1. Introduction

Supercontinuum (SC) generation in the mid-infrared (MIR) has increasingly become a focus for research because bright MIR light sources could be used for efficient detection of almost any biological, toxic or dangerous material or pollutants via molecular fingerprint spectroscopy [1,2]. There are also many other applications demanding such sources including metrology [3], tomography [4,5] and microscopy [6]. As a result, there have been a growing number of experiments aimed at generating MIR SC using a wide variety of nonlinear materials including silicon [7]; bismuthides [8,9]; single crystal sapphire [10]; LiNbO₃ [11]; media with rare-earth dopants [12–15]; tellurites [16–21]; fluorides [22–27]; and chalcogenides [28–30]. The potential of SC sources in the MIR is exemplified by their brightness compared with that of a typical infrared beam line on a synchrotron that produces 10^16-17 photons/s/mm²/sr/0.1% bandwidth within the 1-10 µm band. Assuming 100mW of average power can be generated in a diffraction limited beam in a SC spanning from 3 to 10 µm, the SC source brightness would be ≈10²⁰ photons/s/mm²/sr/0.1% bandwidth – up to four orders of magnitude larger than the synchrotron [31].

Nowadays, the mechanisms of SC generation in different dispersion regimes and under various pumping conditions are well understood and the numerical methods to solve the nonlinear Schrödinger equation (NLSE) taking into account linear and nonlinear losses, dispersion and nonlinear effects such as self-phase modulation (SPM), stimulated Raman scattering (SRS), soliton fission, self-steepening (SS) are also mature [32]. The challenge for extending SC far into the MIR lies elsewhere and, in particular, with the choice of nonlinear material with very broad MIR transparency and pump configuration that will allow the MIR SC to extend across as much of the molecular fingerprint region (2.5-25 µm) as possible.

In this paper, we first review the range of achievements to date by groups working in this field. We then focus on the prospects for mid-IR SC generation in chalcogenides glasses describing some of our recent experiments and simulations which we have used to define the device and pump conditions required to produce SC from around 2.5 µm to ≈12 µm in a single waveguide device. In experiments using chalcogenide waveguides we found that the inevitable reduction of the waveguide nonlinear parameter, γ = 2π n₂/λ A_eff where n₂ is the nonlinear refractive index and A_eff is the area of the waveguide mode, with increasing pump wavelength, λ, means that the high pump powers required to generate a SC can cause damage to the chalcogenide waveguides if picosecond or longer pump pulses are employed. This suggests that sub-picosecond duration pump pulses will be essential. In addition, to achieve sufficiently long wavelength extension, the pump wavelength will also need to be large (4-5 µm). This allows the waveguide dispersion to remain anomalous far into the MIR, leading to rapid extension of the spectrum to longer wavelengths. Whilst we have not yet been able to make the low loss broadly transmitting chalcogenide waveguides required to create a broadband waveguide SC source, we have been able to demonstrate SC generation in a Ge_11.5As₂₄S_64.5 bulk chalcogenide glass pumped by ≈150 fs pump pulses at 5.2 µm where the dispersion is anomalous. We could generate a flat SC extending from ≈2.5 µm to beyond 7.5 µm and this corresponded to the complete wavelength range of our mercury cadmium tellurite (MCT) detector. The measured spectra agree qualitatively with our simulations for waveguides.

2. Overview of previous experiments

There are two basic geometries that have been employed to generate SC: bulk materials and dispersion engineered waveguides. In the former case, self-focusing due to the Kerr nonlinearity of an intense pump pulse is generally required to produce the extreme intensities that generate a SC via self-phase modulation (SPM). This technique is commonly used, for example, to produce near IR seed pulses in commercial femtosecond tunable optical parametric amplifiers pumped by Ti:sapphire lasers and has been extended into the MIR by using bulk samples of tellurite or fluoride glasses as well as an yttrium aluminum garnet (YAG) crystal [33]. As examples, recently there was a demonstration by Liao et al. which reported the generation of SC up to the transmission boundary of the tellurite at about 6 µm using femtosecond pulses with a low repetition rate of 1 kHz through filamentation [21]. The same group also generated SC in bulk fluoride glasses up to their transmission limit of 8 µm, thereby covering the whole functional group region (FGR) [26]. By irradiating a YAG crystal with 85 fs pulses at 3100 nm, Silva et al. generated a SC spanning over 3 octaves from 450 to 4500 nm [33].

