Part 1: Basics and Cloud Vendors

Hybrid Classical/Quantum Computing Conceptual Diagram. Credit: DARPA

by Amara Graps

Hybrid Classical-Quantum Computing is one approach to meeting the end of Moore’s Law, the empirical relationship of the number of transistors in an integrated circuit (IC) over time, which reached a plateau after the early-2010s. The hybrid concept, which, today, dominates quantum computing implementations, joins quantum approaches, based on qubits, with classical computers, based on bits, and delegates specific computations to components in the classical-quantum system to which it is best suited. As an upcoming QNALYSIS series will highlight hybrid classical-quantum computing Use Cases, and hybrid classical-quantum computing has evolved considerably in the last four years, we describe the hybrid concepts and its variations, first.

Introduction to Hybrid Classical-Quantum Computing

Hybrid quantum computing works by creating an algorithm where portions of the algorithm are processed on a classical computer while other portions of the algorithm are performed on the quantum computer. This approach can be very useful for creating optimization loops to solve optimization problems. The quantum computer is first set up to calculate the output of a complex parameterized function that would be too difficult for a classical computer to evaluate efficiently. Then the classical computer will feed in an educated guess for the parameter (called an ansatz) to the quantum computer. After that, the quantum computer calculates the algorithm using the parameter it has received and feeds back to the classical computer the results. This loop will repeat over and over again with the classical computer feeding in different ansatz values as to find increasingly more optimum values from the results that the quantum computer is calculating. This loop can repeat hundreds or thousands of time until the optimum value has been found.

A successful hybrid computation has a fast control system and the right control points that allow the quantum and classical portions of the algorithms to effectively interact. One example of a fast control system is that by Quantum Machines for control of classical NVIDIA hardware with a variety of other quantum computer vendors. Each algorithm has its own control points for generating a quantum circuit with an n-qubit state with parameterized components. The classical computer subsequently measures the qubits and processes the measurement outcomes, to then instruct the quantum processor to alter slightly how the n-qubit state is prepared.  This cycle is repeated until the desired outcome is reached.

This approach is primarily intended for use with a high performance computer (HPC) and a real quantum processor unit (QPU). However, the training and experimentation of the quantum computer can also be emulated with a classical computer which can call vQPUs or virtual QPU. It is advantageous for the HPC to be colocated with the QPU in order to reduce latencies, but this approach can work if they are not colocated with some degradation in performance.

This approach accomplishes several goals:

  • The vQPUs, or virtual QPUs, allow novice users to become acquainted with quantum systems, prior to embarking on the entire quantum computer programming process.
  • Quantum emulation is, additionally, useful for investigating the challenges and prospects of quantum computing for one’s specific problems prior to using a quantum computer. 
  • Four qubits? Or forty qubits? The quantum simulator also provides a strategy to check the runtime and memory needed to run one’s application as compared to a real QPU. This feature is common for many cloud services, for example, by Dell for a singleQPU (IonQ) or by Amazon AWS Braket for multiple QPUs.
  • With today’s results in artificial intelligence (A.I.), hybrid classical-quantum systems can increase their performance by identifying workload characteristics via machine learning.

Cloud Vendors: The Right QPU?

You see the value that A.I. can bring to quantum computing. Once the workload features are identified, A.I. can match the appropriate QPUs or vQPUs for the intended workload output in time.  Single QPU cloud services with simulators have become a standard offer by quantum computing vendors. Cloud providers for Multiple QPU services plus simulators is a fast-evolving business ecosystem.  On the timescales of one year, you’ll see different QPUs, different simulation or workflow software.

Together, Single QPU and Multiple QPU cloud vendors have a niche knowledge for which QPUs are suitable for solving which problems. The two kinds of cloud services are sometimes referred as Quantum Computing as a Service (QCaaS). 

Examples.

  • The work by Karalekas et al, 2020, in the early years of the hybrid concept, demonstrated a cloud architecture for hybrid classical-quantum computing, using the Rigetti Aspen-4 16Q device as the system’s QPU. Their research investigated architectural bottlenecks in quantum-classical cloud for runtime efficiency, presented a framework for benchmarking the platform’s runtime performance, and they showed two enhancements: parametric compilation and active qubit reset.

In part 2 of The Many Faces of Hybrid Classical-Quantum Computing, we’ll continue with colocation, integration with high performance computers, and co-design.

August 24, 2024