# Fileset [s41467-023-42194-6.pdf](https://mdr.nims.go.jp/filesets/8176cfec-2759-4e89-86fa-bb4379fe7c50/download) ## Creator Alberto Montanaro, Giulia Piccinini, Vaidotas Mišeikis, Vito Sorianello, Marco A. Giambra, Stefano Soresi, Luca Giorgi, Antonio D’Errico, [K. Watanabe](https://orcid.org/0000-0003-3701-8119), [T. Taniguchi](https://orcid.org/0000-0002-1467-3105), Sergio Pezzini, Camilla Coletti, Marco Romagnoli ## Rights [Creative Commons BY Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/) ## Other metadata [Sub-THz wireless transmission based on graphene-integrated optoelectronic mixer](https://mdr.nims.go.jp/datasets/05e604a1-a414-4731-ac75-b45bb7869128) ## Fulltext Sub-THz wireless transmission based on graphene-integrated optoelectronic mixerArticle https://doi.org/10.1038/s41467-023-42194-6Sub-THz wireless transmission based ongraphene-integrated optoelectronic mixerAlberto Montanaro 1,2 , Giulia Piccinini3,4, Vaidotas Mišeikis4,5,Vito Sorianello 1, Marco A. Giambra6, Stefano Soresi6, Luca Giorgi7,Antonio D’Errico 7, K. Watanabe 8, T. Taniguchi 9, Sergio Pezzini 10,Camilla Coletti 4,5 & Marco Romagnoli 1Optoelectronics is a valuable solution to scale up wireless links frequency tosub-THz in the next generation antenna systems and networks. Here, wepropose a low-power consumption, small footprint building block for 6Gand 5G new radio wireless transmission allowing broadband capacity (e.g.,10–100Gb/s per link and beyond). We demonstrate a wireless datalink basedongraphene, reaching setup limited sub-THzcarrier frequency andmulti-Gbit/s data rate. Our device consists of a graphene-based integrated optoelectronicmixer capable of mixing an optically generated reference oscillatorapproaching 100GHz, with a baseband electrical signal. We report >96GHzoptoelectronic bandwidth and −44dBupconversion efficiencywith a footprintsignificantly smaller than those of state-of-the-art photonic transmitters (i.e.,<0.1mm2). These results are enabled by an integrated-photonic technologybased on wafer-scale high-mobility graphene and pave the way towards thedevelopment of optoelectronics-based arrayed-antennas for millimeter-wavetechnology.The fast riseof data-hungry andperformance-demanding applications,e.g., digital twins, AI computing, autonomous driving, robotic surgery,digital biology, call for theneed for novel access technology. Advanced5G and 6Gwill make use of New Radio (NR) access technology1. While5 G keeps its promises, advanced 5G has more potential toimplement2. 5 G NR access operates in the sub-1GHz to 100GHzspectrum2, and supports various use cases, including time-criticalservices for consumers, enterprises, and public institutions acrossmultiple sectors1,2. NR is currently a standard in the 3rd GenerationPartnership Project (3GPP)2, and a new generation of antenna systemsis being developed3. In the last decade, mobile networks haveincreased carried trafficbyalmost 300 times, andnetwork speeds haveincreased hundreds of times1. In the journey toward 6G, new extre-mely demanding and immersive applications will require a furtherupgrade of signal processing functions with improved signal integrity,frequency, and capacity2. The need for broader bandwidth capacitiesin wireless networks can be satisfied by increasing the carrier fre-quency (CF). Therefore, the sub-THz frequency range (90–300GHz) isbecoming more and more appealing4–6 and, given the potential toprovide hundreds of Gbit/s over short distances, it will be a key enablerfor 6G systems4. Figure 1 depicts a scenario including three types ofantenna systems, envisioned for sub-THz transmissions. The first is asingle antenna transmitter, used in high throughput point-to-pointradio links with application in the backhaul transport network7 (Fig. 1,Received: 29 March 2023Accepted: 3 October 2023Check for updates1Photonic Networks and Technologies Lab – CNIT, Via G. Moruzzi,1, 56124 Pisa, Italy. 2TeCIP Institute, Scuola Superiore Sant’Anna, via G. Moruzzi 1, 56124Pisa, Italy. 3NEST, Scuola Normale Superiore, Piazza San Silvestro 12, 56127 Pisa, Italy. 4Center for Nanotechnology Innovation @NEST, Istituto Italiano diTecnologia, Piazza San Silvestro 12, 56127 Pisa, Italy. 5Graphene Labs, Istituto Italiano di Tecnologia, Via Morego 30, 16163 Genova, Italy. 6Inphotec,CamGraPhIC srl, viaG.Moruzzi 1, 56124Pisa, Italy. 7Ericsson Research, viaG.Moruzzi 1, 56124 Pisa, Italy. 8ResearchCenter for Electronic andOpticalMaterials,National Institute for Materials Science, 1-1 Namiki, Tsukuba 305-0044, Japan. 9Research Center for Materials Nanoarchitectonics, National Institute forMaterials Science, 1-1 Namiki, Tsukuba 305-0044, Japan. 10NEST, Istituto Nanoscienze-CNR and Scuola Normale Superiore, P.zza S. Silvestro 12, 56127Pisa, Italy. e-mail: amontanaro@cnit.itNature Communications | (2023) 14:6471 11234567890():,;1234567890():,;http://orcid.org/0000-0002-2844-0433http://orcid.org/0000-0002-2844-0433http://orcid.org/0000-0002-2844-0433http://orcid.org/0000-0002-2844-0433http://orcid.org/0000-0002-2844-0433http://orcid.org/0000-0003-2204-0778http://orcid.org/0000-0003-2204-0778http://orcid.org/0000-0003-2204-0778http://orcid.org/0000-0003-2204-0778http://orcid.org/0000-0003-2204-0778http://orcid.org/0000-0003-3165-8465http://orcid.org/0000-0003-3165-8465http://orcid.org/0000-0003-3165-8465http://orcid.org/0000-0003-3165-8465http://orcid.org/0000-0003-3165-8465http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0003-3701-8119http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0002-1467-3105http://orcid.org/0000-0003-4289-907Xhttp://orcid.org/0000-0003-4289-907Xhttp://orcid.org/0000-0003-4289-907Xhttp://orcid.org/0000-0003-4289-907Xhttp://orcid.org/0000-0003-4289-907Xhttp://orcid.org/0000-0002-8134-7633http://orcid.org/0000-0002-8134-7633http://orcid.org/0000-0002-8134-7633http://orcid.org/0000-0002-8134-7633http://orcid.org/0000-0002-8134-7633http://orcid.org/0000-0002-4274-5620http://orcid.org/0000-0002-4274-5620http://orcid.org/0000-0002-4274-5620http://orcid.org/0000-0002-4274-5620http://orcid.org/0000-0002-4274-5620http://crossmark.crossref.org/dialog/?doi=10.1038/s41467-023-42194-6&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1038/s41467-023-42194-6&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1038/s41467-023-42194-6&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1038/s41467-023-42194-6&domain=pdfmailto:amontanaro@cnit.itred links). The second uses multiple phased-array antenna used in thewireless access networks to improve antenna gain in a chosendirection3 (Fig. 1, green links). A further case is the sub-THz short-rangeRADAR imaging with increased CF allowing spatial resolution 100GHz is themajor bottleneck limiting theperformances of sub-THz systems16, since phase noise (PN) grows withfrequency. Indeed, to generate the high CF, a relatively low frequency(tens of GHz) electronic LO is upconverted, i.e., multiplied, amplified,and filtered multiple times17,18 (Fig. 2a). This technique is power-consuming due to nonlinear signal processing and electrical signalamplifiers19, and the resulting signal exhibits modest performance interms of PN. A frequency multiplier increases the PN by at least20Log(N), where N is the multiplication factor20, and a non-negligiblecontribution comes from the noise figure of cascaded amplificationstages21. Moreover, by increasing f x l (where f is the operating fre-quency and l is the typical length of the transmission lines inside thecircuit), electronic circuit design becomes critical15, and mismatch inlength and loadingbetweendifferential lines can cause skew and signaldegradation15. As CF increases, it is thus necessary to use compensa-tion techniques for PN reduction, transmission line parasitics mitiga-tion, and protection from spurious electromagnetic fields. As a result,the system’s complexity increases21–24.The second scheme is based on common microwave-photonicstechniques, advantageously employed to perform heterodyne mixingand overcome the issues related to a fully electronic approach9–12.Highly pure LO’s can be obtained from two optical wavelengths spacedby the required frequency25,26, coupled to a fast photodetector (PD).These two optical wavelengths can generate very stable electrical toneswith low PN. E.g., by using optical combs, ref. 25 reports <100 dBc/Hz at10 KHz from the 300GHz CF and ref. 26 reports −118 dBc/Hz at 10MHzfrom the 331GHz CF. Several wireless links employing photonic trans-mitters have been reported in recent years (see Table 1). These schemesuse optically generated LO’s, that are converted from the optical to theelectrical domain by making use of fast photodiodes27. One of the twooptical wavelengths composing the optical LO is modulated with thedatastream using a modulator27 (Fig. 2b). This scheme does not requireelectronic upconverters. As summarized in Table 1, this is themost usedtechnique to obtain photonic-based wireless transmitters. Despite thelimiteduseofhigh-frequency electronics, theoverall systemcomplexityis only partially reduced compared to the fully electronic approach,since two discrete high-frequency optoelectronic devices are required,i.e., a modulator and a detector, often inter-connected using opticalfibers (Fig. 2b), see Table 1. In this configuration, the electrical basebanddatastream is first converted in the optical domain by an optical mod-ulator and then it is converted again in the electrical domain using aPoint-to-pointPoint-to-mul�pointAutomo�ve RADARλ1 + λ2THz Upconverted datastreamBaseband Fig. 1 | Antenna systems for sub-THz transmission. In these scenarios a grapheneoptoelectronic mixer could be used inside a transmitter to perform frequencyupconversion. In red, point-to point links, allowing directional communicationsbetween two fixed network nodes. In green, point-to-multipoint links usingphased-array antenna allowing to reach multi-users through beamforming,including objects in movement. In blue, automotive RADAR sensing for, e.g.,autonomous driving. Each node of the network contains a frequency converterthat allows the upconversion of a baseband information in the sub-THz frequencyrange. A graphene optoelectronic mixer (G-OEM), shown on the right, is a minia-turized object that can be potentially used in all these scenarios to perform fre-quency upconversion of baseband signals in the sub-THz range, in a very effectiveway. The device uses an optical local oscillator (LO) constituted by two opticalwavelengths (λ1,λ2) instead of an electrical LO, thus performing excellent stabilityand easy tunablity of the sub-THz carrier frequency (CF), allowing, e.g., the reali-zation of compact phased-array antenna systems based on optoelectronics.Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 2photomixer. Each one of these two steps is characterized by conversionlosses. In addition, the sub-THz antenna element is of the order of amm2, while the footprint of such transmitters is large since two discreteactive devices (a modulator and a PD) are reported in most of theliterature, as shown in Table 1. Instead, the next-generation phased-array antenna systems such as massive MIMOs28 requires the use ofintegrated devices, and electronics is currently the only commercialsolution for high-density integration29. As a matter of fact, most of thedemonstrated photonic transmitters present in literature are based onbulk discrete systems, with few examples of integrated or partiallyintegrated solutions (see Table 1) always exhibiting footprint >1mm2.These dimensions are still incompatible with high-density integrationand preclude the realization of antenna array systems based on thisapproach. Indeed, one must consider a minimum amount of mm-waveelectronic circuitry that would still be necessary for signal conditioning,thus the total area of the single antenna element would easily exceedthe mm2 limit.We propose a third, different approach to implement a photonictransmitter, offering significant advantages in terms of power con-sumption, bandwidth capacity, footprint size, and PN. It is based on aself-mixing device30, i.e., an optoelectronic mixer based on graphene(G-OEM). An OEM is a fast photomixer able to generate the high-frequency LO through optical detection, and directlymix the LOwith abaseband electrical datastream30 (see Fig. 2c and inset of Fig. 1) withoutthe use of an optical modulator. Compared to the common photonic-aided wireless transmitters (Fig. 2b), only one G-OEM is needed toupconvert the electrical signal by internally mixing the self-generatedLOwith the electrical datastream30, this last never being encoded in theoptical domain during the upconversion process. Graphene hasremarkable optoelectronic properties such as broadband opticalabsorption31, short photocarriers lifetime32–34 and high carriermobility35,36. Graphene photonics and optoelectronics technologiesare evolving toward wafer-scale production37–39 and are compatiblewith the standard silicon technological platforms37. OptoelectronicTable 1 | Comparison of the most recent implementations of sub-THz photonic transmittersRef. Year Scheme Carrier (GHz) Datarate1 Distance Total datarate FootprintThis work 2022 Int. Optoelectronic mixer 93 4Gbit/s 2m 4Gbit/s <0,1mm2100 2014 Int. modulator + UTC-PD 90 0,1Gbit/s 0,2 cm 0,1 Gbit/s ~1mm248 2018 modulator + UTC-PD 150 264Gbit/s 3m 1 Tbit/s bulk50 2022 modulator + UTC-PD 230 192Gbit/s 115m 192Gbit/s bulk50 2022 modulator + UTC-PD 230 240Gbit/s 5m 240Gbit/s bulk27 2013 modulator + UTC-PD 237.5 100Gbit/s 40m 100Gbit/s bulk101 2022 modulator + UTC-PD 275 50Gbit/s 1m 50Gbit/s Si mod/UTC-PD2102 2020 modulator + UTC-PD 300 40Gbit/s 1.4m 40Gbit/s Si mod/UTC-PD3103 2018 modulator + UTC-PD 300 100Gbit/s 50 cm 100Gbit/s bulk104 2020 modulator + UTC-PD 300 115Gbit/s 110m 115Gbit/s bulk105 2019 modulator + UTC-PD 310 10Gbit/s 58m 30Gbit/s bulk106 2018 modulator + UTC-PD 350 100Gbit/s 2m 100Gbit/s bulk49 2020 modulator + UTC-PD 380 300Gbit/s 2,8m 600Gbit/s bulk107 2019 modulator + UTC-PD 408 107Gbit/s 10m 107Gbit/s bulk108 2019 modulator + UTC-PD 450 132Gbit/s 1,8m 132Gbit/s bulk109 2016 modulator + UTC-PD 500 20Gbit/s 50 cm 160Gbit/s bulk1(single channel, single polarization). 2Silicon modulator (3mm) + bulk UTC-PD. 3Silicon modulator (3mm) +bulk UTC-PD.Electronic transmi�erPhotonic transmi�er based on Graphene optoelectronic mixerConven�onal technologies for sub-THz datastream upconversionPhotonic transmi�er based on modulator + detectora) c)b)<0,5 mm100-200 μm<0,5mm2-3 mm<0,5mmλ1 + λ2Data IN(baseband)UpconverteddatastreamElectrical sub-THz LOSub-AmplifierFrequencydoublerFrequency chainData IN (baseband)Upconverted datastreamElectricallow-frequency LODatastream Up-conversionMMICx2Upconverteddatastreamλ1λ2Fast PDLOMZM Datastream Up-conversion( )Optoelectronic conversionData IN (baseband)Fig. 2 | Sub-THz transmitters technologies. a, b Conventional implementationsof sub-THz transmitters: a Fully electronic approach, comprising frequency mul-tipliers for sub-THz LO generation. Once generated, the sub-THz LO is routedtowards an electronic upconverter implemented using a Microwave MonolithicIntegrated Circuit (MMIC) composed by electronic components, operating in thesub-THz frequency range. b Photonic-based transmitter, comprising an opticalmodulator and a fast photodetector (PD). The two components are rarelyintegrated into the same chip, and aremost often inter-connected through opticalfibers (see Table 1). c Our proposed implementation, comprising only one smallwaveguide-integrated optoelectronic mixer based on graphene (G-OEM). It allowsus to reduce the complexity compared to the conventional approach shown ina, b, and to obtain full integration compared to the approach shown in b), thusenabling the realization of compact phased-array antennas based onoptoelectronics.Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 3mixing using graphene PDs has been also demonstrated40–43 andoperations up to 67GHz have been reported42,43, but the bandwidth ofG-OEMs can be much larger44–46 as can be deduced in analogy withgraphene ultrafast PDs43–45,47. Recent direct bandwidth measurementshave shown >500GHz bandwidth46, according to theory predicting a3-dB bandwidth exceeding 500GHz44. An important figure of merit ofupconverters is the conversion efficiency. This is defined asPconv = PRF/PIF, where PIF is the input electrical power, while PRF theupconverted power. The conversion efficiency of a G-OEMdepends onthe charge carrier mobility43, i.e. by the graphene quality35,36. Conver-sion efficiencies exceeding −20 dB have been predicted by theory43.This value, together with high-frequency operation, makes G-OEMspromising to realize sub-THz wireless links with performances poten-tially exceeding state-of-the-art photonic-based systems27,48–50. In thiswork we report a wireless datalink based on graphene, reachingmulti-Gbit/s datarate and sub-THz CF. This result was enabled by a G-OEMused as frequency upconverter at the transmitter stage of the link. Theproposed photonic transmission scheme relies on a silicon nitride(SiN) waveguide-integrated G-OEM embedded with a sub-THz copla-nar waveguide (CPW). The G-OEM exhibits high conversion efficiency(−44 dB) and record (92–96GHz) frequency operation compared tostate-of-the-art G-OEMs41–43. The device was fabricated using hBN-encapsulated single crystal CVD graphene36,51,52 with high carriermobility (27000 cm2 V−1 s−1 at room temperature). We report a 4Gbit/swireless link at sub-THz (93GHz) CF, by directly mixing a basebandelectrical signal (2 GHz CF) with a 91 GHz optical LO using one singledevice. The device reveals high potential in terms of speed, far beyondthe limits of our experimental setup. The photonic upconverter foot-print is <<0.1mm2, considering the active area, without access pads foron chip probing. This active area is of the sameorder as the active areaoccupied by electronic up/down converters14,53,54, and is much smallerthan any other integrated-photonic transmitter based on the schemein Fig. 2b. The small footprint opens to application scenarios in whichan integrated-photonic transmitter can be used. Figure 3 shows aphased-array antenna system implementation approach, based onintegrated G-OEMs. In the example, the antenna comprises 256 ele-ments. Each antenna element would contain an optoelectronic mixerthat converts an optically distributed LO in the electrical domain, anamplifying electronic stage, and an antenna. Phase shifting function-ality may be required for beamforming/steering implementation. Thisis done at the IF signal level using analog or digital approaches55,56, or atthe LO level56,57. In the latter case, optical phase shifters need to beintegrated in the antenna element58. The sub-THz upconversion func-tionality is currently realized using sub-THz electronics, comprisingmulti-stage mixers and amplifiers14, schematically shown in Fig. 2a.With this conventionalmethod, signals withCF of >100GHz are routedto each antenna element, before being transmitted59. The proposedsolution based on our G-OEM only requires baseband (IF) electronics(BW <10GHz), while the upconversion is carried out by mixing the IFwith an optically generated LO, routed to each antenna element usingphotonicwaveguides insteadof sub-THzelectrical transmission lines59.The distribution of the sub-THz LO bymeans of electrical transmissionlines can be in general power hungry since it requires the use of activesub-THz elements e.g., buffers and amplifiers59, which are necessary tocompensate for transmission lines losses and keep the SNR at accep-table levels15,16,21,22 all along the path through the antenna elements.Using the integrated G-OEM the sub-THz signal distribution is avoidedin favor of optical routing of the LO to each antenna element. Suchsolution drastically reduces the electronic circuit complexity and los-ses since the sub-THzupconverted electrical signal is generatedonly atthe very end of the path, where an amplifying stage is present inproximity to the antenna element.ResultsDevice design and fabricationFigure 4a shows a schematic of the G-OEM. It consists of a CPW inground-signal-ground (GSG) configuration, operating in the sub-THzrange. The signal electrode (S) width is 74μm and the gap between SSub-THz signalSub-THz antennaG-OEMAmplifying electronicsElectrical IFOp�cal LOb)Op�cal fiber (LO)Electrical baseband informa�on(IF)Op�cal LO distribu�on Electrical Rou�ng (IF)a)Fig. 3 | Proposed optoelectronic antenna array system. a The antenna is fed byan optical local oscillator and by a baseband electrical signal. A dual-wavelengthoptical LO is distributed inside the chip toward each element by means of anoptical distribution layer. The IF signal is distributed as well towards each elementusing low frequency (<10Ghz bandwidth) electrical transmission lines. No sub-THzelectrical signals or sub-THz LOs are present in the antenna. b is a zoom of oneantenna element: each element is solely constituted by a low Intermediate Fre-quency (IF) port, and an optical port. The sub-THz signal is generated just beforebeing transmitted through the wireless link.Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 4and the ground metals (G) is 17.5μm. This geometry provides a 50Ωcharacteristic impedance to match the baseband electrical signalgenerator and the antenna impedance (see Supplementary Informa-tion Section I.B), with negligible ohmic losses. Standard SOI technol-ogy is based on Si substrateswith low resistivity, typically lower than 15Ωcm. This is detrimental for RF performances, since ohmic losses insuch substrates are non-negligible60–62. RF losses can be mitigatedusing Si substrates with high resistivity >10 KΩ cm61. An alternativesolution is the use of a sufficiently thick oxide to avoid interactionbetween the electromagneticRF propagatingmodewith the substrate.Our SiN technology is based on a 15μm thick box oxide, allowing todecouple the RF mode from the Si substrate, thus minimizing ohmiclosses. The electrical CPW embeds at the center of the signal electrodean hBN-encapsulated single-crystal graphene layer, transferred on topof a SiN photonic waveguide. Figure 4b, c shows a schematic detail ofthe active area of the device (respectively, top view and section view),indicated with a dashed circle in Fig. 4a. The graphene bottom layer(graphene channel) dimensions are L = 4μm andW = 50μm. A secondgraphene layer acts as electrostatic gate on top of the hBN-encapsulated graphene and is contacted with a metallic electrode(indicated as “Gate electrode” in Fig. 4a, b). The thickness of the hBNseparating the graphene gate and the graphene channel is ~30 nm. Thegraphene gate is used to set the operating point of the device: thechemical potential μc of the bottom graphene layer must be set farfrom the Charge Neutrality Point (CNP), i.e. μc ~ 150meV for optimaloperation, since at such electrostatic doping the photobolometriceffect, and so the phoresponsivity, is maximized43 (see also Supple-mentary Information, Section I.D-E). The hBN/graphene/hBN/gra-phene stack is placed on top of a quasi-transverse electric (quasi-TE)SiN photonic waveguide of 1500nm width (WGw) and 270 nm thick-ness (WGt), allowing single mode operation at 1550 nm optical wave-length. It is worth noting that Si waveguides can be used as a validalternative to SiN waveguides to perform OEM using graphene.Nevertheless, silicon waveguides suffer from two-photon-absorptionat relatively low-power density. In the proposed perspective (optoe-lectronic antenna array) a relatively high input power could be coupledat the input of the chip, beforebeing routed and split toward the singleelement. The SiN technology supports high optical power densitieswithout suffering from two-photon absorption63. Moreover, the rout-ing of the local oscillator toward each antenna could require a rela-tively long path. SiN waveguides offer ultra-low propagation losscompared to Silicon waveguides63. Another advantage of SiN tech-nology is the lower thermo-optic coefficient at telecom wavelengths,making SiN photonic circuits more stable against temperature64. Thedevice comprising the graphene stack and the optical waveguide(“device section” in Fig. 4d) has the optical input from one side of thephotonic chip, via edge coupling. An adiabatic inversed taper(“inversed taper” in Fig. 4d) has been designed to access the photonicwaveguide through butt-coupling using a lensed single mode fiberwith mode field diameter ~3μm (see Supplementary Information,Inversed taper sec�onStraight sec�onDevice sec�on (50 nm cladding)400 nm to 1500 nm WG width transi�onxyzWGW1500 nmWGt270 nmWGt270 nmWGW400 nmMetal contactsGraphene topGraphene bo�omLensed fiber INd)tcladdthBNWGwWGtSiNwaveguideGraphene channel (bo�om layer)Graphene gate (top layer) hBNSilicaMetalAc�ve area Cross sec�onc)Graphene channel(bo�om layer)LWb)Op�cal waveguideGraphene gate electrodeAc�ve area Top viewGraphene gate (top layer)a)Fig. 4 | G-OEM design. a Schematic representation of the graphene-integratedoptoelectronic upconverter. The metals constituting the electrical CPW are indi-cated in yellow. A graphene layer is embedded in the middle of the Signal (S)electrode of the coplanar waveguide (CPW) and is positioned on top of a SiNphotonic waveguide, indicated in blue. A second graphene layer (black) acting as atop-gate is present on top of the first layer, for electrostatic doping adjustment. Abaseband electrical signal at an Intermediate Frequency (IF) is fed to the device. Anoptical LocalOscillator (LO) is coupled to theoptoelectronicupconverter bymeansof the photonic waveguide. The resulting upconverted Radio Frequency (RF)electrical signal is present at the output of the device. b Top and c section view ofthe active region, indicated with a dashed circle in a, comprising the photonic SiNwaveguide, a graphene channel encapsulated with hexagonal Boron Nitride (hBN),and the graphene gate. L is the channel length, W is the channel width. WGt is theoptical waveguide thickness, WGw is the optical waveguide width, tcladd is thethickness of the waveguide cladding and thBN is the thickness of the (hBN).d showsthe top view of the entire structure, comprising the graphene device on top of theopticalwaveguidewith thin cladding, an inversed taperwaveguide section allowingaccess to the photonic waveguide through butt-coupling using a lensed single-mode fiber, and a straight waveguide section with thick cladding to route theoptical signal from the tapered section to the device section. For each one of thethree sections, the simulated optical mode is shown.Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 5Section I.C). The adiabatic taper gives access to a straight waveguidesection (“Straight section” in Fig. 4d). The taper and the waveguidestraight section are covered by a silica top cladding, with thickness of~900 nm. The cladding is then selectively thinned down to 50nm ontop of the waveguide in the region where the hBN/graphene/hBN/graphene stack is transferred (“device section” in Fig. 4d). In this way,the optical waveguide TE mode is evanescently coupled to the gra-phene layer placed on top. The obtained effective absorption in thegraphene active layer is ~0.057 dB/μm (see Supplementary Informa-tion, Section I.C). We used graphene as top gate to reduce the opticalpower losses compared to a metallic gate. From simulations, we esti-mated <3 dB total optical power lossdue to absorption in the graphenegate andmetal contacts (see Supplementary Information, Section I.C).Figure 5a shows an optical microscopy image of the device. Thefull fabrication flow is presented in detail in Supplementary Fig. 11 anddescribed in the Methods section. We use a van der Waals ‘pick-and-flip’ technique51 to expose a CVD-grown graphene crystal (previouslytransferred from Cu to SiO2/Si36,52) on top of a ~30 nm thick hBN flake(mechanically exfoliated frombulk crystals). After depositingNi/Au (7/60 nm) top contacts by thermal evaporation37, we place a secondgraphene/hBN stack (obtained via the same pick-and-flip procedure)onto the first one for both full encapsulation of the device channel andelectrical isolation of the top graphene layer. Relatively thick hBN(>15 nm according to ref. 35) is known to crucially screen the rough-ness of the underlying substrate, resulting in enhanced carriermobilityin graphene65, with optimal results upon full encapsulation in dryconditions66.Halfway through the assembly, we use standard scanning micro-Raman spectroscopy to evaluate the quality of the hBN-supportedgraphene channel. Figure 5b shows the Raman spectrum from a spatialmapping over the active area (50 × 4μm2). We observe the typicalspectral features of high-quality single-layer graphene67,68, such as anarrow Lorentzian 2D peak and large intensity ratio between the 2Dand G peaks, in addition to a sharp E2g peak from the hBN flake. As astandard indicator for the quality of graphene/hBN stack69, we analyzethe full width at half maximum (FWHM) of the 2D peak, obtaining astatistical distribution peaked at 19.5 cm−1 (see Fig. 5b inset). Thedecisive improvement over the response of nominally identical gra-phene crystals placed on conventional substrates (~23 cm−1 on SiO2/Siin ref. 37) points toward a reduction of strain fluctuations70, which isknown to crucially enhance the carrier mobility71,72.We measured the two-terminal resistance (R) of the sample as afunction of the top-gate potential (Vtg), at roomtemperature and in air.Figure 5c shows the typical ambipolar behavior of graphene field-effect transistors, with a CNP at Vtg ~ 0.5 V. We estimate low dopinglevel of n ~ 3 × 1011cm−2 at Vtg = 0V, (assuming 30nm hBN dielectricwith out-of-plane relative permittivity εr = 3). Using the standard fittingprocedure from ref. 73 (light blue lines in Fig. 5c)we obtain an electron(hole) mobility of 27,000 cm2 V−1 s−1 (23,000 cm2 V−1 s−1) and a contactresistance of 1.7 kΩμm (1.2 kΩμm). The clear asymmetry of the field-effect curve indicates a contribution from gate-dependent contactresistance74,75, which likely affects the mobility estimate (proper four-probe measurements are however not possible in the current devicedesign). Finally, the magnitude of charge inhomogeneity in the CNPregion, estimated via the procedure in Fig. 5c inset71, falls well belowthe 1011cm−2 range, further confirming the expected high devicequality71. Supplementary Fig. 12 described in “Methods” shows that theachievement of these results is promoted by post-fabrication thermalannealing of the device in inert atmosphere.Graphene OEM Operating principle and characterizationThe operating principle of the G-OEM relies on the fundamental phy-sical mechanism of hot carrier generation in a graphene layer coupledto an optical field. The chemical potential (μc) is set far from the CNP(μc > 150–200meV, depending on charge carrier mobility43). At thiselectrostatic doping, the dominant effect allowing photodetection iscarrier heating, which translates into a decrease of the carriermobility,i.e., into a decrease of graphene electrical conductivity42,43. The con-ductivity change can be expressed as:Δσ = σdark + σlight ð1ÞWhere σlight is the conductivity under illumination, while σdark isthe conductivity without optical excitation (dark conditions). TheseμFig. 