Modal Phase Matching in Nonlinear Optics

• Modal Phase Matching (MPM) is a nonlinear optics technique that uses distinct spatial modes to achieve effective index matching and satisfy momentum conservation.
• MPM employs modal dispersion engineering by selecting higher-order modes to compensate for material dispersion, enabling efficient processes such as second-harmonic generation and four-wave mixing.
• Practical MPM design requires balancing exact phase matching with strong nonlinear overlap and optimal tensor access to maximize conversion efficiency in integrated photonic devices.

Searching arXiv for recent and foundational papers on modal phase matching across $\chi^{(2)}$ and $\chi^{(3)}$ platforms. Modal phase matching (MPM) is a phase-matching technique in nonlinear guided-wave optics that uses different spatial modes of a waveguide or resonator so that the propagation constants of interacting waves satisfy the momentum-conservation condition. In the $\chi^{(2)}$ setting emphasized by recent thin-film lithium niobate, Al$_x$Ga$_{1-x}$As, AlN, and III–V microring work, the characteristic design move is to place one field—often the generated wave at the higher frequency—in a higher-order mode whose effective index compensates material-dispersion mismatch (Hansen et al., 2023, Kang et al., 2021, Luo et al., 2018). In $\chi^{(3)}$ multimode four-wave mixing, the same principle appears as the use of modal dispersion to reduce $\Delta \beta$ by assigning pumps, signal, and idler to suitable mode branches (Kernetzky et al., 2022). MPM is distinct from birefringent phase matching and from quasi-phase matching (QPM): birefringent schemes exploit polarization- and direction-dependent material indices, whereas QPM compensates residual mismatch with a longitudinal grating or domain inversion rather than forcing $\Delta \beta = 0$ directly (Hansen et al., 2023, Luo et al., 2018).

1. Definition and formal phase-matching conditions

For guided-wave second-harmonic generation (SHG), MPM is typically expressed through the equality of effective indices between a pump mode at $\omega$ and a distinct guided mode at $2\omega$. In thin-film lithium niobate-on-insulator (LNOI), one formulation is

$\chi^{(3)}$0

with

$\chi^{(3)}$1

Perfect phase matching requires

$\chi^{(3)}$2

which is the MPM condition in effective-index form (Hansen et al., 2023). In more standard SHG notation this is

$\chi^{(3)}$3

The same structure appears in resonant lithium-niobate microrings. A Z-cut LN microring designed for SHG uses a TM$\chi^{(3)}$4 mode near $\chi^{(3)}$5 and a TM$\chi^{(3)}$6 mode near $\chi^{(3)}$7, with phase matching written as

$\chi^{(3)}$8

In a cavity, this is accompanied by resonance matching,

$\chi^{(3)}$9

so efficient conversion requires both spatial/momentum matching and frequency matching (Luo et al., 2018).

For generic $\chi^{(2)}$0 three-wave mixing in waveguides, the mismatch is often written

$\chi^{(2)}$1

with $\chi^{(2)}$2 at phase matching. In SHG this reduces to $\chi^{(2)}$3 (Kang et al., 2021). For multimode $\chi^{(2)}$4 Bragg-scattering four-wave mixing (FWM), the linear mismatch is

$\chi^{(2)}$5

and MPM becomes the use of mode-dependent $\chi^{(2)}$6 to reduce this quantity (Kernetzky et al., 2022).

A recurring implication across these papers is that MPM is best regarded as a modal-dispersion engineering method rather than a purely material-dispersion method. This distinguishes it from birefringent phase matching in bulk crystals and from QPM schemes that rely on a grating vector or periodic sign inversion of the nonlinearity (Hansen et al., 2023, Luo et al., 2018).

