Self-Injection Locking in Optics

• Self-injection locking (SIL) is a technique for frequency stabilization of lasers that uses a delayed, coherent, frequency-selective fraction of the laser's own output, which modifies the laser's oscillation phase and frequency,
• SIL extends from linear linewidth narrowing to nonlinear regimes, enabling applications such as soliton and platicon generation, Raman lasing, and coherent frequency conversion,
• Critical parameters in SIL include coupling efficiency, resonance quality factor, feedback phase, backscattering, and delay, with applications in compact narrow-linewidth sources, spectroscopy, and integrated frequency-comb generation.

Self-injection locking (SIL) is a dynamical frequency-stabilization phenomenon in which an oscillator receives a delayed, coherent, and strongly frequency-selective fraction of its own output. In optical implementations, a semiconductor laser couples light into a high-$Q$ resonator; resonant backscattering, engineered reflection, or nonlinear frequency conversion returns a phase-defined field to the laser cavity. The returned field modifies the laser’s oscillation phase and frequency, pulling emission toward a resonator mode and suppressing frequency fluctuations within a finite locking interval. Unlike conventional electronic stabilization, SIL requires no photodetector, frequency discriminator, high-bandwidth controller, or fast current-modulation path. Its operating regimes extend from linear linewidth narrowing to nonlinear self-injection locking, soliton and platicon generation, gain-switched-comb purification, Raman lasing, dual-laser synchronization, and parametric frequency conversion.

1. Physical principle and terminology

A typical SIL system consists of a semiconductor laser, an optical coupling path, a resonant structure, and a return path. The laser field enters a high-$Q$ cavity or microresonator, where clockwise and counter-clockwise modes are coupled by Rayleigh scattering or by an engineered reflector. A fraction of the counterpropagating field returns to the laser after a delay $\tau_s$. Because the resonator response is narrowband, the returned field has a frequency-dependent amplitude and phase rather than the broadband character of ordinary optical feedback.

The feedback loop can be represented schematically as

$$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$

A generic returned field has the form

$$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$

where $r(\omega)$ is the complex resonator-mediated reflection coefficient and $\phi$ includes propagation, scattering, coupling, and laser phase contributions. The laser remains the active oscillator, but its phase evolution is constrained by the field that has propagated through the resonator and returned to the diode.

The central distinction is between self-injection locking and ordinary external injection locking. In conventional external injection locking, a separate stable master oscillator injects light into a slave laser. In SIL, the injected field is generated from the same oscillator after propagation through its own resonant optical system. The resonator consequently acts simultaneously as a frequency-selective reflector, optical discriminator, and stabilizing feedback element.

For a WGM resonator, the principal parameters are the resonance frequency $\omega_m$, loaded linewidth $\kappa_m$, quality factor

$Q=\frac{\omega_m}{\kappa_m},$

Rayleigh coupling rate $Q$0, coupling efficiency $Q$1, feedback phase $Q$2, and round-trip delay $Q$3. The normalized backscattering parameter is commonly written as

$Q$4

The feedback phase includes the resonator frequency, propagation delay, and semiconductor-laser amplitude–phase coupling. One commonly used form is

$Q$5

where $Q$6 is the Henry linewidth-enhancement factor. The phase determines whether the returned field reinforces the frequency-restoring component of the laser dynamics or partially cancels it.

2. Locking dynamics and frequency stabilization

The free-running laser frequency and the actual emitted frequency are distinct in the locked regime. A convenient pair of normalized detunings is

$Q$7

where $Q$8 is the free-running laser-cavity frequency and $Q$9 is the generated frequency. Far from resonance, $\tau_s$0. Within the SIL regime, a large change in $\tau_s$1 produces only a small change in $\tau_s$2, yielding a nearly horizontal tuning curve.

