III-V on CaF2: a possible waveguiding platform for mid-IR photonic devices

We developed a technique that enables to replace a metallic waveguide cladding with a low-index (n $\sim$ 1.4) material - CaF2 or BaF2 - that in addition is transparent from the mid-IR up to the visible range: elevated confinement is preserved while introducing an optical entryway through the substrate. Replacing the metallic backplane also allows double-side patterning of the active region. Using this approach, we demonstrate strong light-matter coupling between an intersubband transition (lambda $\sim$ 10 microns) and a dispersive resonator, at 300 K and at 78 K. Finally, we evaluate the potential of this approach as a platform for waveguiding in the mid-IR spectral range, with numerical simulations that reveal losses in the 1-10 cm$^{-1}$ range.


Introduction
The mid-IR is currently a spectral range of great technological interest. On one hand, in the field of optoelectronics, the control of material deposition at the atomic scale has revolutionized industrial applications, leading for instance to quantum cascade lasers (QCLs, from the mid-IR up to the THz spectral region) and quantum-well infrared photodetectors (QWIPs); on the other hand, by engineering plasmonic nanostructures, the ability to confine light down to the micro-and nano-scales has further boosted technological applications. To cite a few, enhanced optoelectronics [1] and energy-transfer [2], biochemical spectroscopy [3][4][5][6], electrochemical and photochemical catalysis [7], photovoltaics [8], and more. Novel materials have at last differentiated the possible platforms, allowing great manufacturing flexibility and possibilities [6,9,10].
In this context, and especially for actively controlled optoelectronic devices, such as detectors and emitters, confining and guiding light in high-index semiconductor layers is a basic, yet crucial functionality. When moderate confinements are required, slightly lowerindex dielectric claddings are employed, in the near-IR and in the mid-IR spectral range as well [11], and very low losses are in general achieved. This is especially useful for passive devices for guidonics applications, for instance [12]. When instead elevated confinements are necessary, metallic claddings yield unsurpassed confinement factors, but higher losses must be factored in. However, they are low enough to be tolerated for longer mid-IR [13]and THz wavelengths [14]. This confinement strategy, that is typically based on Au-Au thermocompression waferbonding, is amply used for active devices such THz lasers [15][16][17] and detectors [17][18][19], but also for mid-IR devices operating in the strong light-matter coupling regime [20][21][22].
Nevertheless, the choice of the cladding material does affect the device functionality well beyond the level of propagation losses introduced. For instance, metal-insulator-metal waveguides [23] because of the metallic back-planetotally prevent optical access of the device active core from the substrate, a functionality that would be extremely welcome for pump-probe experiments, especially in the emerging field of polaritonics [24]. Dielectric claddings present similar constraints, since they must be pseudomorphically grown on the active region and their energy gaps are typically larger than the active region one.
Furthermore, the continuous metallic back-plane in traditional metal-insulator-metal architectures, acting as both bonding and cladding layer, constrains all patterning of photonic / electric nanostructures to take place on one side. This de facto blocks one of the two optical accesses to the active region (see Fig. 1(a)). It would be advantageous to employ as cladding material the one that best fits a specific application, independently of its chemical, epitaxial or thermal expansion compatibility with the active device core.
In this letter, we demonstrate an alternative approach to confine light inside high-refractiveindex semiconductors as a possible platform for low-loss mid-IR waveguides. This approach is of interest for photonic devices in the mid-IR, a spectral range important for gas detection, spectroscopy, free space communications. It is based on the direct transfer of III-V material on CaF2 (or equivalently BaF2), and the permanent bonding to the new host substrate. We have chosen CaF2 because it has an extremely low index of refraction in the mid-IR (n 1.3), thus enabling elevated optical confinements, but also because it is transparent in the whole visible range.
On one hand, this latter peculiarity (the CaF2 replaces the metallic back plane) enables pump-probe applications with pump wavelengths well into the visible/UV range, something that is of interest for the study of fundamental phenomena in intersubband polaritonics. In fact, a transparent substrate is a formidable asset to study the temporal construction of polaritonic states [24] and possibly observing dynamic Casimir radiation [25]. For this very reason we present here an experimental demonstration of the technology on grating-based mid-IR polaritonic devices [21].
On the other hand, the same architecture potentially enables other interesting families of devices. The optical access from both sides transforms it into a two-port system [26], with possible applications to devices exploiting coherent perfect absorption [27]. Also the possibility of patterning both sides of the active region, as well as further sequential lithography/bonding steps, may simplify the implementation of multi-layer metasurfaces (chiral structures for instance), or devices with microstructured front and back contacts (detectors, lasers). A last important feature is the possibility to reliably operate at cryogenic temperatures.

