1. Introduction

The wide transparency of chalcogenide glasses (ChGs), which can include most of the important range for vibrational spectroscopy in the mid-infrared (MIR) between 400 and 4000cm⁻¹ (25–2.5μm), makes them a very good candidate for chemical or biochemical sensors [1]. In addition ChG planar waveguides can be compact, highly reproducible, and be monolithically integrated with sources [2] and detectors as well as microfluidic systems [3, 4]. As a result, chalcogenide waveguides are being developed as optical sensors for the mid-infrared (MIR), however, to date their capabilities have been demonstrated only in the near infrared (NIR) [5–7].

Obtaining low optical loss in the MIR has been a recurring theme. So far, several methods of fabrication and post-processing of planar waveguides have been described [8, 9]. Perhaps the simplest method is to exploit photo-darkening in ChGs by illuminating them near their band-edge. This provides a way to form channel waveguides in planar films and using this approach waveguides with losses as low as 0.5dB/cm at 8.4µm have been reported [10]. However, photo-darkening leads to low index contrast in the plane of the film and this means that the waveguides cannot be bent tightly which precludes the use of photo-darkening to create micro-ring resonators which have to be employed if high detection sensitivity is to be achieved [11, 12].

Both plasma etching and lift-off techniques have been used to create high index contrast waveguides [13–15]. In the case of lift-off, sidewall roughness becomes a significant challenge because relatively thick films are required for MIR waveguides. The thermal reflow technique has been used to reduce the sidewall roughness and its associated scattering loss in thin waveguide designed for the near infrared [16] although there have been no reports, so far, of its efficacy when applied to much thicker structures for the MIR. In the case of plasma etching, which produces relatively smooth sidewalls, surface contamination becomes a concern since any residue from the photoresist or from C-H or C-F containing polymers deposited to passivate the sidewalls during etching can introduce loss. ChG waveguides have also been fabricated using solution-based methods [17, 18], which have some potential to simplify device fabrication, but in this case it is difficult to remove the organic solvents that absorb strongly in the MIR.

In 2007, Hu, et al. [6] reported the fabrication of the first microfluidic device monolithically integrated with planar Ge₂₃Sb₇S₇₀ chalcogenide channel waveguides. Using this device, N-methylaniline could be detected using the absorption fingerprint of its N-H bond near 1496nm. Following this a microfluidic sensor based on Ge-Sb-Se glass was developed to detect phenylethylamine molecules absorbing at 1550nm [7]. To improve sensitivity, optical resonators integrated with planar microfluidic systems have also been developed but these again operated in the NIR [19]. Reference [20] reviews progress on these on-chip, low loss planar molecular sensors for biological and chemical sensing, and summarizes the manufacturing processes used to realize the chalcogenide resonators. Recently, a high-Q chalcogenide glass-on-silicon resonator and a photonic crystal cavity operating at 5.2µm have been demonstrated [21, 22]. So far, however, none of these ChG waveguide sensors have made use of either the mid-infrared transparency of the chalcogenides or the very strong fundamental molecular vibrational absorption bands that exist in the MIR. Ideally to distinguish between different molecules, the devices should operate across as much of the functional group and fingerprint bands spanning from, at least, 2.5-11µm (4000-900cm⁻¹) but the general lack of waveguides for this spectral range has so far limited demonstrations of chemical or biochemical sensing [23].

In this paper, we focus on high index contrast waveguides designed for the functional group band spanning from 2.5µm to 6.6µm (4000cm⁻¹ to 1500cm⁻¹). Our aim has been to evaluate what losses can be achieved and how they may be affected by contamination that occurs when using conventional lithography and dry etching to define the structure. Some of the anticipated sources of increased loss are as follows. At the short wavelengths (around 3µm), the main challenge will be to eliminate surface contamination produced by the fabrication process or from the environment. For example, any photoresist residue or adsorbed water or hydrocarbons will increase the loss. At longer wavelengths, losses can rise due to hydrogen impurities in the chalcogenide glass itself which produce H-S or H-Se vibrational bands between 4.0 and 4.2µm and 4.45-4.8µm, respectively. At wavelengths around 7.7µm, losses can rise due to fluorocarbon contamination of the waveguide surfaces. This can be present because the deposition of fluoropolymer during etching is used to passivate the sidewalls in order to achieve a vertical etched surface [24]. Adsorbed water can also raise the losses around 6µm, a region where additional absorption due to double bonds or bending modes of hydrocarbon contamination can also exist. It is difficult a priori to estimate how high the losses may be since the limiting levels of contamination are unknown. Hence, in this work we have attempted to quantify the levels of absorption that are achieved using a standard method of waveguide fabrication which has been used successfully for devices that operate in the near infrared [25, 26].

