A design test for photon-pair sources without periodic poling

Two conditions that must work together

Spontaneous parametric down-conversion, or SPDC, converts a pump photon into two lower-energy photons, conventionally called signal and idler. Detecting one can announce the presence of the other. But if their frequencies are correlated, that announcement alone leaves the remaining photon in a mixture of spectral states. Such mixtures can reduce interference between photons from independent sources.

Spectral filters can improve purity by discarding parts of the spectrum, at the cost of losing photons. The alternative explored here is to engineer the spectrum during generation.

That requires two different kinds of coordination. Phase matching allows generation amplitudes from different positions along a device to reinforce one another. Group velocities—the speeds at which optical wave packets travel—help determine the frequency correlations of the resulting pair.

Periodic poling supplies an additional way to achieve phase matching by repeatedly reversing a material’s nonlinear response. Without that pattern, the optical modes must provide both the phase and group-velocity conditions themselves.

A neighboring guide is not an independent control knob

Qiang Sun, Ke Wang and Andrew D. Greentree study a uniform pair of neighboring waveguides. The main guide supplies the nonlinear response; the auxiliary guide changes how light propagates. In their simplified model, the pump and signal remain mainly in the first guide, while the idler occupies a stationary optical mode spread across both.

This shared mode changes the idler’s propagation constant and group velocity. However, those changes cannot be selected independently. The same mixing also determines how much of the idler mode remains in the nonlinear guide.

The paper’s central contribution is an explicit criterion tying these quantities together. A preliminary screen asks whether the available group-velocity contrast could reach a desired delay relationship. A second condition asks whether phase matching selects exactly the mode mixture needed for that relationship.

Passing the first screen is therefore not enough. Nor can a guide that already phase-matches by itself generally retain the same operating point after coupling is introduced. The main guide may need adjustment too.

There is also a generation trade-off: transferring more modal weight into a linear auxiliary can reduce the nonlinear interaction. The criterion exposes that cost rather than treating dispersion control as free.

A simulated material example

The authors test feasibility with full-vector electromagnetic calculations of an aluminum-nitride main guide beside a titanium-dioxide auxiliary, embedded in silica. The separation and cross-section remain constant over a 5-millimeter interaction length: there is no periodic poling or longitudinal grating.

The example uses a 775-nanometer pump, a 1,300-nanometer signal and an approximately 1,919-nanometer idler. Its calculated carrier phase mismatch is small enough to give essentially the maximum phase-matching factor over that length.

The pump–idler and pump–signal group-delay mismatches have opposite signs, with a ratio of −0.899, close to the symmetric target of −1. A direct nonlinear-overlap calculation also confirms that the selected mode retains a generation pathway in aluminum nitride.

After optimizing the Gaussian pump bandwidth, the calculated unfiltered Schmidt purity is 0.789. Purity measures spectral single-mode character; it is not the fraction of pump photons converted or the probability of detecting a pair.

Why this is not a near-perfect photon source

The purity calculation retains sampled modal dispersion and an approximate frequency-dependent nonlinear overlap. It covers a finite frequency window, extending 12 terahertz on either side of each daughter’s central frequency. Weak spectral sidelobes make the reported value dependent on that window.

The work is numerical, not a fabricated-source demonstration. Titanium dioxide uses a provisional dispersion model, and the design is sensitive to dimensions. Tested gap changes of 30 nanometers and width changes of 20 nanometers sharply reduce phase matching at the original wavelengths. Smaller errors can shift the generated wavelengths while also changing the delay relationship.

The reduced two-mode criterion is consequently a search guide, not a replacement for full-vector refinement. Loss, selective pump-mode excitation, access coupling and actual photon-pair generation still require device-level characterization.

The paper also discusses idealized spectral shaping that could yield unit purity, but reports no physical device realizing it. Its practical contribution is narrower: a way to reject incompatible designs early and identify targets worth detailed calculation.