Methodological approach and fabrication
Today intense research activity is devoted to the development of low-loss waveguiding solutions in the infrared spectral range, mainly driven by the recentyet extremely rapiddevelopment of silicon photonics. In the 3-5 m range, for instance, much effort has been put into realizing on-chip mid-IR biochemical sensors [28,29], mostly exploiting silicon-oninsulator platforms [30,31]. In parallel, germanium-on-insulator has been developed as an alternative, due to its broader transparency range, reaching the mid-IR up to 15 m: Germanium-on-SiO2 [12,32,33] and Germanium-on-CaF2 [34] for example. In most cases, materials are processed to achieve an intimate, direct adhesion between the active semiconductor region and the host substrate, mimicking the ideal case which would come from a direct epitaxial growth. This approach is interesting but limits applications to those materials that can be reliably wafer-bonded together, and hardly would allow a double-sided patterning of the thin active regiona possibility that can open interesting routes for electrically controlled active devices.
In the case we tackle in this work, III-V semiconductor on CaF2 , traditional direct atomiclayer bonding is not straightforward because of the extremely different thermal expansion coefficients: 5.73x10 -6 °C -1 for GaAs, against 18.85x10 -6 °C -1 for CaF2. Plasma-activated wafer bonding also appears difficult, due to the different nature of the Ca-F2 (ionic) and III-V can reliably bond to CaF2 [35] and guarantee permanent adhesion, also when strong shear forces arise from cooling / heating the sample. Furthermore, the solution we look for must be flexible enough to be independent of the materials used, allowing us to bond also surfaces that have been previously patterned with optical nanostructures elements (i.e. for instance plasmonic elements or gratings). A non-flat patterned surface such as in this case can very hardly be pressure-bonded without affecting the quality and the geometry of the realized structures. We thus developed solutions (ii) and (iii), presented hereafter.
Optical grade CaF2 substrates (thickness: 500 nm, surface quality S/D=40/20) were prepared following a standard cleaning procedure (sonication in warm acetone and rinsing in isopropanol, followed by 5' ashing in O2 plasma).
The direct transfer procedure is described in Fig. 1(b). We proceeded with the following steps: (i) patterning of the coupler grating on the active region (standard electron-beam lithography, Ti/Au evaporation and liftoffparameters in figure captions); (ii) spin coating of a protective PMMA A4 electron beam resist layer on the patterned active region (providing a clean interface at the end of the process) and fixing the sample to a temporary support substrate with a thermo-labile mounting adhesive (crystalbond 509); (iii) GaAs growthsubstrate removal by mechanical polishing followed by chemical etch to AlGaAs stop layer; (iv) removal of stop layer in HF (49% conc.); (v) release of the Active Region (AR) in acetone, which dissolves PMMA and mounting adhesive. To obtain a clean and defect-less interface, minimizing destructive turbulence from weakly mixable solvents, we gradually substituted acetone with isopropyl alcohol and the latter with milli-Q water. The active region is at last allowed to rest on the substrate. After lifting it out of the liquid environment and allowing for evaporation, the patterned AR is inherently brought in close contact with the CaF2 substrate by water's strong surface tension. Van der Waals interactions then keep it bonded to the new host substrate. The advantage of this approach is a clean AR/CaF2 interface. In contrast, further postprocessing cannot be reliably performed, and the samples cannot be cooled down to cryogenic temperatures due to the weak interaction forces. We thus developed a more robust bonding strategy that is compatible with heating up to 180 °C (allowing further processing) and cooling down to liquid nitrogen temperature (-196 °C). The (acceptable) price to pay is the use of a thin polymeric bonding agent, whose robust thermo-mechanical properties however permit further cleanroom processing steps after transfer (optical or electron-beam lithography for instance), as well as device operation at cryogenic temperatures. Furthermore, its chemical formulation, based on SH-and OH-groups, makes it suitable for strong chemical bonding also with fluorides.
The intermediate bonding layer procedure is described in Fig. 1(c). The bonding polymer we chose is a commercially available off-stoichiometry thiol-ene agent used for microfluidics and micro-electromechanical systems (MEMS) (Ostemer 322 Crystal Clear, Mercene Labs, Stockholm, Sweden). To demonstrate post-processing compatibility, and double-side access to the active region, we tested two different layouts: grating patterning (via e-beam lithography and metallization) performed before (case 1), or after (case 2) bonding. In the former case, the grating is buried at the AR/CaF2 interface; in the latter case, it lies at the AR/air top surface (after substrate removal). The detailed steps for case 1 are described in Fig. 1(c): (i) patterning of the coupler grating on the AR; (ii) 2-step bonding to CaF2 (UV irradiation + soft bake hardening); (iii) substrate removal through mechanical polishing and chemical etch. Lithography and metal deposition, or any alternative processing step, can be performed at this point, in alternative or in addition to step (i).

