Optical polarizers based on graphene oxide integrated with waveguides and ring resonators

Integrated waveguide polarizers and polarization-selective micro-ring resonators (MRRs) incorporated with graphene oxide (GO) lms are experimentally demonstrated. CMOS-compatible doped silica waveguides and MRRs with both uniformly coated and patterned GO lms are fabricated based on a large-area, transfer-free, layer-by-layer GO coating method that yields precise control of the lm thickness. Photolithography and lift-off processes are used to achieve photolithographic patterning of GO lms with precise control of the placement and coating length. Detailed measurements are performed to characterize the performance of the devices versus GO lm thickness and coating length as a function of polarization, wavelength and power. A high polarization dependent loss of ~ 53.8 dB is achieved for the waveguide coated with 2-mm-long patterned GO lms. It is found that intrinsic lm material loss anisotropy dominates the performance for less than 20 layers whereas polarization dependent mode overlap dominates for thicker layers. For the MRRs, the GO coating length is reduced to 50 µm, yielding a ~ 8.3-dB polarization extinction ratio between TE and TM resonances. These results offer interesting physical insights and trends of the layered GO lms and demonstrate the effectiveness of introducing GO lms into photonic integrated devices to realize high-performance polarization selective components.


Introduction
Polarization control is one of the fundamental requirements in many optical technologies [1][2][3].
To implement polarization-selective devices, a number of schemes have been proposed and demonstrated, including those based on refractive prisms [14,15], birefringent crystals [16,17], ber components [18][19][20], and integrated waveguides [21][22][23][24][25]. Among them, integrated polarization-selective devices based on complementary metal-oxide-semiconductor (CMOS) compatible integrated platforms [26] offer advantages of compact footprint, high stability, mass producibility, and high scalability as functional building blocks for photonic integrated circuits (PICs) [2]. Optical waveguides with metal cladding have been widely used to implement waveguide polarizers [27,28]. Although high polarizationdependent loss (PDL) has been achieved for these polarizers, it usually comes at the expense of high overall propagation loss and requires complicated buffer layers to achieve broadband operation.
However, none of these demonstrations were based on CMOS compatible platforms. Generally, the integration of 2D materials on CMOS compatible platforms requires layer transfer processes [33,36], where exfoliated or chemical vapour deposition grown 2D membranes are attached onto dielectric substrates (e.g., silicon and silica wafers). Despite its widespread implementation, the transfer approach itself is complex, which makes it di cult to achieve precise patterning, as well as exible placement and large-area continuous coating on integrated devices. Accurate control of the layer position, thickness and size is critical for optimizing parameters such as mode overlap and loss for performance. Current methods signi cantly limit the scale of fabrication for integrated devices incorporating 2D materials.
Owing to its ease of preparation as well as the tunability of its material properties, graphene oxide (GO) has become a highly promising member of the 2D family [37,38]. Recently [39], a broadband GO-polymer waveguide polarizer with a high PDL of ~ 40 dB was demonstrated, where the GO lms were introduced onto an SU8 polymer waveguide using drop-casting methods. The GO lm thickness for each dropcasting step was ~ 0.5 µm and the drop coating diameter was ~ 1.3 mm, neither being ideal for achieving precise control of the placement, thickness, and length of the GO lms.
Recently [40,41], we reported large-area, transfer-free, and high-quality GO lm coating on integrated waveguides using a solution-based method with layer-by-layer deposition of GO lms. Here, we use these techniques to demonstrate GO-coated integrated waveguide polarizers and polarization-selective microring resonators (MRRs) on a CMOS compatible doped silica platform. We achieve highly precise control of the placement, thickness, and length of the GO lms coated on integrated photonic devices by using our layer-by-layer GO coating method followed by photolithography and lift-off processes. The latter overcomes the layer transfer fabrication limitations of 2D materials and represent a signi cant advance towards manufacturing integrated photonic devices incorporated with 2D materials. We measure the performance of the waveguide polarizer for different GO lm thicknesses and lengths versus polarization, wavelength, and power, achieving a very high PDL of ~ 53.8 dB. For GO-coated integrated MRRs, we achieve an 8.3-dB polarization extinction ratio between the TE and TM resonances, with the extracted propagation loss showing good agreement with the waveguide results. Furthermore, we present layer-bylayer characterization of the linear optical properties of 2D layered GO lms, including detailed measurements that conclusively determine the material loss anisotropy of the GO lms as well as the relative contribution of lm loss anisotropy versus polarization-dependent mode overlap, to the device performance. These results offer interesting physical insights and trends of the layered GO lms from monolayer to quasi bulk like behavior and con rm the high-performance of integrated polarization selective devices incorporated with GO lms.
