Introduction
All-optical signal processing consists in manipulating the properties of light waves carrying information signals without resorting to electronic means following photodetection. This includes the implementation of functionalities such as signal temporal or spectral conditioning, wavelength conversion, regeneration, switching, modulation format conversion, clock recovery, etc. Generally, the purpose of all-optical signal processing is to avoid electronic bottlenecks by performing the required functionalities directly on optical high-speed modulated signals or by processing a number of (typically wavelength-multiplexed) channels in parallel, thereby overcoming the speed limitation of electronics, or providing cost savings.
Typically, all-optical signal processing functionalities are demonstrated using nonlinear interactions in a wide range of materials and devices, including the use of highly nonlinear optical fibers [ 1]), nonlinear waveguides (e.g., silicon [ 2] or chalcogenide [ 3]), semiconductor optical amplifiers (SOAs) [ 4], periodically-poled lithium niobate (PPLN) [ 5], etc. However, some functionalities can also be realized linearly by appropriate filtering of the signal.
In this paper, we show how silicon micro-ring resonators (MRRs) [ 6] can be used to implement a variety of signal processing functionalities, including multi-channel and ultra-high speed modulation format conversion, multi-channel demodulation of phase modulated signals, signal conditioning and waveform generation. Silicon MRRs present a number of advantages. Their periodic transfer functions can be used for the parallel processing of several wavelength channels, while transfer function engineering enables flexible tailoring of the signal properties. The devices are compact and can be manufactured with potentially low-cost and high yield using well-controlled micro-electronics fabrication processes. The compactness of the devices means that they have potential for applications in telecommunication, data-communication, as well as on-chip communication. For applications where the signal to be processed has been propagating over a length of optical fiber, the inherent polarization sensitivity of silicon waveguides appears first as a limitation. However, we also show in this work that devices with reduced polarization sensitivity can be designed and fabricated on the silicon platform, thereby allowing the demonstration of polarization- independent linear signal processing.
This paper is structured as follows. An introduction to silicon MRRs and their fabrication is provided in Section 2. In Section 3, the use of a single MRR to perform modulation format conversion from the return-to-zero (RZ) to the non return-to-zero (NRZ) modulation format is demonstrated for both on-off keying (OOK) and phase-shift keying (PSK) signals at 40 Gbit/s, as well as for ultra-high speed optical time division multiplexed (OTDM) signals, resulting in the first demonstration of NRZ signal generation at an unprecedented bit rate of 640 Gbit/s. Section 4 deals with the demodulation of differential PSK (DPSK) signals. In Section 5, it is shown how filtering of the adiabatically chirped signal generated from a directly modulated laser (DML) by an MRR can enable such a laser to be operated at a higher bit rate than what is expected from its nominal bandwidth. An add-drop MRR with adjustable coupling coefficients can be used to synthesize an ultra-wide band (UWB) signal, as demonstrated in Section 6. The implementation of linear signal processing with reduced polarization sensitivity thanks to the use of a polarization diversity circuit is described in Section 7. The results are finally summarized in Section 8.
Silicon MRRs
MRRs are very attractive devices for integration [ 7, 8] thanks to their property of supporting traveling wave resonance modes without any reflective facet. Thanks to this great advantage, MRR-based laser sources [ 9], electro-optical modulators [ 10], optical memory units [ 11– 13], optical bandpass filters [ 14], etc., have been widely investigated. The through and drop transfer functions of a standard add/drop MRR, as shown in Fig. 1, can be expressed as [ 15]
where ri and ki are the field transmission and coupling coefficients, respectively, of the coupling regions of the resonator, satisfying the relation ri2 + ki2 = 1 for lossless coupling with i = 1 and 2 denoting the through and drop coupling regions, respectively. q and a are the roundtrip phase shift and field transmission coefficients along the ring waveguide, respectively. If the drop coupling coefficient k2 equals to zero, the MRR is degraded to all-pass type. The free spectral range (FSR) DlFSR of a MRR can be expressed as [ 16]
where ng is the group index, and L is the roundtrip length. The Q value of an add/drop MRR, which is a measure of the sharpness of the resonance relative to its central frequency, can be further expressed as [ 16]
where l0 is the resonance wavelength, DlFWHM is the 3-dB resonance bandwidth of the drop transmission. In Eq. (4), weak coupling is assumed for both through and drop coupling regions. An important feature of MRRs is that their FSR can be easily controlled by designing the size of the micro-ring, and resonance bandwidth or equivalently Q value can be easily engineered by controlling the coupling coefficients of the coupling regions, allowing MRRs to be optimized for wavelength division multiplexing (WDM) linear all-optical signal processing applications.
