1. Introduction

Ultrafast all-optical signal processing devices play a key role for future communications networks [1]. InGaAs/AlAsSb coupled double quantum wells (CDQWs) are a good candidate owing to the merit of only a few picoseconds relaxation time [2]. For such CDQWs, with the intersubband transition (ISBT) caused by the transverse magnetic (TM) pump light, cross-phase modulation (XPM) can be induced to the transverse electric (TE) probe light, which excites interband transition and is immune to ISBT absorption [3, 4]. This ISBT-induced XPM has already been applied in high-speed all-optical switch modules. A Mach-Zehnder interferometer (MZI) gate switch in a free-space layout with an ISBT waveguide chip for XPM placed in one MZI arm was reported [5]. It can be used for all-optical signal processing experiments such as demultiplexing [6], etc. However, this MZI gate switch suffers from the inconvenience for large-scale integration and the environmental change-caused mechanical instability. Thus, a monolithic integration of an ISBT waveguide with a Michelson interferometer (MI) was developed [7], which was successfully applied for delivering real-time video signals for a 172-Gb/s optical-time-division-multiplexed (OTDM) system [8]. Moreover, a compact MZI switch was realized on a single chip recently [9], with the ISBT-induced XPM caused by the area-selective Si ion implantation. However, all these ISBT switches suffer from either low signal transmittance or weak XPM efficiency. An XPM study for different CDQW designs has been reported and a high XPM efficiency usually associates with a high probe signal absorption [10]. There is always a trade-off between the signal transmittance and the XPM efficiency. As a low signal loss is critical in the scaling of large size photonic integrated circuits with stringent power requirement, a gating device with low power consumption and high XPM efficiency is more favorable for practical applications.

Actually, the XPM and signal propagation parts are different functional sections in a gating device, and should have a varying bandgap energy requirement. Selective area epitaxy can realize the bandgap integration [11], but it involves repeated use of expensive epitaxial growth systems, which limit the yield owing to the high cost. Quantum well intermixing (QWI) is another means of integrating different bandgap regions in the same epitaxial layer at the postgrowth level [12]. Phosphorus (P) ion implantation has been used to induce the intermixing of III-V quantum wells [13]. Following the implantation, a rapid thermal annealing (RTA) step is usually carried out to induce diffusion of the point defects through the active region, thus promote QWI while partially heal the damage within the shallow implanted layer. QWI can result in a blueshift of the bandgap profile. Spatial selectivity of this process can be realized by using the implantation mask with appropriate shape and thickness. In this paper, we first compare the transmittance and XPM efficiency of the implanted and non-implanted CDQWs with both annealing at 720 °C for 60 s. Then an MI switch with monolithic integration of an implanted part for signal propagation and a non-implanted portion for XPM is designed and fabricated, and the gating performance is discussed.

2. Signal transmittance improved by implantation induced QWI

The wafer with InGaAs/AlAsSb CDQWs was grown on a semi-insulating InP substrate by molecular beam epitaxy method. The CDQWs were Si-doped to activate ISBT, which were designed to have high XPM efficiency but low signal transmittance. QWI for as-grown and implanted samples were then investigated experimentally. Vacancies were produced in the cladding InP layer, after implantation with 5 × 10¹⁴ cm^–2, 350-keV P⁺ ions. The implantation energy was chosen in order to limit the maximum defect concentration within the upper cladding layer, thus keeping the high optical quality of the CDQWs. The samples were tilted at a 7° angle during implantation to minimize ion channeling effects. The substrate temperature was elevated to 200 °C throughout implantation to avoid the formation of defect aggregates which have higher diffusion activation energy [14]. Following ion implantation, subsequent RTA at 720 °C for 60 s was carried out. The vacancies were diffused to CDQWs layer and induced QWI. Before RTA, a 50-nm SiO₂ protective layer was deposited on the sample surface by electron cyclotron resonance (ECR) plasma enhanced sputtering. In order to measure the XPM and transmittance property of the samples, deep etched straight waveguide structures were fabricated by electron beam lithography and subsequent RIE, ICP etching processes [7]. The mesa was covered with benzocyclobutene (BCB) polymer and cleaved into 1-mm length, and then both facets were anti-reflection coated. From the TE spectra with varying mesa width, it can be inferred that for the width larger than 1.3 μm, higher-order modes will propagate through the waveguide for a 1545-nm wavelength, which is the probe wavelength for the gating operation.

