Cookie Consent by Free Privacy Policy Generator
Search
Close this search box.

Superposition Guy’s Podcast — Florian Neukart, Chief Product Officer, Terra Quantum

“The Superposition Guy’s Podcast”, hosted by Yuval Boger, CMO at QuEra Computing

 

Florian Neukart, Chief Product Officer of Terra Quantum, is interviewed by Yuval Boger. Terra Quantum, a company with a broad approach in quantum technologies, focuses on creating diverse products ranging from software to hardware, cryptography, and quantum algorithms. Florian discusses the launch of TQ42, a software platform designed to make quantum computing accessible to both experts and non-experts, and the company’s breakthrough in achieving room-temperature superconductivity in graphite. He addresses the optimization of quantum algorithms for specific machines, the role of hybrid algorithms in AI, the potential of real quantum hardware, and much more.

Listen on Spotify — The Superposition Guy’s Podcast

Transcript

Yuval Boger: Hello Florian, and thank you for joining me today.

Florian Neukart: Always a pleasure.

Yuval: So, who are you, and what do you do?

Florian: My name is Florian. I’m Terra Quantum’s Chief Product Officer. So what I do is, together with our great research and development teams, take our research and think about products that we could create based on it and then make these products. So whatever that can be, Terra Quantum is a company that’s very broad in the sense of quantum technologies. We do software, we do hardware, we do cryptography, we do quantum algorithms. So it can be many folks. Sometimes, it’s algorithmic description that we bring into a library. Sometimes, it’s a hardware concept that we bring into a chip. So very broad and I have the pleasure to do that together with our great teams.

Yuval: Is that a good strategy to be so broad? I mean, quantum is so complex, to begin with, and now you spread your insights over so many different fields. How does that work for a business strategy?

Florian: Yes, we get that question a lot. I think what we can do well, what we do really well, is reuse our resources and results. So imagine we work on qubits. We can use the same qubits that we use for quantum computing as an antenna for sensing, for example. So that is something that I think we do really well. We also use algorithms all over the place. We use it in industry use cases. But let’s say we have an optimization algorithm that we use for solving logistics problems. Then, our medical device branch may use it to optimize the laser parameters in a medical device that is used for treating calcifications in the arteries in your heart, for example. Which is also something that we are working on. So, as long as we can do that efficiently, I think it makes sense for us to be broad. And as long as we see the potential to bring these things to the market ourselves, then it also makes sense to do it.

Yuval: And I think you had two recent announcements, one a software platform and one a superconducting thing. Could you tell me a little bit more about those, please?

Florian: Yes, absolutely. So that’s just very exciting. We recently launched TQ42 and closed beta. TQ42 is a software platform. It’s more an ecosystem actually designed to make access to quantum computing a lot easier, to facilitate access to quantum computing. The idea is that we target not only the experts, the quantum computing experts who know how to write software, who know how to design quantum algorithms, but also those who are not skilled in writing code, so that have a deep understanding of the problems. But are maybe not deep into technology, but still want to solve these problems, of course. So how can you make it accessible, make quantum computing accessible to those folks? It’s a much broader and much bigger group of users that we target here.

And that is what TQ42, first and foremost is. It’s an ecosystem platform, an SDK to provide access to everyone, be it deep technical experts or those with deep problem understanding, but not necessarily to come up with a quantum algorithm. It’s an SDK-first approach. So it can be integrated into any existing platform, into any existing software system, any environment. It is cloud agnostic, so it runs on any cloud. So wherever we have quantum processors available, wherever we deploy our libraries, our SDK, it can be executed, which is very important too. It comes with guided flows.

So if someone says, for example, I have my data here, but I’m not really interested in the algorithm that solves the problem. I just want to have my results. So then the idea would be, and this is something we’re currently still working on, so it’s a product that’s still under development. So, this algorithm design is taken away from the user if they want it. So we have a set of algorithms, finding the right algorithm based on your data and the problem type that you want to solve. We have a hyperparameter optimization. So, all our algorithms are hybrid, which means there’s a significant part running on classical high-performance computers and another part running on quantum chips. Both have hyperparameters that need to be optimized. So, in terms of classical algorithms for a neural network, for example, it’s those that we know well. It’s the number of neurons, the layers, the connections, the activation functions potentially. For quantum computers, it’s the set of gates. It’s the rotation angles that we want to find. And that is done by the platform, by the libraries that are actually great team development that we integrated to that very platform.

