newsfilter
← Latest News
Earnings Call Transcripts·

Quantum Computing in Practice: From Fundamentals to First Use Cases - ILOG SA

Good morning, everyone. Thank you for joining this webinar. Before we start, I'd like to go over some housekeeping items. At the bottom of your screen, you should see a Q&A box where you can submit your questions for Erik today. Later today, we'll send you an email with a link to the on-demand replay. So without further ado, I'll pass it over to you, Erik. Thank-you very much, Rikke. And welcome to all of you to this webinar quantum computing in practice from fundamentals to first use cases so as the title makes clear we will not only talk about the basic aspects of quantum computing but also make the switch to how you can put it in practice with already some nice examples on quantum advantage but let me tell more before doing so i would like to provide you with an overview of the teams that we are going to address today so we will start with first of all quantum computing how is it a differentiator for business science and for also scientific approaches um Secondly, we will also discuss a lot of use cases of quantum algorithms, give you examples of quantum advantage, and of course, it would not be complete without providing an evolution of the IBM quantum platform, and also perhaps some inspiration for your next steps. There is also the possibility to ask questions, as already said by Rikke. Don't hesitate to do that. On regular moments in time, I will also take a kind of logical stop. If there are interesting questions, I'll address them with the help of Wim. Okay, first of all, when we talk about quantum computing, perhaps it's not a bad idea at all. And we have here a twofold introduction. The left hand side is about technology. When we talk about quantum computing, it is based on a new, I would say, information science. We call it quantum information science and this in itself is based on the possibilities that are offered by quantum mechanics. Without going too much into detail here, quantum mechanics provides us with some phenomena that you find only on a very small particle level. And these phenomena are really implemented in the calculation unit of a quantum computer, and that is a qubit. On the left-hand side, you see here qubits, the term qubit, and you see also that a qubit can have states zero, a state one, but also, as you suggested by the sphere, not only this ground state zero and excited state one but also something between them. You could call it a linear combination of the zero and one. And that state is what we call a superposition state. And superposition is a phenomenon that you find in nature and that is applied in a quantum computer. So it's completely different from our classical bits that can only take values zero or one at a given moment in time that are discrete values. while a qubit can have let's say something between a superposition state and on in the picture you also see how this is used what do you see in the picture you see five qubits q0 q1 q2 q3 q4 and you see also quantum gates what are quantum gates that are these special pictures and the H picture, the plus picture, that are quantum gates. They are the alternative to our classical, I would say, Boolean gates, and that provide these special states. The H is the Hadamard gate, creating superposition. The blue one is connecting qubits, and what is that providing? It is providing entanglement, and entangement is really something very special in quantum mechanics as well. because it allows that or it makes it possible that one particle influences the state of another particle at once. We call that also a spooky action at a distance because it was really something special that was discovered already in the early 20th century and that was discovered recently observed recently in the CERN in Switzerland but in our quantum computer we can program that with what you can see here the blue gate is the CNOT gate so and here you see then a complete small quantum algorithm by the way this is the Bernstein-Vazirani algorithm that is not so important now and what is important is that you see the utilization of qubits quantum gates and in the gray also measurements because we are going to execute this experiment multiple times measure it also multiple times and look at the bit combinations that come out of it And the bit combinations that have the highest probability or the highest values in the histogram are the results of our calculation. So it is about probabilistic computing. Yet this probabilistic competing allows us to solve some real complex problems, as I will come back on in a couple of slides. So to make it short, it's a very new way of doing computing. Yeah. Is this promising? Yes, we will be able to solve very complex problems with quantum computing and a lot of industries, a lot scientists, a lot researchers and also economists believe in it. And you see that on the right hand side, quantum computing has the potential to create up to 850 billion in value at tech maturity. is an expression by Boston Consulting Group. And so really we see a boom in, let's say, the business of quantum computing. And this is, of course, provided by a complete ecosystem on which I also will come back. So it's the right moment to invest with skills and perhaps even with investments in quantum computing When you go to the next slide, what are the application areas of quantum computing? And there are four big application areas. Let me go over them one by one with a couple of examples, without going too much in all technical details, of course. First of all, there is Hamiltonian simulation on the left-hand side. This is really about simulating Hamiltonians. What is a Hamiltonian? It is the description, it's a mathematical description, of the behavior of a system that consists of for instance atoms molecules and all other let's say elements and this is hamiltonian is describing the energy of the system and when you are able to solve the hamilton equation then you can also find let's see the status of certain chemical systems understand chemical reactions better and develop new materials new vaccines and so on so it's a world of chemistry you could say then you have optimization optimization is coming back in quite some industries so solving an optimization problem can be reduced for instance max cut problem traveling salesman problem knapsack problems and things like that often we will transform our problem in a quantum in an excuse me quadratic unconstrained binary optimization form a cubo or a high order unconstraint binary optimization form and then translate it into a quantum algorithm and solve that quantum algorithm an example is i already give it gave it a traveling salesman problem supply chain problems scheduling problems and more then we have the third column with machine learning and AI. And here we will see that part of the training and part of the inference will happen on the quantum computer. That can be for traditional machine learning, that can be from deep learning, it can be even more and more for large language models. The fourth column is about differential equations and this is about solving equations that are relatively complex we can do that already today but we need to make too much simplifications with quantum it is possible to bring the non-linear equation to a higher level in the dimensional Hilbert space in Hilbert's space make it a linear problem and get it solved in a more efficient way and we see this happening for instance in the world of aerodynamics and other engineering problems. The two most promising domains on the short term are Hamiltonian simulation optimization. On the midterm we will see much more happening in the world of machine learning and differential equations. And this brings us to quantum computing. We will not solve all challenges with quantum computing, No, we see in the middle here in the Venn diagram the set of easy problems you could call them. Why do we call them easy? This is based on the fact that we can solve them in a polynomial amount of time, p, meaning that if you have a number of variables, you will be able to solve the problem