Main

Graphene, a single sheet of carbon atoms in a hexagonal lattice, has attracted great interest because of its exceptional electrical and optical properties8,9,10,11,12,13. Graphene couples strongly to light, which enables observation of monolayer graphene under an optical microscope with the naked eye. A pristine graphene monolayer has a constant absorption of πα = 2.3% across the infrared and visible range, where α is the fine-structure constant (e2/c, where e is the electronic charge, is Planck’s constant divided by 2π, and c the velocity of light). Moreover, the broad optical absorption in graphene can be controlled through electrical gating: by shifting the electronic Fermi level, one can controllably change graphene’s optical transitions14,15. The strong electroabsorption effect, which has not yet been observed in bulk materials, originates from graphene’s unique electronic structure and its two-dimensional character. It implies that graphene has the potential to be used as the active medium in an optical electroabsorption modulator. However, one of the challenges involved in a direct graphene modulator is the limited absorption of a monolayer. This can be overcome by integrating graphene with an optical waveguide, which greatly increases the interaction length through the coupling between the evanescent waves and graphene.

A graphene-based waveguide-integrated electroabsorption modulator has several distinctive advantages. (1) Strong light–graphene interaction. In comparison to compound semiconductors, such as those exhibiting the quantum-well with quantum-confined Stark effect (QCSE)6, a monolayer of graphene possesses a much stronger interband optical transition, which finds applications in novel optoelectronic devices such as photodetectors16,17. (2) Broadband operation. As the high frequency dynamic conductivity for Dirac fermions is constant, the optical absorption of graphene is independent of wavelength, covering all telecommunications bandwidth and also the mid- and far-infrared18,19. (3) High-speed operation. With a carrier mobility exceeding 200,000 cm2 V−1 s−1 at room temperature20,21 (this is among the highest known), the Fermi level and hence the optical absorptions of graphene can be rapidly modulated through the band-filling effect. In addition, speed limiting processes in graphene (such as photocarrier generation and relaxation) operate on the timescale of picoseconds22, which implies that graphene-based electronics may have the potential to operate at 500 GHz, depending on the carrier density and graphene quality. (4) Compatibility with CMOS processing. The athermal optoelectronic properties of graphene and its CMOS-compatible integration processes at wafer scale23,24,25 make it a promising candidate for post-CMOS electronics, particularly for high frequency applications. With all these advantages, monolithic integration of a graphene electroabsorption modulator could open new routes to integrated photonics, with a compact footprint, low voltage operation and ultrafast modulation across a broad range of wavelengths.

Here we report the first waveguide-integrated graphene-based electroabsorption modulator, in which modulation is achieved by actively tuning the Fermi level of a monolayer graphene sheet. The gigahertz graphene modulator demonstrates a strong electroabsorption modulation of 0.1 dB µm−1 and operates over a broad range of wavelength, from 1.35 µm to 1.6 µm, under ambient conditions.

The structure of the electroabsorption modulator is schematically illustrated in Fig. 1. A 50-nm-thick Si layer was used to connect the 250-nm-thick Si bus waveguide and one of the gold electrodes. Both the silicon layer and the waveguide were shallowly doped with boron to reduce the sheet resistance. A spacer of 7-nm-thick Al2O3 was then uniformly deposited on the surface of the waveguide by atom layer deposition. A graphene sheet grown by chemical vapour deposition23,24,26 (CVD) was then mechanically transferred onto the Si waveguide. To further reduce the access resistance of the device, the counter electrode was extended towards the bus waveguide by depositing a platinum (10 nm) film on top of the graphene layer. The minimum distance between platinum electrode and waveguide was controlled at 500 nm, so that the optical modes of the waveguide remained undisturbed by the platinum contact. The excess graphene was removed by oxygen plasma, leaving only the regions on top of the waveguide and between the waveguide and the platinum electrode.