In this report we studied waveguides made using Ge_11.5As₂₄Se_64.5 glass as the core on a Ge_11.5As₂₄S_64.5 bottom cladding and this combination provided a refractive index contrast of about 0.4. The best waveguides (which included a thin fluoropolymer top coating ≈10nm thick) had losses for the TE-mode averaging < 0.5dB/cm over the whole measurement range from 3µm – 7.4µm (3000-1350cm⁻¹). However, around 3.43µm these increased to about 0.8dB/cm due to C-H contamination on the waveguide surfaces that could not be removed by either wet chemical or plasma cleaning. Beyond ≈7µm the losses also increased due to the presence of the fluoropolymer coating, whilst the lowest loss of 0.3dB/cm occurred near 5µm where there are generally fewer absorption bands present in any of the most likely contaminants. Since these waveguides were designed for sensing, we also include a simple “sensing” experiment in the MIR where we used our devices to measure the absorption spectrum of soluble Prussian Blue in Dimethyl Sulfoxide (DMSO) solvent placed on the waveguide surface.

2. Design of chalcogenide waveguides for MIR sensing

Amongst the wide range of possible chalcogenide compositions, Ge_11.5As₂₄Se_64.5 has attracted our special interest because this composition has excellent film-forming properties and high thermal and optical stability under intense illumination [27, 28]. Normally ChG waveguides for the near infrared are fabricated on oxidized silicon wafers, however, this substrate is not useful above ≈4.7µm due to rapidly increasing absorption from the silica bottom cladding. In this work we, therefore, employed a lower index chalcogenide glass with composition Ge_11.5As₂₄S_64.5 as the bottom cladding, although MgF₂ crystalline substrates were also employed successfully. This latter substrate has the advantage of providing a larger index contrast between the bottom cladding and the core and this helps push power into the sensing region (liquid or gas) above the waveguide. However, a major issue when using MgF₂ crystalline substrates is their cost and fragility. Additionally MgF₂ would be expected to absorb beyond about 9µm. The alternative solution of using a chalcogenide lower cladding reduces the index contrast and this, in turn, reduces the power in the sensing region, however, it allows the use of low cost substrates such as silicon. In addition, the long wavelength limit of such an all-chalcogenide structure can extend beyond 10µm even when using a sulphur-based glass as cladding. However, for either gas or liquid sensing, the waveguide design has to be chosen carefully to avoid cut-off of the fundamental mode which can occur due to the asymmetry of the structure.

Rib-type waveguides were chosen since these have been shown to produce low losses both in TE and TM modes because of reduced interaction between the mode and the etched sidewall [20]. A number of parameters can be varied to produce the best waveguide design, including the thickness of core layer, the waveguide width and the etch depth. Since our aim was to produce a device operating over a wide wavelength range (from 2.5µm to 6.6µm), some compromises were necessary. For example, to avoid cut-off at long wavelengths the film thickness and etch depth could not be very small, and this meant that below 3µm the waveguides could support TE1 and TM1 modes although their mode indices were very close to that of the slab modes and hence they should be leaky. However, this also meant that the power in the sensing region above the waveguide surface was smaller than desirable. Our compromise design is shown in Fig. 1(a) and consisted of a 2.5µm thick layer of Ge_11.5As₂₄Se_64.5 glass on a 2µm thick Ge_11.5As₂₄S_64.5 bottom cladding on an oxidized silicon wafer. The refractive indices of Ge_11.5As₂₄Se_64.5 and Ge_11.5As₂₄S_64.5 over the entire wavelength range used in the experiments were measured using an infrared spectroscopic ellipsometer (IR-VASE) and interpreted as Sellmeier Eq. (1) and (2) respectively.

 

Fig. 1 (a) Schematic representation of the waveguide for sensing. (b) SEM image of the waveguide.