The main disadvantages of SC involving self-focusing in bulk materials are that the intensities required are often close to the damage limit of the material; multiple filaments can be generated when the power is significantly above the self-focusing threshold and this reduces the brightness and coherence of the source; and long wavelength beams can be highly divergent and difficult to collimate particularly when short pump wavelengths are employed. Most of these disadvantages can be overcome by employing a waveguide geometry either in the form of optical fiber or planar waveguides where propagation is controlled, path lengths can be increased, dispersion can be tailored and the required peak laser powers to create a SC reduced. In recent years there have been several demonstrations of SC generation in fibers and a few demonstrations using planar waveguides.

Kulkarini et al. reported a SC spectrum covering ≈1.9 µm to 4.5 µm in ZBLAN (ZrF₄-BaF₂-LaF₃-AlF₃-NaF) fiber with high average output power beyond 3.8 µm [23]. The fiber was pumped with a thulium-doped fiber amplifier. Qin et al. reported SC generation from 350 nm to 6.28 µm in centimeter-long fluoride fibers pumping with a femtosecond laser at 1.48 µm and this is, to our best knowledge, the broadest mid-IR SC reported in fluoride fibers [22]. At the long wavelength end, no further broadening is expected because of high fiber loss. Furthermore, the nonlinearity of the fluoride itself is rather low and this means very high peak powers were required [22] and this is ultimately a significant disadvantage.

Domachuk et al., reported a SC over 4000 nm wide and extending to 4870 nm in a 8mm long highly nonlinear tellurite photonic crystal fiber (PCF) pumped by 100 fs pulses at 1.55 µm [17]. The short length of the PCF made the SC spectrum smooth, the dispersion low and the material loss small. Nevertheless, it will be difficult for tellurites to produce emission further into the mid-IR again because of the glass transmission is limited to < 6 µm.

Buczynski et al. generated a SC spectrum spanning two octaves covering the 750-3000 nm range using a new type of lead-bismuth-galate (PBG08) photonic crystal fiber with tailored rheological and transmission properties [9]. The PCF was only 2 cm long and was pumped with femtosecond pulses with 10 nJ of energy at 1550 nm. In spite of the delicacy in fabricating this kind of PBG08 PCF, the glass itself hardly transmits beyond 5 μm, which is inadequate for mid-IR SC sources except in the short wave region of the MIR.

In addition, there have been some unique approaches to SC generation. In [10], a 5 cm long single crystal sapphire fiber was pumped with femtosecond laser pulses at 2 µm, and the generated SC spectrum covered from 1.2 µm to 2.8 µm. Doped media have also been used which combine nonlinear effects of the host material with active super-radiative processes of the dopants to generate SC. The long wavelength limits for this approach have been 2.45 µm, 2.6 µm and 2.7 µm in [12], [13] and [14] respectively. It has been proposed that this is a promising approach for SC generation at longer wavelengths if the host materials possesses better mid-IR transmission and the dopants are carefully chosen [15].

In the case of waveguides, MIR SC was recently reported from a silicon nanowire by Kuyken et al. who produced wavelengths from the telecom-bands to the short wave MIR using a 2 cm long silicon-on-insulator (SOI) waveguide pumped with picosecond pulses in the anomalous dispersion regime [7]. However, these SOI waveguides will not transmit further than about 4.2 µm due to absorption in the silica cladding layer. In addition, in [11], Phillips et al. reported SC generation from a periodically poled lithium niobate (PPLN) waveguide using a Tm-doped fiber laser as the pump, with the output spreading from 1.3 µm to 2.8 µm.

