Polarization Diversified 16λ Demultiplexer Based on Silicon Wire Delayed Interferometers and Arrayed Waveguide Gratings

We proposed novel 16λ demultiplexer based on cascaded connection of silicon-nanowire waveguide type delayed interferometric filters and arrayed waveguide gratings, and experimentally verified 100-GHz-spaced bandpass filtering operation with nearly identical insertion loss (<; 6 dB) and low crosstalk (<; -19.5 dB to <; -34.6 dB for a single channel, <; -12.7 dB for entire output channels). We also demonstrated polarization diversified operation with 32 Gbps non-return-to-zero (NRZ) signal transmissions in C-band range. We also verified the operation of 140-GHz-spaced bandpass filtering in O-band spectral range.

SECTION I. Introduction

EXPLOSIVE data traffic growth via cloud networks and 5G mobile communications stimulated to require more advanced optical technologies such as wavelength division multiplexing (WDM) or coherent technologies for long reach datacenter applications [1]–​[4] in order to further enhance the signal communication bandwidth. Usually, together with several optical functional elements (i.e., optical sources, modulators and detectors) in the silicon photonic integrated circuits [5], [6], silicon wire waveguide type multiplexers/ demultiplexers (MUX/DeMUX) are required to process the WDM optical signals [7], [8].

In our previous works, we reported silicon based compact and low loss MUXs/DeMUXs based on cascaded delayed Mach-Zehnder interferometers (DMZIs) [9]–​[12] and arrayed waveguide grating (AWG) [13], [14]. To make WDM transmission capability much better, we need to further increase the channel count (N_Ch) of MUX/DeMUX. Usually, it is easy to increase N_Ch in AWGs. However, insertion loss (IL) and channel uniformity tend to fundamentally degrade as N_Ch increases [15], [16]. Although increasing free-spectral range (FSR) of the AWG can be one of efficient ways of improving the IL and the uniformity, it may cause to make the device size markedly larger to preserve spectral properties such as filter bandwidth and spectral crosstalk etc [15].

To date, to relax the degradation of spectral crosstalk of the AWG, there have been some approaches to additionally connect the interleaver that filters the AWG spectra out into an odd group (λ₁, λ₃, λ₅ etc) and the even group (λ₂, λ₄, λ₆ etc) [17], [18]. As an alternative way to circumvent the aforementioned problem, we propose a novel MUX/DeMUX consisting of flatband DMZI-type interleavers and pair of AWGs. By adopting highly efficient 8λ-AWG and the flatband interleaver for discriminating the two 8λ groups spectrally separated by the FSR of the AWG, we can make the channel scalability of the MUX/DeMUX two times higher (16λ), without sacrificing IL and channel uniformity. In this work, we theoretically verify and experimentally demonstrate the device operability.

Additionally, as it has been well known that since a polarization state of a signal is not maintained when the signal is transmitted through single mode fibers (SMFs), the DeMUX located at a receiver side is normally needed to operate for arbitrarily polarized input signal [14], [19]. In order to cope with this situation, by utilizing polarization diversity scheme, we also verify 16λ demultiplexing together with 32  Gbps modulated non-return-to-zero (NRZ) signal transmission for orthogonally polarized inputs in C-band range. We also identify the device operability in O-band range for extending application areas.

SECTION II. Device Configuration and Analytic Study

Fig. 1 shows the proposed silicon-wire based 16λ DeMUX for operating with a single polarization input (i.e., TE-mode). The proposed DeMUX consists of an 8λ-AWG and a single DMZI-type flatband interleaver filter whose free-spectral range (FSR) is given to just twice of that of AWGs, thus enabling to filtering 16 kinds of wavelength signals out for a TE linearly polarized signal. Based on the cascaded connection of multistage delayed interferometers with different path lengths and optical splitting ratios [8], [9], the DMZI region was designed to have flat-topped spectral response with an extremely low excess loss. As it was already reported in previous works [14], [19], by using the counter-propagating type AWG, we can make the entire scheme much smaller and simpler. Since the AWG does not need to increase N_Ch, and the DMZI interleaver intrinsically has much less insertion loss [9] than AWG type interleaver [20], the proposed device configuration has great advantages in terms of low loss and design simplicity.

Fig. 1.

Proposed silicon wire based 16λ-WDM MUX/DeMUX consisting of DMZI-type flatband interleaver and counter-propagating type 8λ-AWG.

