The optical parametric oscillator (OPO), one of today's most known optical devices, can be functionally defined as a widely tunable coherent source . Like the laser, it is based on the resonant feedback provided by a cavity to an optical amplifier, but it relies on parametric amplification rather than stimulated emission. Unlike the laser, whose heterostructure diode version  has propelled the field of photonics, the quest for an electrically pumped monolithic version of the OPO is still open half a century after its original demonstration . While most of guided-wave OPOs have been demonstrated in LiNbO3 [4-6], its dielectric nature is incompatible with the perspective of full optoelectronic integration, which seems more viable for a direct gap semiconductor with strong nonlinearity.
In this context, a major breakthrough would be the demonstration of an electrically pumped OPO integrated on a chip, which could be used as a coherent variable wavelength emitter in spectral ranges where semiconductor lasers are only poorly tunable. The potential impact of a monolithic OPO is apparent if we compare the wavelength tunability Δλ of an integrated InGaAsP SG-DBR single-mode telecom laser diode (Δλ ≈ 60 nm for a control current change of 90 mA ) and a guided-wave OPO based on periodically poled LiNbO3 (Δλ ≈ 900 nm for a pump tuning of 16 nm )
Let us stress that the perspective of the full convergence at chip-scale level between optics and electronics has been a major driving force for scientific and technological efforts in photonics for the last decades. Of course a large number of applications might benefit from the merging of coherent light sources and the technologically advanced silicon platform . However, the perspective of monolithic silicon photonics is limited by its indirect gap and centrosymmetric crystal structure, with the related lack of electrically injected lasers and second-order nonlinear optical components. This explains the interest drawn by AlGaAs waveguides, which stems from the numerous advantages of this III-V direct-gap semiconductor, well known as an established laser material: high second-order susceptibility (d14 ≈ 100 pm/V), wide transparency window (from 0.9 to 17 μm), good thermal conductivity and mature technology.
Here we demonstrate a near-IR continuous-wave (CW) OPO in a form-birefringent GaAs/AlGaAs waveguide, with a low pump threshold Pth = 210 mW. The development of such a source might radically transform a number of applications in near and mid IR up to 3 µm. Besides two telecom windows of paramount importance (around 1.3 and 1.5 μm), this spectral region contains the roto-vibrational lines of many relevant molecules (pollutants, toxins, etc.), which makes it essential for civilian and military applications such as spectroscopy, material processing, molecular sensing, thermal imaging, and defense.
Because of the key role played by phase matching (PM) in parametric interactions, let us recall that the optical isotropy of GaAs deprives this cubic material of the standard option of birefringent PM, and it has prompted several alternative approaches for fulfilling this condition . To date, the main PM strategies that have been investigated in the near-IR for guided-wave nonlinear interactions in GaAs are quasi-PM , modal PM [11,12] and form-birefringence PM , but none of them has resulted yet in an integrated OPO. In the last years orientation-patterned GaAs in the mid-IR has undergone constant progress with high-power OPOs exhibiting thresholds of ≈2kW in the pulsed regime and ≈7W in the CW regime [14,15]. However, the transposability of this approach to a near-IR integrated platform seems to be hindered by large scattering losses at short wavelengths, and by the lack of GaAs-based high-power pump lasers around 2 µm. Therefore the route to a fully integrated tunable semiconductor source of near-IR light remains open and challenging.
The rest of this paper is organized as follows: Section 2 is devoted to the form-birefringence PM technology in AlGaAs waveguides, including the study of the optical propagation losses; the design, the fabrication and the characterization of integrated mirrors are addressed in Section 3; the experimental demonstration of the OPO is provided in Section 4, along with the discussion; conclusions and perspectives are drawn in Section 5.
