1. Introduction

Planar waveguide spectrometers, including waveguide echelle gratings and arrayed waveguide gratings (AWGs) [1], are essential components in wavelength division multiplexed (WDM) communication networks. Although integrated spectrometer devices for a wider range of applications have been demonstrated [1–12], these devices often have a limited spectral resolution. A compact, monolithic optical microspectrometer capable of the high-resolution required for Raman or infra-red absorption spectroscopy would be a significant advance in the field. Such devices would find applications in chemical and biological sensing, genomics, optical metrology, space-born environmental sensing from micro- and nano-satellite platforms, and medical instrumentation.

The spectral resolution of any grating spectrometer may be enhanced either by increasing the interference order of the grating and/or the number of grating teeth (or waveguides in case of an AWG), and by narrowing the input and output slits (apertures) of the spectrometer if the grating resolution limit is not yet reached. In AWG waveguide spectrometers, the grating order may be increased simply by enlarging the waveguide array section, or by using waveguides with an extremely high group index [13]. However, increasing the grating order alone also reduces the free spectral range and therefore limits the usable bandwidth of the spectrometer. When maintaining small device size and large operating bandwidth are critical, it is advantageous to maximize the resolution by narrowing the input and output apertures. In an AWG spectrometer, the waveguide mode size at the input and output Rowland circles plays the analogous role of the slit width in a bulk optic spectrometer. In glass waveguides, this mode size is limited to a minimum of several micrometers, and mode coupling between closely spaced waveguides also limits the possible wavelength channel density. On the other hand, the high index contrast of silicon-on-insulator (SOI) also makes it possible to confine and engineer the propagation of light at the sub-micrometer scale to achieve novel functionality and enhanced performance. In this paper, we describe the design, fabrication and characterization of a compact AWG spectrometer that takes advantage of the strong optical confinement of SOI to achieve high spectral resolution and channel density.

The large refractive index difference between the Si waveguide core (n ~ 3.5) and the SiO₂ cladding (n ~ 1.45) allows light confinement in waveguides with cross-sections and bend radii two orders of magnitude smaller compared to conventional low index contrast glass waveguides, resulting in markedly reduced device size [14–17]. The first silicon AWG spectrometer was reported by Trinh et al. [18], but this device was still constrained to be relatively large (2.7 × 2.7 cm²) to avoid the polarization dependent spectral response encountered in smaller devices [19]. More recently, the use of stress engineering to manipulate SOI waveguide birefringence [20, 21] has made it possible to take full advantage of the high refractive index of silicon, and fabricate SOI AWG devices that are a few square millimeters in area [22–25], and polarization independent [26].

Further reduction of scale can be achieved using silicon wire technology. Ultra-compact AWGs with silicon wire waveguides have recently been reported [27–29]. One can expect that similar devices may be used in practical applications in the near future as the outstanding issues such as the quality of sub-micron silicon waveguides particularly in terms of the sidewall roughness, strong polarization dependence, and a limited efficiency of light coupling to and from photonics wires, are being resolved [1]. Though silicon wire AWG devices reported up to present have a rather limited number of sparsely spaced wavelength channels, the silicon wire technology appears particularly amenable for making ultra-compact spectrometers with a high wavelength resolution and a large number of channels.

2. Design

The high-resolution SOI AWG spectrometer described in this paper was designed following the methodology by Smit and van Dam [30]. Here we outline the main steps: First, the receiver waveguide width w_r and the spacing d_r at the output Rowland circle are chosen. In our design, w_r = 0.6 μm and d_r = 1 μm, resulting in densely arrayed waveguides yet with a negligible receiver waveguide crosstalk. Fabrication related aspects, including the waveguide width and gap aspect ratios, were also considered when choosing w_r and d_r values. From the maximum loss inhomogeneity L_roll-off = -10log(exp(-2θ_m ²/θ₀ ²)) of the outer receiver channel, the maximum acceptable dispersion angle θ_m is calculated, where θ₀ is the angular half-width of the equivalent Gaussian far field of an arrayed waveguide of width w_a at the array-combiner interface. In our design, w_a = 1.2 μm, the arrayed waveguide pitch is d_a = 1.6 μm, yielding 0.4 μm gaps between the waveguides at the array-combiner interface. We chose this geometry as a compromise between a large waveguide packing density, a low excess loss at the array-combiner interface, and fabrication constraints. The array waveguides are adiabatically tapered to their nominal width of 1.5 μm throughout most of the array length. The length of the combiner region is found from the maximum dispersion angle as f= s_m/θ_m, where s_m is the coordinate of the outer (marginal) receiver waveguide along the focal curve (Rowland circle). The divergence angle between the arrayed waveguides is thus Δα = d_a/f and the linear dispersion of the array is found as D = ds/dλ = d/Δλ, where Δλ. is the wavelength channel spacing. The length increment ΔL between the adjacent waveguides is then calculated from the AWG dispersion formula:

