1. Introduction

Graphene-based optoelectronic devices have gained a lot of attention in the last few years due to the unique properties of the two-dimensional material such as very high carrier mobility, almost wavelength-independent tunable absorption as well as a gapless linear dispersion relation [1,2]. Recently, the commercial availability of large-area CVD-grown graphene with high purity made the fabrication of such devices on wafer-scale possible. Waveguide-integrated graphene-based electro-absorption modulators (EAMs) prove to be especially promising as they utilize the tunable optical absorption in graphene and can be integrated with existing telecom and datacom device concepts. For this kind of devices, modulation speeds of 10 Gbit/s and extinction ratios of 0.16 dB/µm have been demonstrated [3–5].

While the previous work concentrated on graphene-based EAMs on silicon waveguides, here we present the first fabricated graphene-polymer EAM (GP-EAM). The simple spin-coating processes involved in the fabrication of the passive polymer waveguide network enable a unique flexibility in the positioning of the graphene layers relative to the waveguide. The integration of these novel electro-absorption devices with the readily available passive and thermo-optical functions of the PolyBoard photonic platform offers a way towards high-speed, cost-effective integrated devices for telecom and datacom applications [6].

2. GP-EAM structure and simulations

The GP-EAMs presented in this work are based on two graphene layers separated by 35 nm of silicon nitride below a polymer waveguide. Figure 1(a) shows the cross section of the device. In order to avoid mechanical damage to the graphene layers, the surface of the underlying substrate should be sufficiently smooth and without steep edges. As in previous work the graphene was transferred onto previously structured waveguides, either a surface planarization as in [5] or spin-coating of an additional layer as in [3] was needed to obtain a smooth substrate surface. This increases the fabrication effort and, in the latter case, reduces the graphene-light interaction due to the increased distance between waveguide and graphene.

 

Fig. 1 (a) Cross section (not to scale) and (b) top view of a GP-EAM. The micrograph shows the structured and contacted graphene-silicon nitride-graphene stack of the fabricated structures before spin-coating of the waveguide layer.

Download Full Size | PDF

It should be noted that, unlike in previous work, here the graphene layers are located below the waveguide and not above. This enables the transfer of the graphene layers to the unstructured polymer bottom cladding, thus omitting the need for such surface planarization and facilitating a wafer-scale processing of graphene-based optoelectronic devices. Furthermore, the polymer waveguide and top cladding act as a cover for the active graphene layers shielding it from environmental influence.

Figure 1(b) shows the top view of the layout as well as a micrograph of a structure fabricated as outlined in section 4. The light is coupled from an external light source into the polymer access waveguide and guided to the integrated GP-EAM with a cross section as shown in Fig. 1(a). In the GP-EAM section the light propagates in the waveguide parallel to the graphene layers. In this waveguide-integrated approach the interaction length between light and graphene can be increased significantly compared to a perpendicular light transition through a graphene layer, which would yield a relatively low absorption of $\pi \alpha =2.3\%$ [2].

In the GP-EAM section the light intensity is modulated by the optical absorption in the graphene layers, which is tunable by applying a gate voltage V_G to the electrical contacts of the two electrically insulated graphene layers. In this dual graphene layer approach, one layer acts as the back gate for the other layer and vice versa, effectively forming a plate capacitor.

The resulting electrical field perpendicular to the parallel graphene planes induces a surface carrier density in the capacitor section, which can be expressed as a shift of the chemical potential $\mu$ and therefore the Fermi level E_F. Taking a doping-induced Fermi level shift into account and assuming a plate capacitor model for the capacitance C , the following relation holds around the graphene’s K point [7]:

Here $v_F=1.{10}^{6}m/s$ denotes the Fermi velocity in graphene [7], e the elementary charge, A the overlap area of upper and lower graphene layer, n the carrier density in the overlap area, ${\epsilon }_r=7.2$ the relative permittivity of the silicon nitride isolation layer, V₀ the shift in the neutrality point of the graphene due to doping, and d its thickness. As the graphene layer was not intentionally doped we set V₀ to 0 V. We will discuss this when presenting the experimental results. Assuming a time constant of ${\tau }_{inter}=50$ for the interband absorption [8], ${\tau }_{intra}=1.2$ for the intraband absorption [9] and a temperature of 296 K, the refractive index and absorption of a monolayer graphene were calculated with respect to the chemical potential $\mu$ based on the optical conductivity as in [10]. The results served as input for the simulation of the TE00 mode of the GP-EAM section as sketched out in Fig. 1 (a) by means of a finite difference mode solver. All calculations and simulations were carried out for a wavelength of 1550 nm, as this is the most important spectral region for fiber-optical communications. Dimensions and refractive indices are specified in the beginning of section 4. Absorption in materials other than graphene and additional loss contributions were not taken into account.

