1. Introduction

Silicon photonics enjoys benefits of accessible and low-cost integration with CMOS Electronics, allowing large on-chip integration densities. To accommodate increasing bandwidth demand particularly at shorter reaches, on-chip multiplexing techniques such as Mode Division Multiplexing [1–4] and Wavelength Division Multiplexing (WDM) [5–10] may be leveraged. There are 2 types of WDM – Dense WDM (DWDM) and Coarse WDM (CWDM), the specifications of which are standardized by the International Telecommunications Union Telecommunication Standardization Sector (ITU-T). DWDM has higher spectral efficiency and is used mainly in long-haul applications. However, their stringent spectral requirements lead to wavelength trimming and tuning being required to ensure the wavelength alignment [11], which results in higher costs and power consumption. On the other hand, CWDM is used mainly for short-reach metropolitan networks, Fibre-To-The-Home (FTTH), and Ethernet. Furthermore, 1 by 4 CWDM has been adopted by both the CWDM4 Multi-Source Agreement [12] and CLR4 alliance [13] in their 100G interconnect solutions. The ITU-T G.694.2 standard has specified a 20-nm channel-spacing grid with specific center wavelengths. The wider channel spacing compared to DWDM allows better temperature and fabrication tolerance compared to the finer grid DWDM. This allows transmission of high-speed data in the optical fiber domain using low-cost uncooled lasers, thus minimizing the associated link budget.

Various CWDM devices have been proposed such as the arrayed waveguide gratings [14], Mach-Zehnder interferometers [15,16], and directional couplers [17]. However, the devices are not flat-top, are sensitive to phase error or have large footprint. Moreover, as in the majority of integrated photonics, previous work is largely based on standard silicon as the propagating core. However, future all-optical information processing requires harnessing material nonlinearities, needed for next generation of communication systems. Silicon is known to have nonlinear losses such as two-photon absorption and free-carrier effects [18]. This reduces the efficiency of nonlinear processes required for application such as all-optical modulators [19], wavelength converters [20,21], optical switching [22] and supercontinuum sources [23,24].

In this paper, we implement an 8-channel CWDM using FSR-free coupled vertical gratings with its sidewalls sinusoidally corrugated [5] on a bandgap-engineered ultra-silicon-rich nitride (USRN) platform [25–27]. High speed testing with data at 10Gb/s and 30Gb/s yields clear eye diagrams and error free operation with a power of −14dBm and −12dBm respectively. USRN possesses a large Kerr nonlinearity and negligible two-photon absorption losses at the telecommunications wavelengths, rendering it an ideal candidate for combining CWDM with nonlinear optics applications. Consequently, we demonstrate an 8-channel multi-wavelength source using a two-stage device consisting of a nonlinear USRN waveguide generating supercontinuum followed by the 1 by 8 CWDM.

2. Device design

Figure 1(a) shows the schematic of the device implemented using a USRN core (n = 3.1) and SiO₂ cladding (n = 1.46). Two sinusoidally corrugated waveguides are placed parallel to each other. The coupled vertical gratings possess a grating length, L of 450μm and grating pitch, Λ. w₁ and w₂ are the waveguide widths, set at 650nm and 500nm respectively, separated at their closest point by gap g. The gratings are apodized using raised cosine filter, $a=a_{\max }{\sin }^2\left(\frac{\pi z}{L}\right)$, to reduce sidelobes that introduces undesirable crosstalk. a and z refer to grating amplitude and propagation direction respectively. a has a value of zero at the start of the device and increases to its maximum amplitude, a_max at the middle of the gratings. a_max is set to 50nm. The footprint for each CWDM channel is 600μm². When light located at the cross-coupling wavelength is launched into the device, the light will be cross-coupled to the counter-propagating direction in w₂ as shown in the large arrow in Fig. 1(a). This drop wavelength, λ_d, is governed by the cross Bragg coupling condition,

 

Fig. 1 (a) Schematic and (b) scanning electron micrograph of a single CWDM block. (c) Calculated effective index versus waveguide width and (d) sum of propagation constants versus wavelength. The dashed lines denote wavelengths at which Eq. (2) is satisfied and the corresponding CWDM wavelengths.