In a stationary resonant-feedback model, the tuning curve can be expressed as

$\tau_s$3

with

$\tau_s$4

The zero-point stabilization coefficient is

$\tau_s$5

where $\tau_s$6 is the laser output-coupling rate and $\tau_s$7 describes spatial mode overlap. A pronounced locking plateau generally requires $\tau_s$8. In the small-backscattering, zero-phase limit, the normalized locking range is approximately

$\tau_s$9

The stabilization coefficient is the inverse slope of the tuning curve,

$$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$0

Inside a stable locking plateau, $$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$1. A frequency perturbation is approximately transformed according to

$$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$2

so frequency-noise power is reduced approximately by $$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$3, provided that resonator and external-path noise are not dominant. This relation describes short-term suppression rather than complete long-term frequency stabilization.

The locking range and stabilization coefficient are different optimization objectives. Critical coupling often maximizes the locking bandwidth, whereas maximum frequency stiffness can occur in the undercoupled regime. For small $$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$4, theoretical and experimental studies identify an optimum coupling coefficient near $$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$5 for stabilization, while critical coupling remains useful for robust capture and large modulation excursion (Galiev et al., 2020, Shitikov et al., 2022).

Backscattering has a nonmonotonic role. Weak scattering produces insufficient returned field; moderate scattering increases frequency discrimination and feedback; strong scattering produces a split resonance doublet and can reduce the useful discriminator slope. The short-delay optimum is approximately

$$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$6

A “semi-split” resonance is therefore generally preferable to either an almost unsplit resonance or a strongly split doublet. Optical delay can increase the calculated stabilization slope, but long paths also increase sensitivity to vibration, thermal drift, phase fluctuations, metastable fringes, and dynamical instability (Galiev et al., 2020).

3. Feedback architectures and optimization

Rayleigh-scattering feedback

In conventional WGM SIL, microscopic surface or volume imperfections couple clockwise and counter-clockwise waves. The backward wave returns through a prism, tapered coupler, bus waveguide, or butt-coupling interface. The scattering amplitude is determined by fabrication and material properties and is not generally independently tunable.

Experiments with crystalline MgF$$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$7 and related resonators demonstrate that weak Rayleigh feedback can produce substantial stabilization. A high-$$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$8 WGM can convert feedback powers of approximately $$\text{laser}\rightarrow\text{resonator}\rightarrow\text{backscattered or engineered reflected field}\rightarrow\text{laser}.$$9–$$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$0 of the incident power into strong spectral collapse and stable locking (Kondratiev et al., 2022). The same mechanism has been applied from the visible and near-infrared to the mid-infrared, including a $$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$1 laser stabilized to a crystalline-silicon WGM resonator with $$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$2 (Kondratiev et al., 2020).

Deliberate reflection and synthetic reflection

The uncontrolled feedback channel can be strengthened or replaced by an engineered reflector. A reflection from a BBO Brewster facet increased the capture range of a frequency-doubling laser from approximately $$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$3 with weak backward circulation to approximately $$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$4 with attenuated deliberate feedback and approximately $$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$5 with stronger feedback (King et al., 2018).

Mirror-assisted SIL uses a drop-port-coupled reflector. If $$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$6 is the drop-port coupling rate and $$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$7 is the complex mirror reflectivity, the effective counterpropagating coupling is

$$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$8

This makes the feedback level adjustable even when intrinsic Rayleigh scattering is weak. The proposed configuration can approach the stabilization performance of an optimally engineered Rayleigh-scattering resonator while operating farther from critical input coupling and therefore preserving more usable output power (Galiev et al., 2021).

A related approach uses a single coated auxiliary prism. The prism extracts a clockwise field, reflects it from a coated surface, and returns it to the resonator in the counter-clockwise direction. In a MgF$$E_{\mathrm{fb}}(t)=r(\omega)E_{\mathrm{L}}(t-\tau_s)e^{i\phi},$$9 WGMR, the auxiliary prism increased the selected TE-mode counter-clockwise feedback intensity by more than a hundred times relative to Rayleigh scattering. The resulting system reached approximately $r(\omega)$0 instantaneous linewidth, an $r(\omega)$1 locking range, and RIN of $r(\omega)$2 at $r(\omega)$3 (Wu et al., 2024).