in a polynomials number of operations and time in terms of this number of variable. And this is what we are doing today. Even more, we are doing that for six decades in the same way. Of course, we have always improved this way of computing. And today we are dealing with fantastic things. Think about Internet of Things, think about AI, Gen AI, Agentic AI, and all this intelligence and other stuff. And this will not disappear, of course. It will also become better and better. But then we have the problems that you cannot solve or will not be able to solve at all with our classical systems and that is where quantum comes in we call that quantum easy and you see between parentheses bqp that stands for bounded quantum polynomial meaning that if you accept a certain error small error you can solve the problem in a polynomial amount of time with your quantum computer and that is where we see these types of problems that their quantum will commit i have put as an example complex supply chain but on the previous chart we saw other domains today i will not talk about the other two examples what do i mean with the other few examples decrypt rsa encryption rsa stands for for Ravist Shamir Adelman. What is that about? That is our classical asymmetric cryptography. So that is what we are using daily in all our IT business today. But it can be potentially decrypted by quantum computers which are sufficiently powerful. I mean by that, which have sufficient qubits and which are equipped with the right algorithm, being the Schorr algorithm or a successor of it. And this is the basis of the quantum threat and the reason of quantum-safe cryptography. And then I go to the left-hand side, the hard problems that are non-deterministic polynomial. There you find algorithms like the Crystal-Skyber encryption, meaning quantum-save cryptography, which cannot be broken by classical and also not by quantum computers, we think. So, and this is the world of quantum-safe cryptography that I will not discuss anymore today. I will really focus on the purple Venn diagram, I would say, in my picture. And a really interesting evolution is the evolution towards quantum-centric supercomputing. Let me illustrate that with a concrete use case. On the top right-hand side, you find here the companies who have been investing in this. That is Cleveland Clinic and IBM, and you also see the type of problem that was addressed. Protein ligand complexes, what is that about? we have as all human beings and also living organisms consist of proteins a protein is a chain of amino acids proteins fold all in the same way all the time and if they don't fold in the right way it is the cause of a disease so the what is really the objective is to find molecules or regions that bind to the proteins so you can influence the protein folding in the right way. So this experiment was about computer simulation of proteins and the molecules that can bind to these proteins. Yeah, that's the starting point. I go to the bottom of the slide and you see an evolution in time one because we were as you can see in 24 and 25 we were able to simulate with a quantum computer this protein legend complexes for 10 atoms 36 atoms 303 atoms not bad at all but when we take 26 early this year we see suddenly more than 11 000 atoms and in the same month even more than 12 000 atoms which is really a huge progress why was that possible it was based on three things first of all the combination of classical computing ai and quantum computing and you see that on the left hand side with ai first of al the complexes were modeled using a mean field approach, meaning you make a little bit of an abstraction of the exceptions and you represent all electrons with their, let's say, mean orbitals. You simplify the orbitals in which the electrons position themselves as a starting point. With this starting point, based on AI, you go to your QPUs, your quantum processing units and then you use a quantum algorithm the local unitary coupled gestural circuit lucj circuit which is really going to the details and let me very shortly explain how that works and you are going to represent the electrons and the correlation between electrons in a very efficient way that is still representative for reality while being able to be deployed locally on a quantum computer with let's say with a certain topology of qubits so you make the link between your chemistry science and your computational science in quantum when this operation is done you are going to use your classical cpu let's say to do the final post processing and to find the right energy values using the sample quantum diagonalization algorithm. To make a long story short, you combine quantum computing, AI, classical computing, and a new type of algorithm to find all these complexes. So that's really the future where we see quite some progress going on. I have a second example that might be complex, but in the end is very easy to understand, I think. Let me explain, it's about tritium production, a very recent example. What is tritum about? We all are talking about fusion reactions as a new way to generate energy. And this is based on the combination of tritium and other materials. And tritum is hydrogen, 3-hydrogen consisting of one proton and two neutrons. And we should be able to produce it because it is a fuel of our, let's say, fusion reaction. Where do we want to find tritur? We want to find it in molten salt, and molten salt as the name tells itself, it's salt brought at a very high temperature. There you find a lot of complexes of molecules and atoms, and when you can, the idea is to extract tritium from it, and as such, find fuel for your fusion reactions. In the middle of the slide, you see the approach starting from data, supercomputing, and quantum computing. We saw that by combining these, a new algorithm was found out with a very promising result. What was the promising result? How quantum computing really is performing better than only using classical computing. Let me explain. On the top you see unfragmented methods and fragmented methods. Unfragmented are the classical methods. Fragmented metals are metals in which you split up the complete problem, you solve the individual problem, and you also embed correlations between different sub-problems. And what you see is the method with the asterisk and the red asterisks, EWFCISQD, is a quantum method. And the black dot is the metal that brings us, let's say, the real values of the energy. The energy that is generated by this tritium. And you see multiple clusters, about nine clusters have been done experiments with, lithium, beryllium, and phosphor clusters, with 21 atoms, as you can see, 6 plus 3 plus 12. The top graph shows the difference between the energy, the zero energy, and the cluster. And you see that there are two clusters with relatively high energy that are very promising to be part of your, let's say, fusion energy, fusion reaction. Two of them are not very good. As you can see, they have a zero energy level. But more importantly, your red asterisk is really doing well in terms of the energy that is calculated. And on the bottom of the slide, you see how the difference of the energy between your quantum method in red and the gold method in yellow are really close to each other. To make a long story short here, quantum computing proves to provide the better estimation of your energy for this tritium that is used in fusion reaction. So just a very scientific example, but very useful. Then, a little bit more easier to consume. And I will pick out a couple of examples, and not all of them, because I already have been addressing some of them. I already had been addressing the Cleveland Clinic example. I already been addressing an example in the energy and utilities. Let me look at financial services with Vanguard on the left-hand side. you see the portfolio optimization example. What is that about? Portfolio optimization is a very big challenge in the financial services sector. Why? You want to compose a portfolio of investments while taking a minimum amount of risk and the highest possible return. That is a contradictory objective, of course. Today, we use the Markowitz