FORM BIREFRINGENT PHASE MATCHING IN AlGaAs WAVEGUIDES
In order to artificially induce a significant amount of birefringence between the fundamental TE00 and TM00 modes of a waveguide that is fabricated with an isotropic material, one can pattern its core at sub-wavelength scale, by repeatedly breaking the refractive index continuity with a two-materials multilayer. The resulting metamaterial behaves as a macroscopic uniaxial crystal, whose birefringence is fully determined by the index contrast and the filling factors of the materials . In the AlGaAs platform, the form-birefringence-PM approach has been developed during the late 90's at Thomson CSF Laboratory (today III-V Lab) .
Actually the first phase-matched interaction of this type had been demonstrated back in the seventies, with the doubling of a CO2 laser emitting at 10.6 μm . In that case, given the weak material dispersion in the mid-infrared range (few 10-2) an AlAs/GaAs heterostructure suffices to meet the PM condition. However, since the material dispersion strongly increases when the frequencies of the interacting waves lie close to the band gap of the material, nonlinear interactions between the visible and the mid-infrared are prevented for the simple AlGaAs platform.
In 1990, the discovery of selective wet oxidation of Al-rich AlGaAs layers drastically broadened the potential of form-birefringent PM, thanks to the density, homogeneity and stability of a new type of aluminum oxide . This material exhibits nice optical properties, such as a wide transparency window and a low refractive index of ~1.6, and is electrically insulating. A few engineering domains rapidly took advantage of these physical properties: electronics, with field effect transistors ; optics, with broadband Bragg mirrors ; and optoelectronics, by combining optical and electrical confinement in vertical-cavity surface-emitting lasers (VCSELs) to improve their yield and ensure single-mode emission .
In the case dealt with here, the GaAs/AlOx system allows accessing a birefringence of several 10-1, i.e. up to one order of magnitude higher than in a GaAs/AlAs heterostructure. This in turn enables to phase-match any nonlinear quadratic interaction with wavelengths spanning from the visible to the mid-IR region [22,23].
The structure that we consider here has been designed to operate as an OPO in the near IR, with degeneracy at 1.064 µm. It relies on a guided-wave type-I FBPM scheme between a TM00 pump mode at wavelength λP and two TE00 signal/idler modes at wavelengths λS and λI, respectively. The corresponding vertical layout, depicted in Fig. 1a along with the interacting mode profiles, consists of GaAs (substrate) / 1 µm Al0.92Ga0.08As (spacer) / 1 µm Al0.7Ga0.3As (bottom cladding) / 4 × [37.5 nm Al0.98Ga0.02As / 273 nm GaAs] / 37.5 nm Al0.98Ga0.02As (core) / 1 µm Al0.7Ga0.3As (top cladding) / 30 nm GaAs (protective cap).
This heterostructure has been grown by molecular beam epitaxy on a semi-insulating (001) GaAs substrate. Guiding ridges have been defined along the crystalline axis by means of UV photolithography, then chemically etched with a CH3COOH:HBr:K2Cr2O7 solution, resulting in waveguides with extremely smooth sidewalls, as it can be appreciated from the SEM picture given in Fig. 1b. Typical waveguides are 3 to 4 µm wide, ~ 3 µm deep and cleaved with a length L = 3 mm. Eventually, Al0.98Ga0.02As has been selectively transformed into AlOx by thermal wet oxidation. This process is performed in a tubular quartz oven at a temperature between 400 and 450°C, exposing the waveguide sidewalls to an oxidizing gas stream . In our setup, the latter is obtained by flowing dry N2 (2L/min) through a water-filled bubbler heated at 70°C.
Optical propagation losses play a key role in all guided-wave phase-matching schemes, and their reduction is a critical issue. For this purpose, we optimized the thermal wet oxidation process using the feedback provided by systematic loss measurements. In addition, we investigated the possible loss mechanisms in order to single out specific technological limitations and initiate an adequate technological development.
After careful calibration of the oxidation apparatus, we confirmed that 1) the average loss level suffers from slow oxidation kinetics (i.e. from low temperature, since the reaction is thermally activated), and 2) the process must be stopped at the exact moment when the oxidation fronts merge at the center of the structure (i.e. when all the Al0.98Ga0.02As has been converted into AlOx). In our case, the sample has been oxidized at 420°C for 14 minutes, which is our best trade-off between speed and reproducibility.
