by environmental noise, and this destroys the delicate quantum states that are used to store and process information. When quantum computing was first proposed, some physicists thought that this problem was insurmountable. But thanks to the development of quantum error correction, practical quantum computers that can solve useful problems could soon be a reality.
This year, we awarded half of the 2024 breakthrough of the year to Mikhail Lukin and Dolev Blufstein, and colleagues at Harvard University, the Massachusetts Institute of Technology, and CUERA Computing. And that's for demonstrating quantum error correction on an atomic processor with 48 logical cubits. I'm very pleased to have Dolev and Mikhail on the line from Cambridge, Massachusetts. Welcome to the podcast, and congratulations on achieving a remarkable breakthrough in quantum computing.
Hello. So, Mikhail, I think my first question is for you. What is quantum error correction, and why is it necessary? Thank you, for having us. So I have to answer this question, I will maybe go back a little bit to a history of, quantum computing and quantum information, which, kind of the early ideas are now by now, maybe 40 years old or so.
And, when people started realizing that, you can use the ideas of quantum superposition and quantum entanglement to build new quantum information processing systems, From the very beginning, there was a concern whether you can actually realize such systems in in practice.
And in particular, maybe around, like, 30 years ago where some of these early ideas started solid to solidify, for example, ideas of quantum simulations, the ideas of Shor's algorithms were put forward, people really started asking, you know, can quantum computers be built practical quantum computers be built be built and how? And already at that time, it was very clear. It will be a very challenging task.
And, in particular, one challenge, I would say, is conceptual is that, it's very hard to put, a big system in a quantum superposition state. So at the kind of microscopic level, like single electrons, you know, single, you know, spins, you know, can be put routinely in a superposition state. But the big objects around us, you know, like, you know, this this span or this table, you know, while they're composed from quantum mechanical particles, you know, there is nothing quantum
about them. Right? So big systems lose quantum character, and this is really fundamental. So in the, field of quantum computation, this specific example, this, issue shows up is that if you start building, for example, quantum computer and you build it from some kind of quantum logic operations, inevitably, there will be a small errors that, you know, will happen during this quantum logic operations.
And these small errors eventually accumulate to make this, you know, the output completely classical. So it basically loses all quantum features. And, because of that, kind of early on, there were, kind of, there was a lot of excitement on one hand about quantum computers, but there was also a lot of skepticism. Right? And so and in particular, if you start looking at what type of error rates you need to really implement, you know, some, you know, interesting quantum algorithms at scale.
These error rates are, you know, extremely low. You know? Well below 1 part per billion. You know? So and, for this reason, I mean, there was a little you know, some of the skeptics were actually very kind of prominent, you know, I would say, leaders on the field. We want you know, thought that, you know, while these quantum computers is really a furious dream, it's experimentally's nightmare. You know, it's really literally kind of impossible. It's impossible to reach such low error
rates. And, and, already early on, there was an idea. Oh, okay. So in classical, computers or quantum classical information processing systems, you can often use some redundancy to, protect quantum information. And, you know, people started asking a question. So can you use this redundancy to basically preserve quantum information? And the answer, even conceptually at the time, was really, non, obvious, because, quantum information cannot be copied.
Also, if you measure so to, like, utilize it done to see classically, you basically need to, measure state to verify where errors happen or not. In quantum mechanics, you basically cannot measure the state without collapsing this. And, you know, for this reason, it kind of seemed challenging that even in principle, you could use error correction to protect quantum information. Nevertheless, about, you know, 25, 30 years ago, there was theoretically shown that you can actually use,
this kind of redundancy. You can actually use entanglement to, store and protect quantum information. And I would say it was really this breakthrough that really kind of jump started this field as we know it now. I see. And, Dolev, an idea that's central to quantum error correction are the concepts of a physical and a logical cubit, and and that's something that was important in your research. Can you can you explain why quantum error correction considers physical
and logical cubits? And and what are they? What what are the differences between the 2? Error correction is this really remarkable process, and, it uses physical cubits. And when we say physical cubits, we mean things like ions or superconductors or, you know, defects or spins, things that have two levels that are like a quantum two level system.