2. Mechanism: modal dispersion engineering and mode selection

The central physical mechanism of MPM is the use of a higher-order guided mode to shift the effective index at the generated frequency. In the 2023 LNOI double-ridge waveguide design, the pump is the fundamental quasi-TE mode at $\chi^{(2)}$7, while the second harmonic at $\chi^{(2)}$8 is a higher-order quasi-TE mode; the geometry is chosen so that the two modes have equal effective index, $\chi^{(2)}$9 (Hansen et al., 2023). In the 2018 LN microring, finite-element simulations showed the effective indices of TM$_x$0 at $_x$1 and TM$_x$2 at $_x$3 crossing near a top width of about $_x$4, which provided the design point for modal phase matching (Luo et al., 2018).

The same logic extends beyond straight waveguides. A thin-film lithium niobate squeezed-light source uses MPM in a doubly resonant ring by matching a higher-order quasi-TM$_x$5 pump mode at $_x$6 to a fundamental quasi-TE$_x$7 signal mode at $_x$8. The phase-matching criterion is stated in effective-index form as

$_x$9

equivalently,

$_{1-x}$0

for the degenerate down-conversion process (Arge et al., 2024). In III–V semiconductor microrings for spontaneous parametric down-conversion (SPDC), the pump is taken in TM0 near $_{1-x}$1, while signal and idler are in TE0 near $_{1-x}$2; the paper gives the “usual phase matching” estimate

$_{1-x}$3

but in the ring geometry this acts together with azimuthal mode-number constraints and built-in QPM (Fontaine et al., 2024).

In $_{1-x}$4 multimode FWM, the role of modal dispersion is equally explicit. One- and two-mode operation are favored because group-delay matching can be expressed compactly, whereas more-than-two-mode FWM requires numerical optimization of the full $_{1-x}$5 and becomes much less practical (Kernetzky et al., 2022). This suggests that the most effective forms of MPM are those in which the number of participating modal branches is kept small enough that the dispersive compensation remains controllable.

3. Nonlinear overlap, tensor access, and the main limitation of MPM

The major limitation of conventional MPM is not the phase-matching condition itself but the nonlinear overlap integral. Because the higher-order mode used to compensate dispersion has sign-changing lobes, the overlap of the three interacting fields can be strongly reduced. In a generic $_{1-x}$6 treatment, the overlap factor is

$_{1-x}$7

and even with $_{1-x}$8, poor $_{1-x}$9 limits efficiency (Kang et al., 2021).

A vectorial formulation appears in the LNOI double-ridge design, where the nonlinear coupling coefficient is

$\chi^{(3)}$0

For the optimal phase-matched mode pair, the paper computes $\chi^{(3)}$1, whereas another phase-matched higher-order SH mode yields $\chi^{(3)}$2, directly illustrating the tradeoff between exact phase matching and strong nonlinear overlap (Hansen et al., 2023). In the LN microring, the analogous cavity single-photon coupling depends on an overlap factor $\chi^{(3)}$3, and the measured strong SHG implies substantial TM$\chi^{(3)}$4/TM$\chi^{(3)}$5 overlap under a $\chi^{(3)}$6-mediated interaction (Luo et al., 2018).

Tensor access is correspondingly central. The thin-film X-cut LNOI double-ridge design uses a type-0 TE–TE interaction so that both pump and second harmonic remain in the quasi-TE family and access the largest second-order tensor element $\chi^{(3)}$7, quoted as $\chi^{(3)}$8 with $\chi^{(3)}$9 and $\Delta \beta$0 (Hansen et al., 2023). The Z-cut LN microring likewise uses quasi-TM modes to exploit $\Delta \beta$1 (Luo et al., 2018). By contrast, the TFLN squeezed-light source uses a quasi-TM$\Delta \beta$2-to-quasi-TE$\Delta \beta$3 interaction and therefore the weaker $\Delta \beta$4 tensor element, one reason a doubly resonant cavity is needed to compensate for the weaker interaction (Arge et al., 2024).