Silicon-on-insulator (SOI) is considered to be one of the most promising platforms for integrated optical devices thanks to its complementary metal-oxide-semiconductor (CMOS) compatible fabrication process. Moreover, the ultra-high refractive index contrast between silicon (n≈3.5) and buried oxide (BOX) layers (n≈1.45) enables the realization of compact SOI devices. In the following work, the silicon devices have been fabricated by standard electron beam (e-beam) lithography-based nano-fabrication processes, as shown in Fig. 2. First, a thin layer of e-beam resist (ZEP520A) is spin coated on the SOI sample. The structures are then defined by e-beam writing (JEOL JBX-9300FS) and subsequent developing. After that, the sample is etched by inductively coupled plasma reactive ion etching (ICP-RIE) to transfer the patterns to the top silicon layer. The final device is obtained by stripping the residual e-beam resist after ICP etching.
Modulation format conversion
Optical modulation format conversion is a potentially important functionality in nodes interfacing optical networks operating with different modulation formats. Format conversion between RZ and NRZ formats is one essential conversion, since both formats are widely used in different parts of optical networks [ 17]. A good modulation format converter requires low power consumption and the ability to support simultaneous multiple channel operation to accommodate widely deployed WDM networks. The use of passive filters such as optical nano-fiber ring resonators [ 18] and fiber delay-interferometers (DIs) [ 17] has been demonstrated for single and multiple channels RZ-to-NRZ format conversion. However, both fiber DIs and nano-fiber ring resonators require stabilization, and integrated implementations would be preferred.
WDM RZ-OOK to NRZ-OOK format conversion
Thanks to the periodic property of the MRR transmission, a single silicon MRR is able to perform multi-channel RZ-OOK to NRZ-OOK format conversion based on optical spectrum transformation [ 19], which consists in transforming an RZ spectrum to an NRZ-like spectrum by optical filtering. The corresponding experimental setup is presented in Fig. 3. Four channels of continuous wave (CW) laser light with wavelengths of 1548.16, 1549.76, 1551.36, and 1552.97 nm (200-GHz channel spacing) are combined in a coupler, amplified by an erbium-doped fiber amplifier (EDFA), and simultaneously modulated at 50 Gbit/s in the 33% RZ-OOK format in two Mach-Zehnder modulators (MZMs). The four modulated RZ channels are amplified using a second EDFA and further de-correlated by 50 m dispersion-compensating fiber (DCF). A set of polarization controllers (PCs) followed by a polarizer (Pol.) and a second PC are used to align the input polarization of all WDM channels to the transverse electric (TE) mode of the waveguide. The WDM RZ signal is then converted to a WDM NRZ signal by the MRR. Finally, the amplified converted WDM NRZ signal is wavelength demultiplexed by an arrayed waveguide grating (AWG) with 3-dB bandwidth of 62 GHz, and detected in a pre-amplified receiver.
Fig.4 (a) WDM RZ signal spectrum and MRR through transmission; (b) converted WDM NRZ signal spectrum; measured eye-diagrams of (c) single RZ signal, (d) converted NRZ signals and (e) original reference NRZ signal; (f) BER measurements of the converted NRZ channels in single-channel and WDM operation, as well as an electrically generated reference NRZ signal |
Figure 4 shows the measurement results. The MRR has an FSR of 100 GHz and an optimized Q value of 7900 with an extinction ratio (ER) as high as 20 dB, as shown in Fig. 4(a). The first order harmonic components in the spectra of all four RZ channels are suppressed simultaneously by the MRR, resulting in an effective spectrum transformation from WDM RZ to WDM NRZ, as illustrated in Figs. 4(a) and 4(b). Accordingly, the RZ eye-diagram (see Fig. 4(c)) is effectively converted to a very clear NRZ format eye-diagram (see Fig. 4(d)), which is very close to the reference electrically modulated NRZ signal (see Fig. 4(e)). Figure 4(f) shows the result of bit error ratio (BER) measurements with a 27–1 pseudorandom binary sequence (PRBS). BER values below 10-9 are obtained for all the four converted NRZ channels with limited power penalty with respect to the reference electrically modulated NRZ signal. The moderate power penalty is introduced by inter-channel crosstalk. It should be mentioned that, for this particular application, the channel spacing should be a multiple of the FSR of the MRR, which is dictated by the spacing between the discrete tones in the spectrum of the RZ signal, hence its bit rate.
