a Optical and scanning electron microscope (OM, SEM) images of the mid-IR ultra-high-Q (UHQ) resonator on a chip. The upper-left figure presents the concept of flip-chip coupling between the bus waveguide and the resonator. The SEM image on the right shows a cross-section of the fabricated multilayer UHQ resonator with false coloring for visibility. The lower figures show the OM and SEM images of the actual resonator with varying radii. b Schematic of cavity-enhanced light-matter interactions that can occur using UHQ mid-IR microresonators. c Examples of possible applications in nonlinear optics and molecular physics using the phenomena listed in b.

In this paper, we present a different approach to realize UHQ resonators in the mid-IR spectral range, obtaining Q-factors of 38 million at 3.86 μm wavelength, which represents over a 30-fold enhancement relative to prior works^{5,26,27,28,29}. Through an approach where light-guiding geometries are defined during the deposition of mid-IR optical materials, eliminating the necessity to etch them, resonators having internal multilayer structures have been seamlessly fabricated and passivated with chalcogenide glasses within a single chamber, thus eliminating interstep contamination. The primary sources of optical loss in the mid-IR are extensively analyzed and mitigated by optimizing the cavity structure and fabrication conditions, including material absorption, surface scattering, and surface-adsorbed airborne chemical absorption whose importance is revealed here. Leveraging the high Q-factor and controllability of FSR afforded by microfabrication, stimulated Brillouin lasing is demonstrated for the first time in the mid-IR, exhibiting a threshold power of 91.9 μW, well below the output power of on-chip OPO lasers and QCLs, and a theoretical Schawlow-Townes linewidth of 83.5 Hz, significantly surpassing that of commercialized QCLs.

a The fabrication steps for the proposed mid-IR UHQ resonator are depicted, with a detailed explanation given in the Methods. The different colored ChG layers represent the bottom cladding - (green), the upper cladding - (pink), and the core layer (orange). The bottom left corner shows an example of the calculated optical mode in a resonator having the same geometric parameters as seen in Fig. 3b. b Surface roughness measured by AFM at the top and wedge surfaces of the trapezoid. Roughness data are shown for the SiO₂ substructure and after sequentially depositing one to three layers of ChG on that structure. The numbers show the root mean square (rms) roughness measured over a 3 × 3 μm² area. The thickness and material of each layer are as follows: first layer: 1.1 μm As₂S₃, second layer: 2.2 μm As₂Se₃, third layer: 0.9 μm As₂S₃.

To adapt this near-IR-proven technique for mid-IR applications, a bottom cladding layer made of ChG is incorporated to counter the Q-factor reduction caused by the excessive material loss of SiO₂ in the mid-IR (460 dB/m at 3.7 μm)³⁹ As shown in Fig. 2a, the optical modes are contained within the core layer, effectively separated from the SiO₂ platform by the bottom cladding layer. For our specific choice of ChGs, with As₂Se₃ (n_core: 2.77 at 3.7 μm) as the core and As₂S₃ (n_clad: 2.41 at 3.7 μm) as the cladding, bottom cladding thicknesses of a few microns were confirmed to sufficiently suppress SiO₂ absorption loss, even at longer mid-IR wavelengths (Supplement 1E). The proposed resonator featuring an internal multilayer structure can be easily fabricated by depositing ChGs successively in a single high vacuum (HV, <5 × 10⁻⁸ torr) thermal evaporation chamber. In addition, the deposition reduces the surface roughness that is initially present on the bottom SiO₂ platform to sub-nanometer levels at the core-cladding interfaces, as confirmed by atomic force microscopy (AFM) measurements shown in Fig. 2b.

To further approach the loss limits of ChG materials, absorption losses resulting from in-process contamination and impurities originating from the source materials, which have been extensively studied in low-loss mid-IR ChG fibers^32,40,41, have been minimized. Chemical contamination during fabrication was inherently blocked by sequentially depositing the internal multilayers in a single HV chamber, in contrast to traditional microfabrication methods that involve multiple steps of deposition and etching. In addition, prior to deposition, the SiO₂ substructure is cleansed with O₂ plasma inside the deposition chamber to remove surface contaminants. Impurities originating from the sources in the deposited films were suppressed by selecting highly purified ChGs, similar to those used for mid-IR fibers, as the deposition sources. The sources are purified to systematically eliminate hydrogen and oxygen-related impurities, as detailed in the Methods. In particular, a double-purified source⁴⁰ was used for core deposition, which led to a notable improvement in the Q-factor, almost doubling from 1 × 10⁷ to 2 × 10⁷, as discussed in Supplement 1C.