Synthetic reflection can also be integrated directly into the resonator. A periodic angular corrugation in a silicon-nitride photonic-crystal ring creates designed coupling between counterpropagating modes at a selected resonance. With normalized coupling $r(\omega)$4, the engineered feedback enabled deterministic operation exclusively in the single-soliton regime (Ulanov et al., 2023).

Parameter identification through SIL

SIL can characterize the resonator and laser in situ. Measuring forward and backward locking ranges as a function of prism–resonator gap yields the coupling rate, intrinsic quality factor, vertical mode index, and effective laser output-coupling parameter. The maximum locking range identifies critical coupling, while the gap dependence distinguishes WGM vertical mode indices. At $r(\omega)$5, this method yielded $r(\omega)$6, in agreement with an independent cavity-ringdown value of $r(\omega)$7. It is particularly useful in the mid-infrared, where conventional linewidth and ringdown measurements may be limited by detector bandwidth and laser noise (Kondratiev et al., 2020).

4. Linewidth, noise, and frequency control

SIL suppresses fast frequency noise by modifying the oscillator dynamics rather than merely filtering the emitted light after generation. The returned field changes the laser phase condition, preferentially reinforces the resonator-selected mode, and can suppress longitudinal-mode competition in multimode Fabry–Perot lasers.

Linewidth must be interpreted together with the noise spectrum and measurement interval. Flicker frequency noise can dominate over white frequency noise, making a single linewidth ambiguous. An effective linewidth obtained from a $r(\omega)$8-separation-line analysis can therefore be much larger than the intrinsic Lorentzian linewidth inferred from a high-frequency white-noise plateau.

Representative results include the reduction of a $r(\omega)$9 DBR linewidth to approximately $\phi$0 in a frequency-doubling laser system, with an estimated intrinsic Lorentzian linewidth below roughly $\phi$1 (King et al., 2018). In WGM systems, measured stabilization coefficients have reached approximately $\phi$2, corresponding to an idealized linewidth estimate near $\phi$3, while direct measurements reported a $\phi$4 Lorentzian contribution, a $\phi$5 Gaussian contribution, and an instantaneous linewidth near $\phi$6 (Shitikov et al., 2022). These values are not contradictory: the Lorentzian component reflects white frequency noise, the Gaussian component includes slower fluctuations, and the instantaneous value emphasizes high-offset noise.

In gain-switched SIL, the resonator stabilizes the common optical seed while current modulation generates sidebands. Comb-tooth Lorentzian linewidths below $\phi$7 and Gaussian contributions below $\phi$8 were reported, with electrically tunable line spacing from approximately $\phi$9 to $\omega_m$0 (Shitikov et al., 2021). The sidebands need not be independently locked; they inherit coherence from the stabilized gain-switched seed.

SIL also permits frequency tuning within the locking interval. Current tuning is compressed by the stabilization factor, but the locked frequency can remain approximately linear over portions of the tuning curve. Frequency modulation from $\omega_m$1 to $\omega_m$2 was demonstrated using injection-current modulation while remaining inside the SIL regime. A coherent FMCW demonstration used a $\omega_m$3-km fiber delay and a $\omega_m$4 delayed self-heterodyne beat corresponding to a $\omega_m$5 chirp (Shitikov et al., 2022).

Slow frequency stability remains limited by resonator and path noise. Relevant mechanisms include thermorefractive noise, mechanical vibration, acoustic disturbances, absorption-induced thermal drift, laser current noise, and optical-path-length fluctuations. A narrow instantaneous linewidth therefore does not imply an absolutely stable optical frequency over long times.

5. Nonlinear self-injection locking and photonic dynamics

At sufficiently high intracavity power, the resonator is no longer a linear frequency discriminator. Kerr self-phase modulation (SPM), cross-phase modulation (XPM), four-wave mixing, Raman gain, and thermal shifts modify both the resonator response and the returned feedback field. The resulting process is nonlinear self-injection locking.

