Shopping the quantum computer supermarket

Realizing that there’s a competitive advantage to be found in speed and accuracy, top players in the financial industry have been investing in quantum computing, in some cases, for many years [1, 2, 3]. After all, quantum computers promise to solve certain classically intractable financial risk management problems such as large-scale derivatives pricing and portfolio optimization, as well as other computationally intensive problems such as Monte Carlo simulation. These companies knew early that developing intellectual property and filing patents would require an understanding of applicable use cases, how to formulate problems for quantum algorithms, how to efficiently encode these problems, how to implement these algorithms, and how to interpret the results.

But quantum algorithms don’t run in a vacuum. Well, technically, some quantum algorithms run in vacuums. But, idioms aside, you need a quantum computer.

Aisles of modalities

When shopping for a digital computer, you consider such properties as the type of processor, its clock speed, the amount of memory available, and the amount of storage available. Of course, your needs depend on what you’ll be using it for; a GPU isn’t necessary for writing articles, but it can be helpful with machine learning tasks. With quantum computers, at this point, unless you’re already partnered with a specific provider, you’re mostly considering the type of qubit and the quantity of qubits. Physical qubits are the quantum analog of the transistors inside our supercomputers and smartwatches, in that we do classical computation with transistors and we do quantum computation with physical qubits. With a few exceptions, however, physical qubits can be radically different from transistors.

If the industry was laid out like a supermarket, you would walk into the store and see aisles labeled with the following modalities:

• Atom, aka neutral atoms and cold atoms
• Cat
• Electron, aka spin and quantum dot
• Ion, aka ion trap and trapped ion
• Nitrogen Vacancy Center (NVC) in diamond
• Nuclear Magnetic Resonance (NMR)
• Photon
• Superconducting
• Topological
• Other

While it might seem like you’ve got a lot of research to do walking up and down these aisles, that’s not actually the case. Topological quantum computers, for example, despite some experimental progress, remain mostly theoretical [4], so those shelves are empty. Spin qubits and NVC qubits have classical software that emulates these devices, but there’s no actual hardware on the shelves.

The aisles that have actual quantum computers are:

• Atom
• Ion
• NVC
• NMR
• Photon
• Superconducting

After a quick walkthrough, you’ll notice that most quantum computers are still quite small. If you look at all the qubit counts, you’ll be able to focus your attention on two aisles:

• Atom
• Superconducting

Figure 1 "IBM Q quantum computer" by Lars Plougmann is licensed under CC BY-SA 2.0.

The two largest publicly-available quantum computers are the 256-atom “Aquila” by QuEra Computing [5] followed by a fleet of 127-qubit superconducting devices by IBM Quantum [6]. It’s important to note, at this stage, that all quantum computers are unique, so selecting a specific device can get a bit technical. However, Aquila is a singular entity and the IBM Quantum portal conveniently sorts the fleet qualitatively. It’s also worth noting that larger quantum computers exist, but these are mostly of engineering interest and they usually have restricted availability.

There are legitimate reasons to explore the other providers. Aquila is currently in analog mode, which will be explained in the next section. Also, Rigetti has an 84-qubit alternative to IBM Quantum that might be worthy of your consideration [7]. However, proofs-of-concept don’t necessarily require these qubit counts. You may be able to run your algorithms on smaller quantum computers, which opens up options regarding cost, availability, dedicated access, and more.

If availability is a concern, there are quantum computers that can be purchased or leased, including from the already-named providers. This is important, of course, for protecting intellectual property. Otherwise, quantum computers are accessed via the cloud. They may be offline when you want to use them, which you can’t do anything about, or they may have long queues from other users. Regarding queues, some providers offer premium access to circumvent this.

 Finally, there is always someone in the store talking about something called “quantum annealing.” Quantum annealers are not “universal” quantum computers, which are intended to solve a wide range of problems. Instead, they are looked at for optimization problems, and problems that can be formulated as optimization problems [8]. NEC and Qilimanjaro, besides superconducting quantum computers, are both building next generation quantum annealers, but they are not on shelves yet.

