The authors report their latest results on II–VI intersubband all-optical switches in which the 10 dB absorption saturation energy is lowered to ~2.0–2.2 pJ for 1.55–1.58 µm by decreasing the thickness of the active layer and increasing the refractive index difference between the core layer and the cladding layers in waveguides. Such low saturation energies greatly improve the switching performance. <7 pJ pump energy at 1520 nm is sufficient for realizing 10 dB switching operation at 1566 nm (switching energy: ~0.7 pJ/dB). To the best of our knowledge, these switching energy and saturation energy values are the lowest reported ones for such ultrafast intersubband all-optical switches at telecommunication wavelengths.
© 2007 Optical Society of America
High-speed all-optical switch modules (>100 Gbps) are fundamental components in optical multiplexing and demultiplexing devices for next-generation fiber communication (λ~1.55 µm) networks. Such ultrafast nature can be satisfied in the intersubband transition (ISBT) of semiconductor quantum wells (QWs) due to its subpicosecond carrier relaxation [1–3]. Therefore, quantum systems with sufficiently large conduction band offsets can be fabricated into ultrafast ISBT waveguide switches for the light of the transverse-magnetic (TM) polarization. So far, such switches have been reported in InGaAs/AlAsSb [4,5], GaN/AlN [6–9], and (CdS/ZnSe)/BeTe [10,11] systems. The switching energy (Es) is the characteristic parameter used for evaluating the switching performance, which currently is still not low enough for above switches (32 pJ , 100 pJ , and 11.3 pJ , respectively) to achieve a 10 dB switching extinction ratio (SER). Therefore, a further reduction in Es is necessary for practical switching operation at >100 Gbps.
In this letter, we present our latest progress in decreasing Es for (CdS/ZnSe)/BeTe-based waveguide devices. In addition to our previous waveguide structure optimizations , we continue to optimize the composition of cladding layers (CLs); further, we adopt ZnSe/BeTe multiple quantum wells (MQWs) as new optical confinement layers (OCLs), instead of the previously used ZnBeSe ternary alloy. Meanwhile, the thickness of the MQW layer is decreased to a third of the previous value. Consequently, the absorption saturation efficiency is expected to be further enhanced, which will significantly decrease the 10 dB saturation energy as well as Es for 10 dB SER.
2. Waveguide fabrication
The waveguide sample comprises a ~2.5-µm-thick Zn0.57Mg0.28Be0.15Se bottom CL (n~2.32–2.34), a ZnSe (10 ML)/BeTe (5 ML) MQW bottom OCL (n~2.52–2.53), a (CdS/ZnSe)/BeTe MQW active layer (AL) (n~2.52–2.54), a ZnSe/BeTe MQW upper OCL, and a ~1-µm-thick Zn0.57Mg0.28Be0.15Se upper CL, all of which were sequentially grown on a GaAs (001) substrates in a dual-chamber molecular beam epitaxy system. The corresponding n was measured at 1.55 µm using an ellipsometer for samples comprising each single layer only. As compared to our previous waveguides , the Mg and Be compositions in CL have been respectively increased to 28% and 15% in order to decrease the n of CL (n CL), while the lattice-match condition is still maintained. The previous ZnBeSe ternary OCL (n~2.45) has been replaced by ZnSe/BeTe MQW. By optimizing the relative thicknesses of ZnSe and BeTe, we can ensure the lattice-match growth and meanwhile increase the n of OCL (n OCL) as high as that of AL. As a result, the separate confinement heterostructure (SCH) layer (including an AL and both OCLs) has a nearly homogenous n, which increases the refractive index contrast (RIC) between SCH and CLs. Each OCL has 86 QWs and the AL has 15 QWs (~0.072 µm) in which each cycle comprises 1 ML ZnSe/~3 ML Cl-doped CdS/1 ML ZnSe as well as a 10 ML BeTe barrier. Therefore, there are two main differences between the previous waveguide and the current waveguide: (1) the RIC between SCH and CL is increased and (2) the thickness of AL was decreased from 0.24 to 0.072 µm, i.e., a reduction of nearly two thirds. A ~1-µm-wide high-mesa waveguide was fabricated with ~3-µm-wide tapered structures at both sides by using a standard device process including high-resolution photolithography and reactive ion etching. The total waveguide length is ~500 µm. The waveguide image can be referred to in Ref. .
