Transcript | Quantum Computing Interconnect, Memory, and Other Engineering Advances

Engineering challenges abound in quantum computing. Technologies like interconnect and improved memory for repeaters will enable these machines to accelerate and power connectivity in the future. Q-NEXT is one of five quantum information science research centers funded by the DOE to help develop these technologies in addition to strengthening the nation’s leadership position in the quantum arms race.

Find out more in this episode of The Post-Quantum World. I’m your host, Konstantinos Karagiannis. I lead Quantum Computing Services at Protiviti, where we’re helping companies prepare for the benefits and threats of this exploding field. I hope you’ll join each episode as we explore the technology and business impacts of this post-quantum era.

Today, we’re talking to the chief technology officer for Q-NEXT, which is a Department of Energy National Quantum Information Science Research Center. He’s also a senior scientist at Argonne National Laboratory and a professor at the University of Chicago. Supratik Guha, thanks for joining me.

Thank you, Konstantinos, and thank you for having me.

It’s good to get to talk to you again, this time on the show. You’ve been in this industry about as long as I have. So I figured it would be fun to give everyone a quick look at your background, because you started at Big Blue, basically.

That’s right. I finished my Ph.D. and joined IBM in 1991, and other than a few years when I left for a bit, I was pretty much at IBM for a long time. I then moved to academia in 2015 and joined Argonne National Labs at the same time.

When you were at IBM, you were working on some of the development of the first quantum computers there, right, like in the management side?

Yes, I was not technically involved. The quantum computing group reported up to me starting from roughly about 2010 to about 2015. I was the director of physical sciences there, responsible for physical sciences research at IBM and worldwide IBM strategy in the physical sciences. That was the time I got involved with quantum computing as a manager, and at that time, we were the first to take a look at asking the question “What would it take at a system level to put together a working quantum processor?”

That’s pretty much one of the most exciting moments in the history of this field, because you guys actually got it working. So you were there when they built the very first 5-qubit processor, right?

No. I believe I had left by then. There were two managers at IBM who were very forward-thinking. One was a guy called Nabil Amer. He managed most of the quantum computing effort at that time — and this will be the early 2000s, maybe. It was mostly a theoretical group, and he really pushed to protect that group. And then there was this gentleman called Roger Cook who was the first line manager, and he began experimental efforts in trying to improve the coherence times of the superconducting qubits.

Roger, unfortunately, passed away, but it was his stimulus. He put together a small group, and then the qubit lifetimes kept increasing — that came around 2009–2010 — and we started feeling that maybe it’s time to start thinking about putting a little processor together. So I, as a manager, put together a few experts and told them, “Look, why don’t you guys go and try to figure out what it takes to be able to build a working quantum processor,” and then the group started growing. We started hiring, and then, before I left, we proposed putting a quantum processor on the cloud, and this was an organically designed idea that came from bottoms-up. To the credit of IBM’s very senior management, they really championed it, and then I left. Then, that project started again, about putting it together — putting it on the cloud and so on.

Yes, and the rest is history. The idea of putting quantum computers on the cloud is now — it’s just the way we work. It’s hard to imagine having these things in everybody’s server room.

Now, Q-NEXT. Do you want to tell us a little bit about that?

Q-NEXT is one of the five Department of Energy National Quantum Institutes. A couple of years ago, the Department of Energy put a lot of investment into the creation of five such national centers — each one had roughly $115 million over five years. That’s a significant major investment. A lot of this came out of the national focus on picking quantum information science as a research imperative for this nation, and so Q-NEXT is one of those centers. It’s directed by David Awschalom.

There are, I would say, about 90 different PIs. It is a multi-institutional collaboration. It is led by Argonne National Laboratory. It has a number of universities, a number of companies, and the idea here is to be able to do the science that can enable quantum technologies. Q-NEXT’s particular focus is on quantum interconnects, and we prefer to use a definition of quantum interconnects as “the ability to distribute and entangle quantum states or quantum information in a system.”

Unlike traditional grid computing, this is different, because you actually have the calculation alive. You have these entangled states, and you’re able to act on these qubits. Would you agree that interconnect is potentially how we’re going to get that first staggering amount of power one day?

It’s something that really is an integral part of any quantum system, be it in quantum sensing or quantum communications or quantum computing. Let’s say you think of a quantum computer. I’m not a quantum physics expert. My background comes out of material science — building hardware technologies and so on and so forth — but it is my understanding that we have an idea of how to build quantum computers that are, let’s say, roughly up to 1,000 qubits in size. We can probably do that if we do everything that we’re doing right now, just a lot better.

