Dispersion-Optimized Frequency Comb

• Dispersion-optimized frequency combs are broadband sources whose resonator parameters are deliberately engineered to control bandwidth, coherence, and spectral flatness.
• Techniques such as geometric tailoring, avoided mode crossings, and synthetic dispersion effectively balance nonlinear phase shifts with dispersive spreading.
• These approaches enable applications across Kerr, χ(2), and semiconductor platforms by achieving wide comb spans, precise mode spacing, and enhanced pulse synchronization.

A dispersion-optimized frequency comb is a comb source in which the resonator morphology, modal coupling, higher-order dispersion, group delay dispersion, or effective phase-matching landscape is deliberately engineered so that the comb operates in a targeted regime of bandwidth, coherence, spectral flatness, tooth power, or synchronization. In Kerr microcombs, this usually means controlling the integrated dispersion $D_{\mathrm{int}}$, the sign and magnitude of $D_2$, and dispersive-wave phase matching; in semiconductor and $\chi^{(2)}$ systems it can also mean compensating group delay dispersion, using cavity phase matching, or creating a synthetic dispersion landscape through additional pumps or active intracavity control (Fujii et al., 2020, Moille et al., 2022, Moille et al., 2021, Villares et al., 2015, Lv et al., 2018).

1. Dispersion metrics and the bandwidth problem

The standard microresonator description expands the resonance frequencies as

$$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$

where $D_1/2\pi$ is the free spectral range (FSR), $D_2/2\pi$ characterizes the variation of the FSR with mode number, and higher-order terms represent higher-order dispersion. A closely related quantity is the integrated dispersion,

$$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$

which measures the deviation of the actual resonance grid from an equidistant comb grid (Fujii et al., 2020, Moille et al., 2022).

These quantities determine whether nonlinear phase shifts from the Kerr effect can be balanced by dispersive spreading. In one convention, anomalous GVD corresponds to positive $D_2$; in another, it corresponds to $\beta_2<0$, with

$D_2=-\frac{cD_1^2}{n}\beta_2.$

For dissipative Kerr solitons, the usual design target is $D_2$0 near the pump together with higher-order dispersion that makes $D_2$1 at remote frequencies, enabling strong dispersive waves and broadband combs (Fujii et al., 2020, Yang et al., 13 Nov 2025, Moille et al., 2022).

A persistent design tension arises because comb span and resonator size are coupled. In wedge microcavities, the comb bandwidth follows the scaling law

$D_2$2

so small resonators are attractive because they provide large FSRs and favor broad line spacing, but tight curvature usually enhances normal geometric dispersion. That tradeoff recurs across platforms: the most useful combs generally require simultaneous control of FSR, second-order dispersion, and higher-order phase matching rather than geometry alone (Fujii et al., 2020).

2. Geometric dispersion tailoring in whispering-gallery and wedge resonators

A major branch of dispersion optimization uses geometry itself as the control knob. In silica wedge microdisks of about $D_2$3 outer radius and $D_2$4 thickness, changing the sidewall angle tunes the cavity between normal and anomalous GVD without significantly changing the FSR. A shallow $D_2$5 wedge gave measured $D_2$6 values of $D_2$7 MHz and $D_2$8 MHz, whereas a steep $D_2$9 wedge gave $\chi^{(2)}$0 MHz and $\chi^{(2)}$1 MHz; the anomalous $\chi^{(2)}$2 device generated a 300 nm bandwidth Kerr optical frequency comb with FSR $\chi^{(2)}$3 (Fujii et al., 2020).