5 | Graphene characterization. aOptical microscopy image of the fabricateddevice active region. The top and bottom hBN flakes are indicated by the con-tinuous and dotted green lines, respectively. The graphene gate and the graphenechannel are indicated by the black and red lines, respectively. b RepresentativeRaman spectrum of the graphene channel (black line, acquired after the first pick-and-flip process). The light blue line is a Lorentzian fit to the 2D peak. Inset: sta-tistical distribution of FWHM(2D) over the device channel (200 spectra in total).cResistance of the device (black line) as a function of the top-gate voltage (bottomaxis; the top axis shows the corresponding carrier density). The measurement isperformed by applying a constant dc voltage bias of 1mV between the source anddrain contacts, while measuring the current in two-terminal configuration. Thelight blue lines are fitted to the resistance to extract the carrier mobility andcontact resistance. Inset: log–log plot of the conductance G as a function of thecarrier density, used to estimate the carrier inhomogeneity in the Charge Neu-trality Point (CNP) region (~8 × 1010cm−2).Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 6two are calculated from the Drude formula of76,77:σ Te,μc� �=D μc,Te� �π Γ μc,Te� �� � ð2ÞWhereD is the Drudeweight and Γ is the transport scattering rate.Both quantities are dependent on the electron’s temperature and onthe chemical potential. These lasts change under optical excitation,and can be calculated from the heat equation formula:�∇ � Lσ Te� �Te∇Te� �+CeτcoolTe � Tph� �=P x, yð Þ ð3Þcoupled to the charge carrier’s conservation formula:2πKBT� �2_vF� �2 Li2 �e�μckBT� �� Li2 �eμckBT� �h i=CoxVGSe� αμcCoxe2ð4ÞIn Eq. (3) L is the Lorenz number, σ(Te) is the electrical con-ductivity at temperature Te, Ce is the heat capacity Tph is the phonon’stemperature and P(x, y) is the optical power density at a specificcoordinate of the graphene channel (see Supplementary Information,Section I.D for details). In Eq. 4, vF is the Fermi velocity, Li2 is thedilogarithm function, VGS the top-gate voltage, α is 1 or 2 for, respec-tively, metallic or graphene gates(in our case, α = 2), and Cox = ϵϵhBNthBNisthe geometrical capacitance, which depends on the hBN dielectricconstant εhBN and on its thickness thBN (see Supplementary Informa-tion, Section I.D for details). The photogenerated carriers’ coolingdynamics are very fast in graphene, with measured relaxation times of~2 ps45. This means that graphene conductivity can be modulated atvery high frequencies (up to ~500GHz) by means of an optical field.Therefore, to obtain optoelectronic mixing two optical wavelengthsseparated by the desired frequency (LO frequency, fLO) are coupled tothe graphene channel (“LO” in Fig. 4a). If this frequency difference (fLO)falls in the sub-THz range, the optical LO induces a time-varying con-ductivity δσ sin 2πf LOt� �of the graphene channel in the sub-THz range.Thus, if an electrical sinusoid fVin sin 2πf elet� �(“IF” in Fig. 4a) is appliedto the portion of the CPW exhibiting time-varying conductivity (i.e.,where the graphene channel is present), this sinusoid is modulated atthe frequency of the LO, i.e., it is upconverted. As detailed in Supple-mentary Information, Section I.A, this can be seen from the resultingvoltage across the device, which contains the terms:fVin sin 2πf elet� �δσ sin 2πf LOt� �=fVinδσ2cos 2π f ele � f LO�� ��t� �+ cos 2π f ele + f LO�� ��t� � ð5ÞIn particular, the second termon the right hand of Eq. (5) containsthe sum of the two frequencies, corresponding to the frequencyupconversion process. The upconversion efficiency Pconv is then usedas figure of merit to evaluate the device performance. It is defined asthe ratio between the output power of the upconverted tone atfUP = fele + f_LO and the power of the input electrical signal at frequencyfele. Equation 5 comprises the particular case in which the photo-bolometer is used in its more conventional way, i.e., as a photo-detector. This corresponds in Eq.5 in substituting the time-varyingelectrical signal with a DC bias. Since graphene photo-bolometersoperated as detectors need a DC bias, this produces a large dark cur-rent in the mA range78. Instead, we stress that in our case, i.e, when agraphene photo-bolometer is operated as OEM, no DC bias voltage isneeded across the graphene channel41 (see the detailed equationsin Supplementary Information, Section I.A).The fabricated OEM was first characterized in terms of photo-detection bandwidth, i.e. to determine the maximum LO frequencythat can be generated by the device. As detailed in “Methods”, we usedtwodifferent experimental setups. Thefirst allowed to characterize thedevice response up to 67GHz using a Vector-Network-Analyzer (Key-sight PNA-X 5247B). We then characterized the response in the range92–96GHz using a calibrated commercial electronic downconverter(Mi-Wave 970W-94/387) which downconverts high-frequency elec-trical signals from the 92–96GHz range down to the 4–8GHz range.We used a 44GHz bandwidth electrical Spectrum Analyzer (AnritzuMS2850A) to measure the downconverted electrical signal. The resultis shown in Fig. 6a and reveals a flat frequency response over the entirefrequency window accessible using our measurement system. No roll-off wasmeasuredup to96GHz,meaning that the operating bandwidthof our device is larger than this value.We then characterized the G-OEM as a mm-wave upconverterusing the same aforementioned receiver (Mi-Wave 970W-94/387). Theexperimental setup is detailed in Methods. Figure 6b reports theupconversion of an IF signal that has been mixed with the optical LO.The IF was swept between 1 and 5GHz, and the LO was kept at a fixedfrequency of 91 GHz. The 92–96GHz frequencywindowwas set by theFig. 6 | G-OEM optoelectronic characterization. a Frequency response of theG-OEM operated as photobolometric detector. The curve is obtained using avector-network analyzer (VNA) up to 67GHz, while the range 92–96GHz isexplored by means of an electronic downconverter which downconverts electricalsignals from the 92–96GHz frequency range to the 4–8GHz frequency range. Afterdownconversion, the signal is measured with an electrical spectrum analyzer (ESA)with a bandwidth of 44GHz. The 3-dB bandwidth of the device is >96GHz, since noroll-off is evidenced up to this frequency. b Frequency response of the deviceoperated asOEM, specifically as anupconverter, in the range 92–96GHz. AnopticalLO is used to generate a CF at 91 GHz and is mixed with an electrical sinusoidalsignal (IF)with frequency tuned in the range 1–5GHz. Thus, the upconversion lies inthe 92–96GHz frequency range. The inset of b shows the Upconverted power vs IFelectrical input power, for IF = 3GHz. High linearity is reached for a signal power of1 dB, while for a 9 dBm input signal, a ~3 dB compression is evidenced compared tothe theoretical linear curve.Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 7downconverter working range. The IF input power was 3 dBm, whilethe input optical power was 13 dBm. The measured upconvertedpower has an average of ~−41 dBm corresponding to an upconversionefficiency of Pconv[dB] = Pout[dBm]−Pin[dBm]~−44 dB (see SupplementaryInformation, Section I.E). Finally, the inset of Fig. 6b shows the char-acterization of the linearity of theG-OEMversus the input (IF) electricalpower. Themeasurementwas performedby sweeping the input powerof a 3 GHz sinusoidal wave that was mixed with a 91 GHz, 13 dBmoptical LO. The 1-dB compression point is measured for an input IFpower of 1 dBm. A characterization of the conversion efficiency versusthe optical LO power is also present In Supplementary Information,Section III.Sub-THz wireless linkFigure 7 shows the wireless transmission experimental setup. A dualwavelength laser source was realized to obtain an optical localoscillator. As detailed in “Methods”, the two-phase locked wave-lengths were obtained starting from a single continuous wave (CW)distributed-feedback (DBF) laser source fed into a 40 GHz band-width Mach-Zehnder modulator (MZM) driven by a sinusoidal signalof 45.5 GHz and operated in double sideband—suppressed carrier(DSB-SC) mode79. The resulting optical signal was then filtered andamplified to obtain two-phase locked optical wavelengths, sepa-rated in frequency by 91 GHz. The central carrier and high orderharmonics were suppressed by >30 dB. The optical signal was cou-pled to the integrated graphene OEM through a lensed fiber buttcoupled to the photonic chip. A quadrature phase shift keying(QPSK) baseband signal with CF of 2 GHz was generated startingfrom two pseudo-random binary sequencies (PRBS), generatedusing a 65 GS/s digital-to-analog converter (DAC) (Fujitzu LEIA55–65GSa/s 8-bit DAC), delivering a digital signal with maximumtotal length of 217 samples. The datarate of the baseband signal wasswept in the range 1–4 Gbit/s. The signal was fed to the G-OEM to the“IF” port indicated in Fig. 7, through a GSG RF probe (MPI T110A-GSG100) with bandwidth 110 GHz. The baseband signal wasupconverted by the mixer and collected by a second probe (MPIT110A-GSG100) (“RF” port, in Fig. 7). The probe was connectedthrough a 1-mm connector to a short (~15 cm) RF electrical cable. A1-mm to WR-10 waveguide transition was used to connect cable to aWR-10 electrical bandpass filter (mi-wave 460W-86/94/387) with8 GHz bandwidth and central frequency 90 GHz. The filtered signalwas then amplified with an amplifier (mi-wave 955WF-35) with 35 dBgain and 75–110 GHz bandwidth, and transmitted through a 2mwireless link using an horn lensed antenna (mi-wave 258W) with30 dB gain and 4 GHz bandwidth, with central frequency 94 GHz.The receiver was composed by a second antenna (mi-wave 258W),connected to a commercial electronic downconverter (mi-wave970W-94) allowing frequency downconversion from the 92–96 GHz(RF) range to the 4–8 GHz (IF) range. A real-time oscilloscope (Agi-lent Infinium VSA80000A) with electrical bandwidth of 12 GHz wasused to acquire the downconverted signal and visualize the receivedQPSK data stream constellation using a built-in software. A pictureof the setup is shown in the Supplementary Information, Section II.To optimize the Error-Vector-Magnitude (EVM) of the receivedsignal we adjusted the working point of the G-OEM by acting on thegate voltage (VG) of the device. We used a LO power of ~13.3 dBm andan input baseband electrical power of ~9 dBm. Then, we transmitted a1 Gbit/s QPSK baseband signal through the wireless link. As shown inFig. 