approach to do that, but the team that has produced this paper that you find below and has been working on a new algorithm combining the quantum approximate optimization algorithm with a new type of cost function and a new way to calculate let's say the conditional value at risk And they found that the portfolio they were able to compose was, let's say, equivalent to what you can do with the classical computing today, even with a small number of qubits. So, another example in the world of quantum services sector. In Boeing, for instance, there they have been using quantum computing to find new types of metals that are lighter, that are very resistant to corrosion, and that allow a search to build new planes that have a longer life while being lighter and using less energy than today. So, quite some examples. Then you might say, how does this map to our roadmap? Because on this slide, you see our roadmap. and you might already have seen that in other presentations for some people in the room and if that i will come back on it by the way later on but what i want to tell at this moment in time is really the vertical aspect you see that in 26 you see on the bottom of the slide nighthawk indeed we have our most recent qpu is the nighthawke processor and it offers 120 qubits that are programmable it has 400 more than 400 physical qubits and 7502 qubit gates at high reliability so really let's say the next step in our hardware evolution but i want to tell also here the other things in blue in purple and in pink what is that about first of all the general idea is that we talk about the platform and a platform is not only hardware but also the supporting let's say tools software we would call it middleware what do you find first of we are working and we have invented what we call mapping tools you see that in What is a mapping tool? A mapping tool is really making the mapping, indeed, between your business or scientific problem and the representation in terms of quantum, qubits, gates, measurements in your quantum algorithm in Qiskit, quantum information science. So it's a mapping between the logical and, let's say, the implementation aspect. We also have then the benchmark, perhaps better, the profiling tools. Profiling tools are more, I would say, tools that allow you to test the performance, the reliability of the circuits that you are going to use. And in Qiskit, we have tools to do so. And then you have the Use Case Benchmark Toolkit, allowing you to make the comparison between your quantum computation and your classical computation, making sure you really create a value with quantum computing. And so these all go together more and more to do something useful with quantum computer. And this is an overview of the milestones. as you know we started in 2016 2016 with five qubits only accessible via the cloud in 23 we have already had some examples where we saw that quantum computation was able to be brute force computation with more than 100 qubits 5000 operations quantum advantage is something that we have demonstrated this year in three and even more examples and i will come back on these i will tell immediately that the advantage is proven on scientific examples not yet really the i would say the global business examples i would call them and then we are going more and more in the direction of full tolerant quantum computers and i will come back on that on that roadmap towards full full tolerant computers which you will have in 29 with logical qubits and 100 million operations. As you might know, today we still have some noisy qubits and we need error mitigation and error suppression techniques to get something useful out of the deeper circuits. If your circuit is relatively short or not very deep, shallow as we call it, shallow circuit, then we do not need really error mitigation, error suppression. techniques, but once your circuit becomes deeper, then it is important to have these techniques. Okay, talking about quantum advantage, this is really an interesting slide to explain. You see three curves. First of all, I will start at the bottom curve and that is showing how in terms of complexity the classical computers will need time number of operations to get the problem solved as you can see it grows exponentially for certain complexities then you have the straight line this is the full full tolerant quantum computers of course for easy problems they will be more expensive but once your problem becomes more and more complex it behaves relatively stable it's a straight line today we already have quantum computers that are still noisy but you can apply error mitigation suppression techniques and do useful things as you can see and as such you have the blue area which provides quantum advantage quantum advantage you could define it in a more general way by saying you have the situation in which quantum computing combined with classical does a better job than only classical and the better job is faster or more accurate but you also have a more scientific definition in the white paper you find on the left hand side it is two essential criteria you must be able to validate the accuracy and secondly it must be clear that quantum computing makes a difference in the algorithm yeah and when we talk about at the quantum advantage we really see two parallel tracks and the bottom to start there you see much more scientific oriented advantage and I will start by explaining that in the next slide. The top is much more about, I would say, the business oriented things, the real application things. What we see is that the first quantum advantage cases are really positioned at the bottom of the slide. Of course, the top of the slide will follow step by step. And let me now explain a couple of examples about the bottom of the slide. And the first one, at the top of this slide, you see here a curve. And don't let it impress you by these pictures. How do you need to read that? You see here magnetization of the Floquet cycle. The y-axis is the Floquet cycle, the y- axis is the magnetization. It's about calculating magnetization of electrons and the Floquets cycle is an execution of your Hamiltonian but you do it repetitively and you see also a straight classical limit line. Classical limit means that as from let's say 14 executions of your Hamiltonian you see that the classical computation which is in black is shifting away is diverting away from the real situation because the real situation is the middle dotted line and where you see the quantum computation is really doing a good job but this is the quantum computation with error mitigation and suppression techniques if you don't apply error mitigation suppression techniques you're at the yellow line so this is an example of quantum advantage where you apply quantum error mitigation and suppression by the way using KSEM provided by KETMA one of our partners in this area so So, it is an example of where classical computing is losing the track, while quantum computing is still doing a good job. Of course, the higher the number of forecast cycles, the more accurate your result is in terms of magnetization. Let me go to the second example. The second example is about scattering strength versus OLE. what is that orbital localization error so it's the measuring of an error of your orbitals the scattering strength is the strength of your radiation and of course the higher the scattering strength the better it is the easier it is to detect it so it is not abnormal that your curves are descending but what you see the black line on the bottom is the quantum let's say behavior quantum computing simulation and the orbital localization error is low and stays consistency low compared to classical methods so it's an example of where the scattering strength is measured much more accurately by quantum computing than by classical computing in the world of course of electrons and other small particles. The bottom of the slide is a real computational example. Let me explain and it is about the comparison between quantum computation which you find in red and classical computation which are the the increasing lines in purple and in blue. What has been done here? You see