IBM Patent Application Targets More Efficient Sharing of Quantum Computers

Introduction

IBM has filed a patent that proposes a software layer for quantum computers capable of running multiple users’ jobs simultaneously while cutting down the effort required for error mitigation. By intelligently grouping compatible circuits and re‑using previously gathered noise models, the system aims to improve hardware utilisation, shorten queue times, and potentially lower costs for users who opt into shared execution.

How the Scheduler Works

The core idea is a quantum queue optimizer that examines the waiting list of jobs and looks for programs that can run together on the same processor. It evaluates several factors:

1. Qubit allocation – Each job may need only a subset of the processor’s qubits. The scheduler checks that the combined jobs fit within the available qubits and that sufficient spacing exists to minimise crosstalk.
2. Noise‑model compatibility – Error‑mitigation techniques often require fresh measurements to characterise how noise affects specific gate patterns. The optimizer stores these measurements as noise models and attempts to match incoming jobs with existing models, avoiding redundant calibration runs.
3. Timing and priority – If a job finishes early, the system can slide another waiting job forward to make use of still‑valid noise data, preventing unnecessary delays.
4. User preferences – Users can specify constraints such as minimum distance from other activity or willingness to accept shared execution in exchange for billing rebates.

The patent describes three levels of sharing:

• Strict reuse – Jobs are combined only when an exact noise model already exists.
• Extended measurement – The system may perform additional calibration steps to create a new model for the combined workload.
• Derived inference – Existing models are extrapolated to approximate the needed noise information.

Machine‑learning heuristics help decide which level to apply based on queue length, processor load, and interference risk.

Practical Benefits and Challenges

Benefits

• Higher throughput – By filling idle qubits with compatible jobs, the processor can perform more calculations per unit time.
• Reduced calibration overhead – Re‑using noise models cuts the number of extra experiments traditionally required for error mitigation.
• Cost incentives – The filing mentions possible billing rebates for users who allow their jobs to be co‑scheduled, encouraging participation in shared execution.

Challenges

• Inter‑job interference – Simultaneous operations can introduce new error pathways, so the scheduler must carefully assess crosstalk and gate interactions.
• Security and data isolation – Results from different users must be cleanly separated to prevent information leakage.
• Dynamic noise – Quantum hardware noise drifts over time; the system must continuously refresh models and discard outdated ones.

Outlook

IBM’s proposal builds on earlier research such as Columbia Engineering’s HyperQ, which partitions a physical chip into virtual machines for concurrent use. While IBM’s approach focuses on coordinating shared execution with the error‑mitigation workflow, both efforts signal a broader industry move toward multi‑tenant quantum computing. If successful, such scheduling techniques could make quantum‑hardware access more affordable and scalable, accelerating the transition from isolated experimental runs to routine, cloud‑based quantum services.

--- The article is based on IBM’s recently published patent application and does not include measured performance data.