Crystalline whispering-gallery platforms extend the same principle to a broader range of cross sections. Systematic ultraprecision machining of $\chi^{(2)}$4 resonators with spheroidal, triangular, rectangular, and trapezoidal geometries showed that resonator diameter, waveguide height $\chi^{(2)}$5, apex angle $\chi^{(2)}$6, and trapezoid angle $\chi^{(2)}$7 can tune the GVD from normal to near-zero, weakly anomalous, or strongly anomalous. In that platform, reduced spatial mode interactions enabled mode-interaction-free soliton generation with a 3-dB bandwidth of 2.61 THz, pulse width 121 fs, and repetition rate 25.88 GHz, and the same design framework was extended to telecom O- and E-bands, the mid-IR, and 1 $\chi^{(2)}$8m-pumped optical parametric oscillation with predicted signal/idler output spanning roughly 700 nm to 2000 nm (Yang et al., 13 Nov 2025).

In fluorite WGM resonators, dispersion optimization can be dominated by mode-family selection rather than only by external shape. Analytical and experimental work showed that the major radius $\chi^{(2)}$9 and especially the radial mode index $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$0 strongly reshape total GVD. For $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$1 with $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$2 mm and $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$3 mm, the $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$4 family had GVD about $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$5 ps/km/nm at $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$6, while $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$7 dropped to about $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$8 ps/km/nm, and different radial families produced experimentally distinct primary comb spacings of $$\omega_\mu=\omega_0+\mu D_1+\frac{1}{2}\mu^2 D_2+\frac{1}{6}\mu^3 D_3+\ldots,$$9 GHz, $D_1/2\pi$0 GHz, and $D_1/2\pi$1 GHz in the same resonator. The same study identified $D_1/2\pi$2, with zero-dispersion wavelength at $D_1/2\pi$3, as the most promising crystal for pushing Kerr comb generation into the mid-IR (Lin et al., 2015).

Micro-structured crystalline resonators combine high $D_1/2\pi$4 with waveguide-like dispersion control. In a z-cut $D_1/2\pi$5 resonator with an axially symmetric micro-structured waveguide, flattening the dispersion increased the comb span to over 700 nm with 60 mW pump power at 1560 nm, corresponding to nearly 2000 lines separated by 46 GHz. Earlier morphology engineering in CaF$D_1/2\pi$6 had already shown that a truncated spheroidal WGM resonator could generate a comb centered at 794 nm, even though the material dispersion of CaF$D_1/2\pi$7 is normal there, by selecting a vertical mode family with geometry-induced anomalous dispersion (Grudinin et al., 2014, Savchenkov et al., 2010).

3. Avoided crossings, hybridized supermodes, and localized anomalous dispersion

Avoided mode crossings are often treated as spectral nuisances, but the literature also treats them as an intentional dispersion-engineering resource. In few-moded normal-dispersion $D_1/2\pi$8 microresonators with waveguide cross section $D_1/2\pi$9, mode coupling between nearby transverse families produced avoided crossings that strongly modified the local FSR and could even change the sign of the local dispersion. Frequency-comb-assisted spectroscopy showed mode interactions near 1532 nm, 1542 nm, and 1562 nm, and the initial comb sideband was observed to remain pinned near a mode-crossing frequency even when the pump was shifted substantially. Under these conditions the resonators generated coherent Type-I combs and bandwidth-limited pulses at repetition rates down to 75 GHz without passing through a chaotic intermediate state (Liu et al., 2014).

A more deliberate version of this strategy was demonstrated in concentric racetrack resonators fabricated in 300 nm-thick $D_2/2\pi$0, a thickness that ordinarily has high normal dispersion around $D_2/2\pi$1 ps/nm/km. By phase matching inner and outer racetrack modes, the device created symmetric and anti-symmetric supermodes; the anti-symmetric branch acquired anomalous dispersion even though the uncoupled modes were normally dispersive. The resonator retained intrinsic $D_2/2\pi$2 in the $D_2/2\pi$3–$D_2/2\pi$4 million range, and mode-selective coupling allowed coherent comb generation with evidence suggestive of soliton-like pulse formation and a Cherenkov radiation peak near 1496 nm (Kim et al., 2016).

