Microwave waveguide for a quantum sensor

A fabrication-ready microwave waveguide for spin sensors in a 2D material, designed in simulation.

2.7× vs first sweep · sim.

The junction of the final microwave design (four 100 µm copper traces meeting at the hBN flake), and the running best simulated circularly polarized field across 313 designs, ending 2.7 times the best design of the agent's first size sweep. — Simulation · junction from the preprint's Fig. 4b; plot from the campaign's records (Fig. 4c)

≈119 MHzsimulated spin Rabi frequency at 16 W

2.7×vs best of the agent's first size sweep

418simulations over six days

The problem

Near zero field, boron-vacancy spins in hBN have two overlapping resonances, so linearly polarized drive gives no first-order signal. Circular polarization restores it but needs two perpendicular in-plane fields 90° out of phase.

What we did

Researchers froze the score (simulated circular field at the sample per √W) and the lab's build limits, such as one 16 W amplifier. The collaborative agent proposed, ran and scored 418 simulations in six days; researchers twice corrected its physics.

Result

A symmetric 45 mm copper cross. In simulation: ≈119 MHz Rabi frequency at 16 W (4.2 ns π pulse), 2.7× the best of the agent's first size sweep. It reflects 81% of the power (counted in that figure); none of 4,896 matching networks reached spec.

Keep in mind

All results are simulations; fabrication and testing of the device are under way. The 2.7× compares with the agent's own first board-size sweep, not a published device.

In the preprint

Sec. IV.C, “Microwave device design,” and Fig. 4 of arXiv:2610.17193 describe the campaign and its checks in more detail.

Fig. 4 of the preprint, full size — Fig. 4 of the preprint: (a) the final board design, a 45 mm copper cross; (b) the junction region and the layer stack; (c) the simulated circularly polarized field at the flake per √W for the 313 distinct designs (418 simulations), with the running best and the best design of the agent's initial board-size sweep, 2.7× below the final design (all simulation). Click to enlarge.

Data: Simulation (Ansys HFSS finite-element solver)
Models (as stated in the preprint): Claude Opus 5
PhysKit version (as stated in the preprint): 0.6.0, an earlier version
Where: Sec. IV.C, Fig. 4

We list the versions the preprint reports so readers can trace each result to its tools, not as a recommendation: for new work, use the latest release.