at the bottom t-count. What is t? It's a t-gate. It's an non-clifford gate. So a non-Clifford gate is really making the difference in quantum computation. If you would not use Clifford gates at all and only pilot all the gates would only use Clifford gates makes my apologies then you would not see really a difference the T gate is a non-clifford gate and makes a difference and you see that with quantum computation you can simulate you can execute it that says consistently in time but classical computing needs an exponential amount of time to execute the t gates. You might say to me yeah but it is not exponential. Yes but this is because the scale is logarithmic. It's a logarithmatic scale so it's an exponential increase in amount of time required by the classical computers to execute the T gates. So it's not a real I would say type of it's really a quantum it's a really computational type of proof. By the way on the left hand side you also see quantum advantage benchmarking. We have a quantum advantage tracker tool in which scientists can document their quantum advantage evolution. It goes so fast that it goes even faster than the publishing of the papers themselves. So the publication of the The paper takes more time than the publication of quantum advantage to date in this advantage tracker. So if you are interested in that, don't hesitate to consult it. So these are the three typical examples. Three other examples, which are perhaps a little bit much more oriented towards, let's say, the time required for the execution. One of our partners, you see here four partners of IBM that are startups that are investing in quantum computation, that are working with us to solve certain problems. And Kibu has, for instance, done the quantum optimization algorithm in only 0.2 seconds compared to classical computing where it was required to use 17.5 seconds. or 30 QPU minutes compared to 3,360 hours for perturbed mirror circuits. What is a mirror circuit? It's a circuit that inverts itself. It is used to detect, let's say, noise in the circuits. Then another example is in the Hamilton-inspired VQE, you really think in chemistry and where you find also a sufficient considerable reduction in required qpu times compared to the cpu times and same for material simulation other examples so to make a long story short we see that quantum advantage but really on scientific level and we of course we want to make progress in the future as well in the more business oriented level um okay that was a little bit perhaps deep dive material but now how can we um and perhaps are there questions that i should address at this moment is there something that is interesting at this movement to tackle as an answer from a question I have one question that perhaps is interesting. The question is related to opportunities to go into the field of quantum computing. If this should be more vertically industry specific or overall horizontal or both perhaps? that's a good question I think it could be both if I take an example of course if we talk about chemistry it's really a vertical type of problem but there are problems that can also be if I understand the question considered to be horizontal and gave the example of the tritium improvement in simulation of tritum. The nice thing for instance is that this is of course in the world of energy generation of energy but it was inspired by the healthcare sector which national labs had come on the idea of doing this experiments based on their experience in the healthcare sector so i would say the problems are potentially vertical problems are certain certainly let's say important and very relevant but sometimes you need to look horizontally i would say. So a problem that was solved in certain industry can be reused in an other situation and that can also be for instance in the world of quantum machine learning where you do anomaly detection and the anomaly can be in multiple types of situations. It can be in finance but can also been in observations of natural phenomena or even observations of gravitational waves. So yeah, so I think both you have understood the question well. Yes, that was the only question for now. Yeah, don't hesitate, the participants are certainly welcome to ask more questions, certainly no problem at all. When you see this slide, what can i tell about that that is that um when we look at the evolution of quantum computing there are two parallel tracks the left one is of course about the systems themselves how are we going to configure them how are you going to improve your qubits your gates your error mitigation you are going to error correction that's important but the right hand side of course is also very important the algorithmic discovery and both go together and we are still in a situation where both influence each other the hardware influences of course how the algorithm is designed but also vice versa algorithm design has still a very big impact on how the computer hardware is built certainly if if you think about quantum centric supercomputing. So both go together. And when I go to the left-hand side of my previous screen, some people in the room might know that, but we have, of course, two types of deployments, you could say. The top is really on-site deployments. I would call them on-premise deployments and in certain campuses. for instance in europe we have in san sebastian a deployment a new deployment is announced in switzerland where we will also have a new quantum computing system of course this deployment is not this not so easy as a classical supercomputer you would say because you need to be prepared for special equipment like cryogenic equipment to cool down the qubits and to make the system operational. Do you need to do this investment which can be relatively huge? No. On the bottom of the slide, you see two quantum data centers and one of them in the US, one of them in Einingen in Germany and by the way also in the US we will have an installation that will be publicly available on another place in Chicago and in Europe of course we have also a quantum computing center and customers can opt for that and they can say I do all my quantum computation in the environment of Einingen in Germany, that is certainly an option. And so you can access these quantum computers by using an API token, by using a cloud resource name, and then you can access them in your applications. I will come back later on also on what the rules are, the plans to do that. As already said, One of the key components in the stack is Qiskit, Quantum Information Science Kit, open source but to a large extent developed also co-developed by IBM. I pick out some important elements in the Qiskits evolution as well. Today we have Qisk it 2.6.1 if i'm welcome so it already evolved for quite some years and it is intended to be really a an enterprise level enterprise grade type of section you see that on the bottom of the slide some constructs like the circuits you can create the transpiler let's say the compiler that transforms your circuit in an executable circuit for a certain quantum device you have um runtime primitives to do the sampling to do the estimation you also find the samplematic and i will come back and samplematic later on in the session because it's something very interesting something new the right hand side of the screen on the bottom you see that we can access multiple technologies so not only the ibm quantum technology but also competitive technologies to IBM, you would say, that also support Qiskit. So it's a kind of lingua franca, we say, for quantum computing. The layer on top of the black one is the blue one, and there you see that combination of quantum and high-performance computing. Let me explain a couple of these keywords. Slurm, simple Linux resource management, uniform resource management. It's a kind of solution that can be used in high performance computing to access high performance computing resources. Yeah. And we also have now the QRMI, the quantum resource management interface that is compatible with Slurm and that allows you to access the quantum resources, making abstraction of the specific implementation. So, suppose you have a quantum