Download Full Size | PDF

A scanning electron micrograph of a coupled vertical grating is shown in Fig. 1(b). n_eff1 and n_eff2 refer to the effective indices of w₁ and w₂ respectively, and their calculated values are shown in Fig. 1(c). By multiplying Eq. (1) with $\frac{2\pi }{\lambda }$, we obtain [28]

The full CWDM device incorporates 9 cascaded blocks of coupled vertical gratings. Nine coupled vertical gratings with different drop wavelengths were designed and cascaded such that the corresponding values of λ_d cover the central CWDM wavelength grid specified by the ITU-T. The drop wavelengths span from 1451nm to 1611nm, with 20nm intervals; This spans S + C + L bands. The first block multiplexes light centered at 1611nm (Port1), the second, 1591nm (Port2), and so on, and further as described in Table 1. The last block drops 1471nm (Port 8). We note also that the gratings are reciprocal – they can function as a multiplexer or a demultiplexer. We make use of Eq. (2) to plot (β₁ + β₂) as a function of wavelength in Fig. 1(d). The intersection point of each dropped wavelength λ_d corresponds to the calculated $\frac{2\pi }{\Lambda }$ value on the vertical axis.

Table 1. Results of designed and measured center wavelength and 3dB Bandwidth

View Table

During fabrication, 300nm thick USRN is first grown on a silicon substrate with 10μm thermal SiO₂ layer using Inductively Coupled Plasma Chemical Vapor Deposition at a low temperature of 250°C [26]. Next, the CWDM device was fabricated using electron-beam lithography, followed by reactive ion etching, and plasma enhanced chemical vapor deposition of silicon dioxide as the over-cladding.

During measurement, quasi-TE polarized light from a tunable laser (TL) source spanning from 1450nm to 1630nm was launched into the central waveguide (w₁). The individual drop ports were measured using a TL-synchronized power meter. Figure 2 shows the measured CWDM drop port spectra. Table 1 shows the designed and measured center wavelength and 3dB bandwidth.

 

Fig. 2 Transmission spectrum of the 1 by 8 CWDM device.

Download Full Size | PDF

The designed λ_d are the same as CWDM standard wavelengths stipulated in ITU-T G.694.2. It is observed that the measured CWDM channels effectively cover the standardized wavelengths and the designed center wavelengths match the measured ones well. The average channel crosstalk and 3dB channel bandwidth are −25dB and 7.7nm respectively. The insertion loss for each add/drop filter is small, at 0.15dB. There is some loss from the access waveguides for each individual drop port. The waveguide loss is 3dB/cm, and therefore loss for each channel including the filter loss and access waveguides is about 1dB. The insertion loss can be further reduced by reducing the access waveguide lengths for each add/drop. The passband edge is observed to have a steep roll-off. At Port 1 (1611nm band), the transmission drops 8dB in 0.3nm. This is due to the highly wavelength selective feature of coupled vertical gratings [28]. It is noted that Port 7 (1491nm) has a small 3dB bandwidth of only 4nm. This is likely attributed to the fabricated gap, g being wider than designed, as studied in Ref [5]. In Fig. 2, there are some oscillations observed within the passband of Port 4 (1551nm), Port 3 (1571nm) and Port 2 (1591nm). Imperfection apodization is likely to be the cause of these oscillations. As further documented in section 3, we were still able to successfully measure the open eye diagram and measure the bit error rate as shown in Figs. 3 and 4. Part of the ninth WDM block is outside the wavelength range of our tunable laser, and therefore could not be fully characterized. Consequently, we claim the demonstration of a 1 by 8 CWDM device.