This limitation has motivated nonlinearity-engineered variants of MPM. In Al$\Delta \beta$5Ga$\Delta \beta$6As and related systems, one can modify $\Delta \beta$7 itself so that the sign of the material nonlinearity matches the sign pattern of the higher-order mode, converting destructive contributions in the overlap integral into constructive ones (Kang et al., 2021). A plausible implication is that MPM performance is often bounded less by the existence of a phase-matched mode crossing than by the degree to which the chosen mode set can sustain a non-canceling overlap integral while still accessing a large tensor element.

4. Material platforms and structural implementations

Recent MPM literature is concentrated in thin-film lithium niobate, lithium-niobate microrings, III–V semiconductors, AlGaAs, and AlN. The structures differ, but all use geometry to align modal effective indices.

Representative implementations

Platform	Interaction and mode set	Reported structural features
X-cut LNOI double-ridge waveguide	Fundamental quasi-TE pump at $\Delta \beta$8 to higher-order quasi-TE SH at $\Delta \beta$9	$\Delta \beta = 0$0 LN on $\Delta \beta = 0$1 silica; $\Delta \beta = 0$2, $\Delta \beta = 0$3, $\Delta \beta = 0$4 (Hansen et al., 2023)
Z-cut LN microring	TM$\Delta \beta = 0$5($\Delta \beta = 0$6) to TM$\Delta \beta = 0$7($\Delta \beta = 0$8) SHG	$\Delta \beta = 0$9 LN on $\omega$0 BOX; $\omega$1; top width about $\omega$2 (Luo et al., 2018)
Z-cut TFLN ring for squeezing	TM$\omega$3($\omega$4) pump to TE$\omega$5($\omega$6) signal/idler	$\omega$7 Z-cut LN on $\omega$8 SiO$\omega$9; $2\omega$0 radius; phase matching near top width $2\omega$1 (Arge et al., 2024)
Al$2\omega$2Ga$2\omega$3As ridge waveguide	Type-II MPM: TE$2\omega$4, TM$2\omega$5 at $2\omega$6, TE$2\omega$7 at $2\omega$8	Deeply etched ridge, width $2\omega$9; propagation along 110
AlN-on-insulator ridge	TM$\chi^{(3)}$00($\chi^{(3)}$01) to TM$\chi^{(3)}$02($\chi^{(3)}$03)	AlN laterally sandwiched by Ta$\chi^{(3)}$04O$\chi^{(3)}$05 to suppress destructive overlap (Kang et al., 2021)

Two additional structural variants modify standard MPM in a targeted way. First, a reverse-polarization dual-layer crystalline thin-film nanophotonic waveguide in lithium niobate on insulator experimentally observed SHG at $\chi^{(3)}$06 under excitation at $\chi^{(3)}$07, with an ultrahigh conversion efficiency of $\chi^{(3)}$08; the scheme uses two bonded layers with internally reversed polarizations to realize highly efficient SHG via MPM (Wang et al., 2021). Second, “layer-poled modal phase matching” in TFLN selectively poles the bottom part of the waveguide all along its length so that second harmonic is efficiently generated on a higher-order waveguide mode; this is presented as 5 to 10 times more robust towards fabrication uncertainties and theoretically more efficient than conventional QPM (Hefti et al., 6 May 2025).

In III–V semiconductor microrings, MPM does not appear in isolation but as part of a hybrid scheme combining modal dispersion, cavity resonance, and built-in QPM arising from azimuthal rotation of the zincblende $\chi^{(3)}$09 tensor. Efficient structures use TM0 near $\chi^{(3)}$10 and TE0 near $\chi^{(3)}$11, with optimization over ring radius and cross-sectional dimensions (Fontaine et al., 2024).

