Graphene Phase Modulator

TL;DR

This paper presents a compact, high-efficiency graphene-based phase modulator capable of 10Gb/s data transmission, significantly outperforming traditional silicon modulators in efficiency and integration potential.

Contribution

The work introduces a graphene phase modulator with unprecedented modulation efficiency and compactness, suitable for high-speed optical communication systems.

Findings

Achieved 10Gb/s modulation speed.

Demonstrated 0.28Vcm modulation efficiency.

Operates with 2V peak-to-peak voltage in a push-pull setup.

Abstract

We demonstrate a 10Gb/s Graphene Phase Modulator (GPM) integrated in a Mach-Zehnder interferometer configuration. This is a compact device, with a phase-shifter length of only 300$\mu$m, and 35dB extinction ratio. The GPM has modulation efficiency of 0.28Vcm, one order of magnitude higher compared to state-of-the-art depletion p-n junction Si phase modulators. Our GPM operates with 2V peak-to-peak driving voltage in a push-pull configuration, and it has been tested in a binary transmission of a non-return-to-zero data stream over 50km single mode fibre. This device is the key building block for graphene-based integrated photonics, enabling compact and energy efficient hybrid Si-graphene modulators for telecom, datacom and other applications

Full text

Optical communication systems for telecom and datacom implement high-order digital modulation schemes, such as amplitude or phase-shift keying, in order to enhance spectral efficiency and increase the data transmission capacity of telecommunication networksSeimetz2009 . This is possible by encoding several bits of information in fewer symbols, leading to the same amount of data, but transmitted at lower speed, resulting in reduced spectral bandwidth. Spectral narrowing is a key technique for optimizing data capacity in modern wavelength-division multiplexing (WDM)Nakazawa2010 allowing higher data rates, improved spectral efficiency and robust tolerances to transmission impairments, such as chromatic dispersionSeimetz2009 .

The modulator is a key component in optical communicationsAgrawal2010 . To encode the information in complex modulation formats, phase modulation (PM) is neededSeimetz2009 . In integrated Si photonics, this is achieved by exploiting the free-carriers plasma dispersion effectSoref1987 , whereby changes in electron and hole densities result in changes in the Si refractive index and absorption. In state-of-the-art Si modulatorsXiao2013 ; Streshinsky2013 ; Denoyer2015 ; Reed2013 ; Dong2012 ; Fresi2016 ; Milivojevic2013 the complex index change can be driven by p-n junctionsXiao2013 ; Streshinsky2013 ; Denoyer2015 ; Reed2013 ; Dong2012 ; Fresi2016 or capacitorsMilivojevic2013 built in Si waveguides, and applying a voltage to deplete or accumulate free-carriers interacting with the propagating light. High-speed for simple and complex modulation formats has been demonstratedXiao2013 ; Streshinsky2013 ; Denoyer2015 ; Reed2013 ; Dong2012 ; Fresi2016 ; Milivojevic2013 either with a depleted p-n junction under reverse biasXiao2013 ; Streshinsky2013 ; Denoyer2015 ; Reed2013 ; Dong2012 ; Fresi2016 , or with a Si-insulator-Si (SIS) capacitorMilivojevic2013 operating in the accumulation regime. However, the Si free-carrier effect in the depleted p-n junctionSoref1987 typically requires mm-size devices for accumulating a $\pi$-phase shiftReed2013 with a 1-3V driving voltage. This limits the modulation efficiency, defined as the product of the $\pi$-phase-shift voltage and length (VπL)Reed2008 , and increase energy consumption (energy per bit)Miller2012 . Another challenge is that plasma dispersion is always accompanied by absorptionSoref1987 . As a result, there is a tradeoff between optical loss, VπL and footprint, since higher free carriers densities with stronger plasma dispersion come at the expense of increased optical absorptionReed2008 or limited dynamic range, i.e. the maximum intensity or phase change for a given driving voltageReed2008 . Therefore, novel solutions with increased VπL, reduced optical loss, and miniaturized footprints are needed.