computer from IBM, or you have a quantum computers from IonQ, or from Pascal, or whatever. So, when they support QRMI, you can access them in your, let's say, your workflow of quantum and high-performance computing. And Prefect is exactly such a solution which allows you to define these hybrid workflows of quantum classical. We also see that Qiskit has a tradition of Python of course but also supports more and more other languages and this will only extend starting you might say Fortran an old language yes but still very important in the scientific world and on the top we see the purple layer and there you see at once what are at once that are extensions of qiskit with specific purposes that most of them are freely accessible or um yeah you need sometimes yeah for some of them you need to be in sufficiently high tier in your quantum access plan which i come back and let me pick out a couple of examples aqc is an interesting one i like to tell about it because it is the approximate quantum compilation let me explain we have a quantum circuit but it seems to be very noisy okay with this atom you can replace the original circuit with almost the same circuit logically speaking but performing much better as far as the noise is concerned so we have a low level noise circuit that is logically doing the same as your original circuit Then in the middle, PNA. Let me talk about PNA a little bit. It's Pauli noise amplification. It is the equivalent to a probabilistic error cancellation, but on the level of the Pauli gates. gates what does it mean we have a noisy gate you are going to amplify the noise you measure the outcome and extrapolate back to zero noise on the gate level on the power gate level not on the level of the circuit but on the individual gate level so a way to tackle the noise you could say and then i also would like to address sqd in the final column sample based quantum diagonalization in the world of chemistry allowing you to find the eigenvalues on the diagonal of a matrix of your quantum chemistry problem. So really evolutions that enrich your quantum Qiskit stack you could say. Functions also the last thing I want to tell here we also are developing more and more functions that extend the Qiskit stack so that you can address specific problems and business partners that are really specialized in these functions of which I already gave examples that can enrich the environment and support the complex use cases. So it's really more than technology it's becoming also a kind of business model. I already talked about the different plants. You see that on the bottom of the slide. Yes, we have commercial plants and... Can I just interrupt you because I have a question on the high performance computing and the quantum centric supercomputing. They ask how does this work in a realistic way? Is it like a combination of on-premise and public or both on-remise or are we looking for a solution in either? That's a very good question. It can be both. It depends on your use case and the use case comes with an acceptable delay of computation between your classical and quantum computer for some algorithms it doesn't matter that there is some it's not in real time that there's a little bit delay in the computation between quantum and classical then this classical computer the super computer and the quantum computer could even be on a distance of several kilometers even hundreds of kilometers but if your use case and there are of them that requires much more real-time interaction between your quantum and classical computing then they should be co-located and this would mean in reality that yeah you would ideally have your classical and quantum computer on your own premise yeah so um it depends on the use case and the delay you can accept in your use case we have a quantum centric supercomputing reference architecture paper that explains this very clearly so how you need to collocate depending on the delay depending on your use case and what difference is between real time, near time yeah so yeah so I need an answer now in general terms, but all configurations are possible, depending on the use cases you want to support. I hope it's completely sufficiently clear. I should show you the gentleman already who asked the question. We can send her or him the reference architecture paper, which is clearly explained. when to bring them together on-premise, or when it is accepted, they are in different data centers. And when they are different data center, it can be a cloud data center of course, and your own data center. Or even multiple data centers, yeah. I hope it was sufficient as an answer. I think so, Eric, I think. Please continue. Thanks. Thanks. Yeah. Okay. Good. And then we go to the bottom of this slide, you see the different plans. Interesting for people who would not yet know that you can get access for free to our quantum computers for 10 QPU minutes a month. Of course, I will be very honest, when you start doing serious things, that is not sufficient. And then you can take your credit card and pay $96 per minute you use ad hoc. But of course, a serious company likes to plan and likes to make budgets. What are we going to spend on quantum computing? And for these types of organizations, we have the flex plan and the premium plan. And you can see the flex pan is a little bit more expensive, $72 per QPU minute versus $48. Why is that? Because with the flex pen, you have all flexibility. You buy, for instance, I would say 400 QPU minutes, 30K, and during one year, you can use them whenever you want and at every moment. Of course, at the end of the year, only one year to use them, but you have all flexibility. When we use a premium plan, you have, let's say, you pay less, but there is a rolling, I would call, time window of 28 days. So each 28 days, you get 1,600 minutes that you can use. But if the month is over, then you lose your 2,600 units. So it is much more for the organizations who use the quantum computer in a more consistent way during the year. For both models, there are ideal customers, you could say. And then, of course, we have the on-premise plan. The on-remise plan is much better for the organization who wants to invest in their own quantum computer For reasons of, as we already addressed in the previous question, high performance computing with small delay, or for, I would say, really top security and having been the only tenant on the machine, because the other machines are, the other plants are multi-tenant. On top of that, you have the quantum platform with Qiskit, of course. The quantum network is not a physical network, but a combination of coordination, collaboration better between business partners and IBM and companies to do something useful with quantum computing. And there we find consultants and systems integrators. We find all types of organizations collaborating. And on top of the screen you see three types of examples. We have on the left and on the right hand side, I will start with the system integrator partners, really organizations that understand how to integrate quantum computing in the current application workflow, application and data workflows. So how can I leverage quantum computing for my business? So they come up with consulting. In the middle, you see software and SaaS partners that are really these partners that can develop, for instance, quantum functions, quantum algorithms. And on the left-hand side, quantum industry and technology services by IBM. So these are really our experts who know an industry and how to convert an industry problem into a quantum algorithm. And they work closely with people like Wim and myself, being the quantum ambassadors. Nice to know is also that in the world of quantum, of course, a lot of research is going on. And what you see here is a graph of what are the typical quantum computing systems where this research is tested out. And at this moment in time, you see that IBM is one of these leading platforms for the world of the scientists. But my goal of this slide is to show you that in the world of quantum computing, we have these traditional players like Google, Microsoft, Amazon, IBM. But you also have the other