Based on coupled mode theory and transfer matrix method [8]–​[11], [21], we analytically calculated spectral characteristics of the proposed 16λ DeMUX. We assumed a silicon-on-insulator (SOI) wafer where 200-nm-thick Si layer and 2-μm-thick buried oxide (BOX) layer. For operating in C-band range, basic access waveguide width (W_WG) was commonly set to be 480-nm for satisfying single mode excitation condition. Meanwhile, we designed the waveguide width at delayed region (W_DL) both for the DMZI-type interleaver and AWGs to 2-μm-wide for reducing random phase errors (δϕ) during top-down-based waveguide fabrication process [12]. Such δϕ was took into account in analytic calculations of each component. As depicted in Fig. 1, we adopted double filtering scheme in DMZI-type interleaver to relax the crosstalk degradation by δϕ [9].

Table I shows the parameters used in analytic calculations. In the DMZI-type optical interleaver, the directional coupler (DC) based optical coupling ratio (κ_DC) was set for spectral flatness [8], [9]. We also took into account the wavelength sensitivity of κ_DC(λ) that was estimated from the experimental result [11]. In AWGs, we assumed the light in a star coupler spreads out with a Gaussian-function-shaped manner along the propagation distance and undergoes multiple optical interference via 32 waveguide arrays at the output star coupler.

TABLE I Parameters Used in Analytic Calculations

Fig. 2 shows the calculated spectra for the proposed DeMUX with TE mode input at (a) Ch09-Ch16 and (b) Ch01-Ch08 (see Fig. 1) when the two center wavelengths for the DMZI interleaver and the AWG are exactly matched with each other. Each output channel exhibits 100  GHz-spaced 8λ filter response, while the unnecessary filter spectra given by the FSR of 12.8 nm were suppressed by the DMZI interleaver. In Fig. 2, for clarity to interact between the two components, we assumed one-stage DMZI interleaver with κ_DC(λ) at each DC and neglected δϕ.

Fig. 2.

Calculated spectra for the proposed 16λ DeMUX with TE-mode input at (a) Ch9-Ch16 and (b) Ch01-Ch08 when the center wavelengths for DMZI interleaver and AWG are exactly matched with each other.

As can be seen in Fig. 2, the interleaver has a response widely flat enough to preserve AWG response at each spectral regime, which can relax a portion of deviation of the center wavelength for each component by fabrication imperfections. Needless to say, accurate adjustment of the center wavelengths for the two components is very important from the viewpoint of device operability. Thus, each component should be carefully designed by considering equivalent indices and group indices for the optical waveguides in the two components.

Fig. 3 shows the calculated spectra for the proposed DeMUX with TE mode input at (a) Ch09-Ch16 and (b) Ch01-Ch08 when the two center wavelengths for the DMZI interleaver and the AWG are deviated by +π/2 radian with each other. As clearly seen in the Figure, the phase mismatch of +π/2 radian between the DMZI interleaver and the AWG tends to completely break double filtering mechanism and does not work as a WDM filter any longer. This result indicates the two aspects to be required in terms of device design and waveguide fabrication; 1) care in a design level must be taken to equalize each center wavelength for the two components, and 2) phase controllability for the optical waveguides should be accurate for each center wavelength to be matched within ±0.05π radian.

Fig. 3.

Calculated spectra for the proposed 16λ DeMUX with TE-mode input at (a) Ch9-Ch16 and (b) Ch01-Ch08 when the center wavelengths for DMZI interleaver and AWG are deviated by +π/2 radian with each other.

On the other hand, as mentioned in a previous section, in order to operate as a DeMUX in a receiver side, we also adopted polarization diversity into the proposed scheme. Fig. 4 shows the proposed polarization diversified 16λ DeMUX. The device scheme is similar to the case shown in Fig. 2. For a polarization diversity, the proposed DeMUX needs polarization beam splitter/ polarization rotator (PBS/PR) based on partial rib-type waveguide and asymmetric DC region [14] at the entrance of the device, and the two identically designed two DMZI-type flatband interleavers and two counter-propagating type 8λ-AWGs with 32 output channels.

Fig. 4.

Proposed polarization diversified silicon wire based 16λ-WDM DeMUX consisting of a PBS/PR and two pairs of identically designed DMZI-type flatband interleavers and counter-propagating type 8λ-AWGs.