Simultaneous RZ-OOK to NRZ-OOK and RZ-DPSK to NRZ-DPSK format conversion
Apart from OOK modulation, the use of an MRR for format conversion is also compatible with DPSK modulation, allowing simultaneous RZ-OOK to NRZ-OOK and RZ-DPSK to NRZ-DPSK format conversion by a single MRR [ 20]. The experimental setup for simultaneous RZ-to-NRZ format conversion of OOK and DPSK is shown in Fig. 5. Two CW lights at 1549.35 and 1551.36 nm are modulated at 41.6 Gbit/s in the 33% RZ-DPSK and 33% RZ-OOK format, respectively, by two sets of two MZMs. The PRBS length is 231–1 for both channels. The OOK and DPSK signals are then combined in a 3-dB coupler and amplified by an EDFA. A set of PC-Pol.-PC is used afterwards to adjust the states of polarization of the signals to the transverse magnetic (TM) mode of the MRR. The converted NRZ-OOK and NRZ-DPSK signals after the MRR are filtered by an optical bandpass filter (OBPF) and finally detected in a pre-amplified receiver. A 1-bit fiber DI followed by balanced detection in a pair of 45-GHz photodiodes is used in the DPSK receiver, while the OOK signal is detected using a single 45-GHz photodiode.
The fabricated MRR has an FSR of 83 GHz, a Q value of 5200, and an ER of 25 dB, as shown in Fig. 6(a). The spectra transformation is also shown. Since the central wavelengths of the two input RZ signals are tuned to the centers between two notches of the MRR through transmission, specific spectral components of both RZ-OOK and RZ-DPSK signals are effectively suppressed after the MRR and their spectra are successfully transformed to the spectra of NRZ signals. Accordingly, the eye-diagrams of RZ-OOK (see Fig. 6(b)) and RZ-DPSK (see Fig. 6(c)) are effectively converted to NRZ-OOK (see Fig. 6(e)) and NRZ-DPSK (see Fig. 6(f)) eye-diagrams, simultaneously. Figure 7 shows the results of BER measurements for both single channel and simultaneous two-channel format conversion. It can be seen that BER values below 10-9 can be obtained for both OOK and DPSK channels after format conversion. In addition, there is nearly no cross talk between the two channels.
Fig.6 (a) MRR through transmission and spectra of the two-channel input RZ signals and the converted NRZ signals; measured eye-diagrams of (b) input RZ-OOK (single channel), (c) input RZ-DPSK (single channel), (d) demo-dulated signal of the input RZ-DPSK after balanced detection, (e) converted NRZ-OOK (two-channel), (f) converted NRZ-DPSK (two-channel) and (g) demodulated signal of the converted NRZ-DPSK after balanced detection |
OTDM format conversion
OTDM is an essential multiplexing technique potentially allowing symbol rates beyond the limits currently achievable by electronic multiplexing [ 21]. OTDM makes use of RZ modulated signals with small duty cycles allowing temporal interleaving of the channels. However, traditional OTDM signal demultiplexing using an optical gate is very sensitive to timing jitter because of the fairly narrow pulse widths involved. Furthermore, OTDM signals are extremely sensitive to dispersion due to their broad spectra. Being able to generate ultra-high speed signals modulated in the NRZ format would alleviate those issues, resulting in increased timing jitter tolerance in the demultiplexing process thanks to their flat top, as well as better resilience to dispersion and enhanced spectral efficiency [ 22]. Ultra-high speed NRZ signals can be generated using conventional OTDM techniques followed by all-optical modulation format conversion from RZ to NRZ, which can be performed in a silicon MRR [ 23].
