1. Introduction

With the continuous growth of data traffic at the optical communication systems as well as high-performance computing and storage area networks, optical packet switching (OPS) has been recognized as a viable approach to solve the problems of power consumption and bandwidth limitation imposed at electronic routers [1]. To fully benefit from the vast bandwidth and small latency of OPS, the wavelength-division-multiplexed (WDM) OPS has been studied intensively recently [2–5]. In WDM-OPS, a broadband payload is encoded on multiple wavelengths and routed using wavelength-independent space switches. Since the routing is performed without wavelength demultiplexing and multiplexing, the number of switches and other optical components can be reduced significantly. With the mature WDM transmission technologies exceeding several Tbps of optical bandwidth, single packet can be compressed into an extremely fine temporal granularity, which greatly relaxes the requirement for optical buffers. In addition, by coding the optical label on a separate wavelength, we can easily realize label extraction and reinsertion using an optical bandpass filter or add-drop multiplexer.

For the implementation of WDM-OPS, large-scale space switches with wide optical bandwidth and nanoseconds response will be indispensable. Conventionally, large-scale 1×N and N×N integrated switches have been constructed by cascading 1×2 switches in tree or crossbar architectures [6, 7]. The accumulation of insertion loss in the switching elements, however, leads to increasing optical loss with N, which may limit the scalability of these approaches. Alternatively, broadcast-and-select-type 1×N switches with semiconductor-optical-amplifier (SOA) gate switches have been demonstrated [2, 4]. They offer large extinction ratio and small net insertion loss, which are highly attractive for practical implementation. On the other hand, the degradation in optical signal-to-noise ratio (OSNR) and nonlinear crosstalks caused by gain saturation in SOAs are the drawbacks of these switches. The nonlinear effects in SOA would particularly be detrimental for WDM packets, where the four-wave mixing and cross-gain modulation may cause undesired inter-channel crosstalks.

We have recently demonstrated a novel type of monolithically integrated InP photonic switch based on optical phased array [8]. The phased-array switch enables 1×N switching in a single phase-modulating stage, thus may offer potential advantage over the tree switches in terms of size and optical insertion loss for large N [9]. Unlike the SOA-based switches, it consists of only phase-modulating sections for switching, thus is free from nonlinear signal distortion and inter-channel crosstalks. Simple device structure without requirement of active/passive integration also contributes significantly to the high-yield low-cost fabrication. Our proof-of-concept 1×5 InP switch has successfully demonstrated the potential of wideband switching covering the entire C-band (1520–1580 nm) with nanoseconds dynamic response [8].

In this paper, we present the first subsystem demonstration of WDM-OPS using the InP optical phased-array 1×5 switch. By injecting a total current of less than 25 mA, a 320-Gbps (8×40-Gbps) multi-colored payloads are switched to the five output ports with lower than 1.3 dB of penalty, free from any nonlinear signal distortion or inter-channel crosstalks.

2. Structure and fabrication of 1×5 InP/InGaAsP optical phased-array switch

The photograph and the layer structure of the fabricated 1×5 optical phased-array switch are shown in Fig. 1. The mask design is identical to the one reported previously [8], while the total thickness of the upper InP cladding is increased to 1.1 μm to reduce the absorptive loss in the contact p-InGaAs layer at the phase-modulating sections. Input light is dispersed spatially at Slab 1, and directed to the eight arrayed phase shifters. By modulating the optical phase linearly (with modulo 2π) across the array, we can dynamically control the focusing position of light at Slab 2. By designing the array structure so that each array has equal optical path length, the switch operates inherently independent on the wavelength, enabling ultra-broadband switching of WDM packets [8]. The phase modulation is achieved through current injection to the un-doped bulk InGaAsP core layer (Q1.3), having the energy bandgap at 1.3-μm wavelength.

The switch was fabricated by metal-organic vapor-phase epitaxy (MOVPE) growth, followed by wet-chemical etching processes. The fabricated switch was cleaved, bonded on an AlN chip carrier, and placed on a temperature-controlled stage, which was set to 20 °C during the experiments. The eight electrodes were wire-bonded and connected to an 8-port function generator (FG). From the measurement on a test phase modulator, we confirmed that the phase shift of 2π was obtained with less than 7 mA of current injection.

