Transcript | Honeywell’s Roadmap to Quantum Value

How does a 100-year-old company known for keeping your home comfortable surprise the entire quantum computing industry by providing the most powerful quantum computer in 2020, then making it 10 times more powerful within a year? We’re talking with Honeywell Quantum Solutions 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 the post-quantum era.

Our guest today is the chief commercial officer for Honeywell Quantum Solutions, and as you know, Honeywell’s made quite a splash, so we’re really happy to have him here on The Post-Quantum World. Welcome to Justin Ging.

Yes. Thank you, Konstantinos. Happy to be part of this.

Yes, thank you so much for coming. I could still remember sitting in my office —we were still in offices at that time — and all of a sudden, I hear, “Wait, Honeywell is doing a quantum computer? What is going on?” It kind of blew my mind for a moment there. Then, I looked into it and saw it was quite an interesting approach. I don’t know if you want to let our listeners know a brief history of how Honeywell went from being a well-known brand in just about everything to this hyperfocused industry.

Yes. Honeywell is not necessarily a name you think of when you think of next-generation computing. Those companies are players in it where you just say, “Yes, that’s naturally their strategic imperative to go after,” but Honeywell — people pause, and they’re like, “Well, why is Honeywell on it?” A big part of it is that quantum computers are really different than classical computers. They’re room-size devices that look more like big physics contraptions — which, in a sense, they are: We’re levitating individual ions and training lasers on them to do quantum operations. So, it requires a lot of the things that Honeywell, over decades of experience, has a really strong background in. So, we think lasers and optics, cryogenics, vacuum systems, and really complex control systems — all these things are Honeywell’s bread and butter.

Our start was almost a decade ago, where there were some programs where people were looking at doing trapped ions for a number of reasons and looked to Honeywell to help build the ion trap, which is kind of at the heart of the system — the device that does the holding and levitating of these ions. We got into some of that background and said, “Actually, there’s more we could do. We could be trapping these ions ourselves, and we could put the other pieces to the system together.”

After several years of proving it to ourselves, we reached a point about five years ago where a more concerted investment was made. We stood up an organization where we have our headquarters in Broomfield, Colorado, just outside of Boulder, and another facility in Golden Valley, Minnesota, outside of Minneapolis, and built a dedicated team to go after this market. We were doing it in stealth for a while — that’s why people didn’t hear. The reason for that was kind of what you got at the beginning, which is, Honeywell doesn’t necessarily get the benefit of the doubt of doing advanced computing. We wanted to be ready to say not, “Here’s what we’re going to do” but “Here’s what we’ve done already,” and we wanted to be able to have that ready.

When I joined about three years ago, my mandate was to pull back the curtain and show off what we’re doing. I imagine that’s when you had your eye-opening experience in your office, and we stood up our website and reached out to ecosystem partners who built the customer pipeline. Last year, we launched two commercial systems, H0 and H1. With our H1 system, we have set multiple records on pushing quantum volume, the measurement of the capability and system. We have a pipeline of customers using that, and engagements with enterprise customers — moving it forward in the industry and getting the word out that “yes, we’re here, we’re doing it and we’re leading the way.”

Our play is really providing the very best, highest-fidelity systems with the most capability so that for algorithm writers and people developing what quantum computers are going to do — developing these killer apps — we give them the best tools they can have so they can push it forward. In the industry, we’re all dreaming of the day when we know exactly what killer apps quantum computers are good for, how they’re going to provide value for customers and really become a part of business processes. We think that day is pretty close.

Yes, that was an awesome overview, and we’re going to dive into just about everything that you said in this episode. It was so funny because, way to launch, right? You guys didn’t come out and say, “Hey, we’re going to have this computer.” Instead, it was something modest like, “We’re going to come out with the world’s most powerful quantum computer, and, in a year, we’re going to increase upon order of magnitude.” That was like, “Whoa — OK.” Just make a small claim, right? It was pretty exciting stuff. So far, it looks like it’s tracked pretty much right to the minute, too, but we’ll get into all that with the road map. Did you want to talk a little bit more about how trapped ion differs from, let’s say, transmon and the other types of qubits out there?

