Surface Protection and Activation of Mid-IR Plasmonic Waveguides for Spectroscopy of Liquids

748 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 42, NO. 2, JANUARY 15, 2024 Surface Protection and Activation of Mid-IR Plasmonic Waveguides for Spectroscopy of Liquids Mauro David , Ismael C. Doganlar , Daniele Nazzari , Elena Arigliani , Dominik Wacht , Masiar Sistani , Hermann Detz ,GeorgRamer , Bernhard Lendl , Walter M. Weber , Member, IEEE, Gottfried Strasser , and Borislav Hinkov Abstract—Liquid spectroscopy in the mid-infrared spectral range is a very powerful, yet premature technique for selective and sensitive molecule detection. Due to the lack of suitable concepts and materials for versatile miniaturized sensors, it is often still limited to bulky systems and offline analytics. Mid-infrared plasmonics is a promising field of current research for such compact and surface-sensitive structures, enabling new pathways for much-needed photonic integrated sensors. In this work, we focus on extending the concept of Ge/Au-based mid-infrared plasmonic waveguides to enable broadband liquid detection. Through the implementation of high-quality dielectric passivation layers deposited by atomic layer deposition (ALD), we cover the weak and watersoluble Ge native oxide. We show that approximately 10 nm of e.g. Al2O3or ZrO2can already protect the plasmonic waveguides for up to 90 min of direct water exposure. This unlocks integrated sensing schemes for broadband molecule detection based on midinfrared plasmonics. In a proof-of-concept experiment, we further demonstrate that the ZrO2coated waveguides can be activated by surface functionalization, allowing the direct measurement of diethyl ether at a wavelength of 9.38 µm. Index Terms—Germanium, liquid spectroscopy, mid-infrared plasmonics, photonic integrated circuits, surface functionalization, waveguide passivation. Manuscript received 22 January 2023; revised 20 July 2023 and 14 September 2023;accepted25September 2023. Date of publication 29 September 2023; date of current version 16 January 2024. The work of Borislav Hinkov was supported by the Austrian Science Fund FWF under Grant M2485-N34. This work was supported in part by the EU Horizon 2020 Framework Program under Grants 780240 and 828893 and in part by CzechNanoLab Research Infrastructure, MEYS CR under Grant LM2018110. (Corresponding author: Borislav Hinkov; Mauro David.) Mauro David, Ismael C. Doganlar, Daniele Nazzari, Elena Arigliani, Masiar Sistani, Walter M. Weber, Gottfried Strasser, and Borislav Hinkov are with the Institute of Solid State Electronics & Center for Micro-and Nanostructures, TU Wien (Technische Universität Wien Wien), 1040 Vienna, Austria (e-mail: [email protected]; [email protected]; daniele.nazzari@ tuwien.ac.at; [email protected]; masiar[email protected]; walter. [email protected]; [email protected]; borislav.hinkov@ tuwien.ac.at). Dominik Wacht, Georg Ramer, and Bernhard Lendl are with the Institute of Chemical Technologies and Analytics, TU Wien (Technische Universität Wien Wien), 1040 Vienna, Austria (e-mail: [email protected]; [email protected]; [email protected]). Hermann Detz is with the Institute of Solid State Electronics & Center for Microand Nanostructures, TU Wien (Technische Universität Wien Wien), 1040 Vienna, Austria, and also with the CEITEC, Brno University of Technology, 601 90 Brno, Czech Republic (e-mail: [email protected]). This article has supplementary material provided by the authors and color versions of one or more figures available at https://doi.org/10.1109/JLT.2023.3321034. Digital Object Identifier 10.1109/JLT.2023.3321034 I. INTRODUCTION THE mid-infrared (mid-IR) spectral region is highly suitable for a wide variety of applications including optical free-space communication [1],[2],[3] and spectroscopy of molecules in the gas [4],[5], liquid [6],[7],[8] and solid phase [9]. Investigating liquids recently gathered particular interest, sparked by significant progress in the understanding and control of the spectral properties of quantum cascade (QC) devices [1],[10],[11],[12] leading to novel high-performance QC lasers (QCLs) and detectors (QCDs) [13],[14],[15],[16], [17]. They include high-power QCLs, enabling the penetration of thicker sample films on the hundreds of micrometer scale [18] and novel-wavelength and high-performance QCDs [19].This allows monitoring wide concentration ranges characterized by sensor linearity and high saturation thresholds [20]. But even those new and other comparable improved devices still require ratherbulkyexperimentalsetups relying on externalcomponents including flow cells [18],[20],[21] and in some cases also need specific sample preparation techniques [7],[18],[20]. The next generation of mid-IR liquid sensors characterized by compact footprint and robust operation is still in its infancy. It demands for breakthrough integration concepts for realizing miniaturized photonic integrated circuit (PIC) devices. Previous investigations could demonstrate that novel mid-IR plasmonic approaches are suitable for fingertip-sized monolithic PICs [22]. They can be further exploited in terms of lab-on-a-chip sensors allowing highly-sensitive and -selective liquid spectroscopy measurements [8],[22],[23], when merging same-wavelength operating quantum cascade lasers and detectors (QCLDs) with suitable mid-IR plasmonic materials and waveguide geometries [22]. While, similar to regular QCLs, the concept of integrated QCLDs has no fundamental limitations concerning its wavelength coverage throughout the entire mid-infrared spectral range, suitable (broadband) plasmonic materials and concepts in this wavelength range are rather scarce. Therefore, finding alternativepathwaysto make existingplasmonic approaches and material systems suitable for a much wider range