8a, a minimum EVM of 22% was found for VG ~ −1.8 V, corre-sponding to ~2.3 V from the CNP voltage (Vtg ~ 0.5 V), i.e., µc~130meV.Figure 8b, c shows the EVMas a functionof theoptical LOpower andofthe input baseband electrical power for VG ~ −1.8 V.We then tested the EVM as a function of the datarate of theinput baseband QPSK signal. Figure 9 shows the received con-stellation, together with the extrapolated eye diagram of one of thetwo quadratures (Q quadrature). The EVM for 1 Gbit/s is 22%. Thisvalue increases up to 24% for the 2Gbit/s data stream and to 27% forthe 4 Gbit/s signal. We estimated80 a bit error rate of 4 × 10−6,1.3 × 10−5 and 1.3 × 10−4 at 1 Gbit/s, 2 Gibit/s and 4 Gbit/s datarate,respectively.PLO~13dBm X APRF~-40dBmPOUT ~-5dBmQPSK generator 10 1100 01Δλ~0.73nmLO RFIF92-96 GHz downconverter~ 7 dBm2 m Wireless link2GHzcarrierBASEBAND93GHz carrierSUB-THz5GHz carrierBASEBANDG-OEMPIF ~8dBmLensed fiberRF probeIFLORFRF probeG-OEMFig. 7 | Wireless link experimental setup. The first input is a dual wavelengthoptical signal, indicated as LO. A baseband quadrature phase shift keying (QPSK)datastream (IF) with CF 2GHz is generated using a Digital-to-Analog Converter(DAC). Thedata rate of the signal hasbeen tunedbetween 1 Gbit/s and 4Gbit/s. Theoptical and electrical inputs signals are applied to the G-OEM, as indicated in theoptical image in the inset: the LO is coupled to the chip using a lensedfiber. The LOfrequency is 91 GHz The IF is applied using an RF probe. The output port (RF)provides the sub-THz CF carrying the upconverted information. Since the LO fre-quency is 91 GHz and the IF carrier is 2 GHz, the upconverted signal is carriedaround a carrier frequency of 91 GHz+ 2GHz= 93GHz (upper sideband). Thiselectrical signal is collected by a second RF probe. After being filtered and ampli-fied, it is transmitted through a wireless link using a horn antenna, and then isdetected by a receiver composed by a second antenna and a commercial down-converter. A real-time oscilloscope is finally used to analyze the received QPSKsequence. On the bottom, the schematic electrical spectrum of the datastream isdepicted, as it flows along the various elements of the system: the baseband signalis first upconverted by the G-OEM in the sub-THz domain, and then down-converted again in baseband after the wireless link, to be visualized on the oscil-loscope. At each Input/Output stage of thewireless link chain, the electrical (black)or optical (red) maximum powers used during the experiment are indicated.Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 8DiscussionThe reason for the EVM deterioration as a function of datarate shouldbe attributed to the complexity of the experimental setup that affectsthe signal-to-noise-ratio (SNR) of the wireless link as bandwidthincreases. First, the electrical bandwidth of the full transmission chain,resulting from the electrical bandwidth of all the used components(transmissionfilter + antenna), is 2 GHz (with center frequency93GHz)which determines the maximum electrical bandwidth of the inputsignal (max. data rate of 4Gbit/s). Second, the electrical signal gener-ated by the DAC exhibits a degradation in terms of SNR whileincreasing the data rate, see Supplementary Information, Section IV.Consequently, the link performance is most likely setup limited. Theuse of antennas with higher gain would further boost these results,increasing the SNR and thus the EVM of the received signal. We usedantennas with ~30dB gain, but antennas performing up to 48 dB gainat 100GHz are commercially available (https://www.miwv.com/,https://www.eravant.com/48-dbi-gain-88-to-100-ghz-12-diameter-0-094-dia-circular-wg-w-band-gaussian-optics-antenna). This wouldsubstantially improve the performance of our wireless link, enablinglonger link distances. In general, sub-THz antennas can be made moreefficient bymaking use of on axis and off-axis elements along the pathof the link81–83. Moreover, the performances of complex telecom radiolinks such asmassiveMIMOcan be further optimized by implementingadvanced techniques using, e.g., reconfigurable reflecting/-2.5 -2 -1.5 -1Gate voltage (V)2025303540MVE)%(10 11 12 13 14Optical LO power (dBm)222426283032MVE)%(0 5 10Electrical input power (dBm)222426283032MVE)%(Fig. 8 | Error-vector-magnitude (EVM) characterization. a EVM as a function ofthe gate voltage applied to the G-OEM. We get an optimal working point at ~−1.8 V.The corresponding measured constellation is shown in the inset. b EVM as afunction of the optical LO power. c EVM as a function of the applied electricalbaseband power. b–c curves are obtained for VG = −1.8 V, with a 1 Gbit/s input IFdatastream.Fig. 9 | Data transmission results.Constellation diagramof the QPSKmodulationafter transmission through the sub-THz wireless link. For each datarate (from 1 to4Gbit/s) the eye diagram of the Q quadrature of the complex signal is shown, withthe corresponding EVM. The maximum measured datarate is limited by the elec-trical bandwidth of the electronic components constituting thewireless link, that isfixed to 2GHz, with central (carrier) frequency of 93GHz.Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 9https://www.miwv.com/https://www.eravant.com/48-dbi-gain-88-to-100-ghz-12-diameter-0-094-dia-circular-wg-w-band-gaussian-optics-antennahttps://www.eravant.com/48-dbi-gain-88-to-100-ghz-12-diameter-0-094-dia-circular-wg-w-band-gaussian-optics-antennatransmitting surfaces84, and employing forward-error correctiontechniques (which have not been used during our experiment).The intrinsic performances of the photonic upconverter are dis-cussed in Supplementary Information, Section I.E, where the linkbetween the upconversion power efficiency and the mobility of gra-phene is analyzed. Our simulations show that high mobility(27,000 cm2 V−1 s−1) is beneficial compared to low mobility(<10,000 cm2 V−1 s−1), in agreement with experimental results presentin literature41,42. We then show that a reduction of contact resistancedown to 500Ωµm can allow a performance boost. The resulting totalupconversion efficiency improvement can be at least ~11 dB, meaningthat the graphene photonic upconverter can attain at least ~−33 dBconversion efficiency with small design efforts. This performanceboost, together with the possibility to operate at frequencies of500GHz46, would allow to reach and surpass state-of-the-artperformances5,6,11,12,18. Indeed, the sole implementation of the dis-cussed two improvements (use of 48dB gain antennas and optimizedmixing conversion efficiency up to −33 dB) would allow to increase thepower at the receiver by >40dB in the 2m wireless link, which woulddrastically improve the SNR allowing higher modulation formats (i.e.,higher channel capacity), or equivalently enable >100mwireless links,keeping the same performances reported in the present work15 interms of datarate.We now briefly quantitatively discuss the proposed photonic-aided antenna arraypresented in Fig. 3. The current electronic antennaarray transmitters operating at ~100GHz carrier frequency require apower consumption in the range 1–10W and deliver an output powerin the range 10–15 dbm85–87. Each antenna element based on our pro-posed scheme contains a G-OEM providing ~−40dBm. To attain totaloutput power of 15 dBm, a 30 dB amplifying stage is needed for eachelement. Each amplifier should require few tens of mW power to keepthe power consumption within some W. Available electronics88 allows~20mWpower consumption for each element, to obtain ~30 dB poweramplification. Concerning the optoelectronic mixer, it should requirenegligible power compared to the amplifying stage (i.e., <<10mW forIF and LO signals). The G-OEM power consumption can be improvedacting on the mobility of the graphene detector. This is quantitativelydiscussed in Supplementary Information, Section I.E. Also, high pho-toresponsivity enhancement can be achieved using plasmonic detec-tors and/or stacking more than one graphene layer. ref. 89 reported0.5 A/W external responsivity with <0.6 V applied to the detectorcontacts, and compact size (<6μm2 active region). Thesedevices couldbe advantageously used in antenna arrays. Indeed, the detector pre-sented in89 provides a responsivity allowing comparable performancesas the one presented in our work in terms of conversion efficiency, butthanks to plasmonic enhancement it requires only an optical LO inputpower <0 dBm and an IF input power ~0 dBm if used as G-OEM, thusmatching the power consumption requirements of the single antennaelement. Finally, an antenna array with such characteristics shouldpreserve high gain to keep the same performances as the one pre-sented in our work, in terms of link distance and data rate. At thisregard, 40dB gain have been already demonstrated at carrier fre-quencies >100GHz for 16 × 16 antenna arrays90.Finally, let us comment on the perspective for scalability of theproposed G-OEM. We stress that the CVD graphene employed aschannel and gatematerial is obtained via growth and transfermethodswith established wafer-scale capabilities37. The carrier mobility neces-sary for effective G-OEM operation (>104 cm2 V−1 s−1 at room tempera-ture), crucially requires decoupling of the graphene channel fromextrinsic scattering sources91. As demonstrated both for exfoliatedgraphene flakes65,66, and CVD-grown graphene single crystals92,93,screening of substrate roughness is paramount. The only reliablestrategy currently available to achieve this requires the use of thickhBN35. Although large-scale growth of single or few-layer single-crys-talline hBN is well established, the few-layer thickness limit does notguarantee adequate screening, resulting in carrier mobility in line tothat of graphene on SiO2 (<104 cm2 V−1 s−1)94. The synthesis of scalablecounterparts of thick hBN flakes (i.e., providing an adequate screeningeffect) is a sensitive topic. Relevant advances in crystals dimensionscan be found in ref. 95, where a 2 × 5 cm2 trilayer hBN crystals growthhas been demonstrated. Similarly, the growth of large area (5 cm lat-eral dimension) thick (5 nm) hBN have been shown in96, with graphenemobility values approaching ~104 cm2 V−1 s−1. We showed that the pre-sented G-OEM fabrication protocol is compatible