Yes, I’m going to talk about it in the sense of what our qubit is. We use an ytterbium ion, and we’re using the spin of the electrons as the states of the zero and one. What that means for us is that we’re starting from an atom, which, because of nature, it’s perfect. It’s predictable, it’s very well characterized and it’s known. So, all of our quantum operations are working within a realm of “Try not to mess up — try not to add error to it.” Any errors that we see — the lack of fidelity and infidelities — are really about the conditions in which that atom sets, and noise from things that we apply to it.

We know where the noise comes from. We can characterize that, and we can make every effort to reduce that. That’s what drives our approach to it, which is the trapped ion approach. And particularly, within trapped ion, we’re doing something different than others, which is, we’re actually moving the ions into their own separate areas. Our current trap, it’s like a railroad track, a linear trap where we can slide the ions around. We’ll take a pair of them and move them far away — relatively, on an atomic scale, it’s very far away from the others — and do the gate interactions with the lasers there. That reduces the crosstalk with the other ions and has a very clean interaction.

Once they’ve had their gate interaction, they’ve been untangled or other things, they can be separated back out and intermingled with the other qubits. That’s the unique piece of what we’re doing, and at the heart of it is called the quantum CCD architecture, which was originally envisioned by Nobel Prize winner Dave Wineland, and that was 20 years ago. Everyone said, “That’s great — it seems really hard to do,” and so it was not tackled.

Yes, it is hard. That’s the technical debt that we’ve been paying over the past several years — to knock down the challenges and really get a hang of all the pieces that have to go into this. The movement of ions, how do you keep control of them, how do you move them past each other in a swap operation — not logical swap, but physically move the two ions around each other so they can get out each other’s way, like two people going on the airplane aisle? All these pieces had to come together.

We actually had separate teams tackling each one of those problems, so there was a team focused on getting just the highest fidelity gate that you can. Another team: How can you move them most efficiently? Another team: How can we have parallel zones so that we can do more than one operation at the same time? All these pieces were done in parallel in the research side and then integrated into the system. Yes, it was a hard concept — that was 20 years ago — but we’ve done it, and it’s what makes ours unique, and that’s our capabilities.

Yes. I remember when he and Serge Haroche won the Nobel Prize for that work — I think it was in 2012. I was taking about quantum, and I mentioned it in there. People —their eyes lit up. It was like, “Whoa.” If people are actually winning prizes in this field, maybe you’re not crazy, and quantum computing is coming. I told people, “Yes, it really is.” Then, here we are, right? It took a little while. Do you see limitations with that approach? Do you think that you’ll still be able to use that same kind of separation system to keep having qubits ramp up, and up, and up and up as we need?

Yes. That’s always been one of the critiques of trapped ion. People say, “Well, it doesn’t scale,” but I think that’s based on other older approaches to trapping ions. We actually think we have a really good path to scale, and that is by providing the roads and streets that these ions can travel on. The actual gate interaction happens separately, as I said. Often, it’s a quiet area, and really, it’s about, where do you put all the other qubits during this time? Then, how do you get the qubits to meet up with each other? The scaling operation doesn’t change that fundamental fidelity, so we keep our same high fidelity regardless of how many qubits we have.

Then, we have different trap designs — essentially, different roadways to be able to transport these ions around. We’ll go from a linear to a racetrack type of operation, where it’s two linear sections, to then going to 2D, like more of a city street grid, and you can imagine it just like cars driving around the city. It’s more efficient when you have different paths to go, so we have a plan to scale up over the next decade through this approach.

