We have developed a compact gate switch with monolithic integration of all-optical cross-phase modulation (XPM) in a Mach-Zehnder interferometer (MZI). XPM is caused by intersubband transition (ISBT) in InGaAs/AlAsSb coupled double quantum wells (CDQWs) by area-selective silicon ion implantation and rapid thermal annealing (RTA). While injecting pump light through a transverse electric/transverse magnetic (TE/TM) beam combiner, XPM is induced in one MZI arm and gating operation can be realized. The RTA condition is optimized, and the sample is annealed at 780 °C for 8 s with an implantation dose of 5 × 1013 cm–2. Dependence of XPM efficiency on the length of the implanted mesa is also analyzed, and there exists an optimum implantation length to fulfill both high efficiency of ISBT modulation and low loss of the probe and pump signals.
© 2012 OSA
Ultrafast all-optical phase modulation has received much attention due to its key role in high-speed optical signal processing. Cross-phase modulation (XPM) in doped InGaAs/AlAsSb coupled double quantum wells (CDQWs), which was discovered recently, originates from the intersubband transition (ISBT) induced interband dispersion [1,2]. Different from silica-based planar lightwave circuits (thermal-optic) and LiNbO3 phase modulators (electro-optic), this XPM works in an all-optical manner. When the transverse magnetic (TM) pump light excites ISBT, phase modulation is induced in the transverse electric (TE) probe light, which is immune to ISB absorption. This ISBT-induced XPM has a short relaxation time of a few picoseconds , and it is free from the pattern effect beyond 100-Gb/s operation . It also has advantages such as easy integration and device miniaturization, high stability, and low switching power.
A Mach-Zehnder interferometer (MZI) gate switch constructed from bulk optical components in a free-space layout has been reported with an ISBT waveguide chip for XPM placed in one MZI arm . It has been used for all-optical signal processing experiments such as demultiplexing (DEMUX) from 160 to 40 Gb/s , wavelength conversion at 160 Gb/s , and NRZ-RZ conversion at 40 GHz . However, this MZI gate switch suffers from mechanical instability of the interferometer due to environmental changes and the inconvenience of large-scale integration. A monolithic integration of an ISBT waveguide with a Michelson interferometer (MI) on an InP chip has also been developed, and full switching of the π-rad nonlinear phase shift was achieved with a pump pulse energy of 8.6 pJ at a 10-GHz repetition rate and an extinction ratio of 36 dB . This device was successfully applied to the DEMUX of a 172-Gb/s optical time domain multiplexing (OTDM) signal delivering real-time ultra-high definition video signals . However, in this monolithic device, TE probe light suffers high signal loss due to the doped wafer with an interband absorption edge designed to be near the working wavelength of 1550 nm for high XPM efficiency  and the half-reflection coating at the pump input facet. Moreover, this method is difficult to fulfill other multi-functional integrated devices. ISBT-induced-XPM can also be achieved through silicon (Si) ion implantation into undoped CDQWs. Conventionally, ion implantation and subsequent rapid thermal annealing (RTA) are postgrowth techniques for achieving quantum well intermixing in order to realize multiple bandgaps . The generation of ISB absorption through Si+ ion implantation and RTA has already been reported [12,13]. In fact, we can achieve monolithic integration using an MZI structure with only the phase modulation area implanted. In the InGaAs/AlAsSb CDQW without ion implantation, light from both TE and TM has low propagation loss because it contains no electrons and the interband absorption edge can be tailored away from the working wavelength. By using ion implantation and subsequent RTA to activate the carriers, area-selective ISB absorption can be realized. However, RTA optimization has not yet been studied. Moreover, a monolithically integrated ISBT-MZI gate switch has not been demonstrated as yet.