Digital vs analog

As Quantum computing is usually presented in a digital model, which features quantum circuits and quantum gates. While substantially different from classical computing, it is the relatively easy way to learn quantum computing. The digital model abstracts away much of the complexity of how the quantum computer hardware functions so that users can focus more on solving problems.
 

“Nature isn’t classical, dammit,
and if you want to make a simulation of nature,
you’d better make it quantum mechanical,
and by golly it’s a wonderful problem,
because it doesn’t look so easy.”

- Richard Feynman
 

Quantum systems, though, are inherently analog. As the quote above attests, Richard Feynman's challenge to the scientific community to build quantum computers was based on quantum systems not being classical, and therefore hard to simulate with digital computers. While the digital model is not classical computation, nature doesn't use circuits and gates. Therefore, there are performance reasons to use the analog model. The learning curve is steeper, however, because in addition to understanding algorithms, your developers need a deeper understanding of the hardware.

There is, therefore, a trade-off. The digital model sacrifices performance for ease of use, while the analog model sacrifices ease of use for performance. However, like classical computing, these models can be used together. One common example is to rapidly prototype software in the Python language, and then later translate it to C++ to optimize performance. Similarly, the digital model can be used for relatively rapid prototyping, and then the analog model can be used to optimize execution.

It is important to note that the use cases are mostly the same. After all, the digital model gets compiled into the analog model. However, the analog model may offer functionality that isn't found in the digital model. All digital models, for example, use two-level qubits, which ultimately represent either a 0 or a 1. However, analog models may allow higher levels, enabling algorithms that use 2, 3, 4, or higher. Furthermore, for just one example, the Rydberg state of analog model neutral atom quantum computers has been found to naturally solve Maximum Independent Set (MIS) problems [9], which are classically intractable problems.

Figure 2 "Ionenfalle - Quantencomputer" by Mnolf is licensed under CC BY-SA 3.0

Aisles of software

On the second floor of the supermarket, above the aisles of hardware, are aisles of software. Like the difference between buying an Android smartphone and an iPhone, you’re probably interested in what you can do with the hardware.

It is important to note that some software providers are specifically targeting the financial industry. That might make your software selection easier. Additional considerations include:

• Features. Like the differences between Microsoft Word and Notepad, there’s software that makes algorithm development easier and there are niche, singular-function utilities.
  • Application developers may be interested in platforms with APIs, such as QCentroid and aQuantum’s QPath, allowing quantum computation to be called from classical applications.
  • Algorithm developers may be interested in frameworks with code libraries, such as Qiskit and the Classiq Platform, which have building blocks for developing large algorithms.
  • Low-level developers may be interested in software such as Intel’s SDK and TU Delft’s LibKet, which allow exploration of quantum computing with C++.
• Languages. The three most commonly found quantum computing languages are Python, Julia, and C++, but there are others; the Python stack, in terms of libraries, is similar to that of classical machine learning.
• Compatibility. The software you’d like to use may work only with certain hardware providers. It may work with only one provider.

Selecting your software provider(s), therefore, can help you narrow down your hardware choices. If done in reverse, you may find your teams need to develop basic functionality from scratch. That’s not a bad thing, and it might be necessary anyway, but it can divert attention and resources away from algorithms, which is where competitive advantage is going to come from.

Shopping carts

To use hardware, you need software. It can be minimalistic, such as a quantum computing framework. However, you may want to consider putting together ensembles of software. For example, you may want a framework to develop your algorithms along with a utility software that improves the performance of your algorithms. You may also want a higher-level software, that builds upon some framework and makes it faster and easier to develop algorithms. You may even want pre-packaged cloud-leveraging ensembles, such as AWS Braket, Azure Quantum, qBraid, Strangeworks, Covalent, and/or QPath that either bundle multiple quantum resources for you or help you integrate various classical resources.

