Introduction

Chalcogenide glasses contain one or more chalcogen elements (S, Se or Te) as a major constituent covalently bonded to network formers, such as Ge, As, Ga, or Si to form a glass with unusual and sometimes remarkable properties. Chalcogenides have found widespread application as phase change materials for optical data storage media (DVD-Rs) and non-volatile random access memories (PRAM); as lens materials for thermal infrared imaging; as photovoltaic cells; and due to their large non-resonant non-linearity and low linear and non-linear losses, see e.g [1–3], as promising materials for integrated devices for all-optical processing [4–6]. The extraordinarily wide transmission of chalcogenides (out to 20μm in some cases) has also sparked interest in their use to create integrated devices for mid-infrared sensing and defense applications since most chemical or biological contaminants or toxins have their spectral fingerprints in this region.

To unleash the potential of chalcogenides for these applications, high performance, low loss waveguide “chips” are needed that are fabricated using simple, low-cost processing methods, especially for applications where the “chips” have to be disposable, e.g. chemical sensors. Many methods have been demonstrated for fabricating chalcogenide waveguides but the best results so far come from using standard photolithography combined with dry plasma etching. Using this approach waveguide losses of 0.05dB/cm have been reported for As₂S₃ waveguides with ~7μm² mode areas, and ~0.2dB/cm at 1550nm for devices with ~1.7μm² mode areas [7]. These results were, however, not easy to achieve because chalcogenides are aggressively attacked by the alkaline chemicals used in photolithography as well as most gases in plasma form. Thus very careful attention needed to be paid to process design and this introduces complexities such as the use of protective layers [8]; optimized annealing conditions [9]; and tailored process chemistries [10]. This increases the number of processing steps and leads to less than ideal process control.

Thermal nanoimprint technologies on the other hand are extremely fast and low cost, requiring only a single step to fabricate the device layer. They are also well known for their ability to produce even nanometer sized features [11] with no use of chemicals. Further, they can be used unmodified with any glass composition which has suitable softening characteristics, which means it is not necessary to change the processing steps as the material composition changes. Several groups have demonstrated the possibility of molding chalcogenide glasses [12–15], but to date the best waveguides exhibited unacceptably high propagation losses of 2.9dB/cm at 1550nm [15].

In this paper we report the fabrication of the first low loss thermally nanoimprinted chalcogenide waveguides. Remarkably, this was accomplished with a soft Polydimethylsiloxane (PDMS) stamp in contrast to all previous results, which used hard stamps. We show that a simple, low cost PDMS soft stamp can produce waveguides with excellent surface morphology and losses as low as 0.26dB/cm at 1550nm (limited by Rayleigh scattering in the glass films). The soft stamp allows the production of waveguides at the wafer scale with no stamp release issues and excellent uniformity. It also requires only a single “expensive” master to be fabricated to yield many low cost production stamps by simple casting techniques, thereby ameliorating the potential cost issues that can arise from stamp damage. We applied the soft stamp method to produce devices up to 8cm long in a highly non-linear, low softening temperature glass – As₂₄S₃₈Se₃₈. Imprinting of As₂S₃ glass into suitable waveguide geometries at much higher temperatures was also demonstrated.

Fabrication

Films 1μm thick of As₂₄S₃₈Se₃₈ glass were deposited by thermal evaporation onto 100mm <100> silicon wafers with 1.5μm of thermal oxide as under cladding. The deposition was carried out in a chamber evacuated to 2 × 10⁻⁷ Torr and at a source to substrate distance of ~40 cm using a resistively heated Molybdenum boat.