5. Efficiency, bandwidth, and experimental demonstrations

MPM has been used for efficient SHG, optical parametric generation, SPDC, and squeezed-light generation. In the 2023 thin-film LNOI double-ridge design, numerical solution of the coupled equations gives a conversion efficiency

$\chi^{(3)}$12

with $\chi^{(3)}$13 for a $\chi^{(3)}$14 waveguide at $\chi^{(3)}$15 on-chip pump power; the same design reaches a peak conversion efficiency of $\chi^{(3)}$16 at $\chi^{(3)}$17 and $\chi^{(3)}$18, and the simulated SH spectral response for the $\chi^{(3)}$19, $\chi^{(3)}$20 case has $\chi^{(3)}$21 (Hansen et al., 2023). The 2018 LN microring observed SHG at the phase-matched resonance pair $\chi^{(3)}$22 and $\chi^{(3)}$23, with measured conversion efficiency

$\chi^{(3)}$24

and also showed difference-frequency generation in the telecom band (Luo et al., 2018).

In a reverse-polarization dual-layer crystalline thin-film LN waveguide, excitation at $\chi^{(3)}$25 produced SHG at $\chi^{(3)}$26 with experimental conversion efficiency $\chi^{(3)}$27 (Wang et al., 2021). In the TFLN squeezed-light source, MPM enabled degenerate down-conversion in a compact passive microring and yielded a measured shot noise reduction of $\chi^{(3)}$28, with measured squeezing of $\chi^{(3)}$29 and anti-squeezing of $\chi^{(3)}$30 at $\chi^{(3)}$31 sideband and $\chi^{(3)}$32 on-chip pump power (Arge et al., 2024). The paper estimates generated on-chip squeezing of about $\chi^{(3)}$33, while identifying coupling loss and photorefraction as dominant limits on the collected value (Arge et al., 2024).

Nonlinearity-engineered MPM can amplify these efficiencies substantially. In Al$\chi^{(3)}$34Ga$\chi^{(3)}$35As, modifying $\chi^{(3)}$36 laterally to match the higher-order mode sign structure increases the nonlinear overlap factor by $\chi^{(3)}$37 in one SHG design, corresponding to a lossless efficiency enhancement of $\chi^{(3)}$38. With realistic losses, the same design improves normalized SHG efficiency from $\chi^{(3)}$39 to $\chi^{(3)}$40, an enhancement of $\chi^{(3)}$41 (Kang et al., 2021). In a DFG example, the normalized efficiency improves from $\chi^{(3)}$42 to $\chi^{(3)}$43, an enhancement of $\chi^{(3)}$44 (Kang et al., 2021).

For $\chi^{(3)}$45 multimode FWM, practical performance is more restrictive. Few-mode fibers and nano-rib waveguides both show that one- and two-mode FWM give the best bandwidths and idler powers, while four-mode FWM is not feasible and three-mode FWM in nano-ribs has much reduced bandwidth and reduced idler power (Kernetzky et al., 2022). This reinforces the view that MPM is most powerful when modal participation is limited enough that phase matching and overlap remain jointly favorable.

6. Robustness, fabrication tolerance, and relation to QPM

A common misconception is that MPM is automatically simpler or more tolerant than QPM because it avoids periodic poling. The literature does not support such a blanket statement. Instead, it shows a tradeoff: MPM removes the need for longitudinal domain inversion but can be highly sensitive to geometry because the phase-matching condition depends on precise equality of effective indices (Hansen et al., 2023, Luo et al., 2018).

The LNOI double-ridge study gives one of the clearest tolerance analyses. For a $\chi^{(3)}$46 device, the conversion efficiency is much more sensitive to upper-ridge height $\chi^{(3)}$47 than to horizontal misalignment $\chi^{(3)}$48. Reported full-width at half-maximum tolerances are $\chi^{(3)}$49 for upper-ridge height and $\chi^{(3)}$50 for misalignment, and the authors conclude that upper-ridge height is the dominant critical parameter (Hansen et al., 2023). They further show that by implementing multiple waveguides on a chip with $\chi^{(3)}$51, one can obtain a chip design tolerant over $\chi^{(3)}$52 from the ideal etching depth (Hansen et al., 2023). In the Z-cut LN microring, a $\chi^{(3)}$53 change in waveguide width shifts the phase-matched pump wavelength by approximately $\chi^{(3)}$54, far larger than the $\chi^{(3)}$55 and $\chi^{(3)}$56 cavity linewidths; this explains much of the gap between the theoretical and measured SHG efficiencies (Luo et al., 2018).