Your ensemble may have a limited set of hardware that it will support, and this may further help you choose a hardware provider. Also, keep in mind that classical software runs on classical processors, and this can be a bottleneck. All of the software on your laptop, in the cloud, on the control systems, and so forth, all use classical processors to ultimately operate quantum computers. You may want to save space in your cart for high-performance computing (HPC), in which case NVIDIA is a visible provider [10], to accelerate, where possible, any classical processing that might be slowing down overall performance.
 

Name brands

You can buy a generic digital computer and it will work. But some name brands have been around for several decades, and that longevity might be a factor in your purchase. While there are no quantum computing companies with such track records, we can look to the three major portals to speculate which hardware providers might be enjoying the most commercialization at the time of this writing. Extending our supermarket analogy, AWS Amazon Braket and Microsoft Azure Quantum can be thought of as minimarts of quantum computing providers, whereas IBM Quantum is a farm stand featuring primarily their own produce, albeit a lot of it.

• AWS Amazon Braket [11]
  • IonQ 
  • Oxford Quantum Circuits (OQC) 
  • QuEra 
  • Rigetti
• IBM Quantum
• Microsoft Azure Quantum [12]
  • IonQ
  • Pasqal
  • Quantum Circuits, Inc. (qci)
  • Quantinuum
  • Rigetti

Some of these providers are publicly traded, and you can gain insights from the available information. However, most hardware providers are still private and you can only find their funding information. A proposed list of quantum name brands can’t be limited to publicly traded companies because that would exclude the first two companies named in this article. QuEra is not publicly traded, and IBM Quantum is a division of IBM and doesn’t file separately.

These are far from being the only providers. You'll see many other names on the shelves, and several are worthy of your consideration. However, the reasons why the quantum computers of other providers are not available through these portals include:

• Some are not publicly available, and may never be. 
• Some are not publicly available but have been announced in advance.
• Some are available for sale or lease but are not accessible via the cloud.
• Some have their own public clouds. 
• Some require research proposals.
• Some have private clouds limited to select partners.
• Some are not really quantum.
• Some don't actually exist. 

It’s worth noting that some quantum computers are available via the cloud for free. They are mostly small, which limits what can be done with them, and they may have various restrictions. However, they allow non-committal exploration of quantum computing while any major decisions are pending

Figure 3 “Industrie Standard Quantum Computer” by Dieter Kühl is licensed under CC BY-SA 4.0.

Checkout lanes

Quantum computer pricing is not uniform. You can walk into an electronics store and find comparable laptops and comparable smartphones at comparable prices, but that is not the case yet in the quantum industry. The range of options you'll encounter includes:

• Free. As previously noted, some quantum computers are cloud-accessible for free, although there may be licensing limitations regarding commercial usage, etc.
• Pay-As-You-Go (PayGo). The fee structures vary wildly, so running one algorithm could cost from dollars to thousands of dollars.
• Lease. These numbers are not public, but some providers offer on-site leasing.
• Buy. Publicly available numbers range from USD 1 million to USD tens of millions. Software is mostly open source and free. There are some exceptions, for which fees are rarely disclosed. But even much of the proprietary software is currently available for free.

Figure 4 "Quantum dot circuit (24964113547)" by UCL Mathematical & Physical Sciences from London, UK is licensed under CC BY 2.0.

Reusable shopping bags

One additional consideration when considering costs, especially for on-site hardware, is that quantum computers are more energy-efficient than supercomputers. And they’re not just marginally more efficient, but they’re potentially significantly more energy efficient [13]. Therefore, while emphasis is understandably on speed and accuracy, a lesser-known “quantum advantage” is lower operating costs, at least in terms of your electric bill. Extending that further, some quantum computers have smaller footprints than GPU clusters and space savings can convert into cost savings. So, in all cases where quantum computers may be advantageous, they may be even more advantageous than you thought.
 