In the blue and ultraviolet, where III-nitride materials exhibit strong normal material dispersion, avoided-crossing engineering becomes a route to regimes that are otherwise inaccessible. A vertically coupled $D_2/2\pi$5/AlN hybrid waveguide was designed so that a supermode near $D_2/2\pi$6 acquired a strong anomalous-dispersion dip through the coupling term in

$D_2/2\pi$7

With representative geometry $D_2/2\pi$8 nm, $D_2/2\pi$9 nm, $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$0 nm, $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$1 nm, and $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$2 nm, split-step Fourier simulations of the Lugiato–Lefever equation predicted a bright Kerr soliton and a comb bandwidth exceeding $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$3 at the $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$4 dB level, covering the 436 nm and 467 nm $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$5 transitions (Dorche et al., 2020).

Hybridization can also be used to insert localized anomalous pockets into an otherwise normal-dispersion background. Concentric dual-ring microresonators in chalcogenide glass exploit symmetric and antisymmetric supermodes formed by inner and outer concentric rings. In a representative design with radius 30 $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$6m, height 650 nm, inner ring width 1200 nm, outer ring width about 1050 nm, and gap around 825–850 nm, hybridization near the optical-path-length crossing created multiple zero-integrated-dispersion points and multiple dispersive waves. Numerical LLE simulations yielded an octave-spanning dissipative Kerr soliton comb spanning 1227.25 nm to 2912.87 nm with $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$7 dB bandwidth 1265.84 nm (99.82 THz) and comb spacing about 620 GHz; the same work reported that the maximum phase mismatch was reduced from about 70 GHz in a single ring to about 20 GHz in the dual ring (Wang et al., 2022).

Visible extension in thin SiN combines both cross-section engineering and coupled-resonator dispersion engineering. An over-etched 417 nm SiN waveguide with $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$8 nm and over-etch depth $$D_\mathrm{int}=\sum_{k>1}\frac{D_k}{k!}\mu^k=\omega_\mathrm{res}-(\omega_0+D_1\mu),$$9 nm shifted the anomalous-dispersion window toward $D_2$0 nm, and subsequent coupled-racetrack designs tuned the integrated dispersion to strengthen soliton-induced Cherenkov radiation. In simulation, a single-ring device produced a Cherenkov peak at 741 nm and a comb spanning 715 nm to 1070 nm at the $D_2$1 dB level, while MZI-enhanced coupled racetracks increased the $D_2$2 dB bandwidth to 0.77 octave and the $D_2$3 dB bandwidth to an octave spanning 643 nm to 1279 nm (Dorche et al., 2017).

4. Mode-by-mode synthesis and synthetic dispersion

A more granular approach abandons global geometry as the only design handle and instead reshapes individual resonance frequencies. In photonic-crystal microrings, the spatial modulation

$D_2$4

is obtained from an inverse discrete Fourier transform of the desired modal coupling envelope. The resulting CW/CCW mode splittings can flatten the lower branch of $D_2$5 over many modes, allowing broadband frequency combs with stronger dispersive waves. Experimentally, the method reproduced designed modal envelopes over more than twenty modes in TM polarization, with total splitting up to about 100 GHz, coupling $D_2$6 GHz, and intrinsic $D_2$7 around $D_2$8. The same work showed that TE polarization requires a more advanced model because the direct perturbative treatment fails at the modulated interface (Moille et al., 2022).

Dual-pump Kerr combs generalize dispersion optimization into an effective, pump-defined phase-matching landscape. In a silicon nitride microring with thickness 775 nm and radius 23 $D_2$9m, a primary pump at 1063 nm and an overview pump at 1557 nm were used to generate a single low-noise comb spanning 137 THz to 407 THz, about 1.6 octaves. The central concept is synthetic dispersion: the second pump drives four-wave-mixing Bragg scattering so that the translated comb behaves as if it were governed by an effective dispersion landscape $\beta_2<0$0, a shifted version of the physical $\beta_2<0$1. New dispersive waves then appear at frequencies that are hard to reach by geometry alone, while beat-note measurements at 970 nm, 1270 nm, 1420 nm, and 1520 nm verified phase coherence across the broadened spectrum (Moille et al., 2021).

