Machine learning models bring fast quantum gate control to edge hardware

Researchers have shown that compact machine learning models can compute control settings for quantum gates in microwave cavity qudits fast enough to run on edge hardware, achieving gate trace infidelity near 10^-3 while using programmable logic resources efficiently. The work, published in Quantum Machine Intelligence, addresses a long-standing bottleneck in quantum optimal control: the algorithms that find the best control pulses are powerful, but CPU- or GPU-based implementations often carry delays too long to be applied in practice on live quantum hardware.

The team, led by Flor Sanders and Gaurav Agarwal with colleagues including Luca Carloni, Giuseppe Di Guglielmo, Andy C. Y. Li and Gabriel N. Perdue, focused on SNAP gates — Selective Number-dependent Arbitrary Phase gates — used to prepare states in microwave cavity qudits. Qudits are multi-level quantum systems that serve as elementary computation units for quantum computing, and controlling them accurately requires precisely shaped control pulses whose parameters must be determined quickly and with high fidelity, where fidelity is defined as the overlap with a target quantum state.

According to the abstract, the methodology involved four stages: generating training data using classical optimization techniques, developing machine learning models on that data, exploring the design space of possible models, and quantizing the models for hardware implementation. The quantization step is what makes deployment on edge hardware feasible, since it converts the learned models into forms that can run on programmable logic rather than on general-purpose processors with their associated latency.

The reported results demonstrate the efficacy of the approach, with optimized models achieving low gate trace infidelity near 10^-3 and efficient utilization of programmable logic resources. The study is published as a review article in Volume 8 of the journal, article number 116, dated 21 October 2026. Because the source material is limited to the abstract, the full experimental details, the specific hardware platform used for deployment, and any comparisons with alternative control approaches are not available in the supplied evidence, and readers should consult the complete article for those specifics. What the abstract does establish is a proof of the pipeline: classical optimization to create labeled data, machine learning to learn the mapping from gate targets to pulse parameters, and hardware-oriented quantization to bring the resulting inference latency down to a regime where real-time quantum control becomes practical.

Subject of Research: Machine learning for fast quantum optimal control of microwave qudits on edge hardware

Article Title: Fast machine learning for quantum control of microwave qudits on edge hardware

Article References: Sanders, F., Agarwal, G., Carloni, L., Di Guglielmo, G., Li, A. C. Y., & Perdue, G. N. (2026). Fast machine learning for quantum control of microwave qudits on edge hardware. Quantum Machine Intelligence, 8(2), Article 116. https://doi.org/10.1007/s42484-026-00455-9

Image Credits: AI Generated

DOI: 10.1007/s42484-026-00455-9

Keywords: quantum computing, quantum optimal control, machine learning, qudits, SNAP gates, edge hardware, microwave cavities, quantum gates, programmable logic, quantization, Fast, machine