For counterpropagating modal amplitudes $\omega_m$6 and $\omega_m$7, a representative Kerr model contains

$\omega_m$8

with a corresponding equation for $\omega_m$9. The cubic convolution describes SPM, cross-mode four-wave mixing, and comb generation; the terms proportional to $\kappa_m$0 describe XPM between counterpropagating families. Nonlinear shifts can be represented by an effective detuning and an effective coupling, which must be solved self-consistently with the laser tuning curve.

The nonlinear model explains red-detuned locking, multivalued tuning curves, scan-direction dependence, and soliton formation. A linear SIL model predicts an accessible normalized detuning near $\kappa_m$1, insufficient for many dissipative Kerr soliton states. Kerr-induced resonance shifts displace the nonlinear locking curve toward larger red detuning, allowing the emitted pump frequency to enter the soliton domain (Voloshin et al., 2019).

Dissipative Kerr solitons and platicons

In anomalous group-velocity dispersion, nonlinear SIL can access dissipative Kerr solitons, soliton crystals, breathing states, and multi-soliton states. The SIL pump follows the nonlinear resonator response, allowing the effective detuning to remain nearly fixed while the free-running laser detuning changes. Experimental demonstrations with integrated silicon-nitride resonators produced single-soliton microcombs at approximately $\kappa_m$2 and $\kappa_m$3 repetition rates; a $\kappa_m$4 state exhibited repetition-rate phase noise near $\kappa_m$5 at $\kappa_m$6 offset (Voloshin et al., 2019).

In normal dispersion, SIL and backscattering can enable platicon formation. A platicon is a flat-top localized state bounded by switching fronts and is related to a pair of dark solitons. Numerical work found a backscattering-dependent threshold, with $\kappa_m$7 at $\kappa_m$8 and $\kappa_m$9 at $Q=\frac{\omega_m}{\kappa_m},$0 (Kondratiev et al., 2020). Experimentally, a fully integrated normal-dispersion silicon-nitride system generated a $Q=\frac{\omega_m}{\kappa_m},$1 platicon microcomb with approximately $Q=\frac{\omega_m}{\kappa_m},$2 CW-to-platicon conversion efficiency (Lihachev et al., 2021).

Deterministic single-soliton selection

SIL can reorganize the bifurcation structure in experimentally controlled coordinates. In the weak-backscattering limit, the forward field obeys a Lugiato–Lefever equation coupled to a backward-field average and an algebraic locking relation. Different $Q=\frac{\omega_m}{\kappa_m},$3-soliton states occupy different intervals in free-laser detuning and feedback phase. The maximum stable effective detuning decreases as soliton number increases, producing single-soliton-exclusive regions.

For representative parameters $Q=\frac{\omega_m}{\kappa_m},$4, $Q=\frac{\omega_m}{\kappa_m},$5, $Q=\frac{\omega_m}{\kappa_m},$6, $Q=\frac{\omega_m}{\kappa_m},$7, and $Q=\frac{\omega_m}{\kappa_m},$8, a single-soliton branch was the only stable soliton branch over $Q=\frac{\omega_m}{\kappa_m},$9. Numerical phase sweeps sequentially terminated three-soliton and two-soliton branches, leaving a single soliton at $Q$00 (Deshmukh et al., 29 Jun 2026).

Engineered synthetic reflection provides a related route. A selected coupling $Q$01 produced a substantial overlap between the SIL range and the single-DKS range. Nine consecutive on–off cycles repeatedly generated a single DKS, with reported pump-to-DKS conversion efficiencies of $Q$02 and $Q$03 (Ulanov et al., 2023).

Raman and other nonlinear actions

In aluminum-nitride microresonators, nonlinear SIL has been combined with stimulated Raman lasing. A $Q$04 pump generated a first-Stokes wave near $Q$05, with SRL output exceeding $Q$06 and a fundamental linewidth below $Q$07. Kerr interactions produced anti-Stokes emission near $Q$08 and second-Stokes emission near $Q$09. Thermal relaxation generated pump and SRL frequency excursions of approximately $Q$10 and $Q$11, respectively, while the effective detuning remained nearly constant during turnkey operation (Ding et al., 16 Feb 2025).