with CVD-grownmaterials (in this case, graphene itself), making it ready for the inte-gration of large-scale thick hBN films.In summary, we have reported a wireless link based on graphene,using a hybrid electronic/photonic approach based on an integrateddevice, the G-OEM. This transmitter targets radio applications in thesub-THz band to improve performance parameters such as frequencyaccuracy, PN, upconversion efficiency, and to envisage an integratedsolution that fits in the sub-THz antenna element footprint. The reali-zation of sub-THz radio links enables telecom operators to achievenetwork densification while optimizing overall network capacity andlatency performance (e.g., deploy more links per square km with smalland picocells or new cell-free network topologies28). Approaching100GHz CF, electronic technology is showing its limits, while thecommon photonic implementation is still far from allowing the reali-zation of single antenna elements. The present work assesses the meritof the realization of an integrated optoelectronic sub-THz upconvertersuitable for single or multiple antenna systems, opening new perspec-tives for the use of integrated photonics in next-generation antennaarrays in 5G and 6G NR, potentially solving the issues related to theconcurrent technologies. We show setup limited record frequencyoperation (93GHz) and datarate (4Gbit/s), using high-quality CVDgraphene allowing high values of upconversion efficiency (~−44dB)compared to state-of-the-art G-OEMs42,43,97. Eventually, we describe thepathways towards a substantial increment of both datarate and con-version efficiency by acting on the design at both the system and thedevice level. We use our device as a photonic upconverter, neverthelessit can be used as downconverter as well39,41–43, thus allowing the reali-zation of the receiver using the same graphene technology, in view ofthe realization of a G-OEM-based full-link. This work opens the route tocompact, low-power, cost-effective sub-THz antenna array systems fullybased on graphene as active material, with performances potentiallyovercoming state-of-the-art limitations5,6,11,12,18.MethodsDevice fabricationThe complete flow developed for the fabrication of the G-OEM isdepicted in Supplementary Fig. 11, with steps numbered from 1 to 10.Central to our process is a modified version of the pick-and-flipmethod described in ref. 51. This technique consists of picking up hBNand graphene, respectively, with a polymeric stamp, and then depos-iting it facedown (i.e., with the hBN flake on top) on a second stamp, toobtain graphene exposed on top of hBN when the stack is finallyreleased on the target substrate. Two polymer membranes—poly(bi-sphenol A carbonate)polycarbonate (PC) and poly(vinyl alcohol) (PVA)—are deposited on a polydimethylsiloxane (PDMS) stamp, which isattached to a glass slide. With the aid of a micromechanical stage, thisstructure is used to pick-up an hBN flake and then a portion of agraphene crystal from SiO2 (steps 1 and 2). Subsequently, the resultingstack is put in contact with a second underlying PDMS stamp (step 3).To transfer the graphene/hBN from the first to the second stamp, thesolubility of PVA in water is exploited. Indeed, when a drop of water isreleased between the two stamps, the PVA is dissolved, and the firststamp is detached from graphene/hBN (step 4). Subsequently,the second stamp is flipped and aligned to the SiN photonic wave-guide. The hBN/graphene stack is thus deposited, leaving the gra-phene exposed (step 5). The sample is heated on the assembly stage toArticle https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 10improve the adhesion to the substrate (step 6). The graphene is pat-terned by electron beam lithography followed by oxygen plasmaetching into a 50 × 8 µm2 rectangle. Two contacts are fabricated on topof graphene (with 2 µm superposition on each side, resulting in a finalchannel of 50 × 4 µm2) using electron beam lithography followed bythermal evaporation of Ni/Au and lift-off (step 7). To fabricate thegraphene gate, the pick-and-flip technique is used again (steps 1–4),and a second hBN/graphene stack is deposited on the device (steps 8and 9). The top graphene is finally contacted with the gate electrode(step 10). By doing so, the top hBN flake has the double function ofencapsulant for graphene and of top-gate dielectric.Immediately after fabrication, the DC transport properties of theG-OEM are comparable to those of CVD graphene single crystals onSiO2 (dotted black line in Supplementary Fig. 12)37, indicating poorquality of the graphen/hBN interfaces. We attribute this to thesequential stacking method for encapsulation, which traps polymericresiduals from processing the top contacts. We find that annealing thestructure in inert atmosphere results in a dramatic increase of thecarriermobility (continuous black line in Supplementary Fig. 12), likelydue to the aggregation of interface contaminants98.Optoelectronic frequency response measurementThe optoelectronic frequency response of the G-EOM allows to eval-uate of the optical LO frequency detectable by the device, and corre-sponds to the optoelectronic bandwidth of the device when used asphotodetector. The characterization was performed using twoexperimental setups, covering two different frequency windows. Thescheme in Supplementary Fig. 13 shows the first setup, while the twoimages show the elements indicated in the scheme. The setup con-sisted of a Vector-Network-Analyzer (VNA) (Keysight PNA-X 5247B),connected to a MZM, used to modulate a 1.55μm wavelength CWdistributed-feedback (DBF) laser source. The modulating frequencywas swept up to 67GHz. We accounted for the MZM 40GHz electro-optical bandwidth by measuring its response during the VNA calibra-tion, using a 70GHz bandwidth photodiode (Finisar XPD3120R) withknown frequency response. The modulated optical signal was thencoupled to the G-OEM using a tapered optical fiber. The photobolo-metric response characterization requires a DC voltage bias99 appliedto the graphene channel. This was fed using twobias tees connected ateach side of the G-OEM. A DC needle was used to contact the gate padand set the optimal graphene chemical potential μc optimizing theoptoelectronic response. In Fig. 8a we show an optimal response forVG–VCNP = 2.3 V. The G-OEM was contacted with RF GSG probes (MPIT110A-GSG100). One side of the device was AC coupled to ground,while the other side was connected to the second port of the VNA. Theelectrical power resulting from the photocurrent flowing through the50Ω load of the VNA was measured as a function of the electricalfrequency driving theMZM. The result of the characterization is shownin Fig. 6a, and reveals a flat response over the entire frequency range.Because of the frequency limitation of the VNA, we used a differentsetup to characterize the optoelectronic bandwidth of the G-EOM in thefrequency range used for thewireless transmission experiment. First, weimplemented a phase-locked dual-wavelength tunable laser sourceshown in Supplementary Fig. 14. We used one port of the VNA as signalgenerator, which was manually swept in the range f0 = 46–48GHz. Weoperated the MZM in double sideband—suppressed carrier (DSB-SC)mode79, to obtain two optical wavelengths separated by 2f0. To furthersuppress the central wavelength, we used a band-stop narrow fibergrating filter centered at 1.55μm, with optical bandwidth of~0.241 nm(~30GHz). The resulting signal was then amplified and filteredagain with a 1 nm (~125GHz) optical pass-band filter to reduce theamplified spontaneous emission (ASE) power of the optical amplifier.The resulting spectrum of the optical signal obtained with an opticalspectrum analyzer (OSA) is shown in Supplementary Fig. 14. We cali-brated the modulation depth of the MZM as a function of themodulating frequency by measuring the optical power of the two gen-eratedwavelengths on theOSA compared to the central wavelength andhigher order harmonics. We measured suppression of both carrier andhigher-order harmonics of >30dB, as shown in Supplementary Fig. 14.The twowavelengthswere then coupled to theG-OEM tomeasurethe optoelectronic frequency response of the device in the rangearound the wireless link experiment CF. The setup is shown in Sup-plementary Fig. 15. The beating of the two wavelengths on the deviceproduced a photocurrent in the range 2f0 = 92–96GHz. The G-OEMwas then connected to an electronic downconverter (Mi-Wave 970W-94/387), which downconverted the sub-THz signal in the range4–8GHz. The input RF port of the downconverter had a WR-10 wave-guide connection. A 1-mm coaxial to WR-10 transition was used toconnect the RF probes to the electronic component, as shown in theimage of Supplementary Fig. 15. The downconverted electrical signalwas visualized on an electrical spectrum analyzer (ESA) (AnritzuMS2850A). We corrected the amplitude of the measured signalaccounting for the downconverter frequency response, which oscil-lated by ~2 dB in the considered frequency range, and for the loss ofthe RF cable connected between the downconverter and the ESA. Thefrequency response is shown in Fig. 6a (red dots) and does not presentany roll-off in the considered range.Conversion efficiency measurementThe setup used to characterize the G-OEM conversion efficiency isshown in Supplementary Fig. 16. It is very similar to the one used tocharacterize the optoelectronic response in the range 92–96GHz. Thesame dual-wavelength laser source described above and shown inSupplementary Fig. 14 has been used, without sweeping themodulatingfrequency of the MZI. Specifically, we fixed fLO = 91GHz. When thedevice is used as optoelectronicmixer, there is no need to apply anyDCbias voltage along the graphene channel as can be deduced fromequations S4-S10 in the Supplementary Information. Thus, the DC vol-tage supply used for the “Optoelectronic frequency response mea-surement” is essentially substituted by an AC signal generator,representing the IF frequency fed to the mixer. The RF output carryingthe upconverted frequency was then connected to the same electronicdownconverter presented above. The upconverted frequency was inthe range fRF = fIF + fLO = [92 −96]GHz, and so in the range 4–8GHz afterdownconversion. The signal was visualized on a spectrum analyzer.Data availabilityThe Source Data underlying the figures of this study are available athttps://doi.org/10.5281/zenodo.8341503. All raw data generated dur-ing the current study are available from the corresponding authorsupon request.References1. Cerwall, P. et al. EricssonMobility ReportNovember 2021.EricssonMobility Report 1–40 (2021).2. Rahman, I. et al. 5G Evolution Toward 5G advanced: an overviewof 3GPP releases 17 and 18. Ericsson Technol. Rev. 14, 1–13 (2021).3. von Butovitsch, P. et al. Advanced antenna systems for 5G net-works. Ericsson white paper 1–15 (2018).4. Tripathi, S., Sabu, N. v., Gupta, A. K. & Dhillon, H. S. Millimeter-wave and terahertz spectrum for 6G wireless. In: 83–121. https://doi.org/10.1007/978-3-030-72777-2_6 (2021).5. Rappaport, T. S. et al. Wireless communications and applicationsabove 100GHz: opportunities and challenges for 6G and beyond.IEEE Access 7, 78729–78757 (2019).6. Kumar, A., Gupta,M.