TE transmission spectra for 1.3-μm-wide waveguides were recorded by optical spectrum analyzer (OSA) (Fig. 1(a)). Compared with the as-grown sample without RTA, transmittances are increased for both implanted and non-implanted samples after annealing, and the implanted one has a nearly 6-dB improvement at a 1545-nm wavelength. The implanted sample after RTA has lower loss than the non-implanted one, though not obvious at this probe wavelength. However, 4-dB transmittance increase can be observed at a 1520-nm wavelength. In actual switch device, light pass length is more than 2 mm, thus corresponding transmittance improvement would be more than 8 dB for a 1520-nm probe. We cannot do the experiment at a 1520-nm probe for the present CDQWs due to the limitation of experimental setup, which is determined by the wafer growth. After the CDQWs are re-designed, a similar transmittance improvement as at 1520-nm wavelength can be expected for the working wavelength of C-band (1530-1565nm). Inset of Fig. 1(a) shows the recorded TM spectra and all samples have very strong intersubband absorption, with the TM loss higher than 40 dB.

 

Fig. 1 (a) TE Spectra property (inset: TM spectra) and (b) peak nonlinear phase shift as a function of TM pump power, for 1.3-μm-wide waveguides of the implanted and as-grown samples with or without RTA.

Download Full Size | PDF

XPM efficiency was measured by a system using a delay-line interferometer (DLI) with a 25-ps delay time to convert phase shift to intensity variation [9]. TM Pump light with a pulse width of 2.4 ps, a repetition rate of 10 GHz, and a wavelength of 1560 nm was injected into the straight waveguide. Figure 1(b) shows the peak nonlinear phase shift as a function of the pump power. TM pump power is defined as the power injected into the fiber. The as-grown sample has the highest XPM efficiency of 0.93 rad/pJ. After RTA, the XPM efficiency decreases for either implanted or non-implanted sample. The non-implanted sample can still maintain a relative high XPM efficiency of 0.85 rad/pJ, while that of the implanted one is about 0.78 rad/pJ.

3. Monolithically integrated MI switch

Due to the implantation-induced QWI, the implanted sample can acquire a high transmittance compared with the as-grown one, while the XPM efficiency of the non-implanted sample changes little. Thus an MI gating switch was developed by area-selective ion implantation to monolithically integrate the implanted CDQWs part for signal transmission and the non-implanted portion for XPM. A schematic illustration and a microscope image of the switch undergoing test are shown in Figs. 2(a) and 2(b), respectively. The device is 1-mm-long and based on an MI consisting of multi mode interference (MMI) 3-dB coupler. The left facet is anti-reflection (AR) coated for the input/output of TE probe signal, whereas that on the right is half-reflection (HR) coated for both TM pump light input and TE probe signal reflection. The reflected TE signals from two arms on the right of the MMI coupler will then interfere with each other as an MI configuration. Part of the lower MI arm is not implanted and functions as a nonlinear waveguide for XPM. Due to the strong ISBT absorption of TM light, length of the non-implanted area is chosen to 250 μm, which will not decrease the XPM efficiency too much compared with a 1-mm-long mesa. After fabrication of the photonic integrated circuit, the sample was planarized by BCB polymer and a resistive heater (80-nm Ti and 10-nm Au) is attached to the XPM arm for a static phase bias control between the two MI arms. On the top arm, a variable optical attenuator (VOA) is formed to maintain a power balance between the two MI arms. This is because the signal has an additional loss due to a thermal red shift of the interband absorption edge in the case of high pump power injection [4]. VOA function is achieved by an MZI with a heater for controlling light attenuation [7]. Two U-shaped waveguides and an inverse taper with a tip width of 0.3 μm are connected to the MZI to attenuate signals completely during VOA operation.

 

Fig. 2 (a) Schematic diagram of a monolithically integrated ISBT-MI switch. MMI: multi-mode interference; VOA: variable optical attenuator; MZI: Mach-Zehnder interferometer. (b) Microscope image of the fabricated chip for test.

Download Full Size | PDF

As the MMI 3-dB coupler is a key component for the switch, we experimentally optimized the structure parameters. Two well-worked MMI 3-dB couplers can lead to a maximum MZI cross-port output, thus an MZI configuration was adopted for optimization. The MMI width is fixed at 8.4 μm, taking consideration of the connected waveguide width and the gap between the waveguides for easy of fabrication. That is because a narrow gap area is hard to etch away during the ICP process. It can be seen from Fig. 3(a), with the measured MZI cross-port transmittance normalized by that of a 1.3-μm-wide straight waveguide, the maximum output happens at a length of 107 μm, which is the optimum 3-dB coupler length. The experimental values coincide well the simulation results. The transmission spectra of the MMI 3-dB coupler are shown in Fig. 3(b). In order to clear demonstrate the bandwidth property, the MZI output spectra were measured and normalized by the recorded straight waveguide spectrum. Performance of the MMI 3-dB coupler will deteriorate at a long wavelength region.

 

Fig. 3 (a) Simulated MZI cross-port output and the measured values normalized by the straight waveguide transmittance with varying MMI length. (b) Transmission spectra for the MMI 3-dB coupler (inset: normalized MZI spectra).