Unboxing

Quantum algorithms specifically for the financial industry date at least as far back as 2018 [14]. That's six years that early adopters have had to upskill their workforces, partner with quantum computing companies, and develop algorithms. So, while you're walking up and down the aisles and trying different hardware and software, the items to keep at the top of your shopping list are still use cases and algorithms. Ultimately, the best algorithms for your use cases are your path to quantum advantage, but you'll need compatible hardware and software to run them. Because no one quantum computer has been demonstrated to be the best choice under all circumstances, the quickest way to write a shopping list will be with QFStudio powered by Multiverse Computing or QCentroid.

The algorithms for your use cases will point you to compatible hardware. In the event you have multiple choices, you can do benchmarking right within the portal. Once you have your shopping list in hand, there are minimarts such as Amazon Braket, Azure Quantum, qBraid, and Strangeworks where you can find multiple brands in one place. With qBraid and Strangeworks, in particular, you may find additional brands with which to do continued benchmarking, or other hardware types to potentially adapt your algorithms to. And just like supermarkets that hand out free samples, the quantum computer supermarket has free samples, too. For example, you can try IBM Quantum and QuTech superconducting devices for free. You can also try Quandela and Xanadu photonic devices for free. And the qBraid minimart hands you a 1,000-credit coupon when you enter.

References

[1] "IBM Ramps Up Their Quantum Marketing Activities," Quantum Computing Report, 17 12 2017. [Online]. Available: https://quantumcomputingreport.com/ibm-ramps-up-their-quantum-marketing-activities/. [Accessed 31 1 2024].

[2] G. Lawton, "Barclays Bank takes a crack at IBM's quantum computer," TechTarget, 25 1 2019. [Online]. Available: https://www.techtarget.com/searchcio/feature/Barclays-Bank-takes-a-crack-at-IBMs-quantum-computer. [Accessed 31 1 2024].

[3] "Quantum Computing for Business," QC Ware, [Online]. Available: https://q2b2017.qcware.com/. [Accessed 31 1 2024].

[4] J. Langston, "In a historic milestone, Azure Quantum demonstrates formerly elusive physics needed to build scalable topological qubits," Microsoft, 14 3 2022. [Online]. Available: https://news.microsoft.com/source/features/innovation/azure-quantum-majorana-topological-qubit/. [Accessed 13 2 2024].

[5] "Aquila, Our 256-qubit quantum Computer," QuEra Computing Inc., [Online]. Available: https://www.quera.com/aquila. [Accessed 29 1 2024].

[6] "Compute resources," IBM Quantum Platform, [Online]. Available: https://quantum.ibm.com/services/resources?tab=systems. [Accessed 29 1 2024].

[7] "Scalable quantum systems built from the chip up to power practical applications," Rigetti & Co, LLC, [Online]. Available: https://www.rigetti.com/what-we-build. [Accessed 29 1 2024].

[8] J. Preskill, "Quantum Computing in the NISQ era and beyond," Quantum, vol. 2, p. 79, 2018.

[9] H. Pichler, S.-T. Wang, L. Zhou, S. Choi and M. D. Lukin, "Quantum Optimization for Maximum Independent Set Using Rydberg Atom Arrays," arXiv, 31 8 2018. [Online]. Available: https://arxiv.org/abs/1808.10816. [Accessed 13 2 2024].

[10] "News," Quantum Computing Report by GQI, [Online]. Available: https://quantumcomputingreport.com/news/. [Accessed 31 1 2024].

[11] "Amazon Braket Quantum Computers," Amazon Web Services, Inc., [Online]. Available: https://aws.amazon.com/braket/quantum-computers/. [Accessed 29 1 2024].

[12] "Azure Quantum cloud service," Microsoft, [Online]. Available: https://azure.microsoft.com/en-us/products/quantum/#features. [Accessed 29 1 2024].

[13] T. S. Humble, "Energy—Quantum computing efficiency," Oak Ridge National Laboratory, 1 2 2018. [Online]. Available: https://www.ornl.gov/news/energy-quantum-computing-efficiency. [Accessed 31 1 2024].

[14] P. Rebentrost, B. Gupt and T. R. Bromley, "Quantum computational finance: Monte Carlo pricing of financial derivatives," Physical Review A, vol. 98, no. 2, 2018.