1. Introduction

Hydrogenated amorphous silicon (a-Si:H) is a highly promising material for power efficient nonlinear photonic devices due to its ultrahigh optical nonlinearity [1–6] and low nonlinear loss. Recently, using an a-Si:H nanowaveguide we measured the nonlinear refractive index to be an order of magnitude larger than the value of single-crystal silicon (c-Si) [5]. Furthermore, Kuyken et al [2] measured a nonlinear figure of merit (FOM) of 2.2 for a-Si:H, which is four times higher than the FOM for c-Si indicating that the impact of two-photon absorption is significantly reduced relative to c-Si. Due to these encouraging nonlinear properties, a-Si:H can considerably decrease the power requirements for nonlinear photonic devices. For example, the maximum effective nonlinearity of a-Si:H nanowaveguides is greater than 3000 W⁻¹m⁻¹ [5]; a value that is five orders of magnitude larger than highly nonlinear optical fiber [7] and an order of magnitude larger than ultrahigh-nonlinearity c-Si [8] and chalcogenide waveguides [9]. This extremely large effective nonlinearity indicates that comparable nonlinear efficiency can be achieved in a-Si:H nanowaveguides with greatly reduced power requirements in more compact devices.

A further advantage of this material platform includes the low-temperature deposition of a-Si:H, making it back-end compatible with multilayer integration using current complementary–metal–oxide–semiconductor (CMOS) fabrication processes [10]. In contrast, c-Si requires epitaxial growth or wafer bonding to define the photonic layer thereby limiting integration of photonic components to solely the base layer of a multilayer integrated circuit. The ability to integrate a-Si:H devices into any layer or multiple-layers within an integrated circuit is critical for resource-intensive ultrafast nonlinear photonic systems as it allows for integration of a larger number of photonic devices within a smaller footprint and facilitates tight integration with CMOS electronics.

Nonlinear photonic systems such as those required to demultiplex high-speed optical time-division multiplexed (OTDM) signals are typically extremely resource intensive in terms of size, power and cost. To minimize the resource requirements, integrated nonlinear photonic devices are highly attractive due to the potential for ultra-compact chip-scale systems and their high effective nonlinearity relative to fiber systems. Specifically, chalcogenide (As₂S₃) waveguides have been used to demultiplex 1.28 Tb/s and 160 Gb/s data streams using peak optical control powers of 4 W and 4.4 W and waveguide lengths of 7 cm and 5 cm, respectively [11,12]. Additionally, c-Si nanowaveguides were used to perform demultiplexing of 1.28 Tb/s and 160 Gb/s data streams to 10 Gb/s channels using peak powers of 2 W and 0.5 W and waveguide lengths of 5 mm and 1.1 cm, respectively [13,14]. Semiconductor optical amplifiers (SOA) have been successfully used in demultiplexing [15] with low optical switching powers, but the device itself requires additional power for amplification. Using slow light photonic crystals, the enhancement of the nonlinearity can greatly reduce the length of the waveguide, but the power requirement is still relatively high (~0.9 W) [16]. Silicon nano-crystals can achieve a very high nonlinearity (n₂ > 10¹² cm²/W) and have been investigated in slot waveguide geometries, however these structures typically exhibit much higher propagation losses (20 dB/cm) [17]. Lower switching energies have been achieved in photonic crystal resonant cavities [18] and ring structures [17], but the operating bandwidth is inherently limited due to the resonant cavity structure. An additional approach using a silicon-organic hybrid structure can potentially overcome the two photon absorption (TPA) induced free carrier absorption (FCA), but the high propagation loss and complex fabrication limit the device performance [19] and as a result, error free operation has not yet been demonstrated.

Here, we perform the first experimental investigation of demultiplexing high-speed optical time division multiplexed (OTDM) data signals in an a-Si:H nanowaveguide. Using four-wave mixing, we demonstrate demultiplexing of a 160-Gb/s OTDM data signal to 10-Gb/s with error-free performance for all 16 channels. Due to the ultra-high nonlinearity of the a-Si:H device investigated here, we are able to achieve error-free operation in a 6-mm-long nanowaveguide using ultralow peak powers of only 50 mW for the switching pump pulse. This power requirement is an order of magnitude lower than previous CMOS-compatible integrated device demonstrations for OTDM demultiplexing.

2. Device design and fabrication

In our a-Si:H device, all-optical switching relies on the third-order nonlinear parametric process known as four-wave mixing (FWM). In order to achieve broad-bandwidth FWM, the nanowaveguide is designed to obtain phase matching by having low anomalous group-velocity dispersion (GVD) or near-zero group-velocity dispersion (ZGVD) at the operating wavelength [20]. The cross-section and quasi-transverse electric (TE) mode profile of the designed nanowaveguide is modeled by the finite difference method [21] and is shown in Fig. 1(a) . The designed waveguide is about 198 nm thick by 500 nm wide. The GVD value for the TE mode of the dispersion-engineered waveguide is calculated to be ~10 ps/(nm∙km) at a wavelength 1550 nm with a dispersion slope of −0.43 ps/(nm²∙km) (Fig. 1); corresponding to a FWM conversion bandwidth of more than 150 nm [5]. The ZGVD wavelengths for the designed waveguide are 1500 nm and 1565 nm.

