Vijoy Pandey, who leads Cisco’s Outshift incubation group, and Reza Nejabati, Cisco’s head of quantum research, are interviewed by Yuval Boger. They discuss Cisco’s strategy to enable distributed quantum computing and a future “quantum internet.” Vijoy and Reza explain why Cisco won’t build QPUs, focusing instead on room-temperature, telecom-band entanglement networking. They discuss Cisco’s aim is to make networking a “no-brainer” and push modality complexity to the edge via transducers and even entanglement-mode conversion, the argument that scale-out plus scale-up will reach “million usable qubits” faster, and much more.
Transcript
Yuval Boger: Hello Reza, hello Vijoy. Thank you so much for joining me today.
Vijoy Pandey:It’s a pleasure to be here, everyone. Yes, great to see you.
Yuval:So Vijoy, who are you and what do you do?
Vijoy:Yeah, so I can spend hours on who am I, but I think in the context of Cisco and Outshift, I run this internal incubation engine called Outshift for Cisco. And the idea is for this engine to go after adjacent spaces to the markets where Cisco already is in. So if you think about Cisco, you think about a networking, security, observability, and collaboration player. And so we take ideas all the way to customer traction in spaces that are adjacent to Cisco, both in terms of personas, user and buyer, as well as in terms of timeline and technology. So quantum being one such space and generative AI being the other space.
Yuval:And Reza, how about you? Who are you and what do you do?
Reza Nejabati:Thanks for having me. My background is academic. I have been most of my career, almost all of it, a university professor in the UK. So I used to lead quantum networking activity in the University of Bristol and be a chair professor there. Then doing more than a decade, near two decades on the quantum research, I made a decision now is the time. Let’s commercialize. I’m making something really impactful. I said, okay, where is the better place than Cisco? And there was an opportunity in Cisco so I can cash in on the competency of the big talent that they are there. And they know how to build a network. They made the Internet a reality. So I decided to join Cisco and they had the opportunity. And then to basically make my dream come true to build a real impactful quantum network with useful application. Yeah, I’ll have a company. Some shares in Cisco and quantum lab, which is in Santa Monica.
Vijoy:Yes, Reza leads all of our quantum efforts, and he’s driving both the hardware side as well as the software side. So we’re looking at it in a systematic way, and he’s leading all of that.
Yuval:So the lab is in Santa Monica. I was about to ask that. So why did you choose quantum? What made you choose? I mean, there are so many other things that are probably adjacent to what Cisco is doing. Why quantum?
Vijoy:So like I mentioned, I mean, so Cisco is primarily an infrastructure player. But then we also think of Cisco as distributed computing, distributed systems player. because the networking piece, the software piece, the security piece, the observability piece, all of these are in the service of enabling distributed computing. And so we think of ourselves as a distributed computing infrastructure player. And as I mentioned, the original internet is something that we helped create. Of course, there are many, many other players in the conversation as well. And so we are thinking about what next. And so what are the internets and what are the white spaces that we need to go after? And what are the big problems, some big board problems that we need to go after, as an infrastructure for distributed computing player. And one of them is, of course, going after agentic computing. And we just launched an Internet of Agents thesis. But the next wave, beyond that, is the quantum internet and distributed quantum computing. So if you think about our thesis and our belief in this space, is to go after scale out. And if you think about quantum computing, we are all looking at scale up systems in the quantum domain, and we don’t want to get into the quantum computing per se business. We’re not in the business for making QPUs. We want to build out networks, secure networks, the software and the protocols that enable distributed quantum computing and the distributed internet, quantum internet to happen.
Yuval:And is the focus indeed computer to computer communication over a quantum link, or do you see it also as a, say, sensor to computer?
Vijoy:So it’s both. And so I would say we are building the fundamental technologies that are at the hardware component layer. There are the protocol layer and algorithms layer, and there are the software control plane management layer and the compiler layers. So the entire stack from hardware to software that enable both. the distributed quantum computing, so within the data center, how do you scale out quantum computing, and bring some of those timelines that people have been talking about for a while, bring some of those timelines a little bit closer to reality, because scale out always wins compared to scale up due to a variety of reasons, but also the same building blocks actually help or create the quantum internet, which not only connect distributed quantum computing islands together, for quantum sensing together and have applicability into the classical world as well today. So it’s the same building blocks that we’re looking for quantum computing, sensing, as well as classical use cases.
Yuval:You mentioned scale up versus scale out as someone who works in quantum in the QPU vendor. At the moment, I think that we and other players in the industry think that scale up is much easier than scale out. So how soon before scale out becomes a practicality in your opinion?