players, players you might not have heard about, like IQM from Finland, like IMQ, like D-Wave, perhaps already heard of it, Xanadu, Rigetti. So new types of players on the market that are only focusing on quantum computing. What we see is that almost all of them benefit from relatively high venture capital and investments. So there is a lot of belief in the market, in these companies, and there are even investment funds that invest in these types of companies. So, yeah, it's really, let's say, an ecosystem of suppliers. The market is not yet mature. What I want to say by that is that probably mergers and acquisitions are going to take place in the near future, as is happening with all types of, let's say, IT solutions. But at this moment, we see quite some in the market. And these are more the platforms, but we also have, of course, more and more algorithmic providers. Then this is the roadmap discussed partially. And first of all, interesting to notice is that IBM has been able to work in stock, meaning that IBM IBM has been able to really realize what was on the plan. So also in 26 we are again able to realize what we have been planning. And when we go to the bottom of this screen, you can see here some QPUs and connections between QPU's and further evolutions. So I am going to explain you more on this in the next coming slides. Please stay with me and I will tell you a little bit more what is going to happen. Because on this slide, Nighthawk, Starling, Blue Jay, you see names of QPU, names of birds also. So that is really about the QPUs that will be accessible by, let's say, our user community. But internal and within our own labs, we are also doing additional research and creating additional types of QPU's that are used as preparation for these final QPU'S that we will put in the commercial machines. So, I will show you more about that later on in this presentation, over a couple of minutes, by the way. Yeah, so, Nighthawk, I already addressed this type of processor, which is really the basis of the further evolution. I know this is deep hardware, but interesting to notice. Why? Because in our release 2 of Nighthawk, you see 120 programmable qubits with square connectivity. What is that about? First of all, these qubits, we could call them programmable, so you can really use them. But behind the scenes, there are 458 physical qubits. And so you already see a kind of logical type of qubits before we are in the world of Starlink. Square connectivity. You see that here in the grid. We can combine, we can connect three qubits, excuse me, four qubits of course with each other. In the past it were only three. Why is this important? Yeah, if you can do that, of course, you can accelerate your execution. Why? Because you can transpile your circuit in more efficient physical circuits which use less swap gates. So you can have, to make a long story short here, you kan transpile in more efficient quantum circuits. And by the way, In the future, we will have a connectivity of more, five, six qubits, which allows us to provide quantum error correction. We also see on-chip reset. What is that about? Sometimes, within a circuit, you must be able to reset the value of the qubit to the ground state zero again. For instance, because you have found an error or you want to re-initialize a certain part of your circuit. So since August, we have been doing that. We also see 100,000 circuits per second. It's about speed. And a comparison with the previous generation, the generation of processors. and that brings me to the following slide all details but I will give you a good impression about how we are evolving with our quantum error correction and I start at the left hand side error migration error migration mitigation is what we are doing today with probabilistic error cancellation tailored error mitigation and so on These are techniques that are really adding more samples, we are doing more samples to do more extra executions, adding more gates also, so that you can find patterns and correct the patterns of the errors. and we have dynamic decoupling, you have T-Rex, you have zero-noise extrapolation, and so on, and this is already used today. What we see here are two important elements that allow to do so, samplematic and dynamic circuits. Let me explain. Dynamic circuits is already known for some years. Dynamic circuit is a circuit in which you can program a kind of if-then-else, even do a reset, and, for instance, say, if there is a certain error, then I take action in real time. By the way, on the bottom of the slide, you see how KSM by Ketma is using this, yeah? But you also have another new way, samplematic. What is samplematic about? And this is a really interesting evolution. Let me explain. you have a quantum circuit let's say about you have hundred qubits and thousand gates to make it easy you could say i do error mitigation at the end of my circuit yeah okay or you could do it on the gate level with your pna probably noise amplification but you can also do something between suppose i know there are about 20 qubits under about 100 operations of which i don't want them to be noisy. Then I create, I can put them in a box of 20 qubits, 100 gates, I put them in a Box. I do experiments with them to measure the noise. I'll do that with an executor, they call that. I put a box, an executer, and then I do an annotation by adding quantum gates to eliminate the noise so to simplify it you take an extract of your circuit to improve the noise after doing some experiments of course it's not in real time here it takes time it's near time but it is a new construct that allows to improve current error mitigation techniques we also had exercises on that in the Qiskit Global Summer School 2026 in summer. But it is becoming more and more, let's say, a good way to do your error mitigation. When I go to the extreme on the other side of the picture, on the right hand side, we see there LDPC, low density parity check code. I come back on that in in a couple of minutes but also relay bp bp stands for belief propagation what does it mean it is really using um let's say multiple um code checks and the for each type of error the probability is indicated so you might have a probability on the level of your let's see um the coherence of your qubit, your gate, your measurements, the probability with the highest, excuse me, the error with the high probability is tackled, only that one. So it's a kind of optimization of your quantum error correction based on LDPC. I come back on LDPC later on in this slide, in the slide deck. I will not do it now. What I want to tell is that there are intermediate steps that will improve our error mitigation techniques to go into the direction of quantum error correction. And you see that with post-selected quantum error correction, what is that about? You only look at the results that are correct and the noisy results you eliminate. Of course, this is only possible to a certain level because if your circuit is very deep, everything is wrong, you could say. So it's only applicable for certain situations. On the bottom of the slide you see that it has already been applied by the University of Chicago and with space time codes they call that. So it is already applied. In the middle of the screen you see conditional quantum error correction. So you are going to apply quantum error correction based on an if-then-else logic. um it's based of course on dynamic circuits that's normal concatenated quantum error correction coders what is that about that is a hierarchical level so you have an error correction on level one level two level three level four to eliminate step by step the noise also algorithmic aware full tolerance what is it about yeah you develop your logical algorithm and you make sure that um the part of the algorithm that is uh very crucial in your result is noise free the part the algorithm of which you say yeah the noise okay i forget the qubits so i don't need the qubit anymore if they are noisy okay no problem so you look at the let's say the structure of your algorithm to do your error correction to make it more efficient so So a lot of research going on, as you can see. I'll