In this case, the two kinds of orthogonally polarized lights are split into by the PBS/PR region. Then each light with TE-mode is processed by the same device libraries based on the interleaver and the AWG. To accurately match the filtered wavelength peaks between the two orthogonally polarized inputs, we designed the two kinds of signals to be processed by the same counter-propagating type 8λ-AWG that assures nearly zero polarization sensitive wavelength shift [14]. It is noted that the device inherently includes a cross junction to construct compact polarization diversified configuration. The crossing shape was optimized for minimizing size, excess loss (<0.1 dB) and optical leakage (<−30 dB) [10].

SECTION III. Experimental Demonstrations

Based on the analytic investigations, the proposed polarization diversified 16λ DeMUX was fabricated by ArF-immersion lithography technology on a 300-mm SOI wafer with a 200-nm-thick Si layer and a 2-μm-thick BOX layer [22]. The parameters for the waveguide widths were set to be the same as those used in the calculations. Fig. 5 shows the top view of the fabricated polarization diversified silicon wire based 16λ DeMUX. Entire configuration is the same as described in Fig. 4. Chip size was measured to 1.7-mm-wide and 2.8-mm-long including the PBS/PR, the two DMZI interleavers, and the two 8λ-AWGs with a crossing.

Fig. 5.

Fabricated polarization diversified 16λ WDM DeMUX operating in C-band spectral range.

For the measurement of spectral response, broadband spontaneous emission was used, and the transmitted light through the device under test (DUT) by fiber butt-coupling system was measured by a spectrum analyzer. The polarization state of the input signal was adjusted by in-line type polarization controller through a polarization maintaining fiber (PMF). Fig. 6 shows the measured spectral characteristics. Solid and dotted lines indicate the output response for the TE and TM mode light input. It is noted that the transmittance shown in Fig. 6 indicates the relative transmission ratio of the fabricated device normalized by that of S-bend shaped single mode silicon wire waveguide for each polarization input. That is, the transmittance in Fig. 6 stands for the excessive loss caused by several kinds of filter architectures such as the PBS/PR, the DMZI interleavers and the AWGs. Since the propagation loss of silicon wire waveguide is <0.5 dB/cm by ArF-immersion lithography technology [22], the insertion loss is almost the same as the excess loss of the device.

Fig. 6.

Measured 16λ DeMUX characteristics in C-band spectral range.

Each peak transmittance was estimated to be –5 to –6 dB for all output channels. As reported in our previous works [9], [14], we expect that the entire excess loss is given by 0.5 dB of the PBS/PR, 0.5 dB of the DMZI interleaver and 4.5 dB of the AWG. It is important to note that the excess loss of AWG can be further reduced to 1.2 dB by design optimizations as our previous work [13].

For either of linearly polarized light, we were able to experimentally confirm clear demultiplexing response. Each AWG exhibited nearly 100-GHz (0.8 nm) spaced 8λ spectra at the C-band range. As seen in Fig. 6, periodic filter response given by the FSR of AWGs was suppressed by the interleaver by more than 25 dB, thus enabling to distinguish all 16 kinds of wavelength signals from the device irrespective of the polarization states of input signal. Compared with the simulation result, the center wavelength (1555 nm) of filter spectral group for λ₁-λ₈ (Ch01-Ch17) was deviated by +5 nm. This is due to fabrication imperfections, which needs to be adjusted by optimizing fabrication process or by controlling design parameters. It is noted that the WDM grid wavelengths for each polarization input was almost identical (<0.1 nm) due to the same optical paths by counter propagating AWG [14]. Also, the two separated AWGs showed similar 8λ spectral responses spaced by the FSR of AWG, which is based on the accuracy of 300-mm ArF-immersion lithography [22]. Polarization dependent loss was estimated to be less than 0.5 dB for all output channels.

In order to experimentally estimate spectral crosstalk of the device, we used a tunable laser as an optical source. The butt coupled input and output lensed fibers were controlled to optically couple into the input channel and the output channel Ch01 for λ₁ (see Fig. 4). Then, we measured transmission spectra by changing the input laser oscillation wavelength from λ₁ to λ₁₆.

Fig. 7 shows the measured superimposed transmission spectra when we discretely controlled the wavelengths of the tunable laser. As seen in Fig. 7, each transmittance was normalized by that of the signal with λ₁. In case of an ideal state, the input signals having wavelengths with λ₂ to λ₁₆ do not transmit through the Ch01, which means that the transmittance difference between λ₁ and other wavelengths corresponds to the spectral crosstalk for each wavelength component.