The same logic has been repurposed from bandwidth enhancement to synchronization. In a $\beta_2<0$2 microring with $\beta_2<0$3, $\beta_2<0$4, and $\beta_2<0$5, multi-pumping created a secondary color bound to the primary dissipative Kerr soliton by cross-phase modulation. This secondary wavepacket generated a synthetic dispersive wave at a tunable synchronization wavelength, allowing efficient Kerr-induced synchronization at a mode far from the main pump. The reported color-KIS bandwidth was around 1.53 GHz versus about 130 MHz for the direct-KIS comparison, with available synchronization power improved by about 25 dB to 32 dB and more than 10 GHz of synchronization-window tuning demonstrated (Moille et al., 2024).

These results broaden the meaning of dispersion optimization. It no longer refers only to flattening or sign reversal of $\beta_2<0$6; it also includes mode-by-mode resonance placement, nonlinear spectral translation, and deliberate positioning of strong synchronization teeth.

5. Dynamic dispersion control in normal-dispersion and semiconductor combs

Not all dispersion-optimized combs are built from static geometry. In a pulse-pumped normal-dispersion fiber mini-resonator made from Corning MetroCor dispersion-shifted fiber, pulsed driving lifted the usual requirement for avoided mode crossings. The cavity had FSR $\beta_2<0$7 GHz, quality factor $\beta_2<0$8, $\beta_2<0$9, $D_2=-\frac{cD_1^2}{n}\beta_2.$0, and $D_2=-\frac{cD_1^2}{n}\beta_2.$1. Switching waves formed on the leading and trailing edges of the intracavity field, and the pump-cavity desynchronization

$D_2=-\frac{cD_1^2}{n}\beta_2.$2

became a direct control knob for the dispersive-wave edges. The platform produced a spectrally flat comb of about 26 nm, corresponding to 3.2 THz, with center-frequency tuning by more than 2 THz, selectable line spacing from 0.54 to 10.8 GHz, and, at $D_2=-\frac{cD_1^2}{n}\beta_2.$3 fs, a Raman-assisted anomalous-dispersion cavity soliton with 6 dB bandwidth $D_2=-\frac{cD_1^2}{n}\beta_2.$4 nm, $D_2=-\frac{cD_1^2}{n}\beta_2.$5 THz span, and about 26,000 comb lines (Xu et al., 2020).

Semiconductor FM combs expose a different problem: the existence of dispersion is not enough if it varies strongly over the lasing range. In a 4 mm Fabry–Pérot QCL, the total dispersion

$D_2=-\frac{cD_1^2}{n}\beta_2.$6

was shown to generate reduced bandwidth, nonuniform spectral amplitudes, spectral holes, and piecewise-linear instantaneous frequency. Electrical injection locking by RF modulation of the laser bias at the roundtrip beatnote removed the spectral hole, broadened and flattened the spectrum, and increased the comb bandwidth by about 100%, while restoring a nearly linear chirp (Opačak et al., 2023).

For QCL combs more generally, integrated Gires–Tournois interferometer mirrors supply a frequency-dependent reflected phase that compensates device GDD. In QCL combs centered at 1330 cm$D_2=-\frac{cD_1^2}{n}\beta_2.$7 (7.52 $D_2=-\frac{cD_1^2}{n}\beta_2.$8m), negative-GDD GTI designs introduced minima of about $D_2=-\frac{cD_1^2}{n}\beta_2.$9 at 1258 cm$D_2$00 and about $D_2$01 around 1300 cm$D_2$02. The dispersion-compensated devices operated over a much larger current range, showed no signature of the high-phase-noise regime, produced narrow RF beatnotes, and delivered continuous-wave output power typically $D_2$03 mW with optical bandwidths up to about 70 cm$D_2$04 (Villares et al., 2015).

