1. Introduction

The intersubband transition (ISBT) in GaN/AlN multiple quantum wells (MQWs) has been of great interest for optical devices, such as detectors and switches that operate at optical communication wavelengths, since the conduction-band offset between GaN and AlN is sufficiently large and they can provide a widely tunable wavelength range with a simple quantum well (QW) structure [1, 2]. In addition, owing to the strong interaction with longitudinal optical phonons, the absorption recovery time is extremely short for the ISBT in GaN/AlN MQWs. It is therefore anticipated that ultra-fast all-optical switches could be realized that can be operate at a bit rate of 1 Tbps or higher.

Since Suzuki et al. proposed applying GaN ISBTs to all-optical switches [3], there have been many achievements in this field of research. Iizuka et al. verified ultrafast absorption recovery at a wavelength of 4.6 μm [4]. The first ISBT at optical communication wavelengths was reported by Ng et al. in 2000 [5]. The characteristics of the absorption recovery were investigated by several groups [6–8]. Ultrafast all-optical modulation was demonstrated with an extinction ratio of 2.4 dB by the authors in 2004 [9], and a gate switch with an extinction ratio of more than 10 dB was reported in 2005 [10]. Absorption saturation at a wavelength as short as 1.43 μm was reported for a waveguide with AlN cladding layers [11, 12]

At present, lowering the operation energy is one of key issues for the application of ISBT in GaN/AlN MQWs to practical devices. Li et al., proposed a novel waveguide design to reduce the excess transmission loss with the purpose of decreasing in the switching energy [13]. Another effective way for reducing the operation energy would be to reduce the cross section of the waveguide. To establish a high coupling efficiency between the input fiber and the narrow waveguide, it is necessary to employ a spot-size conversion (SSC).

In this study, SSC structures and their integration with the GaN/AlN ISBT switch were investigated. There are edge dislocations in the nitride semiconductors grown on a sapphire substrate, and the edge dislocations cause a propagation loss for transverse magnetic (TM) -polarized light [14]. In general, the edge dislocation density is higher near the substrate. Then, the downsizing the waveguide thickness for the SSC may lead to the increase in the excess propagation loss due to the higher dislocation density. On the other hand, SiN is proved to be a low-loss material at optical communication wavelengths [15]. Therefore, in this study, SiN instead of AlN was used as an upper cladding.

The SSC consists of the horizontal conversion and the vertical conversion. The horizontal conversion seems to be achieved by a simple lithography technology. Then, designing the vertical conversion has been a key for an efficient SSC. In this study, two types of SSC were proposed and examined. First, stair-like SSC were proposed and fabricated with an ISBT switch. With this structure, the absorption saturation due to the ISBT was verified and characterized. Then, for further reduction of the conversion loss, a triangularly-tapered structure was proposed and the light propagation through the SSC was simulated. Finally, the SiN/AlN waveguides with the triangularly-tapered SSC were fabricated and the low-loss coupling was verified.

2. Simulation of a tapered spot-size converter

First, tapered SSCs were examined with three dimensional finite-difference time-domain (3D-FDTD) method. In the simulation, the waveguides were assumed to consist of AlN (n = 2.03 [16]) and SiN (n = 1.85) on a sapphire substrate (n = 1.74). The refractive index of SiN was determined by the separate experiment. In Fig. 1 , the calculated coupling efficiencies of the SSCs with TM-polarized light are shown as a function of the taper length. The simulations were performed for three-step stair-like SSCs as well as smooth-slope SSCs, because sophisticated technique will be necessary for fabrication of the smooth slope. In the simulations, input facet dimensions of 3.5 (width) x 2.2 (height) μm² were investigated, with input beam diameters of 3 μm. The cross section of the narrow waveguides was 0.8 x 1.2 μm². The results are plotted as a function of the taper length. For the slope-type SSC, the coupling loss is less than 1 dB even when the taper length is as short as 10 μm. The loss is almost unchanged for taper length, beam diameter and cross section of the waveguide (dependency on the beam diameter (1.6 – 3.2 μm) and the cross section of the input facet (2.4 x 2.4 – 3.2 x 3.2 μm²) are not shown in the figure), which indicates that the error tolerance of this structure is excellent. For the stair-like SSC, on the contrary, the coupling loss varies with the taper length. This dependency on the taper length is caused by the reflection at the stairs.

 

Fig. 1 Simulation results for the coupling efficiency. The waveguide consisted of AlN (n = 2.03) and SiN (n = 1.85) on a sapphire substrate (n = 1.74). The cross section of the input facet is 3.5 x.2.2 μm². The beam diameter is 3 μm. (a) Comparison for smooth slopes and step-like slopes. The cross section of narrow waveguide is 0.8 x 1.2 μm². (b) Simulation for the actual fabricated device.