In addition, a mid-IR passivation material that can preserve the ultra-high-Q while providing sufficient degradation resistance was applied and scrutinized. Although fluoropolymer films and atomic layer deposited (ALD) films have traditionally been utilized for passivating mid-IR devices, they introduce significant losses owing to their intrinsic absorption band⁴² or poor chemical purity⁴³. Here, we verified Ge-As-Se, a ChG variant known for its high mid-IR transmittance⁴¹ and stability under ambient conditions, as a passivation layer for mid-IR UHQ resonators. Examination revealed that a 10 nm thick Ge-As-Se film, deposited consecutively within the same HV chamber, effectively prevented crystallization of the ChG core and claddings (Supplement 1B), without causing an observable change in loss (Supplement 3). As a final step, a post-treatment by thermal and light annealing under a medium vacuum condition (<1 mTorr) is applied to minimize local irregularities inside the deposited films, resulting in an additional increase in the overall Q-factor (approximately twice from 5 × 10⁶ to 1 × 10⁷), as presented in the Methods and Supplement 1C.

Post-fabrication loss from surface-adsorbed molecules

The strong absorption associated with fundamental molecular vibrations^31,42 and wavelength proportional longer evanescent fields make mid-IR on-chip devices susceptible to absorption loss from molecules that continuously adhere to the surface from the surrounding air. This post-fabrication degradation requires the addition of upper claddings to be done with more consideration, for instance, compared to near-IR devices where molecular absorptions are negligible, or mid-IR optical fibers with extremely thick claddings. Therefore, besides the aforementioned refinements in device design and fabrication, it is crucial to quantitatively assess the absorption loss resulting from surface-adsorbed airborne chemicals and suppress it with adequately designed upper claddings.

This loss factor was thoroughly analyzed by measuring the Q-factor of resonators with various upper cladding thicknesses across a wide spectral range. Figure 3a depicts the measured Q-factor, covering wavelengths from 3.2 µm to 3.9 µm, for resonators with two contrasting cladding thicknesses, 0 (no cladding) and 800 nm, respectively, as illustrated in Fig. 3b. It is obvious that a cladding thickness of 800 nm is enough to enhance the overall Q-factor spectrum by approximately twenty times. The relatively lower Q-factors of the resonator without cladding are not constrained by the fabrication issues described in the previous section but rather are attributed to molecular absorption, as evidenced by the Q-factor exceeding 10⁷ at 1550 nm, which is discussed in Supplement 2. Airborne impurities adhere to the outermost surfaces of both resonators in similar concentrations because of identical surface properties defined by the Ge-As-Se passivation layer, which remains unaffected by the upper cladding thickness. Consequently, within the regime where surface molecular absorption dominates, the loss (Q-factor) is linearly (inversely) proportional to the outermost surface electric field intensity, diminishing exponentially with increasing cladding thickness. This trend was validated experimentally using resonators with various upper cladding thicknesses (Supplement 3), enabling quantification of surface molecular absorption. By incorporating this loss factor into the previously established loss calculation based on material absorption and surface scattering, the Q-factors of a given resonator geometry could be accurately predicted over wavelengths from 3 to 4 μm, as described in the Methods. This analysis, delineating the contribution of each loss factor, offers insight for further enhancing Q-factors by increasing the upper cladding thickness, though a technical issue coming from our current point-coupling scheme limited it to 800 nm in this study (Supplement 5).