Vijoy:So I think, I mean, it’s a fair question. And the way we think about it is, and Reza, you can jump in as well. The way we think about it is it’s easier for a QPU vendor to think about scale up because it’s a closed system, closed environment, and you’re basically making that environment bigger and bigger. And yes, you are there, I mean a bunch of these QPU providers, and we talk to a whole bunch of them. They are all in the same mindset and it makes sense. So we are not arguing against that. But as that vector improves and has the QPU providers scale up, we want to provide the fabric that also connects these scale up systems to then provide scale out. So a combination of those two vectors, scale up and scale out, will get you to pragmatic scalable quantum computing faster. That is the belief. So it’s not one versus the other. It’s both together that we think is going to bring in some of those timelines and make pragmatic computing of a million usable qubits much easier to achieve. The other two things I would throw in here is scale out is not just about scaling. It’s also about operability. And so if you think about scale up systems at certain points in time and you’ve seen this time and again from history, history teaches us, that scale up systems are harder to upgrade, migrate, manage, deal with downtimes and SLOs. So if you do scale out, it’s actually easier to operate those kinds of environments because you have redundancy, built into the system in a very different way than you would have with a scale-up system. So that’s the second bit. And the third bit, whether we like it or not as vendors, but customers prefer choice and customers prefer interoperability. And there are two things here, especially in the quantum domain, there are two things here that we need to be careful about. One is the technologies that power QPUs are still early stages. And so we don’t know which winners will actually emerge. maybe there’ll be many winners based on the different kinds of use cases. So we want to stay agnostic to that, and there’s that level of interoperability, but also from the customer point of view, you need interoperability between different kinds of even similar technologies. So a scale-out system also enables those attributes.
Yuval:Reza, I think we’re here, or the trigger for this conversation, was a chip that you guys developed and released. Could you tell me a little bit about the chip and what its main design, goals and attributes are?
Reza:So, I mean, as Vijoy said, we wanted to connect quantum computers together. And obviously, if you want to connect quantum computer and your objective is to scale out your qubit capacity, means that you want to run your quantum algorithm or quantum circuit in a distributed way. So it means that you need to be able to your quantum processor to, your quantum processor, to have an entanglement relationship between each other. And what we decided to do that, okay, first thing we need for our network, a source of entanglement resources that can basically pair with these quantum computers and entangle them together and allow them to do distribute quantum computing or what we call it teleportation of the data or telegating or teleportation of the gates. and so to do a two distribution quantum computers. Now also, we were, as Vijoy said, we were very cognizant that we want to build a network that is also from CAPEX and OPEX make sense to operate it. And we want to also cognizant that we are talking about the quantum qubits or they are very fragile or very, prone to loss and all the effect of the network. So we wanted to create something that is a small, is a low power, works at room temperature, and generate a large number of the, or high rates with a high fidelity entangled per photon. So that’s how we came up with our design, which is a quantum entanglement source based on the ring resonator. The ring resonator concept is not a new concept by itself, but the chip design and the way that we are arranging and multiplexing multiple ring resonator allows us to do a very powerful entanglement source, which is a very high rate, is tunable and let us to basically create the network the way that we would like to create them and make a sense.
Yuval:And I think you’re probably aiming at a certain wavelength for this, right?
Reza:Yeah, that’s another thing. We wanted to use the fibers and we wanted to to operate it, a standard fiber, which is used, and we also operated at the lowest loss band. So we decided to our chip to make it to operate at telecom wavelength, which is 1,500 nanometers. And also it can operate on other frequencies if you want to, but that’s the frequency of choice that we have.
Yuval:Vijoy, when I was listening to you earlier, I think you mentioned the word compiler. How does a compiler fit into this picture?
Vijoy:So this is, and maybe Reza will jump in a little bit more into detail here, but I think there’s something that we are currently working on, so it’s not out there. We have not talked about it. But the idea, again, is that if you’re thinking about a heterogeneous system, and to further the point that actually Reza was raising earlier, we believe that the networking part of this should be, a no-brainer. So it should be room temperature, telecom frequency, existing fiber, low cost, and take that with a grain of salt because we are not pricing people yet. It’s not a product. It’s a resource prototype. So when I say low cost, it’s not an advertisement for how much it’s going to cost in the market, but it’s relatively low cost, right? And so those are the goals we are shooting for because what we want to do is make the networking piece a no-brainer. And so similar, and all the complexity of handling the different modalities of these QPUs, we want to push that to the edge so that that complexity is in one place and the network is simpler. And so there’s a similar thesis on the software stack side. What we want to make sure is that while you build out the compilers and the applications on the QPU side, as you start bringing in the network, we want that piece to be as simple as possible. So when you think of it as a distributed system, you think of distributed compilation. You think about protocols to manage things. You think about a network controller. So network awareness is a thing that we want people to just forget about and make that thing so simple, but also enable the benefits that come with it, but in a simple API-driven manner. So the thing that we like to say is we are moving quantum computing from the physics heavy domain to a system’s heavy domain, to a very software-heavy API-led domain. So that’s our thought process. Let APIs play a role. Let quantum computing come into the modern world in so ways.
