1. Introduction

New micromachining technologies have the potential to dramatically enhance the functionality and performance of integrated photonic devices. For example, integrated switches based on deflection of a suspended waveguide have been demonstrated in silicon-on-insulator (SOI) and gallium arsenide (GaAs) systems [1] and in indium phosphide (InP) based systems [2, 3]. However, the optically-active nature of III-V semiconductors has yet to be fully exploited in these micromachined waveguide devices; Lasers, optical gain segments, or electro-optic modulators can be integrated with micromachined III-V waveguides. Not only would the use of electro-optically active heterostructures eliminate the need for off-chip photonic gain or modulation components, but the electro-optic properties could also be enhanced by the unique optical properties of suspended semiconductor waveguides, which include high-index contrast and strong evanescent fields. The high index contrast inherent in suspended waveguides enables small cross-sectional mode areas, which in turn enhance nonlinear optical and electro-optical properties. In addition, the evanescent fields in the air surrounding the waveguide [4] suggest uses in sensing [5] that could also take advantage of integrated gain segments.

Previous studies of suspended quantum well heterostructures have focused on lasing in photonic crystals [6], microcantilever photodetectors [7], and lateral band structure deformation [8, 9]. In addition, a number of research groups are investigating the waveguiding properties of suspended bulk semiconductors [1, 5, 10]. The successful fabrication of a suspended quantum well waveguide in the III-V material system depends critically on the epitaxial strain, bandstructure, and etch selectivity of the different materials. In this work, we show that with a proper choice of materials, alloy fractions, and layer thicknesses within the heterostructure, a suspended multiple quantum well (MQW) waveguide can be fabricated to allow low-loss propogation in the optical L-band, just 125 nm below band gap. Such a design allows strong electro-optic effects in the near band edge regime due to the quantum-confined Stark effect, without adding significant interband absorption loss.

2. Fabrication and characterization

We have used InGaAs quantum wells with InGaAsP barriers grown on InP designed to have a band edge near 1500 nm. Both InGaAs and InGaAsP (with a phosphorous content below about 60%) are also known to have a high resistance to a hydrochloric acid (HCl) etch [11], which is used to selectively remove an InAlAs sacrificial layer. By growing either the InGaAs or InGaAsP layers with a small amount of tensile strain, the suspended waveguide will not crack or buckle upon release from the sacrificial layer. The 20-period In _.47Ga_.53As/In_.82Ga_.18As_.40P_.60 MQW is grown between two 150 nm thick In_.77Ga_.23As_.50P_.50 layers using molecular beam epitaxy. This 590 nm thick waveguide layer is grown on a 1.6 µm lattice-matched In _.52Al_.48As sacrificial layer and a 50 nm thick InGaAs etch stop layer. The waveguides are patterned and etched using electron-beam lithography followed by a chlorine-based inductively coupled plasma (ICP) etch and then released by selectively etching the sacrificial layer [12], resulting in a MQW core surrounded by an air cladding.

The waveguides are attached using tethers spaced at regular intervals along the waveguide length. The tethers vary in width from 2 µm to 4 µm while the waveguide width ranges from 1 µm to 6 µm on a given sample. The waveguide lengths range between 1.2 mm and 2.4 mm, and the tether spacing along the waveguide ranges between 120 µm and 575 µm. An example of such a waveguide is shown in Fig. 1. Figure 1(a) shows a fabricated waveguide 2 µm wide along with two tether pairs of the same width. Figure 1(b) shows a close-up of one of the facets of this waveguide.

 

Fig. 1. (a): An SEM image of a 2 µm wide suspended MQW waveguide. (b): An SEM image of the end facet of the 2 µm wide waveguide.

Download Full Size | PDF

A six-band k·p band structure model predicts a heavy-hole excitonic resonance at 1507 nm and a light-hole excitonic resonance at 1477 nm for our quantum wells. Separate spectroscopic measurements (not shown) indicate that the as-grown heavy hole band to conduction band excitonic resonance is at 1478 nm, and this resonance is shifted to 1470 nm in the suspended waveguide, due to partial relaxation of the tensile strain in the quantum well layers [9].

Light from a tunable laser is coupled to and collected from the cleaved waveguide facets using high-numerical-aperture microscope objectives. A sample normalized transmission spectrum for both TE and TM light for a 4 µm wide, 2.4 mm long waveguide is shown in Fig. 2(a). The measured losses includes propagation loss, tether loss, and insertion loss. The TE (TM) spectrum is dominated by near-bandedge absorption from the heavy-hole (light-hole) band to conduction band transition for wavelengths less than about 1580 nm (1555 nm). At larger wavelengths, the loss shown in Fig. 2 is dominated by insertion loss, as discussed below. The splitting between the heavy-hole and light-hole bands is indicated by the wavelength offset between the TE and TM spectra. Here, the approximately 25 nm splitting is close to the predicted value of 30 nm.