Despite this care, the present level of optical losses in such AlGaAs/AlOx waveguides is still significant: α ~ 0.7 to 1.5 cm-1 around λ = 1.55 µm. As a comparison, typical PPLN waveguide losses are in the range of few 0.01 cm-1, enabling the fabrication of several cm-long OPOs with threshold as low as ~ 10 mW . Thus, in order to develop a low-loss AlGaAs/AlOx waveguide technology, a better understanding of the optical loss mechanisms involved in such selectively oxidized AlGaAs devices is crucial.
Most of the propagation losses clearly originate from the oxidation process, for they are about three times lower before oxidation (~ 0.1 to 0.5 cm-1 around λ = 1.55 µm). However, the relevant loss mechanisms and their link with the AlOx properties remain uncertain. In particular, several aspects of the oxidation process are not yet clearly established, including the exact formation reactions and the fine chemical and structural properties of the oxide.
The characterization of oxidized layers via transmission electron microscopy (TEM) , which was carried out to assess the chemical and morphological properties of AlOx at microscopic scale, has established that:
– AlOx layers are composed of γ-Al2O3 polycrystalline grains, sized between 10 and 20 nm, embedded in an amorphous AlxOy matrix,
– The roughness at AlGaAs/AlOx interfaces is increased by about 40% and 80% for GaAs and Al0.7Ga0.3As respectively,
– Residual oxidation of the neighboring GaAs and Al0.7Ga0.3As layers occurs through the interfaces over 3 and 9 nm respectively,
– Further in the GaAs layers surrounding the AlOx, the material becomes amorphous in the vicinity of the oxidized layers (~20 nm from the interface) and remains mono-crystalline beyond.
For high-index-contrast waveguides, optical losses are very sensitive to imperfections. We recall that the device dealt with here is passive, thus the semiconductor alloys have been chosen to be transparent at all the wavelengths involved in the χ(2) process. In addition, in these waveguides operating in CW with internal pump powers Pp ≤ 250 mW, the contribution of two-photon absorption is not significant as αTPA is ≤ 0.2 cm-1. Hence, assuming a good confinement, i.e. no optical leakage in the substrate by design, losses can only arise from scattering by rough interfaces or volume inhomogeneity, or from absorption by point defects.
Based on the above observations, our starting hypothesis was to ascribe the guided modes propagation losses to scattering. However, the index contrast between AlOx and γ-Al2O3 being negligible, polycrystalline grains are not expected to contribute significantly to losses, thus we focused on the roughness of the (Al)GaAs/AlOx boundaries.
Among the different existing techniques to evaluate optical propagation losses, we opted for the Fabry-Perot fringes method , which is well suited for α □≤ 1 cm-1 and only requires the knowledge of the modal reflectivity (estimated by 3D-FDTD simulations). However, for losses of several cm-1, this technique is less reliable and we switched to transmission measurements. Let us recall that, unlike the Fabry-Perot technique, the accuracy of the latter also depends on the estimations of the coupling and collection efficiencies, which are calculated indirectly.
We measured the losses of the TM00 mode (resp. TE00) around 1.064 µm (resp. 2.128 µm). Around the degeneracy of parametric fluorescence, when the wavelength of the TM00 pump mode is tuned from 950 and 1060 nm, the TE00 signal wavelength ranges from 1060 nm to 2120 nm. Accordingly, we used a set of linearly polarized CW lasers, tunable between 1280 – 1610 nm (Tunics telecom external cavity diode lasers), and around 2120 nm (Nanoplus DFB laser diode) for the Fabry-Perot fringe measurements, and between 950 – 1100 nm (Spectra-Physics Ti:Sapphire laser) for the transmission measurements.