of applications is very compelling. Existing octave-spanning and low-loss mid-IR Ge/Au semiconductor-loaded surface plasmon polariton (SLSPP) waveguides [24], are highly suitable candidates for this task. While showing an excellent trade-off between mode confinement and propagation length, their topmost layer, which © 2023 The Authors. This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ DAVID et al.: SURFACE PROTECTION AND ACTIVATION OF MID-IR PLASMONIC WAVEGUIDES FOR SPECTROSCOPY OF LIQUIDS 749 is Ge oxide, easily dissolves in water [25],[26] even at moderate temperatures, deteriorating any possible sensor performance. In this work, we present an innovative approach to overcome the current limitations associated with the exposure of spectral broadband Ge-SLSPP plasmonic waveguides to water. We demonstrate the suitability of “high-k dielectrics”, such as ZrO2and Al2O3, as protective surface coatings for Ge-SLSPP waveguides against direct water exposure. These protective layers, with a thickness of just around 10 nm, are deposited using atomic layer deposition (ALD). Our research aims to explore the impact of these coatings on the plasmonic properties of Ge-SLSPP waveguides. To evaluate the effectiveness of the protective coatings, we employ a combination of finite element method (FEM)-based simulations and optical measurements conducted in a custom-made waveguide characterization setup. Additionally, we subject both, coated and uncoated Ge/Au SLSPP waveguides, to prolonged water exposure, monitoring changes in waveguide losses, and profile degradation. The obtained results demonstrate the remarkable protective capabilities of the high-k dielectric coatings. As an additional proof-of-conceptfeature,we further exploreon-chipliquid sensing by surface activating the ZrO2-coated waveguides through trimethylsilyl functionalization. This surface activation method enables direct measurements of diethylether (and in principle also other hydrophobic compounds [27],[28],[29],[30])onthe waveguide surface at a wavelength of 9.38 μm, using a “dipstick sensing method”, which is otherwise not possible without functionalization. Following the evaporation of diethylether, the plasmonic surface is fully recovered, enabling repeatability of the measurement with the same chip. Furthermore, our work extends beyond the scope of Ge-loaded plasmonic structures and holds broader implications for research on the development of germanium-based devices in general. II. PLASMONIC CONCEPTS IN THE MID-IR SPECTRAL RANGE Surface plasmon polaritons (SPPs) are the collective electron density oscillations, observed at an interface with a sign-change of the dielectric function, e.g. between a metal and a dielectric [31]. They combine the high-speed capabilities of photonic circuits with the ability of miniaturization below the diffraction limit for visible [32],[33],[34] to near-IR wavelengths [33], [34], provided through plasmonic confinement [35],[36],[37]. SPPs can be confined and guided along electrical structures including wires [32],[37] and waveguides [24],[32],[36],[38]. While this results in relatively high guiding losses [32], it still enables implementation in ultra-confined and high-speed SPP and localized SP (LSP) detectors [33],[34],[39]. In contrast, mid-IR plasmonics is still pretty much in its infancy. The traditional noble metal-based structures show poor mode confinement in the mid-IR, with modes extending far beyond the wavelength scale into the dielectric medium [22], [24],[40]. This makes them highly unsuitable for plasmonic on-chip and mode-guiding applications. Alternatives, such as heavily doped epitaxial group IV and III-V semiconductors, transition metal nitrides, transparent conductive oxides, (metal-)silicides or graphene have been realized in recent years [41],[42],[43],[44]. But they still show certain limitations. For example, highly doped epitaxial semiconductors benefit from tailoring the plasma frequency through adjusting their doping level, but they require high-quality epitaxial growth processes and are limited to suitable underlying substrate materials. A relatively simple way to overcome performance limitations in mid-IR plasmonics is to realize so-called dielectric-loaded surface plasmon polariton (DLSPP) waveguides [22]. They enable mode confinement on the wavelength scale and guiding from 10’s to 100’s of micrometer along the chip surface, as e.g. shown for simple SiN/Au structures [8],[22]. When thin dielectric slabs (∼200–300 nm) are applied, the plasmonic mode mostly (>96%) propagates in the surrounding dielectric medium like e.g. air, making this approach highly suitable for, e.g. real-time in-situ sensing applications in liquids, as previously demonstrated for proteins measured around 6.2 μm wavelength [8]. Unfortunately, the covered wavelength range of SiN as dielectric loading material is limited due to absorptions between 7–16 μm[45]. III. SEMICONDUCTOR/METAL STRUCTURES FOR PLASMONICS IN THE MID-IR SPECTRAL RANGE One possibility to extend the operational range of mid-IR DLSPP waveguides is by exchanging their dielectric-loading layer with another, better suitable material. For this, we recently realized a novel concept based on SLSPP waveguides. We use Ge [24] as a highly transparent material in the entire mid-IR spectral range. In addition, it can be easily deposited in a suitable quality by a regular sputtering process and patterned by following state-of-the-art cleanroom fabrication techniques. This approach enables the realization of octave-spanning plasmonic waveguides from 5.6–11.2 μm[24]. Its experimental demonstration is so far only limited by the used excitation laser source and can in principle be further extended to cover the whole range between 2–14 μm. However, Ge is not very resilient against the exposure