Building bridges between quantum and classical computing

0:08- Hello, and welcome to the Physics World weekly - podcast. - I'm Hamish Johnston, and my guest in this - episode - is Yonatan - Cohen, - who is cofounder - and chief technology - officer - of Quantum Machines. - The company was started in Israel in 2018, - and it now has facilities - worldwide. - As Yonatan explains, - Quantum Machines develops control systems - for quantum computers, - providing a link between the quantum and classical - worlds.

0:53- Hi, Yonatan. - Welcome to the podcast. - Hi, Hamish. Thanks for having me. - So, Yonatan, can you give us a brief - introduction to the company? - When was it founded? How many employees do - you have? And what's the main business focus? - Yeah. So at Quantum Machines, - we are focused on, accelerating the realization of - useful quantum computers - by providing, - the control systems for quantum computers. - We're providing control systems for many, many different - players in the quantum industry.

1:31- The control system is the, - one of the core elements of a quantum - computer. It's a system it's a classical system, - not the quantum system, but the classical system - that operates the quantum hardware, - and orchestrates the the quantum operations that that - we perform on the quantum hardware. - We were founded in 2018. - We're three founders, - physicists. - We all did our PhDs, - together in the same lab at the Weizmann - Institute, in Israel.

2:00- And, - and, - yeah. And today, we are, - almost more than 300 people, - around the world. We're spreading over, 21 countries, - and we, have, more than 500 customers around - the world using our products. And, - yeah, it's very exciting. - And and why is it important to - integrate quantum and - classical systems? - Why do you have to do that to - do effective - quantum computing? And and what are some of - the challenges involved? - Yeah. So to begin with, - you know, we are classical - beings.

2:40- We live in the classical world, and we - need something to that bridges, - you know, our classical language, - and controls quantum quantum - hardware, at the quantum level. So that's that's - the main task of the control systems. Again, - it's a classical system - that but that interfaces the quantum hardware. So - it's really sits in this interface between quantum - and the the quantum hardware and the the - classical world. - So that's the first thing. But besides that, - it's very important these days to also connect - quantum processing, - with classical processing, - because almost everything we do on a quantum - computer involves some classical processing as well. So - it it starts from, - you know, bringing up, booting the quantum computer - where we run a sequence of - quantum programs, but then we take some data - from the quantum processor, and then we analyze - it on classical processors, - CPUs, GPUs, etcetera, to extract, - the parameters, - of of of the quantum system. And based - on that, we, we call it calibration of - the quantum computer. So we calibrate the quantum

3:48- computer. So all the way from calibrations - through then you have, things like error correction, - where you take data from the quantum processor - in real time, and then you, crunch the - numbers on the on the classical processor - to decode the errors, - and understand what what errors you have on - your quantum hardware, - in your quantum algorithm and fix them. And - all the way up to applications where at - the end, - many of the applications are hybrid applications by - themselves where the quantum processor only does, - a subtask, - sub subtask in the entire quantum application or - quantum classical application, - we should say. So everywhere from you see - from the calibration through the error equation all - the way up to the applications, - we need this kind of integration between quantum - and classical processing. We need to do some - quantum processing, - and then the results go to a classical - processor. We do some classical processing, and then - the result go again to the quantum processor.

4:46- So this interleaving of quantum and classical processing, - we see it all over the stack. So - it's very important to integrate - correctly, - these two, - different types of of, compute resources - in a way that we can orchestrate, - workflows, - of these quantum and classical processing tasks, - on the system. - And and I would imagine that there's there's - some big challenges here. I mean, I'm I'm - assuming one of the biggest challenges is the - fact that - quantum, - a quantum processor is a very delicate thing.

5:21- You know, it could be, you know, for - example, based on superconductors - that are cooled to near absolute zero. And, - actually, - just connecting to those things and getting the - information in and out - in an efficient way is - is is a big challenge. I mean, is - that something that you spend a lot of - time - thinking about? Or or are there other - important, - factors as well? - No. That's absolutely right. That's the first, that's - that's what we call the the analog front - end or sometimes we call it the quantum - the QPU front end. The front end of - our system, - where it integrates with the quantum system, that's - one of our our, main, - main, technology - developments and and main challenges.

