How to Build a Quantum Supercomputer: Can We Overcome the Scaling Challenges?

[Labs_Editorial]

Guest Authors: Masoud Mohseni, K. Grace Johnson, Ray Beausoleil

In 2019, the quantum computing industry claimed an important milestone called quantum supremacy, or quantum advantage, which marked the first time a small-size quantum computer was able to solve a problem that would take an impractical amount of time on a classical computer. However, that demonstration was on a highly contrived problem—sampling from a random quantum circuit—that has no known real-world application. Moreover, most of the reported quantum computational advantage was washed away by recent advances in the corresponding classical counterparts.

Given the progress on both quantum and high-performance classical computing technologies since 2019, there are some new questions the quantum community needs to answer: What are the fundamental or technical challenges to build a scalable quantum computer? How can we integrate such machines with state-of-the-art high-performance computing infrastructures? Is it possible to engineer a quantum-centric supercomputer within the next decade that can solve an industrially useful problem in a cost-effective way?

In “How to Build a Quantum Supercomputer: Scaling Challenges and Opportunities,”^[1] we address these questions by bringing together research groups from a diverse set of companies and organizations to design and build a full-stack platform for high-performance hybrid quantum-classical computing. We take a systems engineering approach, where one embraces the idea that many system parameters must be simultaneously optimized for a complex system. We begin with the quantum problem to be solved and the algorithm to be executed, then establish criteria for distributed execution and error correction needed to support the algorithm. Finally, we determine the hardware resources required to solve the selected problem.

The research arising from this systems engineering methodology reveals technical challenges to scaling that have only been discussed piecemeal in the existing literature. Most quantum computing systems to date have been focused on technical challenges at the 100-1000 qubit scale for noisy quantum devices limited by qubit quality. By anticipating technical challenges in scaling from one thousand to one million qubits, our research presents a holistic system that can provide practical quantum advantage. The technical roadmap was developed based on detailed resource estimates of classically challenging quantum simulation leveraging realistic error models and performance characteristics of superconducting qubits.

FIGURE 1: Since the quantum supremacy milestone in 2019, there has been an expectation that scaling the number of qubits will follow a Moore's Law (exponential) growth over time from 50 qubits to a million qubits at the end of this decade. This goal would mark the arrival of a practical fault-tolerant quantum computer. We plot the perceived and projected scaling progress with time including three actual data points, starting from the 2014 early error correction experiment performed at UCSB (nine qubits) [2], through the Google quantum supremacy on random circuits (53 qubits) [3], on to the most recent surface-code error correction experiment (105 qubits) [4]. Although the industry expectation of scaling has not been materialized, we argue that a more practical and cost-effective approach to utility-scale is possible by employing state-of-the-art semiconductor technology and incorporating quantum processors within a high-performance distributed quantum-classical framework.

Our observation is that the current pace of innovation is too slow to reach utility-scale quantum computing in the next decade (see graphic above). The optimistic industry expectation has not materialized; it also appears the double exponential desired trajectory is unlikely to happen. In our paper, we argue that a less ambitious but more practical target of building heterogeneous high-performance quantum-classical coprocessors could be met by employing state-of-the-art semiconductor fabrication and supercomputing. This could allow us to design and test high-quality quantum components and system integration at all intermediate scales with accelerated pace and reduced cost. We take advantage of the atomically accurate fabrication methods and design simulations pioneered by Qolab, Applied Materials, and Synopsys. For software scaling and real-time error correction, we explore how HPE, 1QBit, and Quantum Machines can augment HPC software stacks to take advantage of classical computing infrastructures such as NVIDIA DGX Quantum. The hardware and software systems engineering designs are provided by Qolab and HPE, respectively.

In our paper, we provide a comprehensive list of all known and overlooked technical challenges and the corresponding research and development opportunities that are needed to build a machine that can enable new discoveries. Not all solutions to the technical challenges we outline are addressed by our approach. We hope that publishing the technical challenges openly will build a bridge between the semiconductor manufacturing, classical HPC, and quantum computing communities to address outstanding challenges.

The world will know quantum computing is real when it enables the announcement of a major scientific or technological discovery by academia or industry, and the fact that it was obtained using a quantum computer is not as important as the result itself.

[1] M. Mohseni, et al., arXiv:2411.10406 (2024)

[2] J. Kelly, et al., Nature 519 (2015). 

[3] F. Arute et al., Nature 574, 505 (2019).

[4] R. Acharya, et al., arXiv:2408.13687 (2024).