Figure 1: A graphene-based waveguide-integrated optical modulator.
figure 1

a, Three-dimensional schematic illustration of the device; a monolayer graphene sheet is on top of a silicon bus waveguide, separated from it by a 7-nm-thick Al2O3 layer (not shown). The silicon waveguide is doped, and connected to the electrode (right, shown gold) through a thin layer of silicon defined by selective etching. b, Left, cross-section of the device, with an overlay of the optical mode plot, calculated by finite element simulation. The waveguide was carrying a single optical mode, and was designed so as to maximize the field at the interface between the waveguide and the graphene, to maximize the absorption efficiency. The thin silicon layer and the Pt electrode, which is 500 nm away from the waveguide, have negligible influence on the optical mode. Right, a cross-section through the centre of the waveguide; the purple curve shows the magnitude of the electric field. The actual thicknesses of the graphene sheet and the Al2O3 in the simulation are 0.7 nm and 7 nm, respectively. c, Top-view optical microscope image of the waveguide. The Si waveguide was bent by 90° to change the polarization state between the input and the output light, to improve the signal-noise ratio. d, SEM image showing the boxed region in c at higher magnification, showing the detailed structure of the graphene modulator. False colours are used to highlight the Au electrode (yellow), the Pt electrode (Green), the graphene sheet (blue) and the waveguide (red). The width of the Si waveguide is 600 nm, while the distance between the Pt electrode and the Si waveguide is 500 nm. The bright multi-ring region beside the Au electrodes is due to a charging effect on the SiO2 layer in the SEM.

PowerPoint slide

Full size image

The cross-sectional view of the device structure and the optical field distribution of the guided mode are shown in Fig. 1b. The thin silicon layer and the platinum electrode adjacent to the waveguide have negligible effect on the mode profile. To further improve the electroabsorption modulation efficiency, the silicon waveguide was designed to have the electric field maximized at its top and bottom surfaces, so that the interband transitions in graphene are also maximized (Fig. 1b). As graphene only interacts with the tangential (in-plane) electric field of electromagnetic waves, the graphene modulator is polarization-sensitive, as are conventional semiconductor-based electro-optical modulators3.

Figure 1c shows a top-view optical microscope image of the device, and a close-up scanning electron microscopy image of the active region is given in Fig. 1d. The graphene sheet, highlighted in Fig. 1d by a false blue colour, covers only the waveguide region; this is to minimize the capacitance. The platinum electrode (green) is placed 500 nm away from the 600-nm-wide Si waveguide. Light was coupled in and out of the waveguide through tapered gratings, which contribute most to the overall loss of the system. The Si waveguide was bent 90° to change the polarization state between the input and the output light, to improve the signal-noise ratio.

Figure 2 displays the transmission of 1.53 µm photons through the waveguide at different drive voltages, VD. At low drive voltage (−1 V < VD < 3.8 V), the Fermi level EF(VD) of graphene is close to the Dirac point (EF(VD) < hν0/2), and interband transitions occur when electrons are excited by the incoming photons (hν0). The optical absorption of graphene is determined by the position of the Fermi level. By applying a drive voltage between the graphene and the waveguide, we can tune the Fermi level of the graphene, and therefore modulate the total transmission. With the current waveguide design, the modulation depth is as high as 0.1 dB µm−1, resulting in a graphene electroabsorption modulator with a footprint of merely 25 µm2. At large negative VD (<−1 V), the Fermi level is lowered below the transition threshold (EF(VD) = hν0/2) owing to positive charge accumulation. As a result, there are no electrons available for interband transitions, and hence the graphene appears transparent. On the other hand, at large positive VD (>3.8 V), all electron states are filled up, and no interband transitions are allowed. Ideally, there should be a sharp change in transmission at EF(VD) = hν0/2. In reality, this transition was broadened owing to defects in the graphene, and shifted to higher voltage owing to natural doping from the substrate27. When the modulator is in operation (that is, when no interband absorption is allowed), its insertion loss is negligible as the intraband absorption of graphene is extremely low at near-infrared wavelengths28.

Figure 2: Static electro-optical response of the device at different drive voltages.
figure 2

The main panel shows the modulation depth, normalized to the device length (40 µm), at different drive voltages (VD). Three regions can be seen, and their band structures are shown as insets. In the middle region (with VD in the range −1 V to 3.8 V), the Fermi level (EF, black dashed line) is close to the Dirac point and the interband transition is allowed from electron-occupied regions (red lines) to unoccupied regions (green lines). Thus the graphene sheet is absorptive to incident photons (hν), resulting in a modulation depth of 0.1 dB µm−1 and a miniaturized footprint of the modulator. In the left-hand region (with VD < −1 V), the Fermi level (EF) is lower than half the photon energy (−hν/2, blue dashed line) and there are no electrons available for the interband transition. In the right-hand region (with VD > 3.8 V), all electron states in resonance with incident photons (hν) are occupied, and the transition is forbidden. In both of the last two cases, the transmission increases. The natural doping from the substrate offsets the centre of the absorption curve from zero bias. The transmission is measured at a laser wavelength of 1.53 μm.