to liquids.Simultaneously,thereisalackofother more stable materials with similar optical properties as Ge, especially in the spectrally narrower long-wave infrared (LWIR, ∼8–12 μm) [46]. Therefore, as alternative approach, we enhance the robustness of our Ge/Au plasmonic waveguides by applying appropriate ALD surface passivation coatings to extend the Ge-SLSPPs operational range to liquids. The group IV semiconductor Ge already gathered particular attention in the past, as an intermixed materialwithSilicon(Si) in traditionalMOSFETs[47],[48].Itis fully compatible to complementary metal-oxide-semiconductor (CMOS) fabrication processes. Enhanced performance was enabled by the implementation of high-kdielectrics, substituting SiO2as gate material to prevent limitations from shrinking transistor dimensions [25]. In fact, Ge was used to realize the first working transistor and was also among the first materials being intensively studied for surface states and their impact on the electrical properties of a semiconductor [49],[50],[51].This detailed knowledge is another important benefit of using Ge as plasmonic-loading material in Ge/metal SLSPP structures. 750 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 42, NO. 2, JANUARY 15, 2024 Fig. 1. (a) Sketch of the sensing concept. The figure shows the plasmonic waveguide submerged in deionized (DI) H2O together with an analyte, and measured with an externally coupled QCL source and MCT detector. The individual layers including the one for surface passivation are indicated. (b) Cross-sectional profile of the whole waveguide layer sequence including the typical device dimensions. However, Ge forms a native germanium oxide with low electrical and chemical stability [25],[26], together with a poor interface towards the Germanium underneath. This prevented the widespread use of Ge in integrated (optical) circuits for a long time [52]. The oxide is dominated by the presence of GeO2[53] which is removable in water. Regrettably, it also etches bulk Germanium through a recursive process: first the dominant GeO2is removed from the surface, followed by surface re-oxidation and another oxide removal in the water. The intermediate surface oxidation step includes the oxidation of water-stable and therefore previously remaining suboxides of GeOxwith x <2toGeO 2[53]. We will show, that this mechanism is also removing our Ge-SLSPP waveguides and limits their unprotected application in aqueous liquids. However, there is a possible solution to this problem, coming from transistor devices: for being able to use Ge for improving state-of-the-art MOSFETs, different dielectrics have been extensively studied and implemented to cover or substitute the unsuitable native surface oxide(s) of Ge. While first focused attention was given to HfO2[25],[54] due to its highly suitable characteristics and maturity level [55], other promising dielectricswereinvestigatedas well. Theyincludee.g. Al2O3[25], [26],TiO 2[25] and ZrO2[25],[56] which were all evaluated in our study. Fig. 1(a) shows a sketch of the layer structure of the SLSPP waveguides including surface protection layer (”Passivation layer” in the sketch), inside an external measurement setup. It includes a mid-IR QCL to excite the plasmon in the waveguide and a Mercury-Cadmium-Telluride (MCT) detector for measuring the signal. Fig. 1(b) displays the corresponding cross-section profile of the waveguide including typical device dimensions. Fig. 2. Process flow of the Ge/Au plasmonic waveguide fabrication including thedepositionoftheALD-based passivation layersina post-processingstep.The pre-, post-ALD passivation and post-water-submersion characterization (1–3) of the samples is indicated. The corresponding custom made waveguide setup is shown in Fig. 3. IV. PLASMONIC WAVEGUIDE FABRICATION The Ge/Au SLSPP plasmonic waveguides were fabricated according to the flow chart in Fig. 2on a double-side polished 275 μm thick Si (100) substrate. First, a 100 nm thick Au layer is sputtered on the substrate (7 cycles, P base ≤1e−5mbar), together with a 10 nm Ti sticking-layer between metal and semiconductor. After lithography and development, the 300 nm thick Ge surface-loading layer is sputtered on top of the gold (12 cycles, Pbase ≤1e−5mbar). In contrast to previous work, where the needed Ge-slab was patterned by reactive ion etching (RIE) [24], we defined the 9 μm wide waveguides in this work through a lift-off process. This results in waveguides with slightly structurally degraded sidewalls and therefore increased line edge roughness, as compared to previous similar structures using a RIE etching process for Ge patterning [24] (see e.g. a SEM picture of a typical fabricated Ge ridge in Fig. S2(a) of the supplemental material). Some fabrication defects from the lift-off process can be identified, but they show negligible impact on the losses of the waveguides. The overall ridge smoothness and sidewall quality of the devices are still fairly good, which enables conducting the coating study as intended. This will also be confirmed by the following characterization of the typical waveguide losses, which are on the same level as in our previous work. The devices were cleaved to various waveguide lengths between 1 mm and 2 mm and characterized in the previously mentioned waveguide setup (see. Fig. 3,Characterization 1) to obtain their average waveguide and coupling losses. These reference values are needed for the quantitative analysis of the following steps including the ALD deposition and water submersion experiment. After this initial characterization, individual chips were each coated with nominally 10 nm of one of the four different protective ALD-materials Al2O3(10.2 nm), TiO2(9.8 nm), HfO2 (8.5 nm) and ZrO2(10.1 nm). While aiming for a reasonable layer thickness of about 10 nm