And we have been studying these for roughly, like, almost a 100 years since the beginning of NMR, where people were using these for, you know, various different applications. And, the invention Misha mentioned from 30 years ago was the fact that you can actually put these together to make a logical qubit.
And, the way that it works because you can't copy quantum information is what you do is you take this abstract unit of information, which is our logical qubit, and you use entanglement, these, you know, funny quantum correlations between particles to take this one logical cubit degree of freedom and delocalize it. You spread it across, you know, a large array of physical cubits. And now what happens is that this delocalized information is protected.
Now, you know, if the environment comes in and tries to measure part of the system, if it looks at just one of the physical cubits in the system, it actually will not learn anything about the underlying stored delocalized state. And so that is what the physical mechanism is behind logical cubits, and that's how we take we use entanglement between physical cubits to make robust logical cubits, and it is truly remarkable that it is, you know, physically possible.
That being said, it is clearly a completely different object than these physical cubits are. These physical cubits are these two level systems we've been working with for, you know, almost a century. These logical cubits are these, you know, highly entangled states, and is in in a sense an abstract unit of information. And now that we're starting to work with them in the lab, we see that there's many, many differences to the physical cubits that we're used to working with.
And, perhaps one of the, you know, most key features there in terms of the difference between physical and logical cubits is that in a physical cubit, it's a quantum two level system. You can take this cubit and it can be in you know, you can just rotate it and put it in any state that you want. And that is, you know, for example, what people do when they do NMR is you have, you know, a spin and the spin can process, and it can be in any, you know, point of this, you know, qubit's phase space.
But with logical cubits, it's very different. You can't do things like arbitrary rotations like we do in NMR. You can only do digital operations where the logical cubit and the still localized information can only do things that are exactly the operations that are allowed by this logical qubit. And in a sense, it starts to make quantum information processing more digital.
And this also should parallel, you know, what we do with our classical computers, with classical computers, all of our information processing, the reason it's so robust is because it's digital. It works on bits, and the bits do, you know, precise logic operations that are insensitive to, for example, voltage fluctuations in your computer.
And we have analog classical computers as well that are, you know, can do arbitrary rotations, and these things are much more similar to the physical cubits that we've been working with in the field, before. But now that we're working with logical cubits, we're really starting to explore digital processing, very similar to how we have digital processing with
our classical computers. So there's various, you know, very important key differences, between these physical and logical cubits that we're starting to explore now. I see. And, Mikhail, can you can you explain or describe the the physical qubits that you use in your lab? Well, I suppose, in this specific bit of research that you and Dovlev have done. What what does the, what does the physical cubic comprise?
These experiments utilize, isolated, trapped, neutral atoms, actually, rubidium atoms, which are individually trapped and held, in optical tweezers and tightly focused beams of beams of light. So to kind of take a step back, you know, this, of course, one of the many platforms that people are now exploring. So the reason why, neutral atoms is a promising platform is because, first of all, they
have excellent coherence properties. So while isolated and held in tweezers, you can basically store quantum information for very long time for tens of seconds or kind of minutes. It's kind of almost like unlimited, you know, in increase in principle. And that's one important, feature. For example, atomic clocks, you know, some of the most precise precise instruments, scientific instruments that humankind ever built, you know, utilize now neutral atoms, trapped neutral atoms.