In the present work, we demonstrate for the first time a compact ISBT-MZI gate switch on an InP chip using area-selective Si+ ion implantation and RTA. RTA optimization is introduced first in order to obtain the lowest propagation loss of TE probe signal and the highest XPM efficiency for TM pump signal. Dependence of XPM efficiency on implantation length is also analyzed. An optimum length is selected in consideration of both high modulation efficiency and low propagation loss. Then, the design and fabrication of a gate switch is shown. Performance of the gate switch is also presented and discussed. Finally, some conclusions are derived.
2. Rapid thermal annealing optimization
Intersubband absorption can be generated through Si+ ion implantation and RTA. RTA can remove defects, which trap electrons, produced during ion implantation . A high RTA temperature can facilitate carrier activation, and it exhibits a threshold and saturation temperature dependence . ISB absorption can only be observed when the defects are reduced to a sufficiently low level. However, RTA can cause a structural change in the CDQWs. Moreover, when temperature increases, the InP capping layer slightly evaporates, and the wafer surface becomes rough. Further, although SiO2 can be used to protect the surface, roughness increases the loss due to light scattering. Therefore, there is an optimum RTA temperature that results in low signal propagation loss and high XPM efficiency.
For RTA optimization, the wafer was first grown on an InP substrate by molecular beam epitaxy with 60 cycles of undoped CDQWs and a 50-nm InP capping layer . Si+ ions at a dose of 5 × 1013 cm–2 were implanted at an energy of 250 keV and a 7° tilt angle at room temperature. The samples were covered with a 60-nm SiO2 protective layer before undergoing different RTA conditions. Deep etched mesa structures, 1 mm long and ~2.5 μm wide, were fabricated by standard photolithography processing techniques. The photoresist waveguide pattern was first defined by a mask aligner. Then, after hard baking, the sample with a photoresist pattern was etched directly by inductively coupled plasma (ICP) dry etching using a mixture of Ar and BCl3 working gases. In order to coincide with the final switch fabrication, the waveguide was etched to a large depth of 1.2 μm. Then, the remaining photoresist was removed by O2 plasma ashing, and finally, the sample was covered with benzocyclobutene (BCB) polymer. The cross-sectional scanning electron microscope (SEM) image of the fabricated waveguide (with CDQWs layer slightly etched by a mixture of oxalic acid and hydrogen peroxide for clear visualization) is shown in Fig. 1(a) .
The measured transmittance for TE polarization at a wavelength of 1545 nm with different RTA temperatures (at an annealing time of 10 s) is shown in Fig. 1(b). The transmittance includes the coupling loss from the fiber to waveguide and a 3-dB selective polarization loss. It can be seen that when the RTA temperature reaches 780 °C, the mesa has minimum signal propagation loss. A further increase in temperature causes more structural damage, with the CDQWs gradually evolving into single quantum wells, which deteriorates the performance. Next, we tested different annealing time with the temperature fixed at 780 °C, and the optimum value was about 8 s.
After determining the optimum RTA condition for an implantation dose of 5 × 1013 cm–2, XPM efficiency for different RTA conditions was measured. The optimum RTA condition corresponded to the highest XPM efficiency, as well as the lowest TM transmission loss. Under the optimum RTA condition, the transmittance and XPM efficiency for different mesa lengths were also measured, and the results are shown in Figs. 2(a) and 2(b), respectively. The absorption was very strong, and the output signal was extremely weak, making it difficult to obtain the exact propagation loss for the implanted sample before RTA. Implantation-induced defects could be recovered through RTA. The TM propagation loss was 18.5 dB/mm, and TE loss was 12.4 dB/mm. For an as-grown sample without implantation, RTA caused a blue-shift in the absorption bandgap which improved the propagation property slightly. After RTA for an as-grown sample with the same fabrication process, it showed a loss of 1.1 dB/mm for TM and 1.9 dB/mm for TE (transmittance figure is not presented here). Propagation loss increased excessively in the implanted sample because of the implantation-induced defects, and RTA could not recover the defects completely. TM polarization loss increased more than that of TE polarization due to carrier activation and corresponding ISB absorption.