Fig.9 (a) Spectra of the original OTDM (blue), wavelength converted RZ (purple), and format converted NRZ signals (green), as well as through transmission of the silicon MRR (dashed black); (b)–(d) optical sampling oscilloscope traces of the original incoherent 640-Gbit/s OTDM signal, 640-Gbit/s wavelength converted RZ signal, and the 640-Gbit/s format converted NRZ signal, respectively; (e) BER results with PRBS length of 27–1 for channel 1 demultiplexed from 640-Gbit/s NRZ with 1 and 4 m DCFs, and a demultiplexed tributary from the 640-Gbit/s wavelength converted RZ signal with 1 m DCF |
Figure 8 shows the experimental setup for 640-Gbit/s NRZ signal generation. A 640-Gbit/s RZ OTDM signal with 27–1 PRBS at 1560 nm is first wavelength converted using a Kerr switch with an amplified CW probe at 1535.6 nm in a 200-m highly-nonlinear fiber (HNLF), resulting in the generation of a 640-Gbit/s pulse-to-pulse coherent RZ signal. This phase coherence ensures that the spectrum of the RZ signal contains discrete lines at frequencies that are multiple of the symbol rate, in contrast with the original OTDM signal where this phase coherence is lost in the time-multiplexing process. The converted RZ signal is then amplified and input to a silicon MRR in the TM mode in order to perform format conversion. The MRR has an FSR of 1280 GHz and Q value of 638. An OBPF is then used to reduce the amplitude ripples of the converted NRZ signal. Finally, the 640-Gbit/s NRZ OTDM signal is time-demultiplexed in a nonlinear optical loop mirror (NOLM) [ 24] before being detected in a 10-Gbit/s receiver.
Figure 9(a) illustrates the spectrum transformation of the 640-Gbit/s NRZ signal generation. As anticipated, the original 640-Gbit/s OTDM spectrum (blue) does not contain strong harmonic components at multiples of frequencies corresponding to the bit rate due to the lack of pulse-to-pulse phase coherence between the OTDM tributary channels. However, the 640-Gbit/s wavelength converted RZ spectrum (purple) displays clear and strong harmonic components, which are aligned with the notches of the MRR through transmission (dashed black) in order to perform format conversion. After the OBPF, a 640-Gbit/s NRZ spectrum (green) is obtained, showing a much reduced bandwidth compared to the 640-Gbit/s wavelength converted RZ spectrum. To test the dispersion tolerance, short pieces of DCFs are added before demultiplexing the 640-Gbit/s NRZ and wavelength-converted RZ signals. BER measurement results are shown in Fig. 9(e). Good BER performance with power penalty as low as ~1.5 dB is obtained for the tributary demultiplexed from the 640-Gbit/s NRZ signal after 1-m DCF transmission. However, for the same DCF length, an error floor is present in the case of the 640-Gbit/s wavelength converted RZ signal. A similar error floor appears for the 640-Gbit/s NRZ signal after transmission over 4 m DCF. Hence, it is clear that the 640-Gbit/s NRZ signal exhibits improved dispersion tolerance compared to the wavelength converted RZ signal.
DPSK demodulation
DPSK is a promising modulation format for optical communication networks. Compared to OOK, DPSK exhibits a 3-dB improvement in receiver sensitivity when balanced detection is employed, and is more tolerant to fiber nonlinearities [ 25]. Mach–Zehnder delay interferometers (MZDIs) [ 26] with 1-bit delay are typically used to demodulate DPSK signals. The use of silicon MRRs as DPSK demodulators has been recently proposed [ 27] and demonstrated [ 28] for single channel operation. Thanks to the periodic property of MRRs transfer functions, this scheme is also promising for multi-channel applications [ 29]. The experimental setup is shown in Fig. 10. Four channels of CW laser light with wavelengths of 1548.65, 1550.26, 1551.90, and 1553.54 nm (channel spacing of ~200 GHz) are combined in a coupler, amplified by an EDFA, then simultaneously modulated with a PRBS length of 231–1 at 40 Gbit/s in the NRZ-DPSK format in a MZM. The WDM NRZ-DPSK channels are then amplified by another EDFA and de-correlated in 150-m DCF. A set of PC-Pol.-PC is then used to align the input polarization of all channels to the TM mode of the silicon waveguide. Demodulated AMI and DB signals will be obtained at the through and drop ports of the MRR, respectively. Afterwards, 1-km standard single mode fiber (SSMF) is used to compensate the dispersion introduced by the DCF. The WDM demodulated signal is amplified by another EDFA, de-multiplexed by an AWG and finally detected in a pre-amplified receiver.