Download Full Size | PDF

3. Fabrication of an ISBT switch with a stair-like tapered SSC

As shown in Fig. 1, the coupling efficiency for the slope-type SSC is better than that of the stair-type SSC. To realize such a smooth-slope structure, however, would require rather sophisticated fabrication techniques. On the other hand, the stair-like SSC could be achievable with simpler fabrication and the coupling efficiency is still expected to be more than 60%. Therefore, an ISBT switch was fabricated with the stair-like SSC to examine the applicability of the structure to an actual device. In addition, Si₃N₄ was utilized for a cladding layer with the following anticipated advantages: (1) quality of the dry etching will be better than that for AlN so that smooth sidewall is expected, and (2) there is no transmission loss caused by defects, whereas the edge dislocations in AlN layer lead to the loss. The device structure is schematically shown in Fig. 2 .

 

Fig. 2 Schematic view of the fabricated device. (a) Top view. (b) Side view.

Download Full Size | PDF

After the growth of 0.4-μm thick AlN on a sapphire substrate by metalorganic vapor phase epitaxy (MOVPE), 100-nm-thick GaN, MQWs, and 100-nm-thick GaN were grown with molecular beam epitaxy (MBE). The MQWs consisted of two GaN (1.5 nm) /AlN (1.5 nm) QWs with Si doping in the wells of 2 x 10²⁰ cm⁻³. The GaN and MQW layers were in part etched off, which formed the SSC and the outer waveguide. Then, the semiconductor layers were covered by 1.8-μm-thick Si₃N₄ deposited by sputtering method. The use of SiN made it possible to reform the waveguide rather than re-growing a thick AlN layer. This is a further advantage for using SiN as an upper cladding. The refractive index of the Si₃N₄ layer was 1.85 for a wavelength of 1.55 μm. Vertical SSC was realized by etching of the Si₃N₄ layer three times, which led to stair-like shape. Then, horizontal tapers were formed by etching the SiN layer and the semiconductor layers. Finally the wafer was thinned by polishing and cleaved to chips.

The lengths of the switches were 104 - 424 μm and the lengths of the tapers and the total device lengths were 36 and 1050 μm, respectively. The simulation indicated that a length of 8 μm would be sufficient for a good coupling. It also suggests, however, that the coupling become drastically worse as the taper length is shorter. Then, for the first fabrication, a long taper length was adopted in expectation of a sufficiently safe margin. The width of the input and output facets was 3.5 μm. Figure 3 shows images of scanning electron microscopy (SEM) for the switch. The cross section is shown in Fig. 3(a). The width was 2 μm at the top layer and 2.8 μm at the MQWs. Because the processing technique was not well-matured, the waveguide width was unintentionally wider than designed one (1.6 μm), and narrower waveguides could not be fabricated well. The improvement of the fabrication processing is a subject to be resolved. The total thickness was approximately 1.2 μm. Figure 3(b) shows a top view of the switch. It can be observed that the sidewall roughness is smaller at the SiN layer than at the AlN layer.

 

Fig. 3 SEM images for the ISBT switch device. (a) Cross section of the switch. (b) Top view of the SSC.

Download Full Size | PDF

The devices were characterized by measurements of the insertion loss, absorption spectrum and absorption saturation. The spectra were measured with a wavelength range from 1.3 to 1.75 μm using a super continuum light that was produced with a femtosecond fiber laser and a photonic crystal fiber. The absorption saturation was characterized with an optical parametric oscillator excited by a mode-locked Ti:sapphire laser with a repetition rate of 80 MHz. The pulse width was nominally 130 fs. The polarization of the pulses were controlled with a polarizer and a wide-band waveplate. The light was input to the sample with a polarization-maintaining dispersion-shifted fiber.

Figure 4(a) shows the polarization dependent spectrum, that is, the ratio of transmittance for TM-polarization to that of transverse electric (TE) polarization for the sample with a switch length of 104 μm. The fringe oscillations observed in the trace of Fig. 4(a) were a result of the reflections at the step interfaces of the SSC. A Gaussian fit to this trace indicates an absorption peak wavelength of 1.48 μm and a full width at half maximum (FWHM) of 110 meV. These results verify that large and narrow ISBT absorption spectrum can be realized for a waveguide as short as 100 μm.

 

Fig. 4 (a) The ratio of the transmittance spectra for TM-polarization to that for TE-polarization for the sample with the length of 104 μm. The blue line represents the Gaussian fit. (b) Characteristics of the absorption saturation. Saturation by 5 dB was achieved with a pulse energy of 25 pJ.

Download Full Size | PDF

Since there is no ISBT for TE polarization it was possible to estimate the insertion loss for the device from the losses for TE polarization, which was found to be in the range of 12-13 dB. The dependence of the insertion loss on the length of the switch indicated that the loss was approximately 3 dB/100 μm for the narrow waveguide. Consequently, the loss is 9-10 dB except the marrow waveguide. The causes for the loss may be attributable to the coupling loss at the stair-like SSC. To estimate the coupling loss for the fabricated structure, simulations were performed for TM- and TE-polarized light as shown in Fig. 1(b). According to the results, the coupling efficiency is high for both polarizations. Therefore, the reason for the high insertion loss is unclear at present and it is a subject that needs further investigation.