2. Go-coated Waveguide Polarizer 2.1 Device Fabrication Figure 1(a) shows a schematic of a uniformly GO-coated waveguide polarizer. The waveguides were fabricated from high-index doped silica glass core surrounded by silica via CMOS compatible processes [26,42] with chemical mechanical polishing (CMP) used as the last step to remove the upper cladding, so as to enable GO lm coating on the top surface of the waveguide. GO coating was achieved with a solution-based method that yielded layer-by-layer GO lm deposition, as reported previously [40,41]. Four steps for in-situ assembly of monolayer GO lms were repeated to construct multilayer GO lms on the desired substrate, with the process being highly scalable.
We uniformly coated waveguides with 1 to 10 layers of GO. Figure 1(b) shows the measured Raman spectra of the waveguides without GO and with 2 layers of GO, con rming the integration of GO onto the top surface by the presence of the D (1345 cm -1 ) and G (1590 cm -1 ) peaks of GO. The microscope images of the integrated waveguide with zero and 2 layers of GO are shown in the insets, which illustrate the good morphology of the GO lms. Figure 1(c) shows the thickness of GO lms versus the layer number characterized by atomic force microscopy. The insets show images of 1 and 10 layers of GO coated on a 2.2 cm × 2.2 cm silica substrate with high uniformity. The dependence of GO lm thickness versus layer number shows a nearly linear relationship, with a thickness of ~ 2.18 nm on average for each layer.
In addition to the uniformly coated devices, we selectively patterned areas of GO lms using lithography and lift-off processes. Apart from allowing precise control of the size and placement of the GO lms, this enabled us to test the performance of the GO-coated waveguide polarizers with shorter GO coating lengths but higher lm thicknesses (up to 100 layers). The chip was rst spin-coated with photoresist and then patterned using photolithography to open a window on the waveguide. Next, GO lms were coated on the chip using the method mentioned above and patterned via a lift-off process.
As compared with the drop-casting method that has a GO lm thickness of ~ 0.5 µm and a minimum size of about 1.3 mm for each step [39], the combination of our GO coating method with photolithography and lift-off allows precise control of the lm placement, size, and thickness (with an ultrahigh resolution of ~ 2.18 nm), all of which are critical for optimizing the device performance including the polarization gure of merit (FOM) and four-wave mixing conversion e ciency. Further, our solution based GO coating approach, unlike for example, the sophisticated transfer processes employed for coating 2D materials such as graphene [29,43], is capable of covering large areas (e.g., a 4 inch wafer) on dielectric substrates (e.g., silicon and silica wafers) with relatively few defects. The combination of patterning and deposition control of GO lms along with large area coating capability is critical for large-scale integrated devices incorporated with GO.

Polarization Loss Measurements
We used an 8-channel single-mode ber (SMF) array to butt couple both TE and TM polarized continuous-wave (CW) light from a tunable laser near 1550 nm into the waveguides. The mode coupling loss between the SMF array and the waveguides was ~ 8 dB/facet, which can readily be reduced to ~ 1.0 dB/facet with mode convertors [26]. The propagation loss of the uncoated 1.5-cm-long waveguides was very low (< 0.25 dB/cm) and so the total insertion loss (TE = − 16.2 dB; TM = − 16.5 dB) of the uncoated devices was dominated by mode coupling loss. The slight PDL of the uncoated waveguides resulted mainly from a slightly different mode-coupling mismatch but possibly also polarization dependent scattering loss from the roughness of the polished top surface.