Figure 11 shows the measured MRR through and drop transmissions, and the spectra of the WDM NRZ-DPSK signals, as well as of the WDM AMI and DB signals demodulated by the MRR. The ER of the through transmission of the MRR is as high as 25 dB, its measured FSR is 0.8 nm, corresponding to 100 GHz, and its Q value is 6700. The notches of the through transfer function and the peaks of the drop transfer function of the MRR are aligned to the carrier wavelengths of all the WDM channels, resulting in simultaneous demodulation of all channels to the AMI and DB formats, respectively. Figure 12 presents BER measurements for all channels at the output of the through (AMI) and drop (DB) ports. The inset pictures illustrate very clear eye-diagrams for the demodulated AMI and DB signals. For comparison purpose, the MRR is also substituted with a commercial fiber MZDI with 42.7-GHz FSR used for single channel demodulation. Although the BER for both through and drop port demodulations are worse than that of the MZDI, all the four channels reach bit error rates below 10-9, without noticeable error floor. In addition, the optimized Q value of the MRR leads to similar receiver sensitivities for both through and drop port demodulations.
Modulation speed enhancement
An MRR can be used in order to enhance the modulation speed of a DML, as illustrated in Fig. 13. The operation bears some analogy with the principle of chirped managed lasers (CMLs), which consist of a DML biased at a high current with respect to the threshold, followed by an optical spectrum reshaping (OSR) filter [ 30]. Under these conditions, the laser is operated with a wide bandwidth, however at the expense of a limited ER, and its frequency chirp is dominated by its adiabatic contribution. By using the MRR through transmission to suppress the frequency components corresponding to the modulated zeros, both ER and eye-opening of the signal emitted by a DML operated beyond its nominal bit rate can be greatly improved [ 31].
This concept has been experimentally demonstrated using a commercially available 10-Gbit/s DFB laser modulated at 40 Gbit/s and a discrete custom-made silicon MRR with FSR of 200 GHz and Q value of 3300. The DML was thermally tuned to the short wavelength side of an MRR resonance (detuning of ~0.17 nm), as shown in Fig. 14(a). The MRR resonance was then used to filter the longer wavelength side of the signal spectrum, which corresponds to the emitted zeros. As a result, the eye-diagram at the output of the DML is significantly opened thanks to the MRR and the zero level is largely suppressed, at the expense of some overshoot on the one level. Figure 14(b) shows the BER performance of the signal emitted at 40 Gbit/s by the DML and that of the signal filtered by the MRR, directly after the resonator as well as after different lengths of SSMF. The BER improvement due to the MRR filtering is evident. The sensitivity at 10-9 of the filtered signal was about -17.2 dBm whereas an error-floor of 1.2 × 10-6 was obtained for an average input power of -9 dBm for the DML signal without MRR filtering. The filtered signal could furthermore be transmitted over SSMF lengths compatible with short reach applications, with a power penalty of about 1.5 dB after 3 km and a signal still detected error-free after 4.5 km uncompensated SSMF. Such a scheme has also been shown to be directly applicable to vertical cavity surface-emitting lasers (VCSELs) [ 32], and is promising for the development of compact high-speed sources for short-reach systems.