So and the second thing which is also very important is that in print in principle and also in practice, you know, you can, create a very large, number of this kind of, you know, of the, atoms. And, basically, I should also point out that our experiments make use of this so called laser cooling and trapping techniques. So they're actually room temperature systems, but their atomic motion is slowed down by essentially buffing the atoms in a beam of,
beams of light of certain color. And as a result, they basically come to the standstill. So each of our experiments starts with a cloud which contains, you know, basically many millions of tens of millions of motionless atoms, which are, of course, fantastic, you know, initial, step, you know, to create, a lot of qubits. But the challenge now is not just to create these qubits,
but also to control them. And that's why we use this kind of optical techniques, the techniques from holography, to basically create a large, number of tweezers and trap the atoms there. There is one other challenge, and that is the atoms in the gas phase basically don't interact with each other, don't talk to each other. So in order to entangle them, to make quantum logic, we actually use lasers to promote the atoms into the so called
Rydberg states. So the Rydberg states are the states where, electrons orbit, you know, very far away from the nuclear. So so the atoms, in this state, say, basically, can be thought of having kind of a large size. And then as a result of that, they really start interacting very strongly to each other. And in particular, to entangle these atoms, and, do quantum logic, we utilize something which is called Rydberg blockade. So that's the idea, which is, by now almost 25 years old.
But it kind of, in the recent years, really proved to be kind of remarkably fruitful. And, and the the key idea of this Lydberg blockade is that you can basically consider 2 atoms, which you excite to the Rydberg states. If they sit far away, then, essentially, you can excite them independently. But when you bring them close to each other, then what happens is if one of the atoms is excited to the Rydberg state, the excitation for the second atom is blocked.
And what this mechanism does, it basically makes the interaction of the atoms between the atoms almost digital. So if they're far away, they don't talk to each other. If they're close to each other, the interaction is nearly infinity. And that actually allows us to, entangle, the atoms, with very low errors. And in particular, it also enables us to entangle many pairs of atoms in parallel. So that is something that we will maybe talk
about a little bit later. So, basically, these optical tools combined with this Littburg blockade is what enables, you know, high fidelity or in other words, low error rate, you know, parallel, control and entanglement of 100 of atoms at a time. You know? And this is an essential ingredient which actually allows us to do this kind of logical you know, build these logical processors kind of in a very efficient, way. And can I just ask you, Mikael, how how big is this cloud
of atoms? Just to give our listeners an idea of the size of of, I mean, I know you've probably got a huge vacuum chamber, etcetera, but the actual cloud. The actual you know, the size of the processor, you know, which traps these atoms, you know, and basically allows us to kind of manipulate, the this this atoms is a couple of 100 microns. You know, it's actually, you know, relatively small.
But even this even this, zone is actually kind of divided into small parts, smaller parts, and one of them is like a storage zone. Another one is the zone where we do logic and entangling zone, and another one is on the zone where we do kind of readout. And, actually, one other innovation now going to 2020, which actually really fueled with this development, was actually led by the left, was to realize what we call, the configurable, architecture.
So and to explain it, you know, like, let's think about how the conventional, you know, chips, you know, semiconductor chips. So, you know, you what you do, you know, you design this chip, you design this, you know, transistors, you design that, you know, connectivity, and then you basically, you know, send it to, you know, to to the, you know, factory, you know, where basically, the state of the art technique uses optical
lithography. Basically uses optical tools to define where this chip these transistors are going to be. And then eventually, this, this, you know, things, are made. But, basically, the point what I want to make is that the connectivity, the architecture of this chip is fixed at a design stage. So what happens is that using optical tweezers, we can actually move atoms around, and we move them while preserving the coherence,
while preserving the stored qubit. So this is done by encoding qubits into so called hyperfine states, basically spin states of atoms where they can live for a very long time. And most importantly, you know, by moving the atoms around, you can basically create the, the architecture where the connectivity is like a living organism. It changes during the computation itself. And it, for example, allows us to move atoms between these different zones. It allows us to,
you know, entangle atoms in parallel. You know? And it allows us to basically implement all necessary ingredients, for the logical cubits. So, Dov, in in your work, you created 48 logical cubits using these physical cubits that Mikhail has, described. How how did you do this? I mean, is it possible to to describe it in simple terms? How how you take these physical cubits and sort of blend them together to make logical cubits? Yes. Absolutely. So maybe there's 2 two stages
of answering your question. So one is, like, on a physical level, how does this even happen, like, on a quantum mechanical level? And then the other one is, how did we make how are we able to, you know, really simplify the problem to make it much easier than it has been historically in the field?
