1. Introduction

In the last decade, microresonator-based optical frequency combs have been intensively studied in various materials [1–11], and have successfully be tuned into the state of dissipative Kerr solitons [12–17] for diverse applications, including dual-comb spectroscopy, lidar, telecommunications, and optical frequency synthesizers [18–21]. Recently, stimulated Raman scattering (SRS) and its interplay with frequency combs have been investigated from various optical microresonators [22–28]. While the Kerr frequency comb bandwidth is mainly limited by dispersion, the SRS in microrings are governed by the frequency matching between the cavity resonances and the Raman gain. If the cavity resonance is in the Raman gain bandwidth, frequency comb can be generated at long or short wavelength regions by the Stokes or anti-Stokes scatterings, respectively. High order Stokes and anti-Stokes scatterings enable combs at even longer or shorter wavelength regions which is a good candidate for Mid-IR comb generator from IR input pump.

In this paper, we investigate Raman-assisted frequency comb generations with the help of the Stokes and anti-Stokes Raman scatterings in polycrystalline aluminum nitride (AlN) microring resonators. From our Raman scattering characterization, the optical phonons in polycrystalline AlN are found to exhibit broader Raman gain linewidths (295-351 GHz) than the reported values in single-crystalline AlN [29–33], which makes it favorable for microresonator-based Raman amplification due to the improved alignment tolerance with the cavity resonances. The general mechanism of the interplay between four-wave mixing (FWM) and SRS is discussed and confirmed in our experimental realization of SRS-assisted comb generation, for both TE and TM polarized modes. The observed comb spacing and the free spectral range (FSR) of ring resonators are consistent with our simulation by incorporating thermal effects. We also provide an explanation on the Raman gain and cavity resonance thermal shifts due to the heating from the high circulating power in the microring.

2. Experimental setup and results

Figure 1(a) illustrates the principle of the broadband comb generation assisted by SRS, where Raman or FWM peaks spring up when the circulating pump power in the AlN microring is higher than their respective thresholds. Figure 1(b) depicts a simplified measurement setup, where the pump light from a tunable continuous wave (CW) diode laser is boosted by an erbium-doped fiber amplifier (EDFA, up to 3 W) and is then coupled into the integrated microring resonators. The polarization of pump laser is adjusted to selectively excite TE or TM polarized cavity modes using a fiber polarization controller (FPC). An isolator is placed between the EDFA and the lensed fiber to protect the EDFA from the reflected light from the chip facet or microring. Two different OSAs are used for infrared (IR) and Mid-infrared light detection. A fiber Bragg grating filter (not shown) or 99/1 fiber couplers are used to attenuate the pump laser power to protect PD or OSAs.

 

Fig. 1 (a) Schematics of Raman-assisted broadband frequency combs with a repetition frequency (f_r). (b) Measurement setup and scanning electron microscope (SEM) image of AlN microring. CW: continuous wave, EDFA: Erbium-doped fiber amplifier, FPC: fiber polarization controller, DUT: device under test, BS: beam splitter, OSA: optical spectrum analyzer, PD: photo detector.

Download Full Size | PDF

The AlN microring resonators are fabricated using 700 nm or 800 nm-thick AlN-on-insulator wafers. The fabrication process is detailed elsewhere [34]. The SEM image of a typical AlN microring resonator without PECVD cladding is shown in Fig. 1 (b) as the device under test (DUT). The intrinsic quality factors of the fabricated devices vary from 500,000 – 1,000,000 depending on the waveguide size and ring radius. The basic properties of the Raman scattering in our poly-crystalline AlN film is characterized by a commercial Raman spectrometer, where straight AlN waveguides are employed as the sample at TE- or TM-polarized pump conditions. For TE-polarized input, strong Raman scattering is observed at ~610 cm⁻¹ and ~655 cm⁻¹, which correspond to the excitation of A₁(TO) and E₂(high) phonons, respectively (Fig. 2a). The A₁(TO) phonon is also observed from TM-polarized pump and shows a stronger scattering intensity (Fig. 2b). These numbers are close to the previously reported values in single crystalline AlN film [29–33].

 

Fig. 2 Raman scattering spectra measured using AlN straight waveguides for (a) TE and (b) TM pump, respectively.

Download Full Size | PDF

We then investigate the interplay between SRS and FWM in an AlN microring with TE mode excitation, which is enabled by slowly scanning the input wavelength across one of the fundamental TE mode resonances from blue-detuning side. When the circulating power in the micro-ring reaches the threshold of FWM or SRS, comb or Raman peaks start to be observed. Upon further tuning the pump laser into the resonance, more power circulates in the ring and leads to a broadband Raman-assisted comb with a spacing of ~1 FSR as marked in Fig. 3 (a). The blue comb band near the pump wavelength (1550 nm) is from FWM, while the black and red comb bands at shorter and longer wavelengths arise from the anti-Stokes and Stokes Raman scattering, respectively, as schematically illustrated in Fig. 1 (a). Their comb line spacing are found to be ~1 FSR based on the OSA measurement, which indicates that the Raman comb lines originate from the FWM process with comb at pump.