 

Fig. 1. Top photograph (a) and device structures (b) of the fabricated InP/InGaAsP 1×5 optical phased-array switch.

Download Full Size | PDF

3. Static switching characterization

We first examined static switching properties of the switch. A continuous-wave (CW) light at 1.55-μm wavelength and transverse-electric (TE) polarization state was injected to the switch. Optimal conditions of current injection were found by scanning the DC voltages of FG output while monitoring the optical power from each output port. With the optimized conditions, denoted as State 1–5 hereafter, successful switching to Port 1–5 was achieved with extinction ratio of 15.5 dB (average) and 11.3 dB (worst case). The total current injection was estimated to be less than 25 mA for all states, with the static power consumption below 31 mW. The fiber-to-fiber loss ranged from 16.4 to 18.1 dB depending on the output ports. From the Fabry-Perot propagation-loss characterization, we estimated that the coupling loss from the fiber to the chip accounted for 5–6 dB/facet, which makes the on-chip loss of the device to be around 5–7 dB. The total loss can thus be reduced by applying appropriate anti-reflection (AR) coating and spot-size converters at the input and output waveguides. The polarization-dependent-loss of the device was around 4–6 dB. This is attributed mainly to the polarization-dependent differences in the bending loss and fiber-to-waveguide coupling efficiency, which should be reduced by using dry-etched waveguide [10]

We then investigated the transmittance of 320-Gbps (40-Gbps × 8-channel) WDM signal through the switch. The experimental setup is shown in Fig. 2. In the transmitter, the output from eight WDM laser sources (in 200-GHz grid from 1546.1 nm to 1557.3 nm) were multiplexed and modulated using a LiNbO₃ Mach-Zehnder modulator (LN-MZ) with a 40-Gbps non-return-to-zero (NRZ) pseudorandom binary sequence (PRBS) of length 2⁷-1. Eight different lengths of fiber delay lines (FDL) were employed to decorrelate the data patterns among channels.

Figure 3(a) shows the receiver sensitivity at the bit-error-rate (BER) of 10^-9 measured for all 40-Gbps channels in State 1–5. In each switching state (State 1–5), the output light from the corresponding port (Port 1–5) was measured. As example cases, the BER curves and eye diagrams of Ch1 (1546.1 nm), Ch5 (1552.5 nm) and Ch8 (1557.3 nm) are displayed in Fig. 3(b)–(d). The receiver penalty is less than 1.3 dB for all cases. The fluctuation in the sensitivity and negative penalties at some WDM channels are attributed to the Fabry-Perot optical interference caused by the light reflection at the device facets. We therefore expect that these penalties should be suppressed by the proper AR coating of the device facet.

 

Fig. 2. Experimental setup for 320-Gbps WDM static switching measurement. LN-MZ: LiNbO₃ Mach-Zehnder modulator, PC: polarization controller, EDFA: erbium-doped fiber amplifier, VOA: variable optical attenuator, DEMUX: wavelength demultiplexer, BER: bit-error-rate tester.

Download Full Size | PDF

 

Fig. 3. (a). Receiver sensitivity at BER of 10^-9 for all WDM channels at back-to-back (BB) and at five output ports (Port1–5) measured under static switching conditions (State1–5). (b–d) BER curves and 40-Gbps eye-diagrams of Ch1 (1546.1 nm), Ch5 (1552.5 nm), and Ch8 (1557.3 nm) for all states.