Figure 2(a) shows the simulated TE effective refractive index and absorption as a function of the chemical potential. The gate voltage is calculated from the chemical potential using Eq. (1). At a chemical potential of 0.4 eV, the graphene layers switch from absorbing to the transparent state. As a consequence, the GP-EAM absorption drops from 0.050 dB/µm to 0.001 dB/µm. Hence the extinction ratio is 0.049 dB/µm. When considering gate voltages of 9 V and 12 V, the TE mode exhibits the same effective refractive index of 1.468, resulting in a similar mode distribution as shown in the inset of Fig. 2(b), which is useful for future high-speed modulation of the device. Hence, these working points have been chosen for further simulations of a 25 µm long GP-EAM using eigenmode expansion. At 9 V gate voltage the graphene layers are still highly absorbing, resulting in an attenuation of 1.17 dB inside the GP-EAM section. While increasing the voltage to 12 V, the absorption dramatically decreases due to Pauli blocking of the optical interband transitions in the graphene. Hence, the attenuation inside the 25 µm long GP-EAM decreases to 0.02 dB corresponding to an extinction ratio of 0.046 dB/µm. Taking into account the switching voltage of 3 V, the optical transmission changes with a rate of 0.015 dB/µm/V. For the simulation, a 25 µm long GP-EAM section was chosen, as it corresponds to the length of the device investigated experimentally in section 4. In order to achieve higher total extinction ratios suitable for data transmission, the length could be increased, as the absorption scales with the GP-EAM length. Assuming the extinction ratio of 0.046 dB/µm between ON and OFF state, a 10 dB extinction ratio could be achieved with a just 217 µm long section.

 

Fig. 2 (a) Simulated dependence of the TE effective refractive index and absorption in the waveguide-integrated GP-EAM section as shown in Fig. 1 on the chemical potential for a wavelength of 1550 nm. The corresponding gate voltage for an insulating 35 nm silicon nitride layer as calculated with Eq. (1) is indicated in the top axis. (b) Simulated transmission of the TE mode through a 25 µm long GP-EAM as well as cross sections of the respective mode fields at voltages of 9 V (off) and 12 V (on) at the same wavelength. The light intensity is color coded.

Download Full Size | PDF

3. Raman spectroscopy on graphene transferred to polymer

In order to gain insights on the influence of the transfer process on the electronic and vibrational properties of the graphene layers and assess the quality of the transferred graphene, we performed a Raman spectroscopy analysis on different sets of samples, including the starting material, i.e., CVD-grown graphene on copper, and graphene transferred to polymer. All Raman measurements were carried out in backscattering geometry under ambient conditions using a laser excitation wavelength of 532 nm and a 600 lines/mm grating, yielding a spectral resolution of approximately 2.5 cm⁻¹.

For statistical analysis, Raman mappings were performed using a motorized xyz stage across areas of 15 µm by 50 µm. It has to be noted, that Raman spectroscopy for the given structures was only possible for the lower graphene layer (on top of the polymer) and not for the upper graphene layer (on top of the silicon nitride insulating layer). Figure 3(a) shows two representative Raman spectra of CVD-grown graphene on copper and on polymer. As indicated in the figure, we observe the main characteristic Raman modes of graphene for both samples, i.e., the first-order G mode around 1585 cm⁻¹ and the double-resonant 2D mode around 2680 cm⁻¹. We furthermore note that we do not observe any Raman signal from the defect-induced D mode at approximately 1340 cm ⁻¹, indicating the high crystalline quality of the graphene layers, both, before and after the transfer. Besides the intrinsic Raman peaks of graphene, we observe additional Raman modes in the spectrum of the transferred graphene in the spectral range between 1300 cm⁻¹ and 1800 cm⁻¹. These modes arise from Raman-active vibrations from the polymer layer underneath the graphene. Following the analysis of Lee et al. [11], we present a decomposition of the influence from mechanical strain and doping on the G and 2D mode positions in Fig. 3(b), where each point corresponds to a Raman measurement on different locations on the samples. As can be seen, the data points for graphene on copper span primarily along a line with a slope of approximately 2.6, which is commonly identified with biaxial strain [12]. In contrast, the slope of the data points for graphene on polymer exhibit a slope of approximately 2.2, indicating uniaxial strain [11].

 

Fig. 3 (a) Representative Raman spectra of CVD-grown graphene on copper foil (blue) and graphene transferred to polymer (red) using a laser excitation wavelength of 532 nm. The characteristic G and 2D modes of graphene are labelled. The spectrum of graphene on polymer exhibits additional polymer-related Raman modes in the G mode spectral region. Spectra are vertically offset for clarity. (b) Analysis of the G mode and 2D mode positions for graphene on copper foil and on polymer. Colors are chosen as in (a). More than 150 measurements are taken at different locations on each sample. The inserted parallelogram helps to disentangle strain and doping effects on the Raman frequencies [11,12]. (c),(d) Histograms of the 2D-mode FWHM (full width at half maximum) for graphene on copper and polymer, respectively.