 

Fig. 1 Calculated GVD of the designed a-Si:H waveguide. The waveguide is designed to have low anomalous GVD at 1550 nm for efficient and broadband four-wave-mixing. Inset: (a) Cross-sectional schematic and simulated quasi-TE mode profile of the a-Si:H waveguide. The designed waveguide is 198 nm thick and 500 nm wide. (b) SEM image of the fabricated waveguide with ~100-nm silicon dioxide hard mask on top.

Download Full Size | PDF

The a-Si:H nanowaveguide is fabricated using standard microelectronics fabrication techniques. The a-Si:H film is deposited by plasma-enhanced chemical vapor deposition (PECVD) on a silicon wafer with 3 µm buried oxide (BOX). The substrate is maintained at 300°C during the deposition. A thin layer of silicon dioxide (~100 nm) is deposited as a hard mask to reduce effects from direct etching with organic resists. Electron-beam lithography followed by chlorine-based inductively coupled plasma (ICP) etching is used for waveguide patterning. A thick silicon dioxide layer (~1 µm) was deposited over the waveguide for cladding and for environmental protection. Inverse adiabatic tapers on both ends of waveguide are made for optical coupling [22]. The scanning electron microscope (SEM) image of the fabricated waveguide (Fig. 1 inset (b)) shows the near-rectangular shape and smooth sidewalls to reduce scattering losses in the waveguide. Additionally, all of the fabrication steps utilize CMOS techniques and are back-end-of-the-line (BEOL) CMOS compatible. Using the cut-back method, the coupling and propagation loss of the waveguides are extracted to be 8 dB per facet and 3.5 dB/cm for the quasi-transverse electric (TE) mode, respectively. Additionally, the effective nonlinearity of the waveguide is measured using the method described in [5] to be approximately 3000 W⁻¹m⁻¹.

In previous investigations of a-Si:H devices, a critical concern has been device degradation from exposure to high infrared optical powers. For our devices, no degradation is observed as a result of infrared optical power, however, the linear propagation loss of the samples degrade over time (on the order of months) after long exposure to visible light. This degradation is well-known in the solar cell community as the Staebler–Wronski effect [23] and can be avoided by isolating the sample from visible light or adding an opaque cover for packaging. We found no correlation between degradation and exposure to high average power (300 mW) infrared light in the waveguide for hundreds of hours of testing.

3. Experiment

We demonstrate OTDM demultiplexing from a 160 Gb/s on-off-keying (OOK) data stream to 10 Gb/s with the a-Si:H nanowaveguide device using the experimental system shown in Fig. 2 . An erbium-doped harmonically mode-locked fiber laser (MLFL) set to 1560 nm with a repetition rate of 10 GHz generates both the OTDM test source and the control pulses. An 80/20 coupler splits the test signal and pump sources, respectively. An electro-optic modulator (EOM) encodes the test signal with a 2³¹-1 pseudorandom bit sequence (PRBS) and is then multiplexed up to 160 Gb/s using four highly asymmetric fiber Mach-Zehnder multiplexer (MUX) stages prior to being combined with the control pulses. The MUX stages are set to preserve a 2⁹-1 PRBS sequence in the resulting OTDM signal. During the demux experiment, no change in the BER performance was observed when modulating at either 2⁹-1 or 2³¹-1 PRBS. The 20% side of the 80/20 coupler is spectrally broadened through self-phase modulation in 800 m of highly nonlinear fiber (HNLF) and subsequently filtered at a central wavelength of 1551 nm with a 4-nm bandwidth to generate the control pulses. A tunable delay (ΔT) and polarization controllers (PC) allow the test source and control pulses to be aligned in time and to be matched to the TE polarization of the waveguide. Compressing fibers are inserted at each arm to keep the control and test pulses transform-limited. The control and test sources are then combined using a wavelength division multiplexer (WDM). At the input of the waveguide, a lensed fiber is used for fiber-to-chip coupling. The waveguide output is sent into a receiver assembly, which consists of a 100-GHz optical filter centered at 1541 nm and optical amplifier followed by a second identical bandpass filter and an avalanche photo-detector. The generated idler pulses are isolated and amplified prior to detection. The second bandpass filter is used to reduce the amplified spontaneous emission (ASE) from the amplifier. The detected signal is sent into a bit-error-rate tester (BERT) for error rate measurement. A 10% tap is implemented before the photo-detector for monitoring the received power during the BER measurement. During testing, an optical spectrum analyzer captures the spectrum before and after the waveguide. In the back-to-back (B2B) measurement, the mode-locked fiber laser is tuned to the idler wavelength (1541 nm), and the MUX stages are bypassed. These laser pulses are then sent directly into the receiver assembly of the experiment.