Fig.14 (a) MRR through transmission and spectra of a 10-Gbit/s DML operated at 40 Gbit/s before and after the MRR. The insets show the eye diagrams measured directly at the DML output and after filtering by the MRR; (b) BER performance at the DML output (DML, B2B), after the MRR (DML+ MRR, B2B), as well as after different lengths of SSMF |
Monocycle pulse synthesis
The photonic generation and distribution of UWB signals has been a very active area of research for high data rate wireless applications [ 33]. Thanks to their compact sizes, MRRs have been investigated for UWB monocycle signal generation using differentiation of a signal whose phase has been modulated by a Gaussian electrical signal by phase-to-intensity conversion using the slope of the MRR transfer function [ 34]. In this context, an MRR resonance is traditionally offset related to the carrier wavelength of the phase modulated optical signal. As a consequence, the slope of the MRR transmission performs PM-to-IM conversion of the Gaussian phase modulated optical signal. However, such a scheme relies on the generation of Gaussian electrical pulses, which may not be available from conventional electronic circuits. In contrast, it has been found that UWB monocycle signals can also be synthesized from standard NRZ electronics using an NRZ-DPSK signal generated in a Mach-Zehnder modulator filtered by a single MRR whose resonance is tuned to the carrier wavelength of the optical signal, provided that the MRR through and drop coupling coefficients are properly set [ 35]. The principle of the method is depicted schematically in Fig. 15(a). A CW laser is modulated in the NRZ-DPSK format in a MZM. The modulation results in intensity dips each time the phase of the signal is flipped between 0 and p. The modulated signal is then input to a silicon MRR in add-drop configuration (therefore with two coupling regions). By adjusting the values of the coupling coefficients of the through and drop ports (k1 and k2, respectively), a balanced monocycle signal can be obtained. Figure 15(b) represents the combination of values of k12 and k22 for which A1 = A2 for three signal rise times of 50, 100 and 200 ps. It can be seen that, for each value of the rise time of the phase modulating signal, two sets of (k12, k22) parameters that equalize A1 and A2 can be found. Each set corresponds to a given polarity of the monocycle pulse, as illustrated in Fig. 15(b). Consequently, the optimization of the UWB waveform requires MRRs with tunable coupling coefficients.
To test this new concept, a silicon MRR with adjustable through and drop power coupling coefficients was designed and fabricated. The tunability of the coupling coefficients is achieved by two MZI structures which act as through and drop couplers, as shown in Fig. 16. Thermal tuning enables to control the relative phase shifts between the two arms of each MZI, and consequently the power coupling ratio between the ring and the input and drop waveguides. The device operates in the TM mode with an FSR of 100 GHz. The fabricated device was used in an UWB generation experiment at 625-Mbit/s. The experimental set-up is a direct implementation of the principle diagram shown in Fig. 15(a). The electrical signals driving a MZM in push-pull mode were generated at 625-Mbit/s by programming a 40-Gbit/s bit pattern generator, hence resulting in fast rise and fall times. Low pass filtering of the driving signals with a 4th order Bessel filter having a 7.5-GHz cut-off frequency was performed in order to increase their rise time to ~50 ps. In practice the proposed method will directly accommodate the rise times of lower speed signal generation electronics compatible with UWB applications. The CW laser was precisely tuned to a resonance of the MRR, as shown in the optical spectra of Figs. 17(a) and 17(c). The power coupling coefficients of the through and drop couplers of the MRR were thermally tuned, and balanced monocycle pulses were obtained in both polarities, as shown in Figs. 17(b) and 17(d).
Fig.17 Measured transfer functions at the through port of the MRR, together with the spectra of the optical NRZ-DPSK signals at 625 Mbit/s for generations of (a) negative and (c) positive polarity monocycle signals; waveform of the generated (b) negative and (d) positive polarity UWB monocycle pulse at 625 Mbit/s |
Polarization diversity circuit
In all previous demonstrations, the silicon MRRs were operating on one specific polarization state since standard SOI MRRs are extremely polarization sensitive because of the high index contrast silicon waveguides. However, in a real subsystem deployed over an optical fiber link, the polarization state of the optical signal at the MRR input may change randomly over time, making standard SOI devices no longer compatible with optical processing functionalities. To realize polarization independence, polarization diversity (Pol-D) circuits based on polarization splitter and rotator (PSR) technologies are typically used [ 36], as shown in Fig. 18. Assuming two orthogonal polarization states (TE and TM) are present at the input of the Pol-D circuit, the first PSR1 splits the two orthogonal polarization states into two beams of TE light. The two TE beams are then injected into the MRR, which is designed to have two identical coupling regions. With the same through transmission after the MRR, the two TE beams are combined back to two orthogonal polarization states by the second PSR2, therefore avoiding interference. As a result, the total device exhibits a polarization independent transmission.
