Dispersion-managed octave soliton microcombs in heterostructured microresonators

The paper demonstrates a robust strategy for broadband dispersion management using heterostructured microresonators comprised of adiabatically concatenated waveguides with distinct geometries. By meticulously engineering the local dispersion profiles, the authors flexibly synthesize the global effective dispersion coefficients of different orders, effectively expanding design degrees of freedom beyond conventional limits. As a benchmarking demonstration, they fabricate heterostructured silicon nitride microresonators using a commercial foundry process and successfully generate octave-spanning soliton microcombs with a repetition rate as low as 118 GHz. These microcombs feature deterministically tunable dispersive waves operating across the 290–310 THz range. Such octave-spanning microcombs with detectable repetition rates and carrier-envelope offset frequencies are readily applicable to f-2f self-referencing in optical clocks and frequency synthesizers. The proposed heterostructured architecture establishes a versatile paradigm for generating ultrawideband soliton microcombs with tailorable spectral profiles.

The key innovation is that in conventional uniform microresonators, target values for the free spectral range and second-order dispersion are typically achieved by co-optimizing the bending radius and waveguide width once the waveguide height is fixed. However, during this geometric tuning process, higher-order dispersion coefficients shift inherently with β2, precluding the capability for independent parameter control. In sharp contrast, heterostructured microresonators bypass this rigid coupling constraint by incorporating multiple distinct dispersive segments. By strategically adjusting the fractional lengths of these elements, higher-order dispersions can be engineered independently while anchoring the desired β2, thereby unlocking a vastly expanded and highly customizable design parameter space.

The global effective dispersion is given by the weighted average of the dispersion coefficients of each segment: βm,eff = Σ ai βm,si, where ai = Lsi/Ltotal is the fractional length of segment si. The accessible dispersion space expands to form a complete convex hull, effectively filling and extending far beyond the outermost boundaries of uniform resonators. Quantitatively, the uniform microring space accounts for only 34%, 44%, and 15% of the total area accessible by the heterostructured counterparts in the (β2, β3), (β2, β4), and (β3, β4) planes, respectively.

The fabricated microresonator incorporates three distinct waveguide geometries: a 0.9-µm-wide straight waveguide as segment s1 to suppress avoided mode-crossings between the fundamental and higher-order modes; a 3.0-µm-wide adiabatic bend as segment s2 to ensure broadband coupling with the bus waveguide, featuring a quadratically varying curvature; and a 3.0-µm-wide straight waveguide as segment s3, acting as the tuning element to globally balance the cavity dispersion. The effective second-, third-, and fourth-order dispersion coefficients are evaluated via a multi-objective constraint framework: setting β3,eff = 0, assigning opposite signs to β2,eff and β4,eff, constraining a target dispersive wave to emerge at a specific self-referencing frequency (e.g., 300 THz), and enforcing the normalization condition Σ ai = 1.

For the benchmark device MR1, the measured transmission spectrum and intrinsic Q factors were extracted, and the measured integrated dispersion data agree well with the theoretical curves. The experimental single-soliton spectrum shows dispersive waves near 130 THz and 300 THz, and the microcomb repetition rate was downconverted and detected via a commercial photodetector, confirming a low-noise mode-locked soliton state at 123.96 GHz.

For dispersive wave tuning, under the constraint of zero third-order dispersion, regulating β2,eff enables precise tuning of the dispersive wave positions by systematically varying the weight relationships among s1, s2, and s3. The optimized dispersion parameters lie beyond the reach of conventional uniform architectures. The integrated dispersion curves demonstrate a flexible spectral tuning range spanning from 290 THz to 310 THz. The experimental single-soliton spectral envelopes exhibit excellent agreement with theoretical simulations on the short-wavelength side; on the long wavelength side, the measured comb power undergoes a noticeable roll-off due to elevated propagation losses of the 0.9-µm-wide intracavity straight waveguide segment in the long-wavelength regime.

To characterize the coherence of the generated dispersive waves, an f-2f self-referencing interferometry setup was constructed to detect the carrier-envelope offset frequency. A wavelength division multiplexer splits the broadband microcomb into long- and short-wavelength arms. Within the long-wavelength arm, specific comb lines are selected, amplified via cascaded thulium-doped fiber amplifiers, and focused into a periodically poled lithium niobate crystal for second-harmonic generation. The short-wavelength comb lines are independently amplified by a semiconductor optical amplifier. The fceo beat note was measured at a resolution bandwidth of 300 kHz, exhibiting a signal-to-noise ratio of approximately 15.8 dB.