Fig. 3: Q-factor spectra and measurement set-up.

a Q-factor spectrum of two different resonators, with and without an upper cladding, measured at wavelengths of 3.2-3.9 μm. The grey-shaded region at the 3.33-3.55 μm wavelength indicates the hydrocarbon (C-H) absorption band, with their characteristic peaks indicated with black arrows. The grey dashed line is given as a guideline for the effect of the H₂O absorption tail. b Cross-section drawing of the resonator with an upper cladding used for the Q-factor measurement represented by the red line in Fig. 3a, which has geometric parameters of ?=6.1 ?m,?=28°,?1=1.1 ?m,?2=2.4 ?m,?3=0.8 ?m. The parameters for the ‘without upper cladding’ case, represented by the blue line in Fig. 3a, are ?1=1.0 ?m,?2=2.0 ?m and ?3=0 ?m, with more details provided in Supplement 2.c Lorentzian fit of the record Q-factor measured at 3.86 μm wavelength on the same resonator. The magenta line indicates the resonator transmission spectrum, while the navy line is the interferogram of the calibrated Michelson interferometer (MI, 75.25 MHz), depicted in Figure d, used as a frequency reference. d A drawing of the mid-IR cavity coupling setup. A continuously tunable optical parametric oscillator (OPO) laser was used as the pump source. Details about the setup and interferometer are given in the Methods. BS: beam splitter, VA: variable attenuator, PC: polarization controller, L: lens, Res: resonator, WG: bus waveguide, PD: photodetector, OSC: oscilloscope, OSA: optical spectrum analyzer.

Precise measurement of Q-factor spectra, calibrated against a frequency reference (see Fig. 3c, d and Methods), provides sufficient resolution to discern the distinct contributions of specific molecular groups to absorption loss, based on their unique spectral fingerprints. Within the measurement range shown in Fig. 3a, two primary contaminants are identified: water and hydrocarbons (C – H). The extended tail of the well-known broad H₂O absorption band, centered at approximately 2.9 μm^31,42, imposes a limitation on the overall Q-factor, as indicated by the grey dotted line in Fig. 3a. Hydrocarbons, on the other hand, induce sharp decreases in Q-factors, corresponding to their characteristic absorption peaks at 3.38, 3.42, and 3.52 μm⁴². This highlights the importance of carefully managing airborne chemicals that exhibit significant absorption within the spectral band aimed for UHQ resonator operation, such as regulating surface adhesion and cladding thickness.

It is worth highlighting that the intrinsic Q-factor (Q_int) of 3.81 x 10⁷ at a wavelength of 3.86 μm (Fig. 3c) represents a remarkable breakthrough in the mid-IR range, surpassing all previously reported values by a factor exceeding 30, including those obtained using ChG²⁷ or any other on-chip materials such as silicon⁵, silicon nitride⁴⁴, germanium²⁹, SiGe²⁶, and III−V semiconductors²⁸. This achievement significantly reduces the required pump power for most nonlinearities by approximately a thousand times (∝1/Q²). This in turn enables the realization of numerous optical functionalities on a chip without being constrained by the absence of commercial amplifiers in this region. Furthermore, the propagation loss of 0.52 dB/m converted from this Q-factor is only twice as much as that of the state-of-the-art ChG fiber made from the same ingots, 0.25 dB/m⁴⁰, thus verifying that all analyzed loss factors have been sufficiently minimized to achieve performance comparable to the fiber. Figure 4 shows a comprehensive visual comparison, including the mentioned loss values.

Fig. 4: Comparison of optical losses with different material platforms on a chip.

The propagation losses of diverse mid-IR on-chip resonators and waveguides are compared with this work. Different markers indicate different material platforms: chalcogenide glasses (black, circle), silicon nitride (brown, diamond), silicon (blue, triangle), germanium (purple, square), III−V semiconductors (orange, hexagram), and SiGe (green, reversed triangle). The solid symbols represent resonators, and the open symbols indicate waveguides. The optical losses of this work are converted from the record Q-factors shown as the red line in Fig. 3a. The loss of state-of-the-art As₂Se₃ fiber fabricated using identical source materials in this work is drawn as a grey line. All references are given in Supplement 8.