None of these materials allow the production of SC at wavelengths beyond about 8 µm because of their limited transmission. Chalcogenide or chalco-halide glasses as well as crystalline materials such as germanium are the only readily available materials that can provide much better mid-IR transparency with sulphides transmitting to beyond 8.5 µm, selenides to around 14 µm and tellurides to around 20 µm [34]. Chalcogenides also possess amongst the highest nonlinearities of all glasses [34], making them good materials to generate SC in the mid-infrared. Marandi et.al. reported SC generation from 2.2 µm to 5 µm in a tapered As₂S₃ fiber pumped with femtosecond pulses from a mode-locked Er-doped fiber laser [28]. Shaw et al. reported a SC spectrum covering from 1.5 µm to 5 µm from a single-mode step-index As₂S₃ fiber pumped at 2.5 µm [29]. In our previous work, we generated a SC spectrum from about 2.9 µm to 4.2 µm in a dispersion-engineered As₂S₃ glass rib waveguide with Teflon coating and thermally oxidized silicon substrate [30] and pumped around 3.25 µm. Although current outcomes of chalcogenide-based mid-IR SC sources have not yet achieved the wavelength coverage reported for tellurites or fluorides, the prospects of chalcogenides are much better particularly when emission covering both the functional group (2.5-7.7 µm) and fingerprint (7.7-11 µm) bands are required.

In Table 1 we summarize these representative mid-IR SC experimental results to date.

Table 1. Summary of experimental results on mid-IR SC generation

View Table

3. Supercontinuum generation in chalcogenide rib waveguides

The chalcogenide waveguides described in [30] were fabricated on an oxidized silicon wafer as substrate and this resulted in high losses beyond 4 µm that prevented the spectra extending to longer wavelengths. In [30] we also pointed out that eliminating this absorption as well as changing from 3.28 µm pump to one at 3.7 µm should lead, according to our simulations, to a remarkable increase in long wavelength extension to beyond 9µm. To pursue this possibility we replaced the substrate by a 100 mm diameter single crystal MgF₂ wafer which transmits to 7.7 µm [35]. However, using a 4.7 cm long waveguide in exactly the same apparatus as described in [30], the generated SC spectrum still did not extend beyond about 4.7 µm as shown in Fig. 1 in spite of low waveguide losses (0.5 dB/cm at 5 µm). The problem here, also encountered in [30], was that the input power to the waveguide was limited by damage at the input facets when using relatively long 7.5 ps duration pump pulses and the maximum fluence at the waveguide facet could only be ≈0.2 J/cm² (about 1600 W of coupled power). In addition, MgF₂ wafers are extremely fragile and break easily during processing and this led to a shorter waveguide (4.7 cm c.f. 6.5 cm) compared with that used in [30] and thus more power would be required for SC generation. Hence a better solution is required and we have been exploring devices where both core and cladding are made from chalcogenide glass.

 

Fig. 1 Experimental results of SC generation from an As₂S₃ glass rib waveguide 2.5 µm thick on a MgF₂ substrate and Teflon coating and pumped at 3.8 µm (red); 3.65 µm (blue) and 3.26 µm (green) at a peak power of ≈1000 W.

Download Full Size | PDF

Two chalcogenides, Ge_11.5As₂₄Se_64.5 (GeAsSe) and Ge_11.5As₂₄S_64.5 (GeAsS) have attracted our special interest. Our materials research has shown that these glasses have excellent film-forming properties and high thermal and optical stability under intense illumination [36,37] and we have identified them as good materials for nonlinear photonics. The selenide has a linear index ≈0.4 higher than the sulphide allowing a waveguide with moderate index contrast to be fabricated with the selenide as the core. In terms of dispersion, according to our ellipsometry measurements based on films, the sulphide becomes anomalous around 5µm and the selenide around ≈7 µm. Thus, with the aim of extending the SC to cover as far into the MIR as possible we designed an all-chalcogenides rib waveguide with Ge_11.5As₂₄Se_64.5 as the core and Ge_11.5As₂₄S_64.5 being both the under and upper claddings. In the following section, we will first describe simulation results on MIR SC generation in these particular chalcogenide waveguides.

4. SC simulations in chalcogenide waveguide

In rib waveguides we can modify the properties such as the nonlinear parameter γ and dispersion D by varying the core film thickness T, and the etch depth, E. The geometry considered is shown in Fig. 2. Using the full-vector finite difference method (FDM), we obtained contour plots parameterizing D and γ for the fundamental TM mode for various core film thicknesses and etch depths. Figure 3 shows a typical example when the core width was 4 µm and the core thickness was varied whilst the etch depth was maintained at 40% of the thickness. We chose the TM mode and this specific etch depth after comparison of simulation results for various different configurations and after consideration of practical limitations. If the etch depth is too small, coupling to slab modes can occur and this can make the waveguide very lossy. In addition, reducing the etch depth causes the effective mode area to increase which reduces the nonlinear parameter γ. Alternatively, if the etch depth is large, sidewall roughness on the etched surfaces of the rib will also cause the waveguide losses to increase.

