Mid-infrared integrated silicon–germanium ring resonator with high Q-factor

The mid-infrared (MIR) wavelength range (between 3 and 15 µm) is attractive for sensing applications, as many molecules have strong fundamental absorption lines in this spectral region. Applications that benefit from this spectral range include environmental monitoring, defense and security, health care, and industrial process monitoring.¹ However, so far, the high cost and large size of MIR devices have slowed down the widespread use of this promising technology. Leveraging progress in group-IV photonic integrated circuits² and efforts made to integrate MIR components on chips with reduced cost, power consumption, and size should enable the large scale deployment of MIR technologies.

High quality factor ring (and racetrack) resonators are expected to play a crucial role in on-chip integrated sensing schemes. First, they could be used to generate optical frequency combs,³ and thus broadband MIR light sources for sensing devices, or they could serve as the basis for dual comb spectroscopy.⁴ Second, rings can also be directly used to enhance the interaction between light and the analyte molecules surrounding the ring.⁵ Whether it be via detecting some changes in the resonance wavelength or linewidth, most of the sensing schemes exploiting microring resonators require a high quality factor (Q-factor) to improve the sensitivity and detection limit of the sensor.^6,7

High Q-factors in ring resonators have already been demonstrated in the SWIR range (Short Wavelength InfraRed range, below 3.5 µm) in the silicon on insulator (SOI)⁸ and the silicon nitride on insulator (SiNOI)⁹ platforms with intrinsic Q-factors reaching, respectively, 1.1 × 10⁶ at 3.8 µm and 10⁶ at 2.6 µm, and even in the MIR in silicon on sapphire (SOS)¹⁰ reaching a loaded Q of 151 000 at 4.4 µm. While the SOI, SiNOI, and SOS platforms are very interesting for SWIR and low MIR wavelengths, they suffer from high intrinsic losses due to absorption caused by the lower cladding material beyond ∼3.5 and ∼6.5 µm, respectively.¹ Approaches using suspended Si rings¹¹ and III–V photonics¹² have not yet managed to solve this issue either. Material platforms that have low optical losses across a broad range of the MIR spectrum are thus desired to leverage the full potential of this technology.

In this context, silicon–germanium (SiGe) and germanium (Ge) are promising materials, thanks to their CMOS compatibility and transparency window that extends up to 15 µm.^1,13,14 Low-loss SiGe waveguides operating in the MIR range have been demonstrated,^14,15 and their nonlinear properties were used for supercontinuum generation up to 8.5 µm in SiGe and 13 µm in graded-index SiGe.^16–18 In pure Ge waveguides, supercontinuum generation was achieved up to 6 µm.¹⁹ So far, however, only a few SiGe and Ge ring resonator demonstrations have been reported in the MIR range, with moderate loaded Q-factors between 1700 and 20 000 (see Table I for details).^20–23 Very recently, high-Q resonators on Ge have been reported with an intrinsic Q = 250 000 at 7.8 µm wavelength.²⁴ However, this was achieved using Ge disks on glass pillars, hence hindering the prospects for a full integration within a circuit, as desired for applications.

Here, we report the experimental demonstration of a high Q SiGe ring resonator with a loaded Q-factor of 236 000 at 4.18 µm wavelength. In addition, our measurements on serpentine waveguides show that bending losses become “negligible” beyond 250 µm radius, indicating that intrinsic Q-factors of microrings on this platform could be as high as 1 170 000. Such high Q resonators achieved in a platform that has a very large transparency window (potentially up to 15 µm) should allow us to leverage the full potential of the MIR technology for applications such as sensing and optical frequency comb generation.

Our platform consists of an air-cladded Si_0.6Ge_0.4 core waveguide on a Si substrate [Fig. 1(a)]. Our intended wavelength of operation is 4.18 µm. In addition, the relevance for sensing applications around this wavelength, for example for sensing carbon dioxide,²⁵ this spectral band was shown to be a sweet spot in terms of SiGe nonlinear properties.¹⁷ To determine the optical properties at this wavelength, especially the refractive indices, ellipsometry measurements were performed. The refractive index of the transparent SiGe layer at 4.18 µm was n = 3.57 and, for the Si substrate, n = 3.42. Utilizing these values, we used the finite difference eigenmode (FDE) solver of Lumerical²⁶ to analyze the modal behavior and calculate the dispersion of the waveguides [including the material dispersion n(λ)]. An example for a mode profile (electric field intensity) of the fundamental mode in transverse-electric (TE) polarization is shown in Fig. 1(b).

As our long-term objective is the demonstration of MIR micro-combs, single mode operation and anomalous dispersion of the waveguides, close to zero dispersion, are preferable,^27,28 ideally in a small cross section waveguide to boost nonlinear effects. Cross section waveguide dimensions of w = 3.25 µm and h = 3.3 µm lead to the group velocity dispersion profile (D as a function of wavelength) for the TE0 mode plotted in Fig. 1(c), with a broad anomalous regime between 3.5 and 4.8 µm wavelength and single mode operation.