And, the, to answer the first one, it's we're we're leveraging the fact that we can entangle particles by moving them around entangle these atomic cubits by moving them around and zapping them with the laser pulses when they're next to each other to entangle them.
And, to create these, logical cubit states, similar to what I was describing earlier, we take this, you know, like, one qubit, for example, or a collection of qubits, and then we entangle it with its surrounding atomic qubits in a very structured way that spreads out this information. That creates this logical cubit once you create this entangled state. Now you have to do multiple important things to it that we were
able to explore in our work. One is that you, you know, can now do logic operations between these logical cubits. And, that's actually one of the things that's the hardest to do. When we take, you know, this logical cubit and then spread it out across an array of physical cubits, And we might do this now on 2 different blocks of qubits, and we'll have 2 different, you know, delocalized, degrees of freedom.
They're now protected from their environment because now the environment can't come in and measure this underlying state. But now it's also very hard to get them to interact. And actually in the field before, people had done quite, you know, nice work in creating logical cubits. But the really huge challenge was always getting them to interact Because now there are these, you know, just completely delocalized degrees of freedom.
And for example, imagine that you have these 2 delocalized degrees of freedom, and they're stuck next to each other on a chip or something with a fixed 2 d connectivity. But now it's very hard to get them to interact because they can only interact through some boundary, whereas they're, like, delocalized over space.
So with our ability to move cubits around, what we can now do is we can actually pick up the 2 logical degrees of freedom, put them right on top of each other by interlacing the 2 grids of atomic cubits, And then by entangling all of the pairs of the, you know, underlying logical blocks, that realizes, like, a logical entangling operation. So that is how we do both the creation of the logical cubits as well as their operations. And, that is something that it was an extreme simplification.
So one is it now by doing these gates in this way that we call transversal, where we can take these 2 degrees of freedom and interact them directly. It's an native logical operation that can just be directly done. The other thing that's really special that is related to what Misha said is it now allows us to start controlling things in much simpler ways.
And in particular, you know, modern quantum processors have almost exclusively been built in this way where you have several cubits and you just add more and more cubits and you add more and more controls to control each cubit. But one of the big innovations in this work that made this so much simpler is that once we're starting to do error correction, all of the physical cubits within a logical cubit just need to do the exact same operation in order to realize a logical operation.
So now in this zoned architecture that we're describing, we can also work with all these logical cubit blocks as if there's just one big atom essentially. And we take this one, you know, big atom and put it next to one other big atom and do this entangling logical operation in a single parallel step, and then can go and move these, you know, big qubits and interact them with other
big qubits. And that was one of the things that was really central for us to be able to create such a large number of logical cubits and explore different types of interesting algorithms, with them. We were able to create 48 of these small logical cubits using these approaches and do 100 of logical operations, whereas in the field, people had previously only done 1
or 2. Before this work, we were also able to study things such as improving logic operations as we increase the size of the error correcting code and study a lot of really key features of what does error corrected quantum computation look like due to the fact that we can, you know, do this abstracted, control where we're working with logical cubits as the fundamental units of this processor. I see. And and, Mikhail, you you've got these 48
logical cubits. Are you able to to actually do practical calculations with your system? Is there are there problems, you know, maybe even trivial computational problems that you can solve using it? Or is it very much a sort of proof of principle system? So maybe we'll answer this question in 2 parts. I will start, and then I'll let
Alef complete my answer. So, and, you know, to answer it, I maybe want to make a step back and, you know, mention that in addition to kind of building, you know, large scale quantum computer, another big challenge in the field is to identify what can we use these, you know, devices for. You know? How can they really help humankind? And you could say, well, I mean, it's kind of a, you know, funny question to ask for this field. So active and so on, but this is not
unusual. So when I renew some new tool comes, you know, into play, you know, people often, you know, have hard time anticipating where it's, you know, going to be most, useful. But there is one area where it is very clear that this quantum computers, and quantum simulators, will have tremendous value, And that is in modeling and simulating, systems which have high degree of entanglement. And these type of, you know, systems occur in various areas of science, of physics in particular.
