HIGH PERFORMANCE COMPUTING g
in large quantities in a standard compute environment.’ Another benefit to diamond- based quantum systems is they can be used for edge computing systems, such as in robotics and autonomous vehicles. This is relevant to high performance
computing HPC and research centres in general because it allows quantum computing to be more easily integrated with classical computing architectures. This could help drive the adoption of the technology and allow more scientists and researchers to get access to quantum technology. But scaling these systems to the point of mass adoption is still some way off. There are significant challenges facing today’s quantum computing developers. One significant challenge is getting to the point where you’ve got enough qubits to provide a measurable performance improvement in some way. The second challenge is how to integrate a quantum computer with classical computing? While Quantum Brilliance wants to connect Quantum Processing Units (QPUs) to classical systems directly, some other organisations want to connect these systems via the cloud. However, Mattingley-Scott thinks this is a mistake. ‘Most companies are using the
“If you talk to almost all the hardware vendors, there will not be isolated quantum computers, at least for the foreseeable future – the next few decades”
cloud, so they’re looking at a cloud hybrid execution model. We believe – and I think history bears this out – the QPU needs to be physically as close to the other classical compute devices, like CPUs and GPUs, as possible. ‘We envisage a future in which you’ll
go into your computing centre and pull a blade out – maybe it’ll be a CPU blade, maybe it’ll be a GPU blade, or maybe it’ll be a hybrid, and it’ll hopefully be a Quantum Brilliance QPU sat next to AMD or Nvidia or Intel CPUs and GPUs. ‘Quantum computing must operate in a quantum-classical hybrid – it has to be the case,’ Mattingley-Scott continued. ‘If you talk to almost all the hardware vendors, there will not be isolated quantum computers churning away doing stuff, and then delivering their results, at least for the foreseeable future – the next few decades. It is all going to be hybrid.
6 Scientific Computing World Summer 2022 ‘In which case, bite the bullet and put
your QPU actually in an accelerator card. Next to the GPU, next to the CPU. And then you’re not worried about data throughput, latency times and interaction times,’ Mattingley-Scott concluded.
Quantum in the cloud
Quantinuum, on the other hand, is a company that has embraced the use of the cloud to help facilitate access to its prototype quantum systems. Quantinuum’s H1 generation of quantum computers is already commercially available. The Quantinuum H1 generation, currently consisting of two computers, the H1-1 and the H1-2, are fully accessible over the cloud and compatible with a variety of software frameworks. Tony Uttley, president and chief
operating officer at Quantinuum, highlights the company’s growth from both hardware and software provider. ‘Quantinuum is the combination of Cambridge Quantum with Honeywell Quantum Solutions. Honeywell Quantum Solutions did a lot of work directly with the products Cambridge Quantum Computing was making. ‘What we found as we were working
together as separate companies, was that most people who are developing hardware will extrapolate away from the metal layer,’ Uttley explained. ‘They will make a separation to protect IP, and you can’t get the full integrated benefit if you have that separation layer. We realised we could make fully integrated solutions based upon both the application layer on top of the middleware on top of our hardware.’ However, although the platform is
based on the integration of two distinct companies, they also choose to make the software platform inclusive. ‘The applications, the operating system that we develop, is designed to work on everybody’s hardware,’ Uttley said. ‘And as a real practical example, we are one of the biggest users of IBM’s quantum computers in the world. IBM is also an investor in Quantinuum.’
Making use of quantum While the hardware stack continues to mature, scientists and researchers are now getting access to software development tools to create quantum algorithms and quantum simulators, or emulators, that allow them to simulate how a quantum computer might work in a classical system. This allows researchers to start to develop expertise and test out how applications might benefit them in the future. ‘A lot of the algorithmic work is in imagining this future where you don’t have to worry about qubits and how they
interact. Because all of that has been “taken care of” by universal fault tolerance,’ said Uttley. ‘That’s a decade away. So the key is, what do you do in the intervening time? How do you make progress? Can you do things with some of these earlier systems? And the answer is, yes you can. However, this requires users to begin
to think about their problems differently,’ stresses Uttley. ‘What I mean by that is, don’t think about what the problem is, and how you abstract that into the system. You need to think about what these systems are good at. And how do I use that for these kinds of problems?’
Not all qubits are created equally One critical aspect of the development of these quantum systems, particularly in the early days where coherence and error rates make these systems relatively unstable, is that different hardware architectures are more suited to different problems. A simple example of this would be a large number of qubits with a low coherence versus a much smaller number of qubits with a high coherence time. Another factor is the amount of communication required between qubits. ‘If you’re trying to simulate a molecule, then ostensibly it depends on how that molecule is shaped, believe it or not,’ Uttley said. ‘This is because what happens in a superconducting quantum computer and semiconducting ones are similar, where they have an architectural property that’s called “nearest neighbour”. This means the qubits are physically manufactured on a piece of silicon.’ Uttley gave an example of a grid or
several rows of qubits where the qubits can easily communicate with their nearest neighbour, but where communications from one side of the grid to the other take much longer and adds additional error to the system. ‘There are molecules where the shape of the molecule itself is kind of a nearest neighbour interaction,’ Uttley continued. ‘A nearest neighbour molecule running on a nearest neighbour quantum computer actually can work pretty effectively. But if you have complex molecules, where you need these qubits to talk to every other qubit arbitrarily, then our trapped ion hardware works well. This is because we can physically transport our qubits so that any one can talk to any other one without introducing any additional error. ‘It’s that kind of deep knowledge about both the problem and the way these systems work that allows us to know which hardware or platform will be most suitable for a given problem,’ Uttley concluded.
@scwmagazine |
www.scientific-computing.com
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38