GaAs quantum-wells and strong coupling
We have then explored if the strong light-matter coupling regime can be attained by  The polaritonic splitting is slightly larger in the buried grating sample: this is correctly reproduced in the numerical simulations (Figs. 3(b) and 3(d)). The origin must be sought in the electromagnetic overlap factor, which is larger in the buried grating sample than in the top grating one (see FDTD calculations in Fig. 2), as the field can only leak trough the grating openings, while being confined above from air. It confirms therefore that burying the grating at the III-V/CaF2 interface is advantageous. It also permits to more uniformly excite the AR with a near-IR laser beam for perspective pump-probe experiments [24].
Numerical simulations are in very good agreement with the experiments (optical parameters n, k of ostemer are obtained from the manufacturer or deduced from absorption measurements, as detailed in the following paragraph). Furthermore, the value of the CaF2 permittivity can be fine-tuned thanks to the position of the "CaF2 light line" that crosses the polaritonic branches.

Cryogenic performance, and device-oriented perspectives
A crucial feature is the robustness of the bonded samples against operation at cryogenic temperatures. The choice of the bonding agent impacts considerably on the final use that can be envisioned for the presented platform. Brittle or glassy bonding agents (i.e. most resist polymers) will not withstand cooling to cryogenic temperatures, eventually tearing the active region. Weakly interacting polymers, such as elastomers (i.e. for instance PDMS), are mechanically more deformable, but they suffer from weak chemical interaction, eventually leading to AR detachment after few heating / cooling cycles. The material we used in this    Fig. 3(a)) with normal incidence, TM polarization, at room (above) and liquid nitrogen (below) temperature; (b) Transmission experiment for the same sample, showing an additional purely photonic peak (hidden in reflection) at 790 cm -1 (98 meV, compare to Fig. 3(a)), besides the expected lower and upper polariton excitation. Absorption is calculated as: A = 1-R-T = 0.24 (UP), 0.28 (UP).
Cooling induces a slight linewidth narrowing and spectral blue-shift of the polaritonic resonances. At last, since one of the desired features of III-V-on-CaF2 bonding is substrate mid-IR transparency, we probed the transmittance (2nd port of the system) at normal incidence in order to assess the level of absorption of the system. Figure 4(b) reveals the two polaritonic states in transmission at 300 K, whose spectral position is in excellent agreement with the reflectivity measurements. Most importantly, it is possible to estimate the system absorption as A = 1 -R -T: 24% and 28% of the light is absorbed respectively by the upper and lower polaritonic resonances. This value is in line with what is achieved with systems with a metallic back-plane when not in critical coupling [36], and it can be improved with judicious sample design.
Having experimentally established that this III-V on CaF2 bonding approach suits applications for polaritonic devices, also at cryogenic temperatures, we analyzed with numerical simulations its suitability in the field of guidonics in the mid-IR spectral range. This is of interest for passive and active devices [12,37,38], but also in principle for mid-IR frequency conversion applications [39][40][41].