Fig. 7.

Measured superimposed transmission spectra for the input channel and the output channel Ch01 where the tunable laser was discretely controlled from λ₁ to λ₁₆.

Table II shows the summary of the estimated crosstalk for each output channel. The crosstalk was less than <−20 dB for almost output channels. It is noted that the crosstalk for the output channels of λ₉-λ₁₆ was far less than that of λ₂-λ₈ simply because DMZI interleaver additionally suppress the crosstalk. In other words, the proposed scheme can contribute to reduce the amplitudes of total crosstalk for all channels. As can be seen in Table II, subtotal crosstalk for λ₂-λ₈ was estimated to <−13.25 dB, while that for λ₉-λ₁₆ was <−22.04 dB. In this way, the total crosstalk at the output channel Ch01 can be estimated to <−12.71 dB, which means that the crosstalk is mainly caused in a single AWG (λ₂-λ₈), and the increase in crosstalk for other channels (λ₉-λ₁₆) was around 0.5 dB.

TABLE II Estimated Spectral Crosstalk for Each Output Channel

We also evaluated spectral crosstalk for other output channels by simply summing up the measured spectral responses except for the main output channel. Fig. 8 shows the estimated crosstalk for all output channels. As for the Ch01 (λ₁), the lowest peak value in Fig. 8 (–13.7 dB) corresponds to the total crosstalk value by summing up all of crosstalk values shown in Table II. We confirmed that the estimated value in Fig. 8 was comparable with that obtained in Fig. 7. As can be seen in Fig. 8, the total crosstalk was calibrated to be –12.2 dB to –16.8 dB for all output channels.

Fig. 8.

Evaluated spectral crosstalk by summing up the spectral responses for all neighboring channels.

Meanwhile, if we consider point-to-point optical links in conventional optical interconnects, the system power penalty in WDM links is mainly given by incoherent (out-of-band) crosstalk rather than coherent crosstalk (in-band crosstalk) [23]. Considering the channel spacing (Δν) of 100 GHz, since the crosstalk belongs to the channels lying outside the spectral band occupied by the channel detected [23], the power penalty by the total crosstalk of <−12.71 dB could be less than 0.5 dB for the bit error rate (BER) of 10⁻¹². That is, unless the device is employed in routing of the WDM signal through multiple nodes where the system power penalty is govern by coherent beat noises, the proposed DeMUX could be applicable at the expense of <0.5 dB of system penalty. Moreover, further reduction of spectral crosstalk of the AWG could be possible by design optimization and fabrication process, that could lead to reduce the system penalty.

We were also able to reconfirm the device operability by transmitting 32  Gbps NRZ modulated signals into the fabricated DeMUX. Fig. 9 shows the measured eye diagrams for 32  Gbps modulated signals (PRBS 2³¹–1) with TE and TM modes inputs at each output channel. Tunable single mode laser was used as an optical source. The NRZ modulated signal was generate by commercially available lithium niobate Mach-Zehnder modulator. The polarization state of the signal was controlled by the in-line type polarization controller and was incident on the DUT through a PMF. The output signal was calibrated by an oscilloscope.

Fig. 9.

Eye diagrams for 32  Gbps NRZ modulated WDM signals (PRBS 2³¹–1) of the fabricated 16λ DeMUX for TE and TM mode input signals, together with WDM channel grids and output channel numbers.

As shown in Fig. 9, we were able to reconfirm the device operability by transmitting 32  Gbps NRZ modulated signals into the fabricated DeMUX. Irrespective of the input polarization states, the device exhibited clear eye opening at the proper output channels. It should be noted that the modulation bandwidth was just limited by our experimental setups. In the proposed filter scheme, each 3- dB passband width for all output channels shown in Fig. 6 were measured to be >0.42 nm, which means that the proposed scheme would be able to process the signal modulation bandwidth of >50 GHz. In other words, the currently designed filter scheme has an ability to process the NRZ modulation rate of >50 Gbps. Actually, in our recent work, we were able to observe clear opening of eye diagram for the 50 Gbps NRZ modulated signal through the similar transmission experiments [24].