where: n_g is the group index of the arrayed waveguides, n_s is the effective index of the combiner slab waveguide, and λ_c is the spectrometer center wavelength. The angular half-width θ_a of the array aperture is chosen for a specific level of the field truncation crosstalk and loss penalty that in our design are 25 dB and 1.5 dB, respectively. Finally, the number of arrayed waveguides is obtained as N = 2θ_af/d_a + 1, which in our design yields 125 waveguides. This design was further verified using a commercial AWG design software (WDM Phasar from Optiwave Corporation), which was used to generate the device layout.

 

Fig. 1. (a). An optical microphotograph of a 100-channel SOI AWG spectrometer. (b). SEM images of narrow deep-etched rectangular waveguides joining a slab waveguide combiner at the Rowland circle (bottom), and two adjacent deeply etched 1.5 × 0.6 μm² waveguide apertures (top). Some waveguides (bottom) were intentionally removed to evaluate crosstalk.

Download Full Size | PDF

The design outlined above results in a spectrometer with 50 channels spaced at 0.2 nm (25 GHz at λ = 1.55 μm). The device uses an SOI layer structure with a 1.5 μm thick silicon waveguide layer, a 1 μm thick buried oxide (Box), and has a total chip size after dicing of 8 × 8 mm² (Fig. 1). The spectrometer comprises a waveguide grating array of order m = 160 consisting of N = 125 waveguides of nominal width 1.5 μm. The free spectral range is approximately 10 nm. This waveguide grating design gives a theoretical spectral resolving power of R = λ/Δλ = mN = 20,000, thus Δλ = Δ/R ~ 0.08 nm, where Δλ is the wavelength resolution according to the Rayleigh criterion.

The input and output waveguides and the array grating waveguides are conventional partially etched 1.5 μm wide ridge waveguides with a nominal etch depth of 0.9 μm. However, as the input and output waveguides approach the combiner Rowland circles, each ridge waveguide is adiabatically transformed to a rectangular 1.5 μm × 0.6 μm channel waveguide deeply etched down to the buried oxide, as shown in Fig. 1(b). One such channel waveguide forms the aperture or slit that couples light into the input combiner. The corresponding multiple images of this aperture, which are spatially dispersed in the output combiner according to wavelength, are intercepted by identical slit waveguides densely arrayed along the output focal curve. Both the wavelength resolution and channel density scale with the width of these channel waveguides. Since the width of the mode launched into the input combiner and its corresponding diffracted image are minimized in this design, the maximum receiver waveguide packing density is obtained at the focal region (Rowland circle). The narrow input mode width produces a comparatively large divergence angle in the combiner. As a result, the combiner focal length can be quite short (in our design, f= 166 μm) for a given waveguide array aperture. The use of rectangular channel waveguides has the added advantage that it suppresses excitation of higher order vertical slab modes at the junction of the input waveguides and the slab waveguide combiner region, which would otherwise degrade the AWG crosstalk. This is because rectangular waveguides have a symmetric vertical mode profile closely matching that of the fundamental slab mode, unlike ridge waveguides whose mode is asymmetric with respect to the slab waveguide plane.

