Quantum networking – like quantum computing – holds tantalizing promise. Rather than use classical bits, quantum networks work with quantum bits (qubits) and are expected to deliver completely secure networks, play a key role in linking quantum chips to scale up into quantum supercomputers, and enable practical use of networked quantum sensors. Big ambitions. The Department of Energy, for example, even released a blueprint strategy for building a nation-wide quantum Internet back in 2020.

Obviously, we’re not there yet.

Earlier this month, AWS held a Media Innovation Day at its AWS Center for Quantum Networking (ACQN), showcasing several AWS technology areas, but focused on the company’s quantum networking work. The ACQN opened roughly one year ago, and is based in the greater Boston area where it has a close collaboration with the Harvard Quantum Initiative.

That AWS has a quantum networking effort isn’t surprising. Antia Lamas-Linares, head of ACQN, said at the event, “A lot of people think of AWS as providing compute and data and storage and all of these things. But it’s actually one of the largest networks in the world. All of these components are connected by a huge infrastructure of fiber optics. That’s a big part of what makes AWS such a powerful infrastructure.”

In fact, like most of the giant cloud provider crowd, AWS is making a big on quantum beyond its popular Braket quantum services and third party QPU access portal. It also has a solutions lab working with customers on use cases and applications, a center for quantum computing doing QPU research and quantum system development, and the ACQN. (See HPCwire coverage, AWS Takes the Short and Long View of Quantum Computing)

“In quantum networking, we’re primarily concerned how to move qubits around. How do we transmit, manipulate and store quantum information and use it with applications? We call them flying qubits,” said Lamas-Linares, “They typically are photons, because photons are good at traversing long distances. They also tend to be, naturally, very well isolated from the environment. So they’re great carriers of quantum information. Here at the ACQN we develop the technologies for quantum networks. They are not all fully baked. Many of these technologies have been partially demonstrated in academic labs, but still need a lot of development we get to a fully-fledged quantum network.”

Like quantum computing, quantum networks seek to leverage superposition and entanglement and faces many of the same challenges, such as qubit fidelity and environmental noise. They also must deal with distance. Sending a qubit – photons are the preferred entanglement carrier here – over a long distance while maintaining its state is a major challenge. Interfacing with quantum computers using other qubit types – superconducting, trapped ions, etc. – is another challenge.

No surprise, developing reliable repeaters is a major goal for quantum network developers, and it turns out AWS is making significant strides forward in that area. Likewise, effective quantum memory is an important part of these repeaters; there must be a way to capture individual flying qubits, in this case single photons, and hold them still and coax them to interact with other qubits in a way that doesn’t break the finicky laws of quantum mechanics (no cloning!)

Currently, the limit for optical quantum network is roughly100 kilometers. The largest quantum networks use satellites to cover a bit more than 1000 km. An example of a commercially deployed “long” network is the HSBC (London-based bank) network at 62 km. AWS recently reported successful demonstration of a QKD network in Singapore.

Brief backgrounder:

The simplest quantum communication task involves connecting two parties on different ends of a network. Yes, we’ll call them Alice and Bob in line with convention. Each sends a qubit (actually many) encoded in a photon. Leaving aside the protocol, the flying qubits can only go so far before being lost to the environment. Repeaters are needed to capture and transfer the entanglement (state) of each incoming photon into a memory. This process involves transferring (not cloning) the state of the qubit to memory, and then transferring to another photon for the trip down the next leg of the journey.

There are, of course, many other details and steps generally taken. Alice and Bob usually have shared information known only to them about their qubits. While it’s not possible to “read” the qubit without destroying the content, it is possible to use other techniques – joint measurement is one, for example – that produces classical information which tells Alice and Bob something about the other’s qubit without leaking information about the communication to eavesdroppers. It is also possible to swap entangled states. Repeaters and their memory perform these functions.

