Distributed Quantum Computing Requires New Compilers to Manage Network Constraints

Quantum computing promises revolutionary advances, but building sufficiently powerful machines remains a formidable challenge, prompting researchers to explore distributed approaches that link multiple smaller processors. Leo Sünkell, Jonas Stein, and Maximilian Zorn, at the Institute for Informatics LMU Munich, along with their colleagues, tackle a critical software challenge in this emerging field: efficiently allocating quantum bits, or qubits, across a network of processors. Their work introduces new algorithms that minimise the costly communication between processors, considering not only the circuit’s structure but also the changing connectivity of the network over time. By employing and comparing time-aware qubit assignment and circuit optimisation techniques, the team demonstrates significant reductions in communication overhead, paving the way for more scalable and practical distributed quantum computers.

Distributed Quantum Computing and Scalability Challenges

The pursuit of practical quantum computers faces a significant hurdle: scalability. While quantum computing has made remarkable progress, building a machine with enough qubits to solve real-world problems remains a formidable challenge. Current quantum processors contain hundreds, even thousands, of qubits, but many more are needed to achieve a genuine quantum advantage, especially when implementing error correction. Distributed quantum computing offers a promising path forward, envisioning a network of smaller quantum computers working in concert to tackle problems beyond the reach of any single machine.

This approach aims to leverage the collective power of multiple processors, connected by quantum channels, to execute complex computations. However, connecting quantum computers introduces new challenges. Quantum communication, such as teleportation, is resource-intensive, requiring the creation and maintenance of entangled qubits. Establishing and sustaining entanglement, particularly over long distances, is complex and expensive, often requiring quantum repeaters. Therefore, a critical aspect of distributed quantum computing is minimizing the consumption of these valuable quantum resources, necessitating strategies for efficiently assigning qubits to different quantum processors, minimizing communication, and optimizing computational cost.

Researchers at LMU Munich have been exploring innovative techniques to address this qubit assignment problem. Their work focuses on developing algorithms that consider both the evolving connectivity of a quantum circuit and the specific topology of the underlying quantum network. They investigate methods that can intelligently schedule operations, deciding when to use teleportation or other communication protocols, to minimize overall communication cost. This involves formulating the problem as an optimization challenge, employing classical algorithms like simulated annealing and evolutionary algorithms to find the most efficient qubit assignments.

Beyond simply scheduling operations, the team also proposes a novel approach: directly optimizing the quantum circuit itself. Inspired by techniques used in quantum circuit optimization, they have developed an evolutionary algorithm that modifies the circuit’s structure to reduce the need for communication. Through rigorous testing against various circuit types and network topologies, the researchers demonstrate that their evolutionary algorithms consistently outperform traditional baseline methods, significantly reducing communication costs and paving the way for more scalable and practical distributed quantum computers. This work represents a crucial step towards realizing the full potential of quantum computing by addressing the critical challenge of scalability and opening up new possibilities for tackling complex problems.

Circuit Optimisation via Evolutionary Algorithms

The research includes both an evolutionary algorithm and a simulated annealing approach, alongside a technique which optimises the circuit itself, rather than the schedule. This involves a further evolutionary algorithm inspired by techniques from quantum circuit optimisation, modifying a given circuit to maximise a specified objective. Comparisons to baselines, including graph partitioning and sequential assignment, demonstrate that both evolutionary algorithms significantly reduce communication costs.

Evolutionary Algorithms Reduce Quantum Communication Costs

Results demonstrate performance gains when comparing the proposed methods to graph partitioning and sequential qubit assignment baselines. An evolutionary-based quantum circuit optimization algorithm, which adjusts the circuit itself rather than the schedule, aims to reduce overall communication cost. Evaluations against random circuits and different network topologies reveal that both evolutionary algorithms outperform the baseline in terms of communication cost reduction, suggesting potential for integration into a compilation framework designed for distributed quantum computing.

Reducing Communication Costs in Quantum Networks

Distributed quantum computation over quantum networks promises to scale quantum computing, but introduces challenges due to communication overhead. Novel optimisation problems emerge, alongside the need for new compilers. Researchers approached qubit assignment to quantum processing units from two directions: optimising the schedule through time-aware algorithms and applying an evolutionary-based quantum circuit optimisation algorithm to adjust the circuit itself, minimizing non-local operations. Both evolutionary algorithms significantly reduce communication cost compared to the baselines, ranging from 13% to 70% compared to greedy placement. The simulated annealing and evolutionary algorithm schedule optimisation techniques take advantage of considering the temporal dimension of the circuits, as well as the network topology, which enables improvements over methods that do not consider this information. The proposed methods could be integrated into compilation stacks for distributed quantum computation, and the quantum circuit optimisation approach warrants further investigation.