2D 20 µm modal-temporal SiN/SiO2 atom-gap optimizer#
Live status
paused by user · phase q over area · evaluation 200 · updated 2026-09-04 17:19:54 UTC
Q is intentionally plotted on a linear scale. Green points have passed both independent pole-fit checks.#
The hard 1 µm air gap crosses the complete 1 µm SiO2 stack. Only the 300 nm SiN core density outside it changes.#
Latest pole-frequency-filtered Ey field, captured at evaluation 200. Its phase is rotated so the atom field is real and positive.#
The measured atom response and fixed ideal response use one common amplitude scale. Only the source-off portion enters the independent pole fit.#
What this test means#
This is a true y-invariant Maxwell solve: a short periodic Yee supercell represents an infinite structure along y. A solved fundamental TE mode is launched inward from the fixed density-0.5 guide near +x; x symmetry supplies the coherent launch from -x, while z symmetry retains the correct even Ey family. The full design is 20 µm long and the atom remains at the center of the hard 1 µm air gap.
The physical 2D confinement metric is effective area, not an assumed 3D volume. Stage A differentiates one fixed causal target: the complete modal pulse response at the atom followed by a Q=10,000, fixed-amplitude ringdown. It does not fit away amplitude or phase. Its exact compact adjoint carries the linear atom trace; Q, energy, mode area, and field telemetry are audited by a separate forward solve every five evaluations and are not interpolated on the plots. Once the atom trace contains a trustworthy Q>=100 pole and its temporal progress has clearly slowed, Stage B maximizes the same-pole harmonic Q/Aeff with replay-validated adaptive trust steps. The source and pole-field monitor then follow the accepted pole.
The field panel is a source-off frequency projection at the tracked pole—not the final instantaneous field divided by numerical noise. The 2 ps starting trace grows only when the fitted lifetime would otherwise show less than roughly 1.5% energy decay. There is no binarization or spatial filter.
Full method: campaign methodology.