The choice of memory is an important step. AWS is placing a big bet on silicon-vacancy diamond for memory. It is one of the so-called color center choices in which impurities can be added to diamonds. These tend to change the diamond’s color – hence the name – but more importantly, the impurities can act as qubit memory. Swapping in a silicon atom for two carbon atoms in a diamond crystal creates a situation in which the silicon atom’s quantum state can be changed by interacting with an incoming photon – transferring the information from the photon to the silicon-vacancy diamond. AWS is also investigating other memory options, but is furthest along with silicon-vacancy diamond.

As you might guess there’s a whole lot more to making quantum repeaters work. In a follow-on briefing with HPCwire, Lamas-Linares dug a little deeper into AWS quantum repeater-memory development, touched on advantages offered by silicon-vacancy diamonds and more. (Thanks as well to Mihir Bhaskar, research lead at ACQN, for additional input)

HPCwire: Maybe we should start with Amazon’s current preference for silicon-vacancy diamond memory. What driving that?

Lamas-Linares: The answer is nuanced. There are a few promising platforms out there for memory. One of the most generic and most promising is color centers and diamonds. So, there are a few types of color centers. One of the things that we were looking at is the potential for scalability. Silicon vacancies and diamond have some properties that make them likely to be able to scale, to be integrated into fabrication processes that are that are already better understood. Just to give you an example, another very promising and very commonly used color center is the nitrogen, nitrogen vacancy (NV center). That’s another color center that has a lot of work being done and it’s very, very good for quantum sensors, for example.

But we think that the silicon vacancy has some really good points in terms of its potential for integration into other fabrication processes. At the end of the day, your quantum device is going to have to be integrated with other things, and the performance that you’re looking at is the system performance, and not necessarily the just the quantum performance. That’s not to say that we don’t look at other things, we keep an eye on other things, but you need to make a bet on one. You can’t work a little bit on 100 different things.

HPCwire: How does the photo capture process work to get it into memory?

Lamas-Linares: That’s actually a big part of a big part of the effort here and why we like the silicon-vacancy in diamond method. If you just send a photon towards a diamond that has a silicon vacancy, the probability that it gets captured is really small. How do we increase that probability? What we do is essentially trap the photons between a set of mirrors, so it’s like a like a funhouse mirrors, [and] we just work the photon many, many, many times across the vacancy. [Even] if the probability is really small [on each individual pass], when you go a few hundreds of millions of times, it starts to get significant.

So that’s what we built. Of course, the mirrors that we built are not macroscopic mirrors. They’re mirrors that are fabricated into the diamond. We fabricate a mirror structure in the diamond and that’s a big part of this whole effort: how easy is it to fabricate; how easy is to guide the light so it enters this mirror perfectly controlled, how reproducible is that process? That’s a lot of the engineering that we’re trying to do.

HPCwire: What kind of light is used? Is it the visible light spectrum or something else?

Lamas-Linares: There are two sort of schools of thought or two possibilities, if you like. You can use telecom photons. So just like the ones that go down fibers, typically 1500 nm, that works well. Or you can use the visible photons, so right at the edge of visible and near infrared, so red photons around 800 nanometers. Each approach has advantages and disadvantages. Telecom photons have the great advantage because there’s so much optics, commercial optics well developed for telecom. That’s a big, big plus for that. On the minus side, the detectors are harder. It’s harder to do single photon detection in the telecom, they’re noisier and so on. Visible photons, they have better detectors, they’re a little bit more energetic, things that make it make that a little bit easier. On the other hand, commercial optics is not nearly as well developed for visible photons. So that’s still being debated.

HPCwire: Talk a little about the repeater-memory complex. What are its main function and how does it do them?

Lamas-Linares: One of the things that you want to do with quantum repeaters, these quantum networking devices, is called a two-qubit measurement. You need to take two qubits and do a joint measurement on them. This is a uniquely quantum mechanical thing. But you can think of it as similar to a parity measurement. If you have two qubits, you don’t want to measure them directly, because if you measure them, you collapse the state. But you can, if you have two of them, ask a different question. You can ask things like, are you pointing in the same direction? Which is not asking which direction you’re pointing it? But you’re saying, are you pointing in the same one, right.