It will be a challenge, but it’s more of an engineering challenge rather than a scientific challenge. As we go forward over the next decade, we’ll need to look at how do we shrink some of the systems? Right now, we use bulk optics — large lasers sitting on optics tables, going through lenses, mirrors, prisms, wave guides and things – wave plates, I should say. They all shine in through windows into a chamber that’s about size of a basketball — a stainless steel chamber that’s been evacuated to less density than outer space and cooled down to 10 Kelvin in liquid helium, and providing a really clean environment. We’re shining lasers in through those windows, and over time, it becomes such a large number of laser signals that it makes sense to try to shrink those down and deliver them more closely to the ions.

Anyway, the high-level takeaway is, there is a path, and for at least the next decade, we think that the technology we’ve chosen is the right one to pursue in terms of the best way to push quantum capability forward.

Yes, and we have to find a way to make that giant room-size setup into something more photo friendly that you can capture like an iconic little image. You and I joked about this once before. Everyone thinks that the giant chandelier dilution fridge — that you have a transmon hanging at the bottom. It’s like, “What do you do with a trapped ion?”

Yes, exactly. A lot of companies have their big, black boxes, where all the science is inside. I used to joke with people, “Hey, we’ll just have our big, black box with lightning bolts on the side — their racing stripes,” but, yes, there are a lot of different components to it. The heart of our system is that trap, so it’s a nice gold bow-tie-shaped device where the exciting part is, it’s individual atoms that we’re manipulating. It’s mind-boggling, even the idea that you can see an individual atom, but now I see it — actually operate and control, know exactly where it is at all times, and manipulate them with other atoms.

Yes. If anyone wants to see it, you guys have to check out the artwork by going to the Honeywell site, because it’s pretty cool to see that bow-tie design. How does your implementation differ from IonQ?

The IonQ approach comes down to how you’re trapping, and that is, if you imagine that you’re taking electric field, you’re changing electric field that essentially forms a bowl shape of electric potential. And it’s that potential well that’s keeping that ion trapped, because it’s essentially having to go uphill in any direction, so it can’t go anywhere. It’s lowest energy is to sit at the bottom of this trap, this well.

The IonQ approach is to have a large well where all the ions are together in a chain in one large bowl. Our approach is to use small bowls. My analogy is, you’ve got either a busful of kids or you’re taking the kids off in separate cars to go have that conversation. You can imagine that you’re going to have a quieter conversation between two people in their own car than you are in a bus where two people are talking to each other. The difference comes down to how you address the ions within that configuration and how the crosstalk could interfere or not. Our approach is really about that movement of the ions, rather than having them sitting in one stationary spot.

That’s really easy to visualize, actually. Yes, thanks for that explanation. You have, currently, two machines now — basically, H0 and H1. How can people access these today? We were talking about killer apps before. I know that they are accessible on Microsoft Azure Quantum. Is there another way, too?

Yes. We handle customers directly. One path is just coming to us directly. With H0, it’s a credit-based system, so you essentially buy your gift card of credits, then you draw down from that. When you run a circuit, it uses a certain amount of credits depending on how complex it is. It’s analogous to the amount of time spent on the system. If you’re doing something really simple, it’s fewer credits. If you’re doing something really complex, it’s going to take longer for the system to do. It costs more credits.

With the H1 system, we do it as a subscription model, where users get a certain amount of access throughout the month. That’s really driven by what customers have told us in terms of if you’re really developing algorithms, you’re not necessarily just doing a five-minute experiment: “OK, good. I’ve got some results.” You’re repeatedly running things, and some of these algorithms can take tens of hours or longer to run to get those results. We’re trying to accommodate the different needs among the users.

As you mentioned, Microsoft is another path. We’re part of the Microsoft Azure Quantum Solution, so customers who like to use the Microsoft Q# platform and have an easy way to access our system — perhaps they have an Azure account already — that’s a possibility for customers, and it’s quite convenient. We also have a number of ecosystem partners who have their own software platforms, and by working with them, customers can have access that way as well.

