Heaters, controlling the refractive index of the photonic components by the thermo-optic effect, are indispensable components for integrated photonic systems [1, 2]. They are used as tuning elements in wavelength division multiplexing units (WDM) [3, 4] or for thermal stabilization of resonant structures [5, 6]. The most common design concept for waveguide heaters are metal wires fabricated above or next to the target waveguide [7–10]. As a general rule, the smaller the gap between the heater and the heated element the more efficient the heating process takes place. However, due to the strong interaction between light and metal, the heater and the waveguide must be sufficiently separated, typically in the order of 1 µm [9, 10]. This leads to a low tuning efficiency, high power consumption and slow system response time. A significantly better heating efficiency can be achieved by directly passing a current through the silicon waveguide [11, 12]. However, this method has several constraints, as it requires doping of the silicon, it cannot be used for tuning active devices such as microring based modulators, and it cannot be applied to insulating waveguide materials like SiN or AlN.

Graphene, the two-dimensional carbon crystal [13], has been considered as promising material for transparent electrodes in display applications and for window heating because of its low light absorption at visible wavelengths, and its good electrical conductivity [14–16]. The integration of large scale grown graphene into silicon photonic environments has already been proven for the case of modulators [17], detectors [18] or heat spreading layers [19]. Compared to the metal heater design, graphene’s electrical conductivity paired with the transparency at infrared wavelengths enables a significant reduction of the spacing between heater and waveguide. A proof of principle of a graphene heater on a silicon microdisc was recently shown in [20, 21] and on a silicon microring in [22], however without demonstrating any performance advantage over metal heater devices. Here we demonstrate experimentally and by simulations that graphene can be used as a highly efficient heating element to tune the refractive index of a silicon waveguide significantly outperforming state of the art metal heating devices.

In this work, the temperature dependent resonance shift in a silicon microring is exploited to assess the static and dynamic performance of the developed graphene waveguide heater. The resonance wavelength of a microring resonator depends on the ring circumference L, the effective refractive index n_eff and the order of the resonance m by λ_res = Ln_eff / m. However, as the refractive index of silicon depends strongly on temperature, heaters are needed to thermally stabilize the resonance wavelength of microrings. These rings are commonly used in silicon photonic circuits where they carry out WDM and signal modulation operations. The microrings used here were defined by electron beam lithography on silicon-on-insulator substrates with 2 µm buried oxide and 220 nm top silicon thickness. 350 nm wide rib waveguides were etched 180 nm into the silicon leaving a slab of 40 nm. The microring had a perimeter diameter of 18 µm. A false color SEM image of the photonic layer and the graphene heater is shown in Fig. 1(a). Grating couplers optimized for TE polarization and 1310 nm wavelength were fabricated to couple light into and out of the waveguide. To achieve the required cladding thickness to separate the heater from the waveguide, the chip was coated with a stack consisting of spin-coated HSQ to smoothen the surface and atomic layer deposited Al₂O₃ to define the final thickness. The cladding stack had a total thickness of 240 nm measured from the top surface of the waveguide (40 nm HSQ plus 200 nm Al₂O₃, see Fig. 1(c). For a distance of 240 nm, neither the graphene nor the transferred impurities related to the graphene growth were expected to cause relevant absorption. Single layer graphene grown by CVD on copper foil was subsequently transferred by the standard PMMA method [23], patterned using oxygen plasma and contacted by Ni/Al electrodes. The width of the graphene heater was 3.5 µm. The heaters have a resistance of 4 to 6 kΩ. On reference samples, a charge carrier mobility between 1000 and 3000 cm²/Vs was measured and a Fermi-level in the order of −0.3 to −0.5 eV was estimated. Raman measurements were performed to confirm that most of the graphene is single layer material, see Fig. 1(b) for a typical spectrum of the used graphene. The inset shows the 2D peak of a complete mapping of the heater.

 

Fig. 1 (a) False color micrograph of the graphene heater (orange) and the photonic ring (red). The light travels along the direction of the arrows. The scale bar is 6 µm. (b) Typical Raman spectrum of the graphene. The inset shows the 2D peak of a Raman mapping of the entire heater. (c) False color cross section of the waveguide and the cladding. The scale bar is 200 nm. (d) Transmission before (blue) and after the heater fabrication (red). The inset shows resonances before and after fabrication at 1311,7 nm.

Download Full Size | PDF

The optical properties of the ring resonators were measured by coupling TE-polarized light from a tunable laser source to the input port of the device. The transmitted power was detected at the ring’s through port. In Fig. 1(d), a transmission spectrum of a ring resonator before the graphene heater was fabricated is plotted in blue color. Clear resonances with a state-of-the-art quality factor Q = 20,000 were observed. The spectral distance between two resonances, the free spectral range (FSR), was ∆λ = 7.2 nm. The transmission spectrum after the heater fabrication is also shown in Fig. 1(d) (red line). The quality factor of the ring remained unchanged, demonstrating that the graphene did not significantly contribute to the loss. The observed red shift of 40 pm after fabrication of the heater translates into a change of the effective refractive index ∆n = 1.4 x 10⁻⁴. As this change can be explained already by a temperature change of the measurement environment of only 0.8 K, it is not possible to resolve any parasitic impact of the heater on the effective refractive index. Based on the simulations explained later in the manuscript, the change in effective index due to the graphene heater was expected to be ∆n = 1.2 x 10⁻⁶, well below the observed shift.