3. Measurement setup
Figure 1 schematically shows the main components of the laser system for the absorption saturation and pump-probe measurements. A Ti:sapphire modelocked laser with a repetition rate of ~76 MHz and output power of ~1.2 W pumps an optical parametric oscillator (OPO) that generates two pulse beams. One beam can be varied within a large wavelength region by changing the output wavelength of the Ti:sapphire laser, while the other is normally maintained at ~1566 nm. The pulse durations of both beams can be adjusted to 0.3–0.4 ps by selecting suitable filters. A high-resolution stage and OD filters are controlled by a computer to gradually change the pump-probe delay time and the pulse power coupled to the fiber by a freespace-to-fiber (FF) coupler, respectively. Optical pulses are injected into the waveguide and are received through two lensed dispersion-shifted polarization-maintaining fibers (DS-PMF). For the input fiber between the FF coupler and the waveguide, we introduce a coupler that draws out a small fraction (~one eighth) of the pulse power to monitor the remaining power in the DS-PMF. Therefore, the input pulse energy is defined as the power that flows into the DS-PMF connecting the coupler and waveguide. Considering the fiber-to-waveguide loss, the real saturation energy as well as the real switching energy should be lower than our measured values. The input pulse energy is always monitored by an optical powermeter, while the output pulse has two path selections for power analysis. During the absorption saturation measurement, the output pulse energy is also analyzed by a powermeter to calculate the insertion loss, whereas for the pump-probe measurement, the output pulse in the fiber is coupled to a spectrometer by mirrors, which monitors the probe wavelength. Meanwhile, the probe beam is chopped, and a lock-in amplifier is used to increase the signal-to-noise ratio for the intensity analysis of the probe pulse. Except for aforementioned time-resolved system, a Fourier transform infrared spectrometer was used to measure the static intersubband absorption spectrum at room temperature. For this case, the samples were mechanically polished to the multiple-pass geometry.
4. Results and discussion
4.1 Beam propagation simulation
We calculate the absorption saturation curves by simulating the pulse propagation in the waveguide with the aim of numerically optimizing waveguide parameters and examining the potential improvements based on the two abovementioned differences between this work and the previous work. The simulation is performed using BeamPROP (Rsoft Design Group, Inc. ), which adopts the implicit finite-difference beam propagation method (BPM). The waveguide model is three-dimensional tapered structure, as shown in Fig. 2. In the simulation, h AL, h OCL, n CL, n OCL, and w mesa are variables, while the other symbols are constants. The absorption saturation is assumed following α=α 0/(1+I/Isat) where α (α 0), I, and Isat are the absorption coefficient, the input pump intensity, and the saturation intensity, respectively. The simulation needs to set Isat and α 0 for the AL. To obtain comparable results for different hAL values, we need change h AL. But it is difficult to know, in a direct way, the relation between h AL and I sat as well as α 0. Therefore, a better way is to set a fix thickness with fix parameters as a unit. Such a unit is ideally one QW layer, however, we cannot mesh waveguides into so small size in simulation. To meet such an idea, we introduce a unit of a fix thickness and then use it to extend to other thicknesses. The thickness unit of 0.1 µm with constant Isat=1 W/µm2 and α 0=0.081 µm-1 is used to construct AL. For example, AL has three units if h AL=0.3 µm. This method is reasonable for optimizing h AL only if the carrier density is stable from one QW to another. The full transparent boundary condition, TM polarization, and the semi-vector mode are selected in simulation. The input field adopts the mode input file that is pre-calculated for each structure with different index profiles and the pump wavelength is 1550 nm. The computed steps along x, y, and z are 0.2, 0.01, and 0.5 µm, respectively.