Absorption saturation measurements were carried out at a wavelength of 1.48 μm that corresponding to the absorption peak wavelength of the device. As shown in Fig. 4(b), absorption saturation of 5 dB was achieved for an input pulse energy of 25 pJ. The pulse energy for a saturation of 5 dB is the same as that of a previously reported device [17], where the device structure comprised an AlN lower cladding layer, MQWs and an AlN upper cladding layer. The insertion loss of this device was 7-8 dB. For the present device, on the other hand, the MQWs were in the middle of GaN guiding layer placed between an AlN lower cladding and a SiN upper cladding. Considering that the insertion loss was worse than that of the previous sample, the lower saturation energy was caused in part by better optical confinement in the switch. This strongly suggests the applicability of SiN as an upper cladding.

The device structure has two advantages from the viewpoint of applications. Firstly, the switch is integrated with passive waveguides that mainly consist of SiN. Since the MQWs are removed, the optical loss should be very low in the passive waveguide regions. It will therefore be possible to fabricate additional integrated components such as splitters, filters and reflectors to expand the functionality of this device. Thus, the results shown in Fig. 4 open the way for integration of the ISBT switch with functional waveguide devices. Secondly, a switch as short as 100 μm could be realized owing to the integration with waveguides. So far, ISBT switches have been long due to handling requirements during the experiments and to facilitate the formation of facets by cleaving. Due to the integration process, however, these issues were overcome. By increasing the number of the wells, the switch can be further miniaturized. Excess losses caused by defects in the crystal or by surface roughness of the switch can be reduced for a shorter switch, which leads to further decrease in the operation energy.

Although integration of the ISBT switch was achieved, the large insertion loss is an issue that needs to be overcome. The major causes for the loss are poor quality of the sidewall and low coupling efficiency at SSCs. The sidewall roughness could significantly improved thorough optimization of the etching process used. The improvement of the coupling efficiency is discussed in the next section.

4. Improvement of SSC

The structure of the stair-like SSC was a little troublesome due to the multiple etches required to make the steps. An alternative SSC structure with a simpler fabrication process was considered. This triangularly-tapered structure is schematically shown in Fig. 5(a) . Simulations with three-dimensional beam propagation method (3-D BPM) suggested that a coupling loss of less than 1 dB could be expected.

 

Fig. 5 (a) Schematics of the devices structure designed for the fabrication. SEM images ware shown (b) for bird’s eye view of the SSC and (c) for cross section of the narrow waveguide.

Download Full Size | PDF

Waveguides were fabricated for the low-loss triangularly-tapered SSC. Considering the application to the ISBT switch with SiN upper cladding, the narrow waveguide consisted of SiN and AlN. However, to experimentally confirm that the coupling loss is low for TM-polarization as expected by the simulation, the MQWs were not grown in this experiment because the MQWs absorb the TM-polarized light due to the ISBT [14]. First, 0.6-μm-thick AlN was grown on a sapphire substrate by MOVPE, and SiN was deposited by sputtering method with a thickness of 1.6 μm. By etching the SiN layer by 1 μm, the upper taper was formed. The lower-taper waveguide was fabricated by etching SiN and AlN successively. In the fabrication, the condition of the photo-lithography was retuned from the previous fabrication. In addition, a wet-etching technique was applied as well as a conventional dry-etching for AlN. After the dry-etching of AlN, the device was soaked in thin alkaline solution. By the treatment, the side-wall roughened by the dry-etching could be smoothened to some extent.

A schematic view of the designed structure is shown in Fig. 5(a), and SEM views of the fabricated device are shown in Fig. 5(b) and (c) for top view of the SSC and for cross-section of the narrow waveguide, respectively. Figure 5(b) indicates that the upper taper was well aligned with the lower waveguide. However, the tip of the taper was not as triangular as the designed atructure due to limitations of the photo-lithography. By optimizing the etching process for the waveguide, the roughness of the sidewall was reduced. As shown in Fig. 5(c), the main part of the waveguide was narrower (1.7 μm) than that of the previous one (2.8 μm). The typical insertion loss was as low as ~5 dB for TM polarization. The dependence of the insertion loss on the narrow waveguide length indicated that the propagation loss was approximately 0.75 dB/μm. The coupling loss was estimated to be ~2 dB/facet, which is small as expected.

5. Summary and conclusion

Structures of spot-size converters and their integration with GaN/AlN ISBT switches were studied with the purpose of reducing the operation energy by improving the coupling efficiency between the input fiber and the switch. SiN/AlN waveguide structure was also examined for low propagation loss. With a stair-like SSC, the absorption saturation of 5dB was achieved with a pulse energy of 25 pJ for a switch length of as short as 100 μm. Moreover, to establish a simpler structure and a better coupling efficiency, triangular-shaped converters were investigated. SiN/AlN waveguide with the triangular-shaped converter was fabricated and the coupling loss was as low as 2 dB/facet. These results led us to the conclusion that the switching power of GaN/AlN ISBT switches can be reduced by utilizing SSC, and, furthermore, the switches can be integrated with functional devices employing SiN/AlN waveguide.