In Fig. 2(a) the transmission spectra for different heating powers are shown for drive voltages from 0 to 8 V in 1 V steps. The resistance of this heater was 4 kΩ, giving a power of up to 15 mW for 8 V supply voltage. For increasing heater temperature all resonances shifted to longer wavelengths, in accordance with the positive thermo-optical coefficient of ∆n/∆T = 1.8 x 10⁻⁴ K⁻¹ in silicon [24]. We selected the resonance around 1350 nm because here the grating coupler showed the highest efficiency. The location of the resonance minimum as a function of the heating power is plotted in Fig. 2(b). From a straight line fit to the data, a heating efficiency of 0.33 nm/mW was extracted. This translates into a power of 22 mW required to shift the resonance one FSR. Compared to available metal heaters, this power consumption is comparable to or below the best reported values [7–9, 25, 26]. It should be noted that neither the quality factor nor the extinction ratio of the resonances changed with the heating power, indicating that the heat distribution was uniform across the ring and did not affect the coupling from the waveguide to the ring. The thermal breakdown power threshold of the device was 33 mW, which is 50% above the power required to tune the ring within one FSR.

 

Fig. 2 (a) Transmission spectrum of the ring resonance for a heater power from 0 to 15 mW. (b) Shift of the ring resonance wavelength as a function of the heater power. The dashed line is a linear fit to the data with a slope of 0.33 nm/mW.

Download Full Size | PDF

The long term stability of the graphene heater was studied in a 48 h endurance test, where a constant bias of 4 V was applied to the heater. During the test period the resistance and the optical transmission spectrum were measured repeatedly. After an initial drop of the resistance from 5.5 to 4.8 kΩ, which can be explained by a thermal drift of the doping level in the graphene, the resistance remained constant during the measurement period, see Fig. 3. At 4 V bias the measured resonance wavelengths after 24 h and 48 h were at 1357.89 and 1357.91 nm, respectively. The initial and final resonance without bias were at 1357.00 and 1357.01 nm, respectively. The results show reliable and reproducible operation of the graphene heater.

 

Fig. 3 Resistance of the heater device as a function of time for a constantly applied supply voltage of 4 V.

Download Full Size | PDF

Another important characteristic parameter for thermo-optical tuning systems is their response time, depending on the thermal coupling and isolation. To measure the rise and fall time of the temperature change in the silicon waveguide, a square signal was applied to the heater. To stay in the small signal regime, where the electro-optical response of the microring can be considered linear, a low excitation voltage of 0.1 V amplitude and 0.05 V offsetcorresponding to a peak power of 2 µW at a 60 kHz frequency was chosen. The optical power transmitted by the ring was measured by a photoreceiver connected to an oscilloscope. The modulation frequency of 60 kHz was well above the lower cut off frequency (10 kHz) of the detector and close to the highest frequency at which the maximum and minimum values of the optical transmission were still reached. From the dynamic response in Fig. 4, rise and fall times (10% to 90%) of 3.0 and 3.6 µs were extracted, respectively.

 

Fig. 4 Dynamic response of the heater measured by detecting the transmitted light. The electrical modulation signal has a square shape (0 to 0.1V with 0.05 V offset) at a frequency of 60 kHz. The corresponding peak electric power is 2 µW. The 10% to 90% rise time is 3 µs and the corresponding fall time is 3.6 µs.

Download Full Size | PDF

These results are compared to published values of metal heating systems in Table 1. In addition to the response time, the distance of the heater to the waveguide edge, and the power per FSR as well as a figure of merit (FOM) was also listed. The power required to tune one FSR enabled a geometry independent comparison. To account for different thermal environments of the compared systems, because a better thermal insulation reduces the power consumption but on the expense of a slower response, we compared the systems on the basis of the FOM. This number was derived by multiplying the power per FSR with the average value of the rise and fall time (t_Rise + t_Fall)/2. Compared to conventional tungsten heaters on top of the waveguide without thermal insulation [8], the graphene heater is faster and consumes 40% less power. The device concept based on heaters located on the Si slab next to the waveguide outside the evanescent field on the silicon slab was about a factor of 2 slower and consumed twice the energy compared to the presented graphene heater [26].

Table 1. Comparison of key parameters obtained in this work to different metal heaters on Si waveguides and a graphene heater on a microdisk.