To characterize the performance of the devices, we introduce two gures of merit (FOMs) -one for the excess propagation loss (FOM EPL ) and one for the overall polarization dependent loss (FOM PDL ): (2) where the excess propagation losses, EPL TE (dB/cm) and EPL TM (dB/cm), are GO-induced excess propagation losses for the TE and TM polarizations, respectively. PDL (dB) is the polarization dependent loss de ned as the ratio of the maximum to minimum insertion losses. The excess insertion loss, EIL (dB), is the insertion loss induced by the GO lm over the uncoated waveguide. The EIL only considers the insertion loss induced by GO, while excluding from the overall insertion loss both the mode coupling loss between the SMF array and the waveguide as well as the propagation loss of the uncoated waveguide. In our case, since the TM polarization had the lowest insertion loss, EIL is the excess GO-induced insertion loss for the TM polarization and is given by EIL = EPL TM • L, where L is the GO coating length. Note that FOM EPL only considers the propagation loss difference induced by the GO lms, and so is more accurate for the characterization of their material anisotropy, whereas FOM PDL is more widely used for evaluating the device performance [33] since it also includes the background (uncoated) PDL. FOM EPL equals FOM PDL only when the TE and TM polarized insertion losses of the uncoated waveguide are the same. The TE insertion loss increases much more strongly than TM with layer number, thus yielding a large PDL with low EIL and forming the basis for our high-performance polarization dependent devices. Since GO is a dielectric lm, our TM-pass GO-coated waveguide polarizer is quite different from TE-pass metal-clad waveguide polarizers based on a deeper power penetration of the evanescent TM eld into a lossy metal cladding [27,28]. The PDL reached a maximum of ~ 37.4 dB for a 10 layer uniformly coated device and 53.8 dB for a 100 layer patterned device, with a modest maximum EIL of ~ 5.0 dB and ~ 7.5 dB for the two devices, respectively. By optimizing the waveguide geometry to achieve a better mode overlap with the GO lms [33], the EIL can be further reduced. Moreover, the PDL was still increasing at a rate of 2-3 dB/cm/layer at 100 layers, and so substantially higher PDL can be obtained using layers thicker than 200 nm. Both FOM EPL and FOM PDL increase with a maximum of FOM EPL and FOM PDL of ~ 8.2 and ~ 8.1, respectively, at about 50 layers with the difference between them subsequently decreasing. This is because the impact of the background PDL (~ 0.3 dB) becomes smaller as the EIL increases for increased GO layer numbers.  Fig. 2(a) for both polarizations of the two devices, along with the TE propagation loss calculated (by means of the Lumerical FDTD commercial mode solving software) using ellipsometry measurements (at 1550 nm) for the refractive index n and extinction coe cient k of two samples having 2 layers (Fig. 3(a-i)) and 20 layers (Fig. 3(a-ii)) of GO. Since the out-of-plane (TM polarized) response of the layered GO lms is much weaker [44,45], we could only measure, via ellipsometry, the in-plane (TE polarized) n and k of the GO lms (uncertainty < 3%), which were used (in conjunction with the mode solving software) to calculate the waveguide loss for the TE polarization. The simulations assume constant n, k for different GO layer numbers in each plot. In Fig. 3(a), the experimental TE propagation loss agrees extremely well with simulations for 2 and 20 layers of GO. For other thicknesses the experimental TE loss increased more rapidly with GO layer number, indicating that the intrinsic GO lm loss increases with thickness. This is not surprising, and could be due to any number of effects such as increased scattering loss and absorption induced by imperfect contact between the multiple GO layers as well as interactions between the GO layers.    Fig. 3(a) for both polarizations, to the propagation loss calculated assuming isotropic lm properties (using the extracted k TE in Fig. 3(c) for both TE and TM polarizations). For low GO layer numbers (< 10) the PDL is dominated by the material anisotropy η MA at 75%, despite the overall TE loss only being 1 dB/cm/layer (at 1 − 2 layers). The contribution of the GO material anisotropy steadily decreases, becoming comparable to the mode overlap contribution, η MO , at ~ 20 layers, beyond which, for very thick lms (100 layers), is smaller than η MO , which is about 65%. This could re ect the transition of the lm properties slightly towards a bulk (isotropic) material for very thick lms. However, it is also interesting to note that even for very thick lms the intrinsic lm loss anisotropy is still large enough to form the basis for polarization dependent devices. To recon gure the polarization selection, it would be relatively easy to change the mode overlap, while changing the material loss anisotropy is more challenging. A possible method to implement a TE-pass GO waveguide polarizer would be to conformally coat a high-aspect-ratio waveguide with GO lms on the sidewalls. Finally, we note that although our GO-coated waveguide polarizers were based on a CMOS compatible doped silica platform, these GO lms can readily be introduced into other integrated platforms (e.g., silicon and silicon nitride) [47][48][49], offering polarization selective devices with reduced footprint.