Fig.19 (a) Microscope picture of a Pol-D circuit with a single MRR and two asymmetrical DC based PSRs. The inset shows an SEM image of the asymmetrical DC; (b) detailed transmission around the resonance wavelength of 1532.67 nm for TE, TM and 10 randomly chosen input polarization states |
Figure 19(a) shows a Pol-D circuit built on the SOI platform with a single MRR and two asymmetrical directional coupler (DC)-based PSRs [ 37]. The asymmetrical DC-based PSR relies on the phase match between the TE and TM modes of the narrow and wide waveguides of the DC [ 38], as shown in the inset of Fig. 19(a). Thanks to the simple DC structure, the PSR can be fabricated by a single lithography and etching step, greatly simplifying the fabrication process of traditional multiple layer structure based PSRs [ 39]. Figure 19(b) shows the detailed transmission around the resonance wavelength of 1532.67 nm. Very similar Q values of 6780 and 6578 are measured for TE and TM transmissions, respectively. A remarkably low PDL less than 1 dB is demonstrated. However, the residual polarization dependence still results in a power penalty of ~4 dB at a BER of 10-9 between 20-Gbit/s RZ-DPSK signals demodulated with such a circuit with and without (at optimum polarization) a polarization scrambler [ 37].
Even though the fabrication process is simple for the asymmetrical DC-based PSR, its performance is very sensitive to the dimension of the narrow waveguide [ 38]. It is possible to increase the fabrication tolerance by introducing a taper for the wide waveguide. However, the fabrication tolerance improvement is still limited to 14 nm [ 40]. We have proposed and demonstrated a novel PSR based on a tapered waveguide followed by a 2 × 2 multi-mode interference (MMI) coupler [ 41], as shown in the inset of Fig. 20(a). It is demonstrated that the fabrication tolerance is better than 50 nm with large feature size, allowing the device to be fabricated by other fabrication methods such as deep ultraviolet (DUV) lithography. The device can also be fabricated by a simple process with one single step of lithography and ICP etching. Figure 20(a) shows a photography of the Pol-D circuit based on the novel PSR [ 42]. The device exhibits similar transmissions with FSR of 805 GHz and Q value of 1400 regardless of the input polarization. Figure 20(c) shows the details of the transmission around the resonance wavelength of 1546.52 nm. A lowest insertion loss of 0.5 dB is obtained. A low PDL smaller than 1.6 dB, and a high ER of 38 dB with polarization-dependent extinction ratio (PDER) better than 3 dB are measured. The fabricated Pol-D circuit was then used for NRZ-DPSK demodulation at 40 Gbit/s [ 42]. Figure 21(a) shows BER measurements performed for the signals demodulated by the Pol-D MRR without and with polarization scrambling at its input. The corresponding eye-diagrams are shown in Figs. 21(b) and 21(c), respectively. One can see that clear and open eye-diagrams are obtained in both cases. A power penalty of 3 dB at a BER of 10-9 is found between the signals demodulated with and without (at optimum polarization) the polarization scrambler, which is induced by the residual PDL and PDER. In contrast, using a standard MRR design would result in a completely closed eye diagram when polarization scrambling of the DPSK signal is performed before the resonator, as shown in Fig. 21(d).
Fig.20 (a) SEM image of a fabricated Pol-D circuit. The inset shows an SEM image of the PSR based on a tapered waveguide followed by a 2 × 2 MMI; (b) measured transmission of the Pol-D MRR over a 60-nm wavelength range and (c) details of the transmission around the resonance wavelength of 1546.52 nm for 15 randomly chosen input polarization states |
Fig.21 (a) BER measurements for the AMI signal demodulated by the Pol-D MRR with and without the polarization scrambler; eye-diagrams of the demo-dulated AMI signals (b) without and (c) with polarization scrambler, as well as (d) the signal demodulated by a standard MRR with polarization scrambling |
Conclusion
We have demonstrated a wide range of applications of all-optical linear signal processing using silicon MRRs. We have reviewed our recent work dealing with modulation format conversion, DPSK demodulation, modulation speed enhancement of a DML, and UWB monocycle pulse generation using silicon MRRs as passive filters. We have also described our work on improving the polarization dependence of SOI MRR devices by using Pol-D circuits with novel PSRs. As cost-effective ultra-compact passive devices, silicon MRRs may find applications for linear signal processing in telecommunication, data-communications as well as optical interconnection systems.