 

Fig. 3 (a) Performance Evaluation using Commercial SFP + transceiver at 10Gb/s, (b) Measured eye diagram and (c) Plot of -log(BER) as a function of power

Download Full Size | PDF

 

Fig. 4 (a) Experimental setup for 30Gb/s high speed BER Eye diagram evaluation (b) Measured eye diagram and (c) plot of -log(BER) as a function of power.

Download Full Size | PDF

3. Performance characterization

High speed testing of the 1 by 8 CWDM device was performed using 2 sources – first using a commercial Small Form Pluggable (SFP + ) transceiver, secondly using a Mach Zehnder Optical Transmitter. In order to demonstrate the applicability of our CWDM device to data center applications, we utilized Pseudo Random Bit Sequence (PRBS) 2³¹-1 non-return-to-zero (NRZ) optical data into our device. Figure 3(a) shows a schematic of the first high speed testing setup. The PRBS signal is fed to a commercial test board with a 10Gbps SFP + transceiver (Txver). The transceiver’s optical input is not connected (N.C.) as it contains a limiting amplifier. The generated optical data is transmitted into a fiber polarizer (Pol), followed by an Erbium Doped Fiber Amplifier (EDFA) and a Bandpass Filter (BPF) before being coupled into the CWDM device under test (DUT) using a tapered fiber with quasi-TE configuration. The Port4 output (1551nm) was then amplified before a PIN-TIA Photoreceiver and its Bit Error Rate (BER) and eye diagram measured on a BER Tester (BERT) and a Digital Sampling Oscilloscope (DSO) respectively. Figure 3(b) shows the measured clear and open eye diagram. The BER is plotted at Fig. 3(c). An error free transmission of BER = 10⁻¹² was obtained at a power of −14dBm measured at the output of the CWDM device.

In order to evaluate the performance of the CWDM device with a higher bitrate, a 1550nm Continuous Wave (CW) laser was modulated by a PRBS 2³¹-1 NRZ 30Gb/s signal using a Mach Zehnder optical transmitter as shown in Fig. 4(a). The measured eye diagram and BER were characterized and plotted on Fig. 4(b) and 4(c) correspondingly. An error free (BER = 10⁻¹²) and a clear and open eye diagram were obtained with a received power of −12dBm. Hence, we observe that for incremental bitrates from 10Gb/s to 30Gb/s, the power penalty for a BER of 10⁻¹² is small at 2dB.

4. 8-channel multi-wavelength source

The 1 by 8 CWDM device is further implemented in a two-stage system for realizing an 8-channel multi-wavelength source. Our devices were fabricated on the USRN platform instead of crystalline silicon as the propagating core. USRN possess a large Kerr nonlinearity whilst maintaining negligible two-photon absorption losses. The multi-wavelength source consists of a 6mm long, highly nonlinear USRN waveguide as the first stage and the 1 by 8 CWDM source as the second stage, as shown in Fig. 5(a).

 

Fig. 5 (a) Schematic of Supercontinumm preceding CWDM b) Spectral plots of (blue) 500 fs laser source (blue)., (b) supercontinuum of 6mm waveguide (black) and (c) individual CWDM output with 6mm waveguide preceding the 1 by 8 CWDM.