We have numerically simulated the propagation of a TM00 mode into an etched GaAs passive waveguide bonded onto a CaF2 substrate, as schematized in Fig. 5(a). (c) reflectivity spectrum from a 1.8 m-thick ostemer layer deposited on a gold-coated substrate and deduced imaginary part of the refractive index; (d) propagation losses for mode TM00 as a function of the wavelength, for different GaAs and ostemer thicknesses. The ridge width is 8 m. Above 1800 cm -1 losses are below 2 dB/cm (0.46 cm -1 ). In the region of ostemer ro-vibrational transitions, it is possible to keep the losses below in the 1-3 cm -1 with a reasonably thick AR.
The modal overlap factor with the AR is always above 90%, thanks to the low index of refraction of both the bonding polymer and CaF2. The propagation losses are set by the field penetration in the ostemer layer, that is plotted in Fig. 5(b) as a function of the wavenumber for two different AR and ostemer thicknesses. The calculation reveals that the AR thickness plays a primary role, while decreasing the ostemer thickness has an almost negligible effect above 1000 cm -1 .  m-thick waveguide the peak losses lie in the 3-5 cm -1 range above 1000 cm -1 wavenumber.
If the ostemer thickness is further reduced to 0.2 m, the situation improves with peak losses in the 1-3 cm -1 range. Note also that deep oscillations are present and losses can reach values below 0.5 cm -1 for specific frequency ranges: above 1800 cm -1 for instance losses become negligible.
These results show that this platform is compatible with the implementation of passive waveguides in mid-IR-on-a-chip applications or frequency conversion above 1800 cm -1 .
Below this value, further improvements would be needed to reach propagation losses below 1 dB/cm. On the other hand, the estimated level of loss is compatible with the monolithic integration of mid-IR active devices [38] as high AR overlap factors (> 90%) can be obtained.

Conclusions
We have developed a technique that permits to replace the metallic cladding with a lowindex material (CaF2 or BaF2), which is transparent from the mid-IR up to the visible range: elevated confinement is preserved, while introducing an optical entryway through the substrate along with active region double side micro-structuring potential. With this approach we have demonstrated GaAs-based optical cavities transferred on CaF2 both directly or with an intermediate bonding polymer. The use of judiciously doped semiconductor QWs led to the observation of strong light-matter coupling between an ISB transition and the resonator photonic mode, at 300 K and at 78 K. We have also shown that placing the coupler grating on the air/active region boundary or burying it on the active region/CaF2 boundary leads to equivalent results. In the field of polaritonics such a two port system [27,36], with a near-IR transparent substrate and flexibility in radiation coupling, can represent a precious asset to study the temporal construction of polaritonic states [24] and possibly Casimir radiation.
Especially for the latter application, the active region illumination through the CaF2 offers a homogeneous excitation of the QWs, hence by-passing problems of charge diffusion that occurs when illuminating from the grating side.

Funding
We acknowledge financial support from the European Research Council (IDEAS-ERC) ("GEM") (306661) and from the European Union's Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 748071.