Yes, because some of those solutions won’t be pure quantum. It will be mixing classical and quantum in that hybrid approach. I guess, for some people, they will want to have a cloud environment where they can write a more traditional code, too, and then reach out for that quantum piece. With your subscription model for the H1, is that flat for the month, or do costs go up if you end up running, like as you said, many hours’ worth of —

It’s a fixed amount of access over the course of the month. That’s guaranteed, and we tried to have queued access that — some are on a reserved access. You can have the machine to yourself for certain algorithms. That makes a huge difference — you’re not constantly heading back into the queue. Some algorithms, you just want to do a quick thing and make a change classically, then go back in and do another quantum operation — that kind of back and forth. If you have to wait in line every single time, you can imagine it’s like at the DMV, where, “Hey, you didn’t fill out line three. You need to go fill that out and come back.” Like, “Uh-oh, I don’t want to have to get back in line.” We do have reserved times, and that allows for when you need it, you can have that dedicated access to the system, and other times, queueing works fine. You can set it, forget it, come back and get your results later.

Yes. One day, it will make your head hurt, because a quantum computer will be used to optimize how you’re queued up to use a quantum computer algorithm. It might be possible.

That’s often been talked about. You need a quantum computer to build a quantum computer, yes.

Do you have any kind estimator for depths of circuit and how much would it cost? I was curious about that.

Yes, we do. We have something called a syntax validator, which is not quite a simulator, but it does allow users to try their circuit, in a sense — make sure their code is right so they can run it through, it goes through all the same software stack that you would in a normal usage case and the only piece that doesn’t do that is actually run on the quantum hardware. It does everything else, and it will return data in the right format. It will be filled with all zeroes, and that allows users to make sure that they didn’t have bugs and get their code written and interfaced appropriately. That’s a tool that’s provided free to make sure you’re on track, and as part of that, it also gives an indication what the cost of that particular circuit would be.

That’s super useful.

The cost is a formula, so you could also do it on paper. It’s essentially a function of how many single-qubit gates, two-qubit gates, measurements and shots you’re doing. It’s pretty straightforward to figure out. If you just have it written on paper, and you look with your circuit, you could count it up. But the nice, easy way is to just stick it in and see what the cost is going to be.

Yes, because every once in a while, when we’re doing a POC and we use one of the cloud access routes to get to one of these machines, we have that moment of hesitation. It’s like, “Did we calculate it right, or is this going to cost 10 times as much as we thought?” That’s fun.

Yes, exactly. We try to avoid that. Nobody wants that kind of stress. That’s for sure.

Yes, it’s like, “Oh, it’s going to be $20. Oh, $2,000. Oh, yes. OK, that’s a slight error that we made there.” Now, we have these two machines. I guess, we can probably guess what the name of the next one will be if we went from H0 to H1, unless you guys have a new naming scheme coming, but can you talk a little bit about the road map coming up and the quantum volume you’re expecting to hit?

I’d say those go separately, actually. We will keep increasing our quantum volume based on the H1 system. H1 is a little bit more of an offering rather than a system. There would be more than one physical system serving up that capability. When we launch our system, it’s not fully populated in terms of the maximum number of qubits that can be in there. The trap can handle an order of 40 qubits — the one that’s in H1 — and over time, we’ll be adding qubits to that system. So, it’ll go from 10 to 12 to 20, or some numbers like that, over time.

With that, we will also be pushing on fidelity. You’ll see more qubits, more fidelity, from that H1 offering. With H2, as you have guessed with the naming convention, we will be launching next year, and it will add to where the cap is on qubit count. Again, we’ll be pushing on qubit number and qubit fidelity, gate fidelity. Really, that’s what drives what goes into making a good quantum volume measurement. You have to have both number of qubits and the high fidelity. It turns out that fidelity starts to matter even more than the number of qubits to get the number quite high.

Of course, our qubits are fully connected, which is one of those factors in the quantum volume calculation, but the quantum volume is more than a calculation. It’s actually a measurement thing, so that’s something that we think is very important for everyone who’s quoting numbers to provide the data and its measurement. It’s a test you must pass, essentially, and so we’ll keep testing our system and we’ll keep pushing those numbers up. We are still planning to keep on to what we claimed last year, which is an order of magnitude every year. When we claimed that, we were at a quantum volume of 64 in June of last year, so we’re aiming that by June of this year, that we will have exceeded that 640 number, and then, the year after that, 6,400, and then so on and so forth.