Download Full Size | PDF

Furthermore, from the spread of the data points for both samples in Fig. 3(b), we can infer that graphene on copper presents a highly inhomogeneous strain distribution as compared to graphene transferred to polymer. The strain variation of graphene on polymer is estimated to be $\Delta \epsilon =0.3\%$ [13]; the strain variation for graphene on copper is at least by a factor of two higher. The observation of inhomogeneous strain distributions for graphene on copper is further confirmed by a comparison of the 2D mode full width at half maximum (FWHM) for both samples, shown by histograms in Fig. 3(c) and 3(d). As demonstrated in [14], the 2D mode linewidth is indicative for strain variations in the graphene layer. Hence, by a direct comparison of both histograms in Fig. 3(c) and 3(d), we conclude that the transferred graphene on polymer is very homogeneous within the investigated measurement range of approximately 15 µm by 50 µm.

Finally, we analyze doping effects in the transferred graphene layer on polymer. Since the Raman shift of the G mode is symmetric with respect to $E_F=0eV$ for either n- or p-type doping, we cannot unambiguously identify the type of doping solely by the present Raman measurements. However, we can infer that the graphene on polymer presents a higher doping level as compared to graphene on copper. A precise evaluation of the doping type and level in the transferred graphene as well as a separate Raman analysis for bottom and top graphene layer might be the scope of future studies.

4. Fabrication of GP-EAMs and experimental results

GP-EAM structures as outlined in Fig. 1 have been fabricated on 4-inch wafer level. The first layer of CVD-grown graphene on copper foil is transferred to a 4-inch silicon wafer with 8 µm polymer bottom cladding. The transfer process was still limiting the graphene lengths to about 25 µm. Structuring with oxygen plasma reactive ion etching (RIE) followed. The electrical contacts are formed by evaporation of 120 nm gold. The subsequent insulating layer of 35 nm silicon nitride is deposited via PECVD and structured with RIE before the graphene transfer, structuring, and contacting steps are repeated for the second layer. The wafer fabrication is finished with spin-coating and structuring the polymer waveguide layer (refractive index n = 1.479) as well as the polymer top cladding layer (n = 1.449). The passive polymers used for the fabrication are of the ZPU-12 series manufactured by ChemOptics Inc. The waveguides feature a cross section of 3.2 µm by 3.2 µm ensuring single-mode operation in the O-, C- and L-band, respectively.

From transmission line measurements we extract a sheet resistance of 900 Ω/sq after the fabrication for the lower graphene layer. The resistance is higher (1400 Ω/sq) for the upper graphene layer. This might be attributed to the increased surface roughness that we observe after structuring the first graphene layer and the deposition of the silicon nitride insulation layer. The contact resistance for both layers is below 5 kΩ∙µm and the gold-graphene contact exhibits ohmic behavior which is critical for the RF characteristics of the future devices. The individual devices on the processed wafer were separated by means of dicing, and we measured the voltage-dependent optical transmission through the GP-EAM at wavelengths between 1500 nm and 1600 nm with 0 dBm optical power.

Shown in Fig. 4(a) is the voltage-dependent transmission through a 25 µm-long GP-EAM fabricated as described above. Similar to other graphene devices relying on a field effect, the graphene-polymer EAM structure exhibits a pronounced hysteresis due to charge traps at the interface between graphene and silicon nitride. As these charge traps have large time constants (~ms), such a hysteresis is not expected to affect high-speed modulation performance [3,15]. Note that the curves for upwards and downwards sweep do not intersect at −15 V and + 15 V. This is due a time delay between the upwards and downwards measurements in our setup. As a result of this hysteresis we have not been able to deduce the neutrality point shift V₀. The transmission values include fiber-to-chip coupling losses and propagation losses in the 3 mm long access waveguides.

 

Fig. 4 Measured transmission at a wavelength of 1550 nm (a) and spectra at different voltages (b) of the TE mode for a 25 µm long GP-EAM.