But it’s fair to say that the research community at large does not have a clear idea of what would be the microarchitecture of a quantum processor, let’s say, larger than 1,000 qubits, which is what one would need in order to be able to solve and address the problems that are going to be of real interest to the public and the community that’s outside of the quantum field, and where one hopes that quantum processors will make an impact that they will find it useful.

So, as you look at quantum processors all the way across these different size scales, interconnects become important, and increasingly important. How would a large quantum processor look? It would probably be distributed. You’d have to connect individual quantum processors one to another, so you’re looking at different length scales, and you’ve got to connect them with high fidelity. My own sense is that quantum interconnects will become increasingly important as you go forward.

It’s similar for the quantum communications side. If you want to entangle matter qubits — it’s difficult to entangle that, let’s say, across an optical fiber beyond 100 to 200 kilometers, something like that. For that, you would need a repeater, and in order to build a good repeater, you need a good quantum memory. A good practical repeater has not been demonstrated yet. That’s all part of interconnect research, so, yes, I do think quantum interconnects play an integral role in this entire field.

So, it sounds, like once again, you find yourself guiding the direction of the industry in this role.

This is something where everybody does this together. Today, going forward, there’s a lot of back-and-forth. I think industry and academia need to do it together. Things have changed. Research models have changed. The way technology develops has changed. Going forward, there needs to be far tighter coupling between industry and university, and I think you’ll see, going forward, people swapping between industry and universities in their careers, so that needs to happen too.

You’ve gotten to present at Senate hearings and things like that?

Yes, this was in 2017 and 2018. There were congressional and Senate hearings. When the decisions were being made regarding investments in quantum information science and technology, and I, as a representative of a large Department of Energy national laboratory, Argonne, I was asked to testify at these hearings and discuss the need for and importance of quantum information, science and technology.

Along with that comes the importance of having people able to enter the workforce in this space. Are you guys pushing for education?

Yes, that’s a huge part of Q-NEXT, and I’m glad you raised this point, Konstantinos. Quantum information is an interesting area. So far, it’s been dominated by physicists, but it started its journey now heading toward precompetitive technology — even competitive technology now, if you look at some of the quantum processor work that’s going on, quantum computing work that’s going on. It’s a field that’s going to increasingly require multidisciplinary talent. You cannot make a system with just physicists or chemists or just computer scientists. You need electrical engineers. You need packaging people. You need process-development people. You need vendor supply chains. You need metrology. You need characterization.

Quantum information intimately involves quantum mechanics, and it’s something that we’re trying to do that we’ve never tried to do before. We’re trying to manipulate quantum states. I view it in one way as trying to make artificial molecules that are huge. That’s the way I see a big quantum processor. It’s very challenging, and it requires a workforce where an electrical engineer or a mechanical engineer has some understanding of quantum mechanics on some kind of deep level in order to make this happen. So, it requires a broader, almost new kind of education. It happened, say, for instance, when biology started merging with engineering — bioengineering and so on. You had these new-generation material scientists and electrical engineers who started understanding biology, and that sort of thing needs to happen.

There are new workforce requirements. Workforce is huge for Q-NEXT, and we believe that our workforce development, it can’t be just confined to a university. We hope that our graduate students will do their Ph.D.’s collaborating with people from industry. They’ll spend their time in national laboratories, doing their experiments so that they are exposed to the professional environment that national labs and industry labs can provide. I am a huge champion of this, because when I was at IBM, I had students who did their Ph.D.’s co-advised by me, and they would spend a lot of time in my labs and industry labs, and they would see what were the relevant problems there, what was the work culture there, that sort of stuff, and that’s hugely beneficial to a new generation of workforce that will be developed in this area.

Those early days were pretty great. It was that spirit of sharing everything. Are you worried now that people are starting to close the doors and keep everything almost as, like, IP instead of science now?

That’s another great question. When you start developing the science, turning it into a technology, at one point, clearly, IP barriers start coming up. The proprietary nature of stuff starts rising, and that’s totally understandable, and that’s totally needed. In the end, these need to fit into business models. The question is, when does that begin to happen? If I were to give an analogy from the past, if you look at silicon microelectronics research, there was a lot of stuff that was put out there. Industry researchers used to participate in the academic community openly, and they were like cousins of professors intellectually. That has changed over time. That doesn’t happen as much as it used to. There are some downsides to it. You end up having different companies reinventing the wheel.

Let’s take superconducting qubits, for instance. Right now, it’s a metal and a dielectric, and you build this tank. The coherence is affected by defects, by surface states, by interfaces — that sort of thing. There are material issues. Now, you have different companies all trying to solve this, and each, they have their own solution. If there were a common base of understanding of this sort of thing, then that would be great.