On the other hand, as the setting of the operating wavelength deviates from the optimum grid, the insertion loss and crosstalk for all channels of the proposed filter scheme increase, which makes the system penalty much worse depending on the degree of additional insertional loss and crosstalk. In order to mitigate the deterioration of the system penalty, we can make the AWG filter spectral response much flatter rather than Gaussian shaped, which prevents the system penalty from degrading by the deviation of the operating wavelength from the center of channel grid. We believe that flat-topped spectral response of the AWG can be achieved by modifying the device design at around the star couplers of the input and output sides based on the previously reported works [25]–​[27]. Further experimental clarification of the optical link experiment is our work to be performed in the near future.

SECTION IV. Discussion

A. Spectral Uniformity

Since the proposed device is all-passive type scheme and its characteristics are determined by the phase relation of spatially separated lightwaves at multiple optical paths within the DMZI interleavers and the AWGs, filter spectral response in terms of center wavelength and spectral crosstalk is largely influenced by the fabrication technology. In order to evaluate spectral uniformity that is important figure of merit related to manufacturing tolerance, we characterized multiple 16λ DeMUXs located in a 300-mm SOI wafer. Fig. 10 shows the shop-map on the 300-mm SOI wafer. Each shot size was 26-mm-wide and 33-mm-high, thus 64 shots were lithographically drawn on a single SOI wafer. Hereafter, the position of the 64th inter-dies is defined as i.e., A = [4, 3], B = [3, 6], C = [6, 5]). The spectra shown in the Section III was obtained from the inter-die of A. To evaluate the spectral uniformity, we also characterized similar spectral response at other inter-dies of B and C in which their locations were separated by a few centimeters on the SOI wafer.

Fig. 10.

Shot-map on the 300-mm SOI wafer. Three inter-dies indicated by A, B and C were characterized.

Fig. 11 shows the measured spectra with a TE-mode input at the inter-dies of A [4], [3], B [3], [6], and C [6], [5]. As can be seen in Fig. 11, we observed similar spectral response irrespective of die position on the SOI wafer. Compared with the measured spectra for the three kinds of inter-dies at A, B and C, we confirmed good spectral uniformity. In Fig. 11(b), the spectra with a lower power level are seen to be unclear. This is not due to relatively higher excess loss of the fabricated device, but due to relatively worse sensitivity of the spectrum analyzer. The excess loss of the 16λ DeMUX were maintained to be 5–6 dB. As for the set value of Δν = 100 GHz, each channel grid was estimated to be ±<5 GHz. As for the center wavelength distribution, as roughly indicated by the dotted lines in Fig. 11, the deviation of center wavelengths was within ±<20 GHz.

Fig. 11.

Measured spectral response for the fabricated 16λ DeMUXs with TE-mode input at the inter-dies (a) A [4], [3], (b) B [3], [6] and (c) C [6], [5]. The dotted lines indicate the peak wavelengths at A.

Fig. 12 shows the measured spectral response of the fabricated device with a TE-mode input at (a) Ch17-Ch24 and (b) Ch01-Ch08 when an additional phase shifter of +π/2 radian was intentionally added in the DMZI-type interleaver. As theoretically discussed in Fig. 3, it was experimentally verified that the device is incapable of discriminating each wavelength as the interleaver is deviated from the center portion of the AWG spectral range. These results shown in Figs. 11 and 12 represent that the device design of the interleavers and the AWGs are efficiently reproduced in the wafer level during a fabrication process.

Fig. 12.

Measured spectral response for the fabricated 16λ DeMUXs with TE-mode input at (a) Ch17-Ch24 and (b) Ch01-Ch08 when the DMZI-type interleaver was designed to have an additional phase shifter of +π/2 radian.

We have not characterized the spectral distributions across the whole SOI wafer. Since the center wavelengths are given by random phase errors of fabrication process as well as the silicon core thickness fluctuation of the SOI wafer, the distribution range would be expected to a little more increase of σ = 120 GHz where σ is a standard deviation value [22]. However, since the phase controllability for achieving spectral response can be given by ArF-immersion lithography process, we could maintain spectral quality across the whole SOI wafer. As a matter of course, thermal heater controls for compensating for the inherent phase errors occurring in fabrication process will be an alternative efficient way to further improve the spectral uniformity and production yield [28], [29].