 

Fig. 2 A schematic of the structure of the all-chalcogenide rib waveguide.

Download Full Size | PDF

 

Fig. 3 The (a) dispersion parameter D and (b) nonlinear parameter γ for the fundamental TM mode as a function of wavelength and core film thickness for a rib waveguide with 40% etched depth. The locus of the zero dispersion and the contour for γ = 0.5 W⁻¹m⁻¹ are shown by the white lines.

Download Full Size | PDF

It is clear from Fig. 3 that in this all-chalcogenide waveguide with relatively small refractive index contrast, the net dispersion is normal over a wide range of parameters. Only by using relatively thick films (>3.5 µm) can the anomalous dispersion regime be reached and only at long pump wavelengths (>4 µm). The dispersion parameter, D, also remains comparatively small (< 5 ps/nm/km) within the anomalous region. What is also evident is that in the region of anomalous dispersion, the nonlinear parameter will be quite low (<0.4 W⁻¹m⁻¹), and this means the power required to generate a SC will be high relative to that required for SC generation in the near IR (1.55 µm) where nonlinear parameters are > 10 W⁻¹m⁻¹.

The choice of film thickness becomes a trade-off between dispersion and nonlinearity as well as losses to leaky modes particularly for the long wavelengths contained within the supercontinuum. Based on such considerations, we chose the core film thickness to be 4 µm and the TM mode because of its smaller losses to leaky modes. It is worth noting that the waveguide supports only the lowest order TM mode at the pump wavelengths and in this relatively symmetrical structure, it does not show cut-off at longer wavelengths although losses to leaky modes start to rise beyond ≈12 µm. However, this is well beyond the limit expected due to cladding absorption in this particular waveguide (≈10 µm).

By using the split-step Fourier method (SSFM) we solved the nonlinear Schrödinger equation to model SC generation in a 7 cm long rib waveguide including a linear loss of 1 dB/cm at all wavelengths. We fitted the dispersion data with a Taylor series including up to 10th order dispersion since this accurately reproduced the dispersion map over the whole wavelength range of the calculations. A sufficiently small time window was chosen to ensure the spectral coverage greatly exceeded the spectral width of the generated SC. We varied the pump wavelength, the gamma-power-length product (γPL where P is the peak pump power and L the device length) and the pump pulse duration. We chose two pulse durations: 7.5 ps being the pulses produced from our current mid-IR optical parametric amplifier (OPA) pump source [38] and 250 fs, which we expect to generate with a new OPA by replacing the current Nd:YVO₄ pump laser with a mode-locked Yb laser, or typical of a value achievable from a synchronously pumped optical parametric oscillator. As to pump wavelengths, fiber lasers at 2 µm or 3 µm could emerge as viable pumps for SC pump sources and, hence, performance at these wavelengths is interesting in spite of the fact they lie in the normal dispersion regime for this particular waveguide. For (these and) longer wavelengths OPAs based on PPLN can be used and hence we investigated 4 µm and 5 µm as examples of pumps in the anomalous dispersion regime.

The maximum γPL values are ultimately limited by optical damage of the materials. The mode area ranged from ≈12 µm² at 2 µm to ≈20 µm² at 5 µm and γ values from 1.47 W⁻¹m⁻¹ to 0.13 W⁻¹m⁻¹respectively. If we assume that the maximum fluence at the facet should not exceed 0.2 J/cm² at 7.5 ps, then the maximum γPL drops with increasing pump wavelength from around 550 at 2 µm to 50 at 5 µm. By reducing the pulse duration the damage fluence should drop in proportion to t_p^1/2 or the maximum intensity increase by t_p^-1/2 raising the maximum γPL by ≈5.5 at each wavelength. We calculated spectra for a range of γPL values at each pump wavelengths and pulse durations from as small as 7 up to ≈100. Whilst higher γPL values could have been used with the 250 fs pulses this proved unnecessary. The trends are shown in Figs. 4 and 5 for 7.5 ps and 250 fs pump pulses respectively.