Graphene is appealing for photonics and optoelectronics because it offers a wide range of advantages compared to other materials, such as Si and other semiconductorsBonaccorso2010 ; Kim2011 ; Ferrari2015 ; CastroNeto . In particular, it is ideally suited for integration into Si photonicsBonaccorso2010 ; Hanson2008 ; Mak2008 due to its large optical modulationLiu2011 ; Liu2012 ; Hu2016 ; Phare2015 ; Sorianello2015 , broadband photodetectionKoppens2014 ; Goykhman2016 , high-speed operationPhare2015 ; Koppens2014 , and CMOS (complementary metal oxide semiconductor) compatibilityLiu2011 ; Liu2012 ; Hu2016 ; Phare2015 ; Sorianello2015 ; Koppens2014 ; Goykhman2016 . To date, several graphene based amplitude modulators have been reported based on electro-absorptionLiu2011 ; Liu2012 ; Hu2016 ; Phare2015 . However, graphene phase modulators (GPMs) are necessary for all functionalities requiring phase change, such as binary and complex modulation formats, switching, phased arrays, etc. GPMs are also expected to have an improved efficiency-loss figure of merit compared to Si based modulatorsSorianello2015 . Graphene integration with Si photonics opens a new paradigm for developing compact, efficient and low-loss integrated PMs outperforming state-of-the-art Si devicesXiao2013 ; Streshinsky2013 ; Denoyer2015 ; Reed2013 ; Dong2012 ; Fresi2016 ; Milivojevic2013 . Due to its unique opto-electronic properties, optical losses in single layer graphene (SLG) can be electrically suppressed due to Pauli blockingWang2008 ; Falkovski2008 ; Stauber2008 . This modulates the absorption, with a refractive index change much larger than that the free carriers effect in SiSorianello2015 . SLG on Si waveguides with effective refractive index (i.e. the ratio between the light phase velocity in the waveguide and in vacuum) changes larger than 10*-3* have been demonstratedSorianello2016 , about ten-times larger than state-of-the-art Si p-n junction waveguides in depletion modeReed2008 . Consequently, SLG can provide a unique combination of strong electro-refractive effect and optical transparency when operated in the Pauli blocking regime.

Here we demonstrate a GPM included in a Mach-Zehnder interferometer (MZI),Fig.1a, in which PM is realized in the form of a Si-insulator-SLG (SISLG) capacitor and the Si waveguide is used as a gating electrode to the SLG, Fig.1b). By applying a bias to the capacitor, charge is accumulated on the SLG electrode, shifting its Fermi level, EF, and modifying its complex conductivityWang2008 ; Falkovski2008 ; Stauber2008 ; Sorianello2016 . In this way, both the effective index and the optical loss of a SLG-functionalized Si waveguide can be tunedSorianello2016 . If the SLG is doped beyond the Pauli blocking condition (E${}_{F}>$0.4eV at 1.55$\mu$m), the device operates in the low-loss ”transparency” region, where the only loss contribution is given by the Si waveguide propagation loss without SLG. In this regime, PM is dominant with respect to amplitude changesSorianello2015 , with an enhanced modulation efficiency (VπL$<$0.26Vcm)Sorianello2015 compared to typical Si photonics modulators operating in depletion mode (VπL$>$2Vcm)Reed2008 .

Our device is fabricated in a standard Si photonics platform using a Si-on-insulator (SOI) substrate (see Methods). The photonic structure consists of a balanced MZI with 3dB multimode interference (MMI) couplersSoldano1995 to split and combine the optical beam, and grating couplersRoelkens2006 to couple light in and out of the device, Fig.1a). The Si waveguide is designed to support a single transverse electric (TE) in-plane polarized optical mode. The waveguide is B doped to reduce the Si resistance and achieve high-speed operationReed2008 . SLG is grown by chemical vapor deposition (CVD) and then wet transferred (see Methods) onto 10nm planarized Si dioxide (SiO2) top cladding, Fig.1b). We use different SLG lengths onto the two MZI arms (300 and 400$\mu$m) to intentionally introduce a bias phase difference in the balanced interferometer structure for characterization purposes. The quality and uniformity of SLG after device fabrication are characterized by Raman spectroscopy (see Methods). The spectra show negligible D to G intensity ratio (Fig.2), indicating that no significant degradation and/or defects are introduced during the fabrication processCancado2011 ; Bruna2014 . From the Raman G peak position, Pos(G), full width at half maximum, FWHM(G), and intensity and area ratios of D and G peaks we estimateBasko2009 ; Das2008 E${}_{F}<$100meV ($\sim$200meV) at the longer top (shorter bottom) arms of the MZI.