Download Full Size | PDF

3. Experiments

3.1 Fabrication of low loss chalcognide waveguides for sensing in the MIR

Bulk Ge_11.5As₂₄Se_64.5 and Ge_11.5As₂₄S_64.5 glasses were prepared by the conventional melt-quenching technique from purified starting materials [27]. Using these bulk glasses, thin films were deposited via thermal evaporation onto oxidized silicon (TOx) wafers (or MgF₂ wafers) using an Angstrom Engineering deposition system. A variety of different structures were prepared. These included 2.5µm thick Ge_11.5As₂₄Se_64.5 core layers evaporated directed onto TOx wafers; 2.5µm thick Ge_11.5As₂₄Se_64.5 core layers evaporated directed onto MgF₂ wafers; Ge_11.5As₂₄S_64.5 bottom cladding 2µm thick followed by a 2.5µm thick Ge_11.5As₂₄Se_64.5 core layers deposited onto TOx wafers. In each case straight photo-resist lines 4μm wide were patterned on the film using an i-line (λ = 365nm) contact mask aligner (Karl Suss MA-6) and wet development. The ChGs core layer was then dry-etched using an inductively coupled plasma reactive ion etcher (Oxford Instruments Plasmalab ICP100) with CHF₃ gas. The residual photoresist layer was removed using wet chemical stripping and oxygen/argon plasma. The top surface of the waveguides were either left uncoated or were covered with a layer of fluoropolymer between 10nm and 1µm thick that was deposited at high CHF₃ pressure and gas flow in the ICP100 after oxygen plasma cleaning of the waveguide surface [29]. The deposition rate of the fluoropolymer was measured to be 50nm/min and hence its thickness could be controlled with high accuracy ( ± 1nm) via the deposition time. A cross section image of the waveguide is shown in Fig. 1(b). The refractive index of fluoropolymer coating was also measured by IR-VASE and had a value of 1.35 at 4µm. To measure the optical losses of the waveguides we adopted a cut-back procedure using waveguides with different lengths. The waveguide transmission was determined as a function of wavelength for each sample and the losses deduced from the ratio of the transmissions divided by the length difference. Several different waveguides were measured to allow the results to be averaged but, generally, the scatter between measurements was less than ± 15%.

Several optical sources were available for the loss measurements. The first was an optical parametric amplifier based on PPLN fan-out grating pumped by a picosecond Nd:YVO₄ laser using a similar arrangement to that described in [30]. This system was continuously tunable from ≈3µm to ≈4.22µm and from 4.6µm to 5µm. The band from 4.22µm to 4.6µm could not be accessed because of parasitic high order SHG process for the pump in the PPLN led to crystal damage. In addition attenuation due to CO₂ in the laboratory atmosphere adversely affected some measurements between 4.15 and 4.22µm since the optical path from the OPA to the waveguide was about 5m. The range 5.05-5.45µm was covered by a CW high-resolution tunable quantum cascade laser (QCL) (Daylight Solutions MHF), whilst from 6.2µm to > 7.5µm a pulsed tunable QCL laser was used (Daylight Solutions MIRcat). In each case, light was coupled into the waveguides using a molded chalcogenide lens with NA = 0.85. The light emerging from the waveguides was collected with a NA = 0.56 chalcogenide lens and either imaged onto a Xenics Onca InSb camera which was used to optimize alignment, or an appropriate detector. The OPA source could be scanned at high resolution under computer control to automatically record the transmission spectrum and this feature was later used to record an absorption spectrum with around 2cm⁻¹ resolution.

The primary fluoro-polymer vibrational absorption bands from polymers deposited from CHF₃ gas have been reported to lie around 1300cm⁻¹ [31] and, hence, overtone bands in the middle of the transmission range are anticipated. Since, to our knowledge, there have been no reports of the strength and positions of these overtones, we used waveguides fabricated on TOx wafers and coated with different thicknesses of fluoropolymer to study the effect this coating had on the losses within the transmission band. Sample results are shown in Fig. 2(a).

 

Fig. 2 (a) Loss as a function of wavelength for Ge_11.5As₂₄Se_64.5 rib waveguides on TOx wafer with different thickness of fluoropolymer. These data identify overtone absorption bands between 3.75µm and 5µm in the fluoropolymer. (b) Loss as a function of fluoropolymer thickness.