Instead of following the established hard stamp thermal imprint route we chose to modify the process we used previously for ultraviolet nanoimprint lithography of polysiloxane waveguides [16], taking advantage of the low glass transition temperature of As₂₄S₃₈Se₃₈ at ~120°C. The master for creating the final PDMS stamp was patterned on 1.5 µm of thermal oxide silicon via optical contact lithography and Inductively Coupled Plasma (ICP) reactive ion etching. A ~10nm thick anti-stick layer was then deposited by plasma processing with CHF₃ and ICP power only [16]. A 100mm diameter soft stamp was made by casting liquid PDMS (Corning Sylgard 184) onto this patterned substrate and curing at 100°C for 4 hours. The process was designed to displace the minimal amount of glass possible, adapted from a previously developed UV-NIL process [16]. For each imprinted waveguide, a pair of 0.5 µm deep “cladding ribs” was used to define the core with waveguide widths from 1.3 to 3.3 µm. To ensure flexibility and conformal molding to the substrate surface, the stamp was made with a thickness of 1-2mm. The stamp was vacuum baked at 160°C for ~4 hours to ensure the PDMS was fully cured and would not undergo permanent deformation during imprint. The stamp was used along with the deposited chalcogenide film wafer in a home built thermal imprint tool shown schematically in Fig. 1 .

 

Fig. 1 Schematic diagram of the imprinting chambers.

Download Full Size | PDF

Here the wafer with the stamp in contact sat on a hotplate with an elastic membrane suspended above it, this being vacuum sealed back to the hotplate surface. A second sealed chamber lay above the membrane. By evacuating both chambers but keeping the upper one at slightly lower pressure, the membrane bows up and all the air can be removed from the stamp. The hotplate was then heated to 190°C, and 2 atmospheres of pressure applied to the upper chamber. The membrane elastically deformed applying the pressure isobarically to the stamp to create the imprint. After ~20 minutes the hotplate was flash cooled at ~40°C/min rate by forcing compressed air through a cooling chamber incorporated in the bottom surface of the hotplate. Upon cooling below the glass transition temperature, both chambers were vented and the sample removed. The stamp was released by peeling it off by hand, the radically different compositions of the stamp and the films plus the low surface energy of PDMS ensuring there was no adhesion to the molded glass surface. Figure 2 shows some typical SEM images of the imprinted waveguides from which it is clear that very smooth sidewalls have been obtained.

 

Fig. 2 SEM images of imprinted rib waveguide in As₂₄Se₃₈S₃₈ Chalcogenide glass.

Download Full Size | PDF

Compared to the UV imprinted waveguides made with stamps fabricated by the same process [16], the sidewalls here are much smoother. It is likely that this results from pressure based smoothing of “high points” in the stamp coupled with softening of the stamp at the high process temperatures. A layer of RPO Pty Ltd IPG^TM spin-on Polysiloxane upper cladding was then applied and cured, and end facets hand cleaved on the chip with a diamond scriber leaving a finished rib waveguide with the profile shown in Fig. 3 . Inspection of the imprinted films showed no cracking despite the high thermal expansion coefficient of chalcogenide glasses and the contrasting low expansion coefficient of the silica undercladding.

 

Fig. 3 Cross sectional optical microscope view of the finished imprinted rib waveguide.

Download Full Size | PDF

Measurements and results

Measurements of the losses of the imprinted waveguides were made using the cut-back technique at 1550nm, first measuring the full sample and then cleaving this into two pieces constituting 1/3 and 2/3 of the original length. Measurements were performed using lensed fibers with a 2.5μm 1/e² mode field diameter to couple to the sample, and were taken at each length with an external cavity tunable laser (with frequency dither to average out chip Fabry-Perot resonances) and power meter. A fibre coupled mercury arc lamp source and an Agilent 86142B optical spectrum analyzer were also used to obtain the wavelength dependent propagation loss at each length. The waveguides displayed considerable amounts of mode coupling/beating that manifested itself as a strong random (from waveguide to waveguide) wavelength dependence of the insertion loss [20]. The laser-based measurements were therefore taken by sweeping the laser over ~100nm and reading the minimum loss, and the OSA data were taken with a 10nm resolution bandwidth to try to average out such effects. Figure 4 shows the cut back results for launched TE and TM modes in 5 waveguides of 3.3μm width based on the tunable laser/power meter measurement. Measurement uncertainty was verified at +/−0.1dB via repeated measurements.

 

Fig. 4 Insertion loss of 3.3μm wide waveguides measured using cutback method.