Certainly, you know, many of the condensed matter models, feature, you know, so called strongly correlated systems. You know? You know, feature, you know, high degree of an or expect I expect it to feature feature high degree of entanglement. Another area which is actually also very exciting, is, involves, simulating system where you can build entanglement very quickly. And, this is actually, very interestingly connected to the physics of black
holes. You know? So people believe that black holes, at least, you know, a kind of their quantum description, you know, involves this process, involve, you know, fast scrambling. You know? This is where you basically, you know, create entanglement kind of in the fastest way possible. And so one of the experiments we have done is actually exploring this kind of fast scrambling. And maybe I'll let Daleyf add because, you know, he's he was a mastermind of this, you know, specific. You know?
Yeah. So following up on that. So maybe I will also take a step back about this quantum scrambling and say that one of the biggest open challenges in physics is we do not understand how quantum mechanics and gravity combine. It is, in my view as a physicist, one of the most interesting open questions of our time. And that is actually one of the places where quantum computers can almost certainly be very useful.
And remarkably, we don't fully understand how quantum mechanics and gravity combine, but one of our best guesses in terms of how this might arise in our universe is that there's entanglement on some boundary in our universe, and the entanglement on this boundary gives an emergent gravitational description of the universe, which is amazing. And, however, it's really hard to make progress on some of these types of really complex quantum questions
without a quantum calculator. We only have classical calculators, and we're trying to calculate these extremely complex things about the universe. So what we did here is a really, you know, like, toy study of those types of things, but we were able to, you know, study this complex scrambling. In particular, we entangle everything on hypercubes.
One of the things that's special about hypercubes is they're very, very connected and that scrambles information very rapidly, very similar to a black hole. And, we didn't learn anything here about, you know, emergence of gravity from complex entangled systems, but it does start to give us a bit of a hint in terms of, you know, how we can, you know, what types of systems can we simulate with these logical qubits.
And, if I can just nerd out for a second, one of the things that we did, which was really, you know, very tailored to this logical cubit processor here, is that once we're so we've we've had, you know, several decades of exploring, you know, processing with physical qubits. And physical qubits have very particular rules that we're used to following. And one of them is that, you know, it's very easy to do arbitrary rotations of cubits like we do in NMR, and it's very hard to do entanglement.
In these, you know, first error corrected algorithms that we were doing with these logical cubits, one of the things that we saw is it's very hard to do arbitrary rotation, but it's very easy to create entanglement. So it actually very well suited to something like this black hole scrambling.
And this was just one example of us doing quantum simulation with these air corrected cubits, but it starts to open a new scientific frontier of exploring how to do quantum simulations and quantum computations, in ways that are highly tailored to these, you know, new set of weird rules that we have to work with with logical cubits. And so we're not yet doing any practical calculations that are, you know, curing cancer or
completely changing things like that. But what it is very clear is that in the near term and in all already what we've done and also in the near term, there will be a real value in just learning what these quantum computers can do from these types of experiments. So it's practical in that sense, but, yeah. Oh, that's great. I mean, I have to say, I wasn't expecting that you'd say black holes. So that's that's great. You learn learn something new every day. So what's what's next, for you guys?
Mikhail, what what what do you have planned for the future? Are you are you going to try to create more logical cubits in your system? Or is there some other avenue that you can pursue to, sort of, to to gain your understanding of, of this quantum computer system? Yes. Certainly scaling up this, you know, quantum, you know, computation is definitely very much in our agenda. And, I would say that, you know, definitely one would like to have more logical qubits. You
know? But most more importantly or equally important, one would like to actually improve these logical qubits. Right? Because, you know, even if we encode information, you know, at least up to now, what we and others have done is encoding offers some protection, but it's not, you know, I mean, it's not perfect. So we would like to actually, you know, make this logical qubit better,
and then reduce error rates. And the key goal here is really to start, doing computation, which have computations, which have deeper circuits. So, I would say, you know, over last year, you know, in addition to the work that we have done, there was also some very nice experiments from
across several different platforms. So for example, the very recent work of by Google, actually, demonstrated just one logical qubit, but what they have done is they basically, did experiments which involve multiple cycles of error correction. Alright? And it's kind of it's a little bit like scaling up if you want them on different axis as compared to what
we have done. So but, but in reality, what one needs to do is one needs to really combine these two things, you know, to basically start implementing kind of, you know, deep circuits involving large number of of logical qubits and eventually try to, you know, figure out, you know, how to answer, you know, your previous question in different ways. You know? So the question is, what can we do with this kind of systems, basically?