Download Full Size | PDF

Beyond ≈4.75µm the losses increased rapidly due to absorption in the TOx bottom-cladding since in this range the absorption of silica rises to several hundred dB/cm and around 1% of the power in the waveguide penetrated into that cladding. Apparent in Fig. 2a are a set of three peaks around 3.4µm which proved resilient to oxygen plasma cleaning and did not change significantly with the thickness of the fluoropolymer. These are discussed in more detail below. Between 4.65µm and 4.8µm and between 3.75µm and 4.2µm it appeared that the absorption increased with the thickness of the fluoropolymer layer [Fig. 2(b)]. Thus, it was apparent that the overtone absorption due to the fluoropolymer cladding increased the losses over a wide range of wavelengths by around 1-2dB/cm, and this is unacceptably high. Thus, only a very thin layer of fluoropolymer could be tolerated to achieve the lowest losses.

An obvious solution would be to omit the fluoropolymer coating altogether, however, then an additional problem arose. The surface of a bare waveguide became rapidly contaminated with adsorbed water and adventitious hydrocarbons when exposed to the laboratory atmosphere. An example of this is shown in Fig. 3. In this case no coating was applied and the absorption rose over a short period to > 2dB/cm at the short wavelengths with a tail extending to beyond 4µm. We found, however, that this could be prevented if the waveguide was first cleaned with oxygen plasma and then a thin 10nm layer of fluoropolymer was deposited before exposure to the atmosphere.

 

Fig. 3 Loss as a function of wavelength for a Ge_11.5As₂₄Se_64.5 rib waveguides core on a Ge_11.5As₂₄S_64.5 bottom cladding with no fluoropolymer coating.

Download Full Size | PDF

Our final sample, therefore, included a thin (≈10nm) fluoropolymer coating. We measured the absorption of this sample in detail from 3µm-7.4µm (1350-3334cm⁻¹) (the 2.5µm-3µm region was not covered by our sources). The results shown in Fig. 4(a) and demonstrate that over almost the whole functional group band the losses were below 1dB/cm and averaged < 0.5dB/cm. The lowest loss was 0.3dB/cm around 5µm. There was no indication of H-contamination in the core that would have led to increased losses at 4.7µm. The peaks at 3.4µm were due to residual C-H contamination and here the losses rose to 0.8dB/cm. This band is composed of three peaks characteristic of the C-H stretching modes of alkanes. We, therefore, assigned them to the CH₃ asymmetrical stretch at 2950 ± 15cm⁻¹; the CH₂ asymmetrical stretch at 2910 ± 15cm⁻¹; and the CH₂ symmetrical stretch at 2845 ± 15cm⁻¹. Interestingly comparing Fig. 2(a) and Fig. 4(a), we can see that the relative strengths of these three peaks varied between samples and, in particular, the CH₃ asymmetrical stretch almost disappeared in Fig. 4(a) relative to the CH₂ asymmetric stretch. Thus, it seems that there is some variability in the molecular structure of the C-H contamination produced by our process, although the reason for this is unclear.

 

Fig. 4 (a) Loss as a function of wavelength for Ge_11.5As₂₄Se_64.5 rib waveguide core on Ge_11.5As₂₄S_64.5 bottom cladding with 10nm tick fluoropolymer coating. The gaps in the spectrum were due to the unavailability of sources in these regions. (b) Comparison of long wavelength absorption for a waveguide with 10nm and 50nm thick fluoropolymer coating.

Download Full Size | PDF

The losses remained at ≈0.5dB/cm at the shortest wavelengths suggesting that there was no significant water adsorbed onto the fluoropolymer-coated sample. At the longest wavelengths, beyond 6.5µm, the losses rose as the fundamental C-F absorption of the fluoropolymer located at 7.7µm was approached. This was confirmed by comparing the loss for a waveguide with 10nm coating to that with a 50nm coating that had losses ≈2dB/cm higher at 7.35µm as shown in Fig. 4(b). It is also worth noting that beyond ≈8µm the TE guided mode was approaching cut-off and this could also increase the loss. Nevertheless, for the sample with a 10nm coating, the losses remain ≤ 1dB/cm up to the end of the functional group band at 6.6µm (1500cm⁻¹).