Download Full Size | PDF

The maximum extinction achieved for this GP-EAM is 1.4 dB while sweeping the voltage from + 12.7 V to −9.7 V corresponding to an extinction ratio of 0.056 dB/µm. This is in good agreement with the maximum extinction ratio of 0.049 dB/µm yielded from the simulations in section 2. It is lower than the value of 0.16 dB/µm in silicon-on-insulator platforms [3] due to the lower confinement of the propagating mode. The confinement can be increased by placing the complete graphene-silicon nitride-graphene stack into the center of the waveguide, where the intensity is maximum. This leaves the GP-EAM capacitor section and its voltage-tuning characteristics unchanged, while we estimate to increase light-graphene interaction by a factor of around 3. Due to the spin-coating of the polymer layers, this can be achieved with relatively low further process development effort by coating the lower half of the waveguide layer, followed by the formation of the graphene-silicon nitride-graphene stack and the coating of the upper half of the waveguide layer.

Comparing the measurements shown in Fig. 4(a) with the simulation results from section 2, there are several points to be noted. While the maximum extinction ratios from simulation and measurement (0.049 dB/µm and 0.056 dB/µm respectively) match well, the real device exhibits hysteresis as discussed above. Furthermore, the absorbing state of the graphene is shifted towards gate voltages below −9 V for the downwards voltage sweep and below 0 V for the upwards sweep. This corresponds to a shift in the neutrality point of the graphene due to unintentional doping of the graphene layers, which was discussed in section 3. Furthermore, the voltage-dependence of the transmission is almost linear around −6.0 V (downwards voltage sweep) and + 13.5 V (upwards voltage sweep), which is desirable for the operation of such a device. At these points, the transmission changes with maximum rates of 0.0081 dB/µm/V and 0.0093 dB/µm/V respectively. This is lower than the value 0.015 dB/µm/V yielded from the simulations. We believe that this difference stems from inhomogeneous doping, either laterally within a single graphene layer or, at the same lateral position, between both layers. Either case leads to locally different offset voltages and hence an averaged, reduced absorption state. However, the Raman spectroscopy indicates a relatively uniform doping of the graphene at least for the bottom layer. It would hence be highly desirable to be able to characterize the properties of top graphene layer after the deposition on top of the silicon nitride in the future as well. As loss contributions other than the absorption in the graphene have been neglected in the simulation, a direct comparison of the absolute losses is not feasible. The minimal total fiber-to-fiber losses in the GP-EAM are measured to be 7.2 dB. The propagation losses in the access waveguides presumably contribute a large share to these losses due to the large rms roughness of the polymer substrate in the order of 50 nm after the graphene structuring. The roughness stems from the fact that both, the graphene and the polymer, are being etched efficiently by oxygen plasma. This etching-induced roughness could be decreased by introducing a some nanometers thick silicon nitride film between polymer and graphene acting as an etch stop in future device iterations. However, the available data does not allow a detailed break-down of the loss contributions of the GP-EAM section and the polymer access waveguide.

Figure 4(b) shows the measured transmission spectra at tuning voltages. Again, the transmission does not perfectly match between Figs. 4(a) and 4(b) due to hysteresis and a time delay between the measurements. Note that the transmission spectra are almost flat and that the tuning of the absorption in the GP-EAM does not introduce an additional wavelength dependence. This is due to the almost wavelength-independent opto-electronic properties of graphene [2]. As the device features polymer access waveguides with low refractive index contrast, efficient coupling of optical fibers is possible, avoiding bandwidth-limiting coupling via gratings. In our experiment we used lensed fibers with ~0.3 dB coupling loss per facet. Therefore, GP-EAMs offer a way to make use of this unique feature of graphene so that such devices might be used for modulating light in all optical tele- and data-communications bands.

For the current devices it was not possible to measure the modulation bandwidth and characterize the RF performance due to the design of the electrical contacts. Furthermore, the transfer process and fabrication limited the working device to a graphene length of 25 µm. In order to achieve extinction ratios of around 10 dB suitable for data transmission, the fabrication process has to be optimized to enable device lengths of around 200 µm und additionally reduce the optical losses inside the GP-EAM. Therefore, while the working principle of a graphene-based EAM could be shown, future work will address these important points in the optimization and characterization of GP-EAMs.

5. Conclusion

To the best of our knowledge, graphene-based EAMs integrated with passive polymer waveguides have been fabricated and characterized for the first time. The extinction ratio is measured to be 0.056 dB/µm, and the results are in good agreement with the simulations for this structure. Furthermore, the effect of the transfer process and the polymer substrate on the graphene layers has been investigated by means of Raman spectroscopy indicating a high crystalline quality and reduced strain distribution. The GP-EAMs offer a way to integrate the promising optoelectronic features of graphene with a low index-contrast waveguide platform enabling a use of graphene-based optoelectronic devices from the visible to the near infrared spectral region. Hence, the combination of GP-EAMs with passive optical functionalities in polymer waveguides, such as TFF-based polarizing beam splitters and U-grooves for passive fiber-to-chip coupling can enable the development of novel high-speed, cost-effective and flexible devices for telecom and datacom applications.