 

Fig. 4 Simulated SC spectra for different wavelengths and increasing values of γPL for 7.5 ps pulses. Blue 2 µm; green 3 µm; red 4 µm; turquoise 5 µm. (a) γPL = 49; (b) γPL = 77; (c) γPL = 91; (d) γPL = 105.

Download Full Size | PDF

 

Fig. 5 Simulated SC spectra for different wavelengths and increasing values of γPL for 250 fs pulses. Blue 2 µm; green 3 µm; red 4 µm; turquoise 5 µm. (a) γPL = 7; (b) γPL = 21; (c) γPL = 49; (d) γPL = 77.

Download Full Size | PDF

Figure 4 shows the predicted spectra for different pump wavelengths and γPL values for 7.5 ps duration pulses. It is quite apparent that pumping with the anomalous dispersion region at 5 µm produces the broadest spectrum at the lowest threshold power. At 2 µm where the dispersion is large and normal, the broadening is small. The case of 4 µm is interesting since in spite of lying in the anomalous dispersion regime, the threshold power for SC is high and more broadening is in fact predicted when using a 3 µm pump with normal dispersion.

In the case of the 250 fs pulses, shown in Fig. 5, the spectral broadening actually occurs at much lower values of γPL than at 7.5 ps and γPL from 10 to 20 is all that is needed to create SC using the 4 µm or 5 µm pumps. This corresponds to peak powers of <2200 W at 5 µm. At higher γPL, significant broadening is also observed at 3 µm for normal dispersion. In this case self-steepening of the pulse occurs which enhances the SPM creating the relatively flat-topped steep sided spectrum observed in Fig. 5(d). At 2 µm the broadening remained small in all conditions.

It is particularly interesting to understand why a relatively low threshold is observed for 250 fs pumping at 5 µm. The clue can be found by calculating dispersion lengths and soliton fission lengths for the various conditions. The dispersion at both 4 µm and 5 µm is small: 4.9 × 10⁻⁴ ps²/m at 4 µm and 6.87 × 10⁻² ps²/m at 5 µm. Hence the dispersion lengths (L_D) are very large indeed ≈10⁵ and 10⁴ m respectively when using 7.5 ps pulses and correspondingly 820 m and 0.91 m for the case of 250fs pulses. The soliton numbers (N_S) are then also large ranging from 30 for a 5 µm pump at 250 fs pulse to 11000 for a 4 µm pump and 7.5 ps pulses. As a result, only in one condition is the soliton fission length (L_D/N_S) less than the device length: namely in the case of the 5 µm 250 fs pump. In all other cases, the soliton fission length is in the 1-10 m range, requiring higher γPL values to create a SC. These large soliton numbers also would normally suggest that the modulational instability causes breakup of the pump pulses into multiple low order solitons. Whilst the pulse envelope breaks up into multiple solitons in simulations using 7.5 ps pulses, in the case of the 250 fs pump it is clear that the dominant mechanism is soliton pulse compression and this results in a coherent spectrum in spite of the relatively large soliton number of 30. In Figs. 6(a) and 6(b) we demonstrate this by plotting the input and output pulse profiles in the case of 7.5 ps and 250 fs pulses, respectively. It is easy to see, however, that soliton numbers calculated above are overestimates. A single cycle of the carrier at 5.3 µm lasts 18 fs and hence it would be impossible for breakup of the pulse envelope into pulses shorter than a few cycles, as implied by large soliton numbers. The simulation in Fig. 6(a) demonstrates that the 250 fs pulses are instead compressed down to ≈40 fs, two to three optical cycles.

 

Fig. 6 The time dependence of the power at the waveguide output from simulations. (a) 250 fs pulses for γPL = 7 at 5 µm demonstrating soliton pulse compression. (b) 7.5 ps pulses for γPL = 64 at 5 µm demonstrating break-up of the pulse envelope into multiple solitons.

Download Full Size | PDF

The important point to note is that SC can be generated at low γPL values with a 250 fs pump corresponding to fluences far below the expected damage limit. For the longer pulses, however, the required fluence is close to or above the damage limit encountered in our experiments and generally the spectrum is narrower. We conclude, therefore, that it should be possible to generate a SC well beyond 10 µm by pumping in the anomalous regime with ≈250 fs pulses, but this requires a source operating at 4 µm or longer.