And then it’s also an app store. So applications that find usage with multiple customers, not only one, so it’d be that for solving logistics problems, traffic problems, any chemical simulations. So these are available as applications, as ready-to–design applications in the store or in the app store of the platform. The only thing that a customer needs to do is provide the data in the same way, in the way that we specified in the way that the algorithms have been trained. So that’s the only requirement. But other than that, ready to use and ready to deploy. So that is TQ42 in a nutshell. That’s really a lot of things coming together from Terra Quantum in that one platform.

And as for the second big news that we recently had, it was our announcement of room-temperature superconductivity in graphite. So that is very exciting and has been, so in general, superconductivity and room temperature superconductivity have been in the news a lot. And there were some results that were not always great, doubted, and then sometimes even had to be retracted. In our results now in our experiments, we also make that claim that we can achieve room temperature superconductivity in graphite, I mean, ambient pressure. We’re still waiting for third parties to reproduce the results, but our experiments are very solid. So we’re very confident that we are on the right track here. That has a huge impact. So, in terms of the things that we can do, if we have materials that don’t have electrical resistance, could define and design better sensors, for example. Anytime you have electrons that you move through a material, you heat it up, you create vibrations. And in the radar system, for example, sensing devices, noise is a problem. It’s also very important for quantum chips. So if you know that superconducting quantum chips, they have to be cooled down at the moment to the mini Kelvin regime, which is very problematic if you want to scale these systems. So if you want to scale them up, you need more or more powerful cooling devices to get them still down to the operating temperature. It is challenging, so ideally, at some point, we can get rid of the fridge completely for superconducting devices. Energy transmission in general, is very interesting. If you had cables, superconducting cables that transmit energy or power without loss, that would be fantastic. So, if I were an energy company operating a grid, I would be very interested in that. But we’re not there yet, so it’s just a very fundamental result that we presented. And now it’s again about developing products based on that new result, which we are now starting.

Yuval: Let’s start with the software platform with a silly question. TQ42, I think I understand where TQ comes from, but what’s the 42 part?

Florian: That was an internal code name, actually. We started with that platform, and we wanted to, and still want to, make quantum computing access to it very simple. But then we noticed we bring in more and more components and then the joke came up. It’s almost like the answer to everything from the Hitchhiker’s Guide to the Galaxy. And that’s why TQ42 came as an internal code name, and it stuck. So now it’s also the external name, and that’s how it came about.

Yuval: There’s a school of thought that quantum algorithms need to be optimized for particular machines. Different machines have different numbers of qubits, different error characteristics, and different gates that they do better. And so the knock on the generic run-anywhere platforms is that you lose performance when you do that. How do you address that concern?

Florian: Yeah, that is very true. So that is where we have to optimize the algorithms and train the algorithms based on the system, the quantum system that is used. So if we had a neural network, a hybrid neural network, for example, the same set of rotations and potentially even the same set of gates that we use on one chip and of one architecture cannot be used on another chip. So that is really the idea behind it, that we have a set of algorithms that, based on the underlying hardware, based on the data, finds the optimal parameters and gives you the optimal result. So right now, how that happens, as you know a lot better than I do, because you came up recently, QuEra came up with that fantastic system, and people are now very excited I guess to design algorithms for that system. So it always takes this expert knowledge and experts. I’m spending a lot of time thinking about how to do it. And I think in the future, we still may need experts, but maybe we can make it a little easier when we look for these parameters using algorithms instead of having people look for it. And that would be the goal of this system. But I agree. So it’s definitely depending on the system.

Yuval: You mentioned hybrid algorithms, classical, quantum. Does TQ42 run on on-premises deployment? Does it run just on the cloud? Can the classical and quantum resources be in different places? How does it work from a deployment standpoint?

Florian: Yes, so that is a very good point. It is a cloud-first approach, so it runs in the cloud. Let’s say we deploy on one specific cloud, then the requirement would be that through that specific cloud, either a simulated quantum processor or physical quantum processor is available. So that is the requirement if you want to run it on one cloud environment. In the future, we may think about splitting it up. Say you have high-performance computing resources in one cloud and a quantum processor in another cloud. Through web service calls, yes. But then you want to really minimize the latency. So ideally, these systems sit close together. So, in classical, high-performance computing and quantum computing, so we minimize latency. Even more ideally, that’s our other great company in the family of tera-quantum companies, QMWare. You even have a hardware integration with the quantum processor or quantum system and then minimize the latency even further down to milliseconds. It’s very important for, say, variational algorithms. So if you have web service calls between high-performance compute and classical resources, that may just take too much time, and then you don’t win anything. But if you say and do it in an ideal scenario, only have the latency from the control hardware to the system, then that would be the way to go. And that’s the other platform, the other hybrid cloud that QMWare provides.