Figure 3(a) shows two simulated absorption saturation curves, from which an apparent improvement of saturation efficiency is observed due to the increase in RIC and the decrease in h AL. For quantitative comparisons, the half-saturation intensity (Ish) is defined in Fig. 3(a) as the input power that increases the insertion loss from the unsaturated value (T0) to 0.5T0. Here we use Ish instead of the 10 dB saturation energy since the maximum increase in transmittance is occasionally less than 10 dB. As seen, the increase in RIC and the decrease in h AL by two-thirds results in a ~67% reduction in Ish. In order to separate the effects of these two changes and further optimize the waveguide parameters, we map Ish against h OCL and w mesa for three situations with different h AL, n CL, and n OCL values, as shown in Figs. 3(b)–3(d). By comparing Fig. 3(b) and 3(c), we can understand the role of the increase in RIC, which enlarges the structure region allowing the existence of the optical mode and moves the valley path to a smaller h OCL value, although Ish at the valley paths exhibits no obvious difference. A comparison of Figs. 3(c) and 3(d) reveals a >50% decrease in Ish for any w mesa and h OCL values, which shows the remarkable effectiveness of decreasing h AL in lowering Ish. Ish exhibits a weak dependence on h OCL for w mesa >1.5 µm (Fig. 3(b)) and w mesa >1 µm (Fig. 3(c)). Unlike that in Figs. 3(b) and 3(c), the decrease in h OCL in Fig. 3(d) results in an apparent decrease in Ish for each w mesa value. Moreover, the valley path shown in Fig. 3(d) continues to move down to h OCL=0.3 for w mesa >1 µm. In all these three figures, Ish always decreases with decreasing w mesa for any h OCL value due to the improved optical confinement. Therefore, we can make the following conclusions to guide future waveguide design: (1) decreasing h AL can significantly decrease the saturation energy; (2) decreasing h OCL is more effective for thinner ALs; and (3) the lowest saturation intensity occurs close to the edge of the mode-inhibited region that depends on RIC and w mesa. The numerical results indicate a lower saturation energy in our current waveguide with thinner AL, which is confirmed in the following experimental parts.
Our purpose of the simulation is to find a guidance direction to adjust waveguide parameters in order to increase the pulse power per unit area in AL cross section and to understand how the waveguide structure influences saturation behavior. By relative comparison with experiments, we can know how helpful the simulation is. However, it is necessary to mention that we cannot directly compare the simulated saturation curves with the experimental ones because the continuous wave light is used in simulation. In addition, the nonlinear parameters are different from the real values since it is difficult to know such parameters for a part of AL. So to make the results of different AL thicknesses comparable, we set constant nonlinear parameters in simulation for a thickness unit and only discuss what differences the changes in waveguide structure result in.
Next, we discuss the mechanism for optimizing structural parameters in reducing saturation energy. Generally, the saturation energy depends on the basic properties of quantum wells such as relaxation time and transition moment and waveguide structures. If assuming the nonlinear properties of AL is constant, the difference in the saturation energy should result from the difference in the waveguide structure. As we know, the waveguide structure determines the optical mode profile and the relative position of AL in the mode profile influences the saturation energy. We discuss two cases: (1) vary h AL and h OCL with a constant wmesa, and (2) vary wmesa with constant hAL and hOCL. (1) There should exist an optimum h OCL for each h AL. If h OCL is too large, the fraction of the pump power to excite AL will decrease. If h OCL is too small, either the optical mode will disappear, or the upper and lower areas of AL’s cross-section will approach the edges of the mode profile where the power is low. As a result, these areas will be unsaturated. Therefore, h AL should be decreased for AL to feel the power as large as possible, and meanwhile h OCL should be optimized to increase the pulse power per unit area in AL. As seen in Fig. 3(c) and 3(d), Ish at valley paths decreases by more than a half after optimizing h AL and h OCL. But the decrease in h AL will result in the decrease of on-off ratio. So we may need to adjust the doping contration. (2) Only if the optical mode exists, more smaller w mesa, more lower the saturation energy. This is because the pulse power per unit area in AL increases when horizontal light confinement becomes better with decreasing w mesa. In our case, RIC can not be increased more to get an obvious decrease in the saturation energy due to the composition limitation of CL. So it could be a feasible method to improve waveguide structure by the method introduced above.
4.2 Absorption saturation
Before measuring the absorption saturation, the static intersubband absorption spectra were measured to check its peak position. We use the same thick epitaxial wafer as that for waveguide fabrication and polish a piece of wafer into multiple-pass geometry, as seen in Fig. 4. The absorption peak at ~1565 nm is observed in Fig. 4 for p-polarization, but absent from s-polarization, which is sufficient to attribute it to ISBT due to the quantum well confinement direction. Except for this peak, other peaks and valleys result from interference due to the existence of optical path difference when the light passes epilayers. Such interference does not exist in waveguide samples since the light is parallel to the surface of epilayers. When we change the incident angle, the interference phenomenon shows the same varied behavior for both p and s polarization, however, the ISBT peak position does not change. This difference further supports its intersubband origin. In the inset of Fig. 4, we can determine its peak position and full width at half maximum (FWHM) as ~1565 and 100 nm, respectively.