 

Fig. 3 (a) Raman-assisted broadband comb spectrum from an AlN microring (height = 700 nm, radius = 18 μm, and width = 1.54 μm) at TE mode input. Inset shows the TE mode profile at 1.55 μm wavelength. (b) The measured comb line spacing and the simulated TE₀₀ mode FSRs versus the wavelength at two different temperatures. Inset and yellow dot: measured FSR from the TE₀₀ transmission curve at low input power. (c) Schematics of the resonance thermal shift of the microring. (d) Zoom-in of the spectrum near 1700 nm wavelength in (a), where the modulation instability (MI) comb and Stoke comb lines overlap. (e) The measured spectra after passing TE (black) and TM (magenta) polarizers. (f) The Raman comb lines in low power regime. The Stokes and anti-Stokes peaks from E₂(high) phonon modes are clearly seen. Inset: measured and expected E₂(high) Raman gain at RT (black) and HT (red), respectively.

Download Full Size | PDF

The measured FSR and comb line gaps for each band are plotted in Fig. 3 (b) with simulation results considering the thermo-optic effects due to the local heating from the high circulating power. The FSR of TE₀₀ mode at room temperature (RT) is measured to be 1219.7 GHz at 1550 nm wavelength with 0.1 mW input power (yellow dot and inset), which matches with the simulated FSR at RT (black curve). The comb line gaps extracted from the IR OSA (black circle) show higher error than the Mid-IR OSA (blue circle) due to the limited resolution of the IR OSA we used. Even considering the measurement error, the comb line gap is smaller than the measured FSR at RT.

To explain this discrepancy, we take the thermo-optic effect into consideration, since a portion of optical pump would be absorbed by the poly-crystalline AlN and converted into heat. As schematically depicted in Fig. 3(c), when the laser wavelength approaches the resonance, the absorbed optical power in the AlN microring increases the local temperature, which also increases the refractive index and reduces the FSR and therefore pushes the resonance to the longer wavelength. Based on the resonance thermal shift (dλ) and the AlN thermo-optic coefficient ($\frac{dn}{dT}=$ 2.32 × 10⁻⁵/K) at 1550 nm wavelength [35], the temperature inside the microring is estimated using the equation, $\frac{d\lambda }{dT}=\frac{\lambda }{n}·\frac{dn}{dT}$.In this experiment, the resonance thermal shift is ~9 nm, which corresponds to a temperature increment of 500 ͦ C inside microring. By assuming the intracavity temperature of 500 ͦ C, we found that the simulated FSRs of TE₀₀ mode (HT, blue curve) pass the mean value of comb line gaps (~1210 GHz).

A closer examination of comb lines near the region where pump comb (blue) and Stokes comb (red) meet, shown in Fig. 3(d), reveals a ~95 GHz mismatch between the two. This implies that the Stokes comb lines are initiated by SRS, not from pump comb in cascaded way. Since it is also possible to involve the modes with different polarization, we check the polarizations of the Raman combs by polarizers. As shown in Fig. 3 (e), the power of each comb line after passing TE polarizer is 10 − 20 dB higher than that after the TM polarizer, proving that the input TE mode is conserved after FWM and SRS.

Figure 3 (f) shows the spectrum generated at a lower circulating power from another device with the observation of the 1st, 2nd Stokes and 2nd anti-Stokes comb lines. The measured Raman shifts (purple arrow in the inset) are smaller than that of the E₂ (high) phonon at room temperature (inset, black curve). However, if the temperature dependent Raman shift is considered, the Raman shift is near the center of Raman gain at high temperature (inset, red curve) [30–32]. Here, the first anti-Stokes peak is not observed likely due to the missing cavity resonance within the first Raman gain bandwidth, as frequency matching is essential for SRS in microrings [27].

The Raman scattering within AlN microrings is also investigated with TM excitation. Figure 4(a) shows the spectrum of Raman-assisted combs when the TM pump is injected into the ring resonator with a height of 800 nm, a radius of 40 μm and a width of 2.2 μm. The comb lines around the first order Stokes and anti-Stokes wavelength are both observed. The Raman shifts from this spectrum are slightly smaller than the measured values of A₁(TO) phonon in Fig. 2(b) due to the mismatch between the cavity resonance and the center of Raman gain, but within the Raman gain bandwidth of 9.8 cm⁻¹.