Yuval: You mentioned the platform includes an app store. Are these apps just applications that you guys develop? Or if I’m an application developer, I develop this new algorithm, can I perhaps deploy it on your system and benefit from the usage of other customers that use my application?

Florian: You ask all these questions. The things that I forget to mention this is always spot on. Yes, it is open to third-party developers. So it’s not only ours, it’s also open to our customers. So if a customer says we have this great application now that we either co-developed with Terra Quantum or we developed ourselves even without Terra Quantum, then you can deploy it there and monetize it. So it’s not that you just put it there, and Terra Quantum should earn money with it. It’s really that whoever deploys it, whoever develops it, should monetize it and earn money with it if they want to and sell it out to interested parties. Open to anyone.

Yuval: Quantum is often mentioned in the context of AI. This is just a fundraising trick to say quantum and AI, and everyone’s excited about AI, so they’re going to give me money for quantum. Some people say this is a very serious near-term connection. Where do you stand on that debate?

Florian: Yeah, I think in my personal view and also in view of Terra Quantum, it is a very serious connection. It is in the results that we obtain, achieve with our hybrid algorithms that we already see performance advantages with quantum computers versus algorithms that don’t use quantum computers. A good example is a convolutional network. So it’s a deep neural network you use for image classification, and you can nicely hybridize it. You can take, for example the last layer, the fully connected layer, and express that as a quantum circuit. And even with relatively few qubits, say 13, 14 qubits, you can achieve very high expressive power. Even higher expressive power than you would achieve with a fully connected classical layer. So now, if we take these two, the classical, the purely classical one, and the hybrid one, same architecture except replacing the fully connected layer with a quantum circuit, we obtain better results with the hybrid one. Better results mean we can train to the same performance as the classical one with less data. So that is very important because of the expressivity. Then, we also achieved better performance on test data. So that means we are faster and we are more accurate with less data. And that is more than an indication. We have a last performance benchmarking paper out there that shows exactly how that works and where the performance differences are. But this allows for a conclusion that for the algorithms that we use in machine learning, hybridization is the way forward. So for us, definitely the way to go, not only in the near term, but generally also as quantum chips, quantum systems become more powerful, we’ll see, in my view, a lot more things happening in AI.

Yuval: Given what you said about hybrid algorithms for machine learning, do you think we are at the point of quantum advantage today? Or are you comparing this hybrid quantum-classical algorithm to more of a generic classical implementation? How far are we there yet for quantum advantage?

Florian: Yes, so we are there. We are there. But I need to make a distinction. So, what we do these days is use simulated quantum processors. So, with physical quantum processors, there’s often the challenge of accuracy still. So, I know QuEra made a great experiment with logical qubits recently. So we’re very excited about this tool. That’s exactly what we need: logical qubits. We had that only in simulated quantum processors until recently, and that gave us the accuracy that we needed. But now, moving forward with physical systems, we expect a lot more. We expect faster training. We expect more accuracy, better accuracy as well. So, for us, when I say we are at the advantage today, then the advantage is based on hybrid algorithms as we tried it, using a simulated quantum processor up to 42 qubits. But that we have. So there is only indications, there’s evidence that we can outperform the same classical algorithms, the same configurations, and the same hardware, except this last part where we hybridize.

Yuval: And then we get into a little bit more of a philosophical discussion because if you’re running 42 qubits on a classical system, then where is the advantage? Is the advantage that there is a quantum-inspired algorithm, essentially a simulator, that’s doing that? Or is it real quantum hardware that will be able to deliver even better results if you had 60 or 100, or 256 qubits?

Florian:Quantum hardware will definitely be able to deliver better results. So, we see a performance or a cutoff in terms of the time that it takes for training. So with about between 20 and 25 qubits, simulated qubits, our physical quantum processor becomes better in terms of speed. The problem that we had in the past was always the accuracy, so we’re very excited about recent developments and potentially using logical physical qubits soon to test and verify how far we can get here. In terms of what it is, when you think about a simulated quantum processor, it’s really simulating the quantum effects in that system. It’s simulating the quantum system itself, so the chip is in that case. That then allows us to do things that are inherent to quantum computing. So even if we use the classical hardware only and process purely classically, we would not achieve better results as compared to a simulated quantum processor because if we treat this thing as if it were a physical quantum processor, it’s just a simulated one in that case. But it is quantum.

Yuval: You mentioned that TQ42 is open for closed beta, or maybe that’s a misnomer, open for closed beta, but where do you see traction in terms of the customer base? Is there a particular type of customer or particular industry that are already using it, that are excited more than others about using it?