come back in a couple of minutes on full-tolerant quantum computing. I know this is a little bit advanced, but I will not go too deep in it. If you are really interested in this, we have a nice paper explaining all the details. I just want to tell you that full-tollerant quantum competing is based on a Tanner graph, as you Can see on the top of the slide. oh every qubit is connected with three with let's say four other qubits excuse me not of course but also additional connections you see that in the black line on long distance and what we see these are the bivariate recycle codes with 144 data qubits 12 logical qubits and a distance of 12, meaning you can make 12 errors and you can recover that and you need 144 data qubits for 12 logical qubits. You could say the ratio is 1 over 12 and this is very advantageous compared to the surface code where you have only 1 over 100 ratio of a logical qubit versus physical qubit. If you are really interested how do they do that you can see on the bottom of the slide You combine, you have here 12 qubits, and a lot of CNOTs are used to accomplish this. And you have seven CNOT layers to make that happen. So you need more qubits to provide your error correction. And I go to the top of the previous slide. So you add more qubit to correct your errors. and compared to more samples of or executions of your circle okay and that is what i told a couple of slides before so instead of i would say you might think we only have the nighthawk we only have starling and blue j no within our labs we also are developing other types of qpus that pave the way for the commercial machines. In loom for instance the type of processor you will never access yourself in because it is an internal processor it uses C couplers and a C coupler is really making couplings within the cube U itself. So I go to the previous slide this is a C-coupler, so that black line and the black lines are C- couplers. And we see that these coupling techniques are doing very well, we make a lot of progress with that and they are behaving very well. Cocaburra is then combining, let's say, a logical processing unit and memory, indeed quantum memory, to support the computation. it's not of course long term it's a short-term memory we could call it a quantum ramp in the coca2 we are going to combine these modules with let's say an l coupling long distance coupling and long distance coupling is when we are talking about lower distances also there we see a lot of progress has been made what is the big challenge of course it must be operational also in a low temperature in a very cryogenic equipment and cryogenic situation um on the right hand side of course you also see more magic state distillation what is that about magic state installation that is a processing that is needed for the non-clifford gates and so the long clifford gates made a difference with classical computing because they allow for much more complexity and quantum typical quantum operations and we need quantum state distillation to do that and to do it efficiently and error free that will be part of the starling machines as you can see so continuous hardware evolution some people might already have seen this picture this is part of it does not exist part of this fake you could say part of which is real the real is on the right hand side quantum system two you see there are two machines with special configuration and the middle is the quantum computer with the control units and with also the control electronics in it but then we will see the combination of these machines and And BlueJ and Quantum Starlink and BlueJ are really entire sets of quantum computers connected together, as you can see. And, okay, so the quantum computing data centers of IBM, of course, are prepared for this. Erik, a question perhaps. There is a question more about the storage of the data that we need to process and compute. So, where does a quantum computer get its data or storage the data? And will it be in classical standard storage platforms? And if so, what connectivity are we talking about? What protocols does it support? Yeah, that's a very good question. The first answer is that today, and I think also in the near future, even in the mid-term, quantum computing is really about computing only. so it's not about storing any data storing qubits when i talked about quantum around which is really short term it's really storing the state of the qubit for a short amount of time to support the computation just like in a classical computer but long-term storage of qubits is not what we want to do long- term storage of data will be done from a classical way so and then i come to the answer completely and so how are we going then to use this data so in your application you will also take care of data access as you do today so you will need a database jdbc or whatever technology or you will use kafka to get data from your data store and once you have the classical data what you are going to do you are going to do a kind of conversion in your application from these classical data into qubits and in quantum gates we call that data encoding so but the data storage will be as today so no change in that and the protocol used you could say the protocol of the day to access your data so And the protocol used to access the quantum computer can be over the internet, using an API token, using a cloud resource name, and that can also be done over the networks you are using today. So IBM is not working at this moment on a quantum, I would call it a quantum type of communication network. that might be in the future. We have connections between quantum computers which are maintaining the quantum state. We call that the L-couplers I have been talking about, but there is also relatively short distance. It's a connection between quantum computers also in low-temperature environments. So I hope the answer was still understandable. Yes, thank you, Erik. Okay. By the way, yeah, and I'm going to talk about, yeah. Let's say the economic context. So IBM is really not withstanding everything that happens geopolitically, economically. We are going to invest 10 billion in the next five years in quantum compute and so there is a strong dedication and a strong belief in that this 10 billion of course will go to yeah the systems you build but also the talent you need to to do this you need to recruit right people to do the same so a lot of research talent is recruited all over the world to So, you might have heard about the fact that we are starting up a company called Andero, which is really specialized in creating the foundry, so the quantum chips, the QPUs for the quantum computers. computers and that will not only be the cube use for ibm but might maybe also cube use for other types of quantum computing providers by the way the department of commerce of us has also invested in it with 1 billion following the chips award you might know about the chips award it's an idea to get more chips production in the us um yeah the dream is of course that we would have something similar in europe but yeah okay um ibm is also acquiring hrl hr l is the abbreviation of a huge research laboratories um so the top of the screen here is not you should not mention laboratories again because it's already in hr but so it is a company and that is a little bit, why did IBM acquire it? Because you might say they are specializing in spin qubits and that is not what IBM is working on. IBM is really not superconducting qubits but HRL is also very strong in the cryogenic, let's say the world of cryogenic equipment and The working of qubits in a cryogenic situation, connecting qubits, quantum sensing, in which IBM is not yet active, and of course the talent they bring to the table. But don't be confused, the roadmap on superconducting qubits is not changed, so it will stay as is. The HRL acquisition will only help to make it true. That's how we can say it. Okay, then I go almost to the end of my presentation. This is also a site I always like to present at the end. Where are we with quantum computing, I would say, in the market? And then we take a look to a typical quantum journey and the four high-level phases. So