As for the XPM efficiency measurement, a system using a delay-line interferometer (DLI) with a 25-ps delay time was applied to convert phase shift to intensity variation . Phase bias of the DLI was adjusted to π/2, and the phase shift was deduced from Δφ = sin–1(ΔI/I), where I and ΔI were, respectively, the corresponding intensity and its variation in the time domain recorded by an optical sampling oscilloscope. Considering the propagation loss of TM polarization, the theoretical modulation phase Δφ could be calculated by integrating over the entire mesa length. The simulated and measured XPM efficiency for different mesa lengths can be seen in Fig. 2(b). The dots are experimental measured data, while the solid line is the fitted curve; these coincide well with each other. It is clearly shown that there exists a saturated mesa length of around 700 μm. Increasing the mesa length cannot significantly improve actual modulation efficiency, but it does enhance the propagation loss.
Moreover, we cannot expect higher XPM efficiency from a higher implantation dose. Actually, a high ion dose generates a high density of defects that trap electrons, while the activation efficiency is not increased . Even with a high RTA temperature, defects cannot be completely recovered. For example, we tested an implantation dose of 1 × 1014 cm–2, and the TM loss was around 5 dB/mm higher than in the case of 5 × 1013 cm–2, with an XPM efficiency for a 500-μm-long mesa of only about 0.07 rad/pJ. Hence, in the following, we focused on an implantation dose of 5 × 1013 cm–2 and an RTA condition of 780 °C for 8 s for the final switch fabrication.
3. Switch design and fabrication
A schematic diagram of the monolithically integrated ISBT-MZI all-optical gate switch and a microscope image of the device for testing are shown in Figs. 3(a) and 3(b), respectively. The MZI gate switch consists of two multimode interference (MMI) 3-dB couplers and TE/TM beam combiners (BCs). Only part of MZI arm area is implanted for cross phase modulation. The device length is 2.2 mm, and the XPM part is 0.7 mm. In the implanted area, the refractive index of the CDQW material becomes lower than that in the area without implantation. From the measured cut-off mesa width, with consideration of the fundamental waveguide mode, the effective refractive index can be calculated. For TM polarization, it is 3.31 for an as-grown sample without RTA, and it slightly changes to 3.315 after RTA, while decreasing to 3.295 for an implanted sample after RTA. The waveguide width is selected to be 2 μm, which is near the cut-off width of the implanted mesa for TM polarization. At the input/output ports, the waveguide width is gradually increased to 2.5 μm for better optical coupling with a lensed fiber.
A directional coupler is employed as the TE/TM beam combiner with a mesa width of 1.6 μm, length of 200 μm, gap of 0.4 μm, and an additional 50-μm-long taper for width varying to 2 μm. TM light can be completely coupled to the other port (cross port), while TE light remains mainly in the original port (through port). Figures 4(a) and 4(b) show a schematic illustration and a cross-sectional SEM image at the central part of the fabricated directional coupler, respectively. As the etching rate in the gap area is slower than that in other part during the ICP process, deeper etching is needed to penetrate the CDQWs layer completely in the gap area. Actually, the gap profile has a trapezoidal shape. The measured transmission spectra of the TE/TM beam combiner are presented in Fig. 4(c). The extinction ratio for TM mode at λ = 1559 nm and for TE mode at λ = 1545 nm are 43 and 14 dB, respectively. The combining efficiencies are calculated to be more than 99% for TM polarization and 96% for TE polarization. Insertion loss for TM mode is better than that for TE mode because the coupling efficiency from the input fiber to the CDQW waveguide is higher for TM mode.