to combine proper encapsulation with limited additional losses, especially the HfO2turned out thinner than expected, i.e. ∼8.5 nm. After these coatings were DAVID et al.: SURFACE PROTECTION AND ACTIVATION OF MID-IR PLASMONIC WAVEGUIDES FOR SPECTROSCOPY OF LIQUIDS 751 Fig. 3. Sketch of the custom made waveguide characterization setup using a9.38μm emitting singlemode DFB QCL. The optically chopped CW laser beam is focused onto the front facet of the Ge/Au-based plasmonic waveguide with a suitable lens (effective focal length: 1.873 mm). After propagating along the waveguide surface it is out-coupled at the back facet and collimated with another identical lens. The waveguide is positioned using piezo-actuators with nanometer precision. The beam is then separated with a ZnSe beam splitter, where one beam is analyzed on a mid-IR camera (Tau 2, Teledyne FLIR LLC, USA) for its 2D profile while the other is focused on a thermoelectrically cooled MCT detector. A lock-in amplifier was used as it strongly improved the signalto-noise-ratio. deposited, the waveguides were characterized again (Characterization 2) in the waveguide setup. Finally, the submersion experiment was conducted, followed by another sample characterization (Characterization 3). V. PLASMONIC WAVEGUIDE CHARACTERIZATION A sketch of the custom-made waveguide characterization setupisshowninFig.3. Wecharacterizedthewaveguidesusinga continuous wave (CW) singlemode distributed feedback (DFB) QCL emitting at a wavelength of 9.38 μm (mirSense, France). The laser was operated at an optical output power of 1.2 mW in CW mode. The laser beam is collimated and coupled into the plasmonic waveguide using a focusing (in-coupling) and a collimating (out-coupling) lens optics, after passing through a beam chopper. For nanometer precision alignment, the waveguides are positioned on a piezo-actuator stage and follow a dedicated alignment routine, as described previously [24].The out-coupled beam from the waveguides is then either directed to a mid-IR camera (Tau 2, Teledyne FLIR LLC, USA) for inspecting the beam profile, or directly focused onto a thermoelectrically cooled MCT detector (PVI-4TE-10.6, Vigo Systems S.A., Poland) by a parabolic mirror. The measured signal is filtered for optimized signal-to-noise ratio using a lock-in amplifier. In order to obtain the individual waveguide losses preand post-ALD-coating as well as after the submersion experiment in water, we first characterized the typical coupling losses of our SLSPP waveguides. As described in previous works [24], [57],[58], the effective cut-back method can be used to extract the waveguides losses. The experimental procedure consisted of acquiringreference voltagemeasurementswhilesimultaneously ensuring a Gaussian beam profile on the IR camera. The beam position with respect to the lenses was priorly determined by utilizing specially designed shutters together with the use of a digital microscope. The waveguides were aligned to the focal point, and intensity measurements were taken multiple times to obtain precise values. The average intensity was calculated for each waveguide and length. The measurements were performed on waveguides of different lengths assuming identical coupling and attenuation losses. Only waveguides with good facets and similar Ge-layer parameters were measured to ensure consistency. The measured data was then compared with modelling results: the insertion losses (in dB) were plotted against the length of the corresponding waveguide in millimeters (see Fig. 7(a)). A linear model, y=mx +b, was fitted to the data using a least squares approach. The slope mof the linear model provides the attenuation coefficient of the waveguide in dB/mm, while the y-intercept brepresents the total coupling losses in dB for both facets combined. In end-fire coupling schemes, losses may occur if the input light mode does not perfectly match the supported mode size of the receiving optical component. In our case, the relatively large plasmonic mode area compared to the small focal spot size of the lens (1/e2beam profile radius of ∼4μm) result in a smaller spatial overlap and, therefore, significant coupling losses. This value combines the contribution of the overlap factor and the scattering losses at the waveguide facets. After extracting the coupling losses, discrepancies between predicted and measured waveguide values can be investigated by examining the waveguide structure and potential fabrication defects. Surface roughness, inhomogeneities in the waveguide structure, and material stress at the layer interfaces can all contribute to scattering losses that are added to the material (theoretical) losses calculated from simulations (αT heoretical). As will be shown later in the analysis (see Fig. 7(a))ofthe fabricated waveguides prior coating, the coupling losses amount for6.6dBandthescattering/fabricationlossesforapproximately 4.5 dB. The obtained total waveguide losses (material plus scattering/fabrication losses) amount for 9.7 dB/mm. All these values agree well with our previous findings [24]. VI. PASSIVATION OF MID-IR SLSPP WAVEGUIDES Working with semiconductor-based devices, especially when using heterostructures or other nanometer thin layer sequences, requires a high degree of control of the interfaces between the different materials. This is especially true, when using surfacesensitive structures such as SLSSP waveguides. Simultaneously protecting their surface while maintaining the (plasmonic) functionality poses one of the key challenges. Thus, the main requirements concerning coatings for mid-IR plasmonics used in liquid sensing applications are: a) a layer that prevents “physical” penetration from the surrounding liquid, b) low additional optical losses and c) a preservation of the plasmonic characteristics. All three