In summary, the authors experimentally demonstrated a global dispersion synthesis paradigm utilizing heterostructured microresonators, successfully realizing single-chip, octave-spanning dissipative Kerr soliton microcombs. The custom-engineered dispersion profiles facilitate detectable repetition rates ranging from 118 to 124 GHz, featuring tailorable dispersive waves in the 290 to 310 THz range. Consequently, the direct detection of the fceo frequency via f-2f self-referencing was achieved with a signal-to-noise ratio of 15.8 dB.

Improvements to AI Systems:

1. Multi-Objective Constraint Optimization for Inverse Design

• Improvement: Integrate the paper’s multi-objective constraint framework (e.g., setting β3,eff = 0, opposite signs for β2 and β4, target dispersive wave frequency, normalization Σ ai = 1) into AI-based design tools. This enables simultaneous optimization of multiple, often conflicting, dispersion parameters.

• Capability: An AI system can now automatically propose heterostructured microresonator geometries (segment widths, lengths, curvatures) that meet precise dispersion targets, expanding beyond uniform designs. It can solve for fractional lengths ai to achieve user-defined β2, β3, β4 values, enabling rapid prototyping of custom microcombs for specific applications (e.g., optical clocks, frequency synthesis).

2. Convex Hull Expansion for Design Space Exploration

• Improvement: Use the convex hull concept (where accessible dispersion space is the weighted average of segment coefficients) to train generative models or reinforcement learning agents. The AI can learn to navigate the full convex hull, not just the limited uniform-resonator subspace (which covers only 15–44% of the accessible area).

• Capability: The AI can generate novel heterostructured designs that achieve dispersion combinations previously impossible, such as independent control of β3 and β4 while fixing β2. This unlocks new spectral shapes, enabling ultra-broadband combs with tailored dispersive wave positions across 290–310 THz.

3. Physics-Informed Surrogate Modeling for Dispersion Prediction

• Improvement: Train neural networks on simulated or experimental dispersion data (e.g., from the fabricated Si3N4 devices) to predict global effective dispersion coefficients (βm,eff) from segment geometries. Incorporate adiabatic bend effects, mode-crossing suppression, and propagation losses as inputs.

• Capability: The AI can rapidly predict performance (e.g., integrated dispersion, Q factors, soliton stability) for arbitrary heterostructured designs without full FDTD simulations. This accelerates iteration, enabling real-time design feedback and automated discovery of high-performance microresonators.

4. Autonomous Dispersive Wave Tuning via Reinforcement Learning

• Improvement: Implement a reinforcement learning agent that adjusts fractional lengths (a1, a2, a3) to steer dispersive waves to a target frequency (e.g., 300 THz) while maintaining zero β3 and opposite-sign β2/β4. Use the experimental tuning range (290–310 THz) as the action space.

• Capability: The AI can autonomously calibrate microresonators in situ, compensating for fabrication variations or thermal drift, by dynamically rebalancing segment lengths. This yields deterministic, reproducible microcombs with user-specified repetition rates (118–124 GHz) and dispersive wave positions.

5. Self-Referencing Readiness Prediction

• Improvement: Train a classifier or regression model to predict the feasibility and signal-to-noise ratio (SNR) of f-2f self-referencing (e.g., the 15.8 dB beat note) based on simulated comb spectra, dispersion parameters, and pump conditions.

• Capability: The AI can pre-screen designs for octave-spanning comb generation with detectable carrier-envelope offset frequencies, reducing experimental trial-and-error. It can also suggest optimal pump power, coupling, and amplification strategies (e.g., thulium-doped amplifiers, PPLN SHG) to maximize beat-note SNR.

6. Loss-Aware Spectral Envelope Correction

• Improvement: Incorporate wavelength-dependent propagation losses (e.g., the roll-off in the 0.9-µm-wide segment at long wavelengths) into AI models that predict comb spectra. Use transfer learning from the paper’s experimental data to correct for these losses.

• Capability: The AI can generate realistic spectral envelopes for heterostructured microcombs, including long-wavelength roll-offs, enabling accurate prediction of comb line power distribution. This improves system-level integration with photodetectors, amplifiers, and frequency combs for metrology applications.