For device fabrication, a group of global marks was first formed in the sample surface by a mask aligner and ICP etching. The mask for area-selective implantation was 800-nm-thick SiO2, which was formed by buffered hydrofluoric acid (BHF) wet etching after the window pattern was defined by electron beam lithography. A 5 × 1013 cm–2 dose of Si+ ions was implanted onto the wafer surface, and the implanted sample underwent RTA at 780 °C for 8 s. Then, a photonic integrated circuit pattern was defined by electron beam lithography on the wafer with 300-nm-thick SiO2, with the position determined by the global marks. A Ti/Ni (10/45 nm-thick) metal mask was formed using the metal deposition/lift-off technique. After reactive ion etching of the SiO2 layer, a deep-etched mesa waveguide with a depth of 1.2 μm was fabricated by the ICP facility. Then, the composite mask was removed by BHF. The etched surface was planarized by BCB polymer to protect the mesa for later fabrication processes. The Ti/Au heater (100/10-nm thick, 250 × 10 μm2) was formed on the BCB surface. Both cleaved facets for the TE signal input/output were anti-reflection coated by a double layer of dielectric materials (SiO2 and ZrO2). Finally, the sample was bonded to a metal holder for the performance test.
4. Device performance and discussion
Performance of the heater for phase bias control in the gate switch device was first examined. Signal transmittance of two MZI output ports as a function of the voltage applied to the heater is shown in Fig. 5 . TE cw light at 1545 nm was launched as an input signal. It can be clearly seen that on one output port, i.e., the through port, the on/off switch operation can be accomplished by switching the applied voltage. When the applied heater voltage is 4.2 V, corresponding to the “off” state of the MZI through port, a π phase shift can be generated, while the extinction ratio, i.e., the transmittance difference between the MZI through port and drop port, is 21 dB. Compared with the condition of heater voltage not applied (0 V), a similar extinction ratio can be obtained. We can regard this to be the “on” state for the through port. This indicates that the imbalance in light power between the MZI arms due to fabrication errors is sufficiently small.
Then, the all-optical gate-switching operation was demonstrated. Pump light with pulse energy of 8.7 pJ, width of 2.4 ps, repetition rate of 10 GHz, and wavelength of 1559 nm was injected to an input port of the TE/TM beam combiner. TE probe spectra measured by an optical spectrum analyzer (OSA) for different pump polarization modes are shown in Fig. 6 . It can be clearly seen that only in TM pump mode, does the TE probe spectrum have strong modulation centered at a wavelength of 1545 nm. The sideband induced by XPM is to be noted. The modulation band is caused by the periodic phase shift induced by the 10-GHz pump.
Figure 7 shows the temporal profile of the gated TE probe light measured by the optical sampling oscilloscope. The upper and lower panels show the maximum and minimum signal output, respectively, corresponding to the different voltages applied to the heater for phase bias control. Although the extinction ratio is not significantly high compared with previous reported MI switch , it can be confirmed that all-optical gating operation is realized while the TM pump light is injected from the beam combiner. The full width at half maximum of the gated pulse is about 3.4 ps.
There are many possible reasons for the low extinction ratio of the gate switch. First, the XPM efficiency of the ion-implanted sample is small compared with the doped sample . Tailoring the interband absorption edge can enhance the XPM efficiency . Selecting a different CDQW material composition could improve the device performance. Moreover, the designed mesa width is close to the cut-off width of the TM pump light, and the waveguide edge roughness has considerable influence on the transmission loss of both TM and TE light. As analyzed before, a higher propagation loss causes a decrease in the XPM efficiency . Indeed, we observed that after ICP etching, sometimes, there were some small particles remaining around the waveguide edge that were difficult to remove. These could cause increased loss. Therefore, the ICP etching process needs to be further stabilized. Furthermore, there is a refractive index mismatch between the as-grown and implanted areas, which can reflect the light and decrease the effective pump light energy. It should also be noted that we have obtained a slightly high XPM efficiency in the straight implanted waveguide measurement, where the probe and pump signals are connected into the same input fiber and their mode overlap is always the maximum. However, the pump and probe signal are combined through a beam combiner in the gate switch, and the mesa width is not optimized for the highest mode overlap.