6:08- And as you said, like, this is where - we really produce, - very delicate signals. - They need to be very clean signals. They're - still classical signals, but once they interact with - the qubits, - or the quantum elements on the quantum processor, - then they perform these quantum operations. And if - the signals - that our control system produces, - are not clean enough, - and, not shaped correctly, and and and the - noise level is too high, - then you then the basic quantum operations, - are not accurate enough. We call it the - fidelity of the operation. So the fidelity of - our quantum operations - would, could be actually affected - by the control system if the control system - is not designed, - correctly.

6:52- So and the same in the readout, you - mentioned, you know, both getting data in into - our qubits and then taking data back from - the qubits. How do we how do you - perform the readout - of your qubits? - How do you measure them, - with, the least amount of interference with your - quantum processor? - Again, you need to, have a very, - well thought of, - front end of the system. - And this - has has a lot to do with, - you know, generating and also, - acquiring and and and, - analog signals.

7:26- So this is the way we call it - the analog front end. So that's one big - challenge that we face. And, also, that challenge - becomes, like, 10 times harder, a 100 times - harder as you scale. Because today, you also - wanna scale the quantum hardware. - So, also, the analog front end of the - control system needs to scale dramatically. - And so not only are we trying to - generate very, very clean signals, we're trying to - generate many, many, many of these signals. And - you you wanna lower the cost per quick - qubit control. You wanna lower the power consumption - per qubit control. You wanna, lower the size - per qubit control.

8:01- All the, you know, things that you always, - do when you scale up a a technology, - while at the same time, you have to - keep, - producing very, very accurate signals. - So that's a big challenge. - Other than that, there are other challenges. - For example, how do you connect that control - system, - again, with classical, - high performance compute resources? - So how do you build the architecture - that the the chip that orchestrates - those analog pulses, analog signals that we send - to our quantum processor?

8:34- That chip needs to also, - communicate with classical processors. - What's the architecture of that chip? How do - we sequence those quantum operations? - How do we communicate with other, - compute resources in the in the in the - in the stack - in the right way? - As I mentioned, it will allow us to - orchestrate these these complex workflows. That's another, - whole, different channels challenge, - which is, so there are there are really - challenges across the stack. And then, of course, - you wanna create, like, good software on top - of it and good user experience because a - lot of a lot of what's happening in - the field is still, you know, of course, - research and development.

9:14- And it's very important that we bring, - you know, good, - good, user experience, what we call developer experience - to people that develop quantum computers to shorten - the r and d cycle - and give them a lot of flexibility. - Every day, there is a new paper to - we can do decoding this way. We can - do error correction this way. We can do - calibration that way. So keeping the the system - also very flexible, - is very, very important. - And and you understand, - my understanding is when you were doing academic - research, you were looking - at qubits. Did did you realize then that - when you were doing your research that there's - a gap in the market here? That, okay, - there's people trying to build really good qubits, - but maybe people weren't thinking too much about, - you know, the interface with the real world, - and that's where you and your colleagues - got your ideas - to to to form a company. Is is - that was that the the eureka moment?

10:19- Yeah. So, we are three founders, - and, one of us, - Nissim, my cofounder, - did his post doc at Yale. So he - actually finished his PhD a few years, before - Itamar and I, and he went to do - his postdoc at Yale, - with, probably, you know, the the sort of - most famous, - superconducting - qubits, - group - in in academia, - at Yale, with professor Rob Shulkov and and, - and professor Michel de Verere that just won - the Nobel Prize recently in physics.

10:52- And when he did his postdoc, - that's he worked on, - on the quantum error correction. - And he had to develop, - a lot of things in the control system - over there in order to be able to - enable this kind of experiments. So, - yes. When when when, - when we started, - when when when when we started searching, - for our startup, - Itamar and I, - who we, - we called Nissim at some point when we - start started focusing on quantum. - And then, we interviewed him, and we interviewed - we knew his work really well, but we - interviewed also him and what's what's happening in - the lab. And we interviewed other people that - that did similar works.