The experimental results are given in Fig. 2, where it appears that the error bars for the pump-mode data are significantly bigger that their signal-idler counterparts. As mentioned above, the transmission method is indeed less reliable than the Fabry-Perot one, so that the uncertainty is increased. Compared to the data acquired in the telecom band, also the data at 2.12 µm lack reliability and overestimate the actual loss value. This is due to the limited tunability of the laser diode (Δλ ~ 1 nm) used for this measurement, which does not allow us to refine this estimate based on the Fabry-Perot method.
Two different regimes can be clearly distinguished in Fig. 2:
– Above λ~1.1 µm, losses are of the order of 1 cm-1 and decay as an inverse power law of the wavelength α ∝ λ-N. According to the fits of the upper and lower bounds of the prediction band, we find 1.9 ≤ N ≤ 2.5. The fair agreement between the experimental data and the model prediction confirms that in this spectral region, losses originate from scattering due to the AlOx layers rough interfaces.
– Below λ~1.1 µm, losses are much higher (up to 3 cm-1) and diverge when λ decreases, as highlighted by the phenomenological exponential fit α ∝exp[-λ/141], which clearly deviates from the prediction band.
Very little literature addresses directly the study of optical losses in AlGaAs/AlOx devices beyond 1 µm. In particular, for λ < 1.1 µm, the sole scattering on surface imperfections is not sufficient to explain the propagation loss behavior, and a new loss mechanism, most likely accounting for bulk defects in the vicinity of the oxide layers, had to be considered. While the high level of losses and their exponential decay with wavelength shown in Fig. 2 are reminiscent of Urbach's tail absorption model, the origin of this absorption is still unclear at the moment. Since Al2O3 and GaAs themselves are known to be transparent in the near IR, we tentatively ascribe these additional losses to point defects associated to energy levels in the semiconductor gap and created during the oxidation process in the GaAs layers adjacent to AlOx. Arsenic antisites (AsGa) have already been supposed to introduce donor levels with similar properties in the gap of AlGaAs, in optical devices  as well as in electrical devices . Such defects might arise from trapped oxidation reaction products resulting in an excess of elemental arsenic As0 at the oxide-semiconductor interfaces. Many AsGa would then be generated after the diffusion of As0 into the neighboring semiconductor layers , consistently with the TEM observations recalled above .
Since propagation losses are currently the limiting factor for the demonstration of efficient integrated AlGaAs/AlOx devices, further minimization remains to be tackled. To this purpose four potential solutions are under study:
– Based on existing research on quantum-well luminescence close to oxidized layers, hydrogenation might be tried in order to remove absorbing species trapped at the oxide interfaces ,
– Thermal annealing is also expected to diffuse defects out of the structure, and to restore the crystallinity of the amorphous materials,
– Shifting the coalescence of the counter-propagative oxidation fronts away from the center of the waveguide (i.e. where the fields are maximums) should prevent the formation of additional defects induced by over oxidation .
– The insertion of thin lattice-matched GaInP barriers between the AlOx and GaAs layers seems effective to hamper the diffusion of As-rich oxidation reaction products into the latter, thus preventing the formation of point defects .
We can conclude this section by resuming that, despite the recent results in terms of three-wave mixing efficiencies in AlOx-based waveguides, this material has still considerable margins of amelioration for the most demanding photonics applications, and its fine understanding is a matter at the intersection of device physics and materials science.
THE OPO CAVITY
In view of the parametric oscillation, the choice of the cavity configuration has been made between the two practical optical resonators on which OPOs are mainly based, namely the singly resonant (SR) cavity and the doubly resonant (DR) one. In the first (resp. second) case, the cavity mirrors reflect only one (resp. both) of the signal/idler wavelengths. Since the pump threshold of the latter is typically one order of magnitude smaller than that of the former, we opted for a DR configuration, at the expense of a poorer spectral stability and tunability [1,34]. For a symmetric DR cavity with single pass of the pump, the steady-state pump power oscillation threshold Pth, which is minimum at degeneracy, reads :
where ηnorm is the waveguide normalized conversion efficiency, αP is the pump loss, αS,I is the signal/idler loss, L is the waveguide length, and RS,I is the modal reflectivity at signal/idler wavelength. We remind that for a phase-matched interaction in a lossy waveguide, ηnorm is itself a function of L and αj . Plugging our experimental loss values of in Eq. 1, we find that for the pump threshold to be ≤ 100 mW, the mirrors modal reflectivity at degeneracy should be ≥ 98 % in a waveguide of optimal length L ≈ 2 mm. Therefore, the nominal requirements of our mirrors were: 1) a broad high-reflectivity (HR) stopband centered about λ=2.12 µm with Rmax= 99%; and 2) an anti-reflection (AR) window around λ = 1.06 µm with Rmin = 0.5 %.