&Singh, R. Topological integratedcircuits for5G and 6G. Nat. Electron 5, 261–262 (2022).7. Ronkainen, H., Edstam, J., Ericsson, A. & Östberg, C. Integratedaccess and backhaul - a new type of wireless backhaul in 5G.Ericsson Technology Review 1–12 (2020).Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 11https://doi.org/10.5281/zenodo.8341503https://doi.org/10.1007/978-3-030-72777-2_6https://doi.org/10.1007/978-3-030-72777-2_68. Dickmann, J. et al. ‘Automotive radar the key technology forautonomous driving: Fromdetection and ranging to environmentalunderstanding’. In: 2016 IEEE Radar Conference (RadarConf) 1–6.https://doi.org/10.1109/RADAR.2016.7485214 (IEEE, 2016).9. Ghelfi, P. et al. A fully photonics-based coherent radar system.Nature 507, 341 (2014).10. Dolfi, D. et al. Building blocks and concepts for THz remote sen-sing and communications. In: 2019 44th International Conferenceon Infrared, Millimeter, and Terahertz Waves (IRMMW-THz) 1–2.https://doi.org/10.1109/IRMMW-THz.2019.8874248 (IEEE, 2019).11. Sengupta, K., Nagatsuma, T. & Mittleman, D. M. Terahertz inte-grated electronic and hybrid electronic–photonic systems. Nat.Electron 1, 622–635 (2018).12. Nagatsuma, T., Ducournau, G. & Renaud, C. C. Advances in ter-ahertz communications accelerated by photonics. Nat. Photonics10, 371–379 (2016).13. Dan, I. et al. A superheterodyne 300GHzwireless link for ultra-fastterahertz communication systems. Int J. Micro. Wirel. Technol. 12,578–587 (2020).14. Tsai, J.-H., Yang, H.-Y., Huang, T.-W. & Wang, H. A 30–100GHzwideband sub-harmonic activemixer in 90nmCMOS technology.IEEE Microw. Wirel. Compon. Lett. 18, 554–556 (2008).15. Pozar, D. M. Microwave engineering. (John wiley & sons, 2011).16. Levanen, T., Tervo, O., Pajukoski, K., Renfors, M. & Valkama, M.Mobile communications beyond 52.6GHz: waveforms, numerol-ogy, and phase noise challenge. IEEE Wirel. Commun. 28,128–135 (2021).17. Mazor, N. & Socher, E. X-band to W-band frequency multiplier in65 nm CMOS process. IEEE Microw. Wirel. Compon. Lett. 22,424–426 (2012).18. Kang, S. & Thyagarajan, S. v. & Niknejad, A. M. A 240GHz fullyintegrated wideband QPSK transmitter in 65 nm CMOS. IEEE J.Solid-State Circuits 50, 2256–2267 (2015).19. Cao, T. et al. Frequency multiplier-based millimeter-wave vectorsignal transmitter using digital predistortion with constrainedfeedback bandwidth. IEEE Trans. Micro. Theory Tech. 68,1819–1829 (2020).20. Robins, W. P. Phase noise in signal sources: theory and applica-tions. vol. 9 (IET, 1984).21. Rubiola, E. &Boudot, R. Phasenoise inRF andmicrowaveamplifiers.In: 2010 IEEE International Frequency Control Symposium 109–111.https://doi.org/10.1109/FREQ.2010.5556363 (IEEE, 2010).22. Lee, T. H. & Hajimiri, A. Oscillator phase noise: a tutorial. IEEE J.Solid-State Circuits 35, 326–336 (2000).23. 3GPP TS38.300v. 15.9.0 - NR; NR andNG-RANOverall Description;Stage 2 (Release 15). In: 3GPP - 3rd Generation Partnership Project -Technical Specification Group Radio Access Network (2020).24. Liu, G., Trasser, A. & Schumacher, H. A 64–84-GHz PLL With LowPhaseNoise in an80-GHzSiGeHBTTechnology. IEEE Trans.Micro.Theory Tech. 60, 3739–3748 (2012).25. Tetsumoto, T. et al. Optically referenced 300GHz millimetre-wave oscillator. Nat. Photonics 15, 516–522 (2021).26. Zhang, S. et al. Terahertz wave generation using a soliton micro-comb. Opt. Express 27, 35257–35266 (2019).27. Koenig, S. et al. Wireless sub-THz communication system withhigh data rate. Nat. Photonics 7, 977–981 (2013).28. Interdonato, G., Björnson, E., Quoc Ngo, H., Frenger, P. & Larsson,E. G. Ubiquitous cell-free massive MIMO communications. EUR-ASIP J. Wirel. Commun. Netw. 2019, 197 (2019).29. Hindle, P. Comprehensive survey of commercialmmWavephasedarray companies. (2020).30. Ruff, W. C. et al. Self-mixing detector candidates for an FM/cwladar architecture. In: vol. 4035, 4011–4035 (Spie Proceed-ings, 2000).31. Nair, R. R. et al. Fine structure constant defines visual transparencyof graphene. Science 320, 1308–1308 (2008).32. Breusing, M. et al. Ultrafast nonequilibrium carrier dynamics in asingle graphene layer. Phys. Rev. B 83, 153410 (2011).33. Tomadin, A., Brida, D., Cerullo, G., Ferrari, A. C. & Polini, M.Nonequilibrium dynamics of photoexcited electrons in graphene:Collinear scattering, Auger processes, and the impact of screen-ing. Phys. Rev. B 88, 035430 (2013).34. Brida, D. et al. Ultrafast collinear scattering and carrier multi-plication in graphene. Nat. Commun. 4, 1987 (2013).35. Purdie, D. G. et al. Cleaning interfaces in layered materials het-erostructures. Nat. Commun. 9, 5387 (2018).36. Tyagi, A. et al. Ultra-clean high-mobility graphene on technologi-cally relevant substrates. Nanoscale 14, 2167–2176 (2022).37. Giambra, M. A. et al. Wafer-scale integration of graphene-basedphotonic devices. ACS Nano 15, 3171–3187 (2021).38. Mzali, S. et al. Stabilizing a graphene platform toward discretecomponents. Appl Phys. Lett. 109, 253110 (2016).39. Montanaro, A. Study of graphene-based high frequency optoe-lectronic devices for optoelectronic mixing. (Université Paris sci-ences et lettres, 2019).40. Mao, X. et al. Optoelectronic mixer based on graphene FET. IEEEElectron Device Lett. 36, 253–255 (2015).41. Montanaro, A. et al. Thirty gigahertz optoelectronic mixing inchemical vapor deposited graphene. Nano Lett. 16,2988–2993 (2016).42. Hamidouche, L. et al. Optoelectronic mixing in high-mobilitygraphene. ACS Photonics 8, 369–375 (2021).43. Montanaro, A. et al. Optoelectronic mixing with high-frequencygraphene transistors. Nat. Commun. 12, 2728 (2021).44. Xia, F., Mueller, T., Lin, Y., Valdes-Garcia, A. & Avouris, P. Ultrafastgraphene photodetector. Nat. Nanotechnol. 4, 839 (2009).45. Urich, A., Unterrainer, K. & Mueller, T. Intrinsic response time ofgraphene photodetectors. Nano Lett. 11, 2804–2808 (2011).46. Koepfli, S. M. et al. Metamaterial graphene photodetector withbandwidth exceeding 500 gigahertz. Science 380, 1169–1174(2023).47. Marconi, S. et al. Photo thermal effect graphene detector featur-ing 105Gbit s−1 NRZ and 120Gbit s−1 PAM4 direct detection. Nat.Commun. 12, 806 (2021).48. Li, X. et al. 1-Tb/s millimeter-wave signal wireless delivery atD-band. J. Light. Technol. 37, 196–204 (2019).49. Jia, S. et al. 2 × 300Gbit/s line rate PS-64QAM-OFDMTHzphotonic-wireless transmission. J. Light. Technol. 38, 4715–4721 (2020).50. Horst, Y. et al. Transparent optical-THz-optical link at 240/192Gbit/s over 5/115 m enabled by plasmonics. J. Light. Technol.40, 1690–1697 (2022).51. Wong, D. et al. Cascade of electronic transitions in magic-angletwisted bilayer graphene. Nature 582, 198–202 (2020).52. Miseikis, V. et al. Deterministic patterned growth of high-mobilitylarge-crystal graphene: a path towards wafer scale integration. 2dMater. 4, 21004 (2017).53. Gavell, M., Ferndahl, M., Zirath, H., Urteaga, M. & Pierson, R. AFundamental upconverting gilbert mixer for 100GHz wirelessapplications. In: 2010 IEEE Compound Semiconductor IntegratedCircuit Symposium (CSICS) 1–4 https://doi.org/10.1109/CSICS.2010.5619615 (IEEE, 2010).54. Muller, A., Thiel, M., Irion, H. & Ruob, H.-O. A 122GHz SiGe activesubharmonic mixer. In: European Gallium Arsenide and OtherSemiconductor Application Symposium, GAAS 200557–60 (2005).55. Cao, Z. et al. Advanced Integration Techniques on BroadbandMillimeter-Wave Beam Steering for 5G Wireless Networks andBeyond. IEEE J. Quantum Electron 52, 1–20 (2016).Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 12https://doi.org/10.1109/RADAR.2016.7485214https://doi.org/10.1109/IRMMW-THz.2019.8874248https://doi.org/10.1109/FREQ.2010.5556363https://doi.org/10.1109/CSICS.2010.5619615https://doi.org/10.1109/CSICS.2010.561961556. D’Amato, G., Avitabile, G., Coviello, G. & Talarico, C. DDS-PLLphase shifter architectures for phased arrays: theory and techni-ques. IEEE Access 7, 19461–19470 (2019).57. Natarajan, A., Komijani, A. & Hajimiri, A. A fully integrated 24-GHzphased-array transmitter in CMOS. IEEE J. Solid-State Circuits 40,2502–2514 (2005).58. Porzi, C. et al. Photonic integrated microwave phase shifter up tothe mm-wave band with fast response time in silicon-on-insulatortechnology. J. Light. Technol. 36, 4494–4500 (2018).59. Natarajan, A., Komijani, A., Guan, X., Babakhani, A. & Hajimiri, A. A77-GHz phased-array transceiver with on-chip antennas in silicon:transmitter and local LO-path phase shifting. IEEE J. Solid-StateCircuits 41, 2807–2819 (2006).60. Reyes, A.C., El-Ghazaly, S.M., Dorn, S., Dydyk,M.&Schroder, D. K.Silicon as amicrowave substrate. In: 1994 IEEEMTT-S InternationalMicrowave Symposium Digest (Cat. No. 94CH3389-4) 1759–1762.https://doi.org/10.1109/MWSYM.1994.335101 (IEEE).61. Makita, T., Tamai, I. & Seki, S. Coplanar waveguides on high-resistivity silicon substrateswith attenuation constant lower than 1dB/mm for microwave and millimeter-wave Bands. IEEE Trans.Electron Devices 58, 709–715 (2011).62. Reyes, A. C. et al. Coplanar waveguides andmicrowave inductorson silicon substrates. IEEE Trans. Micro. Theory Tech. 43,2016–2022 (1995).63. Siew, S. Y. et al. Review of silicon photonics technology andplatform development. J. Light. Technol. 39, 4374–4389 (2021).64. Rahim, A. et al. Expanding the silicon photonics portfolio withsilicon nitride photonic integrated circuits. J. Light. Technol. 35,639–649 (2017).65. Dean, C. R. et al. Boron nitride substrates for high-quality gra-phene electronics. Nat. Nanotechnol. 5, 722–726 (2010).66. Wang, L. et al. One-dimensional electrical contact to a two-dimensional material. Science 342, 614–617 (2013).67. Ferrari, A. C. & Basko, D.M. Raman spectroscopy as a versatile toolfor studying the properties of graphene. Nat. Nanotechnol. 