11:36- And then we, yes, we realized that, - not only do we sort of have a - sort of an unfair advantage because this really - brought a lot of knowledge in, the control - systems, - But, - also, we, we realized that, yes, there's going - to be a serious unmet need - for, - next generation control systems, for quantum control, - because people were using, - things that were not dedicated for quantum at - that time. - And even the ones who did who did - develop their own - quantum control systems, - were still basing it on on sort of - very simple architectures that we thought are not - going to give, - enough, - flexibility in what's needed in order to to - to build, for instance, quantum error correction or - other things that that would be needed in - the field. So, yes, we we that that - so it wasn't in our PhD, but it - was in in Nissim's postdoc, - based on on on what he did in - his postdoc. We that was the the the - beginning of - the the, - the understanding that that, that there is a - potential serious potential here to to to build - something which will be very valuable for the

12:47- community. - And you mentioned superconducting qubits, and I I - I think that's - mostly - what, quantum machines - works with. But there are other, you know, - sort of qubit designs. There's trapped ions, - spins in solids. - And, - I don't think it's clear which of these - qubit technologies will ultimately - be useful. - Does this make it difficult to develop - quantum classical integration - technologies, you know, looking to the future, - when, you know, you don't really know, - you know, ten, twenty years down the road, - which - qubits people will be working with? Or or - is there sort of a universal - nature to what you're developing - in terms of qubit technology?

13:39- Yeah. So that's a great question. So first - of all, at Quantum Machines, we are working - with - many different, - QPUs, - many different qubit types. We have customers in - in all of the qubit types that you - mentioned and actually more. So we work with - trapped ions, quantum computers. We work we, we - support - neutral atoms based quantum computers. We have many - customers in in neutral atoms based quantum computers, - superconducting qubits, as you mentioned, spin qubits, - and, - and and more and more, NVCenters and and - and others.

14:13- So, - yeah, when we started thinking about how we - wanna build our control system and our control, - solutions and platforms, - we, yeah, we we one of the goals - we set for ourselves is to provide solutions - to all of these different qubit types. - And as you said, it's it's not a - trivial, decision because, there are differences between those - qubit types. So there are obviously differences also - in the control system, - requirements, - from one qubit to another.

14:41- However, there are also a lot of commonalities. - And so we were able - with, - you know, - a single control platform, - that does have different features and different, - front end modules, for example, for different qubit - types, but still a single control platform - where I think at least 50% of the - development - is is common. At least 50% of the - platform is common, - to to different cubic, - types. We're able to to really provide solutions, - to all of this. So, yes, it cost - in extra development and extra effort, but I - think it also has huge, - huge advantage.

15:22- Besides the fact that we're not betting - on who's going to be the the winner - yet, - which is great from a business perspective at - this stage, I believe, because, - yeah, we don't know yet who's going to - be the the the definite winner or if - there is going to be even a definite - winner. - But besides that, I think that also there - is a lot of value in, - you know, even just, like, understanding - what should be common to all the qubits, - how to build such an architecture, - where are the interfaces that are agnostic, where - are the interfaces that are not - agnostic? And so on and so forth.

15:57- And at the end of the day, also, - everyone to everyone wants to eventually integrate those - different types of quantum processors - to the the general data center. - So by building a platform where, you know, - the the higher, - layers of of our stack are agnostic - sort of agnostic, and the lower layers of - our stack are less agnostic and more specific - per qubit type, We are actually creating this - bridge between all of these, different qubit types - to, how you can integrate with a single - control layer to, the general - data center.

