A buyer’s guide to quantum as a service: Qubits for hire

Braket seems the most fitting prototype for the type of quantum computing service that could become commonplace, should QC essentially work the way we predict it should at present. It presumes its customer is a developer, and that this customer's objective is to use development tools to devise a quantum circuit, which is a simulated device whose schematics follow the principles of quantum physics as we understand them.

At a very deep level, all computer programs may be decompiled into a fundamental form called a Boolean circuit, whose logic may be depicted on a flowchart. A quantum circuit has a graphic form as well, and as with classical, binary computers, there are languages and symbologies that make it easier for a developer to express the logic of such a circuit. For instance, there's the so-called "Bra – Ket" notation that uses characters such as "|v>" and  to represent vectorization and linearization, respectively, when mapping the state of one quantum system to another, created by quantum physics pioneer Paul Dirac in 1934 [PDF], and for which AWS' system is named.

Suppose the broader application you're writing calls upon you to take a very complex audio waveform, and decompose it into a series of constituent, simpler waves that can be described in terms of frequency and amplitude. A classical Fourier transform algorithm can accomplish this, only with recursion -- it needs several iterations before arriving at a result. With a properly constructed quantum circuit, such a transform can be accomplished with just one pass.

Braket starts by giving you the tools you'd need to build the quantum circuit linguistically, and then test that circuit ("printing" it) against AWS' online quantum simulator -- an application running on the Amazon cloud, which garners $0.075 or $0.275 per active minute (first hour free), depending on service grade -- or the local simulator, which is an application you can run locally on your PC. Such a simulator doesn't quite provide a full solution, though it may reveal enough data to tell you whether the circuit is error-free. Note you don't have to spin up your own AWS virtual machine; the staging area provides just the resources you need to "print" the circuit.

After that point, AWS serves as a kind of broker for a handful of other providers' real (not simulated) quantum computers. Through an online marketplace that should look familiar enough to regular AWS users, Braket facilitates transactions that let you stage your tested circuit on one of these real QCs. You settle the price with your choice of QPU provider, at a base price of $0.30 per task, plus a surplus fee per "shot" (a very generalized way of symbolizing a program cycle). 

AWS supplies its own software development kit (SDK) for use with Braket. However, the quantum development community at large has settled upon Qiskit (pronounced "kiss - kit"), the most commonly employed, open source SDK at the time of this writing. Now marshaled by IBM, Qiskit is made to be utilized by other, common high-level libraries - for example, in conjunction with a Python library such as PennyLane.

While Amazon refrains from officially endorsing Qiskit, offering Braket's own SDK instead, Braket's caretakers publicly acknowledge its use, stopping just short of declaring it the de facto SDK. Indeed, an open source Qiskit provider(an interface to the service) for Braket is publicly available, and such providers are widely used.

"AWS' approach is to provide the infrastructure of the Internet," remarked Michael Brett, Braket's Principal Quantum Specialist, in response to an inquiry from ZDNet. "The tools that enable other people to build - if people are bringing their favorite library to connect into Braket, it is an open environment to do that work. So we provide the SDK so people can connect into it. Right now, we're not directly supporting Qiskit ourselves, but we've seen third-party libraries out there where there's a connection between Braket and Qiskit that has been built already. And we think that that's great."

BRAKET AT A GLANCE

Intended market: Academics, experimenters, scientific programmers

Class of service: Quantum circuit design, testing, and staging

Affordability: Affordable for tasks designed only to be run for a handful of minutes; not intended to be run for long durations as part of a public application

• AWS quantum computing service Braket now generally available

Quantum Inspire is an ongoing project by the Netherlands' Delft University of Technology (TU Delft), collaborating with the Dutch research organization TNO and commercial quantum products firm QuTech, to prototype a network of QC systems that can be put to use for commercial and academic purposes. In a March 2020 white paper introducing the platform, TU Delft's researchers explained that the intention of their system is to model not just the full quantum computing stack, but the entire computing economy that will make QaaS available and useful at some point:

Quantum Inspire comprises of a number of layers including quantum hardware, classical control electronics, and a software front-end with a cloud-accessible web-interface. Such a system is called a full-stack. Full-stack systems are essential test beds for understanding this novel computational paradigm. They can act as technology accelerators because only through careful analysis of the individual system layers and their interdependencies is it possible to detect the gaps and necessary next steps in the innovation roadmap and supply chain.

At the time of this writing, in addition to a pair of simulators, Quantum Inspire utilized two genuine QPU-based backends:

• Spin-2, comprised of a two-qubit quantum dot suspended in purified Silicon 28, designed more as an experiment in creating quantum processors using more conventional lithographic techniques [PDF];

• Starmon-5, comprised of five qubits in a X-formation, testing the resonant characteristics of qubits in close proximity to one another.

For this project, TU Delft developed its own quantum assembly language, entitled cQASM. It specifies a sequence of logical operations, more like a notation than a language. When a programmer specifies a quantum circuit using cQASM, the Quantum Inspire editor generates a graph of the circuit, like the simple one above. There, an "H" box represents a two-qubit logic gate, called a Hadamard gate. It's one of the visual tools used to generate an expression of a quantum circuit -- in quantum terms, to generate a Hamiltonian.