8,235–246 (2013).68. Ferrari, A. C. et al. Raman spectrum of graphene and graphenelayers. Phys. Rev. Lett. 97, 187401 (2006).69. Banszerus, L. et al. Identifying suitable substrates for high-qualitygraphene-based heterostructures. 2d Mater. 4, 025030 (2017).70. Neumann, C. et al. Raman spectroscopy as probe of nanometre-scale strain variations in graphene. Nat. Commun. 6, 8429 (2015).71. Couto, N. J. G. et al. Random strain fluctuations as dominant dis-order source for high-quality on-substrate graphene devices.Phys. Rev. X 4, 041019 (2014).72. Wang, L. et al. Mobility enhancement in graphene by in situreduction of random strain fluctuations. Phys. Rev. Lett. 124,157701 (2020).73. Kim, S. et al. Realization of a high mobility dual-gated graphenefield-effect transistor with Al2O3 dielectric. Appl Phys. Lett. 94,062107 (2009).74. Berdebes, D., Low, T., Sui, Y., Appenzeller, J. & Lundstrom, M. S.Substrate gating of contact resistance in graphene transistors.IEEE Trans. Electron Devices 58, 3925–3932 (2011).75. Wilmart, Q. et al. Contact gating atGHz frequency in graphene.Sci. Rep. 6, 21085 (2016).76. Frenzel, A. J., Lui, C. H., Shin, Y. C., Kong, J. & Gedik, N.Semiconducting-to-metallic photoconductivity crossover andtemperature-dependent drude weight in graphene. Phys. Rev.Lett. 113, 056602 (2014).77. Massicotte, M., Soavi, G., Principi, A. & Tielrooij, K.-J. Hot carriers ingraphene – fundamentals and applications. Nanoscale 13,8376–8411 (2021).78. Schall, D. et al. 50Gbit/s photodetectors based on wafer-scalegraphene for integrated siliconphotonic communication systems.ACS Photonics 1, 781–784 (2014).79. Inagaki, K., Kawanishi, T. & Izutsu, M. Optoelectronic frequencyresponse measurement of photodiodes by using high-extinctionratio optical modulator. IEICE Electron. Express 9, 220–226 (2012).80. Schmogrow, R. et al. Error vector magnitude as a performancemeasure for advanced modulation formats. IEEE Photonics Tech-nol. Lett. 24, 61–63 (2012).81. Nellen, S. et al. Experimental comparison of UTC- and PIN-photodiodes for continuous-wave terahertz generation. J. InfraredMilli. Terahz. Waves 41, 343–354 (2020).82. Van Rudd, J. &Mittleman, D. M. Influence of substrate-lens designin terahertz time-domain spectroscopy. J. Opt. Soc. Am. B 19,319 (2002).83. Smith, J., Naftaly, M., Nellen, S. & Globisch, B. Beam profilecharacterisation of an optoelectronic silicon lens-integrated PIN-PD emitter between 100GHz and 1 THz. Appl. Sci. 11, 465 (2021).84. Jamali, V., Tulino, A. M., Fischer, G., Muller, R. R. & Schober, R.Intelligent surface-aided transmitter architectures for millimeter-wave ultra massive MIMO systems. IEEE Open J. Commun. Soc. 2,144–167 (2021).85. del Rio, D. et al. A D-band 16-element phased-array transceiver in55-nm BiCMOS. IEEE Trans. Micro. Theory Tech. 71, 854–869(2023).86. Li, H. et al. W-band scalable 2 × 2 phased-array transmitter andreceiver chipsets in SiGe BiCMOS for high data-rate communica-tion. IEEE J. Solid-State Circuits 57, 2685–2701 (2022).87. Shahramian, S., Holyoak, M. J., Singh, A. & Baeyens, Y. A fullyintegrated384-element, 16-tile, $W$ -bandphased arraywith self-alignment and self-test. IEEE J. Solid-State Circuits 54, 2419–2434(2019).88. Chikkanagouda, R. & Cyril Prasanna, R. P. A Low-noise Front-endAmplifier With 15 dB gain operating at 100–120GHz in 65 nmCMOS technique. In: 2019 International Conference on Commu-nication and Electronics Systems (ICCES) 981–985. https://doi.org/10.1109/ICCES45898.2019.9002541 (IEEE, 2019).89. Ma, P. et al. Plasmonically enhanced graphene photodetectorfeaturing 100Gbit/s data reception, high responsivity, and com-pact size. ACS Photonics 6, 154–161 (2019).90. Gomez-Torrent, A. et al. A 38 dB gain, low-loss, flat array antennafor 320–400GHz enabled by silicon-on-insulator micromachin-ing. IEEE Trans. Antennas Propag. 68, 4450–4458 (2020).91. Rhodes, D., Chae, S. H., Ribeiro-Palau, R. & Hone, J. Disorder invan der Waals heterostructures of 2D materials. Nat. Mater. 18,541–549 (2019).92. Petrone, N. et al. Chemical vapor deposition-derived graphenewith electrical performance of exfoliated graphene. Nano Lett. 12,2751–2756 (2012).93. Banszerus, L. et al. Ultrahigh-mobility graphene devices fromchemical vapor deposition on reusable copper. Sci. Adv. 1,e1500222 (2015).94. Wang, R. et al. A Peeling approach for integratedmanufacturing oflargemonolayer h-BN crystals. ACS Nano https://doi.org/10.1021/acsnano.8b08712 (2019).95. Ma, K. Y. et al. Epitaxial single-crystal hexagonal boron nitridemultilayers on Ni (111). Nature 606, 88–93 (2022).96. Fukamachi, S. et al. Large-area synthesis and transfer ofmultilayerhexagonal boron nitride for enhanced graphene device arrays.Nat. Electron. 6, 126–136 (2023).97. Montanaro, A. et al.OptoelectronicmixingonCVDgrapheneup to30 Gigahertz: analysis at high electrostatic doping. In: Proc. SPIEvol. 9932 (2016).Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 13https://doi.org/10.1109/MWSYM.1994.335101https://doi.org/10.1109/ICCES45898.2019.9002541https://doi.org/10.1109/ICCES45898.2019.9002541https://doi.org/10.1021/acsnano.8b08712https://doi.org/10.1021/acsnano.8b0871298. Uwanno, T., Hattori, Y., Taniguchi, T., Watanabe, K. & Nagashio, K.Fully dry PMMA transfer of graphene on h -BN using a heating/cooling system. 2d Mater. 2, 041002 (2015).99. Freitag, M., Low, T., Xia, F. & Avouris, P. Photoconductivity ofbiased graphene. Nat. Photonics 7, 53 (2012).100. van Dijk, F. et al. Integrated InP heterodyne millimeter wavetransmitter. IEEE Photon. Technol. Lett. 26, 965–968 (2014).101. Lee, W. et al. Coherent terahertz wireless communication usingdual-parallel MZM-based silicon photonic integrated circuits.Opt.Express 30, 2547 (2022).102. Moon, S.-R. et al. Demonstration of photonics-aided terahertzwireless transmission system with using silicon photonics circuit.Opt. Express 28, 23397 (2020).103. Chinni, V. K. et al. Single‐channel 100Gbit/s transmissionusing III–V UTC‐PDs for future IEEE 802.15.3d wireless linksin the 300GHz band. Electron Lett. 54, 638–640 (2018).104. Harter, T. et al. Generalized Kramers–Kronig receiver forcoherent terahertz communications. Nat. Photonics 14,601–606 (2020).105. Harter, T. et al. Wireless THz link with optoelectronic transmitterand receiver. Optica 6, 1063–1070 (2019).106. Liu, K. et al. 100Gbit/s THz photonic wireless transmission in the350-GHz band with extended reach. IEEE Photon. Technol. Lett.30, 1064–1067 (2018).107. Udalcovs, A. et al. 107.1-Gbps net-rate transmission over a joint51km-fibre-and-10.7m-wireless link for terahertz radio accessnetworks. In: 45th European Conference on Optical Commu-nication (ECOC 2019) 265 (4 pp.)–265 (4 pp.) https://doi.org/10.1049/cp.2019.0999 (Institution of Engineering and Technol-ogy, 2019).108. Li, X. et al. 132-Gb/s Photonics-Aided Single-Carrier WirelessTerahertz-Wave Signal Transmission at 450GHz Enabled by64QAM Modulation and Probabilistic Shaping. In: Optical FiberCommunication Conference (OFC) 2019 M4F.4. https://doi.org/10.1364/OFC.2019.M4F.4 (OSA, 2019).109. Yu, X. et al. 160Gbit/s photonics wireless transmission in the300–500GHz band. APL Photonics 1, 081301 (2016).AcknowledgementsThis work was supported by the European Union’s Horizon 2020research and innovation program, through GrapheneCore3 under grant881603. K.W. and T.T. acknowledge support from the JSPS KAKENHI(Grant Numbers 21H05233 and 23H02052) and World Premier Interna-tional Research Center Initiative (WPI), MEXT, Japan.Author contributionsA.M., V.S., M.R. carried out the device simulation and design. G.P., V.M.,M.G., S.P. fabricated the device. G.P., V.M., S.P. characterized the gra-phene samples. S.S. fabricated the passive photonic circuit. K.W., T.T.grew thehexagonal boronnitride crystals. A.M.,M.R., V.S. conceived theexperiment. A.M., M.R, V.S., A.D., L.G carried out the experimental work.M.R., C.C., S.P., A.D. supervised the work. A.M. wrote the paper, withcontributions from all authors.Competing interestsThe authors declare no competing interests.Additional informationSupplementary information The online version containssupplementary material available athttps://doi.org/10.1038/s41467-023-42194-6.Correspondence and requests for materials should be addressed toAlberto Montanaro.Peer review information Nature Communications thanks Xi Xiao, andthe other, anonymous, reviewer(s) for their contribution to the peerreview of this work. A peer review file is available.Reprints and permissions information is available athttp://www.nature.com/reprintsPublisher’s note Springer Nature remains neutral with regard to jur-isdictional claims in published maps and institutional affiliations.Open Access This article is licensed under a Creative CommonsAttribution 4.0 International License, which permits use, sharing,adaptation, distribution and reproduction in any medium or format, aslong as you give appropriate credit to the original author(s) and thesource, provide a link to the Creative Commons licence, and indicate ifchanges were made. The images or other third party material in thisarticle are included in the article’s Creative Commons licence, unlessindicated otherwise in a credit line to the material. If material is notincluded in the article’s Creative Commons licence and your intendeduse is not permitted by statutory regulation or exceeds the permitteduse, you will need to obtain permission directly from the copyrightholder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.© The Author(s) 2023Article https://doi.org/10.1038/s41467-023-42194-6Nature Communications | (2023) 14:6471 14https://doi.org/10.1049/cp.2019.0999https://doi.org/10.1049/cp.2019.0999https://doi.org/10.1364/OFC.2019.M4F.4https://doi.org/10.1364/OFC.2019.M4F.4https://doi.org/10.1038/s41467-023-42194-6http://www.nature.com/reprintshttp://creativecommons.org/licenses/by/4.0/http://creativecommons.org/licenses/by/4.0/ Sub-THz wireless transmission based on graphene-integrated optoelectronic mixer Results Device design and fabrication Graphene OEM Operating principle and characterization Sub-THz wireless link Discussion Methods Device fabrication Optoelectronic frequency response measurement Conversion efficiency measurement Data availability References Acknowledgements Author contributions Competing interests Additional information