Optical Bandwidth and Power Dependence Figures 4(a-i) and (a-ii) illustrate the PDL of both uniformly coated and patterned waveguides versus
wavelength from 1500 to 1600 nm, showing a variation less than 2 dB and con rming the broadband operation of the polarizers. Table I compares the performance of a range of silicon photonic polarizers and 2D material based optical polarizers. The bandwidth of the material anisotropy of GO thin lms is very broad [37] − several hundred nanometers, even extending to visible wavelengths ( Fig. 4(a-i) insets). This is a distinct advantage of GO-coated waveguide polarizers that is extremely challenging to achieve with silicon photonic polarizers [3,21,24]. Moreover, GO based polarizers have simpler designs with higher fabrication tolerance as compared with silicon photonic polarizers. Indeed, the latter require precise design and control of the dimensions [2,3]. It is also interesting to note that the PDL slightly increases at longer wavelengths for both the uniformly coated and patterned devices. This is probably a result of the excitation of high-order modes in the doped silica waveguides at shorter wavelengths, which reduces the strength of the interaction between the GO lms and the evanescent eld leaking from the waveguides, thus leading to a degradation in PDL. Figs. 4(b-i) and (b-ii), for both samples and polarizations, indicating only a slight increase in loss for the thicker layers and only for the TE polarization, possibly due to photo-thermal reduction of the GO lm at higher powers [50]. This might also result from self-heating and thermal dissipation in the multilayer GO lm, being the subject of on-going research. In contrast, since the TM polarized absorption was very low, it did not show any signi cant variation with power. The increase in loss for TE polarized light was reversible − indicating that the optically induced changes were reversible. Note that this slight reversible increase in loss for the TE polarization with power actually enhances the device performance. Finally, we have shown previously [38,51] that the material properties of GO can also be permanently changed by laser-induced photo-reduction but at signi cantly higher power levels than those used here, with femtosecond laser pulses. This is different from the reversible photo-thermal reduction observed here. a) The ILs exclude the ber-to-chip coupling losses.

The insertion loss versus CW power is shown in
b) The device was only characterized at a single wavelength. c) We cannot precisely characterize the bandwidth due to the lack of suitable lasers. In our measurements, we achieve high PDLs of over 52.4 dB from 1500 to 1600 nm (with a variation less than 2 dB) and also a PDL of ~ 25.2 dB at 1064 nm.

Polarization-selective Microring Resonators
We coated GO lms onto integrated MRRs to implement polarization-selective MRRs, for applications such as polarization-handling devices in coherent receivers [53]. Figure 5(a) shows a schematic of the GO-coated polarization selective MRR, with the insets showing schematic atomic structure of GO and scanning electron microscope (SEM) image of the GO lm with up to 5 layers of GO. The unclad MRR made from high-index doped silica glass was fabricated via the same CMOS compatible processes as for the integrated waveguides in Sect. 2 [26, 54]. These devices have been extremely successful at RF and microwave applications as well as nonlinear optics and quantum optics . The ring and the bus waveguide had the same waveguide geometry as in Sect. 2. The radius of the MRR was ~ 592 µm, corresponding to a free spectral range of ~ 0.4 nm (~ 50 GHz). The gap between the ring and the bus waveguide was ~ 0.8 µm. We used the same method to couple CW light to the MRR. We fabricated and tested two types of GO-coated MRR polarizers, uniformly coated with 1 − 5 layers of GO and patterned (50 µm in length) with 10 − 100 layers of GO using the same photolithography and lift-off processes as for the patterned waveguide in Sect. 2. Gold markers, prepared by metal lift-off after photolithography and electron beam evaporation, were used for precise alignment and accurate placement of GO on the MRR (deviation < 20 nm). Microscope images of the integrated MRR uniformly coated with 5 layers of GO and patterned with 50 layers of GO are presented in Figs. 5(b) and (c), respectively. Note that although a number of concentric rings are shown, only the center ring was coupled with the through/drop bus waveguides to form a MRR − the rest were simply used to enable easy identi cation by eye. There