Download Full Size | PDF

Figure 4 shows the results of dynamic power range measurements for single wavelength (Ch5, circles), two wavelengths (Ch5 and Ch4, dots), and eight wavelengths (Ch1–Ch8, triangles). For all cases, the switch was set to State 1 and the output signal from Port 1 was measured with increasing input power. Note, for ease of comparison, the power penalty is plotted as a function of input power of Ch5, which should correspond to the total input power divided by the number of channels. At low input power below 0 dBm, the penalty increases due to the OSNR degradation. On the other hand, there is no increase in penalty for large input power. Also, the penalty curves are almost independent on the number of WDM channels. These facts indicate that the switch is free from any nonlinear effect, on the contrary to SOA-based switches, where the gain saturation and four-wave mixing inside the SOAs typically impose strict limitations to the dynamic range of input signal power [11, 12]. From Fig. 4, the dynamic power range of the phased-array switch was more than 10 dB for 2 dB penalty, which was only limited by the maximum output power of the input erbium-doped fiber amplifier (EDFA) used in the experiment.

 

Fig. 4. Receiver penalty at BER of 10^-9 as a function of input signal power of Ch5 (1552.5 nm) for the cases of single wavelength (Ch5, circles), two wavelengths (Ch5 and Ch4, dots), and eight wavelengths (Ch1-Ch8, triangles). The switch is set to State 1 and the output signal from Port 1 is measured.

Download Full Size | PDF

4. Demonstration of 320-Gbps WDM-OPS testbed

We next performed dynamic switching experiment using segmented WDM payloads attached with optical labels to test the applicability to WDM-OPS. Since a field-programmable gate array (FPGA) circuit equipped with 8-port digital-analog converters (DACs) was not available at the time of experiment, we carried out a simplified OPS experiment; the label consisted of only one bit and the switch was converted between two states whenever the label “1” arrived.

The experimental setup for WDM-OPS is shown in Fig. 5. The 320-Gbps WDM payload was combined with an optical label at 1544.5 nm, segmented into packets using another LN-MZ, aligned to a TE-polarization mode, and injected into the 1×5 switch. The packet length was set to 800 ns with a guard time of 200 ns. At the router, the label was extracted using a narrowband optical bandpass filter and detected with a photodetector (PD). The electric signal from PD was injected directly as a trigger signal to the 8-port FG, which was operated in the sequential mode to control the analog voltages applied to the eight electrodes of the switch. The voltage conditions for State 1 and State 3 were saved in the internal memory of FG, so that the routing port would switch between Port 1 and Port 3 whenever the trigger arrived.

 

Fig. 5. Experimental setup for 320-Gbps WDM-OPS demonstration.

Download Full Size | PDF

Figure 6 shows the packet waveforms observed at the input and output ports of the switch for all WDM channels (before demultiplexing), and Ch1, Ch5, and Ch8 (after demultiplexing). Thanks to the wide bandwidth of the switch, the entire 320-Gbps WDM packet was switched successfully with dynamic extinction ratio larger than 12 dB. By using a fast FPGA circuit with 8-port DACs, complete label recognition and dynamic switching to all the output ports should be possible.

 

Fig. 6. Waveforms of 320-Gbps WDM optical packets (all channels, Ch1, Ch5, and Ch8) at the switch input (left column), output Port 1 (center column), and Port 3 (right column).

Download Full Size | PDF

Although the demonstrated switch has insufficient crosstalk characteristics for practical implementation, it should be improved by using highly confined dry-etched waveguides, more advanced iterative algorithm in deriving the phase shifter conditions, and appropriate apodization to the slab interfaces, which is a common technique used in arrayed-waveguide gratings. We should also note that owing to the unique phased-array architecture, the switch should scale to larger port count with relatively compact dimension [9, 10], making it an attractive building element in constructing large-scale integrated switch matrices.

5. Conclusions

We have fabricated a monolithically integrated 1×5 optical phased-array switch on InP and demonstrated its applicability to broadband WDM-OPS. Owing to the wide operating bandwidth of the switch, error-free forwarding of 320-Gbps (40-Gbps × 8 channel) WDM signal has been achieved with less than 1.3-dB power penalty. Since the device consists of only phase-modulating sections for switching, it is free from nonlinear signal distortion and inter-channel crosstalks, allowing large dynamic range of input power. These results support the potential of phased-array switches in realizing the future large-scale broadband WDM-OPS router on chip.