In addition, the TE signal pulse modulated by the TM pump has an energy loss ΔA due to a bandgap shift , which is proportional to the mesa length. Therefore, there is some intensity imbalance between the modulated and unmodulated signal in the MZI arms. This loss is estimated to be about 0.12 dB/pJ for a straight implanted mesa (1 mm long, 2 μm wide), according to the recorded intensity change before and after the pump is induced. Higher pump energy causes higher probe signal loss. Therefore, the symmetric MZI gate structure needs to be corrected to incorporate a variable optical attenuator for amplitude imbalance control . Taking all these factors into consideration, the device performance can be improved. All these works need further investigation.
In conclusion, a monolithically integrated all-optical MZI-ISBT gate switch was developed by area-selective Si+ ion implantation and subsequent RTA in InGaAs/AlAsSb CDQWs. The RTA condition was optimized, and the sample was annealed at 780 °C for 8 s with an implantation dose of 5 × 1013 cm–2. The dependence of XPM efficiency on the implanted waveguide length was also analyzed, and there was an optimum working length of about 700 μm with consideration for the saturated modulation efficiency and loss of the probe and pump signal. An increased implantation dose did not improve XPM efficiency. For the MZI-ISBT switch, by injecting pump light through a TE/TM beam combiner, XPM was induced in one MZI arm, and all-optical gating operation at a 10-GHz repetition rate was demonstrated. Despite the low extinction ratio, we could confirm that the all-optical gating operation was realized with a response time of 3.4 ps as a proof-of-concept. This device is compact and mechanically stable. Moreover, it can be easily applied for large-scale integration. It can also overcome the drawbacks of the previously reported Michelson-type monolithic device, which suffers from optical loss due to the interband absorption edge and 50% reflection of pump and probe light. This technique also provides a simple means of connecting active and passive optical waveguides in an integrated manner, which shows the feasibility of monolithically integrating the novel XPM with other quantum-well-based functional modules through area-selective carrier activation.
References and links
1. H. Tsuchida, T. Simoyama, H. Ishikawa, T. Mozume, M. Nagase, and J. Kasai, “Cross-phase-modulation-based wavelength conversion using intersubband transition in InGaAs/AlAs/AlAsSb coupled quantum wells,” Opt. Lett. 32(7), 751–753 (2007). [CrossRef] [PubMed]
2. G. W. Cong, R. Akimoto, M. Nagase, T. Mozume, T. Hasama, and H. Ishikawa, “Mechanism of ultrafast modulation of the refraction index in photoexcited InxGa1-xAs/AlAsySb1-y quantum well waveguides,” Phys. Rev. B 78(7), 075308 (2008). [CrossRef]
3. T. Ogasawara, S. Gozu, T. Mozume, K. Akita, R. Akimoto, H. Kuwatsuka, T. Hasama, and H. Ishikawa, “Ultrafast electron dynamics of intersubband excitation concerning crossphase modulation in an InGaAs/AlAs/AlAsSb coupled double quantum well,” Appl. Phys. Lett. 98(25), 251104 (2011). [CrossRef]
4. H. Ishikawa, Ultrafast All-Optical Signal Processing Devices (Wiley, 2008), chap. 5.
5. R. Akimoto, T. Simoyama, H. Tsuchida, S. Namiki, C. G. Lim, M. Nagase, T. Mozume, T. Hasama, and H. Ishikawa, “All-optical demultiplexing of 160- to 10-Gb/s signals with Mach-Zehnder interferometric switch utilizing intersubband transition in InGaAs/AlAs/AlAsSb quantum well,” Appl. Phys. Lett. 91(22), 221115 (2007). [CrossRef]
6. R. Akimoto, G. Cong, M. Nagase, T. Mozume, H. Tsuchida, T. Hasama, and H. Ishikawa, “All-optical demultiplexing from 160 to 40/80 Gb/s using Mach-Zehnder switches based on intersubband transition of InGaAs/AlAsSb coupled double quantum wells,” IEICE Electron Express E92(C), 187–193 (2009). [CrossRef]
7. R. Akimoto, S. Gozu, T. Mozume, K. Akita, G. W. Cong, T. Hasama, and H. Ishikawa, “All-optical wavelength conversion at 160Gb/s by intersubband transition switches utilizing efficient XPM in InGaAs/AlAsSb coupled double quantum well,” in Proceedings of 35th European Conference on Optical Communication (VDE VERLAG GMBH, Berlin, 2009) 1.2.2.