Download Full Size | PDF

In the first stage, 500fs pulses at a repetition rate of 20MHz with a peak power of 90W are launched into a USRN waveguide to generate supercontinuum. The USRN waveguide used for supercontinuum generation is 600nm in width and 300nm in height, and possesses propagation losses of 3dB/cm. Including the 1dB loss of each CWDM filter and the preceding USRN waveguide loss, the loss of each channel for the multi-wavelength source is 2.8dB. The entire footprint of the integrated 8-channel multi-wavelength source is 0.08mm². The effective nonlinear parameter of this waveguide is 550W⁻¹/m and the group velocity dispersion at 1550nm is anomalous with a calculated value of ~0.2ps²/m [29]. The waveguide is then connected using a linear waveguide taper that adiabatically converts the mode from the 600nm waveguide width to the first filter with a central waveguide width of 650nm. The taper region has a length of 100μm, which is sufficiently long to minimize any losses from the adiabatic mode conversion. The generated supercontinuum then enters the 1 by 8 CWDM device where 8 channels of light spanning 1470nm-1610nm separated by 20nm are generated. The spectrum of the 500fs seed pulses is shown in Fig. 5(b). In the first stage, supercontinuum is generated, and the wavelength span at the −20dB level is 1450nm-1650nm, and exceeds the center wavelengths of the CWDM device. Figure 5(c) shows the generated supercontinuum after the seed pulses propagate through the first stage (USRN waveguide). This spectrum was measured through a 6mm long waveguide without the 1 by 8 CWDM device. The CWDM device is then used to spectrally carve out individual channels (Fig. 5(d)), and the spectrum of each channel is shown in Fig. 5(d). It is observed that all 8 generated wavelength channels are within the −20dB level of the channel that’s centered closest to the seed pulse. Aside from the 1550nm channel which overlaps in wavelength with the SCG seed pulse, all the other 7 channels have power levels which are within 10dB of one another. Therefore, the generated powers in each channel are somewhat close, though with further optimization in the SCG flatness, the multi-wavelength source could become even more applicable for CWDM-based systems.

5. Discussion

Because of the large linear refractive index in USRN, on-chip photonic lightwave circuits possess compactness that are almost equivalent to that their silicon counterparts. The CWDM device demonstrated here possesses low crosstalk (−25dB on average) and large channel bandwidth (7.7nm on average). The large channel bandwidth is particularly useful for withstanding large temperature fluctuations during device deployment and fabrication imperfections. This is also in congruence with ITU G.694.2 recommendation for allowing around 7nm of wavelength variation. Because the device does not possess a free-spectral range, it can be potentially scaled up to cover the entire O- to L-bands. High speed testing with data at 30Gb/s yields error free operation at a power of −12dBm, implying an aggregate 240Gb/s data transmission over 8 channels. If the CWDM channels are implemented over 18 channels over the full O- to L-bands, the potential aggregate data rate possible in the CWDM device would exceed 0.5Tb/s. Future work will investigate the ability of the CWDM device to achieve error-free operation at data rates exceeding 30Gb/s, such that the aggregate data rate may be augmented by both channel count and data rate.

In silicon photonics-based devices, the ability to combine linear and nonlinear functionalities for efficient all-optical signal processes is limited by the low nonlinear figure of merit of crystalline silicon [27]. Two-photon and free-carrier absorption at telecommunications wavelengths limit both speeds and efficiencies of nonlinear processes. The low deposition temperature of 250°C ensures that the USRN platform is compatible with backend CMOS processing, whereas the engineered bandgap of 2.1eV further ensures that the two-photon edge is far sufficiently far from 1550nm for nonlinear processes at 1550nm to proceed without detriment. Combining the nonlinear USRN waveguide which serves a nonlinear function for SCG with the CWDM device which is an intrinsically linear element creates a multi-wavelength source which may be used to power CWDM-based photonic systems. The supercontinuum generated also encompasses the entire 8 channel span at the −20dB level, ensuring that there is relatively high power within each of the generated channels.

6. Conclusion

We have designed and demonstrated an on-chip CWDM device using the USRN platform. Eight CWDM channels covering the specifications defined by the ITU-T G.694.2 standard have been demonstrated, spanning from 1470nm – 1610nm. The 1 by 8 CWDM device possesses low crosstalk and large channel bandwidth. We have also tested our device with 10Gb/s and 30Gb/s NRZ data and achieved error free operation and clear and open eye diagrams. Spectral slicing of supercontinuum generated in a highly nonlinear USRN waveguide also yielded 8 individual channels of light that may further be used as a multi-wavelength source in CWDM-based photonic systems.