The ring radius was targeted to be as small as possible to reduce both the device footprint and, eventually, the threshold powers for nonlinear processes, while still being large enough not to be significantly impacted by bending losses. An R = 250 µm value was obtained, corresponding to a free spectral range (FSR) of 53 GHz. Bending losses are expected to be below 0.05 dB/cm from mode solver simulations. A straight bus waveguide was designed to side couple light to the ring resonator [see Fig. 1(a)] with a coupling gap of 240 nm, the smallest gap achievable by deep-UV lithography. We probed our resonators using free-space butt-coupling to the bus waveguide, and due to our sufficiently large mode diameter, mode adaptors on the end-facets were not necessary. A summary of all design parameters is provided in Fig. 1(a).

The ring resonators were fabricated on a 200 mm CMOS pilot line at CEA-Leti. The 3.3 µm thick Si_0.6Ge_0.4 layer was grown by epitaxy on a Si substrate. The choice of a 40% Ge concentration resulted in an alloy with a wide transparency window, high nonlinearity, and low propagation loss because of a limited number of misfit dislocations. Deep ultraviolet lithography was then used to pattern the waveguides followed by deep reactive ion etching. More details on the fabrication can be found in the supplementary material of Ref. 16. In addition, the etching process was optimized for fabricating the small 0.24 µm gap. This was needed to meet the difficulties in fabricating large aspect ratio structures,²⁹ in this case 3.3 µm/0.24 µm ≈ 14. Beside the rings, some serpentine waveguides (waveguides with a large number of subsequent bends) were also fabricated to determine bending losses [Fig. 1(f)]. Scanning electron microscope (SEM) pictures of a fabricated ring are shown in Figs. 1(d) and 1(e). As shown in Fig. 1(e), the gap was reliably obtained.

Before characterizing the ring resonator, the bending loss was measured to estimate the achievable intrinsic quality factor and to validate the choice of bending radius. To characterize the losses, we used a tunable optical parametric amplifier (OPA) pulsed laser source (MIROPA-fs, Hotlight Systems) with 5 mW input power, set at 3.9 µm. The power was chosen low enough to stay in the linear loss regime and not be influenced by the three and four photon absorption present in SiGe at this wavelength.¹⁷ The bending loss was determined by measuring the transmission in TE polarization through waveguides with a large number of subsequent bends (serpentine waveguides) and different bending radii (68–212 bends depending on the radius, adding around 1 cm of length) and comparing this to a 2.2 cm long straight reference waveguide. The total loss (propagation + bending loss) measured on the serpentine waveguides for increasing bend radii is shown in Fig. 2 for a waveguide width of 3.25 µm. We repeated this study for slightly larger waveguide widths (3.50 and 3.75 µm) but did not observe any clear effect on the bending loss, showing that the mode was well confined for these waveguide widths. For R = 250 µm, the loss is around 0.2 dB/cm. In fact, for this radius, the loss contribution caused by the bends becomes negligible as the propagation losses for serpentine waveguides with R = 250 µm and the straight waveguide (R = ∞) are indistinguishable within the measurement accuracy, which is consistent with our simulated bending loss estimation. These measurements are also in line with our previous demonstration of low propagation loss in the range of 0.1–0.4 dB/cm in straight SiGe waveguides for similar waveguide dimensions and wavelengths.³⁰ The loss α and the intrinsic Q-factor Q_in are related through Q_in = (10/ln 10) · (2πn/αλ). Using this, the measured loss of 0.2 dB/cm indicates that an intrinsic Q-factor Q_in = 1 170 000 should be achievable for the ring resonator.

Next, we characterized the ring resonators and measured their Q-factor. Figure 3(a) shows the measurement setup. The ring resonance characterization was performed using a narrow-linewidth continuous wave Adtech distributed feedback quantum cascade laser (QCL) with a small tuning range around 4.18 µm. The out-coupled light was recorded using either a PbSe photodetector or a Thorlabs Fourier transform infrared (FTIR) optical spectrum analyzer. The emission wavelength of the QCL was wavelength modulated between 4177.27 and 4177.89 nm by applying a periodic triangular current modulation, which, as a side effect, also resulted in a power modulation of roughly ±50%. The time-average optical pump power was set at around 20 mW (measured directly before the input coupling lens), which is well below the onset of nonlinear losses as 4.18 µm is beyond the three photon absorption cutoff and four photon absorption is negligible.¹⁷ 

Figure 3(b) shows one period of the triangular wavelength modulation recorded by the FTIR. The triangular power modulation was subtracted (raw data shown in the supplementary material, Fig. 1). Over the whole wavelength range, there are pronounced Fabry–Perot resonances with a FSR of around 17 pm. These are caused by reflections at the chip end-facets that are 2.2 cm apart. Superimposed on these fringes, the ring resonance can be clearly distinguished: it manifests as an additional dip occurring twice, during the wavelength ramp up and ramp down, respectively. Figure 3(c) shows the ring resonance after applying Fourier filtering to filter out the Fabry–Perot fringes. Some low-amplitude oscillations are visible as remainders of the Fabry–Perot resonances. No significant difference is observed between blue and red detuning, indicating that there are no thermal effects.

