Technical Perspective: Applying Design-Space Exploration to Quantum Architectures

Quantum computing machines have made tremendous progress over the last 20 years, with quantum operations improving in fidelity by three orders of magnitude. Machines of non-trivial size are now deployed as cloud services and are available 24/7. Yet systematic design-space studies of how to scale these machines have been lacking. In the following paper, Murali et al. present the first such study on scaling trapped-ion quantum architectures, and their results challenge the conventional wisdom of experimentalists in the field.

Trapped ions are a very promising technology for implementing quantum bits (qubits). They are extremely consistent and reliable. In fact, the technology is like that used for atomic clocks. Charged ions are trapped in an electromagnetic field. Each ion represents a qubit and logical operations on qubits are implemented by hitting the ions with laser pulses at specific frequencies. Qubit readout is also implemented with laser pulses, where ions will emit photons only in the "1" state (and the photons are detected with a camera).

Experimentalists trying to build large trapped-ion systems have generally fallen into two camps—those that focus on building small traps connected by moving ions between them, and those that focus on building large traps with many ions. The primary design goal is to minimize errors. Large traps avoid the errors that can occur from moving ions but can incur more errors than small traps due to interaction of the ions within the trap. The two leading industry players are good examples of these two camps. IonQ has had good success building machines from a single large trap. Honeywell has built some very reliable, smaller machines with very small traps connected by channels that ions can move through.

The following paper shows the optimal solution is somewhere in between and that it depends upon application characteristics. This is perhaps not surprising to classical system designers, but it was no small task to create the software tools and noise models required to map this design space for quantum machines. Moving ions heats them up, and this can cause them to experience higher error rates for the remainder of the computation. Trapping many ions in a trap requires slower operations between any two ions and can increase the error of those operations and upon other ions.

One of the key features of this paper is that it is a collaboration between computer scientists and physicists. To rigorously explore the design space, they built compiler and simulation tools that automatically mapped applications to a family of architectures, and then modeled the noise and resulting error on those architectures. This is exactly the kind of work we need to guide the evolution of scalable quantum systems, and which will be instrumental in accelerating our progress towards practical quantum computing.

One of the key features of the following paper is that it is a collaboration between computer scientists and physicists.

In fact, there is a critical shortage of research and trained researchers in the design of quantum computing systems. In a recent meeting with those in the quantum industry, I spoke with two leading development groups. One group was composed mostly of physicists that had done an excellent job of coming up with good technical solutions for each problem they encountered, but their manager felt they needed a better perspective on how to evaluate the larger design space and consider alternative solutions. The other group was composed of mostly classical control engineers, and their manager felt that they needed more quantum knowledge so that they could understand better what their designs are for. Clearly, some cross-training and interdisciplinary knowledge is needed.

This paper is a great example of what this kind of interdisciplinary work can accomplish. Specifically, neither extreme in the trapped-ion design space is likely to achieve the scalability that is needed in the long run. A happy medium, tailored to the characteristics of a target application domain, is the clear implication of this work. This result will substantially guide the evolution of future quantum machines, and I hope to see many more design-space studies of this nature as the field of quantum computer systems design grows.