To meet these specifications, we opted for Bragg dielectric dichroic mirrors to be deposited onto waveguide facets. Their multilayer structure has been designed using the Essential MacLeod commercial software . The widespread SiO2/TiO2 coating materials have been chosen for the bilayers of both mirrors, owing to their relatively high-index-contrast in the near infrared (Δn ~ 0.8) that enables to keep the thickness of the stack to a minimum. Six such bilayers turned out to be necessary to fulfill requirements.
The fabrication protocol that we have developed is based on Ion beam Assisted Deposition (IAD) that enables the coating at relatively low temperature, hence avoiding to anneal the sample, which could possibly alter the material quality. After oxidation, and prior to cleaving the sample facets, its surface is shielded from dielectric material overspray with a photoresist film (Shipley S1828 baked at 120°C for 5 min) that is eventually stripped.
A SEM picture of a waveguide facet after mirror deposition is shown in Fig. 3 (Left), where the mirror appears in the form of a dielectric “wall”. Besides the morphologic study of the fabricated cavities, we characterized them by measuring the mirrors reflectivity.
In order to validate the multilayer design, a silicon wafer placed in the deposition chamber close to our sample and coated during the same run, served as control sample to check the reflectivity spectrum of one mirror. The related experimental Fourier transform IR (FTIR) spectrum is shown in Fig. 3 (Right), along with modal reflectivities measured at λ = 1.06 µm and λ = 2.12 µm, respectively for the TM00 and TE00 modes. The former value has been inferred from a transmission measurement, resulting in RP = 11 ± 5 %. The latter has been inferred from the study of the Fabry-Perot fringes of a waveguide having one of its facets coated. A reflectivity RSI = 90 ± 5 % has been estimated from the finesse of the Airy function fitting the data.
While the FTIR spectrum meets the theoretical requirements in terms of AR at λP and broad HR stop-band at λS,I around degeneracy, the measured modal reflectivities differ significantly from the target values. Such discrepancy is attributed to the 1D design procedure and the FTIR measurement, which only consider plane waves at normal incidence, while guided-modes also experience diffraction losses at each reflection. The experimental data are in better agreement with FDTD numerical calculations of the TM00 (resp. TE00) mode reflectivity around λ = 1.06 µm (resp. 2.12 µm) accounting for this effect. Let us note that, even if the best agreement is obtained with accurate 3D-FDTD calculations, an acceptable alternative is obtained with the less-time consuming 2D-FDTD calculations (which only accounts for the vertical mode confinement).
EXPERIMENTAL RESULTS AND DISCUSSION
Both SPDC and OPO experiments have been performed with a TE polarized, CW Ti:Sapphire laser, tunable from 950 nm to 1100 nm with a linewidth around 30 GHz. A half-wave plate is used to obtain the TM polarized pump beam that is coupled into the sample with a 60× microscope objective (0.85 NA) mounted on a piezoelectric positioning system. At the output, the pump beam is filtered out with a long-pass interferential filter and the IR beams are collected with a similar objective (a Glan-Taylor polarizer is used to verify the polarization selection rules). The latter are then focused onto an InGaAs photodiode, either directly or after passing through a monochromator. The photocurrent is measured with a low-noise transimpendance amplifier followed by a lock-in.