B. Spectral Discontinuity

Compared with a single DMZI-type filter or an AWG-type filter, the proposed scheme provides a design simplicity and good channel scalability. However, discontinuous wavelength grid separated by the FSR/2 for the DMZI interleaver might not be an ideal, depending on the WDM applications. It should be noted that the spectral gap for the DMZI interleaver could be further reduced by design optimization. In order to efficiently minimize the aforementioned spectral gap, along with further reduction of the FSR of interleavers and AWGs, the interleaver filter design should be upgraded to show 1) much steeper spectral roll-off at both spectral edges and 2) wider flat spectral range preserving low excess loss and low crosstalk.

Two kinds of the aforementioned requirements could be achieved by cascaded connecting additional DMZIs [30] or by using another filter architecture based on microring assisted DMZI scheme where spectral flatness and crosstalk are markedly improved by adjusting phase variation against a wavelength in all-passive type microring resonator [10].

C. Operability in O-band Spectral Range

Functionality of the proposed scheme is not limited in C-band spectral regime. In this work, we were also able to verify the 16λ demultiplexing operation in O-band spectral regime. Based on the identical SOI wafer structure to the case of the C-band option, only the waveguide widths such as W_WG and W_DL described in Section III were modified to operate in O-band spectral range, considering the single mode condition and dispersion relations between the DMZI interleaver and the AWG. FSR_DMZI and FSR_AWG were set to 14.6 nm and 7.3 nm, while Δν_DMZI and Δν_AWG were 7.3 nm and 0.8 nm (∼140  GHz). In this case, the device design was made to the TE-mode input only. Fig. 13 shows the top view of the fabricated silicon wire based 16λ DeMUX operating in O-band spectral regime. The chip size was measured to 0.9-μm-wide and 2.0-μm-long.

Fig. 13.

Fabricated 16λ DeMUX operating in O-band spectral range.

Fig. 14 shows the measured spectral response of the fabricated device with a TE-mode input. As seen in Fig. 14, we experimentally confirmed the similar DeMUX functionality at O-band spectral range. Measured channel spacings were nearly the same as the designed Δν_DMZI and Δν_AWG. Depending on the output channel number, the crosstalk and the excess loss were estimated to be worse than that shown in Fig. 6. This is most likely due to the following reasons: 1) the mismatch of center wavelengths between the two kinds of filters, and 2) somewhat inferior spectral response of the DMZI-type interleaver resulting from insufficient parameter optimization related to optical splitting ratios at multiple delayed interferometers. These drawbacks could be overcome by optimizing the design parameters. Robust DMZI filter design where optical coupling ratio keeps nearly constant within >100-nm-wide spectral range in O-band regime would be a promising way to suppress the crosstalk shown in Fig. 14 [31].

Fig. 14.

Measured 16λ DeMUX characteristics in O-band spectral range.

D. Performance Comparison

Finally, Table III depicts a brief performance comparison of several kinds of silicon wire based WDM filters. Compared with single filtering type devices reported in [16], [19], the proposed device exhibited lower loss with a larger channel count and a lower crosstalk. Also, in case of comparing with the combination type devices [17], [18] where the additional DMZI filter interleaves the AWG spectra into odd and even wavelength groups, only the proposed scheme exhibited polarization diversified operation. If the comparison is made with silicon based device, the proposed one showed superior property on the excess loss in spite of narrower channel spacing. In addition, the proposed scheme has a potential to further increase the channel count. If we adopt 1 × N (N: interger) type flatband DMZI interleaver [8] with 1 × M (M: interger) AWG, the total channel count could be scaled to N × M.

TABLE III Performance Comparison for Multichannel WDM Filters

SECTION V. Conclusion

We proposed a novel polarization diversified silicon wire 16λ DeMUX scheme consisting of PBS/PR, DMZI-type interleaver and 8λ-AWG and experimentally verified optical demutiplexing operation in C-band spectral range both for orthogonally polarized input signals with a low polarization dependent loss of <0.5 dB.

By using the fabrication process based on ArF-immersion lithography technology, 16λ spectral response was reproduced in multiple inter-dies on the 300-mm SOI wafer together with nearly identical insertion loss (<6 dB), spectral crosstalk (<−19.5 dB to <−34.6 dB for a single channel, <−12.7 dB for entire output channels). Polarization diversified filtering operation was experimentally reconfirmed with 32 Gbps NRZ modulated signal transmissions. We also identified the operability of the proposed filter scheme in O-band spectral regime. Further investigations to monolithically integrate active components such as modulators and photodiodes are under way now.

ACKNOWLEDGMENT

This paper is based on results obtained from a project commissioned by the New Energy and Industrial Technology Development Organization (NEDO).