There are several sources of fluctuations present in the measurements shown in Fig. 4 that affect the achievable signal to noise ratio. These originate due to mode-beating and mode coupling, particularly at short wavelengths where the waveguides were multi-mode; from Fabry-Perot effects; and from absorption in the laboratory atmosphere. The OPA had a minimum linewidth of 2cm⁻¹ since the pulses were close to transform-limited and was, therefore, mostly affected by mode-beating in the waveguides. The Daylight Solutions MHF laser was externally modulated to increase its bandwidth to 1cm⁻¹, however, the modulation frequency can be only a few hundred Hz and, hence, the measurements using this laser were also affected by Fabry-Perot resonances in the chip and the optics as the wavelength swept. Particularly strong modulation occurred at some wavelengths due to the presence of sharp absorptions lines from water vapour in the laboratory air that are present in the 5.05-5.45µm band. Whilst the MIRcat has a linewidth specified to be ≤ 1cm⁻¹, its instantaneous linewidth is narrow. As a result, it produced strong time-dependent Fabry-Perot resonances in uncoated optics because the frequency chirps during the 100ns long pulses used in the measurements. These sources created fluctuations in the measurements at about the ± 15% level. We expect some of these could be reduced by eliminating Fabry-Perot resonances by using wedged optics, and employing anti-reflection coatings, as well as operating, for example, in a dry nitrogen atmosphere.

3.2 Setup of MIR sensing platform

To demonstrate the ability of these waveguides to perform spectroscopy, which is the basis of sensing, we employed the apparatus shown in Fig. 5. Here the OPA was used as a tunable source scanned in ≈2cm⁻¹ steps and the chip transmission measured against a reference detector that monitored the input power to the chip. For this experiment the waveguide coupling system and the detectors were enclosed in a dry-nitrogen-purged enclosure. A 1.4cm long waveguide sample was aligned to the beam and its transmission checked at high resolution with no analyte. The solution of Prussian blue (PB: ferric ferrocyanide) was prepared by dissolving about 80mg of soluble PB in 8g of DMSO that had been dried using 4Å molecular sieves. The resulting solution was then diluted by about an order of magnitude to obtain a solution containing around 0.1wt% of PB in DMSO. Firstly, a drop of DMSO that covered about a 4mm length of the waveguide was placed on the chip and the transmission was measured. More DMSO was then added to extend the covered length to about 9mm. The loss introduced by the DMSO was determined from the ratio of the transmissions obtained from the two measurements. This procedure was repeated with the solution of PB in DMSO. Because of its high boiling point, 189°C (372°F), DMSO evaporates very slowly at normal atmospheric pressure.

 

Fig. 5 Schematic of MIR sensing scheme based on a chalcogenide planar waveguides and tunable laser source.

Download Full Size | PDF

The resulting spectra are shown in Fig. 6(a). Here several distinctive strong peaks are evident at 3334nm and 3431nm characteristic of DMSO as is the shoulder at 3550nm and the weak peaks between 3650nm and 4150nm. The main absorption for PB lies at 4775nm and is due to the main C-N stretching vibration of the ferric ferrocyanide molecule. The positions of these absorption peaks are in agreement with reference spectra from the AIST infrared database [32], however, the overall shape of the spectrum is somewhat different. As noted previously, the power in the sensing region varies quite strongly with wavelength being > 3 times higher at 5µm compared with 3µm. Correction for this factor produces the spectra in Fig. 6(b) which is in good agreement with the database reference by in particular producing a flatter background beyond ≈3.6µm where the device sensitivity is increasing rapidly. A remaining difference exists at the short wavelength end of the spectrum where the 3334 and 3431nm peaks are superimposed on a broad background. This is due to the presence of water in our sample. DMSO is extremely hygroscopic and readily absorbs water vapour from the environment. Whilst we had attempted to reduce the water content, this was not completely successful as is evident from the presence of the broad absorption tail characteristic of water.

 

Fig. 6 (a) Loss spectrum of DMSO (blue) and PB in DMSO (red). (b) Loss spectrum of PB in DMSO corrected for wavelength dependence of power in the sensing region.