criteria can be satisfied very well by using high-quality dielectric materials known from CMOS-transistors. They are on the nanometer-scale and: a) homogeneously encapsulate the Ge-slab below, b) show (material-dependent) relatively low losses, and c) do not interfere significantly with the underlying plasmonic structure as they are only nanometer-thick dielectrics. The materials we tested are the well-established and previously mentioned Ge-compatible dielectrics HfO2,ZrO 2,Al 2O3and TiO2. Before conducting the experiments, we evaluated the additional waveguide losses of the nanometer-scale passivation 752 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 42, NO. 2, JANUARY 15, 2024 Fig. 4. (a) Simulated propagation length as function of the Ge-slab width of the plasmonic waveguide for three different slab thicknesses at λ=9.38 μm.The crossing point of 3 dB/mm losses (50% losses) and 7 dB/mm losses (80% losses) are indicated for orientation. (b) Illustration of similar simulations showing the variation of the effective mode area with respect to the Ge-slab. (c) FEM-based simulation of the mode cross-section for a typical Ge/metal based waveguide at λ=9.38 μm. Calculated propagation length LPand effective mode area Aeff are given. layers by FEM-simulations. As a starting point, Fig. 4shows the simulations of the bare Ge SLSPPs at 9.38 μm without protective layer. They rely on ellipsometer measurements of the complex mid-IR refractive index of Ge and Au and are given in the supplemental material A, Fig. S1(a) and (b). The plasmonic properties of the different device structures were calculated using the RF module of the commercial software COMSOL Multiphysics, employing the finite element method (FEM). Second-order scattering boundary conditions are utilized for all the simulations to mimic open boundaries. Convergence analyses have been performed to ensure the accuracy of the calculations by adjusting the simulation region until the effective indices show less than 1 % of variation. The propagation length is calculated from the attenuation constant, which is defined as the travelling distance of a mode before its energy density drops to 1/e (approximately 37 %) of its initial value: Lp=1 2·Im(α)(1) The attenuation in dB scale per meter can be calculated from the propagation constant using the formula: α(dB/m)=20·α·log(e)(2) For the effective mode area (Aeff ) we use the following definition [24],[59]: Aeff =1 Max{W(r)}A∞ W(r)dA (3) where the energy density W(r)at position ris defined as: W(r)=1 2·Re d[ω(r)] dω |E(r)|2+1 2μ0|H(r)|2(4) with E(r), the electric field, H(r), the magnetic field, is the permittivity, and μ0is the permeability. In Fig. 4(a), we present the modal propagation length as a function of the Ge-slab width for three different Ge-thicknesses. In agreement with previous observations [24], we find that increasing the Ge width leads to enhanced mode confinement, resulting in higher modal losses from the Au layer below the Ge. It is worth noting that the propagation length exhibits significant variations with changes in the Ge slab thickness. Increasing the Ge thickness from 275 nm (blue curve) to 325 nm (black curve) also reduces the propagation length due to enhanced confinement. These considerations emphasize the importance of minimizing thickness variations to avoid increased modal and coupling losses during sensing with plasmonic devices. Thinner Ge stripes offer longer propagation lengths; however, this increase in propagation length comes at the cost of a larger mode size, which can introduce coupling losses with other optical elements. This relationship is further illustrated in Fig. 4(b), wheretheeffectivemodeareaisplottedagainsttheGeslabwidth for the same range of thicknesses. While the propagation length monotonously decreases with increase in width and thickness (see Fig. 4(a)), a distinct minimum in the mode size is observed around the operating wavelength, suggesting an optimal device width for a specific device thickness. Therefore, in the context of sensing around 9 μm, achieving a suitable balance between low losses (ensuring reasonable propagation length) and optimal modal overlap with the surrounding probed medium, can be accomplished by utilizing 9 μm wide and 300 nm thick Ge slabs. This choice is particularly significant when considering the characteristics of the lenses employed in our setup, such as the C037TME-F lenses with a 1/e2mode diameter of approximately 8 μm. This approach allows us to achieve an effective compromise in terms of performance and compatibility with the overall system. Generally speaking, by carefully designing the parameters of Ge slabs to suit specific applications, it becomes possible to achieve the desired propagation length, and interaction with the surrounding media, while adjusting the mode area to mitigate coupling losses with other optical elements. Finally, Fig. 4(c) shows the mode cross-section for the geometry selected DAVID et al.: SURFACE PROTECTION AND ACTIVATION OF MID-IR PLASMONIC WAVEGUIDES FOR SPECTROSCOPY OF LIQUIDS 753 Fig. 5. Spectrally dependent mid-IR (a) real part and (b) imaginary part of the refractive index of the four different ALD coatings, i.e. Al2O3(green), HfO2(red),TiO2(black) and ZrO2(yellow),obtained frommid-IRellipsometry measurements. (c) Corresponding additional losses to the waveguides, when simulating the impact of applying 10 nm thick layers of the different coatings at 9.38 μm wavelength. in our study, highlighting the significant overlap (>95 %) of the modewiththesurroundingmedium. In this case,thepropagation length remains almost a millimeter. For simulating the impact of the additional protective coatings and due to the lack of literature on the complex mid-IR refractive index of the coating materials [45], we first performed mid-IR