PowerPoint slide

Full size image

To measure the dynamic response of the graphene modulator, radio frequency signals generated by a network analyser were added on a static VD and applied to the modulator. The same 1.53-µm laser was used to test the modulator, and the out-coupled light was sent to a high-speed photodetector. Shown in Fig. 3 are the VD-dependent r.f. responses of the graphene modulator; gigahertz operation of the device at various drive voltages is obtained. Owing to the exceptionally high carrier mobility and saturation velocity of graphene, the bandwidth is not limited by the carrier transit time, but by the parasitic response of the device. With the platinum electrode placed 500 nm away from the waveguide, the total resistance of the system is reduced to around 600 Ω. This resistance, together with the capacitance (of the order of 0.22 pF), limits the operation bandwidth of the present device to about 1 GHz (see Fig. 3 legend).

Figure 3: Dynamic electro-optical response of the device.
figure 3

The main panel shows the response of the device as a function of frequency. Each curve corresponds to a different drive voltage: −2.0 V, −2.5 V, −3.0 V and −3.5 V (black, red, blue and green curve, respectively). The measured 3 dB bandwidths of the device are respectively 0.8 GHz, 1.1 GHz, 1.1 GHz and 1.2 GHz; the bandwidths are mainly restricted by the parasitic response of the device. Inset, low-frequency device response with different drive voltages, indicating that the device has best performance at a drive voltage of −4 V. The laser wavelength is 1.53 μm in the test.

PowerPoint slide

Full size image

The device response at low frequency (300 kHz) is shown in Fig. 3 inset. At low VD, the modulation response is weak, as the optical transmission is insensitive to VD . When the drive voltage is increased, the r.f. response increases to a maximum at VD = −4 V. As the drive voltage increases further, the modulation efficiency saturates, as graphene is then transparent within the modulation range of the bias voltage.

As the overall optical opacity of graphene is independent of wavelength and the high frequency dynamic conductivity for Dirac fermions is constant, the graphene electroabsorption modulator is therefore intrinsically broadband, unlike modulators that are based on optical cavities or resonant optical effects (such as QCSE). In order to investigate this broadband effect, we study the static response of the device to a white light source from a supercontinuous laser. The response is shown as a function of wavelength and VD in Fig. 4a; we refer to this as a two-dimensional (2D) transmission spectrum. A 3 dB modulation, corresponding to a transmission value of 2 (a.u.) in Fig. 4a, is achieved for a broad band of wavelengths, from 1.35 µm to 1.6 µm. Although a higher modulation depth and broader wavelength range are expected at a higher drive voltage, we chose to use low drive voltage not only to avoid spacer oxide breakdown but also because high drive voltages increase power consumption and violate voltage restrictions in CMOS devices.

Figure 4: Spectrum characterization of the optical modulator.
figure 4

a, The transmission of the device as a function of drive voltage and wavelength (1,350–1,600 nm). The transmission is normalized to VD = 1 V. The red dashed curve shows the trace for maximum transmission change rate, which unambiguously bends to higher drive voltages at shorter wavelengths. b, Squared photon energy versus critical drive voltage for maximum transmission change rate. The red dashed line shows a linear fit to the experimental data, which directly yields the Fermi velocity of vF = 0.9 × 106 m s−1.

PowerPoint slide

Full size image

Two-dimensional transmission spectra also allow us to determine the electronic band dispersion of the graphene. As the graphene electroabsorption modulation is dictated by the optical transition, hν = 2EF, the graphene modulator has a different response at different wavelengths. Operation at higher photon energy (hν) always requires a larger change in the Fermi level (EF). The trace of critical drive voltage (V) for maximum transmission change rate, shown as a dashed line in Fig. 4a, is defined by , where vF is the Fermi velocity, V0 is the voltage offset caused by natural doping, and η = 9 × 1016 m−2 V−1, as estimated using a parallel-plate capacitor model of our device. The relation between critical drive voltage and the square of the photon energy is plotted in Fig. 4b. The linear fit (red dashed line) yields the voltage offset (−0.8 V) and the Fermi velocity (0.9 × 106 m s−1), which agree with other reported values29.

We have demonstrated a graphene-based optical modulator that has broad optical bandwidth (1.35–1.6 µm), small device footprint (25 µm2) and high operation speed (1.2 GHz at 3 dB) under ambient conditions, all of which are essential for optical interconnects for future integrated optoelectronic systems. The modulation efficiency of a single layer of carbon atoms in a hexagonal lattice (graphene) is already comparable to, if not better than, traditional semiconductor materials such as Si, GeSi and InGaAs, which are orders of magnitude larger in active volume. The flexibility of graphene sheets could also enable radically different photonic devices. For example, graphene could be integrated with flexible substrate and plastic waveguides30. Or it could be used in novel geometries, such as a flexible modulator on a nano-optical cable. The recent development of large scale graphene synthesis and transfer techniques23 ensure its compatibility with the existing integrated electronics platform.