5. SC experiments using bulk glasses

Presently we are still developing methods to fabricate the waveguides described above along with the appropriate high repetition rate, moderate peak power (10 kW) fs pump source for pumping in the 4-5 µm range. However, since SC generation is not inherently a waveguide phenomenon, but simply requires sufficient nonlinear phase shift and anomalous dispersion, we should expect similar spectral broadening and SC threshold by pumping a bulk chalcogenide glass at a wavelength where its dispersion is anomalous. This occurs in sulphide glasses with a pump in the 5-6 µm range. In such conditions, the same processes that occur in a waveguide should dominate SC generation: soliton fission, self-phase modulation, parametric amplification and the Raman self-frequency shift, although self-focusing could also occur in this geometry.

To test this we irradiated a ≈5 mm thick bulk sample of Ge_11.5As₂₄S_64.5 glass with up to 20 MW peak power in ≈150 fs duration pulses at ≈5.3 µm. The pulses were generated by a Quantronix Palitra OPA fitted with a difference frequency generator and pumped by a Clark CPA2001 Ti:sapphire laser. The output beam from the Palitra was filtered with a 2.5 µm long-pass filter to isolate the MIR pump beam and block the signal and idler. A beam-shaping aperture was used to select the most uniform region of the beam creating an Airy diffraction pattern at a 100 mm focal length CaF₂ lens that focused the beam into the glass sample. The output beam was imaged using a NA = 0.56 molded chalcogenide lens onto the input slit of a Newport Cornerstone 0.25 m monochromator fitted with a 150 l/mm grating. The output was dispersed onto a Vigo PVI-2TE-6 mercury cadmium telluride detector. Order blocking filters allowed the monochromator to cover the spectral range from ≈2.5 µm-7.5 µm (approximately the maximum spectral coverage of the detector) and the input signal could be attenuated to achieve high dynamic range.

The nonlinearity of Ge_11.5As₂₄S_64.5 glass at 1.55 µm was determined by z-scan measurements to be (2 ± 0.3) × 10⁻¹⁴cm²/W but will be significantly smaller at longer wavelengths. In the case of As₂S₃, for example, our waveguide results [30] suggested that the nonlinearity at 3.28 µm was about half that at 1.55 µm. In the absence of specific measurements, we estimated the nonlinearity to be respectively 2 or 3 times smaller at 3 µm and 5 µm than that at 1550 nm. The maximum power used was ≈20 MW and is about 8 times the self-focusing critical power at 5.3 µm based on a nonlinearity of ≈7 × 10⁻¹⁵ cm²/W. The focal spot diameter was determined by imaging the output onto a Xenics Onca InSb camera from which we found the FWHM beam diameter to be ≈90 µm. Using this value the self-focusing distance is predicted to be ≈2 mm – about half the sample length. Thus, self-focusing could contribute to SC generation making it difficult be sure of the γPL value although measurements showed that in the absence of self-focusing it would be 14. In our experiments the spectra were very stable and evolved smoothly as the power was increased, and there was no indication of any sudden increase in nonlinear response as would be expected were self-focusing to occur.

Sample experimental data for different input powers are shown in Fig. 7. In Fig. 7(a) the black dotted line is the detector responsivity supplied by the manufacturer. With increasing power the spectrum broadens steadily. At about 3 MW it shows evidence of two weak side peaks at 4.1 µm and 7.1 µm that lie symmetrically about the pump at 5.3 µm, which we attribute to four-wave mixing gain suggesting the dispersion is indeed small and anomalous. With increasing power the spectrum broadens and flattens. In fact, correcting the shape of the spectrum at 20 MW with the detector response produced a spectrum flat to ± 5 dB from 2.5 to 7.5 µm as shown in 7(b) mirroring the simulations of Fig. 5. Moving the OPA to 3.425 µm, a region where the dispersion is normal, produced a very different result. As shown in Fig. 7(c) for a pulse power of 20 MW the spectrum did not show the large spectral broadening observed for the 5.3 µm pump and had a width at −10 dB of peak of only ≈1000 nm. This is because in the normal dispersion regime spectral broadening is the result of self-phase modulation along with self-steepening of the pulse. Both these spectra are in good qualitative agreement with the predictions of the simulations for rib waveguides (Fig. 5(a, b)) displaying similar SC thresholds (γPL ≈6) and spectral form.