Yuval:So if I were to describe it simplistically, let’s assume I have a circuit that needs, say, 500 qubits, and I’ve got a computer with, say, 250 or 256 qubits, or I have two of them, the compiler will help me determine which half of the circuit to run on computer one and which half of the circuit to run on computer two. Is that about right?
Vijoy:Yeah, that’s about right, and there are various ways of tackling that, and as I can talk about a couple of them.
Reza:So what we want to give you, or we are developing now, is pretty advanced stage. And let’s go a little bit. If you have a 1 million qubits requirement for a really do a, let’s say, drug simulation discovery or a 10 million qubits. And you know that you want to do your infrastructure planning and based on, let’s say, QuEra quantum computer. You know the roadmap is for 2030 or 2027, what is that? And let’s say they are coming with X,000 number of the qubits. So how many of these processor you need to run your circuit and how they need to be connected together and how you can optimize your algorithm for that? So we allow with our compiler to algorithm to a circuit developer to do a code design. So they optimize their algorithm to reduce the number of the QPU they need or play with how they are connected together. Also, we are the only compiler when we make it available. I mean, our compiler is the only one that has this capability because it works based on the network. You can rearrange the topology of the interconnectivity between your processor on a subroutine basis in your quantum program or quantum algorithms. So you can, this is something that is impossible now. Everything is fixed and you run that. So you can do a very in-depth optimization or algorithm to reduce your network. amount of compute to increase your utilization. So that’s the whole idea behind our compiler to let you really plan if you want to run your real circuit, real algorithm, how does it look like and how does it look like the infrastructure?
Yuval:And you mentioned interoperability between different computers. So as you know, some computers operate on superconducting qubits and others use lasers or photons, or atoms, who does the translation between whatever modality that computer uses to the input to your chip?
Vijoy:Yeah, I mean, I can start, but I think, so the way we’re thinking about this is those are, so think of this as the equivalent of the nick card at the edge. And so that is a hard problem, as you can imagine, it is non-trivial, but, you can, it’s like a device driver problem. So you think about a system and you have varying degrees of complexities that you’re adding into that distributed system. And so you tackle the complexity of translating from your domain to that uniform domain at the device driver layers. So what we’re doing right now is we are working on our own, we’re working with a bunch of startups, we’re working with some academic institutions to work on the different modalities exists out there and then work on those transducers that help us convert from those modalities to this fabric that we’re trying to build. So it’s exactly like a device driver problem in the sense that A, it’s complex, B is going to happen one by one and based on the priority that we see out there in the field and talking to customers and is non-trivial and you need to keep up with that all the time. But the good way to think about this is that you are pushing that to the edge so that the rest of the fabric stays intact. Otherwise, you’re just dealing with all the family.
Reza Nejabati:If I add to that, there are two problems if you want to connect quantum computers to a network together. One is the frequency translation, as Vijo mentioned. So we are working on the transducer or interfaces for a microwave, neutral atom, and charge atoms to our photonics network. So that’s one thing. But the other thing is important. Imagine you want to connect, even more interesting, two different breed of computing platform. So not only frequency become important, but also modality of your entanglement. They work with different modality of entanglement. And our network has a capability to convert from one mode of entanglement to another mode of entanglement. For example, from polarization to time-bin from time to polarization or frequency vice versa. So stay tuned. Very soon we have an announcement how that conversion works on the hardware. It’s a new hardware is planning to come out.
Yuval:What do you do about distance? If I ran a telecom link from Boston to San Francisco, it probably goes through a couple of repeaters. What happens in your network?
Reza:At the moment, our network is we are focusing to design a data center scale. so it’s a short and repeater-less. However, we have some use cases that not quantum computing, but they are looking to use our entanglement network. We can do with our technology at the metro scale up to 100 kilometers range. We have tested that, that works. Beyond that, there are other consideration depends on the use case than the rate they need. However, if you go beyond that, you need a repeater. Now, the interesting thing is that we work on the data center, and repeater is a very small quantum computer, if you like with a memory and a small quantum computing capability. So with a transduction capability and work that we do on interface with quantum computer, we believe the same technology once is mature enough, we can use it as a basic building block for a repeater and extend it.
Yuval:So if I listen to the specs, and say, wow, this is amazing. Room temperature, 1 milliwatt, 1550 nanometers, transducers, compilers. I say, okay, great, where can I buy one? Or how soon can I buy one?