To test the electro-refractive effect, we first perform static characterizations by measuring the MZI output power as a function of bias applied to the GPMs on both MZI arms, Fig.3. A 1550nm laser source is coupled to the MZI input using a SMF, while the optical power at the MZI outputs, i.e the bar and cross ports, is collected with a similar fibre and monitored by a power meter (see Methods). In our balanced MZI the cross port is expected to be at maximum when the phase difference between the two arms is zero, while the bar port tends to be at zero powerBorn1970 . We measure an output power difference$\sim$5dB between the two ports when no voltage is applied. We assign this imbalance to the difference in absorption and phase accumulation between the two MZI arms caused by the different SLG lengths, doping and defects, Fig.2. By applying a bias to the SISLG capacitors, we decrease the bar port power to zero, with an extinction ratio (ER, ratio between maximum and minimum of the transmission)$>$35dB, Fig.3b). This is due to the phase change introduced in the two arms by electrical gating the SLGs. The 35dB ER is due to a phase difference between the two arms approaching $\pi$, and this is evidence of interferometric behaviorBorn1970 , with a considerable electro-refractive effect. In particular, we measure V${}_{\pi}\sim$7.25V on the 400$\mu$m SISLG capacitor, corresponding to VπL$\sim$0.28, very close to the theoretical predictionSorianello2015 . This is at least a 8-fold improvement compared to state-of-the-art p-n junction based Si MZI modulatorsDenoyer2015 . The VπL efficiency can be doubled by using a SLG-insulator-SLG phase modulator taking advantage of the effect of two SLGs on the effective index of the waveguideSorianello2015 .

To test the modulator electro-optical (EO) bandwidth and operation speed, we first characterize the frequency response by using an electrical vector network analyzer (VNA) (see Methods). We measure a 3dB roll-off frequency of 5GHz at 4V, Fig.4a). The bandwidth is limited by the RC time constant primarily due to the SISLG capacitor and the series resistances. The overall capacitance C is the sum the capacitor on top of the waveguide core (C${}_{ox}=\epsilon_{0}\epsilon_{SiO_{2}}w_{wg}/t_{ox}\sim 1.6$fF/$\mu$m, where $\epsilon_{0}$ is the vacuum dielectric constant, $\epsilon_{SiO_{2}}$ is the SiO2 dielectric constant, wwg is the waveguide core width and tox the oxide thickness) and the parasitic capacitor Cp due to the SLG overlap on the Si slab (C${}_{p}=\epsilon_{0}\epsilon_{SiO_{2}}w_{ol}/t_{p}\sim 0.2$fF$\mu$m, where w${}_{ol}\sim 1\mu$m is the overlap and tp the oxide thickness between SLG and Si slab). Cp depends on the alignment between transferred SLG and waveguide core, and it can contribute up to 20% of the overall C. The series resistance, R, stems from different contributions: the Si resistance (Rsi, estimated$\sim$5k$\Omega\cdot\mu$m, see Methods), the SLG lead from the waveguide core to the metal contact, Rg (estimated$<$1k$\Omega\cdot\mu$m, see Methods), the metal to SLG contact resistance, Rc. The latter is estimated as$\sim$10k$\Omega\cdot\mu$m by using a transfer length measurement (TLM) on test samples (see Methods). In our device, the metal/SLG contact resistance is the main factor limiting the radio-frequency (RF) bandwidth. This can be improved by reducing Cp and Rc. Contact resistances down to 100$\Omega\cdot\mu$m have been reportedLeong2014 , such values would increase the RF bandwidth up to 30GHz and beyond, leading to high frequency operations comparable with state-of-the-art modulatorsXiao2013 ; Streshinsky2013 ; Denoyer2015 ; Reed2013 ; Dong2012 ; Fresi2016 ; Milivojevic2013 .