An on-chip Brillouin laser in the mid-IR

Taking advantage of the ultra-high Q-factor of the microresonators, a Brillouin laser was developed for the first time in the mid-IR. The employed resonator has a cross-sectional design identical to that shown in Fig. 3b. As confirmed in Fig. 5a, this particular design confines, within the As₂Se₃ core layer, the fundamental optical mode and the phonon mode that yields the highest Brillouin gain. These modes were identified using the finite element method, treating the device as a straight waveguide, resulting in a modal overlap factor of around 0.96⁴⁵. A full-vectorial model of the Brillouin coupling⁴⁶ predicts an appreciable Brillouin gain, G_B, of 59.9 m−1W−1. This is attributed to the high overlap factor and the inherently large Brillouin gain coefficient of As₂Se₃⁴⁷ exceeding that of conventional on-chip optical materials such as SiO₂⁴⁸, Si₃N₄⁴⁹, and Si⁴⁸ by over 100 times. All parameters for the Brillouin gain analysis are provided in Supplement 7. Throughout this analysis and subsequent experiments, the pump wavelength was set at 3.3 μm, considering the measurement capability of the available optical spectrum analyzer (OSA, Yokogawa AQ6376, with a wavelength range of 1.5-3.4 μm), despite the observed increase in the Q-factors at longer wavelengths, as depicted in Fig. 3a. The radius of the resonator was set to 4.65 mm to match its free spectral range (FSR) with the calculated Brillouin frequency shift, f_B, of 3.70 GHz.

Fig. 5: Mid-IR stimulated Brillouin lasing (SBL).

a Intensity profile of the acoustic and optical eigenmodes responsible for the stimulated Brillouin scattering (SBS) signal generated at 3.3 μm wavelength. Only the left and right halves of each mode are shown, as both modes are bilaterally symmetric. b Optical power spectrum showing SBL operation up to the 8^th order, measured at the transmission port using an optical spectrum analyzer. c Measured and calculated power dynamics curve of the 2^nd and 4^th Stokes wave (S2, S4). The x and y axes of the calculated curves (solid lines) were both multiplied by a scaling factor of β = 1.43 to fit the measured values. The 1^st Stokes wave threshold power of 91.9 μW was extracted from the fit. d Clamped power of the S2 plotted as a function of the pump wavelength, measured to calculate the Brillouin gain bandwidth.

The generation of a Brillouin lasing signal in the microcavity was experimentally verified by observing a cascading process extending to the 8^th Stokes wave, as shown in Fig. 5b. This result was achieved with a waveguide-coupled input pump power of around 14 mW. Throughout the measurement, the forward-traveling signal, composed of even-order Stokes waves, was captured by an OSA at the transmission port of the coupled waveguide since optical circulators were unavailable within this wavelength range. The Q_int of the pumped mode was 2.19 × 10⁷. The f_B, extracted from the optical power spectrum, was 3.66 GHz, which corresponds closely to the calculated value within the OSA resolution limit of 0.003 nm, or equivalently, 82.8 MHz.

The lasing process exhibited a remarkably low threshold, as determined by analyzing the power dynamics involved. In Fig. 5c, the measured power of the second (S2) and fourth (S4) Stokes waves is represented by red and blue dots, respectively, plotted against the input pump power. Power dynamic curves analytically derived by cascade-order Brillouin laser theory⁵⁰, represented by solid lines in the figure, show an excellent match with the measured values. For the best fitting, a scaling factor (β) of 1.43 was multiplied by both the input and output power of the theoretical curves, compensating for the measurement error in waveguide-to-fiber coupling loss, as explained in the Methods. The lasing thresholds, obvious in Fig. 5c, were identified as 0.367 mW for S2 and 1.65 mW for S4. Importantly, the threshold for S1 was calculated to be 91.9 μW from this analysis. This value is comparable to the lowest threshold observed in on-chip Brillouin lasers in the near-IR, which was 40 μW, achieved using a SiO₂ resonator with a Q_int of 3 × 10^8 10.

To further characterize the laser properties, the Brillouin phonon gain bandwidth was measured using the pump wavelength detuning method¹⁰. By detuning the pump wavelength from the Brillouin gain peak wavelength, a discrepancy between f_B and cavity FSR increases, which decreases the gain in accordance with its spectral profile. In Fig. 5d, the clamped power of the 2^nd Stokes wave, which is inversely proportional to the gain coefficient, is plotted against the pump wavelength. The full-width-half-maximum (FWHM) of its quadratic fit, depicted by the black line, is 13.9 nm. The phonon gain bandwidth of 14.5 MHz, or equivalently the phonon Q-factor of 249, was obtained by multiplying this FWHM by the frequency mismatch per pump wavelength detuning (Hz/nm), calculated considering the frequency pulling effect^10,51. Details of this calculation are provided in Supplement 6.