Figure 5(a) shows the TM and transverse-electric (TE) insertion loss versus the input pulse energy at the pump wavelength of 1565 nm that is nearly resonant with the ISBT energy determined by Fourier transform infrared spectrometer. The pulse duration is measured to be ~0.34 ps using an autocorrelator. Absorption saturation obviously occurs for TM polarization with increasing the input pulse energy, while the TE insertion loss remains at a relatively low level (<6 dB). The 10 dB saturation energy is defined as the input energy that can induce a 10 dB increase in transmittance with respect to the unsaturated insertion loss. As shown in Fig. 5(a), the 10 dB saturation energy at 1565 nm is ~2 pJ. In contrast to our previous results with a comparable mesa width , we achieved a reduction of nearly 70% in the 10 dB saturation energy (decreasing from 7 to 2 pJ). This decrease is comparable to that in simulation (~67%), as shown in Fig. 3(a). Based on the above numerical results, the ~70% reduction in saturation energy should be mainly attributed to the decrease in the thickness of AL. Considering the real application, we measure such absorption saturation curves at different wavelengths from 1520–1620 nm and then obtain the wavelength-dependent 10 dB saturation energy, as shown in Fig. 5(b). The figure also shows the unsaturated TM insertion loss versus the wavelength that exhibits the absorption spectral characteristics. Both curves have peak positions at ~1565 nm, indicating that the saturation energy reaches the minimum value when it is resonantly pumped. When the photon energy deviates from the resonant energy, the 10 dB saturation energy gradually increases due to the decrease in pump efficiency, while it still maintains a very low level (2.0–2.2 pJ) for 1.55–1.58 µm. This is beneficial for future device designs.
4.3 Switching performance
Such low saturation energies are expected to lead to correspondingly low values of Es in ultrafast switching operation. To confirm this expectation, the pump-energy-dependent transmission of the signal was measured by the time-resolved two-color pump-probe experiment with a 1520-nm beam as the control pulse and a 1566-nm beam as the signal pulse, both of which were aligned with the TM polarization. Their pulse durations are ~0.34 ps and ~0.44 ps, respectively and the probe pulse energy is as low as 0.1 pJ. Figure 6 shows the temporal signal transmittance for different pump energies. When two pulses coincide, the signal transmittance is obviously enhanced. This can be clearly illustrated by plotting the increase in signal transmittance (i.e., SER) against the corresponding pump energy, as shown in the inset of Fig. 6. With an increase in the pump energy, SER first increases linearly and then exhibits saturation characteristics. As seen, <7 pJ pump energy at 1520 nm induces 10 dB SER, which is currently the lowest reported Es (7 pJ/10 dB) in ISBT switches. A further ~38% reduction is obtained as compared to our previous value of 11.3 pJ/10dB . Furthermore, the change in the signal transmittance nearly returns to zero within ~1–2 ps. At negative delay times, several satellite peaks of transmittance variations are observed for the pump energy > 2 pJ, and their intensities increase with the increase in the pump energy. So these satellite peaks may come from the pulse shape changes induced by the fiber nonlinearity that is apparent under high-energy pump. Judging from the 0.6 ps full width at half maximum of the absorption saturation recovery, the waveguide can work at bitrates of the order of 800 Gbps-1 Tbps.
The absorption saturation is influenced by ISBT broadening mechanism that is related to various intersubband relaxation processes.  So we briefly discuss the potential homogeneous and inhomogeneous origins in our system (EISBT > ħωLO). Well-to-well thickness fluctuations and subband nonparabolicity are fundamental inhomogeneous broadening mechanism. For our sample growth, we did not observe the well-thickness-fluctuation induced peak separation even when the well thickness was systematically increased. Compared to III-V, this system has a weaker subband nonparabolicity based on the energy-dependence effective mass . Furthermore, the sheet density is ~1011–1012 cm-2 in our sample, and the many-body effects will cancel the effect of subband nonparabolicity on inhomogeneous broadening . So homogeneous broadening may be the main broadening mechanism even though the shape of the absorption spectrum (300 K) in Fig. 4 is not pure Lorentzian line. The decay time constant is ~0.2 ps, dominantly due to LO-phonon scattering, as referred to Ref. . Carrier relaxation in our current quantum system is free from the influence of X valley of BeTe , which is different from the observed carrier-relaxation mediation of indirect valleys .