We now discuss the optical response when the MZI modulator is driven by a non-return-to-zero (NRZ) signalSeimetz2009 . We measure the MZI NRZ eye diagram with the two GPMs biased at 3.5V and operating in push-pull configuration, with 2V peak-to-peak and 231-1 pseudorandom binary sequence (PRBS)Agrawal2010 signals. Fig.3b plots an open eye diagram at 10Gb/s with$\sim$4dB ER and$\sim$6dB signal-to-noise ratio (SNR). With a random bit stream, the individual GPM average energy consumption is given byMiller2012 C($\Delta$V)2/4, where $\Delta$V is the voltage variation driven to the contacts in order to charge/discharge the GPM capacitance. In our device, we get$\sim$1pJ/bit. The energy consumption could be reduced with the SLG-insulator-SLG configuration, potentially allowing us to halve the driving voltageSorianello2015 .

We then carry out bit error rate (BER) measurements when the NRZ signal is transmitted in a standard single mode fiber (SMF) over different distances, Fig.5. We measure the BER as a function of optical signal-to-noise ratio (OSNR) in back-to-back configuration, and after propagation in SMF spools of different lengths from 10 to 50km. We have error free operation (BER$<$10*-10*) up to 30km, while at longer propagation distances a BER floor appears. However, considering the state-of-the-art systems currently employing a soft-decision forward error correction (SD-FEC)Infinera2015 , with overhead in the 7-25% range, and enabling a pre-FEC BER threshold up to 3.4x10*-2* or even higherRahman2015 , our modulator exhibits error free operation up to 50km.

In summary, we demonstrated a graphene-silicon phase modulator operating in the GHz regime. We included the GPM in a MZI device demonstrating a static modulation depth of 35dB and modulation efficiency of 0.28Vcm, outperforming state-of-the-art Si-based p-n junctions and comparable to the SIS-capacitor based Si modulators. The modulator operates at 10Gbit/s, showing an open eye diagram and error free transmission over 50km single mode fibre. These results pave the way to the realization of graphene modulators for a wide range of telecom applications where the phase modulation is crucial.

We acknowledge funding from the Graphene Flagship Project ID: 696656, ERC Grant Hetero2D and EPSRC Grants EP/ 509 K01711X/1, EP/K017144/1, EP/N010345/1, EP/M507799/ 5101 and EP/L016087/1. We acknowledge Graphenea for the provision of CVD samples.

I Methods

The Si photonic device is prepared on the IMEC iSiPP25G Si on insulator (SOI) platformAbsyl2015 . The MZI interferometer is based on Si ridge waveguides with 60nm slab and a core cross-section 480nm$\times$220nm. SLG is placed on the Si waveguide separated by a 10nm spacer of high quality thermal SiO2. Graphene is grown by CVD on Cu foils, as described in Ref.Li2009 , and then transferred onto our Si samples following the procedure described in Ref.Zurutuza2015 . Graphene patterning is done by using a bilayer stack of PMMA/IX845, while etching is done using conventional O2-plasmas. The metal contacts on SLG and Si are processed separately in two consecutive steps. SLG is contacted with a single 50nm thick Palladium (Pd) layer, while the metal contact on Si consists of a Titanium (Ti)/Platinum (Pt)/Gold (Au) stack with thickness 20nm/20nm/30nm deposited on the Si surface after HF cleaning to remove the native SiO2. Contact resistance and mobility are evaluated on different samples with equivalent processing. Typically values of 10k$\Omega\cdot\mu$m are extracted at the charge neutrality point using the TLM methodPolitou2016 ; Politou2017 , with mobility$\sim$1500cm2V*-1s-1*.

