Quantum Computer Access Speeds up with New Shared-Resource Operating System

Researchers are tackling the critical issue of resource scarcity in quantum computing, where high user demand routinely exceeds the availability of qubits. John Zhuoyang Ye from University of California, Los Angeles, Jiyuan Wang from Tulane University, Yifan Qiao from University of California, Berkeley, and Jens Palsberg et al. present HALO, a novel operating system designed to enable fine-grained resource sharing. This work is significant because it moves beyond traditional fair-share scheduling, introducing hardware-aware qubit sharing and a shot-adaptive scheduler to maximise hardware utilisation and minimise latency. Evaluations on the Torino computer demonstrate HALO’s potential, achieving up to a 2.44x improvement in hardware utilisation and a 4.44x increase in throughput compared to existing systems, representing a substantial step towards scalable and cost-effective quantum computation.

HALO operating system facilitates dynamic resource allocation and qubit sharing

A new quantum operating system, HALO, is transforming how quantum computers share resources and overcome a critical bottleneck in scalable computing. As quantum computing transitions to a cloud-based model, demand for processing time far exceeds the availability of quantum processors, often leaving users waiting for minutes or even days to begin their calculations.

This imbalance severely limits the potential of quantum technology, but HALO introduces a design that supports fine-grained resource-sharing, fundamentally altering hardware scheduling and enabling greater parallelism. The core of HALO lies in two complementary mechanisms designed to maximise efficiency.

A hardware-aware qubit-sharing algorithm intelligently places shared helper qubits, essential for many quantum algorithms, on the computer in locations that minimise routing overhead and reduce disruptive cross-talk between different user processes. Simultaneously, a shot-adaptive scheduler dynamically allocates execution windows based on each job’s specific sampling requirements, optimising throughput and minimising latency.

Evaluations of HALO on the IBM Torino quantum computer, utilising benchmarks that heavily rely on helper qubits, demonstrate substantial improvements over existing systems like HyperQ. Results show that HALO improves overall hardware utilisation by up to 2.44x, increasing throughput by 4.44x while maintaining fidelity loss within 33%.

These findings confirm the practicality of resource-sharing and represent a significant step towards cost-effective and scalable quantum computing. This innovative approach addresses limitations in current quantum cloud platforms, which typically employ exclusive-use models where a single job monopolises the entire device, even if it only utilises a small fraction of the available resources. HALO’s ability to dynamically share helper qubits and adapt to varying shot demands unlocks a new level of hardware efficiency, paving the way for broader access and faster results in the rapidly evolving field of quantum computation.

Dynamic Resource Allocation and Scheduling for Concurrent Quantum Computation

A hardware-aware qubit-sharing algorithm and a shot-adaptive scheduler form the core of HALO, a new operating system designed to enable fine-grained resource-sharing on quantum computers. The research addresses the imbalance between increasing user demand and limited quantum resources by introducing mechanisms for dynamic space and time sharing.

HALO dynamically allocates helper qubits between different processes, minimising noise through a cost function that jointly optimizes qubit placement, routing cost, and noise isolation during concurrent execution. This cost function considers both data-qubit and helper-qubit zone placement to mitigate cross-talk interference.

On the temporal side, a shot-adaptive scheduler allocates execution windows based on each job’s shot count and circuit depth. This scheduler balances processes with varying shot requirements, reducing idle time and improving overall batch utilisation. The system was evaluated using three helper qubit intensive benchmarks, stabilizer measurement circuits, classical arithmetic circuits, and multi-qubit controlled gates, to demonstrate its effectiveness.

The methodology incorporates careful control of qubit sharing to prevent cross-process entanglement, ensuring each job’s quantum state evolves independently. Before reassigning a helper qubit, HALO verifies complete disentanglement and resets it to the initial state |0⟩. This is achieved through precise coordination of space sharing, where helper qubits are reused, and time sharing, where execution windows are dynamically assigned.

Performance evaluations on the Torino computer reveal that HALO improves hardware utilisation by up to 2.44x and increases throughput by 4.44x, while maintaining fidelity loss within 33% compared to state-of-the-art systems like HyperQ. The developed cost functions explain fidelity degradation observed with increased throughput from experiments, validating the model’s accuracy.

Hardware-aware qubit sharing enhances utilisation and throughput on quantum hardware

Researchers developed HALO, a novel operating system design supporting fine-grained resource-sharing for quantum computers, achieving up to 2.44x improvement in overall hardware utilisation. Throughput increased by 4.44x when evaluating HALO on the Torino computer using helper qubit intensive benchmarks.

Fidelity loss was maintained within 33 percent, demonstrating the practicality of resource-sharing in a computing environment. The study addresses limitations in existing quantum cloud platforms, which typically employ exclusive-use models resulting in low qubit utilisation and long queue times. HALO introduces a hardware-aware qubit-sharing algorithm that strategically places shared helper qubits to minimise routing overhead and mitigate cross-talk noise between user processes.

This placement optimizes the co-location of circuits, addressing the challenge of increased noise when sharing qubits. Furthermore, HALO incorporates a shot-adaptive scheduler allocating execution windows based on each job’s sampling requirements. This scheduler accounts for variations in shot demand, ranging from a few shots for fault-tolerant algorithms like Shor and Grover to thousands or millions for NISQ algorithms such as QAOA and random circuit sampling.

The system successfully balances processes with differing shot counts, reducing idle time and improving batch utilisation. Evaluations using stabilizer measurement circuits, classical arithmetic circuits, and multi-qubit controlled gates demonstrate HALO’s effectiveness. The developed cost function accurately explains fidelity degradation observed with increased throughput from experiments, validating the model’s ability to predict performance. This work presents the first quantum OS design combining dynamic space sharing, time sharing, and helper qubit sharing, advancing the state of quantum resource scheduling.

Hardware-aware qubit allocation and dynamic scheduling optimise cloud computing performance

Researchers have developed HALO, a novel operating system design enabling fine-grained resource-sharing for computing platforms. This system addresses the growing imbalance between user demand and limited hardware resources in cloud computing environments, where multiple users often contend for access to a small number of processors.

HALO achieves improved performance through two key mechanisms: a hardware-aware qubit-sharing algorithm and a shot-adaptive scheduler. The hardware-aware algorithm strategically places shared helper qubits to minimise routing overhead and mitigate noise interference between processes from different users.

Simultaneously, the shot-adaptive scheduler dynamically allocates execution windows based on the sampling requirements of each job, thereby enhancing throughput and reducing latency. Evaluations on the Torino computer demonstrate that HALO improves hardware utilisation by up to 2.44times and increases throughput by 4.44times, while maintaining fidelity loss to within 33 percent.

These results confirm the practicality of resource-sharing in modern computing systems. The study acknowledges that aggressive qubit sharing can lead to fidelity degradation, particularly in benchmarks such as quantum error correction and Cn(X). Analysis reveals a correlation between fidelity loss and the ratio of helper qubit routing cost to introduction cost, indicating that the overhead associated with qubit routing contributes to performance limitations.

Future work could focus on optimising qubit placement strategies and scheduling algorithms to further minimise routing costs and improve fidelity. The findings establish a clear path towards more efficient and scalable computing by demonstrating the benefits of fine-grained resource sharing and providing insights into the trade-offs between throughput and fidelity.