Download Full Size | PDF

4. Device performance and discussions

An MI switch was manufactured based on the optimized MMI structure. Besides of the switch with non-implanted waveguide for XPM (area-selective implantation), the switch with implanted XPM part (homogeneous implantation) was also fabricated simultaneously for comparison, as shown in the top of Fig. 2(b). Static phase bias control function was checked according to the signal output of the two MI switches as a function of the voltage applied to the heater (Fig. 4(a)). It can be clearly seen that the switching operation can be fulfilled by turning the applied voltage. At 2-V voltage, a π-rad phase shift can be generated for the switch with area-selective implantation, and the static extinction ratio is around 21 dB. Maximum transmittance is realized at the zero voltage. This indicates the optical path difference between two MI arms and their light power imbalance due to fabrication errors is sufficiently small. For the homogeneous implanted switch, its maximum transmittance is almost the same. Its extinction ratio is slightly high (around 25 dB) owing to the uniform waveguide property. But the maximum output cannot be fulfilled when the heater is inactive. This is because the refractive index of the XPM part is different from the non-implanted case, and there exists an additional phase bias when no voltage is applied.

 

Fig. 4 (a) Signal transmittance as a function of the voltage applied to the phase bias control heater. (b) Temporal profile of the gated TE probe light.

Download Full Size | PDF

Then all-optical gating operation was investigated with the same TM pump light for the XPM efficiency measurement as mentioned before. Full switching of a π-rad nonlinear phase shift is achieved with pump pulse energy of around 5.6 pJ for the switch with non-implanted portion for XPM, while for the homogeneous implanted switch, about 6.2-pJ energy is needed. Figure 4(b) shows the temporal profile of the gated TE probe light measured by the optical sampling oscilloscope (OSO) for the switch with area-selective implantation. Upper and lower panel are for the maximum and minimum signal output with adjusting the voltage applied to the heater for phase bias control. All-optical gating operation can be realized successfully.

Through comparison of the switches with area-selective and homogeneous implantation, the former cannot only acquire a low signal loss as the latter, but also realize a high XPM efficiency. The static extinction ratio of either switch still has much space to be raised. As it is related with the interference of the reflected probe signals and strongly affected by the waveguide facet cross-sectional profile, the cleaving method needs to be improved. A fine optimization of the 3-dB coupler design and fabrication process will contribute as well. Furthermore, the XPM efficiency of the switch is around 0.56 rad/pJ, more than half of the straight waveguide efficiency of 0.85 rad/pJ, despite of the half reflection coating. This is because the probe signal undergoes XPM twice during co-propagation and counter-propagation in the switching operation, whereas it does so only once as co-propagation case in the transmission configuration for the XPM efficiency measurement of a straight waveguide. The maximum increase factor could be 2. Previous simulations show that this occurs when the ISBT absorption is sufficiently strong [6]. Also, due to that the probe signal passes through the waveguide twice, even a slight decrease on the propagation loss will enhance the device transmittance much.

It should be pointed out that the transmittance and XPM efficiency difference between the implanted and non-implanted samples is slightly small. The merit of this area-selective implanted MI switch cannot be sufficiently demonstrated at the current band edge energy of the as-grown sample. A sample with higher interband absorption for TE signal and thus achieving an enhanced XPM efficiency would be more suitable to show the advantages by using area-selective implantation induced QWI. The improved transmittance for the non-implanted case may be caused by the decrease of built-in crystal defects after RTA, which is produced during a crystal growth and adds to a base line of the absorption. This absorption decrease is not caused by a bandgap shift. Thus, while the implanted section has an enough large band edge shift in order to attain a signal loss as low as possible, the non-implanted area after annealing can have an almost similar XPM efficiency as high as that of the original as-grown sample. Besides, the implantation and annealing process needs a further fine optimization.

5. Conclusion

In summary, the postgrowth band edge control of the InGaAs/AlAsSb CDQWs is realized, by means of P ⁺ ion implantation and subsequent RTA. With annealing at 720 °C for 60 s, the implanted CDQWs can achieve a high signal transmittance around 1520-nm wavelength while the non-implanted sample can maintain a satisfying XPM property close to the as-grown sample. Through area-selective implantation, this can be applied for monolithic integration of a passive signal propagation part and an active XPM portion for a high-speed all-optical gating device. An MI switch is manufactured as a proof of concept. Full switching of the π-rad nonlinear phase shift is achieved with pump pulse energy of 5.6 pJ. This improved switch device not only benefits the gating operation in OTDM systems, but also testifies the technique for monolithic integration of different bandgap energies through implantation induced quantum well intermixing in the ISBT CDQWs. It has the advantages such as low-cost, reliability, and robustness, which would have a vast application prospects for signal processing photonic devices.