Yes, that’s terrific. I was really impressed to see the 512 announcement a while back onto that achievement. I mean, this is initially technology. With transmon, or whatever, it’s however many qubits that you have to set up for. In this case, it’s almost like having a car engine that you could keep adding cylinders too. It’s a new idea.

Yes, I hadn’t heard that announcement before, but I think that’s accurate. Yes, we’re adding capability. Another analogy is, we built an auditorium, and now, we’re filling it. There’s plenty of room to grow as it is.

You guys plan on continuing to use quantum volume as a measurement, because IonQ — they’ve created their algorithm with qubits. I didn’t know if you had any plans to come up with something different or you are happy with that general approach.

We’re definitely aligned with what quantum volume tries to do, which is to provide a single number reference to what is the capability of the system, and it does a good job. The quantum volume does a good job of characterizing that. It’s not only considering how many qubits — it prevents this race to “I have more qubits.” Well, if the qubits aren’t quality, they don’t really count. Actually, that’s the same thing that IonQ is saying with the algorithm of qubits: It’s got to be quality. We think that it’s a pretty good measurement. It’s not perfect, though. I think there are things that people have proposed different kinds of benchmarks: “Well, we think it should be an algorithm that you run and then you see how good it ran, and how well it ran on each different systems.” Perhaps.

We have been contemplating something more along the lines of a quantum decathlon, where you have different tests that you run. I think there are many analogies to classical computing. You’re running certain tests to see how good different pieces of a system are: How fast does it run? How much electricity does it use? How does it run in this kind of testing? What about a daily office environment. What if you’re gaming? Or whatever — there’s different metrics. We think that there’s room for that in quantum computing, too. It’s our goal to be as open and honest with our data about how it performs, and we’re thinking about other ways to measurement —to measure quantum computers and benchmark them. We think quantum volume is definitely a piece of that, but it’s not the end-all and be-all. There will be more.

With this constant growth, a lot of companies and their road maps seem to be indicating that 2023, into 2024, is going to be this magic 12-month period. IBM is claiming Condor, 1,120 qubits or so around then. IonQ is thinking that by 2024, maybe with the help of interconnect, too, that they’ll be achieving similar numbers. When do you anticipate something like quantum advantage happening?

I think it’s yet to be seen. We see this period as being very important from a slightly different perspective, which is, now is the time when we can check our work. We can run simulations. At this stage on a laptop, it won’t be long before we need HPC and supercomputers to do it, but there’s this period where we can gain confidence that quantum is doing where it’s supposed to. We can say, “Here’s what the real result is. Here’s how the quantum computer’s performing.” It won’t be too long before we surpass the capability of simulation.

Somewhere in the 40- or 50-qubit range, probably, is where we cross into that threshold where you will have wanted to already build the confidence that quantum computer is doing the right thing. Will that be quantum advantage? It’s yet to be seen, but that would be the first time frame when we think that that might start to appear.

And there’s a dynamic going on. We’re racing to build the best hardware, the most capable hardware and so forth to push forward that time. How soon can we get that date? At the same time, the algorithm writers are using their creativity to pool that end data. Some of the early estimates with some of these algorithms, they said, “OK, you’ve got to have hundreds of millions of qubits to get there,” and then they’ll come back a year later: “Hey, new breakthrough. We’ve taken orders of magnitude out of what our resource estimations are.” They’re pooling that in from those future dates.

There are certain smaller applications where people are thinking about “How can we leverage the power of quantum?” I think we’re going to be surprised that some innovative people will leverage the tools that we’re providing to pool that quantum advantage to be sooner than people think.

Yes, I feel that way, too, and I love that idea of a surprise, because this industry had so very many of them already. In fact, I’m talking to someone from one of them in a way. We keep getting these pleasant surprises, and I do think that it’ll be a combination of just a little more hardware and a little better work on the stack — and, ta-da! You’ve got something you can’t do with classical.

