When to Go Multinode? A Novel Approach Aids Quantum Computer Designers

Researchers use a co-design approach to quantify performance tradeoffs in multinode superconducting quantum computers.

Dilution refrigerators keep superconducting qubits cold. — Photo by Andrea Starr, Pacific Northwest National Laboratory

The Science

Researchers have developed a method to evaluate and optimize the design of quantum computers that divide the computer into multiple nodes. Multinode architectures create large, powerful computers by linking individual processors together. The processors in these systems are connected via optical links. A particular challenge for superconducting quantum processors is that they must be operated at extremely low temperatures. This involves the use of special dilution refrigerators. These refrigerators cool components to temperatures lower than space and give superconducting quantum computers their unique appearance. Different nodes may be housed in separate fridges. This means the optical links must shuttle fragile quantum information between parts of the computer at different temperatures. The interconnects are currently hindered by noise, which makes communication between nodes difficult. This study provides a model to understand the performance tradeoffs of using a multinode architecture as compared to one that relies only on a single node or that uses conventional connections between nodes.

The Impact

A single superconducting quantum processor cannot be scaled up to larger and larger systems. One solution is to build quantum computers with multiple nodes. This research lays out a map towards distributed multi-processor superconducting quantum computers. Specifically, it evaluates the tradeoffs of internode vs. local communication. It also examines different ways of handling the quantum entanglement that stores information in quantum computers. The research draws on ways non-quantum high-performance computers jointly optimize their hardware and software. This approach could lead to advances in quantum networking and quantum computing. These computing systems could be important to energy and material sciences and other applications.

Summary

This paper investigates multinode quantum computers (MNQCs), where superconducting quantum processors, called nodes, are connected through optical interlinks to form a larger system. A key issue in these designs is that internode gates, which link the nodes, are noisier and slower than the local operations within a single node. The research takes a co-design approach, combining improvements in entanglement generation, distillation, and compilers to analyze and optimize system performance. It proposes a research roadmap for hardware and software advancements in MNQCs, with a focus on balancing internode and local computations.

The study also compared quantum interconnects to classical ones, i.e., links that do not preserve quantum information. The team found that quantum links—even if they are noisy—are more beneficial than classical internode links in most cases. The findings highlight a tradeoff between local and internode operations as well as entanglement generation and distillation, both of which affect algorithm performance and overall system efficiency.

Contact

Samuel Stein
Pacific Northwest National Laboratory
[email protected]

Funding

This material is based on work supported by the Department of Energy Office of Science, National Quantum Information Science Research Centers, Co-design Center for Quantum Advantage (C²QA).