And, what emerged from our work is that answering this question would really real has to rely on this idea, which we sometimes call codesign. So if you have if you want to solve some specific problem. So what you like to do or you would need to do, basically, you need to think about algorithm to solve this problem, kind of codesigned with first error correcting code, which really fits this problem very well together with decoder, with compiler, and eventually with your hardware system.
So this is it's very clear to us that for the next, you know, 5 years, maybe in the next decade, this is a way to make progress. And we are really excited about kind of starting to put these things together to really kind of, you know, you know, build, you know, systems and and come up with more examples where you can really, you know, enable, like, deep circuit kind of useful, quantum computation with, you know, large number of of qubits. Maybe I'll let Dolesv.
Do you have anything to add to that, Dolesv? Yeah. Absolutely. So, yeah, I would say that last year, we learned a lot about how to do error corrected algorithms. Now, you know, it's a really important frontier to learn how to do deeper error corrected algorithms, and improve the performance. There is really exciting progress happening across the field, both, you know, between, you know, neutral atoms, trapped ions, who are gonna think you, but people are really starting to experiment with
these systems. And I think one of the things that has become extremely clear in 2024 is that error correction is, you know, definitely works. And also, it is currently a real inflection point in the sense of you really start to get below characteristic thresholds in the system. You really start to come up with creative ways to do logic operations. And it really the field is now going to start transitioning toward doing algorithms and computations
and simulations with error correction. And that is going to be a, I think, a very dramatic inflection. And so that, in my view, is extremely exciting. I mean, even just in the past few months alone across many different systems, there's been really remarkable error correction progress. But there is, however, one pretty huge elephant in the room, which is that for a lot of the computations that we have in mind, we need things at the scale of tens of millions of cubits.
And we are working to reduce that number, but the main thing is that we are not yet close to that. We do have ideas in terms of how to get there. But currently we're working with systems that have hundreds of cubits at the most. And so, there will be many challenges in trying to get to these much larger systems. But I do think that the field is going to develop in a very different way than it has in the past. And I would say it's for 2 key reasons that are emerging now.
One is that error correction clearly works, and I think that is really starting to be at an inflection point that it was not nearly at the same level 2 years ago. And 2 is that when we build error corrected processors, we can build them differently than we're used to building physical CUDA processors. And so we are not yet close to our end goal. Although, of course, the goal will always evolve. Although we can, you know, start to explore
interesting science in the meantime. But I also think that things are going to start developing across all these very various different systems more rapidly than is being expected because of these two key changes that I think will both be key inflections. So, yeah, many challenges, but also very exciting. Oh, well, that's great. Well, thanks. Thanks, for coming on the podcast. This is one of 2 podcasts with our breakthrough of the year winners.
The other features Google's Hartmut Kniven, and you can find it on the Physics World website or at your favorite podcast provider. You can also read more about our top 10 breakthroughs of 2024 on the Physics World website. This served as the shortlist for our breakthrough of the year, and it covers a range of fantastic research in physics. So do check it out. I'm afraid that's all the time we have for this week's podcast.
Thanks to Mikhail Lukin and Dolev Blufstein for joining me today, and a special thanks to our producer Fred Iles. Physics World's coverage of the breakthrough of the year is supported by Reports on Progress in Physics, which offers unparalleled visibility for your groundbreaking research. You can find the journal at iopscience.i0p.org. We'll be back again next week.