"As we speak, as of today, I think it's safe to say that we're no longer just validating concepts in the lab environment," remarked Richard Versluis, QuTech's lead quantum engineer, for the Inside Quantum Technology New York 2021 virtual conference. "We are stepping out of these dusty, old, noisy labs, and we're entering a new era where prototype products will soon start to deliver value - and after that, pilot production systems - are available in these early lead markets."

QUANTUM INSPIRE AT A GLANCE

Intended market: Materials science researchers

Class of service: Tests meant to test the reliability of systems and system design; also, advancing the art of programming systems

Affordability: Free, for those researchers willing to make this bargain with TU Delft: Everything that users submit to the system must be considered free and open source, so that the TU Delft and TNO teams may make use of it as well, along with the rest of the QC community.

• First multi-node quantum network paves the way for the quantum internet

There continues to be considerable debate in academic circles as to whether quantum annealing qualifies as quantum computing, since its ultimate objective is not so much to solve problems as to optimize or refine problem solving. While that debate goes on, one of the earliest organizations to offer any kind of functional quantum product, D-Wave Systems, offers cloud-based access to its annealing systems, by way of Leap (whose premium service tier is called Leap2).

Although annealing is only suited to modeling certain classes of problems, it's arguably well-suited. Rather than modeling an algorithm as a kind of circuit, as with conventional QC, annealing adopts a probabilistic model. In one sense, this model resembles the weights and biases applied to a neural network.

The trick for the developer is being able to rephrase the mathematical function in question in the form of an "energy model." Since qubits in quantum superposition (in-between 0 and 1) tend to seek a low-energy state, an energy model can leverage this fact of physics by building a kind of subatomic Plinko machine for solutions to fall into. Here, more probable results dig deeper pits for qubits to fall into, while less probable ones leave shallower pits. If you can accept the results of your operation as probabilities rather than facts, this method could yield more reliable probabilistic estimates. It's actually the same leap of faith, if you will, that researchers in neural computing science have made, and their investments have paid off handsomely.

So what really are Leap and Leap2 for? If you're a statistical researcher or mathematician, and you believe that algorithms are the most reliable ways to attain reliable probability measures, then there's a very good argument that annealing can yield the most reliable estimates possible of uncertainty levels. There's little plus-or-minus factor; you know how uncertain you are, and that can be a benefit.

LEAP / LEAP2 AT A GLANCE

Intended market: Small businesses and institutions seeking to build the first live, working models of real-world problem-solving using any kind of quantum technology

Class of service: Statistical analysis

Affordability: This depends on how good a quantum developer you are.  D-Wave charges per minute, for the duration of the annealing process -- not for data storage or transfer, but for how long D-Wave has to turn the crank. If you're an amateur developer, you could be wasting "shots," or production cycles, but if you know what you're doing, your payoff may come in the form of efficiency. (Leap is also available through Amazon Braket.)

• A quantum computer just solved a decades-old problem three million times faster than a classical computer

Until there is an official scorekeeper for this industry, Honeywell Quantum Solutions may have to be taken at its word that its trapped-ion qubit fabrication method enables up to 64 qubits to become entangled. Think of each qubit as a memory component. The number of possible superposition states for each entangled qubit (analogous to the length of a byte in binary computing) grows to 2, raised to the power of the number of qubits. So 64 is a much bigger stretch from 63 than 63 was from 62.

Honeywell's primary interest in QC is to eventually sell commercial hardware, but in the meantime, sell commercial quantum services. Up until recently, its primary means of building a customer base had been direct collaboration -- getting academia involved in the design process. In October 2020, the company took its outreach program one step further, by selling time on one of its H0 or H1 models, for a presumably nominal fee. It's very transparent about its motives: It's seeking greater involvement by potential customers, as a way of inspiring its own engineers to make process improvements for its forthcoming H2 model.

So this isn't exactly a serve-yourself, self-provisioning, cloud-like service: With Honeywell, the idea is that you're entering into a professional, customer relationship with the manufacturer, perhaps with the idea of giving its machines a test-drive.

H0 / H1 SERVICE AT A GLANCE

Intended market: Prospective customers of Honeywell quantum systems

Class of service: Performance testing -- essentially, taking the latest models out for a spin

Affordability: As with the first cloud services in the 2000s, your final price will be a matter of negotiation. For scheduling time on H0, the price you'll pay is based on Honeywell's determination of the size of the job you intend to run. That size is measured in "credits," and you purchase credits in advance in increments of 1,000. For H1, there are flat monthly subscription fees, which Honeywell will set based on how you intend to use it, and how much time that intended use case may consume.

• Quantum computing: Honeywell just quadrupled the power of its computer

In any discussion about a high-performance technology, IBM continues to present itself as the sole center of gravity in the universe.  IBM did make access to a 5-qubit mechanism available through its own public cloud in December 2015, in so doing invoking the phrase "first quantum processor on the cloud." Many took that to mean IBM had created both quantum processors and the quantum cloud. However, successful efforts by researchers at the US National Institute of Science and Technology (NIST) to make quantum processing accessible through classical means — albeit with more rudimentary mechanisms — predate IBM's by over six years.