Based on the composition optimization of the cladding layers and the introduction of a new ZnSe/BeTe MQW optical confinement layer in II–VI-based ISBT waveguide switches, the refractive index contrast between the core layer and the cladding layers was increased to ~0.2 and the thickness of the active layer was reduced to 0.072 µm. Beam propagation simulation revealed that the above changes could greatly improve the saturation properties. We experimentally confirmed that the 10 dB absorption saturation energy of such ultrafast switches was further decreased to 2.0–2.2 pJ for 1.55–1.58 µm. Further, the 10 dB switching operation at 1566 nm was achieved under only <7 pJ pump energy by using a control pulse wavelength of 1520 nm. The above switching energy (<0.7 pJ/dB) is the lowest reported value in ISBT ultrafast all-optical switches. These results are promising for packing waveguides into all-optical signal processing devices.
References and links
1. S. Noda, T. Uemura, T. Yamashita, and A. Sasaki, “All-optical modulation using an n-doped quantum-well structure,” J. Appl. Phys. 68, 6529–6531 (1990). [CrossRef]
2. N. Suzuki and N. Iizuka, “Feasibility study on ultrafast nonlinear optical properties of 1.55-µm intersubband transition in AlGaN/GaN quantum wells,” Jpn. J. Appl. Phys. 36, L1006–1008 (1997). [CrossRef]
3. J. Hamazaki, S. Matsui, H. Kunugita, K. Ema, H. Kanazawa, T. Tachibana, A. Kikuchi, and K. Kishino, “Ultrafast intersubband relaxation and nonlinear susceptibility at 1.55 µm in GaN/AlN multiple-quantum wells,” Appl. Phys. Lett. 84, 1102–1104 (2004). [CrossRef]
4. S. Sekiguchi, T. Simoyama, H. Yoshida, J. Kasai, T. Mozume, and H. Ishikawa, “Waveguide design of InGaAs/AlAs/AlAsSb intersubband transition optical switch,” Technical Digest 2005 Optical Fibre Communication Conference (Optical Society of America, Washington, DC, 2005) paper OFE4.
5. T. Simoyama, S. Sekiguchi, H. Yoshida, J. Kasai, T. Mozume, and H. Ishikawa, “Absorption dynamics in all-optical switch based on intersubband transition in InGaAs-AlAs-AlAsSb coupled quantum wells,” IEEE Photo. Technol. Lett. 19, 604–606 (2007). [CrossRef]
6. N. Iizuka, K. Kaneko, and N. Suzuki, “Sub-picosecond modulation by intersubband transition in ridge waveguide with GaN/AlN quantum wells,” Electron. Lett. 40, 962–963 (2004). [CrossRef]
8. N. Iizuka, K. Kaneko, and N. Suzuki, “All-optical switch utilizing intersubband transition in GaN quantum wells,” IEEE J. Quantum Electron. 42, 765–771 (2006). [CrossRef]
9. Y. Li, A. Bhattacharyya, C. Thomidis, T.D. Moustakas, and R. Paiella, “Nonlinear optical waveguides based on near-infrared intersubband transitions in GaN/AlN quantum wells,” Opt. Express 15, 5860–5865 (2007). [CrossRef]
10. R. Akimoto, B.S. Li, K. Akita, and T. Hasama, “Subpicosecond saturation of intersubband absorption in (CdS/ZnSe)/BeTe quantum-well waveguides at telecommunication wavelength,” Appl. Phys. Lett. 87, 181104 (2005). [CrossRef]
11. K. Akita, R. Akimoto, T. Hasama, H. Ishikawa, and Y. Takanashi, “Intersubband all-optical switching in submicron high-mesa SCH waveguide structure with wide-gap II-VI-based quantum wells,” Electron. Lett. 42, 1352–1353 (2006). [CrossRef]
13. H.C. Liu and F. Capasso, Intersubband Transitions in Quantum Wells: Physics and Device Applications I (Academic Press, 2000), Chap. 1.
14. P. Harrison, Quantum Wells, Wires and Dots: theoretical and computational physics (John Wiley & Sons Ltd, 2000), Chap. 3.
15. R. Akimoto, K. Akita, F. Sasaki, and T. Hasama, “Sub-picosecond electron relaxation of newar-infrared intersubband transitions in n-doped (CdS/ZnSe)/BeTe quantum wells,” Appl. Phys. Lett. 81, 2998–3000 (2002). [CrossRef]
16. C.V.-B. Tribuzy, S. Ohser, S. Winnerl, J. Grenzer, H. Schneider, M. Helm, J. Neuhaus, T. Dekorsy, K. Biermann, and H. Künzel, “Femtosecond pump-probe spectroscopy of intersubband relaxation dynamics in narrow InGaAs/AlAsSb quantum well structures,” Appl. Phys. Lett. 89, 171104 (2006). [CrossRef]