This is what happened in silicon technology. If you have a silicon dioxide dielectric on a MOSFET, you want to passivate it. You want to understand what those types of defects are — and there were people who did that sort of work, and they published it openly, so that balance needs to happen for me. This is also where, I believe, these five national quantum institute centers will play a huge role. This is where the government funding and investment will hopefully play a huge role in providing a reference voltage of knowledge to this field.

Nobody alone can develop this field. Today, business models are different, as I said, and the kind of investment required to do the sophisticated level of research that’s needed today can’t be afforded by anybody. If I put my industry hat on, it’s thinking about a high-risk, high-payoff model where everybody shared the risk.

Yes. To keep all these things motivated and moving, there does tend to be a lot of hype. Without a little bit of hype, the industry stagnates, but then you get into the false-expectation game, and it becomes a give-and-take to try and move along this kind of development. How do you find the balance there?

I agree with you, Konstantinos, and you said it right. You need a little a bit of it, but you can’t have too much of it. Sometimes, maybe you shoot for 100X improvement, but then, from a product point of view, maybe 50%, or 2X improvement, is good enough, and then you say the rest was hype, but it was useful hype anyway, in the end, but sometimes you can tend to overhype things. Today, with all of these social networks and that sort of thing, there is a tendency for things to go overboard, and we need to be very careful as a community to try to not make that happen. I’m not sure that there’s a ready-made formula to do this or accomplish this.

Yes, definitely.

We have to figure that out as we go along. There have been instances of technologies that are good — that was good stuff — but it failed because it was overhyped, and there have been great technologies that have crept up on us without any hype. Solid-state lighting is an example. It’s changed our lives, but there was no hype on this.

Let’s look a little bit at some of your more technical work. I know you’re involved in material science. Were you working on quantum memory?

That’s what my students and I and some colleagues here at Q-NEXT are working on together. This is a new field for me. Quantum information science is a new area for me, but we’re trying to develop oxide-based, solid-state devices with rare-earth atoms embedded in it. These rare-earth atoms are going to be the memory elements. We’re trying to build this, trying to develop the science of this for use as quantum memory for quantum repeaters. We’ve been looking at dielectric oxides that are hosts, solid-state hosts, to embed atoms such as erbium that will be the quantum memory element. Erbium is attractive because it has atomic levels, electron levels, that can be accessed using photons with wavelengths of about 1.5 microns. That’s the C-band optical transmission wavelength that’s used across the world in optical-fiber communications to make it compatible with optical fiber that’s already laid in the ground today.

You’re working on using memory in that context and not something like memory involved in, let’s say, running large search or anything like that, because I know there’ve been some challenges with getting Grover’s to work without a certain type of memory.

Yes, not yet. What we are trying to do is develop these materials and a device where you can store a quantum state and retrieve it. I know these memories, or adaptations of these types of memories, may be used, though, in storing quantum information while, say, you’re loading a large classical set of data onto quantum states. I’m guessing that’s what you’re trying to refer to.

For our work, you ought to focus on one target, and we’ve thought about building a quantum repeater, but I have the sense that these types of memories or these ideas can be used across the board. I think eventually, you will need quantum memories to be developed in any large processing system that you might wish to build, so it’s a device element that’s going to be important.

At that point, you start at look at this as, “Is this just becoming another von Neumann machine?” this general idea of the memory-processor interaction. What’s your view on that general architecture of quantum computing? Do you see anything shifting in the future?

I’m not a great expert in the area of quantum computing architectures, but this is very far from a von Neumann machine, which is extremely sequential, where today you’re limited by what’s called the memory bottleneck, where, to do anything — any logical step — you need to keep shuttling back and forth the memory to do the logical operation. Here, you may need the memory as you load data onto quantum states, but then, once you’ve loaded it onto the quantum states, then, for the rest of the processing — so, this is a data-interfacing, classical-to-quantum-data-interfacing, issue where our sense is that you might need it. It’s a little different.

That’s, of course, the area that I was referring to. Do you think we can get more efficient, then, with the architecture we have now in our implementation of these processors?

You mean the quantum processors?

Yes, because I’m always wondering if there are ways to squeeze a little more out either through software or just more clever usage.

The question that needs to be asked for quantum processors, which I think needs to begin, is a detailed understanding, analysis and projections and designs — scaling of these architectures and larger sizes. There are simple things like, “If I’m building a machine with so many qubits, how many logical qubits does that translate to? What’s that relation between the number of logical qubits needed, the physical qubits?” If I’m a person who’s not an architecture person, but I need some guidelines in the design of a system — performance specifications, performance projections — there are a lot of microarchitectural design elements and understanding elements that need to be worked out and fleshed out, particularly for larger systems. Most of the rigorous system-level analysis, in the sense of microarchitecture-level analysis of these systems, needs to begin to happen.