 

Fig. 4 (a) Raman comb lines from an AlN microring (height = 800 nm, radius = 40 µm and width = 2.2 µm) at TM mode input. Inset: TM mode profile at 1550 nm. (b) Primary Raman peaks without comb line generation at low power from a 2.4 μm-width microring. (c) Raman comb lines from the 2.4 μm-width microring with high optical input power. The Raman shifts in this case are roughly one FSR smaller than the case of the low power input due to the thermal shift of the resonance and Raman gain of the AlN at high power regime. (d) The measured FSR (inset), comb line spacing (black, blue and red circles) and simulated FSRs at RT (black) and HT (green). (e) The measured A₁(TO) Raman shift in Fig. 2(b) (black curve) and the estimated Raman shift at HT (red curve). Red and blue arrows indicate the observed Raman shifts from (a), (c) and (b), respectively. (f) Schematics of the microring resonances and Raman gain at different temperatures. The elevated temperature changes the location of the resonance and Raman gain, which result in smaller effective Raman shift in microring.

Download Full Size | PDF

We further investigate the Raman shifts at different temperatures under TM excitation. Figure 4 (b) shows the first and second order Stokes lines from a 2.4 μm-width AlN microring at a relatively low pump power which also means low intracavity temperature. We observe the Raman shift of ~618 cm⁻¹ which are larger than that in Fig. 4(a). Using the same microring, however, the smaller Raman shift of ~599 cm⁻¹ is recorded when a higher pump power is applied. (Fig. 4 (c)). Note that the Raman shift difference between Fig. 4 (b) and (c) is ~19 cm⁻¹ which is very close to microring FSR (17.7 cm⁻¹). This shows the Raman shift in microring is discrete and highly depends on the cavity resonance and Raman gain overlap.

The extracted Raman comb line spacing and the simulated TM₀₀ FSRs at two different temperatures are shown in Fig. 4(d). The gaps of the Raman comb lines (blue circles for 1st Stokes and red circles for 2nd Stokes) are extracted from high-resolution OSA and show 528.6 ± 0.5 GHz. The yellow inset is the measured FSR using low input power and shows 532.1 GHz which is out of standard deviation of comb line spacing. The difference between FSR at low power and Raman comb spacing is again, due to the cavity heating. The simulated TM₀₀ FSR at RT (black curve) passes the FSR measured at low power, and the FSR at 500 ͦ C also very close to comb line spacing. From these experiment and simulation, we conclude that the heating of microring changes the FSR and comb line spacing.

In Fig. 4(e), the different Raman shifts from Fig. 4(a-c) are compared with the Raman spectroscopy measurement. The red and blue arrows stand for the extracted Raman shifts from Figs. 4(a), 4(c), and 4(b), respectively. The red and blue bands represent the expected resonances location relative to the pump frequency at high temperature (HT) and RT, respectively. The black curve is the measured A₁(TO) phonon spectrum at RT. At high temperature, the Raman gain shifts to lower frequency, as drawn by the red curve [30–32]. When the input power is low or the cavity temperature is near RT, a Raman peak is generated on the overlap region of Raman gain and blue resonances (b). With high input power or cavity temperature, both the Raman gain and cavity resonance shift, but the overlap region still can be found due to the wide Raman gain linewidth of poly-crystalline AlN. As a result, ~1 FSR (17.7 cm⁻¹) smaller Raman shift (a), (c) are observed than that of (b). Due to the additional comb lines at high power, all the cavity resonances near Raman gain produce Raman-assisted comb lines.

Figure 4(f) schematically explains more details of the Raman gain band locations relative to the cavity resonances at two different temperatures (RT and HT). The initial Raman shift in microring (Ω_RΤ + δ) at low power (ie, RT) is observed on the overlap region of Raman gain and blue resonance (b), which is larger than the Raman shift in straight waveguides (Ω_RΤ), but the offset δ is within the Raman gain bandwidth (blue Gaussian). In the high power regime, the Raman shift in microrings (c) is Ω_HT + δ (δ is negative in this case) which is different from other Raman shift values due to the thermally shifted resonances (red curve) and Raman gain (red Gaussian).

3. Conclusion

In conclusion, we experimentally demonstrate Stokes and anti-Stokes frequency combs using AlN microrings. For both TE and TM modes, the first and second order Raman combs are observed. The comb line spacing around the pump, Stokes and anti-Stokes bands match with the simulated FSRs within the measurement error, by taking the thermal shift at high input power into consideration. In addition, we confirm that the measured Raman shifts in ring resonators are different from straight waveguide Raman shift as they require the overlap between Raman gain bandwidth with certain cavity resonances in microrings. The different Raman shifts in microrings at different temperatures are also explained by analysis of the thermal effects in high powered microring to the Raman gain and resonant frequencies. This Raman-assisted comb generation mechanism can expand the comb bandwidth alleviating dispersion limitation. Even though the experimental demonstrations presented in this work are based on incoherent combs, the mechanism of the Raman-assisted comb bandwidth extension is general and expected to be applied to the soliton mode locked combs for future applications [17,22].