After device fabrication, the quality and uniformity of SLG at both arms of the MZI is monitored by Raman spectroscopy using a Horiba LabRam Evolution Raman spectrometer with 514.5nm laser and optical power below 0.1mW. The Raman spectra from the top and bottom arms (GPMs) of the MZI are normalized to the Si Raman peak at 521cm*-1*. The 2D peak is a single sharp Lorentzian with full width at half maximum FWHM(2D)$\sim$36cm*-1* ($\sim$39cm*-1*) at the top (bottom) arm of the MZI, a signature of SLGFerrari2013 . The spectra show negligible I(D)/I(G)$\sim$0.12 ($\sim$0.16), indicating that no significant degradation and/or defects are introduced during the fabrication processCancado2011 . We estimate a defects density$\sim$2.3$\cdot$1010cm*-2* and$\sim$6.5$\cdot$1010cm*-2* at top and bottom GPMs respectivelyCancado2011 ; Bruna2014 . Pos(G) is$\sim$1592 cm*-1* ($\sim$1593cm*-1*), with FWHM(G)$\sim$17cm*-1* ($\sim$19cm*-1*). The 2D peak position, Pos(2D) is$\sim$2691cm*-1* ($\sim$2693cm*-1*), while the 2D to G peak intensity and area ratios I(2D)/I(G) and A(2D)/A(G), are$\sim$3.7 (2.6) and 8.2 (5.3), indicating a p-doping$<$100meV ($\sim$200meV)Basko2009 ; Das2008 at the top (bottom) arm of the MZI. These correspond to carriers concentrations$<$1$\cdot$1012cm*-2* ($\sim$2.3$\cdot 10^{12}$ cm*-2*)Das2008 .

Rsi is evaluated by numerical electrical simulations of the GPMs cross section with a commercial-grade device simulator that self-consistently solves the Poisson and drift-diffusion equationslumerical . We use Si doping and resistivity as obtained from the IMEC iSiPP25G technologyAbsyl2015 . For SLG, we estimate a sheet resistance $<$400$\Omega/\square$ when the SLG is gated beyond the Pauli blocking condition (E${}_{F}>$0.4eV). For E${}_{F}>$0.4eV the carrier concentration exceeds 1.3$\cdot$1013cm*-2* ($\left|n_{s}\right|=\pi^{-1}\cdot(E_{F}/(\hbar v_{F}))^{2}$, where ns is the carrier concentration, $\hbar$ the bar Planck constant and vF is the Fermi velocityWang2008 ), the mobility degradation from the ungated value is negligibleHirai2014 . As the SLG lead from the waveguide core to the metal contact is 2.5$\mu$m, Rg can be estimated$<$400$\Omega/\square\cdot 2.5\mu$m=1k$\Omega\cdot\mu$m.

The input/output optical coupling is obtained through cut single mode fibres with cleaved output surfaces. The input fibre is connected to the laser source with a fibre polarization controller to maximize the input coupled light. We use a tunable external cavity laser fixed at 1550nm. The output fibre is connected to a high sensitivity power meter to measure the static characteristics in Fig.3. Two ground signal (GS) high frequency probes apply DC and RF signals to the two MZI GPMs. For small-signal RF bandwidth measurements we use an electrical VNA. Its output is connected through a 50$\Omega$ matched high frequency cable to the GS probe contacting the GPM. We use a bias-tee to combine RF power and DC bias. We set a DC bias of 4V and an RF power of -17dBm. The light at the output of the MZI is modulated by the RF signal from the VNA and collected by a low noise, high frequency photodetector connected to the VNA input. The signals for the eye diagram and BER measurements are generated by a pattern generator (PG) and collected with a digital sampling oscilloscope (eye diagram) and a BER tester. The PG provides the 231-1 pseudorandom binary sequence (PRBS) at 10Gb/s. The signal and inverted signal are sent to the two GPMs through the RF cable and bias-tee. We use the optical input of the oscilloscope to collect the light out of the MZI GPM and visualize the eye diagram of Fig.4b. For BER evaluation, we used a high frequency photo-receiver collecting the light at the output of the modulator and connected to the BER tester electrical input.