ellipsometrymeasurements onunstructured Si/Ti/Au-stacksand with the coatings. The results are shown in Fig. 5(a) and (b). Fig. 5(c) shows the resulting additional losses at 9.38 μm, simulated including the four different passivation layers. They are displayed as relative increase compared to simulating the bare Ge-slabs on Au without those protective coatings. The different passivation layers add between 2% (Al2O3) and 7% (TiO2)to TABLE I THE FOUR ALD COATED SAMPLES ZRO2,AL2O3,TIO2AND HFO2 AFTER DEPOSITION the losses of a typical Ge/Au SLSPP waveguide, compared to without them. These supplemental losses can be divided into additional absorption losses from the respective imaginary part of the refractive index of the coating shown in Fig. 5(b) and into the mode confinement-dependent Au-losses, which can be derived from the real part of the refractive index n of the coating displayed in Fig. 5(a). As explained above, a higher real(n) leads to a stronger mode confinement and, thus, increased Au-losses. Even though the additional losses are not fully negligible, the coatings are still well suitable for plasmonic applications in the mid-IR. Especially, when considering state-of-the-art plasmonic liquid sensors with their very short needed plasmonic sensing sectionson the orderof ∼10–50 μm[8],[60],[61], our presented propagationlengths ontheorderof500μm arehighlysuitable for such applications. High-kmaterials are typically grown or deposited on the substrate by chemical vapor deposition (CVD) or ALD. The latter intrinsically combines monolayer-thickness control with excellent conformality of the deposited films [62] as well as a high layer uniformity [63], making it the deposition technique of our choice. The four protective coatings HfO2,ZrO 2,Al 2O3 and TiO2in our study were deposited using a commercial ALD reactor (Savannah, Cambridge NanoTech Inc, USA). The samples were inserted into the reactor and thermalized for 10 minutes before starting the process. A silicon substrate was added alongside the waveguide samples and used as a reference for the precise measurement of the deposited layer thickness. A detailed description of the ALD process is given in the supplemental material Section B. Aninitialsurfacecharacterizationofthefabricated(un)coated SLSPPs yields smooth Ge-surfaces with moderate sidewall roughness.Indeed, the roughnessshowsa certain variation along the waveguides and a first analysis indicates that it is somewhat more pronounced for the Al2O3and ZrO2coated waveguides. More details are given in the supplemental material section C and in Fig. S2 therein. VII. SUBMERSION EXPERIMENT IN H2O For analyzing the degree of protection provided by the additional ALD coatings, we performed a submersion-experiment in pureDI-H2O.SincetheGe-slabsofthewaveguidesarerelatively thick (∼300 nm) the dissolution process is expected to take place on the minutes to hour(s) timescale [64],[65]. In total we tested 6 samples: 2 uncoated reference samples with bare Ge waveguides only, and the 4 additional ALD-coated waveguide samples. Details on all samples can be found in Table I. The submersion experiment was performed in 2 steps. First the samples were submerged into DI-H2O for 30 minutes and 754 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 42, NO. 2, JANUARY 15, 2024 Fig. 6. Profilometer measurements for the water submersion experiments (pre: blue, +30 minutes: red, +60 minutes: yellow) of: (a)+(b) uncoated and (c) ZrO2 (d) Al2O3(e) TiO2and (f) HfO2coated waveguides. The uncoated Ge gets clearly dissolved in water, while TiO2and HfO2only encounter less pronounced surface degradation. In contrast, the ZrO2and Al2O3coated SLSPPs do not show any substantial degradation effects from the water exposure. then characterized by optical microscope and by surface profilometry. This was followed by a second submersion for additional 60 minutes (i.e. in total 90 minutes) and the waveguides were analyzed again. Fig. 6compares the surface profiles before (blue), after 30 minutes (red) and after 90 minutes (yellow) of submersion. All samples have an initial thickness of about 300 nm. But already after 30 minutes of water exposure, the uncoated samples show a significant reduction in the Ge layer thickness by about 30 nm, i.e. ∼10% of the initial thickness, while in contrast, all coated samples maintain a constant thickness. After additional 60 minutes, the uncoated samples display another, much stronger, removal of the Ge and sample # 2 (Fig. 6(b)) is even reduced to about half of its initial thickness (∼150 nm). In addition, the profile of the waveguide changes from rectangular waveguides to a round-shaped geometry. The observed differences in etch rates between the two uncoated devices can be attributed to several factors related to surface conditions and material properties. Firstly, variations in surface morphology, as evident from the SEM pictures in Fig. 2(a) of the supplemental information, can create preferential etching sites and influence the reactivity of water molecules. Such irregularities on the surface can influence the accessibility and reactivity of water molecules during the etching process. Thus, etching rates can be enhanced or slowed down, resulting in differences in etch rates between the two devices, as observed in the profile measurements of Fig. 6(a) and (b). Additionally, the presence of surface defects can affect the local concentration of reactants, alter the diffusion pathways of water molecules, and modify the adsorption and desorption processes, also leading to variations in etching rates. Secondly, localized strain within the Ge layers can contribute as well to the observed differences in etch rates. Such strain may arise due to the presence of defects and can influence the bond strength and breakdown of Ge-H bonds during the