In the present design, the 0.6 μm wide channel waveguides are arrayed with a 1 μm pitch along the Rowland circle, and the gaps between the waveguide cores are 0.4 μm (Fig. 1(b), top). The ridge waveguides are connected to these deeply etched channel waveguides through a 100 μm long two level adiabatic taper structure (Fig. 2). This taper structure is an essential design element of this device, and, as discussed above, it is designed to maximize coupling of light from the ridge waveguide to the fundamental mode of the channel waveguide while suppressing excitation of the higher order modes of the channel waveguide, which is inherently multimode. Detailed simulations and experimental characterization results for this taper structure are reported elsewhere [25, 31], but show that the fundamental-to-fundamental mode coupling loss is better than -0.4 dB.

The Rowland circle radius is 83 μm, yielding the combiner length of 166 μm. In the layout of the fabricated spectrometer, a small number of selected waveguides were deleted (see Fig. 1(b), bottom) to allow the experimental assessment of coupling between adjacent waveguides near the Rowland circle.

3. Fabrication

Devices were fabricated on SOI wafers from Soitec, with 1.5 μm Si and 1 μm buried oxide layers, using a self-aligned two-step electron beam patterning and plasma etching process. The first step (shallow etch patterning) defines the overall device layout except for the two level ridge-to-rectangular waveguide tapers. A 100 nm thick SiO₂ hardmask is deposited by Plasma Enhanced Chemical Vapour Deposition (PECVD). The pattern is written by e-beam lithography in ZEP 520A resist and transferred into the hardmask with an Inductively Coupled Plasma (ICP) etch process using Ar and C₄F₈ as etch gases. After stripping the resist the pattern is transferred into the Si with a cryogenic ICP etch, using SF₆:O₂ chemistry at an etch rate of ~1.2 μm/min and a substrate temperature of -120°C. The etch depth was 870 nm, as measured by profilometry after the etch. The second patterning step uses a 1 μm thick polymethylmethacrylate (PMMA) e-beam resist to define the ridge-to-rectangular waveguide mode converters and deep etched regions near the input and output Rowland circles. The PMMA is removed in the deep-etched regions, including the surface area of the waveguides in this region, which are still masked by the original SiO₂ hard mask during the second etch step. The deep etch pattern is thus self-aligned with the first (shallow) etch. The same cryogenic ICP etching process is used for the deep etch, which self-terminates at the buried oxide layer. Finally, the PMMA resist is stripped from the sample, which is then coated with a 1 μm thick SiO₂ cladding by PECVD, and the samples are cleaved and the input and output edges are polished for the optical measurements, yielding a device shown in Fig. 1. The second cryogenic etching process on the already patterned substrate results in minor waveguide sidewall damage in the deep etched regions. Although some further optimization is required, the existing process still produced viable devices.

 

Fig. 2. (a). Schematic of an adiabatic ridge-to-slit waveguide mode transformer. (b). A SEM view of the fabricated mode transformer.

Download Full Size | PDF

4. Experimental results and discussion

The spectrometer optical performance was measured by coupling polarized light from a tuneable external cavity semiconductor laser via a half-wave plate polarization rotator and a polarization maintaining fiber terminated with a conical tip, into one of the AWG input waveguides. The output signal was collected from each output channel waveguide by a microscope objective and focussed onto a photodetector, and the wavelength spectrum was recorded.

The advantage of using deeply etched rectangular slit waveguides near the Rowland circle is demonstrated in Fig. 3 which shows several channel spectra for two fabricated spectrometers, for one of which the deep etch step was omitted. Spectra A are for a spectrometer with the deep etched aperture waveguides as in Fig 1(b). Spectra B and C are for a spectrometer with only partially etched ridge waveguides near the Rowland circles. Spectrum B is for an isolated channel with no nearest neighbours, while the spectrum C is for a waveguide with the same channel spacing as the spectrometer channel spectra A, and shows the effect of coupling between adjacent waveguides. The spectra in Fig. 3 demonstrate that for a given waveguide grating, the deeply etched channel waveguide apertures improve resolution by more than a factor of three, and significantly reduce adjacent channel crosstalk.