These are called joint measurements and are at the heart of what constitutes a quantum repeater. To perform these joint-measurements we take the flying qubits – these qubits that are encoded in a photon and traveling at the speed of light – and we need to hold them still, literally, and get them to interact. Photons don’t like to interact with each other. So that’s hard. There are some tricks, but it’s one of the reasons that the flying qubits are so good [for networking because] photons don’t interact a lot with the environment which protects them. That’s great for traveling. That’s hard for interaction, right?

So first, we need to absorb the photon into some material qubit that keeps the properties. And we need to get two of them, two photons, one from one side and one from the other, keep them still and perform the joint operation. One of the key things that the repeaters allow us to do, is to do this operation asynchronously. A photon comes from one side, gets held in a quantum memory, and it can stay there for a little while until another one arrives from the other side. It gives us many chances to try because most photons [or] a large fraction of photons will get lost in transit. If you depended on two photons arriving at the same time to do this, it would be extremely hard or very low probability. But because you have a quantum memory, you get to try try, try, hold, try, try, try, try, try, many times, depending on how long that memory lasts. And now you can perform your operation, and this allows you to route and to connect different spokes of a spoke and hub kind of infrastructure as well. Does that make sense?

HPCwire: It still sounds a bit like cloning or copying. Is it?

Lamas-Linares: In cloning, you make a copy and you keep the original. You’re not allowed to do that, right. If you tried to quantum clone, you end up with, with two imperfect copies that do not have the fidelity to the original. I did my thesis on this, if I get excited, please forgive me. What we are doing in this case is transferring the state from one kind of qubit to another. But the original photon gets destroyed or all the information that was contained in that original flying qubit disappears. So, you’re transferring the information from one qubit to another qubit.

Quantum mechanics sort of keeps track of all of these things. The information just transfers, and the original is left in a sort of blank state.

HPCwire: What about the entanglement between the incoming photon and its entangled partner when at the other end where it was sent out? Is the entanglement between that original qubit and its photon partner back at the start of the journey disrupted?

Lamas-Linares: It’s not, and that’s good insight. That’s what we what we mean by transferring the information coherently. It means all of all of the correlations and all of the information, whether it’s contained in that qubit or it’s contained in the relationship of that qubit to other qubits is properly transferred down the down the line into the memory. So the diamond qubit, it would be entangled with whatever was left behind by the first qubit so you still just have the two qubits entangled.

HPCwire: This is what’s meant by distributing entanglement.

Lamas-Linares: That’s right. You can do even do really fun things with this. You can do something called entanglement swapping, and that’s one of the main roles of repeaters. Imagine that you have a repeater in the center, and you have Alice and Bob (commonly used fictional characters in a two-way quantum communication). You have Alice and Bob and then you have a repeater in the center. You can share entanglement between Alice and the repeater. You have something shared between Alice and the memory in the repeater. Then you share an entangled pair between Bob and the repeater.

Now you have these two members of a pair of entangled photons held in the repeater in the memory. Now you do this joint measurement that I was talking about, and you can transfer the entanglement all the way to the edges. So now you have entanglement between Alice and Bob. And there’s nothing left in the repeater. There’s no information left, nothing, which is at the core of why we say we can use a repeater to distribute entanglement between any edge of the network, and also why we say you don’t need to trust the repeater. Because no information is left in the repeater. Nothing. This is related to this no cloning thing that we were talking about; there is nothing left behind and you can very rigorously prove that.

[Editor’s Note: There’s a lot going on here and its best to review the approach more rigorously. Two papers covering the ground well are 1) Experimental demonstration of memory-enhanced quantum communication, Nature, and 2) Cryogenic optical packaging of nanophotonic devices with coupling loss < 1 dB, arXiv preprint. AWS’s Mihir Bhaskar is actually the lead author on the first paper. The second more recent paper is joint work from AWS and Harvard.]

HPCwire: The actual transmission is a classical thing. It’s kind of what we’ve been doing all along, correct?