8. T. Kurosu, S. Namiki, R. Akimoto, H. Kuwatsuka, S. Sekiguchi, N. Yasuoka, K. Morito, H. Hasama, and H. Ishikawa, “Demonstration of 172-Gb/s optical time domain multiplexing and demultiplexing using integratable semiconductor devices,” IEEE Photon. Technol. Lett. 22(19), 1416–1418 (2010). [CrossRef]
9. R. Akimoto, S. Gozu, T. Mozume, and H. Ishikawa, “Monolithically integrated all-optical gate switch using intersubband transition in InGaAs/AlAsSb coupled double quantum wells,” Opt. Express 19(14), 13386–13394 (2011). [PubMed]
10. T. Kurosu, K. Tanizawa, S. Namiki, R. Akimoto, H. Kuwatsuka, T. Hasama, H. Ishikawa, T. Nakatogawa, K. Oyamada, Y. Tanaka, S. Ide, H. Onaka, and T. Asami, “Dynamic optical path switching of 172-Gbit/s OTDM ultra-high definition video signals using fast channel-identifiable clock recovery and integratable devices,” The Optical Fiber Communication Conference and Exposition and the National Fiber Optic Engineers Conference (OFC/NFOEC) (Los Angeles, California 2012) PDP5C.10.
11. V. Aimez, J. Beauvais, J. Beerens, D. Morris, H. S. Lim, and B. S. Ooi, “Low-energy ion-implantation-induced quantum-well intermixing,” IEEE J. Sel. Top. Quantum Electron. 8(4), 870–879 (2002). [CrossRef]
12. G. W. Cong, R. Akimoto, S. Gozu, T. Mozume, T. Hasama, and H. Ishikawa, “All-optical Cross-phase Modulation Generation by Ion Implantation in III-V Quantum Wells,” IEEE Photon. Technol. Lett. 22(24), 1820–1822 (2010). [CrossRef]
13. G. W. Cong, R. Akimoto, S. Gozu, T. Mozume, T. Hasama, and H. Ishikawa, “Simultaneous generation of intersubband absorption and quantum well intermixing through silicon ion implantation in undoped InGaAs/AlAsSb coupled double quantum wells,” Appl. Phys. Lett. 96(10), 101901 (2010). [CrossRef]
14. E. H. Li, Semiconductor Quantum Well Intermixing, Optoelectronic Properties of Semiconductors and Superlattices Vol. 8 (Gordon and Breach Science, Singapore, 2000) chaps. 1 and 7.
15. S. Gozu, T. Mozume, R. Akimoto, K. Akita, G. Cong, and H. Ishikawa, “Cross phase modulation efficiency enhancement in In0.8Ga0.2As /Al0.5Ga0.5As/ AlAs0.56Sb0.44 coupled double quantum wells by tailoring interband transition wavelength,” Appl. Phys. Express 2, 042201 (2009). [CrossRef]
16. M. Nagase, Y. Shoji, S. Suda, K. Kintaka, H. Kawashima, R. Akimoto, H. Kuwatsuka, T. Hasama, and H. Ishikawa, “Ultrafast all-optical gating operation using Michelson interferometer for hybrid integration of intersubband transition switch on Si platform,” IEEE Photon. Technol. Lett. 23(24), 1884–1886 (2011). [CrossRef]