It can also be seen from Fig. 3(b) measurements that the resonance does not reach the noise floor and, instead, we have a transmission around 50% at resonance. This means that we are not at critical coupling (Q_in = Q_ext, where Q_ext is the external or coupling quality factor) and that the intrinsic quality factor of the ring (Q_in) is actually larger than the measured loaded Q-factor (1/Q_loaded = 1/Q_in + 1/Q_ext). A detailed discussion on the coupling regime can be found in Sec. II of the supplementary material.

This result represents a Q-factor improvement by more than one order of magnitude among Ge-based integrated microcavities, as can be seen from Table I. Our rings could, thus, be readily used for a range of sensing applications. This result is still a factor 3.4 smaller than the current record in the MIR range across all integrated platforms, held by SOI with a loaded Q around 800 000⁸ (derived value from data provided in reference). However, the measured loss is almost identical so the difference in Q-factor is likely because we are further away from critical coupling. Moreover, our demonstration was obtained at slightly longer wavelengths (4.18 µm against 3.8 µm in Ref. 8). Most importantly, our Ge-based microresonators should also work deeper into the MIR range, as the SiGe core waveguide is transparent up to 15 µm while record-holding SOI cavities have increasing losses for wavelengths beyond 3.5 µm. Our operation at 4.18 µm wavelength was, indeed, only limited here by the wavelength of our laser, and these results should, thus, be easily extendable to longer wavelengths. According to our propagation loss measurements on this platform,³⁰ we know that reducing the losses by another factor of 2 should be possible for the exact same material platform by optimizing the ring design (for example by choosing a lower loss wider waveguide and compensating for the decrease in coupling by Pulley or racetrack coupling schemes). This was also confirmed with our bending loss measurements, with possibly intrinsic Q-factors of 10⁶ in a future fabrication run.

To conclude, we have reported a SiGe ring resonator operating around 4.18 µm, with a loaded Q-factor of 236 000. This is the highest Q-factor achieved on this material platform and among the highest achieved around this wavelength range on a CMOS-compatible platform in general. Considering the demonstrated potential of the SiGe material platform, these rings should exhibit similarly high Q-factors at longer wavelengths, deeper in the MIR range. Measured losses in bent waveguides as low as 0.2 dB/cm for 250 µm bend radius show that even higher Q-factors should be achievable on this material platform. These results set the basis for various sensing applications in the MIR spectral range. We also target further improvements of the Q-factor to be able to achieve threshold powers low enough for experimentally demonstrating nonlinear processes. This would be another milestone at these wavelengths.

See the supplementary material for the raw FTIR scan and the discussion on the coupling regime.

We acknowledge the support of the Horizon 2020 research and innovation program under the Marie Sklodowska-Curie Actions (Grant No. ECLAUSion, 801512), the Agence Nationale de la Recherche (ANR) (Grant Nos. MIRSiCOMB, ANR-17-CE24-0028; MIRthFUL, ANR-21-CE24-0005), and the International Associated Laboratory in Photonics between France and Australia (LIA ALPhFA). This work benefitted from a France 2030 government grant managed by the French National Research Agency (Grant No. ANR-22-PEEL-0005).

The authors have no conflicts to disclose.

R.A. and M.P. contributed equally to this work.

Rémi Armand: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – review & editing (equal). Marko Perestjuk: Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – original draft (lead); Writing – review & editing (lead). Alberto Della Torre: Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Visualization (equal); Writing – review & editing (equal). Milan Sinobad: Conceptualization (equal); Visualization (supporting); Writing – review & editing (supporting). Arnan Mitchell: Conceptualization (supporting); Funding acquisition (equal); Supervision (equal); Validation (supporting); Writing – review & editing (supporting). Andreas Boes: Supervision (equal); Validation (equal); Writing – review & editing (equal). Jean-Michel Hartmann: Methodology (equal); Writing – review & editing (equal). Jean-Marc Fedeli: Methodology (equal). Vincent Reboud: Methodology (equal); Writing – review & editing (supporting). Pierre Brianceau: Methodology (equal). Alfredo De Rossi: Formal analysis (equal); Software (equal); Validation (supporting). Sylvain Combrié: Software (supporting); Validation (supporting); Writing – review & editing (equal). Christelle Monat: Conceptualization (equal); Funding acquisition (equal); Supervision (equal); Validation (equal); Writing – review & editing (equal). Christian Grillet: Conceptualization (equal); Funding acquisition (lead); Supervision (equal); Validation (equal); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.