We started by measuring a parametric gain coefficient below threshold of 4.1 W-1/2 cm-1 at degeneracy, in agreement with our calculations . The huge waveguide dispersion at pump wavelengths results in an extremely large tuning range for signal and idler (>2800 nm for a 70 nm variation in pump wavelength). Fig. 4 shows the SPDC tuning curve in the narrower spectral region of interest here.
Based on our experimental values of single-pass SPDC efficiency (ηnorm = 1500% W-1cm-2), reflectivity (Rs,i = 90%) and losses (αp ≈ 3.0 cm-1 and αs,i ≈ 1.0 cm-1), Eq. 1 predicts a minimum for L=2 mm, given by P = 210 mW. Reaching this threshold has been a challenging task, mainly due to the unconventional mirror deposition onto the facets of our highly confining ridge waveguides. Despite this technological problem, some of our waveguide cavities have allowed us to observe a clear signature of OPO threshold behavior. This is shown in Fig. 5 (Left), where a distinct threshold is apparent for Pth ≈ 210 mW (internal value), in full agreement with the above predictions. In Fig. 5 (Right) we also show the total SPDC power versus pump wavelength measured near degeneracy and below threshold, showing a good agreement with our simulations.
With the measured value of Pth, which corresponds to a mode intensity of 0.8×107W/cm2, we can estimate two-photon absorption losses, obtaining αTPA ≈ 0.19 cm-1, negligible with respect to the linear absorption at pump wavelength.
The maximum pump power Pp ≈ 230 mW is presently limited by an irreversible thermal damage of the mirrors that occurs after about thirty minutes of CW operation. This restricts the ratio Pp/Pth (i.e. the number of times the pump power exceeds the threshold) to 1.09, with of course an impact on the present maximum OPO power Ps,i ≈ 190 nW.
CONCLUSION AND PERSPECTIVES
Beyond the present achievement of the OPO threshold, our future efforts will conveniently be devoted to increase Pp /Pth. To this end, a double-pass-pump scheme could be implemented, leading to a reduction of Pth by a factor between 2 and 4, depending on the relative dephasing of pump, signal, and idler modes at the back mirror of our DROPO.  Such scheme could also be adopted for a singly resonant OPO for which, despite a higher impact of TPA, we estimate a threshold Pth ≈ 0.8 W: a reasonable price for its better stability and tunability performance.
A further optimization of our AlGaAs/AlOx OPO will demand reducing its near-IR linear propagation losses, which are caused by the oxidation of the Al-rich layers. The example of PPLN cavities is instructive because the fabrication of longer waveguides, possible thanks to their extremely low losses (< 0.1 cm-1), enables a drastic reduction of the OPO threshold . In this regard, many aspects of the lateral oxidation of AlGaAs layers are not clear yet, including the exact formation mechanisms and the fine chemical and structural properties of AlOx. Although oxidation is clearly responsible for the deterioration of the multilayer morphology, we have recently reported that the relatively low Rayleigh-like scattering losses due to AlOx/AlGaAs interface roughness only prevail at signal and idler wavelengths, whereas some volume effects must be considered to explain the higher losses at pump wavelengths, which constitute the bottleneck for our present results . A systematic material optimization is currently under way and will hopefully push forward the performance of our AlOx waveguide cavity.
Finally, for future power scaling, we will explore the option of encapsulating the guiding ridge in e.g. benzocyclobutene (BCB), so that the mirrors can be deposited on wider facet areas, with an advantage in terms of uniformity and heat sink.
Besides improving the optically pumped OPO reported here, we also envisage to fabricate its electrically injected counterpart. A concrete design option would be to cascade the laser and OPO sections within one monolithic device, where we would selectively stop the lateral oxidation process before the complete hydrolization of the AlGaAs layers. We estimate that such VCSEL-like engineering of current and light confinement will lead the way to an electrically pumped OPO.
We thank X. Marcadet for MBE growth and the SEAM Labex of Sorbonne Paris Cité for financial support (DOLPHIN project).