etching process. Variations in strain can result in different susceptibilities to chemical etching processes, leading to variations in the etch rates between different devices. Finally, different compositions of the native Ge oxide on the top may also contribute to different etching rates. After 90 minutes of submersion, the ZrO2and Al2O3coatings remain unaffected, while TiO2and HfO2show first degradation effects by a reduced layer thickness of about 50 nm. For HfO2 this can be explained by a thinner deposited layer thickness of ∼8.5 nm, while for TiO2, AFM measurements reveal holes in the coating (see Fig. S4 in the supplemental material). Still the coatings give a certain protection, demonstrated by no thickness reduction within the first 30 min and by the effective etch rate that can be calculated after 90 min from Fig. 6: while for the uncoated Ge-slabs the etch-rate is between 1–1.5 nm·min−1, it is 0.5 nm ·min−1and 0.6 nm ·min−1for TiO2and HfO2, respectively. An additional issue arises from the lack of a stable Ge surfaceoxide, which might explain the etching of the sample with the HfO2coating. It is the diffusion of Ge, known in the case of HfO2, into the dielectric coating layer [66], resulting in charge trapping and fixed charges [67],[68]. Such diffusion processes can oxidize and, thus, contribute to slowly removing the HfO2 passivation itself in water. They can be strongly suppressed by implementing a SiH4layer between the Ge and HfO2[66].But also surface nitridation and S-passivation can help to improve the Ge surface quality [25] as well as simple storage of the sample for some time in air. This could be shown by an improved passivation-quality from storing an Al2O3coated sample for 3 months in air (surface recombination velocity Seff,max reduces by factor 3-4) [26]. A detailed top-view microscope analysis of all samples reveals that the submersion shows no visible effects on the waveguide surface or width (see Fig. S2 in the supplemental material section C). This is in good agreement with the profilometer measurements, which also show no reduction in the waveguide width during the submersion experiment, even for the removed Ge surfaces (see Fig. 6). Finally, we conducted measurements to evaluate the impact of water exposure on the plasmonic properties of the waveguides at a wavelength of 9.38 μm. Fig. 7(a) illustrates the total insertion losses in dB plotted against the sample length of the fabricated waveguides prior coating. The final result of the protection study is displayed in Fig. 7. It shows the evolution of the absolute and normalized experimental losses in Fig. 7(b) and (c), respectively, after the coating process (yellow) and after the submersion experiment (black), measured with the setup shown in Fig. 3. The normalized plot is shown for better comparison of the data and is obtained by normalizing each device to the losses of its uncoated waveguides. It is clearly visible that we measure an approximately DAVID et al.: SURFACE PROTECTION AND ACTIVATION OF MID-IR PLASMONIC WAVEGUIDES FOR SPECTROSCOPY OF LIQUIDS 755 Fig. 7. (a) Results of the cut-back technique measurements. (b) The histogram illustrates the absolute losses at different experimental stages: prior to coating (orange bars), after coating (yellow bars), and after submersion (black bars). The variation in initial devices’ absolute losses is attributed to differences in the thickness of the Ge layer. (c) Normalized experimental waveguide losses before (red), after ALD coatings (yellow) and after the water submersion experiment (black). Depositing the coatings adds roughly 10–20% (on the dB scale) of losses, while the water submersion significantly increases the losses of the uncoated waveguides. But also the losses of the HfO2coated waveguides and to a minor extend also of the TiO2coated ones are increased due to too thin or holey coatings for those materials that led to water penetration. doubled increase in losses when adding the coatings (from +10% for ZrO2to +19% for HfO2) as compared to the expected simulation results displayed in Fig. 5. After 90 min of submersion, we observe that the uncoated samples experience the expected increase in their losses (+72% and +98%) due to the significant decrease in layer thickness. In agreement with the previously measured profile change, this effect is even more pronounced for the sample “uncoated2”. For the ALD-coated samples we demonstrate excellent protective capabilities for ZrO2 and Al2O3with no distinct additional waveguide losses (<1%), while TiO2(+17%) and even more pronounced HfO2(+57%) show increasing losses. This agrees very well with the previous findings in the reduction of total waveguide thickness after water exposure. Based on this data, we observe a significant increase in loss in HfO2waveguides compared to TiO2waveguides, despite similar device degradation (as shown in Fig. 6). This discrepancymay beattributedto amore pronounced degradation at the waveguide facets for the HfO2-coated devices, leading to higher coupling losses, which would explain the observed disparity in additional losses. Additionally, it is important to stress that the device degradation shown in Fig. 6reveals that different devices exhibit slight variations in the nominal thickness of the Ge-layer (300 nm), which directly correlates with the differences observed in the absolute losses prior to the coatings. This observation is clearly depicted in Fig. 4(a), where even slight variations in the Ge-layer thickness lead to significant changes in waveguide attenuation. Such behavior is a characteristic feature of plasmonic waveguides, known for their high sensitivity to the properties of the surrounding medium and makes them particularly valuable for applications requiring highly sensitive sensing capabilities. Here, it is crucial