Normal-dispersion comb formation can also be stabilized by coupling Kerr dynamics to another nonlinear process. In a fiber Fabry–Perot resonator of length about 8.75 cm built from highly nonlinear fiber, coherent interplay of Kerr nonlinearity and stimulated Brillouin scattering produced a stable broadband comb covering more than 10 THz. The cavity had FSR $D_2$05 GHz, resonance linewidth $D_2$06 MHz, finesse $D_2$07, and quality factor $D_2$08 million, yet the selected repetition rate was 10.58 GHz, or $D_2$09 the cavity FSR, because the Kerr-SBS parametric gain selected the 9th cavity mode family spacing rather than the nearest linear SBS overlap (Bunel et al., 5 Feb 2025).

Taken together, these results show that the common assertion that coherent combs require globally anomalous dispersion is too narrow. Bright dissipative Kerr solitons do require anomalous dispersion, but coherent comb states also appear in pulse-driven switching-wave systems, Brillouin-assisted normal-dispersion cavities, and semiconductor combs whose higher-order dispersion is dynamically corrected (Xu et al., 2020, Bunel et al., 5 Feb 2025, Opačak et al., 2023).

6. Dispersion optimization beyond Kerr microcombs

In $D_2$10 micro-OPOs, dispersion optimization often means relaxing the Kerr-style requirement of ultra-flat anomalous GVD rather than reproducing it. A sheet micro optical parametric oscillator based on monolithic y-cut MgO-doped lithium niobate exploited cavity phase matching, in which the $D_2$11 phase shift on mirror reflection compensates nonlinear phase mismatch in a sub-coherence-length Fabry–Perot cavity. A 485.25 $D_2$12m cavity generated a 21.2 THz comb with 133.0 GHz line spacing despite normal dispersion of 275.4 fs$D_2$13/mm around 1064 nm, and the device reached 22.6% slope efficiency and 14.9 kW peak power handling. The shorter 140 $D_2$14m cavity had a calculated CPM bandwidth of 351 nm and produced a comb span $D_2$15 nm, corresponding to 14.3 THz (Lv et al., 2018).

A later cw-driven bulk $D_2$16 OPO introduced a different combination of active and passive control. In a travelling-wave ring cavity with an internal EOM and intracavity dispersion compensation, the resonant signal acquires a quadratic spectral phase

$D_2$17

while the EOM couples adjacent longitudinal modes synchronized to the cavity FSR. For degenerate operation, the scheme produced coherent broadband phase-locked spectra extending over 9 nm at 1064 nm in the normal-dispersion regime and 119 nm at 3100 nm in the anomalous-dispersion regime. In the single-pulse regime, the simulated bandwidths were 3.7 THz and 2.4 THz, corresponding in the time domain to femtosecond quadratic solitons of about 157–160 fs and 212–220 fs (Sanchez et al., 2024).

These $D_2$18 results mark a boundary of the concept. A dispersion-optimized frequency comb need not rely on the same balance as a Kerr soliton microcomb; it may instead use cavity phase matching, intracavity dispersion compensation, or active mode coupling to make a broadband, equidistant, and phase-locked spectrum compatible with a material or wavelength range that would otherwise appear unfavorable (Lv et al., 2018, Sanchez et al., 2024).

Across platforms, the unifying theme is precise control of how the cavity resonance grid departs from equal spacing. That control may come from sidewall angle, resonator diameter, radial mode family, avoided crossings, inverse-DFT photonic-crystal modulation, dual-pump synthetic dispersion, pump-cavity desynchronization, GTI-based GDD compensation, SBS-assisted switching-wave selection, cavity phase matching, or intracavity $D_2$19 dispersion compensation. The shared objective is not a single preferred dispersion value, but a deliberately engineered nonlinear operating landscape in which bandwidth, coherence, spectral placement, and line-to-line power can be optimized for the target comb state.