View Table

Etching thermal insulation trenches into the silicon reduced the power consumption of conventional metal heaters to about the value of the presented graphene heater. However, this improvement was only possible via a significant increase of the response time [25], which in turn is represented by a factor of 5 larger figure of merit. In our graphene heater design, thermal isolation trenches can also be included if for a certain application the response time plays only a subordinated role but energy efficiency is most important. The results of this work are also compared to a recent work on an integrated graphene heater device fabricated on a silicon microdisk [21]. This device requires the highest power per FSR and has a slow response time, resulting in a high FOM not outperforming state of the art metal heaters. In addition, the Q factor in this device is heavily affected by the graphene heater and not constant for different power levels, which is a serious limitation for most applications. A comparison to the results presented in reference [22] is not possible because key parameters are not given.

Absorption loss in metal defines the spacing between photonic components and heaters. In contrast to metals, absorption in graphene can be tuned [27] and hence the effect on photonic structures can be minimized when graphene is transparent. In order to explore this minimum distance, simulations of the light absorption as a function of the cladding thickness for different graphene parameters, doping and scattering rate, were performed. A schematic of the modeled system is shown in Fig. 5(a). The stack consisted of patterned and cladded Si on SiO₂ and graphene. The cladding was hydrogen silsesquioxane (HSQ) and an Al₂O₃ layer on top to reproduce the conditions of the experiment. In the simulation, the total thickness of the dielectric stack on the waveguide was varied between 0 ≤ d ≤ 300 nm. For d ≤ 40 nm, the cladding consisted only of the HSQ because the HSQ is required to smoothen the surface in our fabricated devices. For d > 40 nm the HSQ thickness was kept constant at 40 nm and the Al₂O₃ thickness was varied to achieve the total thickness. It is noted that the exact composition of the cladding stack had a negligible influence on the absorption and nearly identical results were obtained for a pure SiO₂ cladding. The simulations were based on the complex optical conductivity of graphene, depending on the Fermi energy, the scattering rate Г, and the temperature and were carried out using the finite difference method in MATLAB. Details on the simulations are given in reference [27]. In Fig. 5(b), the absorption coefficient of graphene is plotted as a function of the distance d measured from the top of the waveguide to the graphene layer for different doping levels and different scattering parameters of the graphene. Two Fermi levels 0.3 eV and 0.6 eV were selected, where the first one marks a typical value for our fabricated devices and the latter 0.6 eV is a high, but technologically accessible Fermi level [28]. In Fig. 5(b), the blue curve shows the simulation of Г₁ = 13.5 x 10¹² s⁻¹ (µ ≈2000 cm²/Vs) at a Fermi level of 0.3 eV corresponding to the graphene used in our experiments, and the red curve Г₂ = 2 x 10¹² s⁻¹ (µ = 6750 cm²/Vs) at a Fermi level of 0.6 eV. For the latter parameter set, the absorption is about 2 orders of magnitude lower due to Pauli-blocking. In several devices with different cladding thicknesses the absorption in graphene was measured [27, 29] and the values added to Fig. 5(b), indicated by the blue symbols. We observe a very good agreement with the simulations for thin claddings where enough absorption occurs. In the presented microring with a cladding thickness of 240 nm, the absorption α is connected to the quality factor Q by Q = λ / (FSR*α*R) [30]. We did not observe a significant change of the Q factor and hence, could not extract an absorption related to the graphene. The simulated graphene absorption for this device was 5.8 x 10⁻⁴ dB/µm, corresponding to a change of the Q factor of approximately 10%, which is within the measurement tolerance of the Q factors. The simulated absorption for this device with 240 nm spacing is marked in Fig. 5(b) by a grey dot on the blue curve. The grey dashed line connects the same absorption values simulated for graphene used here (blue curve) and for highly doped graphene with a high mobility (red curve). The distance between the graphene heater and the waveguide at which the same absorption occurs could be reduced from 240 nm to approximately 50 nm without affecting the Q factor. If a reduction of the Q factor by 50% can be accepted, the distance can be further decreased and the heater can be placed directly on top of the silicon waveguide. The reduced spacing would lead to a reduced power consumption and faster response time because of a better thermal coupling. Both parameters affect the FOM in a positive way so that the presented graphene heater device has the potential of further improvement as soon as a temperature stable doping is available.

 

Fig. 5 (a) Profile of the TE mode at λ = 1310 nm in the cladded waveguide. The distance between the upper surface of the waveguide and the top of the cladding is labeled d. (b) Simulated absorption of the graphene as a function of the distance d to the waveguide for different Fermi-levels and scattering parameters Г₁ = 13,5 x 10¹² s⁻¹ and Г₂ = 2 x 10¹² s⁻¹. The blue symbols are measured absorption values.

Download Full Size | PDF

In summary, we presented a new concept for realizing waveguide heaters using graphene as infrared transparent heating element. The graphene heater was located at a distance of only 240 nm from the waveguide without detectable light absorption. Compared to available metal heaters, this compact design increases the thermal coupling to the waveguide significantly. A clear advantage in terms of energy consumption and response time is observed, demonstrating the advantage of this new concept for waveguide heaters. Simulations suggest that a placement of the graphene in close proximity to the silicon is possible, enabling the best possible thermal coupling. The heater presented here already outperforms the state-of-the-art with the potential of further significant improvement.