to highlight that the device degradation depicted in Fig. 6serves as a representative value and does not provideacomprehensiveanalysisoftheentirewaveguidelength. Therefore, the measurements obtained from the profilometer should be considered rather indicative than definitive. We can conclude that our approach of covering a Ge-layer by a 10 nm ALD-coating results in a very good protection from Ge-removal over time, when selecting the proper materials (ZrO2and Al2O3) and thicknesses (10 nm coating thickness and above). We believe that by increasing the thickness of the two so far less successful coatings TiO2and HfO2, we will be able to achieve a similar level of protection. This can become relevant for certain spectroscopic experiments, where exposing the Ge-based SLSPP waveguides to the analyte might specifically benefit from using TiO2as a coating [69]. Its additional propertiesasarobustandbio-compatiblecoatingmaterial[70]andseed for activated surfaces [71], can in such a case counter-balance the expected higher overall losses from a thicker coating. VIII. SURFACE ACTIVATION OF ZRO2COATED SLSPP WAVEGUIDES Besides simple protective capabilities from additional surface coatings for SPP waveguides, the activation and functionalization of such surface-sensitive interfaces is a powerful tool to realize next generation liquid sensors, where chip-scale footprint as well as selective and sensitive detection features are needed [72]. The application of surface-modified waveguides in mid-IR sensing has already been shown in literature, where different coating materials were used ranging from mesoporous materialstometal organicframeworks.Thesurfacemodification is used to increase the sensitivity by enriching the analyte in the 756 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 42, NO. 2, JANUARY 15, 2024 Fig. 8. Procedure of the chemical treatment to activate the chip surface of ZrO2coated plasmonic surfaces. Details are given in the main text. surface layer, while repelling unwanted substances like water and keeping them away from the evanescent field [28],[73], [74],[75],[76]. In addition, these coatings can also be applied in catalysis [77],[78]. In this respect, Zirconia, very similar to Titania and Alumina, is a very interesting material, because of its bio-compatibility [70],[79],[80],[81] and ability to serve as a host layer for surface functionalization strategies, e.g. based on mesoporous coatings [30]. Such membranes have been shown to improve the sensitivity of state-of-the-art ATR-sensors by factors of more than 160, which would effectively push recently demonstrated monolithic SPP-based liquid sensors [8],[22] from their current limit of detection on the ppmto the ppb-scale. In addition, ZrO2shows a high chemical and mechanical stability, which is a favorable property when applying such coatings in sensing schemes [82]. In contrast to merely using the implemented passivation coatings as a simple protection layer in our waveguides, we want to show another possible approach for activating the plasmonic surface. The idea is to keep unwanted water away from the evanescent field of the light traveling through the plasmonic waveguide, while simultaneously increasing the selectivity of our surface layer towards certain molecules. As the pristine ZrO2surface is usually hydrophilic, we added a trimethylsilyl functionality to the surface to introduce a higher affinity towards hydrophobic compounds, similar to the shown mesoporous ZrO2layers on Si-ATR crystals [30].Inafirsttest,the surface of one of the SLSPP waveguide samples, which was covered with 12 nm of ZrO2, was functionalized as described in literature [30]. Unfortunately, this waveguide experienced damages to its surface and the procedure had to be repeated without the use of the ultrasonic bath. Instead, prolonged periods of submerging the waveguides in the solvents were used, revealing no damages to the processed surface by microscope inspection. Consequently, the procedure displayed in Fig. 8was usedtofunctionalizethesurfaceoftheSLSPPwaveguides,using the reagents and materials described in D of the supplemental material: The sample was submerged in acetone, ethanol and Fig. 9. (a) Contact angle measurements, using a water droplet on the nonactivated (top, 3 μL, hydrophilic: ϑ<90◦) and activated (bottom, 7 μL, hydrophobic: ϑ>90◦) SLSPP waveguide surface. (b) Results of the surface activation measurements on the activated ZrO2plasmonic surfaces for: no functionalization(Sample 1)andfunctionalization (Sample2). Whilesubmergingthe not functionalized waveguides into diethyl ether results in no additional losses for the plasmonic mode, the functionalized surface shows increased losses of about 14.4 % because of the molecules sticking to the surface which absorb the 9.38 μm probe light. deionized water for 15 min each. After purging the waveguides with dry air, the sample was put into a drying oven at 110 ◦C overnight. Then, the sample was placed in a three-neck round bottom flask with a reflux condenser and bubbler, a N2-inlet and a vacuum connection. It was dried for 2 h at approximately 15 mbar and 150 ◦C. After that, the flask was purged with N2and 20 mL of CHCl3and 400 μL of chlorotrimethylsilane at 22 ◦C were added. This mixture was kept in an inert atmosphere for 24 h. Then, the solution was removed, the waveguide sample thoroughly washed with CHCl 3and subsequently submerged in acetone, ethanol and deionized water for 15 min each. It was finally purged with dry air and then placed overnight in a drying oven at 90 ◦C. The success of the surface modification was determined by comparing the contact angles of a water drop on waveguides with and without surface functionalization. Details on the contact angle measurement routine are given in section D.2 of the supplemental material.