16:35- And that's something that is is very I - think, would become very valuable in the fields. - Right? At the end of the day, when - you integrate quantum computers to high performance compute - centers or to the, you know, big data - centers, big cloud cloud providers, the infrastructure, - You would want to talk to them with - a single language from a single, - platform. - I see. And so, I mean, it almost - sounds like - your vision of the future is that - there there could be you know, we could - be using multiple qubit technologies. You know, for - example, we could be using - trapped ion qubits to solve - a certain part of a problem, whereas - maybe superconducting - qubits are better - at solving another part of that problem, and - I don't know, linear optical computing - can do something else. I mean, is - is that, - I mean, is that something that, - you know, that you see as as as - a possibility in the future?

17:33- Yeah. I think it's definitely a possibility. - There are people that are working on on - these types of of things, and, - yeah, it's it's it's it's definitely a possibility - that we would use different types of quantum - computers, - that would be, - each would be better at at at its - own tasks, - and we would need to integrate all of - them under a single umbrella. - Right? - Where users eventually they might not even know - that they are working on this or that - quantum computer, - and we need to build this entire stack - that can can really integrate to these different - types of of hardware.

18:11- I see. - And and I wanted to talk, a bit - about, you know, sort of specific pro products, - from quantum machines. Early earlier this year, you - launched, - the open - acceleration - stack. - Can you describe this product and and talk - about how it's being used? - Yeah. So the open acceleration stack allows us - to connect - our control system, which essentially means to connect - all the QPUs, the quantum processors that we - are connected with, - to classical accelerators - classical compute accelerators - with very, very low latency links, - with high bandwidth.

18:54- So, - whether it's an NVIDIA GPU - or an AMD FPGA - or an Intel CPU, etcetera, etcetera, - we can now connect - our control system to that accelerator - with microsecond latency, - links - so that we can communicate. So then the - QPU - and the GPU, for instance, can communicate - over this very low latency link, a high - bandwidth, and share information. - And as I mentioned, this is very important, - because we need to do this interleaving between - quantum and classical processing.

19:28- So now we can virtually connect to any - classical accelerator, - from the QPU side, And, and this gives - a lot of value to our customers. They're - building, - calibration workflows, error correction workflows, and eventually application - workflows - using these these hybrid systems. - I see. And and so is that is - that being used using, - a number of different types of of quantum - processor? - Yes. Definitely. We have customers, - with spin qubits, for example. Dirac is a - company from Australia - that is heavily using this kind of, low - latency links - to, - stabilize their quantum processors, calibrate their quantum processors, - and, again, later on do the error correction, - using this link between our control system and - the, - the the accelerators.

20:24- In the same way, we have superconducting qubits - and - neutral atom companies, - doing the same thing. - So so yes. - Again, each each Qubitab has slightly different requirements, - but, I think this part of the stack - is actually very general - to to a lot to to everyone. - And, - another announcement that you made, earlier this year - is that your OPX - 1,000 - platform was used to control - a Rigetti nine qubit processor, - achieving a 99.5 - median - two qubit gate fidelity.

21:03- Can can you explain the significance - of this, - specific achievement? - Yeah. So this is, a great example, - that, - shows - something which is quite new in the field. - You know, there's a lot of, - I mean, so so far in the field - that there is there's been a lot of, - vertical integration. Right? So which is very natural - at the beginning of of of a new - technological field. You have, - you know, companies doing - everything, - the entire stack and doing vertical integration, - before you start seeing more, - sort of horizontal - expertise - being developed, in the stack.

21:49- So, - it used to be, you know, that the - company does both the quantum processor - and the calibration - stack or maybe even the control system themselves - and then the calibration stack and then bringing - up the the qubits and seeing that that - that they can achieve high fidelity operations on - their quantum, - on their quantum processors. - And, this is an example where we we - we came in, - to a Rigetti, - QPU. - Rigetti is now providing their QPUs. They also - provide a full stack on the computer, but - they can also provide you with only the - QPU, only the quantum processor.

22:25- And so we show that - using our control system and actually also our - calibration stack, - you can achieve this high fidelity operations on - the Rigetti QPU. So showing that we can - do that by, taking, - components from, you know, different vendors, the QPU - from Rigetti, the control from QM, and the - the the calibrations from QM. - We can really achieve state of the art - fidelities, - that shows that the the the now we're - really getting into this place where things are - starting to become more standardized.