 

Fig. 7 (a). Spectra recorded at ≈20 MW (red), ≈6 MW (blue) and ≈3 MW (green) peak power. The detector response is shown as a black dotted line. 7(b) the spectrum at ≈20 MW has been corrected for the spectral response of the detector. 7(c) comparison of spectra at ≈20 MW produced using 5.3 µm and 3.425 µm pumps.

Download Full Size | PDF

6. Conclusion

The generation of supercontinuum at wavelengths that span much of the molecular fingerprint region in the mid-IR requires the use of broadly transmitting and highly nonlinear waveguides. Only a few materials offer the transparency to cover both the functional group and fingerprint bands that extend from ≈2.5 µm to beyond 11 µm. Of these, chalcogenide glasses are particularly attractive with selenides, for example, transmitting to beyond 14 µm. They also have highest optical nonlinearities of all glasses and can be readily used to fabricate planar optical waveguides which can be dispersion engineered to achieve small anomalous dispersion at the pump wavelength – a requirement for broadband SC generation.

Many waveguide designs exist and thus a very wide range of dispersion maps can be created and there have already been published on photonic crystal fibers that have been optimized for short wavelength pumps [39,40]. By using low index substrates such as silica or MgF₂, rib waveguides can be designed that have anomalous dispersion at relatively short wavelengths (3 µm or less). However, as we showed previously [30], using too short a pump wavelength limits the long wavelength extension of the SC. In addition most of the viable low index substrates still do not transmit well beyond 7 µm and are very expensive in wafer-sized samples. As a result in this paper we considered a solution based on fabricating both core and cladding made from chalcogenide glasses that are deposited on an inexpensive substrate such as a silicon wafer. This inevitably reduces the index contrast and as a result there does not appear to be designs where the dispersion can be anomalous when the pump wavelength is shorter than about 4 µm.

The device we considered here had a Ge_11.5As₂₄Se_64.5 core and Ge_11.5As₂₄S_64.5 upper and lower claddings with an index contrast of ≈0.4. For this structure the pump had to be longer than ≈4 µm to fall in the region where dispersion was anomalous. However if that condition is met, the simulations indicated that quite lower powers are required to generate a SC 3.5 octaves wide which extended to the transmission limit of the core glass at 14 µm, although this would not likely be achieved in practice because of cladding absorption beyond about 10 µm and losses to leaky modes at the longest wavelengths. Nevertheless, this is very encouraging prediction.

Since we have not successfully fabricated waveguides with low losses over the required range, we generated a broad SC in a chalcogenide bulk glass by irradiating with ≈150 fs duration pulses at ≈5.3 µm. This demonstration used a Ge_11.5As₂₄S_64.5 glass where the material dispersion is anomalous beyond ≈5 µm. In our experiments 20 MW peak power pulses were focused into a 5 mm thick glass sample and produced a flat SC extending from at least 2.5 to 7.5 µm, which was the full range of our available detector. On the long wavelength end this was certainly not the limit. Changing to a pump in the normal dispersion regime produced a spectrum only 1000 nm wide. In such conditions, the physics of SC generation should be the same as in a dispersion-engineered waveguide although in a bulk glass self-focusing may also be present. Our experiments reproduced qualitatively the waveguide simulations in terms of γPL threshold for SC generation as well as spectral properties.

An obvious issue is the choice of pump. No lasers have yet been demonstrated that can generate ultra-short pulses at >4 µm, however, there is a well-established route using parametric devices. Commercial synchronously pumped fs OPOs exist which produce idler outputs that can reach the lower wavelength limit. Perhaps a better solution is to extend the OPA technology that we currently use in our picosecond source into the few hundred fs region by replacing a Nd pump laser with a commercial Yb laser capable generating ≈500 fs pulses at 1050 nm. This latter system is currently under development. With such a laser we expect average powers in the SC to be in the 100-200 mW range. Produced in a diffraction limited beam, such a source would provide much higher brightness and significantly more average power than the alternative synchrotron MIR sources at least in the 3-11 µm band from a compact bench top device.