Vijoy:Yeah, so that’s a great question. This is why I was saying earlier that the low cost is just a technologist statement here. It’s not a Cisco product statement. But it’s a fair question. And I think the way you think, about this is businesses need to be pretty large for Cisco to then scale out. And so we are at the point where, yes, we have a research prototype. We have a software stack that is brewing. Like Reza said, we’ll be announcing a bunch of those software capabilities as well pretty soon. And what we’re working with is a set of pretty big design partners. And we don’t call them customers because it’s not a profitable scaled out business yet. It’s a is a design partner. So we’re working with some very large design partners on certain use cases, both in the quantum realm as well as what is truly exciting to us is that the same technologies that we are building are relevant in some classical use cases today. So things like decision coordination or time synchronization and security use cases. And so those are use cases that are valid today, here and now, which is what a company like Cisco likes to hear, is like, yeah, there’s money here today. So we are working with some pretty big design partners then. Once we prove those use cases out and we see a roadmap which requires us to go at scale, that’s the time I think we can talk about general availability. So in the meantime, we are building these things out. They are available for quote unquote purchase, but only for design partnerships that we want to prove out certain use cases.
Yuval:And you’re right, if I put on a marketing head, time synchronization or sensing maybe much more mature markets, relatively speaking, to computing. So why are you describing this primarily as a computing chip? Why are you leading with computing? Is that because it’s just a bigger market ultimately?
Vijoy:No, so when we talk about this, we talk about this as a network entanglement chip. And when you see our narrative, you’ll see that we want to build out the quantum internet and we want to build out the distributed quantum computing environment. And we are coming at it from the networking and security viewpoint. So that is our goal. And the goal is to, again, yes, go into quantum for quantum use cases. But the here and now, a subset of that goal will solve for these classical use cases. But they are sort of one and the same. Because the building blocks are the same. It allows us to prove out things today in the real world at scale as we move towards the ultimate goal, which is create super computing environments or high-performance computing environments where you’ll have a mix of CPUs and all kinds of classical accelerators, DPUs, GPUs, et cetera, plus QPUs, quantum accelerators, all together, working together on a problem that we need to solve. So that is the ultimate goal, but these are the roadmap steps. Classical now, distributed quantum computing at scale next, and a mixed mode environment in the future.
Yuval:I have a question I’d like to ask each of you individually, and that is, you know, Reza, you’ve been, sounds like you’ve been in quantum for many years. What have you learned about quantum networking or that you didn’t know, say, a year ago? And then maybe Vijo, you can answer the same question.
Reza:What I am, I, is in, the quantum is interesting space. Both I believe computing goes, what I want to say, computing and networking, is dominating. by physicists. And physicists by training, I am one of them is not, they are not engineer. And really, no, I think we passed the science part. And it has to be basically taken by engineer and mature by engineer. So what I learned when I entered the quantum domain, quantum networking, how amazing is that there is a little bit of awareness is about what does it take to build a network and how a real system works in a real environment. And that’s really surprised me. And I think that’s also case a lot of opportunities and a lot of lesson to be learned. I mean, if you want to make this a success.
Yuval:And Vijoy, how about you?
Vijoy:Sorry, I think I was a mute, but I think there’s a lot of learnings for me. I’m not a quantum researcher by training. So I learned a lot from Reza and the team that he runs. The thing I would say is what was surprising to me was how real this thing is, this field is, because especially on the quantum networking side, was it surprising to me that quantum networking is pretty much here and now. And the narrative that had, I used to hear, like two years ago, three years ago, when we weren’t even looking at quantum, it’s like, yeah, that’s 10 years out, that’s 20 years out. I mean, take your pick, X years out. And now working with the resident team and some of the other team members that are building out these systems, it’s like we can actually, there are problems to be solved today that we can actually solve. Systems can be built today and money can be made. today. So that was actually one surprising thing. The other not so surprising thing is like, I’ll never be a quantum expert. So it’s too hard. So I’ll always be learning. There’ll always be stuff that I’ll never know about, which is a lot bigger bubble than the stuff that I know about. So maybe that’s not surprising, but it’s eye-opening.
Yuval:So as we come to the end of our conversation today and you were speaking about quantum experts, A question I’d like to ask is a hypothetical. If you could have dinner with one of these quantum experts or quantum greats dead or alive, who would these people be? Maybe VJoy first.
Vijoy:Yeah, actually, I would actually go back to maybe Einstein and dig a little bit more as to why he was skeptical. Is his skepticism better today compared to where he was in the 30s? or is he still a skeptic and just pick his brain as to, are we down the wrong path entirely on the quantum theory side of things? And do we need to start from scratch? Or is he convinced and can you convince me as a result that we’re on the right path?
Yuval:And Reza?
Reza:I would go. I would love to meet or have a dinner with Feynman. Richard Feynman, I mean, that’s my hero. As he says that if you claim you understand quantum, you don’t understand it. So that’s his famous saying. And obviously, because he is the fault, I mean, without his work, we wouldn’t have a quantum computers as we have it today or the concept of it.
Yuval:Wonderful. Reza, Vijoy, thank you so much for joining me today.
Vijoy:It was a pleasure having this conversation. Thank you.