22:57- You can really build this horizontal, - expertise - and, and and and and build a a - more, rich, - ecosystem - of players - focusing on different parts of the stack. - You you've mentioned two collaborative - projects, - that Quantum Machines is involved with. - Can you can you talk about, - what what else you're doing worldwide? - Well, we're doing a lot of different things. - We - have many, many different partnerships. We run here - the Israeli Quantum Computing Center in, in Tel - Aviv, which, by itself is hosting different chips - from around the world. We have a collaboration, - for instance, with Colab.

23:42- It's a startup, - from The US, that, - what like, the the the one of the - founders there is John Martinez, who's, - also one of the Nobel Prize laureates, - in physics, - from, you know, '25. And I - we've been working very, - closely with John to, - we we're hosting the the collab chips in - our our Israeli Quantum Computing Center here, demonstrating - different types of hybrid quantum classical - applications on on that chip. - We, - have, of course, our collaboration with with NVIDIA - where we, for the first time, created these - low latency links between the the control system - and the NVIDIA GPUs.

24:28- We're working now with AMD, integrating with their, - FPGAs. - We're working with, - with many other partners on this, - quantum classical hybridization, - part of the stack. - So, - yeah, it's kind of across the board. - Quantum Machines has recently acquired the Dutch company - QHarbor, - which has developed a data storage and visualization - software platform for scientists. - On the face of it, this seems like - a strange - acquisition for a quantum computing company. Why why - did you buy QHarbor?

25:07- So QHarbor, - we we we bought them for a for - a few reasons. - The the first, - is really, as we always say, is just - the team. - QHarbor, I mean, it's a small team, but, - very, very strong, - team on the technical level, and we just - clicked with the people, you know, and, felt - that there is a, you know, we want - these guys on board. - So that that's always, I think, one of - the key key aspects, - especially when the team is small. Right?

25:38- It's it's really about bringing, - you know, fantastic people to the team. Besides - that, we keep building more and more software - tools for our customers. And the software that - QHarbor built - seemed to us like, it can provide a - lot of, extra value to our customers. - I mean, how they how they store, manage, - standardized data and data - viewing, - and data analysis. - As I mentioned, quantum computing, - there's a lot of, - need to do data analysis and, and, data - management.

26:12- And so this tool was just, you know, - something that we thought could integrate really nicely - with our software platform and provide more value - to our customers. - Other than that, we are also, growing in - Delft specifically. We, just opened an office in - Delft, and we are growing there. So this - sort of aligned also with our our - our growth ambitious, - ambitious in Europe in general, but, - also, - specifically in in in the in The Netherlands.

26:40- So that was kind of everything clicked together, - and - and we we made the acquisition. - And, - Yonatan, I was hoping that we could sort - of finish off this interview by talking a - bit about - careers. - And and first, I'd like to ask you - about, you know, your own - career. - You, - you did a PhD, and I believe you - looked at Mariana - cubits. Is that right? - Can can you talk a bit about, - your, you know, your your reese the research - that you did? You know, why why you - did it? Why it's important - for, - for the development of quantum computers?

27:22- Sure. Yeah. So my research, - was, I need at least some of my - my research work was, around - searching for topological - quantum states of matter and, specifically, - my run of fermions in - semiconducting, - superconducting - hybrid - devices. - This was sort of when this business of - superconducting, - semiconducting, - hybrid devices - started to really gain a lot of attention - by the physics community, - because of a a couple of, important, - you know, works that found the signature of - these Majorana fermions. And, - and, - I'm now nowadays, we're seeing more and more - results from Microsoft around this, this this this, - this, this, - you know, physical phenomena that - are promising in the sense that maybe one - day we could use that to create, topological - quantum, - computing, - which means that basically you, the so those - those topological quantum states and specifically Majorana fermions - can form qubits - that, are protected from, you know, environments or - protected from noise, - in the environment, - in a in an inherent way in the - in the hardware. So it's just almost as

28:37- if your your hardware is doing the quantum - error correction for you. Right? - And it's very hard to make those, - qubits, - and, I think that there is still a - long way to go, - which is still a long research, - way to go to really - build - robust qubits out of these, - these type of of of devices. - But I think there's a lot of interesting - progress that's happening, and I hope that one - day we we will see this as a - as a platform that really competes to to - become a, you know, a a quantum computer.