The etaloning evident in the transmission spectra is due to reflections from the end facets that produce Fabry-Perot cavity modes within the waveguide. The inset of Fig. 2 shows a TE transmission spectrum obtained over a smaller wavelength range, indicating the large contrast of the fringes. The free spectral range (Δλ) of these fringes is determined by the waveguide length (L) and group index (n_g):

where

where ω=2πc/λ and β=2πn _eff/λ. β is the waveguide propagation constant, and n _eff is the effective modal index of the waveguide. Thus, measurements of the free spectral range of the Fabry-Perot fringes give the group index of the waveguide for a given waveguide length.

A compilation of these measured group indices for a range of nominally 4 µm wide waveguides at wavelengths between 1570 nm and 1640 nm is shown in Fig. 2(b). The error bars represent the scatter in measurements within the population of waveguides. The group index was calculated for the fundamental mode using a two-dimensional finite-element model that solves for n _eff for a given cross-sectional geometry and wavelength. The calculated n _eff includes effects from both material dispersion and waveguide dispersion. Equation 2 is then used to obtain values for n_g from the calculated values of n _eff. The only free parameters in the model (the excitonic resonance wavelengths and the waveguide cross-sectional dimensions) are determined by independent measurements. The calculated group index differs from the measured values by about 1%, indicating that the propagation remains single mode within the waveguide with negligible polarization mode coupling.

 

Fig. 2. (a): Measured transmission through a 4 µm wide waveguide. The loss includes propagation loss, tether loss, and insertion loss. (b): Measured and calculated group index for a 4 µm wide waveguide.

Download Full Size | PDF

The contrast, K, of the Fabry-Perot fringes can be used to estimate the losses incurred during propagation between the end facets [13]:

where R is the facet reflectivity and α is the net loss (in cm^-1) during propagation over a waveguide length L. By measuring the contrast for waveguides over a range of lengths (the cutback method), Eq. 3 gives an estimate of both α and R. Our analysis using the cutback method yields a facet reflectivity that agrees within error with the calculated reflectivity based on n _eff and the Fresnel equations: 0.27 for TE and 0.26 for TM. However, the α obtained using this method does not distinguish between losses due to tether scattering and those simply due to propagation in the absence of tethers.

To separate these losses, we have measured the contrast ratio of waveguides having the same length and width, but having a range of number of tether pairs along the waveguide. In this way, a plot of the loss, α, vs. number of tether pairs can be fit to a line such that the y-intercept is the zero-tether propagation loss, and the slope is the loss per tether pair. An example of this data for TE propagation in a set of 2.4 mm long, 4 µm wide waveguides is shown in Fig. 3. The loss for each waveguide is deduced from Eq. 3 using the average contrast measured between 1600 nm and 1640 nm and an assumed facet reflectivity of 0.27. In addition to the loss deduced from the fringe contrast, this plot also shows the loss from the normalized transmission, which is also an average over the wavelength range of 1600 nm to 1640 nm. The slope of the linear fit to this data, like the fringe contrast data, yields the loss per tether pair, which provides a check to the fringe contrast method. The y-intercept of the linear fit to the normalized transmission data is the zero-tether propagation loss plus insertion loss. Thus the difference between the y-intercepts of the linear fits in Fig. 3 is the insertion loss.

 

Fig. 3. (a): Measured TE loss through a set of 4 µm wide waveguides in sample A, as a function of the number of tether pairs in the waveguide. The y-intercept of the fringe contrast loss is the propagation loss, whereas the y-intercept of the normalized transmission loss includes both the propagation loss and insertion loss. The slope of both curves is the loss per tether pair.

Download Full Size | PDF

This loss analysis is performed on three samples, A, B, and C, each processed using slightly different etching conditions. For each sample, the analysis provides the (zero tether) propagation loss, the loss per tether pair, and the insertion loss. These parameters are measured for both TE and TM polarizations, and for waveguides 2 µm and 4 µm wide. The results for the 4 µm waveguides with 4 µm wide tethers are shown in Table 1. The error bars originate from the standard error for each parameter found by a least squares fit.

3. Discussion

In two of the samples, the measured TE propagation loss includes zero, implying that we can only put an upper bound on the loss in those cases. Nevertheless, the average TE loss for the three samples is 4.1 dB/cm, showing that for wavegeuide lengths of a few mm (which is

Table 1. Measured loss parameters of 4 µm wide waveguides from three different samples for wavelengths between 1600 nm and 1640 nm.