QPM remains more flexible in some respects. The 2023 LNOI paper explicitly positions MPM as a “scalable alternative” rather than a universally superior replacement for periodically poled lithium niobate (PPLN), noting that QPM can compensate residual mismatch through an engineered grating vector (Hansen et al., 2023). The TFLN squeezed-light source likewise motivates MPM by the fabrication difficulty of periodic poling in TFLN, but the design uses the weaker $\chi^{(3)}$57 tensor element and therefore requires stronger pumping, which aggravates photorefraction (Arge et al., 2024). In III–V microrings, by contrast, modal dispersion is combined with built-in QPM from the tensor rotation around the ring, showing that MPM and QPM are not always mutually exclusive categories (Fontaine et al., 2024).

Layer-poled MPM is a recent attempt to combine some advantages of both. The 2025 TFLN work proposes selectively poling the bottom part of the waveguide all along its length so that SH is generated efficiently on a higher-order mode and reports that this is 5 to 10 times more robust towards fabrication uncertainties and theoretically more efficient than conventional QPM (Hefti et al., 6 May 2025). This suggests a broader trend: rather than treating MPM and QPM as strict alternatives, current work increasingly hybridizes modal-dispersion engineering with tailored nonlinear-coefficient distributions.

7. Extensions, related concepts, and current research directions

MPM now spans classical frequency conversion, quantum light generation, and hybrid resonant architectures. In classical SHG, it has been demonstrated in thin-film lithium niobate waveguides and LN microrings (Hansen et al., 2023, Luo et al., 2018, Wang et al., 2021). In quantum photonics, the same phase-matching relations underlie SPDC and optical parametric amplification. The TFLN double-ridge SHG design explicitly states that it “has the potential for generating pairs of entangled photons in the infrared C-band by SPDC” (Hansen et al., 2023), and the TFLN microring squeezing experiment demonstrates that MPM can support a compact, passive, integrated squeezed-light source without periodic poling (Arge et al., 2024).

A further extension appears in III–V semiconductor microrings, where modal dispersion is used together with built-in QPM and cavity resonance. The key azimuthal selection rule is

$\chi^{(3)}$58

equivalently,

$\chi^{(3)}$59

so MPM-type effective-index engineering operates within a ring-specific momentum-conservation structure (Fontaine et al., 2024). This suggests that in resonators, the natural generalization of MPM requires simultaneous matching of effective indices, resonance frequencies, and azimuthal mode numbers.

It is also important to distinguish true MPM from multimode-beating-induced QPM. In high-order-harmonic generation, two excited waveguide modes can create longitudinal modulation of intensity, phase, and polarization, yielding QPM when one Fourier component of the modulation compensates the HHG coherence-length mismatch. That scheme uses modal propagation constants, but it is not true MPM because it does not enforce continuous $\chi^{(3)}$60; rather, it is quasi-phase matching by modal beating (Liu et al., 2013). The distinction is conceptually significant because it separates exact modal propagation-constant matching from grating-based compensation generated by multimode interference.

Across platforms, the main design lesson is consistent. MPM is most successful when three conditions are met simultaneously: the interacting modes satisfy an exact or near-exact effective-index matching condition; the chosen polarization family accesses a favorable tensor element; and the sign-sensitive nonlinear overlap remains sufficiently large despite the use of a higher-order mode (Hansen et al., 2023, Luo et al., 2018, Kang et al., 2021). Current work on reverse-polarization dual-layer LN, layer-poled MPM, and transverse nonlinearity engineering indicates that the field is moving from pure modal-dispersion matching toward more elaborate control of both mode structure and nonlinear-coefficient distribution (Wang et al., 2021, Hefti et al., 6 May 2025, Kang et al., 2021).