Download Full Size | PDF

Losses for 1µm high by 3.3µm wide As₂₄Se₃₈S₃₈ waveguides were 0.26dB/cm for TM polarization and 0.27dB/cm for TE polarization. Given the estimated 1.5μm² mode area from R-Soft FEMSIM calculations, these values are comparable with the best we have obtained when fabricating waveguides using standard photolithography and dry etching [7] and indicate that this simple single step thermal imprinting approach can produce high quality ribs with dimensions matching those required for dispersion engineering [17]. However given the physical quality of the waveguides seen in Fig. 2, the measured losses are surprisingly high. To seek understanding of this, the wavelength dependence of the loss was studied.

A typical result for the wavelength dependence of the propagation loss is shown in Fig. 5 . The propagation loss was obtained by subtracting the measured spectrum at 1.9 cm from that at 6.7 cm, which removes all the other wavelength dependent effects (e.g. lamp spectrum, coupling, etc). Also shown in Fig. 5 are curves following 1/λ² and 1/λ⁴ dependence fitted by pinning the loss at the 1550nm point to the measured cut back loss value. The propagation loss does not follow the 1/λ² dependence expected for sidewall scattering induced losses [18], but rather the 1/λ⁴ dependence exhibited by scattering off nanoscale inhomogeneities as Rayleigh scattering. Note that the input fibre goes multimode below 1100nm causing additional loss in the measured results below that wavelength. This also fits with the SEM images of Fig. 2 which show no appreciable sidewall roughness. The question then arises whether the losses resulted from the film itself or were induced in the film by the high temperature processing, e.g [19].

 

Fig. 5 Measured Optical propagation loss spectrum of 3.3μm wide waveguide.

Download Full Size | PDF

To investigate the raw film losses, light was coupled into a Chalcogenide glass film before the thermal imprinting process using a Metricon prism coupler and the film loss measured by imaging the top surface scattering loss at various wavelengths with a Xenics cooled InGaAs camera. The results are shown in Fig. 6 . The as-deposited thin film has a similar wavelength dependence of loss to the imprinted waveguide. This suggests that Rayleigh scattering, most likely due to phase separated clusters (known to occur in chalcogenide glasses) in the as-deposited films, was the source of the 1/λ⁴ dependence. The losses at 1550nm are ~0.1dB/cm lower in the films that the waveguides, and we believe this is a result of increased phase separation after high temperature treatment [19]. This implies that much lower losses should be possible in imprinted waveguides by using compositions that lead to more homogeneous films.

 

Fig. 6 Measured wavelength dependence of film propagation loss.

Download Full Size | PDF

One such composition is As₂S₃ where ultra low loss waveguides have been demonstrated [7] and where phase separation can be largely eliminated by thermal annealing of the as-deposited films [8–10]. In contrast to the As₂₄S₃₈Se₃₈ glass, As₂S₃ has, however, a glass transition temperature of ~180°C thereby necessitating a significantly higher imprint temperature. Previous studies [14,15] used a temperature of 240°C for imprinting with hard stamps. We determined that a temperature of 250°C was superior for the stamping method used here, and sought to fabricate devices at this temperature. Figure 7 shows an optical micrograph cross-section of an imprinted As₂S₃ waveguide with ~1μm² mode area. Despite the high temperature, the soft stamp was unaffected and imprinted well. However, severe material decomposition and crystallization occurred at such high embossing temperatures, this phenomenon having been reported previously [14,19] and being the subject of ongoing study.

 

Fig. 7 Optical micrograph of 1.7μm wide 0.85μm high imprinted As₂S₃ waveguide.

Download Full Size | PDF

Conclusion

We have demonstrated the first low loss chalcogenide glass waveguides fabricated by thermal nano-imprint lithography in As₂₄S₃₈Se₃₈ glass. A soft PDMS stamp was used in contrast to previous work using hard stamps and was shown to be capable of providing excellent performance. Losses as low as 0.26dB/cm at 1550nm were demonstrated in small mode field devices – values comparable with the best obtained in waveguides fabricated by photolithography and dry etching. The loss appears to be limited by Rayleigh scattering in the thin film itself rather than from the imprint process, promising further improvements. Imprinting of As₂S₃ glass using a soft stamp was also successfully demonstrated, however film decomposition issues need to be resolved before low loss waveguides can be fabricated.