Florian: Not a specific group of customers that is more excited than others. We just chose a group of friendly customers that we think are most interested based on the use cases they had in the past. And those we granted access, and they range from automotive to aerospace, so it’s not limited. We collect feedback, and soon once we’re done with this feedback phase, implement changes and then, later in the year, make it generally available. That will be then exciting to see how the broader user base will accept it and receive it. It’s a platform that is always growing, so it’s not something that is done at some point. It will grow with the applications, it will grow with new algorithms, it will grow with third parties bringing libraries in. It’s really a basic ecosystem, that’s why I like to call it an ecosystem, that we provide and give to the user base. And then it’s also about creating and maintaining a community. So we’ll do that. We’re very excited about that, but we’ll see where that goes.

Yuval: Now, going back to the room temperature superconducting, when you announced it, what was the reaction? Would people say, “Oh no, yet another room temperature superconducting report?” Or is there true excitement about it?

Florian: Yeah, it’s true excitement, but of course, there are also other mixed reactions. I think everyone is very, very hesitant to say this is what it is and confirm what it is because people may have burned their fingers the last time the results were announced, and then it turned out that maybe it’s not really the case. So that’s understandable. So we’re excited and positive about our own experimental results and now look forward to third parties to reproduce these results. And then we’re confident that it will be more and more positively received. But we already have very positive coverage already. But as I said, some mixed reactions as well, which is understandable.

Yuval: Where was the experiment performed? Is it TQ Labs? Is it some university that you’re partnering with?

Florian: Yes, it’s a university, two universities, actually. One is in South America and Brazil, and then it’s in the United States. Our chief scientist, who was our chief technology officer for the US and who was leading the research, is based in Chicago. That is the group that was doing that experiment. And to say it’s easy to reproduce would probably be too much. But what it is is based on graphite. So it’s not some mystery material. So, it’s something that people understand and have been researching for a long time. And there have been results in superconductivity using graphite already 10, 20 years ago. And it’s been well documented. So the hope would really be that people would soon be able to reproduce it and then confirm the results.

Yuval: How large is TerraQuantum today? How many people roughly are in the organization, and where are they based?

Florian: So it’s about 250 people, and we are based in a couple of locations. So across Europe, we have our headquarters in St. Gallen in Switzerland. We have our co-headquarter in Munich. We have a large office in the United Kingdom in London. We have one in San Francisco, and we have a couple of research locations. We have a hardware location in Finland and we have collaborations with universities in Italy, where we also share laboratories and lab space. So it’s really distributed across Europe and the United States.

Yuval: So you find yourself in a superposition of being in London and Geneva and St. Gallen at the same time?

Florian: I wish I could be in a superposition in all these locations, but I think we manage well. So we found ways. So for example, in software development, when people check in code in Europe, we just continue in the US. So, time zones can actually be an advantage if you do it right. And I think also COVID of course, taught all of us to work remotely. So it’s going well, but being together in the office is required. So it’s not that we all stay at home all the time. And especially if you have hardware locations, as you know, then staying at home is not an option. So, I really got to be there and then do the lab work.

Yuval: So, let’s talk about dinner. If you could have dinner with one of the quantum greats, dead or alive. Who would that be?

Florian: Max Planck.

Yuval: And why is that?

Florian: So because of the work he did, the research he did, the Planck function that he came up with as being the first one to find out there’s something going on. And I think what I did once was dig up an old, so there are not too many out there, but a video of Max Planck being interviewed. And I mean, it’s a like white video, but I already could see this is a fascinating or was a fascinating person. So I would love to have the chance to talk to him.

Yuval: Wonderful. Florian, thank you so much for joining me today.

Florian: Thank you very much, Yuval. As always, it was very nice.

For more market insights, check out our latest quantum computing news here.

 

The Future of Materials Discovery: Reducing R&D Costs significantly with GenMat’s AI and Machine Learning Tools

When: July 13, 2023 at 11:30am

What: GenMat Webinar

Picture of Jake Vikoren

Jake Vikoren

Company Speaker

Picture of Deep Prasad

Deep Prasad

Company Speaker

Picture of Araceli Venegas

Araceli Venegas

Company Speaker

Yuval Boger

Share this article:

Relevant

The Future of Materials Discovery: Reducing R&D Costs significantly with GenMat’s AI and Machine Learning Tools

When: July 13, 2023 at 11:30am

What: GenMat Webinar

Picture of Jake Vikoren

Jake Vikoren

Company Speaker

Picture of Deep Prasad

Deep Prasad

Company Speaker

Picture of Araceli Venegas

Araceli Venegas

Company Speaker

Keep track of everything going on in the Quantum Technology Market.

In one place.

Join Our Newsletter