 

Fig. 2 Experimental setup for 160-to-10 Gb/s all-optical demutiplexing using an a-Si:H waveguide (MLFL: mode-locked fiber laser. EOM: electro-optical modulator. TBPF: tunable bandpass filter. MUX: Mach-Zehnder multiplexer. PC: polarization controller. HNLF: highly nonlinear fiber. WDM: wavelength division multiplexer. PD: photo-detector. BERT: bit-error rate tester.). Inset (lower left): auto-correlation traces of 160-Gb/s signal and 10-GHz pump at the input of the waveguide.

Download Full Size | PDF

The optical spectra before and after the waveguide are shown in Fig. 3(a) . The data stream and pump laser are separated in wavelength by 10 nm. The average power for pump and signal inside the waveguide are 1.2 mW (0.8 dBm) and 0.8 mW (−0.9 dBm), respectively. The pulse widths of the pump and data pulses are measured through autocorrelation to be 1.9 ps and 2.1 ps (Inset of Fig. 2), respectively, corresponding to 63 mW peak power for the pump in the waveguide assuming a Gaussian shape pulse. At the output of the waveguide, the 160 Gb/s data is demultiplexed to 10 Gb/s at a wavelength of 1541 nm through FWM. The on/off conversion efficiency of the four-wave mixing process is measured to be −13 dB, a value competitive with state-of-the-art demonstrations in c-Si. The on/off conversion efficiency is defined by the ratio between idler output power with the pump on and signal output power with the pump off, taking into account the duty cycle difference between the signal (160 Gb/s) and the idler (10 Gb/s) of 12 dB. The BER performance of this device for all 16 channels is shown in Fig. 3(b) demonstrating error-free operation with a BER of less than 10⁻⁹ and a power penalty ranging from 4 to 5 dB relative to the back-to-back (B2B) measurements. For reference, a demultiplexed eye diagram when the BER = 10⁻⁹ is shown in the inset of Fig. 3(b).

 

Fig. 3 (a) Input and output spectra of the demultiplexing process in the 6-mm-long a-Si:H waveguide showing the input 160-Gb/s signal, the 10 GHz pump, and the generated 10 Gb/s idler. (b) The BER measurement of the 160 Gb/s to 10 Gb/s demultiplexing of all 16 channels and 10 Gb/s back-to-back (B2B). Error-free operation (10⁻⁹) is achieved with 4- to 5-dB power penalty. Inset: demuxed eye diagram.

Download Full Size | PDF

To study low power operation of the device, we investigate the minimum switching power necessary for error free demultiplexing (Fig. 4 ). While keeping the coupled OTDM signal power in the waveguide constant (0.8 mW average power), the pump is attenuated to determine the minimum required switching power. As the pump is attenuated the conversion efficiency drops and the signal-to-noise ratio of the signal coupled out of the device suffers leading to increased error rate. Intriguingly, for this measurement error free operation is maintained with peak switching powers as low as 50 mW (17 dBm). This is an order of magnitude lower than previous CMOS-compatible integrated device demonstrations [11–14]. Furthermore, with improvements to the output coupling efficiency of the waveguide (currently 8 dB) an even lower minimum switching power is possible.

 

Fig. 4 Bit-error rate as a function of coupled peak pump power. Error-free operation (10⁻⁹) is achieved with a 17-dBm (50 mW) peak switching power.

Download Full Size | PDF

4. Conclusion

Here we demonstrate 160 Gb/s to 10 Gb/s all-optical demultiplexing via four-wave mixing in a 6-mm long highly nonlinear hydrogenated amorphous silicon waveguide with error free operation at telecommunication wavelengths using ultralow peak pump powers of 50 mW. This represents the first demonstration of OTDM demultiplexing in a a-Si:H device and the switching power of our device is, to the best of our knowledge, the lowest among all CMOS-compatible platforms demonstrated to date. Excitingly, this material platform can also directly benefit from recent advancements in OTDM demultiplexing such as slow-light enhancement to further reduce the device size [16] and direct OTDM-to-WDM (wavelength division multiplexing) conversion to reduce the number of physical switches [24–26]. Furthermore, the low deposition temperature of the a-Si:H material used here allows such devices to be fabricated at the back-end-of-the-line of a CMOS process, enabling their seamless integration with microelectronics. The demonstrated ability to manipulate ultrahigh data-rate signals using ultralow powers in a low temperature deposited waveguide offers exciting prospects for sophisticated multilayer on-chip all-optical signal processing circuits.