I think we’re seeing a lot more people and organizations jumping into quantum. They’re starting to recognize the potential advantages, and the more there’s participation in this industry, there’s more thought behind it. That mindshare is going to lead to some major breakthroughs. That’s what I’m hoping for, and I’m hoping to be surprised by some of these applications that emerged.

Yes. On that note, as we wait for one of these use cases that will show advantage, what kind of use cases are you seeing now as customers prepare for that big quantum lift-and-shift that’s going to happen eventually? Do you have any visibility into what they’re looking on?

Yes. The major categories of problems that people are tackling are in the optimization space, the acceleration of machine learning and the chemical-molecular simulation. It’s in those three categories that people from a variety of industries are starting to see how it can apply to their business. Interestingly, there’s crossover — like a pharmaceutical company, you’d say, “OK, well, they’re just going to use it for molecular simulation, do some drug development,” but they’re actually also looking at quantum from the point of view of “How can I improve my drug distribution or my supply chain around it?” An automotive company would say, “Well, yes, I’m interested in the transportation logistics, but also, how can I build a better battery for electric vehicles?”

I think there’s a lot of crossover among those three concepts, but those are the kinds of problems that people are tackling. Really, the involvement right now is to formulate some of those basics to get proof of concept that, yes, quantum can provide a path on these certain kinds of industry algorithms, problems that they care about. Right now, the problems are small — they’re scaled down to the point that with a pen and paper, you could probably figure out what the optimal solutions are, but it won’t be long before that because of that exponential type of problem, it becomes really impossible with classical computers to be able to solve.

It’s really about getting in, learning, participating and understanding how to use systems and starting to apply them. As the hardware scales up, and as these algorithms get more innovative, there’ll be a path to, hopefully, in the near term, providing that business value where we go to “Yes, we are talking about how many percent savings are, or is this going to be off our supply chain or off of the current processes?”

When this finally happens — that it actually does makes sense to run it on a quantum machine — do you anticipate a bottleneck in access to your systems, or do you think that this little leeway of a couple of years, you’ll have enough live online that should be able to meet the initial demand?

There’s a little bit of a trade-off, and we’re trying to make sure that we can balance the trade-offs. One is, do you put your efforts into scaling up the capabilities we have now, or do you put your efforts into putting that next-generation system out? It’s definitely a balance. It’s not an either/or. It’s “How do we best get that right mix of enough capacity at this capability versus providing that next system?” We think, right now, the emphasis is leaning more toward getting that next-generation system and the one after that, and so forth. Really, because we want to reach that point of value as soon as possible — but, yes, as more players come in, we’re looking at expanding that capacity as well, so we’re trying to walk a fine line.

That sounds great. It will be exciting to see you guys at the next few markers, too, on that orders-of-magnitude road map.

Well, we certainly hope so. I hope we can bring that news out and put the results on our page — show that we are measuring these things and that we’re advancing as we thought we would.

Yes. I’ve been tweeting about it all along, and you haven’t disappointed yet, so I hope that continues to be the case. All right. Yes, I really enjoy this chat. Thanks so much for coming on the show.

Yes, I appreciate the opportunity.

I don’t know if there’s anything you wanted to plug before we close — and, of course, there will be show notes, too, where I’ll be putting info.

I would say that if there are companies interested in talking about their journey through quantum, we’d be happy to engage. Our email is [email protected], and we’d encourage conversations about it.

Awesome. Thanks again, Justin.

Thanks very much.

That does it for this episode. Thanks to Justin Ging for joining today to discuss Honeywell Quantum Solutions, and thank you for listening. If you enjoyed the show, please subscribe to Protiviti’s The Post-Quantum World, and maybe leave a review to help others find us. Be sure to follow me on Twitter and Instagram @konstanthacker. That’s “konstant with a k, hacker.” 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 us on Twitter and LinkedIn. Until next time, be kind, and stay quantum curious.