Download Full Size | PDF

4. Discussion

We have demonstrated that high index contrast chalcogenide waveguides can have quite low optical losses across most of the functional group band in the MIR between 1500 and 4000cm⁻¹. The losses were as low as 0.3dB/cm at 2000cm⁻¹ (5µm) and typically 0.5dB/cm over most of the measurement range. In particular, we found that the level of contamination introduced by a standard fabrication process did not raise the losses to unacceptable levels anywhere in the band. The main identifiable absorptions due to contamination, that also proved resilient to plasma cleaning, could be attributed to C-H stretching modes of CH₃ and CH₂ near 3.4µm where, in a fresh sample, the losses were increased by about 0.4dB/cm above the background.

In a bare waveguide, however, losses between 2500 and 3300cm⁻¹ (4µm-3µm) were large when the waveguide was exposed to a normal laboratory atmosphere for a short period of time. This is most likely due to adsorption of water and adventitious hydrocarbons onto the surface. A few hours of exposure caused the losses to increase by 0.5-2dB/cm depending on the wavelength. At one level this indicates the efficacy of these waveguide devices as sensors, although any increase in background loss would ultimately reduce the sensitivity to other molecules. We found, however, that the build-up of surface contamination could largely be prevented by coating the waveguide surface with a very thin (10nm) layer of fluoropolymer deposited at high pressure and gas flow rate from CHF₃ plasma. In fact the waveguide measured in Fig. 3 was the same one used to produce the data of Fig. 4a after it had been oxygen-plasma cleaned and coated with ≈10nm of fluoropolymer. The losses in this waveguide were stable in the 2500-3300cm⁻¹ range over about a month with the exception of the narrow C-H stretching bands around 3.4µm which approximately doubled in strength over this period. No increase in the background was observed at the water peak at 3µm. This suggests that adventitious hydrocarbons can bond to the fluoropolymer surface but water is largely inhibited by the hydrophobic nature of the coating. 10nm of fluoropolymer was thin enough to have little effect on the power in the sensing region above the waveguide and did not increase the losses significantly at other wavelengths.

If, however, the fluoropolymer layer thickness was increased to 50nm, the losses both near the fundamental absorption band and overtone bands started to rise significantly. Beyond 6.5µm the spectrum qualitatively reproduced the wavelength dependence of the main vibrational absorption band at 7.7µm that has been reported for a fluoropolymer deposited from CHF₃ plasma in [31]. From our measurements we identified a broad overtone absorption in the 4-5µm (2500-2000cm⁻¹) range estimated to have a strength of ≈100dB/cm attributable to the fluoropolymer. For a 1µm thick layer this increased the losses in this region of up to 1.5dB/cm whilst a 50nm layer still added about 0.5dB/cm. Thus, even a 10nm layer raises the absorption in the overtone band by about 0.1dB/cm but this seems acceptable because the coating largely inhibits the build up of surface contamination that otherwise rapidly increases the loss. It is possible that different fluoropolymers can be deposited with lower loss. For example, Reference [31] reports that polymers deposited from CF₄, C₂F_6, and C₃F₈ plasma have quite different MIR absorption characteristics which suggests they should be trialed as coating materials.

MIR sensors made from alternative waveguide materials could also benefit from our thin fluoropolymer coating. For example, we have measured the losses for bare 400nm thick SOI waveguide between 3.2 and 4.2µm and these also showed large additional losses attributable to surface contamination by adsorbed water and adventitious hydrocarbons. Because these silicon waveguide had much more of the field in the sensing region, surface contamination increased the losses by 5 - 10dB/cm above the background measured at 3.8µm. Care must, therefore, be taken to record the spectral dependence of waveguide loss in the MIR since single wavelength measurements may be markedly affected by (or miss the effect of) surface contamination.

5. Conclusions

We have presented measurements of the optical losses of chalcogenide rib waveguides designed for sensing applications that operate over the functional group band from 1500 to 4000cm⁻¹. The build-up of surface contamination from the environment has been identified as a significant issue for bare waveguides but has been shown to be controllable if a thin fluoropolymer coating is applied to the waveguide. Fluoropolymer produced from CHF₃ plasma does introduce absorption across a wide range of the functional group band but we found that only a 10nm layer is required to prevent water adsorption. We estimated this layer added about 0.1dB/cm to the losses in the 4-5µm range. This is small compared with the waveguide losses which averaged 0.5dB/cm over most of the functional group band. The high index contrast and low optical losses indicate that high-Q MIR micro-resonators can be successfully fabricated to increase the sensitivity for molecular sensors.