The improvement in spectral purity achieved through the Brillouin lasing process was estimated theoretically using established models for Schawlow–Townes (ST) narrowing in Brillouin lasers. Since the measured phonon bandwidth (?=2?×14.5 MHz) is comparable to the photon decay rate (?=∼2?×7.20 MHz) of the cavity, the ST equation without adiabatic approximation (?/??1) was employed to calculate the linewidth⁵¹, Δ?0=(??+?)2???3???4??????????. Using the measured parameters for optical and acoustic damping, as described in the Methods, yielded an ST linewidth for the clamped Stokes wave of 83.5 Hz. Remarkably, this value is more than a thousand times narrower than the linewidth of the pump laser source (Argos SF-15 module C, linewidth <1 MHz) or commercialized quantum cascade lasers (Alpes Lasers, ~1 MHz). In conjunction with quantum cascade lasers, this advance offers significant potential to develop ultra-narrow linewidth lasers on a chip encompassing a broad range of mid-IR spectrum, which will open opportunities to various research that requires frequency precisions unachievable with previous mid-IR lasers, as outlined in Introduction.

The total Q-factor (Q_tot) was predicted by adding its individually calculated loss components, namely SiO₂ absorption (∝1/???????), roughness scattering (∝1/??????), and absorption from surface-adsorbed water (∝1/??????), and taking its inverse. The Q_meas values and the used resonator are the same as those in Fig. 3a and b.

Optical power measurement in the mid-IR SBS experiments

Accurately measuring the optical power dynamics at mid-IR, especially for low threshold power on-chip devices, is difficult due to the lack of sensitive power meters (detectable power >~1 μW) and low loss fiber components. To resolve this issue, we instead measured the waveguide in-coupled power using an ultra-sensitive photodetector (PD, VIGO Photonics PVI-4TE-5) placed at the waveguide transmission port (Fig. 3d). The obtained PD voltage was converted to waveguide in-coupled power by the following steps. First, the PD responsivity was calibrated by comparing the PD voltage with a reference mid-IR power meter (Thorlabs S180C) readout. The PD voltage measured behind a pre-calibrated OD2 ND filter was plotted as a function of the optical power measured in front of the ND filter, giving a responsivity 1.73×104 V/W. Next, the net optical loss from the waveguide output facet to the photodetector, including the waveguide-to-fiber out-coupling loss and loss from free-space optics, was measured. The waveguide in-coupled power was calculated by dividing the PD transmission voltage by the calibrated responsivity and multiplying it with the measured loss factor. The output Stokes wave power was measured by multiplying the OSA power readout with the waveguide-to-OSA loss, obtained in the same manner.

During measurement, the waveguide-to-PD and waveguide-to-OSA losses constantly change due to the drift in waveguide-to-fiber out-coupling efficiency. This causes these two loss factors to deviate from the calibrated values on the same scale. Consequently, the measured input and Stokes power (x and y axes in Fig. 5c), obtained using these two factors, will also deviate from reality on the same scale. This causes the discrepancy between the measured and theoretical Stokes power dynamics in Fig. 5c, represented by the scaling factor β = 1.43.

Estimation of the Schawlow-Towns linewidth

The Schawlow-Towns linewidth for the clamped Stokes wave was calculated using the following two equations^45,51:

where ? is the Plank constant, ω_L is the pump frequency, ???=1693 is the number of thermal phonon quanta calculated from the Bose-Einstein statistics, ??(???) is the total (external) Q-factor of the Brillouin cavity, and ???? is the power of the generated Stokes wave. The power of the clamped Stokes wave is inversely proportional to the Brillouin amplification rate per pump photon d?d??=?????2??/?=1.39 Hz, where v_g is the group velocity and L is the physical cavity length⁴⁵. Inserting the clamped power into the linewidth equation leads to the following simplified form:

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.