It’s fascinating how different all the major producers of these machines are when it comes to going from physical to logical, right? Google recently talked about how to get truly error-free computing. In their mind, they would need about 1,000 qubits for one logical qubit, and then just a week later, Honeywell said, “Well, we can give you four logical qubits if we just use three for error correction.” It’s like, “Well, that’s a very big difference.”

Then, there needs to be a consensus as to how you benchmark these, how you define these things. This translation between logical and physical qubits, the material properties of the qubits, also comes into play, and their fidelities and the fidelities with which you can connect one subcircuit to another subprocessor across systems, those things start coming into play. I view this right now as there’s this elephant, and it’s like a bunch of people viewing the elephant from different directions. We need almost like a closed-form analysis where everything is put out together on a subject basis. It’s not going to happen right away, but this is the direction where people need to start adding toward comprehensive analysis that’s also free of hype.

Yes, that’s important. Of course, we still see these systems working with classical systems hand in hand, hybrid solutions. In the future, I think, we’ll still be using classical systems to shepherd the performance along while you do a calculation. Do you see any other type of computing besides quantum playing a big role in the future of this industry?

There is a feeling that a quantum processor will initially be like an accelerator piece in a large classical system. That is one school of thought that sees that that’s how this stuff is going get in, but the specific functions, just like an accelerator does today, will be done by a quantum processor. That connection, the classical-to-quantum interface, becomes really important.

Going forward for the future of computing, in addition to quantum, I see a couple of other things: One is what a lot of people have been talking about: Because computing has become very data analytics, AI, they’ve changed the demands put on a hardware machine. Scientific precision isn’t that important anymore. Classifications and patterns, that sort of stuff, becomes important, so there’s a lot of interest in neuromorphic-type computing, where you try to do a bunch of things together without getting into this memory-bottleneck problem.

There’s going to be a lot more integration of memory and processing — essentially, trying to fuse the two together, if possible — and then there’s another piece of computing that is emerging, which is coming from the direction of sensor networks. That’s a big area that’s emerging. You can call it the Internet of Things, cyber-physical systems, where it boils down to a bunch of sensors doing some processing work locally, and then information being shuttled across this network and some processing being done centrally. I view this as where distributed computer is heading, so I view that as another trend in computing. Those are the three things that I see: quantum, large-scale, sensor-driven distributed computing with this neuromorphic-type approach that starts departing from von Neumannian computing in the sense that it starts fusing memory and logic closer and closer.

It starts to simulate more of the human-brain approach, maybe.

Yes, that’s the ultimate, eventual hope.

It makes you wonder: Will these technologies merge and give us that AGI one day? It’s possible we’ll get this superpowerful artificial intelligence one day with a different approach. We can hope. I think you’re going to keep seeing quantum computing come up in science fiction as a result.

I am hopeful that there will be some practical products that will come out of quantum computing, maybe in a 10- to 20-year time scale. Ten to 20 years passes fast.

Ten to 20 years — I don’t know. I’m a little more hopeful than that.

When I say a product, I mean something that has large-scale impact. That’s what I mean. It’s a magnitude of an impact.

Yes, but luckily, way before that, I think we’ll be seeing real advantage in certain discrete use cases.

You can already buy — there are products sold in quantum links that are not repeater-enabled, so maybe they’re a 1,500-kilometer links. Those products are already coming out. There are small companies in the U.S., Europe. Stuff is already happening. For stuff to become big in terms of large user base, big impact, I think it will take about a decade.

We’re calling this a decade of quantum, so why not? We have a nice little window here for things to manifest. So, I will be putting information on Q-NEXT in the show notes for anyone who’s interested in learning more about this, and yes, Supratik, thanks so much for joining. I really appreciate you taking time to come in. I know you have a lot of hats that you wear.

You know, I’m bald, Konstantinos, so I wear a lot of hats.

Yes. Same here, yes.

Thank you very much, Konstantinos, for having me. I enjoyed the conversation.

That does it for this episode. Thanks to Supratik Guha for joining today to discuss Q-NEXT, and thank you for listening. If you enjoyed the show, please subscribe to Protiviti’s The Post-Quantum World and leave a review to help others find us. Be sure to follow me on Twitter and Instagram @KonstantHacker. You’ll find links there to what we’re doing in Quantum Computing Services at Protiviti. You can also find information on our quantum services at www.protiviti.com, or follow Protiviti Tech on Twitter and LinkedIn. Until next time, be kind, and stay quantum curious.