Lamas-Linares: Yes. You just send you a single photon down a fiber and off it goes. The problem is, we are not allowed to use those very convenient amplifiers that the classical optics people do, right. Their signal gets attenuated and when it gets too weak, they just stick in an amplifier and boost it back up. We can’t do that because that would be a quantum cloner. So we need to use these other techniques, like repeaters, which allow us to subdivide a very long transmission into small sections that we can treat a synchronously and with entanglement swapping operations and these kinds of things.

HPCwire: So each new photon qubit is able to reliably travel the transmission distance between repeaters. Is it because it’s a new fresh photon?

Lamas-Linares: No. All photons are the same thing. In some sense. It doesn’t matter whether it’s fresh or not, because it’s a single photon. Unlike the classical light, where it’s a field that just gets reduced in amplitude. A photon is kind of like a particle, it either gets absorbed or it doesn’t. Although we talk of the repeater as being functionally equivalent to an amplifier, it’s not really the same. So the image is not, I get a signal that is weakened, you know, a photon that is weakened and we amplify it. The photon is the same. If it arrives, it arrives. If it doesn’t, it doesn’t. What the repeater does, is it allows us to break the long transmission lines into very small sections, and perform these operations that move the (quantum) correlations to the edges of the network, I realized that’s not very easy to explain or to understand, but functionally it allows us to go farther, but it works in a in a different way.

HPCwire: That suggests the distance between repeaters is just the distance at which the photon has the most likelihood of not being lost.

Lamas-Linares: Let’s think of the scenario that I did before. I want to distribute entanglement between Alice and Bob and they’re pretty far away. If I tried to distribute it directly, the probability that the two photons make it to both Alice and Bob is really small. Let’s say, for the sake of discussion, this wouldn’t be so small, but for the sake of discussion, let’s say that the probability that Alice get the Alice gets the photon is 50% and the probability that Bob gets the photon is 50%. So each time I send an entangled pair, the probability that they both get there 0.5 times 0.5 or a quarter (0.25). You can see how that scales pretty badly.

If I put a repeater in the middle, and I distribute entanglement between Alice and the repeater, now it’s a much shorter distance, so maybe the probability that both Alice and the repeater get the photon is maybe a 90%. And I do the same for Bob. So it’s 90%. Now, I don’t need both photons to succeed at the same time at arriving at the repeater, because I can hold them in a memory, right? So now the probability that I’m able to do this entanglement distribution via the repeater to Alice and Bob is 90% times 90% or 81%, which is higher than 25%. That’s the principle. By being able to perform these operations asynchronously with the help of a quantum memories, I’m able to greatly increase the probability of success at the end.

HPCwire: The quantum memory is clearly key. How long can you store a qubit in a diamond memory?

Lamas-Linares: I may get this wrong, but I believe it’s milliseconds now, which for light propagation is a long time. We have plenty of time in the memory.

HPCwire: We’ve not talked much about application and goals for AWS’s quantum effort. Quantum key distribution is maybe one nearer-term example. What’s your view?

Lamas-Linares: Quantum key distribution is an application of quantum networks, it’s something you can do if you have a very simple quantum network; key distribution does not require a lot of the more sophisticated things. A QKD network is not necessarily able to do things like entanglement distribution. It’s a bit of a lower tier thing.

Back to your sort of, to the core of your question. What we’re aiming to do – if you had to describe it in one single statement – is to be able to distribute entanglement between any two points, and those two points might be at the edge of the between edges of the network or it might be between the network edge and the core. Or it could be between quantum resources that are part of the network like quantum computers and quantum sensors. That’s the core capability we’re aiming for now. Achieving that is still fairly distant.

HPCwire: Are you working with the AWS quantum computer group as part of their efforts to scale up? Many believe scaling up quantum computers the millions-of-qubits size required for fault tolerant quantum computing will require networking smaller systems together.

Lamas-Linares: Yes, we’re working with them, but it takes a while for the technical roadmaps of quantum computing and quantum networking to converge. They’re very focused right now on quantum error correction, which is one of the big next things that needs to happen in quantum computing. So right now, they haven’t reached the limits –  I don’t think anybody has – of what can be held in a single fridge. Until you reach that limit, and you’ve done what you need to do, you might look into quantum networking, but it’s not a critical capability yet.