 

Fig. 3 Measured microspectrometer channel spectra. (A) three adjacent channel spectra for input and output waveguide apertures etched down to the buried oxide layer [Fig.1(b), top), and (B) and (C) a single channel spectrum for a device with only the shallow ridge etch. (B): spectrum for an isolated channel with no nearest neighbours. (C): spectrum for waveguides with same waveguide pitch (1 μm) at the Rowland circle as the microspectrometer with channel spectra in (A).

Download Full Size | PDF

The measured spectra for 100 spectrometer channels with the slit waveguides are shown in Fig. 4 for TM polarized light. These 100 channels actually span two free spectral ranges of 20 nm total bandwidth. The crosstalk is 10 dB or better. The measured channel spacing is 0.2 nm, with a 3 dB channel bandwidth of 0.15 nm that should allow spectral lines separated by ~ 0.1 nm to be resolved, which is close to the theoretical Rayleigh resolution limit of 0.08 nm. The dashed lines indicate the positions of output channels that were intentionally omitted to evaluate the crosstalk between the receiver waveguides. The measured channel inhomogeneity (roll-off) within the 10 nm free spectral range (50 channels) is 5 dB. The insertion loss, including fiber to waveguide coupling, is -17 dB for the channels near the central wavelength λ ~ 1.45 μm. Since the loss of the individual ridge-to-rectangular waveguide mode transformer was measured to be better than -0.4 dB, we believe the waveguide scattering loss in the ridge waveguides is the largest contribution to the insertion loss, and also the probable cause of the relatively high cross-talk. For TE polarization, the loss and crosstalk were significantly worse than for TM light, again suggesting influence of the scattering by sidewall roughness. We expect that both insertion loss and cross-talk can be improved by further optimization of our fabrication process to reduce sidewall roughness. This could be achieved by using a different etch chemistry for the deep etch step to avoid sidewall damage and by introducing a sidewall smoothing step [32] to reduce the roughness of the ridge waveguides.

Although the spectral resolution of a spectrometer can be increased by increasing the grating order or number of grating elements (teeth or waveguides), this spectrometer design achieves the maximum spectral resolution for the given grating dispersion and at the same time minimizes device size, particularly the combiner length. A further reduction of the input waveguide width below the present 0.6 μm may be impractical because the increasing angular width of the beam in the input slab waveguide requires combiners with low f-numbers for which aberrations become difficult to control. The fundamental lower limit to the waveguide width is due to the mode delocalization effect. When the waveguide width is reduced below a critical value, the mode starts expanding from the core into the cladding. This demands increasing the separation between the neighbouring waveguides at the Rowland circle to avoid crosstalk between the adjacent spectral channels. According to our calculations [33] the minimum waveguide separation d along the focal curve in order to achieve a crosstalk level of less than 40 dB is:

where Δn is the refractive index contrast. For SOI waveguides (Δn ~ 2), d ~ 0.75 μm for λ ~ 1.5 μm

 

Fig. 4. Measured spectra for the SOI AWG spectrometer. The 100 channels span two free spectral ranges of total bandwidth 20 nm. The dashed lines indicate the position of output channels that were intentionally omitted to evaluate the crosstalk.

Download Full Size | PDF

5. Conclusion

We have described the design, fabrication, and characterization of a 50-channel SOI spectrometer that employs high aspect ratio apertures at the Rowland circle to achieve a spectral resolution of ~ 0.1 nm and 0.2 nm channel spacing. The overall device size is 8 × 8 mm², which is two orders of magnitude smaller than previously reported for comparable silica AWGs [1, 34]. The resolution of our spectrometer is the highest reported to date for an SOI AWG. The narrow waveguide aperture spectrometer design imposes no fundamental constraint on the Si waveguide thickness, and hence may be adapted to relatively thick Si waveguides to improve coupling to optical fibers, or to Si photonic wire devices. It may even be extended to low index contrast platforms (e.g. silica-on-silica) by etching air trenches [35] to increase the mode confinement near the Rowland circles. Although the present spectrometer is designed for high resolution over a relatively small spectral range, the suggested approach of using a dense array of sub-micrometer waveguide apertures can easily be adopted for other configurations, for example to fabricate high channel number spectrometers in SOI covering a spectral range as large as several hundred nanometres. Such devices are of interest for numerous applications in spectroscopy and sensing.