Nonlinear SIL can also use a gain-narrowed Brillouin Stokes mode as the feedback reference. In this architecture, the Stokes wave is shifted back to the pump frequency with an electro-optic modulator and reinjected into the semiconductor laser. Recursive interaction between the cleaner pump and the Brillouin oscillator produced an integrated linewidth near $Q$12 and a fundamental linewidth near $Q$13 (Bishop et al., 2023).

6. Applications, extensions, and limitations

SIL has been applied to laser cooling, compact narrow-linewidth sources, photonic microwave generation, frequency-modulated continuous-wave ranging, spectroscopy, optical communications, coherent lidar, and integrated frequency-comb generation. A frequency-doubling enhancement cavity simultaneously generated ultraviolet light and stabilized a $Q$14 DBR laser for the $Q$15 cooling and repumping transitions of $Q$16. The system produced two ultraviolet beams separated by $Q$17 and reduced short-term laser instability from the MHz level to approximately $Q$18 (King et al., 2018).

Multiple lasers can be stabilized by a common resonator. Dual-laser SIL to different silicon-nitride modes produced spectral collapse and linewidths of approximately $Q$19 and $Q$20 for two lasers. Locking both lasers to the same mode produced mutual frequency and phase stabilization, with coherent addition of their outputs (Chermoshentsev et al., 2022). Two counterpropagating SIL lasers locked to different modes of a crystalline magnesium-fluoride resonator generated non-degenerate four-wave-mixing sidebands with a pump separation of $Q$21, below $Q$22 of the $Q$23 free spectral range (Matsko et al., 10 Jun 2025).

SIL also supports correlated nonlinear generation. In a Kerr-nonlinear resonator pumped by two SIL lasers, the pump separation is pulled toward the difference between Kerr-shifted cavity modes. Counterpropagating phase matching enabled spatially separated signal and idler channels. The reported experiment established classical sideband generation and pump-power correlation; direct measurements of entanglement were not reported (Matsko et al., 10 Jun 2025).

The principal limitations are:

• Feedback-phase sensitivity: propagation length, vibration, and thermal expansion alter the phase of the returned field.
• Finite capture range: the laser must remain within the locking interval; excessive current or temperature excursions can cause unlocking or branch switching.
• Backscattering trade-offs: weak scattering gives inadequate feedback, whereas strong scattering produces split resonances, multimode states, or instability.
• Thermal drift: SIL suppresses fast laser noise but does not by itself define the long-term absolute frequency.
• Nonlinear instability: high intracavity power can induce four-wave mixing, Raman lasing, hyperparametric oscillation, breathing, or chaotic dynamics.
• Mode competition and dispersion: transverse-mode families, avoided crossings, higher-order dispersion, and mode-dependent coupling affect locking and comb formation.
• Output-power trade-offs: critical coupling can maximize capture range while reducing transmitted power; undercoupling may provide greater stabilization and more useful output.
• Model limitations: reduced theories may omit laser carrier dynamics, thermal dynamics, sideband depletion, pump-power variation, backward comb generation, or delayed-feedback instabilities.
• Integration constraints: butt coupling, phase control, thermal packaging, engineered scattering, and low-loss nonlinear materials remain important implementation requirements.

SIL is therefore not a single fixed operating condition but a family of coupled oscillator–resonator regimes governed by resonator quality factor, coupling, backscattering, feedback phase, delay, detuning, pump power, nonlinear response, and thermal dynamics. In the linear regime it provides passive optical stabilization and substantial linewidth narrowing. In the nonlinear regime it becomes a mechanism for selecting and maintaining optical states—including dissipative Kerr solitons, platicons, Raman lasers, gain-switched combs, and parametrically generated frequencies—while retaining the compactness and electrical tunability of semiconductor lasers.