are several factors that can limit the minimum pattern length, such as the thickness of the GO lm, lithography resolution, size of the GO akes, and thickness of the photoresist. By using oxidation and vigorous ultrasonics [41], we achieved ultrasmall GO ake sizes down to ~ 50 nm. For thin GO lms (< 10 layers), the pattern length was mainly limited by the lithography resolution and GO ake size, whereas for thick GO lms (> 50 layers), the thickness itself becomes the dominant factor. By using e-beam lithography to write patterns on a 300-nm-thick photoresist, we achieved short pattern lengths of ~ 150 nm and ~ 500 nm for 2 layers and 30 layers of GO, respectively. This con rms the high quality and resolution of the GO deposition and patterning process as well as the good adhesion between the GO lm and the integrated devices.  Fig. 8(a). The uncoated MRR had high extinction ratios (> 15 dB) and relatively high Q factors (180,000) (although signi cantly less than for buried waveguides [42]) for both polarizations. Those values decreased with GO layer number -particularly for the TE polarization, as expected. For the patterned MRR with 50 layers of GO, a maximum polarization extinction ratio (de ned as the difference between the extinction ratios of the TE and TM polarized resonances) of 8.3 dB was achieved. This can be further improved by optimizing the waveguide geometry, GO lm thickness, and coating length to properly balance the mode overlap and material anisotropy. The propagation loss of the GO hybrid integrated waveguides for TE and TM polarizations was obtained using the scattering matrix method to t the measured spectra in Figs. 6 and 7 and is shown for uniformly coated (0 − 5 layers) and patterned rings (0 − 100 layers) in Figs. 8(b-i) and (b-ii), respectively, along with the waveguide propagation loss obtained from the waveguide experiment (i.e., the experimental propagation loss in Fig. 3(a)). Since different resonances did not show a signi cant variation over small wavelength ranges, we only t one resonance around 1549.5 nm in each measured spectrum. The t coupling coe cients between the ring and the bus waveguide for TE and TM polarizations are ~ 0.241 and ~ 0.230, respectively. The close agreement re ects the stability and reproducibility of our layer-by-layer GO lm coating method. We also note that the propagation loss obtained from the ring resonator experiment is slightly higher than that obtained from the waveguide experiment, especially for the TE polarization. This probably results from photo-thermal reduction of GO in the resonant cavity at higher intensity. Combined with the capability of GO to enhance the nonlinear optical performance of photonic chips, [121][122][123][124][125][126][127][128][129][130][131] 2D GO lms represent an extremely attractive material platform for linear and nonlinear integrated photonic circuits.

Conclusion
We demonstrate waveguide polarizers and polarization-selective MRRs incorporated with layered GO lms. We achieve precise control of the placement, thickness, and length of the GO lms using layer-bylayer coating of GO lms followed by photolithography and lift-off. We achieve a high PDL of ~ 53.8 dB for the patterned GO-coated waveguide, and for the GO-coated integrated MRR an ~ 8.3-dB polarization extinction ratio between the TE and TM resonances. We nd that the PDL is dominated by material loss anisotropy of the GO lm for thin lms, and by polarization dependent mode overlap for thick lms. These integrated GO hybrid waveguide polarizers and polarization-selective MRRs offer a powerful new way to implement high-performance polarization selective devices for large-scale PICs.

Declarations
Competing interests: The authors declare no competing interests.     by using in-plane (TE polarized) n and k for (i) 2 and (ii) 20 layers of GO obtained from ellipsometry measurements in conjunction with mode solving software (assuming constant n, k for different layer numbers in each plot). (b) Experimental loss per layer given by the experimental loss in (a) divided by the number of layers for both polarizations. (c) Material loss of the GO lms for TE (kTE) and TM (kTM) polarizations as well as their ratio (kTE / kTM). The two yellow data points (labelled as "SE") show the inplane k measured by spectral ellipsometry (SE) for 2 and 20 layers of GO. (d) Fractional contribution to PDL from mode overlap (ηMO) and material loss anisotropy (ηMA). In (b) − (d), the solid data points refer to the results for a uniformly coated device, whereas the hollow points are for a patterned device.