As a portal for the curious, IBM Quantum Services is rich and well-documented — as one would expect from the company that produced FORTRAN. QS offers two development environments for individuals building their first quantum circuits, and who aren't quite certain what those are. Quantum Composer (depicted above) lets users build circuits out of dragged-and-dropped parts, like placing fingers on a guitar fret to produce a chord. Quantum Lab is the next step up, offering a notebook-like environment for producing circuits with Qiskit, a programming library designed to accompany an existing high-level language — most often Python, which is a versatile, interpreted language. 

While IBM is far from alone in having pioneered the QC space, the company can and should be credited for its very dogged determination to forge a path towards a viable commercial business model, somehow. With respect to services, IBM perceives two groups of customers, and is using its cloud effort to address them both simultaneously:

1.    Industries that are willing to make deep investments in supplementing its existing data centers with not only quantum mechanical systems, but the cryogenic refrigeration and superconductive power systems necessary to maintain them. For this group, IBM QS serves as the company's showroom for certain models capable of interfacing with classical cloud portals.

2.    Academics who naturally associate computing research with IBM, and who need a major brand name to help lift their projects out from the depths of obscurity.

With some nations still mired in their worst days of the pandemic, and others having just emerged from them, these two prospective customer groups are decidedly narrow. What's more, there's not a natural convergence point between the two.

Recently, researchers from corporations such as German automaker Daimler have collaborated with IBM in the establishment of what Daimler calls a "Quantum Initiative." According to a January 2020 white paper co-produced by the two [PDF], the 5-qubit IBM Q Valencia machine was used in the simulation of complex properties of electrolytic molecules used in lithium compound batteries.

But Q Valencia is being made available to a select, premium group of customers it calls Q Network. This makes Quantum Services not exactly the showroom for IBM's premium models, but rather more of a gathering point for academics outside the select Q Network community. For now, QS is something of a fluid entity, or what Werner Heisenberg might have called an uncertainty.

"What we have learned in the last five years, in a very practical sense," explained IBM Chief Quantum Exponent Dr. Bob Sutor, during the recent Inside Quantum Technology New York conference, "is that, while we are improving the quantum facilities themselves, increasing integration on the cloud of classical systems with the quantum systems, is going to be more and more important. . . In the direction we're going, there's going to be much closer integration of classical semiconductor computing capability, with the quantum systems. So there's the question of, how close to the quantum systems will we push which classical hardware? What will it do, and where will it live?"

IBM QUANTUM SERVICES AT A GLANCE

Intended market: Academics, researchers, and other interested parties who may be willing to participate in a proof-of-concept of Qiskit. Quantum circuits developed on IBM Quantum Services, which include both true and simulated QCs, may be shared among a broader community, and may be put to use in testing IBM's more premium Q Network systems.

Class of service: Experimental, from the most shallow toe-dips into the water, to deep-dive explorations.

Affordability: At the time of this writing, basic access to IBM's 23 full-time active quantum systems, and 5 simulators, is free. One does not have to be a IBM Cloud commercial customer or subscriber. However, one does have to have an IBMid account, which renders the account holder effectively an IBM Cloud user. Commercial services are available through the IBM Cloud, of course, and it's from there that the company expects to generate some revenue. But mostly, Quantum Services is a brand-building exercise for Q Network.

LEARN MORE

• Quantum computing: IBM's new tool lets users design quantum chips in minutes

Essentially, Strangeworks has built a collaboration platform for quantum researchers, encouraging them to come together and share their work with each other -- and the emerging community -- under an open source license. The platform was launched with a QC simulator only, though its executives have plans to facilitate direct access to a working QC facility. In the meantime, the platform serves as a marketplace where participants may serve as vendors.

The effort stemmed from founder William "Whurley" Hurley's work as chair of the IEEE's Quantum Computing Working Group. As his day job, Hurley built applications platforms either for, or to be acquired by, major enterprises. For instance, in 2014, Hurley built an investment advice expert system called Honest Dollar, which was acquired two years later by investment firm Goldman Sachs. Experiences such as this informed Hurley that scientific computing entrepreneurs need exposure to investment capital first and foremost. His objective for Strangeworks has been to bring together the class of people he's worked with at IEEE, and make their services and expertise accessible to potential benefactors, thus germinating the kind of ecosystem that QC has yet to attain under its own power.

QUANTUMCOMPUTING.COM AT A GLANCE

Intended market: Practitioners or prospective participants in the quantum computing research field -- people who expect to be in the industry, as opposed to its customers

Class of service: Experimental, and to the extent feasible, even recreational QC research

Affordability: Strangeworks QC is a free tier available to individual researchers, though with a 100 MB storage limit (thankfully, quantum circuit project files can be quite small) and no access to collaboration tools. These restrictions are lifted for the Strangeworks EQ "enterprise" tier, whose fees are negotiable. Some institutions may be skeptical about paying a platform to release its work as open source.

• Why we need to humanize quantum -- A discussion between Strangeworks CEO William Hurley and ZDNet's Tonya Hall