In October 2019, Google unveiled the first proof of quantum supremacy, demonstrating that a quantum computer could solve certain mathematical problems faster than a classical computer.
In March 2020, Honeywell announced that it would launch the first commercial quantum computer and in June announced the creation of the most powerful quantum computer yet. Coming in fast succession, these milestones indicate how far we’ve come since this technology was first postulated by Richard Feynman back in the 1980s.
In the next three to five years, such milestones will be reached with increasing frequency. Eventually, due to the inherent principles of the technology itself, the performance of quantum hardware will accelerate exponentially.
While future developments will unleash the full power of quantum computing, the reality is we can already harness some of that power today through the innovative orchestration of classical computers and existing quantum hardware.
John Preskill, professor of theoretical physics at the California Institute of Technology, called this existing hardware “noisy intermediate scale quantum.” Preskill called the technology “noisy” because we cannot yet adequately control the qubits – the “bits” used in quantum computing – however physically implemented. In the absence of greater control, the error rates involved when executing an algorithm across quantum gates – the logical circuits operating on a set of qubits – can be persistent and relatively high.
Preskill called the technology “intermediate scale” due to the number of qubits currently available on quantum devices. To achieve sustainable quantum supremacy, researchers estimate that we will need machines running between 208 and 420 qubits, depending on the type of circuit used. To put that in perspective, the most powerful machine unveiled by IBM boasts 53 qubits. Honeywell’s latest machine only has 6. This machine, however, has a stated “quantum volume” (a standard for measuring quantum power introduced by IBM) of 64, twice that of its closest competitor.
The question is: For organizations looking to build quantum computing capabilities in the NISQ era, what does the NISQ stack look like? In the following article, we will describe various aspects of this stack and provide a high-level overview of its implementation.
Hybrid by Necessity
Given the limitations of NISQ technology, the quantum stack will be hybrid by necessity, consisting primarily of classical computing components. These classical elements will handle a range of tasks from data preparation and parameter selection to post-processing and data analysis. The quantum elements of the workflow will be limited to very specific—albeit powerful—acceleration or co-processing roles for particular problems.
For the foreseeable future, quantum devices themselves will tend to be fairly specialized, with different types of devices (superconducting, trapped ion, photonic, and so on) particularly well-suited for different types of problems. The challenges posed by the hybrid nature of the stack require the implementation and management of workflows for the effective orchestration of the various components.
Future Compatible
Quantum technology will only continue to evolve, so the NISQ stack requires built-in flexibility to adapt to future innovation. The algorithms and IP developed today must both maximize the capabilities of NISQ and quantum-inspired devices while remaining open to emerging technologies, devices and approaches.
The quantum tools that industry and academia use today must be architected in a way that anticipates and accounts for this inevitable evolution. Creating high level workflows that can implement quantum algorithms on any hardware type represents one specific way to ensure future compatibility.
Replicable, Modular and Flexible at Scale
Working with quantum computing technology today involves trial and error. It is naturally iterative. Algorithms developed in the NISQ era, even those that theoretically can work on the “universal” quantum computers of the future, are heuristic in nature. As researchers and others refine their algorithms and workflows over time, they need to be able to replicate their current efforts on new technology and to experiment with evolving approaches.
The NISQ stack must support this iterative experimentation. Containerization has emerged as one way to provide flexibility, modularity and scalability, while also allowing plug-and-play options on backend devices (both classical and quantum).
The Importance of Workflow Management
The need to orchestrate both classical and quantum capabilities while accounting for their inherent differences benefit from containerization. The execution and composition of containers can be managed with workflows. This in turn calls for a comprehensive workflow management system to efficiently coordinate tasks and processes across the NISQ stack.
Isomorphic with the stack itself, these workflows must be future compatible (i.e., able to run across emerging hardware configurations). They must also be modular to facilitate experimentation and allow for ongoing optimization. Zapata Computing built Orquestra, a unified quantum operation environment, expressly for managing quantum workflows.
Visualizing the NISQ Stack
When thinking about the NISQ stack, it’s best to separate it into three separate functions.
On the front end are tools needed to create workflows along with the frameworks and libraries required to build quantum circuits (Cirq, Qiskit, PyQuill, etc.). Here you will also find specialized tools focused on the problem you are trying to solve (machine learning, optimization, modeling, chemical and molecular dynamics, and so on).
This part of the NISQ stack will be connected to your local infrastructure (e.g., your laptop for writing tasks and workflows in an editor, as well as managing workflows from your command line) through your workflow lifecycle management tool.
The next layer is where the hardware lives. This layer can include any of the existing quantum implementations – superconducting qubits, photonic qubits, ion traps – as well as quantum annealers. You will also find dedicated classical hardware here along with classically-based quantum circuit simulators.
Access to quantum hardware today is primarily cloud-based. For this reason, you will want to have containerized execution tools that connect to the relevant cloud environment. Your workflows will execute across this layer.
Finally, you need an analytics or data layer to analyze intermediate and final data from the workflows you run. This data will in turn inform iterations and replication of your workflows at scale.
From a workflow perspective, this layer will first and foremost house the data aggregation and correlation services responsible for collecting and organizing all the data created from a workflow run. It will also house your analytics tools, most commonly Jupyter Notebooks running Pandas in Python.
The last component consists of plotting and visualization tools: Matplotlib, Tableau or even Excel.
For the purposes of data management, this layer will also need to connect to a database, be it cloud-based or on prem.
Workflow Management: The Continuous Thread
While one might assume the quantum stack will change dramatically as quantum devices evolve, that is probably not the case. The quantum stack will be a quantum/classical hybrid for the foreseeable future. Existing technologies, from analytics and data visualization tools to high-performance computers, are and will continue to be perfectly suited to handle significant aspects of the quantum computing process.
Precisely because of its hybrid nature, the quantum stack will always require workflow management/orchestration. This layer will provide the necessary level of abstraction so that users can repeat, repurpose, and scale quantum processes while employing different quantum frameworks, languages, or hardware types. Given the central role that workflow management plays in the NISQ stack and beyond, it’s fair to say that it will serve as the fundamental enabler of the coming quantum revolution.
Tim Hirzel has a BA in Computer Science from Harvard University and an MS from MIT’s Media Lab. He brings extensive experience in managing teams working on performing data science, machine learning, quantum chemistry, and device simulation. Since 2005, Tim has been a software engineer and architect in science-based technology startups. Today he is focused on delivering a best in class quantum computing platform for Zapata and its customers.