View Table

sufficient for a π phase shift in an electro-optic modulator) propagation loss will be negligible. However, the TM propagation loss is significantly higher than that of the TE mode. Since the TM mode has a stronger overlap with the sidewalls, this suggests that sidewall scattering may be a source of propagation loss, even for waveguides as wide as 4 µm. In addition, in waveguides with widths of 2 µm and less, the measured propagation loss is significantly higher: typically about 30 - 40 dB/cm. Such a dramatic increase in loss in narrower waveguides suggests that sidewall scattering plays a dominant role in waveguide loss at these widths [14]. Indeed, closer examination of Fig. 1 indicates a sidewall roughness possibly as high as 100 nm in some areas, which is consistent with excess scattering loss even for waveguides as wide as 2-4 µm. The relatively low loss values for the TE mode in 4 µm wide waveguides suggest that scattering due to roughness of the top and bottom waveguide surfaces is insignificant, despite the thin (590 nm) core. This is due to the ability of epitaxial growth to provide atomically smooth interfaces. The suspension of the waveguide 1.6 µmabove the substrate and the use of waveguide materials with an index larger than that of InP prevent any substrate leakage loss.

We have used two-dimensional finite element analysis to model the scattering losses at tethers. The results show, not surprisingly, that narrower tethers and wider waveguides decrease the tether loss. Also, we found that the exact amount of loss depends sensitively on the shape of the tether-waveguide joint (the so-called “fillet”), which depends on the specific details of the lithography and etching. The resulting calculated loss per tether pair for the TE mode in 4 µm wide waveguides with 4 µm wide tethers is 0.23±0.03 dB, which agrees with the average TE value for all of our samples, 0.21±0.04 dB. The calculated TE value for 2 µm wide waveguides with 2 µm wide tethers is 0.51±0.21 dB, which also agrees within error with our measured values of 0.70±0.23 dB. The larger area bars for the thinner waveguides and tethers is due to the increased sensitivity to the fillet shape in narrower structures. This amount of tether loss, especially for the wider waveguides, is not likely to have a significant impact on many of the proposed uses of these devices. For example, an application that requires a small modal cross section (e. g. less than 1 µm wide) can use adiabatically widened waveguides near the tethers. An electro-optic quantum well device may require much more densely spaced tethers to minimize contact resistance, but most of these devices can be designed to operate with wider multimode waveguides (e. g. 4-6 µm wide) for which tether loss is small.

Our coupling scheme uses a microscope objective to produce a beam waist approximately 2 µm in diameter at the input facet. A second objective is used to collect the light from the output facet. Mode matching considerations for a 4 µm wide waveguide combined with the facet reflectivity give approximately 7 dB of coupling loss at each facet, for a total insertion loss of 14 dB. This value agrees within error with all six of the measured values reported in Table 1. The measured insertion losses increase only slightly for 2 µm wide waveguides compared with 4 µm wide waveguides, consistent with the assumption that the 590 nm waveguide thickness sets the primary mode-matching limitation. Improving the mode-matching for decreased insertion loss could be accomplished using inverted taper couplers [15] or facets etched with photonic crystal patterns [16].

4. Conclusions

We have demonstrated that suspended MQW waveguides can exhibit a propagation loss that is approximately 4 dB/cm at wavelengths only 125 - 150 nm below bandgap. In addition, tether losses can be as low as 0.15 dB per tether pair, permitting their use in devices that require multiple tethers in an active area. Epitaxial tensile strain ensures that the waveguides remain flat upon release. Our propagation loss is believed to be limited by scattering due to sidewall roughness, which is expected to decrease with the use of an improved plasma etch. However, even though our waveguides are only 590 nm thick, they show no evidence of scattering losses due to top or bottom surface roughness, which is evidence of the high quality of the epitaxial surfaces.

These waveguides, when fabricated with the MQW between p-doped and n-doped layers, should exhibit extremely high electro-optic coefficients due to the strong overlap between the small optical mode and the electrical bias field. Coupled or asymmetric MQW’s [17] should increase the electro-optic coefficients even further, enabling extremely low-V_π modulators in lengths of a mm or less. Forward biased p-i-n MQW’s with band edges near 1550 nm could be used as electroabsorptive modulators, lasers, or amplifiers that could be integrated with suspended waveguide switches or sensors. In addition, since the waveguides can be surrounded by a low-index chemoselective coating after the sacrificial etch, the waveguides can act as chemical sensors.