29:15- So to me, it's very exciting. - In the doing my PhD, it was - an extremely interesting topic on the sort of - fundamental - physics side of things. - How can you create these - exotic, - systems that are - by the nature, by the mathematics of the - the the Hamiltonian, the, - the the way that we sort of engineer - the Hamiltonian, - from different materials, from magnetic fields, we can - actually create these very exotic phenomena that, - that, - that that that that, - that, - that, you know, - that that doesn't just appear like that in - nature, if you don't engineer a system. So - I think that's a fascinating topic, - and, - yeah, especially for research. And I hope that - one day it will also become a become - a a qubit platform, - that can can really scale.

30:12- And and were you working in the lab, - or are you were you a theorist, or - was it a bit of both? - No. I was an experimental physicist. So I - did a lot of, - fabrication work in the clean room, - fabricating devices, - both indium arsenide nanowires coupled to superconductors. - I worked also in gallium arsenide, - quantum all effect, fractional quantum all effect. So - I was, you know, working - a lot on, - making new type of devices and then measuring - them in dilution refrigerators, - doing, all kinds of different, - measurements to explore different phenomena.

30:54- And so, yes, I was an experimental physicist, - looking for this exotic, - phenomena - in topological quantum, quantum states of matter. - And some of our listeners, - I I can guarantee you will be interested - in careers - in quantum computing. - What advice would you give to a physics - graduate - who, - would like a career? - Maybe at quantum machines or at another quantum - computing company? - I think that's doing some serious hands on - work, - on what really interests you, - is the most important thing.

31:34- You know, really going down to the details - of, of the of of of the thing - of interest in doing you know, really getting - sort of your hands dirty with with stuff, - and going to the really nuts and bolts. - That's the best, first thing to to start - from. - And, - and besides that, you know, it's the usual, - do do do something that really, - interests you, - Because you have to succeed in this field, - you have to also work very hard. - And, the field is moving very, very fast - these days.

32:09- There's a lot that's happening. - You wanna succeed in this field. You have - to work very hard. So it better be, - in something that, that you find very interesting - and, and and it gives you, the motivation - to - to wake up in in every morning and - do and and get your hands dirty. - I would also say that that that, - I think, you - know, you know, it's, - as things progress, - you have to be a little sort of - multidisciplinary - in the sense that, - you you okay. So if you're a physicist, - of course, you're coming from the physics, but - it's critical that you, - know, - how to program. You understand, - software. You understand, - technology stacks in general, - how the hardware is built, how the software - is built, how things integrate, - understand a little bit of the engineering language, - both on in hardware and software. Mhmm. And - I think it's it's becoming very, very important - that - that people sort of understand - systems, - especially the physicists, because I think that's one - of the I think that's the the that's - one of the strengths of physicists is that

33:20- they can understand - systems sort of from the, you know, physics - all the way up to to software. - So, you know, if you're a physicist, you're - probably not going to be the the the - best programmer, - in the world already here because you haven't - done that, But you can understand how everything - integrates, - pretty well. So I think that that's something - that physicists should I should look at. - Well, that's great, Yonatan. Thanks so much for - coming on the podcast.

33:48- Thank you so much, Hamish. Thank you. - That was Yonatan - Cohen of Quantum Machines. - Thanks to Yonatan for his insights - into the challenges and opportunities - in interfacing - quantum and classical computing systems. - I'm afraid that's all the time we have - for this week's podcast. - I'm Hamish Johnston, and our producer is Fred - Ailes. - The music in Physics World's podcasts - is called - one three seven, - and it was composed and performed by the - physicist - Philip Moriarty.

34:32- We'll be back again next week.