quantum awareness is something what we are doing today. What is quantum about? What can it do? What are use cases? What is interesting for me? Then you have quantum evaluation, which you say, I'm interested in a couple of use cases, and I already want to do some experiments. And I nominate a quantum champion, someone who gets a little bit of time or even a lot of time to invest in quantum computing. And suppose one of the use cases makes sense. Then we can go to a quantum application, really applying it for a certain, I would say, a type of use case where quantum computing can bring benefit. Where are we today? We see most of the companies in the light blue and in the purple. But we see that a set of companies will make the transition to the pink one in the short to mid-term, of course. What we do not see yet is the green one, quantum transformation. Yes, it is really about business transformation based on quantum computing, as we see today with AI. In the past, we had plus AI. Now we see AI plus. and ai is a starting point for new business models we are not so far yet and that's too early today we are more in the pink with perhaps the first modest steps into the direction of uh yeah it should be the purple into the reaction of the pink yeah okay by the way um if you say i have heard here some interesting things I would like to learn more. Wim and I, we are very happy to come to your office or do an online session on quantum on certain domain with a lot of pleasure. We also have a very interesting quantum learning platform that you can do their studying at your own pace, going from very easy to really really advanced. So don't hesitate to look at it. I have put here the url on the bottom of the slide. Also it allows you to work with the real quantum systems as I already told you and using the free tier of 10 qubits a month. Okay so let's create the future of quantum computing together and I would like to conclude with one important statement that you see on the top of the slide here value of quantum of course will be created or determined by two things frontier circuits sample economics that seems complicated let me explain frontier circuits are circuits that are making a difference circuits in quantum computation with qubits quantum gates that do a better job than classical computing can do so and that is algorithmic research not so easy of course of course on the longer term we will see that we will reuse algorithms invented by some smart people and startups companies that have the time to do that but this is important that the logical circuits that i have been showing from the very start but then i come to my second term and this might be a little bit strange for you sample economics what is that about i told you at the start of my presentation that quantum computation is deterministic meaning you execute your experiment multiple times we call that a shock suppose i have thousand shots to do my probabilistic computing each shot costs money be it some yeah euros or euro cents but suppose now i have a fantastic algorithm but in order to get value out of it i need to do 1 billion shots and each shot cost 10 euro yeah then my algorithm is economically not okay you understand so the economics of my sample is too high if i on the other hand have a quantum algorithm that with 10 shots at the cost of one euro per shot and i find the solution with 10 euro that is economic acceptable so let's say that the the typical nature of quantum computation with the probabilistic aspect and the execution of number of shots is also important in your consideration for the value of quantum computation it's the logical aspect of your circuit and the execution aspect the sample economics i hope it is clear what i told but that's important for the future of quantum computation as well i give one final example there is a bernstein vazirani algorithm it's a theoretical example i know it is not really high value i know but with one shot you find a solution. And I know about practical algorithms that need a hundred thousands of shots to find a reasonable acceptable solution. So just to show you the difference. Okay, so thank you for attending, I would say, this session. Are there more questions, Wim, that show up and that we can tackle now yes perhaps shortly there is one question that's more related to pqc quantum safe cryptography so um but we of course it's an interesting topic and we need to also watch a bit the time for answering so i will read it so for public administrations that keep sensitive data for decades how should they prioritize the mitigation to quantum safe cryptography today given the harvest now decrypt later risk where would you start an inventory of the cryptographic assets or starting with the most exposed systems i have two more questions so we'll see if we can answer yeah yeah yeah that's a that's already an advanced question of course um of course and i will answer with one thing and perhaps we need to do that offline also what we try to create is what we call an application heat map what is a heat map that is a kind of two-dimensional structure in which we put together the criticality of your application and the effort needed to migrate to quantum safe and if you have an application that takes a lot of time to migrate to quantum save and that is really business critical it's a red flag it should be addressed at once if you haven't the other hand an application there is no business critical very fertile that is really easy to migrate to quantum safe you can wait with it so you can do that later on when you find some time so that is defining your priority but yeah you can tell a lot about it but perhaps we can reach out to the person and then yes yes and perhaps plan a more in-depth discussion yes sure because it's another topic yeah i have a final question that comes back on the let's say the economics slide that you just showed earlier so business outcomes um so i think business outcomes is more important for any customer and gets translated to price economics total cost of ownership return on investment etc hopefully the companies works towards this so that was one of the questions yes yes um companies can be considered Yeah, I would also, and I dare to say even, suppose that you are working today with your classical environment, you are doing a very good job, and you estimate that quantum cannot really bring a lot of additional absolute value. should not start with but what we say with quantum is try to find the problems you cannot solve today and you would really like to solve or also the problems where you see that a small improvement in your results brings a huge value i give an example fraud detection in finance suppose you can improve your fraud detection rate with one percent and you will you can make a difference of a billion on the air yeah then it's worthwhile to invest in content you understand so it depends on the the ROI you can find with it if this ROI is not high enough yeah you could say yeah why would I invest in it you're right yeah yeah that's certainly a consideration made by organizations and I support that consideration of course okay that was it eric yeah i just want to put here one slide on the the screen we also are organizing a second webinar on the 10th of december this will really focus on the financial services sector so today it was much more about technology about hardware, software, use cases in multiple sectors, quantum advantage, then we will focus really on financial services sector. It can make a difference, what are typical examples, which customers are already doing it, and so on. So don't hesitate to enroll. You will also receive an email, perhaps, or yeah just do it now with you have an enrollment url don't hesitate to do it okay then um there are no questions anymore then i want to thank you very much for your attention I hope it was sufficiently interesting and I hope to see you again on 10th of December for some people who are interested in the financial services sector. Of course, as I said, it might also be inspiring for other sectors. Thank you very much and then we can close the session, I think.

Company
ILOG SA
Keywords
ILOG SA